CN115933008A - Strong convection weather forecast early warning method - Google Patents

Abstract

The invention discloses a strong convection weather forecast early warning method, which comprises the following steps: a strong precipitation forecast early warning method, a strong wind forecast early warning method, a thunder forecast early warning method and a typhoon system forecast early warning method; the strong precipitation forecast early warning method comprises short-time strong precipitation disaster identification, short-time strong precipitation 0-2 hour close forecast and short-time strong precipitation 0-6 hours short-time forecast based on real-time high-frequency radar information; the strong wind forecasting and early warning method comprises a strong wind short-term forecasting method and a strong wind middle-term forecasting method; the lightning forecast early warning method comprises a lightning potential forecast method and an approaching lightning forecast method; the typhoon system forecasting and early warning method comprises typhoon positioning and typhoon strength setting. The invention provides technical support and guarantee service for operation and maintenance of a power grid, construction infrastructure, load prediction, power first-aid repair and personnel safety by forecasting and early warning on various strong convection weathers.

Description

Strong convection weather forecast early warning method

Technical Field

The invention belongs to the technical field of weather forecast early warning, and particularly relates to a strong convection weather forecast early warning method.

Background

At present, short-time strong precipitation becomes one of main disasters seriously threatening the safe and stable operation of a power grid, the distribution range of the short-time strong precipitation accounts for as high as 80%, the short-time strong precipitation process is accompanied by thunderstorm, strong wind, hail and other disaster weather, the locality is strong, the space scale is small, the life cycle is short, the destructive power is strong, more easily-occurring secondary disasters exist, power system equipment is often tripped, power is cut off, tower falling is carried out, even accidents such as personal casualties are caused, the short-time strong precipitation has the characteristics of strong explosiveness, concentrated precipitation, short life history and the like, and the short time strong precipitation is only a few minutes, generally about one hour to several hours. Because the short-time heavy precipitation process belongs to a small microscale weather system, most of the existing small microscale numerical modes are boundary layer diagnosis modes, and no cloud accumulation convection scheme exists, so that the heavy precipitation cannot be forecasted; the mesoscale numerical model and the conventional meteorological observation network have limited capturing capability, which brings difficulty to the forecast of heavy rainfall.

For many years, disastrous strong wind threatens the safe operation of a power transmission line of a power grid, and accidents such as pole and tower falling, line breakage, wind deflection, pollution flashover, insulator string separation, hardware breakage and the like of the power transmission line caused by strong wind occur, so that the safety of the power transmission line is guaranteed to be extremely important. The prevention of the accidents needs to establish a strong wind early warning mechanism, which objectively requires that a power grid company must establish a weather monitoring network along the power grid, and can accurately predict the future weather conditions along the power grid, especially the strong wind weather, and make a timely early warning.

In south China, the high winds that cause harmful effects on transmission facilities of 25m/s or more are mainly from two extreme weather conditions: firstly, strong wind brought by a typhoon system on the sea is generated; the second is the generation of strong winds from the inland squall line system. Typhoon has attracted extensive attention and key defense, but the crossing of squall line system is accompanied by sharp wind direction and sharp wind speed, which often reach 20m/s, some even 50m/s, and the destructiveness is equivalent to typhoon wind power. It is also important to investigate the early warning and monitoring of squall line systems causing inland gusts. The squall line wind forecasting method mainly aims at carrying out manifold identification or element forecasting based on a plurality of methods of sounding, ground and radar observation data, doppler weather radar inversion, numerical simulation and the like. The forecasting method is single, and the forecasting result has certain uncertainty.

Lightning can generate huge damage action instantly due to physical effects such as strong current, hot high temperature, strong electromagnetic radiation, violent high-pressure shock waves and the like, so that lightning disasters are caused. Lightning is one of the most frequent natural disasters in the power grid, generally causes line flashover and tripping, and can also cause permanent faults such as insulator falling and wire breakage. The thunder and lightning activity is mainly concentrated in spring and summer and shows certain periodic characteristics along with the change of years. In each region, the south China is most seriously influenced by lightning, particularly, the influence of an ultrahigh voltage important power transmission channel is great, so that the influence of the lightning on a power grid facility needs to be paid high attention.

Disclosure of Invention

The invention mainly aims to overcome the defects of the prior art and provide a strong convection weather forecast early warning method, which can forecast strong precipitation, strong wind, thunder and typhoon, thereby providing technical support and guarantee service for operation and maintenance of a power grid, construction of projects, load prediction, power rush repair and personnel safety.

In order to achieve the purpose, the invention adopts the following technical scheme:

the invention provides a strong convection weather forecast early warning method, which comprises a strong precipitation forecast early warning method, a strong wind forecast early warning method, a thunder forecast early warning method and a typhoon system forecast early warning method;

s1, identifying a short-time heavy precipitation disaster based on real-time high-frequency radar information, carrying out close prediction on the short-time heavy precipitation within 0-2 hours and carrying out short-time prediction on the short-time heavy precipitation within 0-6 hours;

the short-time strong rainfall disaster identification based on the real-time high-frequency radar information is as follows:

acquiring image data of a weather radar and preprocessing the image data to obtain preprocessed radar image data;

extracting radar image data of a target site and nearby from the preprocessed radar image data, and acquiring the space-time scale characteristics of the cloud cluster by combining meteorological observation data;

constructing a strong precipitation identification model based on the cloud cluster space-time scale characteristics and relevant information in radar image data, and obtaining a short-time strong precipitation identification method by combining different cloud cluster space-time scale characteristics of strong precipitation disasters;

the 0-2 hour nowcast of the short-time strong precipitation is as follows:

tracking the moving track of the target cloud cluster in 0-2 hours in the future;

inputting real-time radar data into a preset convolutional neural network model for training to obtain a falling area and precipitation amount of short-time strong precipitation, and tracking by combining the moving track of the target cloud cluster to obtain forecast information of the short-time strong precipitation within 0-2 hours in the future;

the short-term forecast of the short-term heavy precipitation for 0-6 hours is as follows:

identifying a cloud cluster band by using a WRF mode, obtaining space-time scale characteristics and a life and consumption change process, and tracking the cloud cluster band by using an SIFT cloud cluster trajectory technology;

extracting 0-6 hours of short-time heavy precipitation characterization factors exceeding a threshold value and related important factors in a forecast field;

inputting the characterization factors, the relevant important factors and input data into a preset deep learning model to obtain short-term heavy precipitation information, and obtaining short-term heavy precipitation forecast information of 0-6 hours by combining the SIFT cloud cluster track technology;

s2, the strong wind forecasting and early warning method comprises a strong wind short-term forecasting method and a strong wind middle-term forecasting method; the method for forecasting the short term of strong wind comprises the following steps:

respectively predicting the squall line wind based on a radar and predicting the squall line wind based on meteorological station observation data to obtain a squall line wind prediction result based on the radar and a squall line prediction result based on the meteorological station observation data;

according to the squall line wind forecast result based on the radar and the squall line wind imminence forecast result based on the meteorological station observation data, adopting a machine learning algorithm to distribute weight, and performing intensity and space fusion on the two forecast results to realize accurate squall line short-term forecast;

the strong wind mid-term forecasting method comprises the following steps:

acquiring a WRF (squall line wind force) numerical mode to obtain the future changes of various physical quantities in the area;

further analyzing the changes of the physical quantities to obtain the changes of the physical quantities in a lattice point form, and giving a 24-hour forecasting result of the squall line wind in the lattice point form;

correcting a 24-hour forecast result of the squall line wind;

s3, the lightning forecast early warning method comprises a lightning potential forecasting method and an approaching lightning forecasting method;

the lightning potential forecasting method is based on establishing a regression equation of lightning related physical parameters to obtain the occurrence probability of thunderstorms;

correcting by using lightning forecast of real-time data of power grid lightning observation according to the occurrence probability of the thunderstorm, and finally obtaining a lightning potential forecast result;

the method for forecasting the approaching lightning comprises the following steps:

performing lattice localization on the hourly lightning positioning data and the hourly lightning discharge time, and extrapolating the lattice lightning discharge time for 0-6 hours to obtain lattice lightning positioning data, lattice lightning discharge time data and lightning discharge time extrapolated data for 0-6 hours;

extrapolating data of 0-6 hours and short-term prediction results and lattice lightning positioning data according to the lightning discharge time, and establishing a regression model to obtain a lightning short-term prediction result;

correcting the lightning short-term prediction result by using a neural network;

s4, the typhoon system forecasting and early warning method comprises typhoon positioning and typhoon strength setting; the typhoon positioning is to acquire the position of the typhoon through warm center positioning, and the typhoon strengthening is to strengthen the typhoon through a wind cloud 4A new algorithm;

the warm heart is positioned as follows:

judging the data of the loaded channel in the remote sensing picture to obtain remote sensing data information, and acquiring the geographical position near the specified longitude and latitude point;

and acquiring the remote sensing data information of the corresponding range according to the geographical position near the longitude and latitude points, finding the point with the maximum value in the remote sensing data of the corresponding range through calculation, displaying the point with the maximum value on a map, and simultaneously recording the coordinate of the point with the maximum value.

Preferably, the acquiring and preprocessing the image data of the weather radar to obtain the processed radar image data specifically comprises:

respectively extracting image data of a Doppler weather radar, a wind profile radar and a power grid self-built X-band radar, and respectively performing inversion on horizontal wind speed data with different heights, data of a three-dimensional wind field and cloud characteristic data according to the radar image data; performing mutual verification on relevant meteorological element data obtained by radar image inversion and meteorological observation data of an observation point, and removing or correcting radar image data with large errors; integrating the data passing the quality check, selecting a fusion algorithm, carrying out standardized processing according to a uniform format, and finally realizing the storage of structured data of the radar image;

the method comprises the following steps of extracting radar image data of a target site and nearby from preprocessed image data, and acquiring space-time characteristics of a cloud cluster by combining meteorological observation data, wherein the method specifically comprises the following steps:

acquiring radar image data of a target site and nearby sites from the radar image data, and selecting radar images with uniform resolution and observation contents from the radar image data; combining meteorological observation data, searching for coordinate correlation characteristics among samples by a template matching mode by utilizing overlapped parts existing among local radar image samples, splicing a series of radar images in a small range to radar images in a wider space, acquiring space-time scale characteristics of radar reflectivity factors of short-time strong rainfall disasters from the whole radar images, and further acquiring space-time characteristics of clouds;

the strong precipitation identification model is constructed by the following steps:

the method comprises the steps of extracting reflectivity factor information from radar image data in a large number of heavy precipitation processes, researching the relation between radar reflectivity factors and precipitation intensity and cloud cluster characteristics, constructing a heavy precipitation identification model, analyzing different space-time scale characteristics of heavy precipitation disasters, and providing a short-time heavy precipitation disaster identification method based on the different space-time scale characteristics.

Preferably, the tracking of the target cloud cluster movement track in 0-2 hours in the future is based on the Taylor freezing hypothesis, the SIFT technology is applied, the time prediction problem of the cloud cluster is converted into the local space prediction problem, and the tracking of the target cloud cluster movement track in 0-2 hours in the future is realized;

inputting real-time radar data into a preset convolutional neural network model for training, specifically:

because the elements of the precipitation, the wind field and the cloud cluster of the atmosphere are continuous functions related to space-time, the three-layer network can approach any continuous function with any precision according to the Robert Hecht-Nielsen theorem; therefore, a hidden layer, namely a 3-layer network is selected to complete the nonlinear mapping of the precipitation, the wind field and the cloud cluster elements; initializing all mapping weight initial values by using different small random numbers to ensure that the network does not enter a saturation state due to overlarge weight;

in the training process, a radar image part is subjected to multilayer convolution pooling, then vectors are leveled to one dimension to obtain radar image characteristics, the radar image characteristics are combined with other non-image characteristics in a full connection layer, and the radar image characteristics and the other non-image characteristics are input into a neural network of 3 hidden layers together;

in the backward propagation process, as the depth of the network increases, the amplitude value of the gradient from the output layer to the first few layers of the network is sharply reduced, namely, the overfitting problem is solved, so that dropout is adopted to prevent overfitting, and an Adam optimization algorithm is adopted for gradient descent.

Preferably, the WRF pattern is used to identify the cloud band, specifically:

extracting multilayer cloud water, cloud ice, rainwater, snow and aragonite data of more than 850hPa in a WRF mode 0-6 hour forecasting field, adding grid points, setting a first threshold value of water-substance ratio quality in a cloud area, wherein the cloud area water-substance ratio quality is larger than the first threshold value and is judged to be cloud, otherwise, the cloud area is judged to be cloud-free, so that a cloud cluster band is identified, space-time scale features and a life and consumption change process are obtained, and then the cloud cluster band is tracked by using an SIFT method;

the extraction of the short-time strong precipitation characterization factors exceeding the threshold value in 0-6 hours in the forecast field and related important factors is based on high-resolution forecast results in a historical WRF mode, and characterization factors of short-time strong precipitation in a deep WRF mining mode, such as local wind speed and violent rise of thunderstorm potential indexes; extracting several important elements of the short-time heavy precipitation, including a vertical wind field and a water vapor field, combining a characterization factor and a second threshold value of the important elements in the short-time heavy precipitation process obtained by statistics of data in a historical forecast result, and extracting the short-time heavy precipitation characterization factor exceeding the second threshold value in 0-6 hours in the forecast field and relevant important elements;

the characterization factors and the relevant important factors comprise short-time strong precipitation characterization factors, relevant important factors, space-time feature scales and life cycle matching features.

Preferably, the radar-based prediction for the squall line wind includes:

radar base data filtering processing: radar base data are adopted, and multilayer echoes with different elevation angles in the radar base data are utilized to carry out three-dimensional mathematical characteristic filtering processing on radar images, so that clutter filtering of the radar is more effective; meanwhile, a spatial interpolation extension method of multilayer elevation echoes is adopted to make up missing data in a weather radar image to obtain a first radar image; and (3) radar echo prediction: quantitatively analyzing the motion trail of the radar echo by adopting a leading-edge optical flow method technology for the first radar image, calculating an optical flow field of the radar echo to obtain a motion vector field of the echo, extrapolating the radar echo based on the motion vector field so as to infer the mobile evolution condition of the squall line wind area, and performing 0-2 hour proximity prediction on the squall line wind area to obtain a predicted second radar image;

radar data post-processing: the predicted second radar image has certain loss in smoothness and continuity, so that post-processing operation is required to be carried out on radar data; firstly, a closing operation is needed to complete cracks and holes, and secondly, radar echoes are processed to a certain extent according to different large area forecasts; if the weather system is a single thunderstorm strong convection weather system, the echo edge is clear and obvious in gradient, the edge diffusion is reduced through corrosion operation, otherwise, the radar echo range is enlarged through expansion operation, and low threshold information filtered out in the preprocessing of the edge is supplemented, so that the forecasting accuracy is improved;

identification of narrow-band echo and convergence line: calculating the two-way gradient of each echo point in the radar image after radar data processing to reserve linear echo, but designing a plurality of models with different included angles in a specific algorithm because the included angles between narrow-band echo and radial line cannot be known; most precipitation echoes of the intensity field after the bidirectional gradient processing are filtered, narrow-band echoes are completely reserved, but a plurality of short lines still exist in the image; in order to remove the short lines, firstly thinning the image, namely only keeping the central point of each intensity segment, recording the width of the lower segment, then calculating the length of each short line by using a recursive algorithm, and filtering the short lines which do not meet a length threshold; in the process, the azimuth angle and the radial library number of each effective point are saved, and finally, only a short line with a certain length is reserved, and the previously recorded segment width is restored into a band shape to form a final narrow-band echo identification image;

the squall line wind intensity mapping and lattice time series: respectively establishing a machine learning regression model for squall line wind in different areas at different moments according to the formed final narrow-band echo identification image to obtain an echo-wind real-time mapping relation matched with climate characteristics of each area, and realizing prediction of the squall line wind area based on the radar;

the squall line wind forecasting method based on meteorological station observation data specifically comprises the following steps:

weather station data quality control: carrying out quality control on observation data in a meteorological station, and carrying out machine learning on an air pressure surge parameter, an air pressure distance flat field parameter, a temperature parameter, a false equivalent temperature parameter, a humidity change discontinuous parameter and a rain group activity parameter in the observation data to obtain data after multi-parameter learning;

multi-parameter learning: performing lattice localization on the data after the multi-parameter learning to obtain lattice-localized multi-parameters;

and (3) wind field extrapolation: extrapolating the grid multi-parameters in a wind field to infer the mobile evolution condition of the squall line wind area to obtain predicted squall line wind data based on the meteorological station;

and (3) data post-processing: the data post-processing comprises the following specific steps:

(1) And (3) meteorological site data arrangement: inputting field names including site names, x longitude lon, y latitude lat, average air temperature, average wind speed, relative humidity and average sunshine hours; wherein, the longitude and latitude need to be converted into a form of degree, and other data are converted into corresponding units;

(2) Carrying out interpolation analysis on the vector point data converted into shape format;

(3) Exporting point data in shape format;

(4) Setting an Arcgis environment;

(5) Data interpolation of meteorological stations;

predicting a grid point wind speed time sequence: predicting an hourly lattice point wind speed sequence, and finally realizing the prediction of squall line wind based on meteorological station observation data;

the method for distributing the weight by using the machine learning algorithm and performing intensity and space fusion on the two prediction results to realize accurate squall line wind short-term prediction comprises the following steps of:

keeping consistent data in the results of the squall wind position prediction based on radar and meteorological station observation data, and taking intersection of the results of the differences of the two prediction methods; a binary tree search algorithm is adopted to accelerate data search so as to improve the fusion speed; the binary tree search algorithm is to generate a binary search tree for the data elements to be searched, then compare the given value with the keywords of the root node, if the given value is equal to the keywords of the root node, the search is successful, otherwise, the search is continued in the left sub-tree and the right sub-tree according to the keywords of the given value smaller than or larger than the keywords of the root node until the search is successful or the left sub-tree or the right sub-tree is an empty tree.

Preferably, the squall line strong wind short prediction method further comprises squall line strong wind identification and location, wherein the squall line strong wind identification and location specifically comprises:

the squall line strong wind is represented as the summation of radial speeds on a radar velocity map, which is mainly represented by the summation of wind speeds, and the speed value of the radial speed is converted from a higher value to a lower value; therefore, according to the characteristic, a group of adjacent distance libraries with continuously reduced radial velocity values are searched along the increasing direction of the radial distance to obtain a one-dimensional convergent section;

to describe the intensity of each one-dimensional convergent section, the following physical quantities are calculated: velocity gradient g, momentum f, orientation, radial center;

if the speed gradient g or momentum f of a certain convergent section is lower than a set lower limit value, deleting the one-dimensional convergent section, and otherwise, carrying out high value inspection on the one-dimensional convergent section; if either the value of g or f of the converged segment is greater than the set upper limit, the one-dimensional converged segment is saved;

judging the difference between the radial center and the position of the current one-dimensional convergent section and the next one-dimensional convergent section in all the stored one-dimensional convergent sections, if the difference is greater than a set radial distance threshold or a set position threshold, deleting the one-dimensional convergent section, and otherwise, keeping the one-dimensional convergent section;

and combining all the one-dimensional convergent sections meeting the conditions into two-dimensional characteristics, filtering the characteristics with too small number of convergent sections, and forming the finally identified convergent line switching line.

Preferably, the 24-hour forecasting result of the squall line wind is given in a lattice form, and specifically includes:

adopting NCAR, NCEP and FSL/NOAA to jointly develop a developed fine grid mesoscale WRF numerical mode; the WRF numerical mode continuously integrates from the initial moment to the next moment according to the initial condition and the boundary condition and the given physical parameterization process, the change of each future physical quantity in the area is obtained through forecasting, the concerned wind field data is extracted from the changes and further analyzed, and finally the 24-hour forecasting result of the squall line wind is given in a lattice mode;

correcting the 24-hour forecast result of the squall line wind, including: the results of the live observation data and the wrf numerical prediction mode are fused and corrected, and the correction is carried out based on the remote sensing terrain coefficient;

after the live observation data and the results of the wrf numerical prediction mode are fused, correcting the results into two aspects of phase correction and intensity correction of the numerical prediction wind field; the phase correction of the wind field comprises the steps of acquiring a total phase translation field by adopting fast Fourier transform and adjusting the regional phase by utilizing a multi-scale optical flow variation method; the correction of the wind field intensity mainly utilizes a cumulative distribution function of the Weber distribution;

the remote sensing-based terrain coefficient correction specifically comprises the following steps: combining a meteorological numerical simulation algorithm of a remote sensing terrain correction coefficient, matching a high-precision topographic map on the basis of traditional numerical weather forecast, and improving the forecast accuracy and resolution of strong wind by using a dynamic or statistical downscaling technology or CFD simulation;

the CFD simulation calculation process is mainly divided into three parts, namely a pretreatment part, a simulation calculation part and a post-treatment part; the pre-processing part comprises acquisition of topographic data, topographic modeling and grid generation, which are prerequisites for wind field CFD simulation, the WRF simulation result is used as a CFD boundary condition, the simulation calculation part comprises boundary condition setting, turbulence model setting, discrete format setting, solver setting and initialization solving steps, which are main solving processes of wind field CFD simulation, and the post-processing part comprises acquisition and analysis of simulation results.

Preferably, the establishing of the regression equation based on the physical parameters related to the lightning obtains the probability of occurrence of the thunderstorm, which specifically includes:

through the relation of the correlation coefficients, 7 physical quantities with the highest correlation coefficient with lightning are finally screened out, wherein the physical quantities are k index, sand index SI, A index, 700hPa temperature dew point difference, 850hPa temperature dew point difference, 925hPa temperature dew point difference, 850hPa temperature difference and 500hPa temperature difference; according to the mesoscale strong convection generation condition and the statistics of a large amount of data, the threshold values of the physical parameters are processed into 0 and 1, and the threshold values are set as follows:

when the k index is more than or equal to 33, marking as 1, otherwise, marking as 0;

when the sand index is less than or equal to 0, recording as 1, otherwise recording as 0;

when the index A is larger than or equal to 10, marking as 1, otherwise, marking as 0;

when the temperature dew point difference is less than or equal to 3, the temperature dew point difference is marked as 1, otherwise, the temperature dew point difference is marked as 0, and 700hPa, 850hPa and 925hPa are all used as threshold values;

when the temperature difference between 850hPa and 500hPa is not less than 23, the value is marked as 1, and otherwise, the value is marked as 0;

finally, 7 indexes subjected to 0 and 1 chemical treatment are used as core physical parameters of potential prediction, and a regression equation of the lightning potential prediction is established; the regression equation of the lightning potential forecast is as follows:

y(j，i)＝0.11+0.154×k1(j，i)+0.142×k2(j，i)+0.061×k3(j，i)+0.146×k4(j，i)+0.131×k5(j，i)+0.03×k6(j，i)+0.097×k7(j，i)

wherein y is the probability of thunderstorm occurrence at a certain site, K1 \8230andK 7 is the physical parameter after 0 and 1;

correcting by using the lightning forecast of the real-time data of power grid lightning observation according to the occurrence probability of the thunderstorm to finally obtain a lightning potential forecast result, which specifically comprises the following steps:

taking the maximum value of each grid point of the radar echo fields at 4 moments corresponding to 3-hour time intervals one by one to obtain the maximum echo field MR within 3 hours;

smoothing the MR, the CP and the LSP for multiple times to obtain parameters of the SMR, the SCP and the SLSP, wherein the CP represents 3-hour convective precipitation, and the LSP represents 3-hour large-scale precipitation;

and calculating a prediction result by adopting a decision tree model according to the SMR, the SCP and the SLSP, marking as P _ dt, and finally smoothing the prediction result P _ dt for dh times to finally output a lightning potential prediction result SP _ dt.

Preferably, the extrapolation of the data of 0-6 hours according to the lightning discharge time and the short-term prediction result and the lattice lightning location data is used for establishing a regression model, and the extrapolation of the data of 0-6 hours according to the lightning discharge time and the short-term prediction result SP _ dt are used as input, and the lattice lightning location data of hour by hour is used as output;

the correcting is carried out on the lightning short-term forecast result by utilizing a neural network, and specifically comprises the following steps:

extracting basic weather information: firstly, extracting basic meteorological elements according to historical sounding data, wherein the basic meteorological elements comprise temperature, dew point temperature, wind direction and wind speed and relative humidity;

strong weather convection index: calculating a strong convection index from the numerical pattern results, comprising: the system comprises a Sas index, a convection effective potential energy, a strong weather threat index, a total index and a convection inhibition index;

characteristic screening and model building: before fitting of a neural network model, selecting a characteristic, selecting the characteristic by using a regression model, and fitting the model based on a neural network method;

identifying and tracking the thunderstorm: based on radar and lightning data, extracting morphological characteristics of the thunderstorm, tracking and extrapolating and forecasting the evolution and motion trail of the thunderstorm area; combining lightning positioning data, focusing on a set area, and judging auxiliary indexes of horizontal gradient and vertical liquid water content of an echo;

and (3) generating a thunderstorm nowcasting: and (3) forecasting the lightning occurrence probability of all grid points in the area based on the neural network model, and correcting the area where the lightning is located according to the storm identified by the radar chart through a weighting algorithm to obtain the final lightning short-term prediction.

Preferably, the typhoon is strengthened through the wind cloud 4A new algorithm, which specifically comprises:

finding a value b which is smaller than a set range in the range of r in the satellite data at the time t, wherein the maximum value bmax and the minimum value bmin meet the number n of the conditional effective points;

acquiring a distance dmin and a distance dmax of the farthest point from the nearest point of (lon, lat) and an average value davg of distances between all points and (lon, lat) in a condition range, wherein lon is longitude and lat is latitude;

solving a curve coefficient dindex and a wind speed v;

calculating an air pressure value according to the relation between the wind speed and the air pressure; and (4) according to the calculated air pressure value, strengthening the typhoon.

Compared with the prior art, the invention has the following advantages and beneficial effects:

according to the method, various strong convection weathers are forecasted, each strong convection weather comprises a plurality of forecasting modes, and forecasting results can be corrected by combining different forecasting modes, so that forecasting is more accurate, and technical support and guarantee services are provided for operation and maintenance of a power grid, construction, load forecasting, power rush repair and personnel safety; meanwhile, the instability of the tower foundation landslide is comprehensively considered from two aspects of time and space, the risk coefficient of the landslide where the tower is located at present can be obtained, and guiding suggestions are provided for landslide disaster early warning and prevention.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the description below are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.

FIG. 1 is a flow chart of meteorological data acquisition and preprocessing according to an embodiment of the present invention;

FIG. 2 is a block diagram of a radar-based squall wind prediction in accordance with an embodiment of the present invention;

FIG. 3 is a block diagram of a squall wind imminence prediction based on meteorological station observation data in accordance with an embodiment of the present invention;

FIG. 4 is a block diagram of squall wind imminence prediction based on radar and meteorological station prediction fusion corrections, in accordance with an embodiment of the present invention;

FIG. 5 is a design diagram of a lightning potential forecasting algorithm according to an embodiment of the invention.

Detailed Description

In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. It is to be understood that the embodiments described are only a few embodiments of the present application and not all embodiments. All other embodiments obtained by a person skilled in the art based on the embodiments in the present application without making any creative effort belong to the protection scope of the present application.

Reference in the specification to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the specification. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.

The embodiment provides a strong convection weather forecast early warning method, which comprises a strong precipitation forecast early warning method, a strong wind forecast early warning method, a thunder forecast early warning method and a typhoon system forecast early warning method, and the specific steps are as follows:

s1, identifying a short-time heavy precipitation disaster based on real-time high-frequency radar information, carrying out close prediction on the short-time heavy precipitation within 0-2 hours and carrying out short-time prediction on the short-time heavy precipitation within 0-6 hours;

s11, identifying the short-time heavy rainfall disaster in the S1 as follows:

further, please refer to fig. 1, image data of a doppler weather radar, a wind profile radar and a power grid self-built X-band radar are respectively extracted, and horizontal wind speed data of different heights, data of a three-dimensional wind field and cloud feature data are respectively inverted according to the radar image data; performing mutual verification on relevant meteorological element data obtained by radar image inversion and meteorological observation data of an observation point, and removing or correcting radar image data with large errors; integrating the data passing the quality audit, selecting a fusion algorithm, carrying out standardized processing according to a uniform format, and finally realizing the storage of structured data of the radar image;

acquiring radar image data of a target site and nearby sites from the radar image data, and selecting radar images with uniform resolution and observation contents from the radar image data; combining meteorological observation data, searching for coordinate correlation characteristics among samples by a template matching mode by utilizing overlapped parts existing among local radar image samples, splicing a series of radar images in a small range to radar images in a wider space, acquiring space-time scale characteristics of radar reflectivity factors of short-time strong rainfall disasters from the whole radar images, and further acquiring space-time scale characteristics of clouds;

classifying and sorting relevant observation data of the short-time strong precipitation process in historical observation data of Guangzhou area according to the duration time of the short-time strong precipitation disaster process and the space range of the cloud cluster; through statistical analysis of a large number of samples, different space-time scale characteristics of the short-time heavy rainfall process are researched; extracting reflectivity factor information from radar image data in a large amount of heavy rainfall processes, researching the relation between radar reflectivity factors and rainfall intensity and cloud cluster characteristics, identifying a model for heavy rainfall, analyzing different time-space scale characteristics of the heavy rainfall disaster, and providing an identification method for the short-time heavy rainfall disaster based on the different time-space scale characteristics;

s12, the method for 0-2 hour nowcasting of short-time strong precipitation in the S1 specifically comprises the following steps:

based on the Taylor freezing hypothesis, the SIFT technology is applied to convert the time prediction problem of the cloud cluster into the local space prediction problem, and the target cloud cluster movement track tracking of 0-2 hours in the future is realized;

inputting real-time radar data into a preset convolutional neural network model for training to obtain a falling area and precipitation information of short-time strong precipitation, and tracking by combining a moving track of a target cloud cluster to obtain short-time strong precipitation forecast information within 0-2 hours in the future; the radar data comprise precipitation, wind field and cloud cluster elements;

the real-time radar data are input into a preset convolutional neural network model for training, and the training method specifically comprises the following steps:

because the elements of the precipitation, the wind field and the cloud cluster of the atmosphere are continuous functions related to space and time, any continuous function is approached with any precision according to the Robert Hecht-Nielsen theorem, namely a three-layer network implementation; therefore, a hidden layer, namely a 3-layer network is selected to complete the nonlinear mapping of the precipitation, the wind field and the cloud cluster elements; initializing all mapping weight initial values by using different small random numbers to ensure that the network does not enter a saturation state due to overlarge weight;

in the training process, a radar image part is subjected to multilayer convolution pooling, then vectors are leveled to one dimension to obtain radar image characteristics, the radar image characteristics are combined with other non-image characteristics in a full connection layer and are jointly input into a neural network with 3 hidden layers;

in the backward propagation process, as the depth of the network increases, the amplitude value of the gradient from the output layer to the first few layers of the network is sharply reduced, namely, the overfitting problem is solved, so that dropout is adopted to prevent overfitting, and an Adam optimization algorithm is adopted for gradient descent.

S13, short-term forecast of the short-term strong precipitation for 0-6 hours is specifically as follows:

extracting multi-layer cloud water, cloud ice, rainwater, snow and aragonite data of more than 850hPa in a WRF mode 0-6 hour forecasting field, adding grid points, taking the minimum mass of water/kg as a cloud area judgment threshold value, taking the mass of water matter of 0.02g/kg as cloud area judgment threshold value, taking the mass of water matter of more than 0.02g/kg as cloud, and taking the mass of water matter of 0.02g/kg as cloud area judgment threshold value, and taking the mass of water matter of no cloud area judgment threshold value as cloud area judgment threshold value, so as to identify cloud cluster bands, obtain the space-time scale characteristics and the generation and elimination change process of the cloud cluster bands, and then tracking the cloud cluster bands by using an SIFT method.

Based on the high-resolution forecast result in the historical WRF mode, characterizing factors of short-time strong rainfall in the deep WRF mining mode, such as local wind speed and violent rise of thunderstorm potential index; extracting several important elements of the short-time heavy precipitation, including a vertical wind field and a water vapor field, combining a characterization factor and a second threshold value of the important elements in the short-time heavy precipitation process obtained by statistics of data in a historical forecast result, and extracting the short-time heavy precipitation characterization factor exceeding the second threshold value in 0-6 hours in the forecast field and relevant important elements;

inputting the short-term strong precipitation representation factor, relevant important factors, space-time scale features, life cycle matching features and input data into a preset deep learning model to obtain short-term strong precipitation information, and then combining a cloud cluster track and a life-saving process forecasting technology based on WRF to obtain short-term strong precipitation forecasting information within 0-6 hours. The input data includes other relevant data that can enhance the predictive algorithm model.

S2, the strong wind forecasting and early warning method comprises a strong wind short-term forecasting method and a strong wind middle-term forecasting method;

s21, the strong wind short-term forecasting method comprises the following steps:

respectively predicting the squall line wind based on a radar and predicting the squall line wind based on meteorological station observation data to obtain a squall line wind prediction result based on the radar and a squall line prediction result based on the meteorological station observation data;

according to the squall line wind forecasting result based on the radar and the squall line wind forecasting result based on the meteorological station observation data, distributing weight by adopting a machine learning algorithm, and performing intensity and space fusion on the two forecasting results to realize accurate squall line short-term forecasting;

further, referring to fig. 2, the radar-based prediction of squall wind may include:

radar base data filtering processing: radar base data are adopted, and multilayer echoes with different elevation angles in the radar base data are utilized to carry out three-dimensional mathematical characteristic filtering processing on radar images, so that clutter filtering of the radar is more effective; meanwhile, a spatial interpolation extension method of multilayer elevation echoes is adopted to make up missing data in a weather radar image to obtain a first radar image; and (3) radar echo prediction: quantitatively analyzing the motion trail of the radar echo by adopting a leading-edge optical flow method technology for the first radar image, calculating an optical flow field of the radar echo to obtain a motion vector field of the echo, extrapolating the radar echo based on the motion vector field so as to infer the mobile evolution condition of the squall line wind area, and performing 0-2 hour proximity prediction on the squall line wind area to obtain a predicted second radar image;

radar data post-processing: the predicted second radar image has certain loss in smoothness and continuity, so that post-processing operation is required to be carried out on radar data; firstly, a closing operation is needed to complete cracks and holes, and secondly, radar echoes are processed to a certain extent according to different large area forecasts; if the weather system is a single thunderstorm strong convection weather system, the echo edge is clear and has obvious gradient, the edge diffusion is reduced through corrosion operation, otherwise, the radar echo range is enlarged through expansion operation, and low threshold information filtered out in the preprocessing of the edge is supplemented, so that the forecasting accuracy is improved.

Identification of narrow-band echo and convergence line: calculating the two-way gradient of each echo point in the radar image after radar data processing to reserve linear echo, but designing a plurality of models with different included angles in a specific algorithm because the included angles between narrow-band echo and radial line cannot be known; most precipitation echoes of the intensity field after the bidirectional gradient processing are filtered, narrow-band echoes are completely reserved, and a plurality of short lines still exist in the image; in order to remove the short lines, firstly thinning the image, namely only keeping the central point of each intensity segment, recording the width of the lower segment, then calculating the length of each short line by using a recursive algorithm, and filtering the short lines which do not meet a length threshold; in the process, the azimuth angle and the radial library number of each effective point are saved, and finally, only a short line with a certain length is reserved, and the previously recorded segment width is restored into a belt shape to form a final narrow-band echo identification image.

Squall line wind intensity mapping and lattice time series: and respectively establishing a machine learning regression model for squall line winds in different areas at different moments according to the formed final narrow-band echo identification image to obtain an echo-wind real-time mapping relation matched with climate characteristics of each area, so as to realize prediction of the squall line wind area based on the radar.

Further, referring to fig. 3, the predicting the squall wind based on the meteorological station observation data specifically includes:

weather station data quality control: carrying out quality control on observation data in a meteorological station, and carrying out machine learning on an air pressure surge parameter, an air pressure distance flat field parameter, a temperature parameter, a false equivalent temperature parameter, a humidity change discontinuous parameter and a rain group activity parameter in the observation data to obtain data after multi-parameter learning;

the air pressure surge parameter: based on the characteristic that squall lines are followed by high pressures from a mesoscale thunderstorm, it is known that the barometric pressure perturbation in such a system may be up to 1hPa per ten minutes, whereas the change in barometric pressure at the weather scale may well exceed 1hPa per hour unless a typhoon is imminent. Therefore, among the various changes induced in the squall line as the wind passes, discrete surges in air pressure associated with steep temperature drops and wind and humidity changes may be readily identified, and the squall line position may be determined using such changes in air pressure surges.

The air pressure distance flat field parameter: taking the average value of the air pressure of the measuring station in the squall line wind influence period in each hour as a reference to obtain an air pressure distance flat field in each hour; the movements of the squall line in the barometric positive pitch flat zone are substantially the same as the movements of the squall line, and the movements of the thunderstorm high pressure after the squall line can be well reflected only after the squall line is slightly deviated from the position.

The temperature and the pseudo-equivalent temperature parameter are as follows: the rear portion of the squall line typically has low tongue wedging that is at a temperature or a false equivalent temperature that is more sensitive than the temperature.

The humidity variation discontinuity parameter: the relative humidity change curves of the substations before and after the squall line wind crosses the border will form a root shape, namely, a small valley appears first, which is caused by the sinking airflow at the front part of the thunderstorm high pressure, and then the relative humidity is sudden due to the influence of precipitation.

The rain group activity parameters are as follows: tracking the activities of the rain masses with a 10mm rainfall contour per hour may discover that there are most rain mass activities in the area near the period of squall line wind activity.

Multi-parameter learning: performing lattice localization on the data after the multi-parameter learning to obtain lattice-localized multi-parameters;

and (3) wind field extrapolation: extrapolating the grid multi-parameters in a wind field to infer the mobile evolution condition of the squall line wind area to obtain predicted squall line wind data based on the meteorological station;

and (3) data post-processing: the data post-processing comprises the following specific steps:

(1) And (3) meteorological site data arrangement: and inputting field names including site names, x longitude lon, y latitude lat, average air temperature, average wind speed, relative humidity and average sunshine hours. Wherein, the longitude and latitude need to be converted into a form of degree, and other data are converted into corresponding units;

(2) Carrying out interpolation analysis on the vector point data converted into shape format;

(3) Exporting point data in a shape format;

(4) Setting an Arcgis environment;

(5) Data interpolation of meteorological stations;

predicting a grid point wind speed time sequence: predicting hourly grid wind speed sequences, such as: currently 9 points, forecasting the wind speed of 3h in the future, wherein 10; finally, forecasting squall line wind based on meteorological station observation data is realized;

the method adopts a machine learning algorithm to distribute weight, and performs intensity and space fusion on two prediction results to realize accurate squall line wind short-term prediction, which specifically comprises the following steps:

referring to FIG. 4, since the results of the squall line wind prediction based on radar and the results of the squall line wind prediction based on the weather stations each employ different data sources, the two sets of predictions may differ in result; fusing results of squall wind forecasting based on a radar and squall wind forecasting based on meteorological station observation data in order to fully utilize each effective data as much as possible and improve the accuracy of squall wind forecasting; under different meteorological conditions, the forecasting effects of the radar data and the meteorological station observation data on the squall wind have certain difference, the machine learning algorithm is adopted to distribute the weight, and the intensity and the space of the two forecasting results are fused, so that the uncertainty of a single method is reduced, and the forecasting accuracy is improved. The radar and the automatic station have two conditions of consistency and inconsistency in the results of squall line wind position prediction, the consistent data in the two prediction results is retained, and the intersection of the results with the difference of the two prediction methods is taken. The fusion of the two different observation data sets can involve a large amount of information search, and a high requirement is put on the fusion speed; in order to solve the problem of rapid fusion of two different types of big data, a binary tree search algorithm is adopted to accelerate data search so as to improve the fusion speed; the squall wind concept based on meteorological station observation data is that a binary search tree is generated for data elements to be searched, then a given value is compared with keywords of a root node, if the given value is equal to the keywords of the root node, the search is successful, otherwise, the search is continued in a left sub-tree and a right sub-tree according to the keywords of the given value smaller than or larger than the root node until the search is successful or the left sub-tree or the right sub-tree is an empty tree.

S22, the squall line strong wind short forecasting method further comprises squall line strong wind identification and positioning, wherein the squall line strong wind identification and positioning specifically comprises the following steps:

the squall line strong wind is represented as the summation of radial speeds on a radar velocity map, which is mainly represented by the summation of wind speeds, and the speed value of the radial speed is converted from a higher value to a lower value; thus, according to this feature, a set of contiguous distance bins of successively decreasing radial velocity values is found along the direction of radial distance increment, resulting in a one-dimensional convergence segment.

To describe the intensity of each one-dimensional convergent section, the following physical quantities are calculated: velocity gradient g, momentum f, orientation, radial center.

If the speed gradient g or momentum f of a certain convergent section is lower than a set lower limit value, deleting the one-dimensional convergent section, and otherwise, carrying out high value inspection on the one-dimensional convergent section; if either the value of g or f for the summed segment is greater than the set upper limit, the one-dimensional summed segment is saved.

And judging the difference between the radial center and the position of the current one-dimensional convergent section and the next one-dimensional convergent section in all the stored one-dimensional convergent sections, if the difference is greater than a set radial distance threshold or a set position threshold, deleting the one-dimensional convergent section, and otherwise, keeping the one-dimensional convergent section.

And combining all the one-dimensional radial sections meeting the conditions into two-dimensional characteristics, and filtering out the characteristics with too small number of radial sections to form the finally identified radial line switching line.

S23, the strong wind middle-period forecasting method specifically comprises the following steps:

a fine-grid mesoscale WRF numerical mode developed by adopting NCAR, NCEP and FSL/NOAA in a combined development manner; the WRF numerical mode continuously integrates from the initial moment to the next moment according to the initial condition and the boundary condition and a given physical parameterization process, forecasts and obtains the change of each future physical quantity in the area, extracts concerned wind field data from the changes for further analysis, and finally gives a 24-hour forecasting result of the squall line wind in a lattice point form;

correcting the 24-hour forecast result of the squall line wind, including: the results of the live observation data and the wrf numerical prediction mode are fused and corrected, and the correction is carried out based on the remote sensing terrain coefficient;

and correcting after fusing the results of the live observation data and the wrf numerical prediction mode: in the process of forecasting through the wrf numerical forecasting mode, errors relative to a real field are generated every time when the mode carries out the next time stratification time, so that the forecasting result needs to be further corrected to meet the actual business needs. Therefore, on the basis of improving numerical prediction, the results of live observation data and numerical prediction are fused and corrected. The post-fusion correction of the numerical prediction can be divided into two aspects of phase correction and intensity correction of a numerical prediction wind field. The phase correction of the wind field comprises the steps of acquiring a total phase translation field by adopting fast Fourier transform and adjusting the region phase by utilizing a multi-scale optical flow variation method. And the correction of the wind field strength is mainly by utilizing the cumulative distribution function of the Weber distribution. Comparing the extrapolation method and the numerical prediction result which only use radar data, and based on the fusion post-correction method of numerical prediction and real-time observation data, the strong convection prediction result is closer to the actual situation in scope and strength.

The remote sensing-based terrain coefficient correction specifically comprises the following steps: the accuracy and the resolution of the forecast of strong wind are improved by combining a meteorological numerical simulation algorithm of a remote sensing terrain correction coefficient, matching a high-precision topographic map on the basis of the traditional numerical weather forecast and utilizing a dynamic or statistical downscaling technology or CFD simulation.

The CFD simulation calculation process is mainly divided into three parts, namely a pretreatment part, a simulation calculation part and a post-treatment part; the pre-processing part comprises acquisition of topographic data, topographic modeling, grid generation and the like, which are prerequisites for wind field CFD simulation, the WRF simulation result is used as a CFD boundary condition, the simulation calculation part comprises boundary condition setting, turbulence model setting, discrete format setting, solver setting and initialization solving steps, which are main solving processes of wind field CFD simulation, and the post-processing part comprises acquisition and analysis of simulation results.

S3, the lightning forecast early warning method comprises a lightning potential forecasting method and an approaching lightning forecasting method;

s31, establishing a regression equation based on physical parameters related to thunder and lightning to obtain the probability of occurrence of the thunderstorm;

correcting by using lightning forecast of real-time data of power grid lightning observation according to the occurrence probability of the thunderstorm, and finally obtaining a lightning potential forecast result;

further, referring to fig. 5, the establishing a regression equation based on the physical parameters related to lightning to obtain the probability of occurrence of thunderstorms specifically includes:

through the relation of the correlation coefficients, 7 physical quantities with the highest correlation coefficient with lightning are finally screened out, wherein the physical quantities are k index, sand index SI, A index, 700hPa temperature dew point difference, 850hPa temperature dew point difference, 925hPa temperature dew point difference, 850hPa temperature difference and 500hPa temperature difference; according to the mesoscale strong convection occurrence condition and a large amount of data statistics, the threshold values of the physical parameters are processed into 0 and 1, and the threshold values are set as follows:

when the k index is more than or equal to 33, marking as 1, otherwise, marking as 0;

when the sand index is less than or equal to 0, marking as 1, otherwise, marking as 0;

when the index A is larger than or equal to 10, marking as 1, otherwise, marking as 0;

when the temperature dew point difference is less than or equal to 3, the temperature dew point difference is marked as 1, otherwise, the temperature dew point difference is marked as 0, and 700hPa, 850hPa and 925hPa are all used as threshold values;

when the temperature difference between 850hPa and 500hPa is not less than 23, the value is marked as 1, and otherwise, the value is marked as 0;

finally, 7 indexes subjected to 0 and 1 chemical treatment are used as core physical parameters of potential prediction, and a regression equation of the lightning potential prediction is established; the regression equation of the lightning potential forecast is as follows:

y(j，i)＝0.11+0.154×k1(j，i)+0.142×k2(j，i)+0.061×k3(j，i)+0.146×k4(j，i)+0.131×k5(j，i)+0.03×k6(j，i)+0.097×k7(j，i)

wherein y is the probability of thunderstorm occurrence at a certain site, K1 \8230andK 7 is the physical parameter after 0 and 1;

correcting by using the lightning forecast of the real-time data of power grid lightning observation according to the occurrence probability of the thunderstorm to finally obtain a lightning potential forecast result, which specifically comprises the following steps:

taking the maximum value of each grid point of the radar echo fields at 4 moments corresponding to 3-hour time intervals one by one to obtain the maximum echo field MR within 3 hours;

smoothing the MR, the CP and the LSP for multiple times to obtain parameters of the SMR, the SCP and the SLSP, wherein the CP represents 3-hour convective precipitation, and the LSP represents 3-hour large-scale precipitation;

calculating a prediction result by adopting a decision tree model according to the SMR, the SCP and the SLSP, marking as P _ dt, and finally smoothing the prediction result P _ dt through dh times to finally output a lightning potential prediction result SP _ dt;

s32, the lightning forecasting method comprises the following steps:

performing lattice localization on the hourly lightning positioning data and the hourly lightning discharge time, and extrapolating the lattice lightning discharge time for 0-6 hours to obtain lattice lightning positioning data, lattice lightning discharge time data and lightning discharge time extrapolated data for 0-6 hours;

extrapolating data of 0-6 hours and short-term prediction results and lattice lightning positioning data according to the lightning discharge time, and establishing a regression model to obtain a lightning short-term prediction result;

correcting the lightning short-term prediction result by using a neural network;

further, the establishing of the regression model according to the extrapolated lightning discharge time 0-6 hours of data and the short-term prediction result and the lattice lightning location data takes the extrapolated lightning discharge time 0-6 hours of data and the short-term prediction result SP _ dt as input, and takes the hourly lattice lightning location data as output;

the correcting is carried out on the lightning short-term forecast result by utilizing a neural network, and specifically comprises the following steps:

extracting basic weather information: firstly, extracting basic meteorological elements according to historical sounding data, wherein the basic meteorological elements comprise temperature, dew point temperature, wind direction and wind speed and relative humidity;

strong weather convection index: calculating a strong convection index from the numerical pattern results, comprising: the system comprises a Sas index, a convection effective potential energy, a strong weather threat index, a total index and a convection inhibition index;

establishing a neural network model: before fitting the neural network model, selecting a characteristic; the feature selection has many bases, and the algorithm utilizes a regression model to select features after data normalization and then performs model fitting based on a neural network method.

Identifying and tracking thunderstorms: based on radar and lightning data, extracting morphological characteristics of the thunderstorm, and tracking and extrapolating and forecasting the evolution and motion trail of the thunderstorm area; in addition to lightning positioning data, the area with the echo intensity reaching 40dBz at the height of a-10 ℃ layer and the area with the echo top height exceeding 9km are focused, and meanwhile, some auxiliary judgment indexes are provided: the echo horizontal gradient is more than 4dBz/km, and the vertical liquid water content is more than 25 kg.m < -2 >.

And (3) generating a thunderstorm nowcasting: and (3) forecasting the lightning occurrence probability of all grid points in the area based on the neural network model, and correcting the area where the lightning is located according to the storm identified by the radar chart through a weighting algorithm to obtain the final lightning short-term prediction.

S4, the typhoon system forecasting and early warning method comprises typhoon positioning and typhoon strength setting

S41, the typhoon positioning is based on warm core positioning, and specifically comprises the following steps:

and judging the data of which channel the loaded remote sensing picture belongs to, wherein the loaded data can be carried out only if the loaded data is in an infrared band because the loaded data is the point with the highest temperature, and relevant information is obtained from a file header of the remote sensing data.

The method comprises the steps of obtaining geographic positions near specified longitude and latitude points (with radius of 0.6 longitude dimensions), finding data in the range in remote sensing data according to the geographic positions near the specified longitude and latitude points, calculating the rows and columns of the data in the range in the remote sensing data through a registration file of a remote sensing picture, and then traversing the data to find the point with the maximum value.

And displaying the point with the maximum value on a map, and recording the coordinates of the point with the maximum value.

The typhoon positioning also comprises sending a typhoon center extrapolation positioning and comprehensive positioning;

the center of the reported typhoon is extrapolated and positioned: the basic idea is to interpolate between two adjacent points by a cubic polynomial using the position coordinates of the two points and the direction of the curve of the two points. I.e. a total of five points (itself and two points to the left and right) are used to determine the curve direction of the point. Because two points on the left and right are needed, for the non-closed curve, the two end points need to be additionally processed; the solution is to extrapolate two points outward using a quadratic polynomial.

The comprehensive positioning comprises the following steps: after the central positions of the typhoons are determined by adopting various positioning methods, a judgment result needs to be obtained by finally integrating a plurality of positions, wherein the comprehensive positioning position is obtained by averaging the positions, namely, the longitude and latitude of the positioning positions are averaged to obtain a comprehensive positioning position.

S42, the typhoon strength determination comprises typhoon objective strength determination based on a wind cloud 4A new algorithm;

the typhoon objective definition of the wind cloud 4A new algorithm is as follows:

finding a value b with the value less than 253K in the range of r in the satellite data at the moment t, wherein the maximum value bmax and the minimum value bmin meet the number n of the condition effective points;

obtaining a distance dmin from the nearest point and a distance dmax from the farthest point of (lon, lat), and an average value davg of distances between all points and (lon, lat) within a condition range, wherein lon is longitude and lat is latitude;

finding out the curve coefficient dindex, dindex = ((dmax ^2+ davg ^ 2)/2) ^ (1/2)

Wind speed v, v = -2.873+ n × 0.041+ lat 0.519+ (bmax-bmin) (-0.145) + davg × 3.142+ lon (-0.044) + dmax 5.594+ dindex (-8.851) + dmin

Calculating an air pressure value according to the relation between the wind speed and the air pressure; and (4) according to the calculated air pressure value, strengthening the typhoon. The lower the central air pressure of the typhoon, the greater the air pressure gradient in the typhoon range, so the wind is naturally stronger; in other words, the lower the central air pressure is, the larger the difference between the central air pressure and the air pressure at the edge of the typhoon is, the larger the wind speed is; because the wind flows from the place with higher air pressure to the place with lower air pressure, just like the water flows from the high place to the low place, the larger the difference of height, the faster the water flow, so the lower the central air pressure of the typhoon, the larger the wind speed, the stronger the typhoon.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), rambus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (10)

1. A strong convection weather forecast early warning method is characterized by comprising a strong precipitation forecast early warning method, a strong wind forecast early warning method, a thunder forecast early warning method and a typhoon system forecast early warning method;

s1, identifying a short-time heavy precipitation disaster based on real-time high-frequency radar information, carrying out close prediction on the short-time heavy precipitation within 0-2 hours and carrying out short-time prediction on the short-time heavy precipitation within 0-6 hours;

the short-time strong rainfall disaster identification based on the real-time high-frequency radar information is as follows:

acquiring image data of a weather radar and preprocessing the image data to obtain preprocessed radar image data;

extracting radar image data of a target site and nearby from the preprocessed radar image data, and acquiring the space-time scale characteristics of the cloud cluster by combining meteorological observation data;

constructing a strong precipitation identification model based on the cloud cluster space-time scale characteristics and relevant information in radar image data, and obtaining a short-time strong precipitation identification method by combining different cloud cluster space-time scale characteristics of strong precipitation disasters;

the 0-2 hour nowcast of the short-time heavy precipitation is as follows:

tracking the moving track of the target cloud cluster in 0-2 hours in the future;

inputting real-time radar data into a preset convolutional neural network model for training to obtain a falling area and precipitation amount of short-time heavy rainfall, and tracking by combining the moving track of the target cloud cluster to obtain short-time heavy rainfall forecast information within 0-2 hours in the future;

the short-term forecast of the short-term heavy precipitation in 0-6 hours is as follows:

identifying a cloud cluster band by using a WRF mode, obtaining space-time scale characteristics and a life and consumption change process, and tracking the cloud cluster band by using an SIFT cloud cluster trajectory technology;

extracting short-time heavy precipitation characterization factors exceeding a threshold value in 0-6 hours in a forecast field and relevant important factors;

inputting the characterization factors, the relevant important elements and input data into a preset deep learning model to obtain short-term heavy rainfall information, and combining the SIFT cloud cluster track technology to obtain short-term heavy rainfall forecast information of 0-6 hours;

s2, the strong wind forecasting and early warning method comprises a strong wind short-term forecasting method and a strong wind middle-term forecasting method; the strong wind short-term forecasting method comprises the following steps:

respectively adopting a squall line wind forecasting based on a radar and a squall line forecasting based on meteorological station observation data to obtain a squall line forecasting result based on the radar and a squall line forecasting result based on the meteorological station observation data;

according to the squall line wind forecasting result based on the radar and the squall line wind forecasting result based on the meteorological station observation data, distributing weight by adopting a machine learning algorithm, and performing intensity and space fusion on the two forecasting results to realize accurate squall line short-term forecasting;

the strong wind mid-term forecasting method comprises the following steps:

acquiring a WRF (squall line wind force) numerical mode to obtain the future changes of various physical quantities in the area;

further analyzing the changes of the physical quantities to obtain the changes of the physical quantities in a lattice point form, and giving a 24-hour forecasting result of the squall line wind in the lattice point form;

correcting a 24-hour forecast result of the squall line wind;

s3, the lightning forecast early warning method comprises a lightning potential forecasting method and an approaching lightning forecasting method;

the lightning potential forecasting method is based on establishing a regression equation of physical parameters related to lightning to obtain the probability of occurrence of thunderstorms;

correcting by using lightning forecast of real-time data of power grid lightning observation according to the occurrence probability of the thunderstorm, and finally obtaining a lightning potential forecast result;

the method for forecasting the approaching lightning comprises the following steps:

performing lattice localization on the hourly lightning positioning data and the hourly lightning discharge time, and extrapolating the lattice lightning discharge time for 0-6 hours to obtain lattice lightning positioning data, lattice lightning discharge time data and extrapolated lightning discharge time data for 0-6 hours;

extrapolating data of 0-6 hours and short-term prediction results and lattice lightning positioning data according to the lightning discharge time, and establishing a regression model to obtain a lightning short-term prediction result;

correcting the lightning short-term prediction result by using a neural network;

s4, the typhoon system forecasting and early warning method comprises typhoon positioning and typhoon strength setting; the typhoon positioning is to obtain the position of the typhoon through warm center positioning, and the typhoon strength setting is to perform strength setting on the typhoon through a wind cloud 4A new algorithm;

the warm heart is positioned as follows:

judging the data of the loaded channel in the remote sensing picture to obtain remote sensing data information, and acquiring the geographical position near the specified longitude and latitude point;

and acquiring the remote sensing data information of the corresponding range according to the geographic position near the longitude and latitude points, finding the point with the maximum value in the remote sensing data of the corresponding range through calculation, displaying the point with the maximum value on a map, and simultaneously recording the coordinate of the point with the maximum value.

2. The strong convection weather forecast early warning method according to claim 1, wherein the image data of the weather radar is obtained and preprocessed to obtain preprocessed radar image data, and specifically the method comprises the following steps:

respectively extracting image data of a Doppler weather radar, a wind profile radar and a power grid self-built X-band radar, and respectively inverting horizontal wind speed data of different heights, data of a three-dimensional wind field and cloud characteristic data according to the radar image data; performing mutual verification on relevant meteorological element data obtained by radar image inversion and meteorological observation data of an observation point, and removing or correcting radar image data with large errors; integrating the data passing the quality check, selecting a fusion algorithm, carrying out standardized processing according to a uniform format, and finally realizing the storage of structured data of the radar image;

the method comprises the following steps of extracting radar image data of a target site and nearby from preprocessed image data, and acquiring space-time characteristics of a cloud cluster by combining meteorological observation data, wherein the method specifically comprises the following steps:

acquiring radar image data of a target station and nearby stations from the radar image data, and selecting a radar image with uniform resolution and observation content from the radar image data; combining meteorological observation data, searching for coordinate correlation characteristics among samples by a template matching mode by utilizing overlapped parts existing among local radar image samples, splicing a series of radar images in a small range to radar images in a wider space, acquiring space-time scale characteristics of radar reflectivity factors of short-time strong rainfall disasters from the whole radar images, and further acquiring space-time characteristics of clouds;

the strong precipitation identification model is constructed by the following steps:

the method comprises the steps of extracting reflectivity factor information from radar image data in a large number of heavy precipitation processes, researching the relation between radar reflectivity factors and precipitation intensity and cloud cluster characteristics, constructing a heavy precipitation identification model, analyzing different space-time scale characteristics of heavy precipitation disasters, and providing a short-time heavy precipitation disaster identification method based on the different space-time scale characteristics.

3. The strong convection weather forecast early warning method as claimed in claim 1, wherein the tracking of the target cloud cluster movement track 0-2 hours in the future is based on taylor freezing assumption, and the SIFT technology is applied to convert the cloud cluster time prediction problem into the local space prediction problem, so as to realize the tracking of the target cloud cluster movement track 0-2 hours in the future;

inputting real-time radar data into a preset convolutional neural network model for training, specifically:

because the elements of the precipitation, the wind field and the cloud cluster of the atmosphere are continuous functions related to space and time, any continuous function is approached with any precision according to the Robert Hecht-Nielsen theorem, namely a three-layer network implementation; therefore, a hidden layer, namely a 3-layer network is selected to complete the nonlinear mapping of the precipitation, the wind field and the cloud cluster elements; initializing all mapping weight initial values by using different small random numbers to ensure that the network does not enter a saturation state due to overlarge weight;

in the training process, a radar image part is subjected to multilayer convolution pooling, then vectors are leveled to one dimension to obtain radar image characteristics, the radar image characteristics are combined with other non-image characteristics in a full connection layer and are jointly input into a neural network with 3 hidden layers;

in the backward propagation process, as the depth of the network increases, the amplitude value of the gradient from the output layer to the first few layers of the network is sharply reduced, namely, the overfitting problem is solved, so that dropout is adopted to prevent overfitting, and an Adam optimization algorithm is adopted for gradient descent.

4. The strong convection weather forecast early warning method as claimed in claim 1, wherein the cloud cluster band is identified by using WRF mode, specifically:

extracting multilayer cloud water, cloud ice, rainwater, snow and aragonite data of more than 850hPa in a WRF mode 0-6 hour forecasting field, adding grid points, setting a first threshold value of water-substance ratio quality in a cloud area, wherein the cloud area water-substance ratio quality is larger than the first threshold value and is judged to be cloud, otherwise, the cloud area is judged to be cloud-free, so that a cloud cluster band is identified, space-time scale features and a life and consumption change process are obtained, and then the cloud cluster band is tracked by using an SIFT method;

extracting the short-time strong precipitation characterization factors exceeding the threshold value in 0-6 hours in the forecast field and relevant important factors are based on the high-resolution forecast result in the historical WRF mode, and the short-time strong precipitation characterization factors in the deep WRF mining mode, such as local wind speed and violent rise of thunderstorm potential indexes; extracting several important elements of the short-time strong precipitation, including a vertical wind field and a water vapor field, and extracting the short-time strong precipitation characterization factor exceeding a second threshold value in 0-6 hours in the forecast field and related important elements by combining the characterization factor in the short-time strong precipitation process and the second threshold value of the important elements obtained by statistics of data in historical forecast results;

the characterization factors and the related important elements comprise short-time heavy precipitation characterization factors, related important elements, space-time feature scale and life cycle matching features.

5. The strong convection weather forecast early warning method as recited in claim 1, wherein the radar-based forecasting for squall line wind is specifically:

radar base data filtering processing: radar base data are adopted, and multilayer echoes with different elevation angles in the radar base data are utilized to carry out three-dimensional mathematical characteristic filtering processing on radar images, so that clutter filtering of the radar is more effective; meanwhile, a spatial interpolation extension method of multilayer elevation echoes is adopted to make up missing data in a weather radar image, and a first radar image is obtained; and (3) radar echo prediction: quantitatively analyzing the motion trail of the radar echo by adopting a leading-edge optical flow method technology for the first radar image, calculating an optical flow field of the radar echo to obtain a motion vector field of the echo, extrapolating the radar echo based on the motion vector field so as to infer the mobile evolution condition of the squall line wind area, and performing 0-2 hour proximity prediction on the squall line wind area to obtain a predicted second radar image;

radar data post-processing: the predicted second radar image has certain loss in smoothness and continuity, so that post-processing operation is required to be carried out on radar data; firstly, a closing operation is needed to be carried out to complete cracks and holes, and secondly, radar echoes are processed to a certain extent according to different forecasts of a large area; if the weather system is a single thunderstorm strong convection weather system and the echo edge is clear and has obvious gradient, the edge diffusion is reduced through corrosion operation, otherwise, the radar echo range is enlarged through expansion operation, and low threshold information filtered out in the preprocessing of the edge is supplemented, so that the forecasting accuracy is improved;

identification of narrow-band echo and convergence line: calculating the two-way gradient of each echo point in the radar image after radar data processing to reserve linear echo, but designing a plurality of models with different included angles in a specific algorithm because the included angles between narrow-band echo and radial line cannot be known; most precipitation echoes of the intensity field after the bidirectional gradient processing are filtered, narrow-band echoes are completely reserved, but a plurality of short lines still exist in the image; in order to remove the short lines, firstly thinning the image, namely only keeping the central point of each intensity section, recording the width of the lower section, then calculating the length of each short line by using a recursive algorithm, and filtering the short lines which do not meet a length threshold; in the process, the azimuth angle and the radial library number of each effective point are saved, and finally, only a short line with a certain length is reserved, and the previously recorded segment width is restored into a band shape to form a final narrow-band echo identification image;

squall line wind intensity mapping and lattice time series: respectively establishing a machine learning regression model for squall line wind in different areas at different moments according to the formed final narrow-band echo identification image to obtain an echo-wind real-time mapping relation matched with climate characteristics of each area, and realizing prediction of the squall line wind area based on the radar;

the squall line wind forecasting method based on meteorological station observation data specifically comprises the following steps:

weather station data quality control: carrying out quality control on observation data in a meteorological station, and carrying out machine learning on an air pressure surge parameter, an air pressure distance flat field parameter, a temperature parameter, a false equivalent temperature parameter, a humidity change discontinuous parameter and a rain group activity parameter in the observation data to obtain data after multi-parameter learning;

multi-parameter learning: performing lattice localization on the data after the multi-parameter learning to obtain lattice-localized multi-parameters;

wind field extrapolation: extrapolating the grid multi-parameters in a wind field to infer the mobile evolution condition of the squall line wind area to obtain predicted squall line wind data based on the meteorological station;

and (3) data post-processing: the data post-processing comprises the following specific steps:

(1) And (3) meteorological site data arrangement: inputting field names including site names, x longitude lon, y latitude lat, average air temperature, average wind speed, relative humidity and average sunshine hours; wherein, the longitude and latitude need to be converted into a form of degree, and other data are converted into corresponding units;

(2) Carrying out interpolation analysis on the vector point data converted into shape format;

(3) Exporting point data in shape format;

(4) Setting an Arcgis environment;

(5) Data interpolation of meteorological stations;

predicting a grid point wind speed time sequence: predicting an hourly lattice point wind speed sequence, and finally realizing the prediction of squall line wind based on meteorological station observation data;

the method adopts a machine learning algorithm to distribute weight, and performs intensity and space fusion on two prediction results to realize accurate squall line wind short-term prediction, which specifically comprises the following steps:

keeping consistent data in the results of the squall wind position prediction based on radar and meteorological station observation data, and taking intersection of the results of the differences of the two prediction methods; a binary tree search algorithm is adopted to accelerate data search so as to improve the fusion speed; the binary tree searching algorithm is to generate a binary tree for the data elements to be searched, then compare the given value with the keywords of the root node, if the given value is equal to the keywords of the root node, the search is successful, otherwise, according to the keywords of which the given value is smaller than or larger than the keywords of the root node, the search is continued in the left subtree and the right subtree until the search is successful or the left subtree or the right subtree is an empty tree.

6. The strong convection weather forecast early warning method as recited in claim 1, wherein the strong wind short forecasting method further comprises squall line strong wind identification and location, and the squall line strong wind identification and location is performed in the following specific process:

the squall line strong wind is represented as the summation of radial speeds on a radar velocity map, which is mainly represented by the summation of wind speeds, and the speed value of the radial speed is converted from a higher value to a lower value; therefore, according to the characteristic, a group of adjacent distance libraries with continuously reduced radial velocity values are searched along the increasing direction of the radial distance to obtain a one-dimensional convergent section;

to describe the intensity of each one-dimensional convergent section, the following physical quantities are calculated: velocity gradient g, momentum f, orientation, radial center;

if the speed gradient g or momentum f of a certain convergent section is lower than a set lower limit value, deleting the one-dimensional convergent section, and otherwise, carrying out high value inspection on the one-dimensional convergent section; if either the value of g or f of the converged segment is greater than the set upper limit, the one-dimensional converged segment is saved;

judging the difference between the radial center and the position of the current one-dimensional convergent section and the next one-dimensional convergent section in all the stored one-dimensional convergent sections, if the difference is greater than a set radial distance threshold or a set position threshold, deleting the one-dimensional convergent section, and otherwise, keeping the one-dimensional convergent section;

and combining all the one-dimensional radial sections meeting the conditions into two-dimensional characteristics, and filtering out the characteristics with too small number of radial sections to form the finally identified radial line switching line.

7. The strong convection weather forecast early warning method as recited in claim 1, wherein the 24-hour forecast result for the squall line wind is given in the form of lattice points, specifically:

adopting NCAR, NCEP and FSL/NOAA to jointly develop a developed fine grid mesoscale WRF numerical mode; the WRF numerical mode continuously integrates from the initial moment to the next moment according to the initial condition and the boundary condition and the given physical parameterization process, the change of each future physical quantity in the area is obtained through forecasting, the concerned wind field data is extracted from the changes and further analyzed, and finally the 24-hour forecasting result of the squall line wind is given in a lattice mode;

correcting the 24-hour forecast result of the squall line wind, including: the results of the live observation data and the wrf numerical prediction mode are fused and corrected, and the correction is carried out based on the remote sensing terrain coefficient;

after the live observation data and the results of the wrf numerical prediction mode are fused, correcting the results into two aspects of phase correction and intensity correction of the numerical prediction wind field; the phase correction of the wind field comprises the steps of obtaining a total phase translation field by adopting fast Fourier transform and adjusting the regional phase by utilizing a multi-scale optical flow variation method; the correction of the wind field intensity mainly utilizes a cumulative distribution function of Weber distribution;

the remote sensing-based terrain coefficient correction specifically comprises the following steps: combining a meteorological numerical simulation algorithm of a remote sensing terrain correction coefficient, matching a high-precision topographic map on the basis of traditional numerical weather forecast, and improving the forecast accuracy and resolution of strong wind by using a dynamic or statistical downscaling technology or CFD simulation;

the CFD simulation calculation process is mainly divided into three parts, namely a pretreatment part, a simulation calculation part and a post-treatment part; the pre-processing part comprises acquisition of topographic data, topographic modeling and grid generation, which are prerequisites for wind field CFD simulation, the result of WRF simulation is used as the boundary condition of CFD, the simulation calculation part comprises the steps of boundary condition setting, turbulence model setting, discrete format setting, solver setting and initialization solving, which is the main solving process of wind field CFD simulation, and the post-processing part comprises acquisition and analysis of simulation result.

8. The strong convection weather forecast early warning method as claimed in claim 1, wherein the probability of occurrence of thunderstorms is obtained by establishing a regression equation based on the physical parameters related to the thunderstorms, specifically:

through the relation of the correlation coefficients, 7 physical quantities with the highest correlation coefficient with lightning are finally screened out, wherein the physical quantities are k index, sand index SI, A index, 700hPa temperature dew point difference, 850hPa temperature dew point difference, 925hPa temperature dew point difference, 850hPa temperature difference and 500hPa temperature difference; according to the mesoscale strong convection generation condition and the statistics of a large amount of data, the threshold values of the physical parameters are processed into 0 and 1, and the threshold values are set as follows:

when the k index is more than or equal to 33, marking as 1, otherwise, marking as 0;

when the sand index is less than or equal to 0, marking as 1, otherwise, marking as 0;

when the index A is larger than or equal to 10, marking as 1, otherwise, marking as 0;

when the temperature dew point difference is less than or equal to 3, the temperature dew point difference is marked as 1, otherwise, the temperature dew point difference is marked as 0, and 700hPa, 850hPa and 925hPa are all used as threshold values;

when the temperature difference between 850hPa and 500hPa is not less than 23, the value is marked as 1, and otherwise, the value is marked as 0;

finally, 7 indexes subjected to 0 and 1 chemical treatment are used as core physical parameters of potential prediction, and a regression equation of the lightning potential prediction is established; the regression equation of the lightning potential forecast is as follows:

y(j，i)＝0.11+0.154×k1(j，i)+0.142×k2(j，i)+0.061×k3(j，i)+0.146×k4(j，i)+0.131×k5(j，i)+0.03×k6(j，i)+0.097×k7(j，i)

wherein y is the probability of thunderstorm occurrence at a certain site, K1 \8230andK 7 is the physical parameter after 0 and 1;

correcting by using the lightning forecast of the real-time data of power grid lightning observation according to the occurrence probability of the thunderstorm to finally obtain a lightning potential forecast result, which specifically comprises the following steps:

taking the maximum value of each grid point of the radar echo fields at 4 moments corresponding to 3-hour time intervals one by one to obtain the maximum echo field MR within 3 hours;

smoothing the MR, the CP and the LSP for multiple times to obtain parameters of the SMR, the SCP and the SLSP, wherein the CP represents 3-hour convective precipitation, and the LSP represents 3-hour large-scale precipitation;

and calculating a prediction result by adopting a decision tree model according to the SMR, the SCP and the SLSP, marking as P _ dt, and finally smoothing the prediction result P _ dt for dh times to finally output a lightning potential prediction result SP _ dt.

9. The strong convection weather forecast warning method as set forth in claim 1, wherein said extrapolating 0-6 hours of data and short-term forecast results from the lightning discharge time and the rasterized lightning location data, and establishing a regression model with the extrapolated 0-6 hours of lightning discharge time and the short-term forecast results SP _ dt as inputs and the rasterized lightning location data hourly as outputs;

the correcting is carried out on the lightning short-term forecast result by utilizing a neural network, and specifically comprises the following steps:

extracting basic meteorological information: firstly, extracting basic meteorological elements according to historical sounding data, wherein the basic meteorological elements comprise temperature, dew point temperature, wind direction and wind speed and relative humidity;

strong weather convection index: calculating a strong convection index from the numerical pattern results, including: a Sauter index, a convection effective potential, a strong weather threat index, a total index and a convection inhibition index;

characteristic screening and model building: before fitting of a neural network model, selecting a characteristic, selecting the characteristic by using a regression model, and fitting the model based on a neural network method;

identifying and tracking the thunderstorm: based on radar and lightning data, extracting morphological characteristics of the thunderstorm, and tracking and extrapolating and forecasting the evolution and motion trail of the thunderstorm area; combining lightning positioning data, focusing on a set area and auxiliary judgment indexes of horizontal gradient and vertical liquid water content of an echo;

and (3) generating a thunderstorm nowcasting: and (3) forecasting the lightning occurrence probability of all grid points in the area based on the neural network model, and correcting the area where the lightning is located according to the storm identified by the radar chart through a weighting algorithm to obtain the final lightning short-term prediction.

10. The strong convection weather forecast early warning method as claimed in claim 1, wherein the typhoon is strengthened by a wind cloud 4A new algorithm, specifically:

finding a value b with a value smaller than a set range in the range r in the satellite data at the moment t, wherein the maximum value bmax and the minimum value bmin meet the number n of the condition effective points;

obtaining a distance dmin from the nearest point and a distance dmax from the farthest point of (lon, lat), and an average value davg of distances between all points and (lon, lat) within a condition range, wherein lon is longitude and lat is latitude;

solving a curve coefficient dindex and a wind speed v;

calculating an air pressure value according to the relation between the wind speed and the air pressure; and (4) according to the calculated air pressure value, strengthening the typhoon.

Images