CN117471575A - Typhoon wave height forecasting method based on BO-LSTM neural network model - Google Patents

Typhoon wave height forecasting method based on BO-LSTM neural network model Download PDF

Info

Publication number
CN117471575A
CN117471575A CN202311834527.XA CN202311834527A CN117471575A CN 117471575 A CN117471575 A CN 117471575A CN 202311834527 A CN202311834527 A CN 202311834527A CN 117471575 A CN117471575 A CN 117471575A
Authority
CN
China
Prior art keywords
typhoon
neural network
network model
typhoons
lstm neural
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.)
Granted
Application number
CN202311834527.XA
Other languages
Chinese (zh)
Other versions
CN117471575B (en
Inventor
陈永平
秦知朋
徐晓武
刘畅
韩韬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hohai University HHU
Original Assignee
Hohai University HHU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hohai University HHU filed Critical Hohai University HHU
Priority to CN202311834527.XA priority Critical patent/CN117471575B/en
Publication of CN117471575A publication Critical patent/CN117471575A/en
Application granted granted Critical
Publication of CN117471575B publication Critical patent/CN117471575B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G01W1/10—Devices for predicting weather conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • 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
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Environmental & Geological Engineering (AREA)
  • Tourism & Hospitality (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Environmental Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • General Business, Economics & Management (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Atmospheric Sciences (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Medical Informatics (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Ecology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a typhoon wave height forecasting method based on a BO-LSTM neural network model, and belongs to the technical field of typhoon wave forecasting. The method comprises the following steps: constructing a large number of virtual typhoons in the sea area of a research site by adopting an empirical path method based on a nuclear density estimation method; screening typhoon data affecting a research site, inputting the typhoon data into a Holland typhoon empirical model, generating typhoon wind fields corresponding to typhoons of each site, and constructing a typhoon wave height data set of the research site by combining with a SWAN wave numerical mode; and optimizing and selecting super parameters of the LSTM neural network model based on the BO algorithm, and performing model unfolding training and testing to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model for typhoon wave forecasting. The invention solves the problem of insufficient sample data volume in the typhoon wave intelligent prediction, effectively reduces the time and effort consumed by the training of the neural network model, and improves the accuracy and timeliness of typhoon wave high prediction.

Description

Typhoon wave height forecasting method based on BO-LSTM neural network model
Technical Field
The invention relates to the technical field of typhoon wave forecasting, in particular to a typhoon wave height forecasting method based on a BO-LSTM neural network model.
Background
The typhoon wave is usually several times higher than the common sea wave, and the wave height can reach tens of meters in extreme cases. In an area sufficiently affected by typhoons, typhoons may have a great disastrous effect on maritime shipping, marine drilling, naval operations, offshore construction, and the like. Therefore, the method for forecasting the stormy waves of the platform rapidly and accurately has very important significance for guaranteeing the life and property safety of human beings.
The current typhoon wave forecasting method is mainly divided into an experience forecasting method, a numerical mode forecasting method and an intelligent forecasting method. The experience forecasting method is simple and feasible, but has obvious regionalization and lacks wide applicability; the numerical mode forecasting method has higher forecasting precision and good universality, but a large amount of computing resources and time are consumed due to the fact that the input items are more and the boundary conditions are complex; the intelligent forecasting method gives consideration to forecasting precision, calculation power consumption and model universality, and provides a new development direction for typhoon wave forecasting. However, the current intelligent forecasting method represented by the neural network model still has some problems, such as insufficient sample data volume required by model training and extremely complicated selection of super parameters in the model training process, which also causes some limitations and disadvantages of the intelligent forecasting method in practical application.
Disclosure of Invention
Aiming at the problems of insufficient sample data volume and the like in the intelligent typhoon wave prediction, the invention provides a typhoon wave height prediction method based on a BO-LSTM neural network model, which effectively solves the problem of insufficient training sample data volume in the intelligent typhoon wave height prediction; the time and effort consumed by training the neural network model are effectively reduced, and the accuracy and timeliness of typhoon wave height forecasting are improved.
In order to achieve the above object, the present invention is realized by the following technical scheme:
the typhoon wave height forecasting method based on the BO-LSTM neural network model comprises the following steps:
s1: collecting historical typhoon data of a sea area of a research site, and constructing virtual typhoons of the sea area of the research site by adopting an empirical path method based on a nuclear density estimation method based on the historical typhoon data;
s2: typhoon data influencing a research site are screened from historical typhoons and virtual typhoons, and are input into a Holland typhoon experience model to generate typhoon wind fields corresponding to each typhoon; the typhoon data comprise longitude, latitude, central lowest air pressure and near-central maximum wind speed of typhoons;
s3: inputting the typhoon wind field into a SWAN wave numerical mode, generating a typhoon wave height data set of a research site during typhoon, and screening out an effective wave height value of the research site which is larger than 1.25 m from the typhoon wave height data set;
s4: integrating the effective wave height value of the research site with typhoon data and research site meteorological data at corresponding moments to jointly construct a sample database; the research site meteorological data comprise wind speed and air pressure of a research site;
s5: dividing data corresponding to the virtual typhoons in the sample database into a training set and a verification set according to the proportion of 7:3, wherein the data corresponding to the historical typhoons are used as a test set;
s6: selecting typhoon data, research site weather data and research site effective wave height values in the time period from t-n to t as input sequences, respectively selecting research site effective wave height values at the moments of t+1h, t+3h, t+6h and t+12h as target results, and constructing a nonlinear mapping relation among the typhoon data, the research site weather data and the research site effective wave height values by utilizing an LSTM neural network model;
s7: optimizing and selecting the super parameters of the LSTM neural network model by adopting a BO algorithm; the super parameters comprise the number of neurons, the learning rate, the batch size and the iteration times;
s8: training the LSTM neural network model based on the data in the training set and the verification set to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model; and evaluating and testing the typhoon wave height forecasting model based on the data in the test set, and forecasting the typhoon wave height through the typhoon wave height forecasting model.
Preferably, in the step S1, an empirical path method based on a nuclear density estimation method is used to construct a virtual typhoon of a sea area where a research site is located, and the specific method includes:
s11: based on historical typhoon data, statistics and analysis are carried out on tropical cyclones of the sea area where the research site is located, probability distribution of annual occurrence frequency and initial characteristic parameters of the historical typhoons is obtained, and parameter combinations are formed by random sampling from the annual occurrence frequency and the initial characteristic parameters of the historical typhoons, so that the number of the virtual typhoons occurring in each year and the initial characteristic parameters of each virtual typhoons are determined; the initial characteristic parameters comprise typhoon occurrence time, initial moving speed, initial moving direction and initial near-center maximum wind speed;
s12: dividing the sea area range of a research site into grids with the angle of 5 degrees multiplied by 5 degrees, counting the initial and final probabilities of historical typhoons in the grids, determining the initial position of a virtual typhoons based on the initial probability of the historical typhoons in each grid, and fitting the probability distribution of the parameter variation of the historical typhoons in each grid by adopting a nuclear density estimation method, wherein the parameter variation comprises a shift speed variation, a shift direction variation and a near-center maximum wind speed variation;
the formula of the nuclear density estimation method is as follows:
；
in the method, in the process of the invention,is a probability density function;irepresent the firstiA number of samples of the sample were taken,mis the total number of samples;hfor the bandwidth to be available,Kis a kernel function; />Represents a sample value, and->Obeying independent same distribution; />Representing the sample value +.>Is the average value of (2);
s13: for the virtual typhoons with initial characteristic parameters, calculating the next node position of the virtual typhoons by taking 3h as a time step, randomly extracting a group of parameter variation combinations based on probability distribution of parameter variation of the historical typhoons in grids where the node positions are positioned, and updating key characteristic parameters of the virtual typhoons, wherein the key characteristic parameters of the virtual typhoons comprise typhoons moving speed, moving direction and near-center maximum wind speed; and (5) circulating the above process until the virtual typhoon termination condition is met.
Preferably, the virtual typhoon termination condition includes:
when a termination command is extracted from the grid where the virtual typhoon is located, the maximum wind speed near the center of the typhoon is lower than 5.2m/s or the sea temperature at the position where the typhoon is located is lower than 10 ℃, the virtual typhoon is immediately terminated;
assigning a value to the central lowest air pressure of the virtual typhoons based on the Copula joint distribution function, and indicating that the simulation of one virtual typhoons is completed;
the Copula joint distribution functionThe formula of (2) is as follows:
；
in the method, in the process of the invention,、/>respectively indicate->Direction(s) (i.e. the directions of the eyes)>An edge distribution function of the direction; />Representing a Copula function;
when the simulation year reaches 1000 years, it indicates that all virtual typhoons are simulated.
Preferably, in the step S2, the identification criteria influencing the research site is specifically that the distance between the typhoon center and the research site is less than 250km.
Preferably, the Holland typhoon experience model consists of an air pressure model and a wind field model, and the formula of the air pressure model is as follows:
；
in the method, in the process of the invention,Pis the air pressure value at the calculation point,is typhoon center air pressure, < >>Is the peripheral air pressure->Is the maximum wind speed radius>Is to calculate the distance of the point to the typhoon center, < >>Is the Holland barometric pressure profile parameter;
the formula of the wind field model is as follows:
；
in the method, in the process of the invention,is +.>Gradient wind speed at>For coriolis force, ->Is the air density.
Preferably, in the step S3, the control equation of the SWAN wave numerical mode is expressed in the rectangular coordinate system as follows:
；
in the method, in the process of the invention,is wave action volume density; />For time (I)>、/>For geographical coordinates>Is relative wave frequency>Is wave direction; />And->For wave propagation speed at +.>And->Component in direction, +_>And->For waves +.>And->Propagation velocity of the space;Sis wave energy density;
wherein the wave energy densitySExpressed as:
；
in the method, in the process of the invention,representing the input of wind energy->、/>Representing wave energy exchanges resulting from three-and four-order nonlinear interactions, respectively, < >>、/>、/>Representing wave energy dissipation due to white waves, bottom friction and wave break-up, respectively.
Preferably, in the step S6, the front time sequences n of the input sequences selected when the wave heights at the time points t+1h, t+3h, t+6h and t+12h are predicted are 4h, 6h and 9h, respectively.
Preferably, in step S7, the optimization selection of the super parameters of the LSTM neural network model by using the BO algorithm specifically includes:
s71: model initialization: the LSTM neural network model comprises an input layer, a hidden layer and an output layer, wherein Tanh is selected as an activation function, adam is selected as an optimization algorithm, and super parameters are randomly initialized within a preset range;
s72: evaluating sampling points: taking the super-parameter combination as a sampling point to be brought into an LSTM neural network model, calculating the Root Mean Square Error (RMSE) of the output result of the model on a verification set, and evaluating;
s73: gaussian process: calculating probability distribution of a functional relation between the hyper-parameter combination and the RMSE based on a Gaussian process;
s74: acquisition function: combining the probability distribution in the Gaussian process, calculating a hyper-parameter combination corresponding to the RMSE minimum value based on an acquisition function, and updating a sampling point by using the hyper-parameter combination;
s75: steps S72 to S74 are repeated until the RMSE evaluated in step S72 is lower than a preset value or a specified number of cycles is reached.
Preferably, in the step S8, a typhoon wave height prediction model based on a BO-LSTM neural network model is obtained, and the specific method includes:
s81: initializing weight parameters of an LSTM neural network model, including an input gate, a forget gate and an output gate;
s82: the input sequence is transmitted forwards through an LSTM neural network model, in each time step, the LSTM neural network model receives the input of the current time step and the hidden state of the previous time step, then the output of the current time step and the new hidden state are calculated, and the output of the current time step and the new hidden state are transmitted to the next time step;
s83: comparing the output of the LSTM neural network model with a target result, selecting a Root Mean Square Error (RMSE) as a loss function, and calculating a loss value of the LSTM neural network model;
s84: according to the loss value, calculating gradient through a back propagation algorithm, and updating the weight parameter of the LSTM neural network model to reduce the loss value;
s85: training the LSTM neural network model based on the data in the training set and the verification set, and repeatedly executing the steps S82 to S84 until the preset iteration times are reached, so as to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model;
s86: and after training, calculating Root Mean Square Error (RMSE) and a Correlation Coefficient (CC) between the BO-LSTM forecast value and the SWAN simulation value based on a test set, and performing evaluation test on the typhoon wave height forecast model.
Compared with the prior art, the invention has the beneficial effects that: by simulating a large number of virtual typhoons and generating typhoon wave effective wave height values corresponding to the typhoon wave effective wave height values at research sites, the problem that the data quantity of training samples is insufficient in intelligent typhoon wave height prediction is effectively solved, and data support is provided for improving model prediction precision and prolonging model prediction duration; compared with the common LSTM, the BO algorithm intelligently selects the super-parameter configuration, so that the model can be converged to the optimal super-parameter combination more quickly, the time consumed in the training process of the neural network model is effectively reduced, and the accuracy and timeliness of the typhoon wave height prediction are further improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art. Wherein:
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram of an unstructured grid employed in the SWAN wave number mode in an embodiment of the present invention;
FIG. 3 is a graph showing the comparison between the BO-LSTM forecast values and the SWAN model values based on the test set in the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. It will be apparent that the described embodiments are some, but not all, embodiments of the invention. All other embodiments, which are obtained by a person skilled in the art based on the described embodiments of the invention, fall within the scope of protection of the invention.
As shown in fig. 1, in one embodiment of the present invention, a typhoon wave height forecasting method based on a BO-LSTM neural network model is provided, which includes the following steps:
s1: collecting historical typhoon data of a sea area of a research site, and constructing virtual typhoons of the sea area of the research site by adopting an empirical path method based on a nuclear density estimation method based on the historical typhoon data;
in one embodiment, an empirical path method based on a nuclear density estimation method is adopted to construct a virtual typhoon of a sea area where a research site is located, and the method specifically comprises the following steps:
s11: based on historical typhoon data, statistics and analysis are carried out on tropical cyclones of the sea area where the research site is located, probability distribution of annual occurrence frequency and initial characteristic parameters of the historical typhoons is obtained, and parameter combinations are formed by random sampling from the annual occurrence frequency and the initial characteristic parameters of the historical typhoons, so that the number of the virtual typhoons occurring in each year and the initial characteristic parameters of each virtual typhoons are determined; the initial characteristic parameters comprise typhoon occurrence time, initial moving speed, initial moving direction and initial near-center maximum wind speed;
s12: dividing the sea area range of a research site into grids with the angle of 5 degrees multiplied by 5 degrees, counting the initial and final probabilities of historical typhoons in the grids, determining the initial position of a virtual typhoons based on the initial probability of the historical typhoons in each grid, and fitting the probability distribution of the parameter variation of the historical typhoons in each grid by adopting a nuclear density estimation method, wherein the parameter variation comprises a shift speed variation, a shift direction variation and a near-center maximum wind speed variation;
the formula of the nuclear density estimation method is as follows:
；
in the method, in the process of the invention,is a probability density function;irepresent the firstiA number of samples of the sample were taken,mis the total number of samples;hfor the bandwidth to be available,Kis a kernel function; />Represents a sample value, and->Obeying independent same distribution; />Representing the sample value +.>Is the average value of (2);
s13: for the virtual typhoons with initial characteristic parameters, calculating the next node position of the virtual typhoons by taking 3h as a time step, randomly extracting a group of parameter variation combinations based on probability distribution of parameter variation of the historical typhoons in grids where the node positions are located, and updating key characteristic parameters of the virtual typhoons, wherein the key characteristic parameters of the virtual typhoons comprise typhoons moving speed, moving direction and near-center maximum wind speed; and (5) circulating the above process until the virtual typhoon termination condition is met.
In a specific embodiment, the virtual typhoon termination condition includes:
when a termination command is extracted from the grid where the virtual typhoon is located, the maximum wind speed near the center of the typhoon is lower than 5.2m/s or the sea temperature at the position where the typhoon is located is lower than 10 ℃, the virtual typhoon is immediately terminated;
assigning a value to the central lowest air pressure of the virtual typhoons based on the Copula joint distribution function, and indicating that the simulation of one virtual typhoons is completed;
copula joint distribution functionThe formula of (2) is as follows:
；
in the method, in the process of the invention,、/>respectively indicate->Direction(s) (i.e. the directions of the eyes)>An edge distribution function of the direction; />Representing a Copula function;
when the simulation year reaches 1000 years, it indicates that all virtual typhoons are simulated.
S2: typhoon data influencing a research site are screened from historical typhoons and virtual typhoons, and are input into a Holland typhoon empirical model to generate typhoons corresponding to each typhoon; typhoon data includes longitude, latitude, center lowest air pressure and near center maximum wind speed of typhoons;
in a specific embodiment, the Holland typhoon empirical model consists of an air pressure model and a wind field model, and the formula of the air pressure model is as follows:
；
in the method, in the process of the invention,Pis the air pressure value at the calculation point,is typhoon center air pressure, < >>Is the peripheral air pressure->Is the maximum wind speed radius>Is to calculate the distance of the point to the typhoon center, < >>Is the Holland barometric pressure profile parameter;
the formula of the wind field model is as follows:
；
in the method, in the process of the invention,is +.>Gradient wind speed at, e.g. calculating the distance r=10 km from the point to the typhoon centre, +.>Then the gradient wind speed at 10km from the tropical cyclone center is indicated; />For coriolis force, ->Is the air density.
In one preferred embodiment, the identification criteria that have an impact on the research site is specifically that the typhoon center is less than 250km from the research site.
S3: inputting a typhoon wind field into a SWAN wave numerical mode, generating a typhoon wave height data set of a research site during typhoon, and screening out an effective wave height value of the research site which is larger than 1.25 m from the typhoon wave height data set;
in a specific embodiment, the control equation of the SWAN wave numerical mode is expressed in rectangular coordinates as follows:
；
in the method, in the process of the invention,is wave action volume density; />For time (I)>、/>For geographical coordinates>Is relative wave frequency>Is wave direction; />And->For wave propagation speed at +.>And->Component in direction, +_>And->For waves +.>And->Propagation velocity of the space;Sis wave energyA density; />Representing the rate of change of the wave action density over time; />、/>Representing the propagation of the wave action density in space, < +.>Indicating the effects of the frequency shift caused by the water depth and the water flow change,representing refraction caused by water depth and water flow variation; />Sources and sinks representing all the generation, dissipation and redistribution of wave energy;
wherein the wave energy densitySExpressed as:
；
in the method, in the process of the invention,representing the input of wind energy->、/>Representing wave energy exchanges resulting from three-and four-order nonlinear interactions, respectively, < >>、/>、/>Representing wave energy dissipation due to white waves, bottom friction and wave break-up, respectively.
S4: integrating the effective wave height value of the research site with typhoon data and research site meteorological data at corresponding moments to jointly construct a sample database; the research site meteorological data comprise wind speed and air pressure of a research site;
s5: dividing data corresponding to the virtual typhoons in the sample database into a training set and a verification set according to the proportion of 7:3, wherein the data corresponding to the historical typhoons are used as a test set; the data herein includes typhoon data, research site weather data, and research site effective wave height values.
S6: selecting typhoon data, research site weather data and research site effective wave height values in the time period from t-n to t as input sequences, respectively selecting research site effective wave height values at the moments of t+1h, t+3h, t+6h and t+12h as target results, and constructing a nonlinear mapping relation among the typhoon data, the research site weather data and the research site effective wave height values by utilizing an LSTM neural network model;
preferably, the front time sequences n of the selected input sequences are 4h, 6h and 9h respectively when the wave heights at the time points t+1h, t+3h, t+6h and t+12h are predicted.
S7: optimizing and selecting super parameters of the LSTM neural network model by adopting a BO algorithm; the super parameters comprise the number of neurons, the learning rate, the batch size and the iteration times;
in one embodiment, step S7 specifically includes:
s71: model initialization: the LSTM neural network model comprises an input layer, a hidden layer and an output layer, wherein Tanh is selected as an activation function, adam is selected as an optimization algorithm, and super parameters are randomly initialized within a preset range;
s72: evaluating sampling points: taking the super-parameter combination as a sampling point to be brought into an LSTM neural network model, calculating the Root Mean Square Error (RMSE) of the output result of the model on a verification set, and evaluating;
s73: gaussian process: calculating probability distribution of a functional relation between the hyper-parameter combination and the RMSE based on a Gaussian process;
s74: acquisition function: combining probability distribution in the Gaussian process, calculating a hyper-parameter combination corresponding to the RMSE minimum value based on the acquisition function, and updating sampling points by using the hyper-parameter combination;
s75: steps S72 to S74 are repeated until the RMSE evaluated in step S72 is lower than a preset value or a specified number of cycles is reached.
S8: training the LSTM neural network model based on the data in the training set and the verification set to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model; and evaluating and testing the typhoon wave height forecasting model based on the data in the test set, and forecasting the typhoon wave height through the typhoon wave height forecasting model.
In one embodiment, a typhoon wave height prediction model based on a BO-LSTM neural network model is obtained, and the specific method comprises the following steps:
s81: initializing weight parameters of an LSTM neural network model, including an input gate, a forget gate and an output gate;
s82: the input sequence is transmitted forwards through an LSTM neural network model, in each time step, the LSTM neural network model receives the input of the current time step and the hidden state of the previous time step, then the output of the current time step and the new hidden state are calculated, and the output of the current time step and the new hidden state are transmitted to the next time step;
s83: comparing the output of the LSTM neural network model with a target result, selecting a Root Mean Square Error (RMSE) as a loss function, and calculating a loss value of the LSTM neural network model;
s84: according to the loss value, calculating gradient through a back propagation algorithm, and updating the weight parameter of the LSTM neural network model to reduce the loss value;
s85: training the LSTM neural network model based on the data in the training set and the verification set, and repeatedly executing the steps S82 to S84 until the preset iteration times are reached, so as to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model;
s86: and after training, calculating Root Mean Square Error (RMSE) and a Correlation Coefficient (CC) between the BO-LSTM forecast value and the SWAN simulation value based on the test set, and performing evaluation test on the typhoon wave height forecast model.
The method of the present invention will be further described below by taking the mountain and boat stations as examples to construct a model for predicting the wave height of the typhoon, as shown in fig. 1 to 3.
S1: collecting historical typhoons in the North-west pacific region of the sea area where the Zhoushan station is located in 1979-2022, and constructing a large number of virtual typhoons in the North-west pacific region by adopting an empirical path method based on a nuclear density estimation method based on the historical typhoons, wherein the method specifically comprises the following steps:
s11: carrying out statistics and analysis on 1128 tropical cyclones in North-west Pacific regions among 1979-2022 to obtain probability distribution of annual occurrence frequency and initial parameter characteristics of historical typhoons, and randomly sampling from the annual occurrence frequency and initial parameter characteristics of the historical typhoons to form parameter combinations so as to determine the number of the virtual typhoons occurring in each year and the initial characteristic parameters of each virtual typhoons; the initial characteristic parameters comprise typhoon occurrence time, initial shift speed, actual shift direction and initial near-center maximum wind speed;
s12: dividing a North Pacific ocean area into grids with the angle of 5 degrees multiplied by 5 degrees, counting the initial and final probabilities of historical typhoons in the grids, determining the initial position of a virtual typhoons based on the initial probability of the historical typhoons in each grid, and fitting probability distribution of parameter variation such as the movement speed variation, the movement direction variation and the near-center maximum wind speed variation of the historical typhoons in each grid by adopting a nuclear density estimation method;
the formula of the nuclear density estimation method is as follows:
；
in the method, in the process of the invention,is a probability density function;irepresent the firstiA number of samples of the sample were taken,mis the total number of samples;hfor the bandwidth to be available,Kis a kernel function; />Represents a sample value, and->Obeying independent same distribution; />Representing the sample value +.>Is the average value of (2);
s13: for a virtual typhoon with initial characteristic parameters, calculating the next node position of the virtual typhoon by taking 3h as a time step, randomly extracting a group of parameter variation combinations based on probability distribution of parameter variation of historical typhoons in grids where the node positions are located, and updating three key characteristic parameters (typhoons moving speed, moving direction and near-center maximum wind speed) of the virtual typhoons; and (5) circulating the above process until the virtual typhoon termination condition is met.
The virtual typhoon termination conditions include:
when a termination command is extracted from the grid where the virtual typhoon is located, the maximum wind speed near the center of the typhoon is lower than 5.2m/s or the sea temperature at the position where the typhoon is located is lower than 10 ℃, the virtual typhoon is immediately terminated;
because the lowest central air pressure of typhoons has a negative correlation with the maximum near-central air speed, the lowest central air pressure of the virtual typhoons is assigned based on a Copula joint distribution functionThe formula of (2) is as follows:
；
in the method, in the process of the invention,、/>respectively indicate->Direction(s) (i.e. the directions of the eyes)>An edge distribution function of the direction; />Representing a Copula function;
when the simulation year reaches 1000 years, it indicates that all virtual typhoons are simulated.
S2: when the distance between the typhoon center and the boat mountain station (122.42 DEG E,29.89 DEG N) is less than 250km, the typhoon is considered to have an influence on the boat mountain station; screening typhoon data affecting the mountain station from historical typhoons and virtual typhoons according to the identification standard; typhoon data includes longitude, latitude, center lowest air pressure and near center maximum wind speed of typhoons;
inputting typhoon data into a Holland typhoon experience model to generate typhoon wind fields corresponding to typhoons in each field;
the Holland typhoon experience model consists of an air pressure model and a wind field model, wherein the formula of the air pressure model is as follows:
；
in the method, in the process of the invention,Pis the air pressure value at the calculation point,is typhoon center air pressure, < >>Is the peripheral air pressure->Is the maximum wind speed radius>Is to calculate the distance of the point to the typhoon center, < >>Is the Holland barometric pressure profile parameter;
the formula of the wind field model is as follows:
；
in the method, in the process of the invention,is +.>Gradient wind speed at>For coriolis force, ->Is the air density.
S3: inputting a typhoon wind field into a SWAN wave numerical mode, generating a typhoon wave height data set of a Zhoushan station in a typhoon period, and screening out effective wave height values of the Zhoushan station which are larger than 1.25 meters from the typhoon wave height data set;
s31: building a network: the SMS software is used for building the unstructured grids in the east China sea, as shown in fig. 2, the grid range is 117.5 DEG E-131.5 DEG E,23.8 DEG N-41.0 DEG N, the number of the grids is about 7.5 ten thousand, and the grid resolution is the finest at the coastal position of Zhejiang;
s32: calculating wave height: inputting the generated typhoon wind field into a SWAN numerical wave mode, and calculating the effective wave height value of the Zhoushan station during typhoon; the control equation of the SWAN wave numerical mode is expressed in a rectangular coordinate system in the following form:
；
in the method, in the process of the invention,is wave action volume density; />For time (I)>、/>For geographical coordinates>Is relative wave frequency>Is wave direction; />And->For wave propagation speed at +.>And->Component in direction, +_>And->For waves +.>And->Propagation velocity of the space;Sis wave energy density; />Representing the rate of change of the wave action density over time; />、/>Representing the propagation of the wave action density in space, < +.>Indicating the effects of the frequency shift caused by the water depth and the water flow change,representing refraction caused by water depth and water flow variation; />Sources and sinks representing all the generation, dissipation and redistribution of wave energy;
wherein the wave energy densitySExpressed as:
；
in the method, in the process of the invention,representing the input of wind energy->、/>Representing wave energy exchanges resulting from three-and four-order nonlinear interactions, respectively, < >>、/>、/>Representing wave energy dissipation due to white waves, bottom friction and wave break-up, respectively.
S4: integrating the effective wave height value of the screened Zhoushan station with typhoon data and Zhoushan station meteorological data at corresponding moments to jointly construct a sample database; the meteorological data of the mountain station comprise wind speed and air pressure of the mountain station;
s5: dividing data corresponding to the virtual typhoons in the sample database into a training set and a verification set according to the proportion of 7:3, wherein all data corresponding to the historical typhoons are used as test sets;
s6: the typhoon data, the Zhoushan site meteorological data and the Zhoushan site effective wave height values in the time intervals from t-n to t are selected as input sequences, the Zhoushan site effective wave height values at the moments of t+1h, t+3h, t+6h and t+12h are respectively selected as target results, and a nonlinear mapping relation among the typhoon data, the Zhoushan site meteorological data and the Zhoushan site effective wave height values is constructed by utilizing an LSTM neural network model;
the front time sequence n of the selected input sequence is 4h, 6h and 9h respectively when wave heights at the moments t+1h, t+3h, t+6h and t+12h are predicted.
S7: optimizing and selecting super parameters of the LSTM neural network model by adopting a BO algorithm, wherein the super parameters comprise the number of neurons, the learning rate, the batch size and the iteration times;
s71: model initialization: the LSTM neural network model comprises an input layer, a hidden layer and an output layer, wherein Tanh is selected as an activation function, adam is selected as an optimization algorithm, and super parameters are randomly initialized within a preset range;
s72: evaluating sampling points: taking the super-parameter combination as a sampling point to be brought into an LSTM neural network model, calculating the root mean square error RMSE (Root Mean Square Error) of the model output result on the verification set, and evaluating;
s73: gaussian process: calculating probability distribution of a functional relation between the hyper-parameter combination and the RMSE based on a Gaussian process;
s74: acquisition function: combining probability distribution in the Gaussian process, calculating a hyper-parameter combination corresponding to the RMSE minimum value based on the acquisition function, and updating sampling points by using the hyper-parameter combination;
s75: repeating steps S72 to S74 until the RMSE evaluated in step S72 is lower than a preset value or reaches a specified number of cycles;
after the calculation is completed, the selection of the super-parameter combinations is completed, and the specific super-parameter combination selection scheme in this embodiment is shown in table 1:
TABLE 1 LSTM optimal superparameter selection scheme under different forecast time length conditions
S8: training the LSTM neural network model based on the data in the training set and the verification set to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model; evaluating and testing the typhoon wave height forecasting model based on the data in the test set, and forecasting the typhoon wave height through the typhoon wave height forecasting model, specifically comprising the following steps:
s81: initializing model parameters: initializing weight parameters of an LSTM neural network model, including an input gate, a forget gate and an output gate;
the calculation formula of the three gating is as follows:
；
；
；
；
；
；
in the method, in the process of the invention,、/>、/>forgetting coefficient, input coefficient and output coefficient respectively; />、/>、/>Respectively corresponding to the weight matrixes of the gating; />、/>、/>Bias vectors corresponding to the gates respectively; />Input information at the current moment; />A candidate state matrix at the current moment; />、/>The new and old memory state matrixes are respectively; />、/>Is a corresponding neuron matrix; />、/>Respectively outputting information at the current moment and the last moment;
s82: forward propagation: the input sequence is transmitted forwards through an LSTM neural network model, in each time step, the LSTM neural network model receives the input of the current time step and the hidden state of the previous time step, then the output of the current time step and the new hidden state are calculated, and the output of the current time step and the new hidden state are transmitted to the next time step;
s83: calculating loss: comparing the output of the LSTM neural network model with a target result, selecting a Root Mean Square Error (RMSE) as a loss function, and calculating a loss value of the LSTM neural network model;
s84: back propagation: according to the loss value, calculating gradient through a back propagation algorithm, and updating the weight parameter of the LSTM neural network model to reduce the loss value;
s85: training the LSTM neural network model based on the data in the training set and the verification set, and repeatedly executing the steps S82 to S84 until the preset iteration times are reached, so as to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model;
s86: after training, based on the test set, calculating Root Mean Square Error (RMSE) and a Correlation Coefficient (CC) between the BO-LSTM forecast value and the SWAN simulation value, and performing evaluation test on the typhoon wave height forecast model, wherein the calculation result is shown in Table 2:
TABLE 2 RMSE and CC on test set for BO-LSTM model under different forecast time length conditions
The comparison between the BO-LSTM forecast values and SWAN simulation values, counted based on the test set, is shown in FIG. 3.
According to the embodiment, the typhoon wave height forecasting model for the Zhoushan station is built based on the BO-LSTM neural network model, the typhoon wave height forecasting for the Zhoushan station is only shown, and similar forecasting effects can be obtained for other coastal stations by adopting the technical scheme.
In summary, the invention simulates a large number of virtual typhoons and generates the effective wave height value of typhoons corresponding to the research sites, thereby effectively solving the problem that the data volume of training samples in the intelligent typhoons wave height prediction is insufficient, and providing data support for improving model prediction precision and prolonging model prediction duration; compared with the common LSTM, the BO algorithm intelligently selects the super-parameter configuration, so that the model can be converged to the optimal super-parameter combination more quickly, the time consumed in the training process of the neural network model is effectively reduced, and the accuracy and timeliness of the typhoon wave height prediction are further improved.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will appreciate that: the technical scheme described in the foregoing embodiments can be modified or some of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (9)

1. The typhoon wave height forecasting method based on the BO-LSTM neural network model is characterized by comprising the following steps of:
s1: collecting historical typhoon data of a sea area of a research site, and constructing virtual typhoons of the sea area of the research site by adopting an empirical path method based on a nuclear density estimation method based on the historical typhoon data;
s2: typhoon data influencing a research site are screened from historical typhoons and virtual typhoons, and are input into a Holland typhoon experience model to generate typhoon wind fields corresponding to each typhoon; the typhoon data comprise longitude, latitude, central lowest air pressure and near-central maximum wind speed of typhoons;
s3: inputting the typhoon wind field into a SWAN wave numerical mode, generating a typhoon wave height data set of a research site during typhoon, and screening out an effective wave height value of the research site which is larger than 1.25 m from the typhoon wave height data set;
s4: integrating the effective wave height value of the research site with typhoon data and research site meteorological data at corresponding moments to jointly construct a sample database; the research site meteorological data comprise wind speed and air pressure of a research site;
s5: dividing data corresponding to the virtual typhoons in the sample database into a training set and a verification set according to the proportion of 7:3, wherein the data corresponding to the historical typhoons are used as a test set;
s6: selecting typhoon data, research site weather data and research site effective wave height values in the time period from t-n to t as input sequences, respectively selecting research site effective wave height values at the moments of t+1h, t+3h, t+6h and t+12h as target results, and constructing a nonlinear mapping relation among the typhoon data, the research site weather data and the research site effective wave height values by utilizing an LSTM neural network model;
s7: optimizing and selecting the super parameters of the LSTM neural network model by adopting a BO algorithm; the super parameters comprise the number of neurons, the learning rate, the batch size and the iteration times;
s8: training the LSTM neural network model based on the data in the training set and the verification set to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model; and evaluating and testing the typhoon wave height forecasting model based on the data in the test set, and forecasting the typhoon wave height through the typhoon wave height forecasting model.
2. The typhoon wave height forecasting method based on the BO-LSTM neural network model according to claim 1, wherein in the step S1, an empirical path method based on a nuclear density estimation method is adopted to construct a virtual typhoon of a sea area where a research site is located, and the method specifically comprises the following steps:
s11: based on historical typhoon data, statistics and analysis are carried out on tropical cyclones of the sea area where the research site is located, probability distribution of annual occurrence frequency and initial characteristic parameters of the historical typhoons is obtained, and parameter combinations are formed by random sampling from the annual occurrence frequency and the initial characteristic parameters of the historical typhoons, so that the number of the virtual typhoons occurring in each year and the initial characteristic parameters of each virtual typhoons are determined; the initial characteristic parameters comprise typhoon occurrence time, initial moving speed, initial moving direction and initial near-center maximum wind speed;
s12: dividing the sea area range of a research site into grids with the angle of 5 degrees multiplied by 5 degrees, counting the initial and final probabilities of historical typhoons in the grids, determining the initial position of a virtual typhoons based on the initial probability of the historical typhoons in each grid, and fitting the probability distribution of the parameter variation of the historical typhoons in each grid by adopting a nuclear density estimation method, wherein the parameter variation comprises a shift speed variation, a shift direction variation and a near-center maximum wind speed variation;
the formula of the nuclear density estimation method is as follows:
；
in the method, in the process of the invention,is a probability density function;irepresent the firstiA number of samples of the sample were taken,mis the total number of samples;hfor the bandwidth to be available,Kis a kernel function;represents a sample value, and->Obeying independent same distribution; />Representing the sample value +.>Is the average value of (2);
s13: for the virtual typhoons with initial characteristic parameters, calculating the next node position of the virtual typhoons by taking 3h as a time step, randomly extracting a group of parameter variation combinations based on probability distribution of parameter variation of the historical typhoons in grids where the node positions are positioned, and updating key characteristic parameters of the virtual typhoons, wherein the key characteristic parameters of the virtual typhoons comprise typhoons moving speed, moving direction and near-center maximum wind speed; and (5) circulating the above process until the virtual typhoon termination condition is met.
3. The typhoon wave height forecasting method based on the BO-LSTM neural network model of claim 2, wherein the virtual typhoon termination condition comprises:
when a termination command is extracted from the grid where the virtual typhoon is located, the maximum wind speed near the center of the typhoon is lower than 5.2m/s or the sea temperature at the position where the typhoon is located is lower than 10 ℃, the virtual typhoon is immediately terminated;
assigning a value to the central lowest air pressure of the virtual typhoons based on the Copula joint distribution function, and indicating that the simulation of one virtual typhoons is completed;
the Copula joint distribution functionThe formula of (2) is as follows:
；
in the method, in the process of the invention,、/>respectively indicate->Direction(s) (i.e. the directions of the eyes)>An edge distribution function of the direction; />Representing a Copula function;
when the simulation year reaches 1000 years, it indicates that all virtual typhoons are simulated.
4. The typhoon wave height forecasting method based on the BO-LSTM neural network model according to claim 1, wherein in the step S2, the identification criterion influencing the research site is that the distance between the typhoon center and the research site is smaller than 250km.
5. The typhoon wave height forecasting method based on the BO-LSTM neural network model according to claim 1, wherein the Holland typhoon experience model consists of an air pressure model and a wind field model, and the formula of the air pressure model is as follows:
；
in the method, in the process of the invention,Pis the air pressure value at the calculation point,is typhoon center air pressure, < >>Is the peripheral air pressure->Is the maximum wind speed radius>Is to calculate the distance of the point to the typhoon center, < >>Is the Holland barometric pressure profile parameter;
the formula of the wind field model is as follows:
；
in the method, in the process of the invention,is +.>Gradient wind speed at>For coriolis force, ->Is the air density.
6. The typhoon wave height forecasting method based on the BO-LSTM neural network model according to claim 1, wherein in the step S3, a control equation of the SWAN wave numerical mode is expressed in a rectangular coordinate system in the following form:
；
in the method, in the process of the invention,is wave action volume density; />For time (I)>、/>For geographical coordinates>Is relative wave frequency>Is wave direction;and->For wave propagation speed at +.>And->Component in direction, +_>And->For waves +.>And->Propagation velocity of the space;Sis wave energy density;
wherein the wave energy densitySExpressed as:
；
in the method, in the process of the invention,representing the input of wind energy->、/>Representing wave energy exchanges resulting from three-and four-order nonlinear interactions, respectively, < >>、/>、/>Respectively representing wave energy caused by white waves, bottom friction and wave breakageDissipation is achieved.
7. The typhoon wave height forecasting method based on the BO-LSTM neural network model according to claim 1, wherein in the step S6, the pre-timing n of the input sequence is selected to be 4h, 6h and 9h when forecasting wave heights at the moments t+1h, t+3h, t+6h and t+12h respectively.
8. The typhoon wave height forecasting method based on the BO-LSTM neural network model according to claim 1, wherein in the step S7, the optimization selection of the super parameters of the LSTM neural network model by the BO algorithm is performed, and the method specifically comprises the following steps:
s71: model initialization: the LSTM neural network model comprises an input layer, a hidden layer and an output layer, wherein Tanh is selected as an activation function, adam is selected as an optimization algorithm, and super parameters are randomly initialized within a preset range;
s72: evaluating sampling points: taking the super-parameter combination as a sampling point to be brought into an LSTM neural network model, calculating the Root Mean Square Error (RMSE) of the output result of the model on a verification set, and evaluating;
s73: gaussian process: calculating probability distribution of a functional relation between the hyper-parameter combination and the RMSE based on a Gaussian process;
s74: acquisition function: combining the probability distribution in the Gaussian process, calculating a hyper-parameter combination corresponding to the RMSE minimum value based on an acquisition function, and updating a sampling point by using the hyper-parameter combination;
s75: steps S72 to S74 are repeated until the RMSE evaluated in step S72 is lower than a preset value or a specified number of cycles is reached.
9. The typhoon wave height forecasting method based on the BO-LSTM neural network model according to claim 1, wherein in the step S8, a typhoon wave height forecasting model based on the BO-LSTM neural network model is obtained, and the specific method comprises the following steps:
s81: initializing weight parameters of an LSTM neural network model, including an input gate, a forget gate and an output gate;
s82: the input sequence is transmitted forwards through an LSTM neural network model, in each time step, the LSTM neural network model receives the input of the current time step and the hidden state of the previous time step, then the output of the current time step and the new hidden state are calculated, and the output of the current time step and the new hidden state are transmitted to the next time step;
s83: comparing the output of the LSTM neural network model with a target result, selecting a Root Mean Square Error (RMSE) as a loss function, and calculating a loss value of the LSTM neural network model;
s84: according to the loss value, calculating gradient through a back propagation algorithm, and updating the weight parameter of the LSTM neural network model to reduce the loss value;
s85: training the LSTM neural network model based on the data in the training set and the verification set, and repeatedly executing the steps S82 to S84 until the preset iteration times are reached, so as to obtain a typhoon wave height forecasting model based on the BO-LSTM neural network model;
s86: and after training, calculating Root Mean Square Error (RMSE) and a Correlation Coefficient (CC) between the BO-LSTM forecast value and the SWAN simulation value based on a test set, and performing evaluation test on the typhoon wave height forecast model.
CN202311834527.XA 2023-12-28 2023-12-28 Typhoon wave height forecasting method based on BO-LSTM neural network model Active CN117471575B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311834527.XA CN117471575B (en) 2023-12-28 2023-12-28 Typhoon wave height forecasting method based on BO-LSTM neural network model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311834527.XA CN117471575B (en) 2023-12-28 2023-12-28 Typhoon wave height forecasting method based on BO-LSTM neural network model

Publications (2)

Publication Number Publication Date
CN117471575A true CN117471575A (en) 2024-01-30
CN117471575B CN117471575B (en) 2024-03-08

Family

ID=89624247

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311834527.XA Active CN117471575B (en) 2023-12-28 2023-12-28 Typhoon wave height forecasting method based on BO-LSTM neural network model

Country Status (1)

Country Link
CN (1) CN117471575B (en)

Citations (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2010049433A (en) * 2008-08-21 2010-03-04 Chugoku Electric Power Co Inc:The System for supporting prediction of damage caused by typhoon
CN109862585A (en) * 2019-01-31 2019-06-07 湖北工业大学 A kind of dynamic heterogeneous network flow prediction method based on depth Space-time Neural Network
CN110113119A (en) * 2019-04-26 2019-08-09 国家无线电监测中心 A kind of Wireless Channel Modeling method based on intelligent algorithm
WO2019189501A1 (en) * 2018-03-27 2019-10-03 旭化成株式会社 Design device, method, program, planning device, control device, and hydrogen production system
CN111027752A (en) * 2019-11-19 2020-04-17 浙江大学 Crop yield estimation method based on deep spatiotemporal feature joint learning
CN111199270A (en) * 2019-12-30 2020-05-26 福建省海洋预报台 Regional wave height forecasting method and terminal based on deep learning
US20200348448A1 (en) * 2019-04-30 2020-11-05 ClimateAI, Inc. Methods and systems for climate forecasting using artificial neural networks
US20200356839A1 (en) * 2019-05-09 2020-11-12 ClimateAI, Inc. Systems and methods for selecting global climate simulation models for training neural network climate forecasting models
CN112785034A (en) * 2020-12-22 2021-05-11 河海大学 Typhoon path forecasting method, system, medium and terminal based on fusion neural network
CN113887787A (en) * 2021-09-15 2022-01-04 大连理工大学 Flood forecasting model parameter multi-objective optimization method based on long-time and short-time memory network and NSGA-II algorithm
CN114445634A (en) * 2022-02-28 2022-05-06 南京信息工程大学 Sea wave height prediction method and system based on deep learning model
US20220374681A1 (en) * 2022-07-08 2022-11-24 Zhihan Lv Method for predicting wave energy based on improved GRU
CN115759371A (en) * 2022-11-08 2023-03-07 国网吉林省电力有限公司电力科学研究院 GCN-LSTM-based short-term load prediction method for power system
US20230214703A1 (en) * 2021-12-30 2023-07-06 SparkCognition, Inc. Predicting energy production for energy generating assets
CN116611270A (en) * 2023-07-20 2023-08-18 国家***北海预报中心((国家***青岛海洋预报台)(国家***青岛海洋环境监测中心站)) Typhoon wave real-time aggregate forecasting method, medium and system
US11733427B1 (en) * 2017-04-11 2023-08-22 DataInfoCom USA, Inc. Short-term weather forecasting using hybrid data
CN116736409A (en) * 2023-05-29 2023-09-12 武汉轻工大学 Automobile safety early warning method, device, equipment and storage medium
CN116976227A (en) * 2023-09-22 2023-10-31 河海大学 Storm water increasing forecasting method and system based on LSTM machine learning
CN117198330A (en) * 2023-11-07 2023-12-08 国家海洋技术中心 Sound source identification method and system and electronic equipment
CN117371303A (en) * 2023-09-13 2024-01-09 河海大学 Prediction method for effective wave height under sea wave

Patent Citations (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2010049433A (en) * 2008-08-21 2010-03-04 Chugoku Electric Power Co Inc:The System for supporting prediction of damage caused by typhoon
US11733427B1 (en) * 2017-04-11 2023-08-22 DataInfoCom USA, Inc. Short-term weather forecasting using hybrid data
WO2019189501A1 (en) * 2018-03-27 2019-10-03 旭化成株式会社 Design device, method, program, planning device, control device, and hydrogen production system
CN109862585A (en) * 2019-01-31 2019-06-07 湖北工业大学 A kind of dynamic heterogeneous network flow prediction method based on depth Space-time Neural Network
CN110113119A (en) * 2019-04-26 2019-08-09 国家无线电监测中心 A kind of Wireless Channel Modeling method based on intelligent algorithm
US20200348448A1 (en) * 2019-04-30 2020-11-05 ClimateAI, Inc. Methods and systems for climate forecasting using artificial neural networks
US20200356839A1 (en) * 2019-05-09 2020-11-12 ClimateAI, Inc. Systems and methods for selecting global climate simulation models for training neural network climate forecasting models
CN111027752A (en) * 2019-11-19 2020-04-17 浙江大学 Crop yield estimation method based on deep spatiotemporal feature joint learning
US20220405864A1 (en) * 2019-11-19 2022-12-22 Zhejiang University Crop yield estimation method based on deep temporal and spatial feature combined learning
CN111199270A (en) * 2019-12-30 2020-05-26 福建省海洋预报台 Regional wave height forecasting method and terminal based on deep learning
CN112785034A (en) * 2020-12-22 2021-05-11 河海大学 Typhoon path forecasting method, system, medium and terminal based on fusion neural network
CN113887787A (en) * 2021-09-15 2022-01-04 大连理工大学 Flood forecasting model parameter multi-objective optimization method based on long-time and short-time memory network and NSGA-II algorithm
US20230214703A1 (en) * 2021-12-30 2023-07-06 SparkCognition, Inc. Predicting energy production for energy generating assets
CN114445634A (en) * 2022-02-28 2022-05-06 南京信息工程大学 Sea wave height prediction method and system based on deep learning model
US20220374681A1 (en) * 2022-07-08 2022-11-24 Zhihan Lv Method for predicting wave energy based on improved GRU
CN115759371A (en) * 2022-11-08 2023-03-07 国网吉林省电力有限公司电力科学研究院 GCN-LSTM-based short-term load prediction method for power system
CN116736409A (en) * 2023-05-29 2023-09-12 武汉轻工大学 Automobile safety early warning method, device, equipment and storage medium
CN116611270A (en) * 2023-07-20 2023-08-18 国家***北海预报中心((国家***青岛海洋预报台)(国家***青岛海洋环境监测中心站)) Typhoon wave real-time aggregate forecasting method, medium and system
CN117371303A (en) * 2023-09-13 2024-01-09 河海大学 Prediction method for effective wave height under sea wave
CN116976227A (en) * 2023-09-22 2023-10-31 河海大学 Storm water increasing forecasting method and system based on LSTM machine learning
CN117198330A (en) * 2023-11-07 2023-12-08 国家海洋技术中心 Sound source identification method and system and electronic equipment

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
MARTINA M P P 等: "Forecasting Significant Wave Height using RNN-LSTM Models", 2020 4TH INTERNATIONAL CONFERENCE ON INTELLIGENT COMPUTING AND CONTROL SYSTEMS (ICICCS), 31 December 2020 (2020-12-31) *
徐晓武 等: "基于非稳态调和分析和长短时记忆神经网络的河口潮位短期预报混合模型", 海洋通报, vol. 41, no. 4, 31 December 2022 (2022-12-31) *
徐高扬 等: "LSTM网络在台风路径预测中的应用", 计算机与现代化, no. 5, 31 May 2019 (2019-05-31) *
赵勇;苏丹;邹丽;王爱民;: "基于LSTM神经网络的畸形波预测", 华中科技大学学报(自然科学版), no. 07, 14 July 2020 (2020-07-14) *

Also Published As

Publication number Publication date
CN117471575B (en) 2024-03-08

Similar Documents

Publication Publication Date Title
CN111199270B (en) Regional wave height forecasting method and terminal based on deep learning
CN111915058B (en) Flood prediction method and device based on long-time memory network and transfer learning
Saxena et al. A review study of weather forecasting using artificial neural network approach
CN107480781A (en) The nuclear accident Source Term Inversion method of neutral net adaptive Kalman filter
Yao et al. Wave height forecast method with multi-step training set extension LSTM neural network
CN115877483A (en) Typhoon path forecasting method based on random forest and GRU
CN115759445A (en) Machine learning and cloud model-based classified flood random forecasting method
CN116976227B (en) Storm water increasing forecasting method and system based on LSTM machine learning
CN111914487B (en) Data-free regional hydrological parameter calibration method based on antagonistic neural network
CN104834966A (en) Fishery forecasting method based on ant colony classifying algorithm
CN111914488B (en) Data area hydrologic parameter calibration method based on antagonistic neural network
CN114676621A (en) Method for improving accuracy of land water reserve abnormity based on deep learning weight load
Kisi et al. Forecasting sea water levels at Mukho Station, South Korea using soft computing techniques
Wang et al. Superiority of a convolutional neural network Model over Dynamical models in Predicting Central Pacific ENSO
CN117471575B (en) Typhoon wave height forecasting method based on BO-LSTM neural network model
CN109960882A (en) Forecast that tropical cyclone generates method, apparatus, equipment and the storage medium of frequency
Wang et al. Xihe: A data-driven model for global ocean eddy-resolving forecasting
Jiang et al. Discharge estimation based on machine learning
Shi et al. A Grey Model for Short-Term Prediction of the Ionospheric TEC
Khanarmuei et al. Calibration and assimilation in hydrodynamic model of a micro-tidal estuary and comparison with Lagrangian drifter data
LU504816B1 (en) Interpretable deep learning model for extreme climate prediction
Shashikanth et al. Fine resolution Indian summer monsoon rainfall projection with statistical downscaling
CN116992781B (en) Multi-step multi-element storm forecasting method based on deep learning
SALIKA Assessing the beach vulnerability to storm-induced erosion using an index-based approach
Zhao et al. A Periodically Updated Adaptive Sampling Framework for Marine Mobile Observation Platforms

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