CN113191568B - Meteorological-based urban operation management big data analysis and prediction method and system - Google Patents
Meteorological-based urban operation management big data analysis and prediction method and system Download PDFInfo
- Publication number
- CN113191568B CN113191568B CN202110558687.0A CN202110558687A CN113191568B CN 113191568 B CN113191568 B CN 113191568B CN 202110558687 A CN202110558687 A CN 202110558687A CN 113191568 B CN113191568 B CN 113191568B
- Authority
- CN
- China
- Prior art keywords
- model
- event
- weather
- data
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 74
- 238000007405 data analysis Methods 0.000 title claims abstract description 28
- 230000000694 effects Effects 0.000 claims abstract description 17
- 238000001556 precipitation Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims description 73
- 238000005065 mining Methods 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 abstract description 15
- 238000010801 machine learning Methods 0.000 abstract description 6
- 238000003066 decision tree Methods 0.000 description 22
- 238000007726 management method Methods 0.000 description 22
- 238000004458 analytical method Methods 0.000 description 14
- 230000008569 process Effects 0.000 description 12
- 238000005516 engineering process Methods 0.000 description 9
- 238000010276 construction Methods 0.000 description 7
- 238000007781 pre-processing Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000002790 cross-validation Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 230000002354 daily effect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 230000008439 repair process Effects 0.000 description 3
- 238000005096 rolling process Methods 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 241001123248 Arma Species 0.000 description 2
- 230000004888 barrier function Effects 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012731 temporal analysis Methods 0.000 description 2
- 238000000700 time series analysis Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- JCCNYMKQOSZNPW-UHFFFAOYSA-N loratadine Chemical compound C1CN(C(=O)OCC)CCC1=C1C2=NC=CC=C2CCC2=CC(Cl)=CC=C21 JCCNYMKQOSZNPW-UHFFFAOYSA-N 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000013180 random effects model Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000008080 stochastic effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000029305 taxis Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Software Systems (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- General Business, Economics & Management (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Educational Administration (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a city operation management big data analysis and prediction method and system based on weather, which are characterized in that an event quantity prediction model is established, the model is based on a classical machine learning algorithm gradient lifting model, and a two-step modeling, a random intercept model and a random effect model are used for improving the gradient lifting model, so that the occurrence rule of an event can be captured better. In actual operation of the model, the street town is divided in 48 hours, and the occurrence number prediction and the corresponding risk early warning level of the events in every 12 hours are realized by integrating multiple types of data sources (weather automatic station data, grid weather element prediction data, grid event data, hot line event data and 110 weather disaster data). On the basis of the event quantity prediction model, the influence of three meteorological elements, namely wind speed, precipitation and air temperature, on the event quantity is obtained by calculating the contribution value of the meteorological elements to the event quantity prediction value.
Description
Technical Field
The invention relates to the technical field of weather, in particular to a city operation management big data analysis and prediction method and system based on weather.
Background
In recent years, the development of big data technology is rapid, and the big data technology becomes a hot spot for research and gradual change of the world. On the one hand, the application range of the big data technology is very wide, and especially the big data technology is outstanding in the fields of medical treatment, finance, security protection, automobiles and the like. Meteorological applications have also been an important area of high performance computing, and big data technology brings about an unprecedented opportunity for the development of observation, forecasting, service, etc., and also brings about a great challenge. Therefore, the characteristics of the development of big data technology have a great influence on the weather service. On the other hand, the important big data technologies such as machine learning, model training processing and computer vision have profound effects on weather in different fields and influence lives of audiences to different degrees.
Disclosure of Invention
In order to solve the defects in the prior art, the invention provides a weather-based urban operation management big data analysis and prediction method and system, which can establish a prediction model of the total number of long-time rolling accumulated events and specific event scenes.
The invention provides the following technical scheme:
in a first aspect, a weather-based urban operation management big data analysis and prediction method includes:
Collecting data sources and generating an event quantity prediction model training set;
based on the training set, modeling the occurrence probability and the occurrence number of the events by adopting a two-step method;
in the event occurrence probability model, predicting whether an event occurs in t time by adopting a gradient lifting model, and taking the probability of whether the event occurs in t time by adopting each object as output;
in the event occurrence quantity model, predicting the quantity of events occurring in t time when each object is once the event occurs by adopting a gradient lifting model;
multiplying the expected occurrence probability of the event by the expected number of the event once to obtain the number value of the event occurrence, and establishing an event number prediction model.
As a preferable technical scheme of the prediction method, the data sources comprise weather automatic station data, grid-point weather element prediction data, grid event data, hot line event data and 110 weather disaster data.
As a preferable technical scheme of the prediction method, the gradient lifting model is a gradient lifting model based on the LightGBM, and a random intercept model and a random effect model are introduced on the basis to optimize and upgrade the model.
As a preferable technical scheme of the prediction method, the method also comprises the steps of carrying out feature mining on time data, historical and live meteorological element data and historical and live city operation data (grid data, hot line data and 110 meteorological disaster data), and adding the model into the model training as the input of the comprehensive event number prediction model.
As a preferable technical scheme of the prediction method, in the mining of the historical and live city operation data characteristics, the delay of the occurrence number of the events in time is judged by adopting a partial autocorrelation coefficient.
As a preferable technical scheme of the prediction method, the instantaneous and historical influences of each meteorological element on the occurrence number of the event are considered in the mining of the data characteristics of the historical and live meteorological elements, and the lag relation between the occurrence number of the event and different meteorological elements in time is calculated by adopting a cross correlation coefficient in the influence of the historical meteorological elements on the occurrence number of the event.
As a preferable technical scheme of the prediction method, the method also comprises the steps of modeling and selecting event scenes closely related to meteorological influence by adopting the two-step method while establishing the event quantity prediction model, and establishing a scene model.
As a preferable technical scheme of the prediction method, the method also comprises the steps of establishing an event risk early warning model: and formulating event early warning standards by combining the historical event occurrence number score and an absolute threshold value of the event occurrence number.
As a preferable technical scheme of the prediction method, the method further comprises the steps of establishing a weather influence index model:
calculating the percentage contribution degree of each meteorological element feature to the event quantity prediction based on the event quantity prediction model;
and taking the percentage contribution value of the meteorological element characteristics in the range of the identified meteorological element as a reference, calculating the absolute deviation of the percentage contribution value of the current meteorological element characteristics and the reference, and taking the absolute deviation as an influence index of the meteorological element characteristics.
In a second aspect, a weather-based urban operation management big data analysis and prediction system is used for executing the weather-based urban operation management big data analysis and prediction method.
By adopting the technical scheme, the invention has the following beneficial effects:
the event number prediction model is based on a classical machine learning algorithm gradient lifting model (Gradient Boosting Machine), and the gradient lifting model is improved by using a two-step modeling method, a random intercept model and a random effect model, so that the occurrence rule of the event can be captured better. In actual operation of the model, the street town is divided in 48 hours, and the occurrence number prediction and the corresponding risk early warning level of the events in every 12 hours are realized by integrating multiple types of data sources (weather automatic station data, grid weather element prediction data, grid event data, hot line event data and 110 weather disaster data). On the basis of the event quantity prediction model, the influence indexes of three meteorological elements, namely wind speed, precipitation and air temperature, on the occurrence of the event quantity are obtained by calculating the contribution value of the meteorological elements to the event quantity prediction value.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a general flow chart of the city operation management big data analysis and prediction method based on weather.
FIG. 2 is a model technology roadmap of the weather-based urban operation management big data analysis and prediction method of the invention.
FIG. 3 is a flow chart illustrating an embodiment of a weather-based urban operation management big data analysis and prediction method according to the present invention.
FIG. 4 is a schematic diagram of a characteristic engineering flow of the weather-based urban operation management big data analysis and prediction method of the invention.
Detailed Description
In order to make the above objects, features and advantages of the present invention more comprehensible, embodiments accompanied with figures are described in detail below.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than as described herein, and therefore the present invention is not limited to the specific embodiments disclosed below.
Referring to fig. 1 to 4, the embodiment of the invention provides a weather-based urban operation management big data analysis and prediction method and system, which uses a method for integrating a plurality of statistical machine learning models, and respectively establishes a prediction model of total number of events sliding for 12 hours and a specific event scene for urban street and town grids/hotlines (such as Shanghai grids, pudong grids, xuhi grids, pudong hotlines and Xuhi hotlines) with prediction timeliness of 48 hours; the 110 weather disaster establishes a storm and strong wind 1 hour event number prediction model, and the prediction aging is 48 hours. According to the event quantity prediction model, a weather influence index formed by three weather element characteristics of wind speed, precipitation and air temperature is designed and used for quantifying the influence degree of different weather elements on urban running conditions. Finally, according to the historical event data distribution, an event early warning standard is designed, and a risk early warning system is formed by matching with an event quantity prediction model.
The event number prediction model is based on a classical machine learning algorithm gradient lifting model (Gradient Boosting Machine), and the gradient lifting model is improved by using a two-step modeling method, a random intercept model and a random effect model, so that the occurrence rule of the event can be captured better. In actual operation of the model, the street town is divided in 48 hours, and the occurrence number prediction and the corresponding risk early warning level of the events in every 12 hours are realized by integrating multiple types of data sources (weather automatic station data, grid weather element prediction data, grid event data, hot line event data and 110 weather disaster data). On the basis of the event quantity prediction model, the influence of three meteorological elements, namely wind speed, precipitation and air temperature, on the event quantity is obtained by calculating the contribution value of the meteorological elements to the event quantity prediction value. According to the data type and source, a whole market, pudong and Xuhe grid event quantity prediction model, a Pudong and Xuhe hot line event quantity prediction model and a whole market 110 weather disaster quantity prediction model are deployed.
Specifically, the city operation management big data analysis and prediction method based on weather comprises the following steps:
step 1: collecting data sources and generating an event quantity prediction model training set;
step 2: based on a training set, modeling the occurrence probability and the occurrence number of the events by adopting a two-step method;
step 3: in the event occurrence probability model, predicting whether an event occurs in t time by adopting a gradient lifting model, and taking the probability of whether the event occurs in t time by adopting each object as output;
step 4: in the event occurrence quantity model, predicting the quantity of events occurring in t time when each object is once the event occurs by adopting a gradient lifting model;
step 5: multiplying the expected occurrence probability of the event by the expected number of the event once to obtain the number value of the event occurrence, and establishing an event number prediction model.
The data sources in the step 1 comprise city operation management big data such as weather automatic station data, grid-point weather element forecast data, grid event data, hot line event data, 110 weather disaster data and the like.
The gradient lifting model in step 2 to step 4 is a LightGBM based gradient lifting model. The object may be a whole city, administrative district, or street town.
The model also performs feature mining during training and real-time online prediction, performs feature mining for time data, historical and live meteorological element data and historical and live city operation data (grid data, hot line data and 110 meteorological disaster data), and adds model training and real-time online prediction. The method and features of feature mining will be described in detail below.
In addition, when the event quantity prediction model is established, a two-step modeling method, a random intercept model and a random effect model are adopted to upgrade and optimize the model to select event scenes closely related to meteorological influence, and a scene model is established. And by combining the historical event occurrence number score and the absolute threshold value of the event occurrence number, an event early warning standard is formulated, and an event risk early warning model is established. A weather effect index model is also established by adopting the percentage contribution degree of each weather element feature to the event quantity prediction.
Details of the weather-based city operation management big data analysis prediction model of the present invention are further described below for specific model purposes as follows:
model purpose: the method comprises the steps of establishing a prediction model of total number of events sliding for 12 hours and a specific event scene by using Shanghai grids, pudong grids, xufu grids, pudong hot lines and Xufu hot lines, and establishing a prediction model of the specific event scene for 1 hour by using 110 weather disasters.
1. Data sources for model:
1. meteorological automatic station live data
(1) Data source structure
One of the main sources of weather data used by the model in training and in real-time on-line prediction is the all-market weather automatic station data. The historical data of each site specifically comprises fields of automatic site names, longitude and latitude coordinates of the automatic site, places and streets where the automatic site belongs, automatic site elements and the like. The meteorological element data monitored in real time by each automatic station comprises fields such as temperature, rainfall, wind direction, wind speed, wind direction of 2 minutes, wind speed of two minutes, maximum wind direction, maximum wind speed and the like. And when a specific event prediction model is trained and the quantity is predicted on line in real time, the temperature, the rainfall and the maximum wind speed are selected as main inputs of the model.
(2) Data preprocessing process
Corresponding live meteorological element data of each street town is generated through the following calculation process:
for the case where there is an automatic station in the district/street town:
(1) space matching is carried out on the weather automatic station and the street town;
(2) calculating average rainfall, maximum wind speed, minimum air temperature and maximum air temperature of automatic stations in each district/street town;
(3) and (3) taking the rainfall, the wind speed and the highest and lowest air temperature obtained in the step (2) as meteorological element characteristics of the corresponding time of the current street and town.
For the case where there is no automatic station in the district/street town:
the street town obtains the automatic stations with the nearest distance and corresponding observation values by calculating the minimum linear distance between each automatic station and each district/street town;
and (3) calculating the steps (1) and (2), and calculating to obtain the weather element characteristics of the current time of the current district/street town.
2. Latticed weather element forecast data
(1) Data source structure
Another major source of meteorological element data used when the model predicts online in real time is the grid-tied meteorological element forecast data. The main form of the data is formatted weather element forecast grid data, the size of the grid is 51 multiplied by 57, and the urban area and suburban area range of Shanghai are covered. The data storage format is multi-channel raster data of NetCDF. The updating frequency of the grid weather element forecast is twice a day, namely eight morning points and eight evening points of each day. Each forecast outputs hour-by-hour grid point weather element forecast data within 48 hours. The grid-point weather element forecast data comprises cloud cover percentage, 10-meter-height wind field, ground precipitation, relative humidity and air temperature. In the actual online running process of the model, a 10-meter-height wind field, ground precipitation and air temperature are selected as input features of the model.
(2) Data preprocessing process
When the model runs online in real time, the following steps are used for preprocessing grid-point weather element forecast, and generating rainfall, air temperature and maximum wind speed characteristics of each street town in the whole city:
(1) according to geographic raster data in the grid point forecast data, space connection is carried out on boundary data of the street town, and a corresponding table of the grid points and the street town is formed;
(2) reading 3-dimensional (time, longitude and latitude) raster data of ground precipitation, wind field and air temperature elements in a NetCDF format;
(3) calculating the maximum wind speed of the corresponding grid point according to the U vector and the V vector in the wind field;
(4) converting the units of the air temperature elements from Kelvin to the temperature, and unifying the units of the air temperature elements with the units of the weather automatic station for monitoring the weather elements in real time;
(5) calculating average rainfall, maximum wind speed, minimum air temperature and maximum air temperature of automatic grid points in each county/street town according to the grid point-street town corresponding table in the step 1;
(6) and (5) taking the rainfall, the wind speed and the highest and lowest air temperature obtained in the step (5) as meteorological element characteristics of the corresponding time of the current street town.
Through the calculation process, the information of the strong wind, the rainfall and the air temperature, which are obtained by the grid-point weather element forecast of the whole town, can be obtained and used for the subsequent weather characteristic input of the model.
3. Grid live event data
(1) Data source structure
Shanghai grid data uses data from 2019/01/01 00:19:18 to 2020/02/01,2020/05/01-2020/10/19 as a training set, together with 4352245 pieces of data. The Shanghai grid original data has 124 fields in total, and five fields including DISCOVERTIME, INFOSCNAME, STREETNAME (the name of the belonging street) and TASKID (task number) are selected for final analysis according to analysis requirements.
The Pudong grid data uses the data from 2020/01/01 00:19:18 to 2020/02/01,2020/03/01-2020/10/19 as a training set, together referring to data 1077230. The original data of Pudong grid has 124 fields, and six fields including DISCOVERTIME, INFOSCNAME, ISFAST (whether the quick treatment is performed or not), STREETNAME (the name of the street to which the quick treatment belongs) and TASKID (task number) are selected for final analysis according to analysis requirements.
The creep grid data uses 2018/01/15 to 2020/10/15 data as a training set, together with 1,054,935 pieces of data. The raw data of the creep grid has 17 fields, and according to analysis requirements, six fields of CREATETIME (discovery time), second_type (subclass name), THIRD_type (subclass name), STREETNAME (street name), TASKID (task number) and CASEATTRIBUTTE (event TYPE) are selected for final analysis.
(2) Data preprocessing process
When training the grid scene model, the event types are classified by using the subclass name field in the data, and the prediction and training targets in the grid scene model are determined by the method.
When the corresponding label of the non-quick treatment model is determined, an isfast (whether quick treatment class event) field is selected as a screening condition, and the selected predicted event object is filtered as the predicted event object of the non-quick treatment class model. And combining with the scene model standard reaching rule, outputting the total quantity of the non-rapid treatment model and the prediction model label of the scene model.
4. Hotline live event data
(1) Data source structure
Pudong hotline data uses data from 2020/01/01 00:19:18 to 2020/02/01,2020/03/01-2020/10/19 as the training set, together with 586869 pieces of data. 136 fields are used in the original data of Pudong hotline, and six fields including DISCOVERTIME, INFOSCNAME, INFOZCNAMME, STREETNAME (the name of the street) and TASKID are selected for final analysis according to analysis requirements.
The creep hotline data uses 2018/01/15 to 2020/10/15 data as training set, together with 144,297 pieces of data. The raw data of the creep hotline has 17 fields, and according to analysis requirements, six fields of CREATETIME (discovery time), second_type (subclass name), THIRD_type (subclass name), STREETNAME (street name), TASKID (task number) and CASEATTRIBUTTE (event TYPE) are selected for final analysis.
(2) Data preprocessing process
In training the hotline scene model, the event type is classified using the subclass name field in the data, and by this method, the prediction and training goals in the hotline scene model are determined.
5. 110 event weather disaster data
(1) Data source structure
110 Meteorological disaster data uses 2020/01/00:00 to 2020/07/29/24:00:00 data as a verification set, together with 1,998 pieces of data.
110 weather DISASTER real-time raw data has 13 fields, and according to analysis requirements, an OBJECID, a DATETIME_DISATER, a TELEPHONE, a LONTITUDE, a LATITUDE, a CASE_ADDR, a CASE_DESC, and DISACT are selected for final analysis.
(2) Data preprocessing process
When the 110 weather disaster scene model is trained, the district and county fields are simply cleaned, and the weather disaster content fields are structurally processed. The disaster type field in the data is used for classifying the weather disaster types, the prediction and training targets in the scene model are determined through the method, and finally, the storm event with the largest event number is selected for prediction modeling.
2. Event number prediction model
1. Principle of model
(1) Decision tree model
The decision tree is an algorithm for solving the classification and the problems, and adopts a tree structure, and the final classification is realized by using layer-by-layer reasoning. The decision tree is made up of several elements:
root node: containing a complete set of samples
Internal node: corresponding feature attribute testing
Leaf node: representing the outcome of the decision
At each state node, the decision tree takes the maximum information gain obtained in each data latitude (feature) by traversing the feature as the direction of growth of the decision tree and calculating which specific node in that direction splits the data, which results in the maximum information gain. And constructing the internal nodes of the next level until the highest growth height of the preset decision tree is reached, or the information gain can not be brought by growth in any direction and data segmentation. The final decision tree will be a classification and regression model with a combination of binary condition decision rules. And the different information gain functions also determine the difference in decision tree growth algorithm. The commonly used algorithms are ID3/C4.5 algorithms based on information entropy, and CART trees based on the coefficient of Kerning for calculation. In the city operation management big data analysis and prediction model based on weather, the decision tree algorithm is CART tree.
Decision trees possess a number of advantages: the method is an algorithm which is easy to intuitively understand the internal structure of the model, can directly embody the characteristics of data, and intuitively interprets the logic of the model to make predictions. Moreover, the decision tree is very robust in terms of data preparation, and can handle both numeric and entry type features. However, in practical applications, the disadvantages of decision trees are also prominent. When the data dimension is large and the number of samples is unbalanced, the decision tree model can generate an excessively complex but unstable judgment structure under the condition of no constraint. And when the data has a sample unbalanced amount item type characteristic, the decision tree model is very easy to grow a deep and unbalanced structure. In practice, therefore, an upgrade algorithm of the decision tree is used, and a gradient upgrade model is used to alleviate the disadvantages of the decision tree mentioned above.
(2) Gradient lifting model
The gradient lifting model (Gradient Boosting Machine) is a classical machine learning model algorithm that integrates a plurality of simple decision trees to obtain a flexible predictive model with excellent fitting capability. In the model, taking the residual error of each decision tree fitting the target as the learning target of the next decision tree, and reciprocating until the model converges or makes a decision. The specific gradient lifting model implementation used in the event prediction model is based on LightGBM. This is a gradient lifting model framework of Microsoft (Microsoft) open source. The method has the advantages of high training speed, high memory use efficiency, high prediction precision, support of various prediction scenes (support of classification, regression and sequencing) and the like, and is widely applied in the industry.
The LightGBM has two special advantages over the general gradient lifting algorithm implementation. First, lightGBM uses the growth logic generated by leaf nodes as opposed to the logic of layer-wise growth of conventional decision trees. At each growth of the decision tree, one leaf node with the largest splitting gain (typically the largest data size) is found from all the current leaf nodes, and then split, and so on. This growth strategy would bring about more information gain improvement while maintaining the same number of leaf nodes. On the processing of the category characteristics, the LightGBM can find out the optimal cutting of the category characteristics, namely, the splitting mode of the many-vs-many. This also solves the phenomenon that the tree model is prone to grow imbalance when the number of categories in the item type feature is large.
Compared with other decision tree algorithms which need to traverse data and sort the data, the LightGBM adopts a histogram algorithm to preprocess the data, so that the memory use efficiency of the model in operation is improved, and the sensitivity of the tree model to abnormal points in growth is further improved due to the fact that the histogram algorithm is used for discretizing the data.
The invention changes the gradient lifting model introduced above, uses a two-step modeling method, a random intercept model and a random effect model to resist the sparsity problem existing in the sample and the unbalanced condition of the samples of each street and town, and finally establishes a grid event quantity prediction model, see fig. 2 and 3.
(3) Two-step modeling
Two-step modeling is a modeling method that is used to address the impact of data sparsity (there are a large number of zero values in the data) on the predictive model. When the grid hotline data is amortized to street town/12 hours, more than 70% of the sample events in the data occur with zero, i.e. no events occur. Fitting the model using this data would make the prediction of the model less significant. In order to reduce the influence of the problem on the model, the model is classified into an event occurrence probability model and an event occurrence number model. In the event occurrence probability model, a gradient lifting model is used to predict whether a corresponding grid hotline event will occur within 12 hours of each street town, and the probability of whether a corresponding event occurs within 12 hours of each street town is taken as a main output. In the event occurrence number model, a gradient lifting model is used for predicting the number of events occurring within 12 hours once the events occur at each street town. Multiplying the probability of expected occurrence of the event by the expected number of events once occurring, and finally obtaining the number value of occurrence of the grid hotline.
Two-step modeling is widely used in economics, sociology and medicine. In economics, two-step modeling is also commonly referred to as a barrier model. In economics, he is used in subdivision pricing practice, where the supply is generally set with some hard settings (i.e. fences). Customers meeting the criteria are discounted, and the economist is referred to as the price discrimination barrier model (Hurdle Model of Price Discrimination). Because of the presence of such fences, the purchase data of users often also show a sparse phenomenon, i.e. the actual consumer population often occupies a small part of the population used. This is also consistent with the time sparsity encountered in the grid hotline scene model. The use of two-step modeling herein mitigates the phenomenon of data sparseness.
(4) Random intercept model
In training the event occurrence number Model, in order to solve the problem that the Model is easy to be overfitted due to the imbalance of street and town samples (the event samples are concentrated on a small part of street and town) in the characteristics, a Random-Intercept Model (Random-inter Model) is used for estimating the reference level of each street and town corresponding to the occurrence of the event, and the estimated reference level is used as the pre-input of the event number prediction Model in the two-step modeling.
(5) Random effect model
The stochastic effect model (random effects models) is a generalization of the classical linear model, which regards fixed regression coefficients as random variables, typically assumed to be from normal distributions. If some of the coefficients in the model are random, others are fixed, commonly referred to as mixed models (mixed models). The introduction of random effects allows for a certain correlation between individual observations and is therefore a suitable choice for fitting data from non-independent observations. The grid hotline quantity data is also data of an independent observation as described above. When the mixed/random effect model is used, the estimated value of an individual with fewer samples during model fitting can be' closed to the middle value of the group due to the compression (kringing) phenomenon during random effect fitting, and the phenomenon also limits that the street town with fewer partial samples but more abnormal occurrence is not easily influenced by abnormal values which occur sporadically during estimation.
2. Feature mining
In conjunction with fig. 4, three types of features, time, historical live data, and weather elements, are used as inputs to the model in the mining of the model inputs.
In terms of time characteristics, hours, months, and holidays are used to capture the relationship between grid event occurrence number and time.
In mining of historical live data features, a statistical index of Partial Auto-Correlation (Partial Auto-Correlation) is used to determine the postponement of the number of event occurrences over time. By this index, it was found that the number of events occurring before 12 hours, 24 hours, 36 hours and 1 week had a statistically significant correlation with the number of events occurring at the present time. These features then also serve as inputs to the model.
The instantaneous and historical influences of three meteorological conditions, namely precipitation, air temperature and wind speed, on the occurrence quantity of the event are mainly considered in the excavation of the characteristics of meteorological elements. The instantaneous weather element uses the current hour cumulative rainfall, the current hour maximum wind speed, the current hour maximum temperature, and the current hour minimum temperature as the inputs of the model. In addition to the statistics commonly used on gas phase standards (e.g., 12 hours and 24 hours accumulated rainfall used to determine whether to storm or not), the hysteresis relationship between the number of events and different meteorological elements in time is calculated by using Cross-Correlation coefficient (Cross-Correlation), and finally, statistical indexes (such as maximum, minimum, average and accumulated value) of the strong wind, precipitation and air temperature elements in 6 hours, 36 hours and 48 hours are added as the input of the model.
Partial autocorrelation coefficients are common statistical indicators in time series analysis and modeling. It is typically used to find out whether a sample on a sequence has a correlation with a sample before a period of time. And in a specific mathematical definition he measures X t And X is t-k And the correlation coefficient after the middle k-1 interference term variables are removed. Typically, in the classical model autocorrelation-averaging (ARMA) model of time series. The partial autocorrelation coefficients are used to determine the number of autoregressive terms in the ARMA model. In the grid hotline event number prediction model, a time delay term of the grid hotline event number is taken as a characteristic to be put into the gradient lifting model. The cross-correlation coefficient is also a statistical index which is more commonly used in time series analysis and modeling. Unlike partial autocorrelation coefficients, tadine is used to mine correlations between different long-term sequences. If sequence A and sequence B have a significant correlation coefficient with respect to the lag term K, then the time lag of sequence A and sequence B can be said to be K time units. Due to grid hotlineThe time sequence of the occurrence number of the events, the strong wind, the rainfall and the air temperature which are equal in length from hour to hour can be used for digging out the historical meteorological element data which is most relevant to the number of the grid hotline events and is used as the characteristic to be put into the input of the grid hotline event gradient lifting model.
3. Grid quantity prediction model construction and scene model
The model construction process changes the gradient lifting model, and uses a two-step modeling method, a random intercept model and a random effect model to resist the sparsity problem existing in the samples and the unbalanced condition of the samples of each street and town, so as to finally establish a grid event quantity prediction model.
In addition to predicting the total amount of grid events, 6 grid event subclass scenes closely related to meteorological influences are selected, and a grid scene model is established. The following are selected subclass scenes: street trees, public greenhouses, vehicle moving, help seeking, overhead line falling, greenhouses guard bars and community greening. For grid scenarios (e.g., cell greening, street tree, etc.) where there is a small number of occurrences and sample imbalance between the street and town, by using two-step modeling and random intercept models, the phenomenon of grid scenario model overfitting can be significantly reduced.
Unlike the aggregate model of the grid event number prediction model (i.e., the number of event occurrences), the phenomenon of sparse tag data in the grid scene model is particularly pronounced. And the sparseness of the events of different scene types is not nearly the same. Different model parameters are used for fitting and training the model for different scene models. And also to some extent in the training strategy of the model. In the aggregate model, the center of the two-step modeling is placed on the event occurrence number model, i.e., how much of the event occurrence number is predicted once it occurs. However, in the scene model, due to the sparse data, the number of the event cases is often only the same. Therefore, in the scene model, the training center is shifted towards the occurrence direction of the event (namely, the center is placed on the event occurrence probability model), namely, the occurrence probability of the subclass event is changed under the current weather conditions, in the time and in the street town. According to such model training strategy changes, training parameters of the scene model are also modified. The training parameters of each scene model are determined by combining the cross-validation and the manual validation, and the following training parameters are the training parameters of each scene model:
Street tree: the number of training wheels of the event occurrence probability model is 600, and the number of training wheels of the event occurrence probability model is 300;
public green land: the number of training wheels of the event occurrence probability model is 500, and the number of training wheels of the event occurrence probability model is 500;
the number of training wheels of the event occurrence probability model is 400, and the number of training wheels of the event occurrence number model is 600;
overhead line fall: the number of training wheels of the event occurrence probability model is 700, and the number of training wheels of the event occurrence probability model is 300;
green land guard bar: the number of training wheels of the event occurrence probability model is 500, and the number of training wheels of the event occurrence probability model is 500;
and (3) greening a cell: the number of training rounds of the event occurrence probability model is 400, and the number of training rounds of the event occurrence probability model is 300.
4. Hot line quantity prediction model construction and scene model
The model construction process changes a gradient lifting model, and uses a two-step modeling method, a random intercept model and a random effect model to resist the sparsity problem existing in the samples and the unbalanced condition of the samples of each street and town, so that a hot line event quantity prediction model is finally established.
In addition to predicting the total quantity of hotline events, 5 hotline sub-class scenes of taxis, road maintenance, fault repair, greenbelt greening, water drainage and pollution discharge management are also predicted and modeled, and a scene model is established. For grid scenarios (e.g., drainage and blowdown management, road maintenance, etc.) where there is a small number of occurrences and sample imbalance between towns, by using two-step modeling and random intercept models, the phenomenon of model overfitting can be significantly reduced.
Unlike the aggregate model, the phenomenon of tag data rareness is particularly evident in the hotline scene model. And the sparseness of the events of different scene types is not nearly the same. So for different scene models we also use different model parameters for fitting and training the model. And also to some extent in the training strategy of the model. In the hot-line total model, the center of the two-step modeling is placed on the quantity prediction model, i.e., how much of the quantity of events occurs is predicted once the events occur. However, in the scene model, due to the sparse data, the number of the event cases is often only the same. Therefore, in the scene model, the training center shifts to the direction of whether the event occurs, namely the current weather condition and the time, and the probability of occurrence of the subclass event is changed in the street town. According to such model training strategy changes, training parameters of the scene model are also modified. The training parameters of each scene model are determined by combining the cross-validation and the manual validation, and the following training parameters are the training parameters of each scene model:
taxi: the number of training wheels of the event occurrence probability model is 600, and the number of training wheels of the event occurrence probability model is 300;
And (3) road maintenance: the number of training wheels of the event occurrence probability model is 600, and the number of training wheels of the event occurrence probability model is 400;
and (3) fault repair: the number of training wheels of the event occurrence probability model is 500, and the number of training wheels of the event occurrence probability model is 500;
green land greening: the number of training wheels of the event occurrence probability model is 600, and the number of training wheels of the event occurrence probability model is 500;
and (3) drainage and pollution discharge management: the number of training rounds of the event occurrence probability model is 500, and the number of training rounds of the event occurrence probability model is 600.
5. 110 quantity prediction model construction and scene model
The model construction process changes the gradient lifting model, and uses a two-step modeling method, a random intercept model and a random effect model to resist the sparsity problem existing in the samples and the unbalanced condition of the samples of each street and town, so as to finally establish a 110 weather disaster quantity prediction model.
Different from the hot line grid event quantity prediction model, quantity prediction modeling is only carried out on two types of disaster scenes, namely, the events of strong wind and heavy rain. In the area range, the method is different from the split street town prediction of grid hotlines, and the data sparseness problem of the 110 weather disaster quantity is more serious, so that the method only performs the split street quantity prediction. Over a time horizon, an hour-by-hour prediction is selected.
For scenes where there is a small number of occurrences and sample imbalance between street and town, the phenomenon of model overfitting can be significantly reduced by using two-step modeling, random intercept model and random effect model. For the difference of the distribution rule of 110 weather disasters and the distribution rule of grid hotline events, different model parameters are also used for fitting and training the model. According to such model training strategy changes, training parameters of the scene model are also modified. The training parameters of each scene model are determined by combining the cross-validation and the manual validation, and the following training parameters are the training parameters of each scene model:
stormwater scene: the number of training wheels of the event occurrence probability model is 500, and the number of training wheels of the event occurrence probability model is 500;
high wind scene: the number of training rounds of the event occurrence probability model is 600, and the number of training rounds of the event occurrence probability model is 500.
3. Event risk early warning model
According to statistical analysis of the occurrence number of the historical events and consideration of actual use scenes, an event early warning standard is formulated by combining the occurrence number of the historical events with the absolute threshold value of the occurrence number of the events. The early warning is now 5 grades, and the grading rule is as follows:
(1) No early warning exists: less than 80% quantiles or less than 9;
(2) blue early warning: 80% quantiles and 9 inclusive;
(3) yellow early warning: 90% or more quantiles and 15 or more;
(4) orange early warning: 95% fraction or more and 25% or more;
(5) red early warning: 99% or more and 40% or more.
For 110 weather conditions, because the number of events is sparsely distributed, the early warning standard is modified as follows:
(1) no early warning exists: less than 80% quantiles or less than 9;
(2) blue early warning: 80% or more and 5 or more;
(3) yellow early warning: 90% or more and 10 or more;
(4) orange early warning: 95% quantiles and 15% or more;
(5) red early warning: 99% fraction or more and 25% or more.
According to this method, the number of events predicted to be increased or decreased in comparison with the average day may be calculated using the median of the number of cumulative events occurring 12 hours in history as the reference for comparing the events.
4. Meteorological impact index model
1. Grid event weather impact index
Since the LightGBM is a gradient lifting model based on a tree model, the information gain of each feature in different value ranges for the whole model can be obtained in the model training process. The percentage contribution degree brought by different features can be calculated by carrying out aggregation and addition calculation on the information gains brought by the different features and normalizing the contribution degree of all the features. From the model explanatory point of view, the percentage contribution degree real quantifies the decision logic in the LightGBM model, i.e. which features bring the greatest gain to the model under the current input conditions, and the importance degree of the features from the side.
Based on the grid event prediction model, the percentage contribution degree of the event to the event quantity prediction by calculating each feature is calculated. Because three main meteorological features of strong wind, precipitation and air temperature are included in the input of the model, the percentage contribution value of the strong wind, precipitation and air temperature to the event is obtained; and calculating the absolute deviation of the percentage contribution value of the current meteorological feature and the wind and daily standard by taking the percentage contribution value of the three meteorological features under the wind and daily conditions (no rainfall, wind speed of 2 levels and air temperature of 25 ℃) as a standard, wherein the deviation is the influence index of the three meteorological features.
Through the weather influence index analysis of the grid scene model, the influence of precipitation on public greenbelts and vehicle moving seeking help can be found to be more, the influence of strong wind on the guardrails of the greenbelts when overhead lines fall can be found to be more, and the influence of air temperature on the greening of the communities can be found to be more. The street tree has a close relationship with strong wind, precipitation and air temperature.
2. Hot line event weather impact index
Similar to the calculation method of the grid model weather impact index, the weather impact index of the hotline model is based on the hotline event prediction model. And based on the deviation method, the influence of three weather features of strong wind, precipitation and air temperature on the total quantity of the hot wire and sub-class events is calculated.
Through the weather influence index analysis of the hot line scene model, the situation that the water and pollution discharge management and the road maintenance are more influenced by water fall, the fault repair and the greenbelt greening are more influenced by strong wind, and the taxi is more influenced by air temperature can be found.
5. Model verification mechanism
In order to simulate the scene of the real online running of the model, a rolling type cross-validation mode is used for validating the prediction capability of the model. The following is a rolling cross-validation flow:
(1) let the present time be t. Then use the data of (t-1, t-2, …, t-12) as a training set and train the event prediction model M t ；
(2) Using model M t Making and recording a prediction result on a data set with time t;
(3) update time t, i.e., t=t+1;
(4) repeating the step (1) until t is the latest time point, and counting the error rate of all the predicted results.
6. Model update optimization
The model adopts an automatic update training strategy, and the update frequency is the daily level, namely the model is retrained by using the latest update data every day. If adjustments are required to the update frequency of the model, then modifications can be made to the cron syntax in the line of commands in the file model_crontab.
If the model needs to be updated and replaced, the updated homonymous model can be directly put into a corresponding folder. If the model needs to be manually trained in the deployed server, the following commands can be executed to manually update the model.
The invention further provides a city operation management big data analysis and prediction system based on the weather, so as to support the realization of the city operation management big data analysis and prediction method based on the weather. The system can be stored in a computer, and when the computer runs the system, the steps of the weather-based urban operation management big data analysis and prediction method are executed.
Likewise, it should be noted that in order to simplify the presentation disclosed herein and thereby aid in understanding one or more inventive embodiments, various features are sometimes grouped together in a single embodiment, figure, or description thereof. This method of disclosure, however, is not intended to imply that more features than are presented in the claims are required for the subject application. Indeed, less than all of the features of a single embodiment disclosed above.
While the present application has been described with reference to the present specific embodiments, those of ordinary skill in the art will recognize that the above embodiments are for illustrative purposes only, and that various equivalent changes or substitutions can be made without departing from the spirit of the present application, and therefore, all changes and modifications to the embodiments described above are intended to be within the scope of the claims of the present application.
The technology, shape, and construction parts of the present invention, which are not described in detail, are known in the art.
Claims (8)
1. A city operation management big data analysis and prediction method based on weather is characterized by comprising the following steps:
collecting data sources, and generating an event quantity prediction model training set, wherein the data sources comprise weather automatic station data, grid-point weather element prediction data, grid event data, hot line event data and 110 event weather disaster data;
based on the training set, modeling is carried out on the occurrence probability and the occurrence quantity of the events respectively, and then a two-step modeling method is adopted to form an event quantity prediction model;
in the event occurrence probability model, predicting whether an event occurs in t time by adopting a gradient lifting model, and taking the probability of whether the event occurs in t time by adopting each object as output;
in the event occurrence quantity model, predicting the quantity of events occurring in t time when each object is once the event occurs by adopting a gradient lifting model;
multiplying the expected occurrence probability of the event by the expected number of the event once to obtain the number value of the event occurrence, and establishing an event number prediction model;
the gradient lifting model is a gradient lifting model based on a LightGBM, a random intercept model and a random effect model are introduced on the basis of the gradient lifting model to optimize and upgrade the model, and the object is a whole city, a administrative district or a street town.
2. The weather-based urban operation management big data analysis and prediction method according to claim 1, further comprising feature mining for time data, historical and live weather element data, and historical and live urban operation data, and adding model training as input of the event quantity prediction model, wherein the historical and live urban operation data comprises grid data, hot line data, 110 weather disaster data.
3. The weather-based urban operation management big data analysis and prediction method according to claim 2, wherein in the mining of the historical and live urban operation data characteristics, a partial autocorrelation coefficient is used to determine a delay in the number of occurrence of events over time.
4. The weather-based urban operation management big data analysis and prediction method according to claim 2, wherein instantaneous and historical influences of each weather element on the occurrence number of the event are considered in mining of the data characteristics of the historical and live weather elements, and the time lag relation between the occurrence number of the event and the different weather elements is calculated by using a cross correlation coefficient on the influence of the historical weather elements on the occurrence number of the event.
5. The weather-based urban operation management big data analysis and prediction method according to claim 1, wherein the method further comprises the steps of modeling and selecting event scenes closely related to weather influences by adopting the two-step method and establishing a scene model while establishing the event quantity prediction model.
6. The weather-based urban operation management big data analysis and prediction method according to claim 1, further comprising the steps of establishing an event risk early warning model: and establishing an event risk early warning model by combining the historical event occurrence number score and an absolute threshold value of the event occurrence number, and formulating an event early warning standard.
7. The weather-based urban operation management big data analysis and prediction method of claim 1, further comprising establishing a weather-influencing index model:
according to the event quantity prediction model, weather element features comprise weather influence indexes formed by three types of weather element features including wind speed, precipitation and air temperature, the weather influence indexes are used for quantifying influence degrees of different weather elements on urban running conditions, and based on the event quantity prediction model, the percentage contribution degree of each weather element feature to the event quantity prediction is calculated;
And calculating the absolute deviation of the percentage contribution value of the current meteorological element characteristic and the standard by taking the percentage contribution value of the meteorological element characteristic under the conditions of no rainfall, 2-level wind speed and 25 ℃ of air temperature and Japanese as the standard, and taking the absolute deviation as an influence index of the meteorological element characteristic.
8. A city operation management big data analysis prediction system based on weather is characterized in that: the system is used for executing the weather-based urban operation management big data analysis and prediction method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110558687.0A CN113191568B (en) | 2021-05-21 | 2021-05-21 | Meteorological-based urban operation management big data analysis and prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110558687.0A CN113191568B (en) | 2021-05-21 | 2021-05-21 | Meteorological-based urban operation management big data analysis and prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113191568A CN113191568A (en) | 2021-07-30 |
| CN113191568B true CN113191568B (en) | 2024-02-02 |
Family
ID=76984763
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110558687.0A Active CN113191568B (en) | 2021-05-21 | 2021-05-21 | Meteorological-based urban operation management big data analysis and prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113191568B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116862071B (en) * | 2023-07-18 | 2024-02-13 | 杭州博晟科技有限公司 | Urban operation index data prediction method based on mixing model |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105260603A (en) * | 2015-10-14 | 2016-01-20 | 成都信息工程大学 | Climatic event risk evaluation method and system |
| CN109359770A (en) * | 2018-10-11 | 2019-02-19 | 中国疾病预防控制中心环境与健康相关产品安全所 | A kind of model and method based on machine learning prediction heatstroke generation |
| CN109472283A (en) * | 2018-09-13 | 2019-03-15 | 中国科学院计算机网络信息中心 | A kind of hazardous weather event prediction method and apparatus based on Multiple Incremental regression tree model |
| CN109978222A (en) * | 2019-01-04 | 2019-07-05 | 国网山东省电力公司电力科学研究院 | A kind of wind-powered electricity generation climbing probability of happening prediction technique and system based on Bayesian network |
| CN110290120A (en) * | 2019-06-12 | 2019-09-27 | 西安邮电大学 | A kind of timing evolved network safe early warning method of cloud platform |
| CN110706823A (en) * | 2019-11-15 | 2020-01-17 | 广州地理研究所 | Method for predicting respiratory system disease morbidity based on lag analysis and LSTM |
| CN111612899A (en) * | 2019-02-25 | 2020-09-01 | 中国石油化工股份有限公司 | Geological modeling method for fractured-vuggy carbonate reservoir |
| CN111612233A (en) * | 2020-05-13 | 2020-09-01 | 云南电网有限责任公司电力科学研究院 | Method and device for obtaining importance scores of power distribution network line re-jump influence factors |
| CN111929748A (en) * | 2020-09-17 | 2020-11-13 | 南京浦蓝大气环境研究院有限公司 | Meteorological element forecasting method and system |
| CN112285807A (en) * | 2019-11-12 | 2021-01-29 | 北京嘉韵楷达气象科技有限公司 | Meteorological information prediction method and device |
| WO2021077729A1 (en) * | 2019-10-23 | 2021-04-29 | 国网电力科学研究院武汉南瑞有限责任公司 | Lightning prediction method |
-
2021
- 2021-05-21 CN CN202110558687.0A patent/CN113191568B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105260603A (en) * | 2015-10-14 | 2016-01-20 | 成都信息工程大学 | Climatic event risk evaluation method and system |
| CN109472283A (en) * | 2018-09-13 | 2019-03-15 | 中国科学院计算机网络信息中心 | A kind of hazardous weather event prediction method and apparatus based on Multiple Incremental regression tree model |
| CN109359770A (en) * | 2018-10-11 | 2019-02-19 | 中国疾病预防控制中心环境与健康相关产品安全所 | A kind of model and method based on machine learning prediction heatstroke generation |
| CN109978222A (en) * | 2019-01-04 | 2019-07-05 | 国网山东省电力公司电力科学研究院 | A kind of wind-powered electricity generation climbing probability of happening prediction technique and system based on Bayesian network |
| CN111612899A (en) * | 2019-02-25 | 2020-09-01 | 中国石油化工股份有限公司 | Geological modeling method for fractured-vuggy carbonate reservoir |
| CN110290120A (en) * | 2019-06-12 | 2019-09-27 | 西安邮电大学 | A kind of timing evolved network safe early warning method of cloud platform |
| WO2021077729A1 (en) * | 2019-10-23 | 2021-04-29 | 国网电力科学研究院武汉南瑞有限责任公司 | Lightning prediction method |
| CN112285807A (en) * | 2019-11-12 | 2021-01-29 | 北京嘉韵楷达气象科技有限公司 | Meteorological information prediction method and device |
| CN110706823A (en) * | 2019-11-15 | 2020-01-17 | 广州地理研究所 | Method for predicting respiratory system disease morbidity based on lag analysis and LSTM |
| CN111612233A (en) * | 2020-05-13 | 2020-09-01 | 云南电网有限责任公司电力科学研究院 | Method and device for obtaining importance scores of power distribution network line re-jump influence factors |
| CN111929748A (en) * | 2020-09-17 | 2020-11-13 | 南京浦蓝大气环境研究院有限公司 | Meteorological element forecasting method and system |
Non-Patent Citations (4)
| Title |
|---|
| Short-Term Wind Speed and Wind Power Prediction Based on Meteorological Model Modification;Liu, Tianze;《9th IEEE Data Driven Control and Learning Systems Conference (DDCLS)》;全文 * |
| 基于沉积模拟的虚拟井建立与深水峡谷相控建模策略——以琼东南盆地中央峡谷陵水段为例;付超;《海相油气地质》(第2期);全文 * |
| 风电场功率预测的研究进展及发展趋势;李根银;《排灌机械工程学报》(第3期);全文 * |
| 龙川县森林火险等级预报方法的建立及应用;曾钦文;曾思亮;王辉;刘东瑶;;广东气象(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113191568A (en) | 2021-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Dong et al. | Hourly energy consumption prediction of an office building based on ensemble learning and energy consumption pattern classification | |
| CN107610469B (en) | Day-dimension area traffic index prediction method considering multi-factor influence | |
| Serrano-Guerrero et al. | A new interval prediction methodology for short-term electric load forecasting based on pattern recognition | |
| CN112749904A (en) | Power distribution network fault risk early warning method and system based on deep learning | |
| CN106845080B (en) | Based on the modified Scene Tourist meteorological disaster intelligent Forecasting of difference | |
| CN112285807A (en) | Meteorological information prediction method and device | |
| CN114676822B (en) | Multi-attribute fusion air quality forecasting method based on deep learning | |
| CN113496314B (en) | Method for predicting road traffic flow by neural network model | |
| CN114202103A (en) | Machine learning-based holiday scenic spot passenger flow prediction method | |
| CN112149887A (en) | PM2.5 concentration prediction method based on data space-time characteristics | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN106526710A (en) | Haze prediction method and device | |
| CN117494034A (en) | Air quality prediction method based on traffic congestion index and multi-source data fusion | |
| CN117935561B (en) | Intelligent traffic flow analysis method based on Beidou data | |
| CN112668821A (en) | Distribution line risk analysis method based on insulator fault probability of sand blown region | |
| CN114331206B (en) | Point location addressing method and device, electronic equipment and readable storage medium | |
| CN113191568B (en) | Meteorological-based urban operation management big data analysis and prediction method and system | |
| CN114580762A (en) | Hydrological forecast error correction method based on XGboost | |
| Hou et al. | Spatial distribution assessment of power outage under typhoon disasters | |
| Cai et al. | Exploratory analysis of massive data for distribution fault diagnosis in smart grids | |
| CN117332909B (en) | Multi-scale urban waterlogging road traffic exposure prediction method based on intelligent agent | |
| CN111145535B (en) | Travel time reliability distribution prediction method under complex scene | |
| JP2003090887A (en) | Predication system and prediction method of instantaneous voltage drop by thunderbolt | |
| CN107977727B (en) | Method for predicting blocking probability of optical cable network based on social development and climate factors | |
| CN115936792A (en) | Power transmission and transformation project cost prediction method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |