Deep learning fusion-based precipitation grading dynamic prediction model design method and device

Abstract

The invention discloses a precipitation grading dynamic prediction model design method and device based on deep learning fusion. The method comprises the steps that S1, precipitation is graded according to the magnitude; s2, respectively identifying climate driving factors for different grades of rainfall; s3, establishing a time-varying cumulative distribution function of different levels of rainfall and climate driving factor sequences; and S4, based on the time-varying cumulative distribution function, establishing a deep learning fusion model, and predicting future rainfall. According to the method, the credibility of climate system prediction factors can be improved, different levels of rainfall climate driving factors and response rules thereof are disclosed, and the rainfall prediction capability and accuracy are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydrological detection, and specifically relates to a method and device for designing a precipitation classification dynamic prediction model based on deep learning fusion. Background Art

[0002] Water is the source of life. Human daily life and industrial and agricultural production cannot do without water. Precipitation is the main source of water resources. Appropriate precipitation can bring convenience and benefits to human production and life, but excessive precipitation may cause floods and secondary disasters such as landslides; too little precipitation may cause regional drought problems, causing great inconvenience to human production and life, and even affecting human survival. Accurate and rapid precipitation forecasts can not only provide scientific references for human production and life, but also serve as warnings in the fields of water conservancy, agriculture, industrial production, etc., to minimize the losses caused by disasters. In order to measure the regional precipitation conditions, rainfall is generally used to express it. Changes in the climate of a region can be more intuitively reflected in the changes in rainfall in the region.

[0003] In recent years, with the intensification of climate changes, the fluctuation of rainfall has become larger, and extreme rainfall events have begun to occur frequently in various urban areas. The regional flood disasters caused by this have also become increasingly prominent. Therefore, the accurate prediction of urban rainfall in the future is of great research significance. The concept of time series analysis method was first proposed by the British GU.Yule in 1927. GU.Yule proposed the concept of autoregressive AR model in the stationary time series model. Later, the British GT.Walker used the moving average (MA) model and the autoregressive moving average (ARMA) model in practical applications. With the repeated research and application of AR model and MA model by scholars, ARMA model and ARIMA model have been slowly developed, and they are all core models in time series. Now in 2006, Su Li used the relevant knowledge of the yule-walker autoregressive equation group in the AR (p) model in the article "Application of Time Series Analysis and Forecasting Method in Precipitation Forecasting". Establish the yule-walker autoregressive equations for precipitation in a certain period of time, find the autoregressive coefficients and bring them into the original equations to establish the corresponding autoregressive forecast model for the area. In 2009, Xu Wenxia et al. considered the great influence of seasonal factors on precipitation in "Dynamic Prediction of Precipitation Based on Time Series Analysis" and chose to analyze and predict the precipitation data of the same month in the past few years. Jin Xuanxuan et al. established different models for the same group of samples in the article "Research on Time Series Process of China's Real Estate Market", and explained that when exploring specific samples, specific model forms should be established for them according to the characteristics of each series. Only by matching the model with the right one can the prediction ability of the model be brought into play to the optimal level.

[0004] Although the above traditional time series analysis methods have certain ability to cope with seasonal factors and climate change, the accuracy of time series prediction is still not high. Therefore, it is urgent to propose relevant solutions to the problem of low accuracy of time series prediction by traditional time series analysis methods. Summary of the invention

[0005] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and to provide a method and device for designing a precipitation classification dynamic prediction model based on deep learning fusion. By constructing a partial-rank Kendall prediction factor selection method based on precipitation clustering level division and a variable-weight deep learning precipitation fusion model, the credibility of climate system prediction factors can be improved, and the climate driving factors of precipitation at different levels and their response laws can be revealed, thereby greatly improving the precipitation prediction ability and accuracy.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for designing a precipitation classification dynamic prediction model based on deep learning fusion, comprising the following steps:

[0008] S1. Classify precipitation according to its magnitude;

[0009] S2, identify climate driving factors for different levels of precipitation;

[0010] S3. Establish time-varying cumulative distribution functions of precipitation and climate driving factor sequences at different levels;

[0011] S4. Based on the time-varying cumulative distribution function, a deep learning fusion model is established to predict future precipitation.

[0012] As a preferred technical solution, step S1 includes:

[0013] The K-means++ algorithm is used to select the Euclidean distance, and cluster analysis is performed to divide the precipitation frequency levels; the Euclidean distance is selected to measure the distance of rainfall Y in different periods:

[0014]

[0015] Based on the Euclidean distance method, clustering is performed with k points in space as the center, and the closest objects are classified. Through an iterative method, the value of each cluster center is updated one by one. The objective function achieved is:

[0016]

[0017] Where c1, c2, ..., c k are the center points of k classes respectively; C(Y i ) indicates Yi is the center point of the class to which it belongs; D 2 Y i With C(Y i )The square of the distance between two points.

[0018] As a preferred technical solution, step S2 includes:

[0019] The Kendall rank correlation coefficient is used to quantify the nonlinear correlation between climate factor X and rainfall Y, and the calculation formula is:

[0020]

[0021] Where n is the number of observation pairs, P n and Q n Represents the number of consistent pairs of X and Y and the number of inconsistent pairs of X and Y respectively; τ 1,2 The distribution of is close to 0 and has a variance of If the normal distribution of , the condition for significant correlation between climate factor X and tree-ring chronology Y at the set confidence level is:

[0022]

[0023] Whether the climate factor is the main driving factor of rainfall is determined based on whether the significant correlation conditions are met.

[0024] As a preferred technical solution, the confidence level α=5%.

[0025] As a preferred technical solution, step S3 includes:

[0026] The three-parameter gamma distribution is used to fit the rainfall, denoted as Y~Γ(α,β,ε), and its probability density function is:

[0027]

[0028] Where Γ(·) is the gamma function, α>0 is the shape parameter, β>0 is the scale parameter, and ε is the location parameter. Based on the rainfall, a generalized additive model is established to simulate the time-varying frequency distribution. Assuming that the sequence has n observations and its time series is t (t=1,2,...,n), then the non-uniformly varying rainfall y t The distribution function of t ,β t ,ε t ), define θ=(α T ,β T ,ε T ) represents the time-varying distribution function parameter matrix, let k = 1, 2, 3, and the values ​​represent the three parameters of shape, scale, and position respectively, then the time vector of the kth distribution function parameter is θ k=(θ 1k ,θ 2k ,...,θ nk ) T ; Remember g k (·) represents θ k and the corresponding explanatory variable X k The monotonic functional relationship between and the random effect term is expressed as follows:

[0029]

[0030] In the formula is the explanatory variable matrix; For I k The length of the regression parameter vector; h jk (·) represents the distribution parameter for the influencing variable e jk relationship function.

[0031] As a preferred technical solution, step S4 includes:

[0032] Establish a LSTM neural network based on LSTM And the convolutional neural network f CNN The variable weight fusion prediction model F t , the formula is:

[0033] F t =θ t ·f LSTM +(1-θ t )·f CNN (7)

[0034] where f LSTM for:

[0035] Forget gate: f t =σ(W f ·[h t-1 ,x t ]+b f ) (8)

[0036] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i ) (9)

[0037]

[0038] Where h t-1 is the value of the hidden layer at the previous moment, x t is the current input at time t, σ is the sigmoid function, W f and b fare the forgotten weights and bias terms, respectively, and f t is the output of the forget gate; W i and W c is the input gate weight, b i and b c is the bias term of the input gate, tanh() is the activation function, i t is the input gate output, is a candidate status;

[0039] f CNN for:

[0040]

[0041] Where b is the deviation, Z l and Z l+1 represents the convolution input and output of the l+1th layer, k is the number of channels, f, s0 and p are the convolution layer parameters, corresponding to the convolution kernel size, convolution step size and number of padding layers.

[0042] In a second aspect, the present invention provides a precipitation classification dynamic prediction model design system based on deep learning fusion, which is applied to the precipitation classification dynamic prediction model design method based on deep learning fusion, including a classification module, a climate driving factor identification module, a time-varying cumulative distribution function establishment module and a precipitation model establishment module;

[0043] The classification module is used to classify precipitation according to its magnitude;

[0044] The climate driving factor identification module is used to identify climate driving factors for different levels of precipitation;

[0045] The time-varying cumulative distribution function establishment module is used to establish the time-varying cumulative distribution functions of precipitation and climate driving factor sequences of different levels;

[0046] The precipitation model building module is used to build a deep learning fusion model based on a time-varying cumulative distribution function to predict future precipitation.

[0047] In a third aspect, the present invention provides an electronic device, the electronic device comprising:

[0048] at least one processor; and,

[0049] a memory communicatively connected to the at least one processor; wherein,

[0050] The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the precipitation classification dynamic prediction model design method based on deep learning fusion.

[0051] In a fourth aspect, the present invention provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the method for designing a precipitation classification dynamic prediction model based on deep learning fusion.

[0052] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0053] (1) Based on the precipitation clustering level classification, the present invention analyzes the mechanistic relationship between precipitation of different levels in the basin and various climate system indicators, and specifically compares and analyzes the significance of different climate system indices on precipitation levels at all levels in the region based on the partial rank Kendall correlation coefficient, which can fully improve the credibility of climate system prediction factors and reveal the climate driving factors of precipitation of different levels and their response laws.

[0054] (2) Based on the results of the joint distribution model of climate system prediction factors and regional precipitation, the present invention tests the results of different time series analysis models, adjusts the optimal weights and parameter combinations according to the accuracy of each model, and then integrates multiple time series analysis algorithms to reduce costs and failure rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0056] Figure 1 A flowchart of a method for designing a precipitation classification dynamic prediction model based on deep learning fusion according to an embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of the structure of a deep learning fusion model according to an embodiment of the present invention;

[0058] Figure 3 A schematic diagram of the structure of a precipitation classification dynamic prediction model design system based on deep learning fusion according to an embodiment of the present invention;

[0059] Figure 4 Schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0061] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0062] See also Figure 1 In this embodiment, a method for designing a precipitation classification dynamic prediction model based on deep learning fusion is provided, comprising the following steps:

[0063] S1. Classify precipitation according to its magnitude;

[0064] Furthermore, the step S1 is specifically as follows:

[0065] The K-means++ algorithm is used to select the Euclidean distance, and cluster analysis is performed to divide the precipitation frequency levels; the Euclidean distance is selected to measure the distance of rainfall Y in different periods:

[0066]

[0067] Based on the Euclidean distance method, clustering is performed with k points in space as the center, and the closest objects are classified. Through an iterative method, the value of each cluster center is updated successively until the best clustering is achieved. The objective function achieved is:

[0068]

[0069] Where c1, c2, ..., c k are the center points of k classes respectively; C(Y i ) indicates Y i is the center point of the class to which it belongs; D 2 Y i With C(Y i )The square of the distance between two points.

[0070] S2, identify climate driving factors for different levels of precipitation;

[0071] Furthermore, the step S2 is specifically as follows:

[0072] The Kendall rank correlation coefficient is used to quantify the nonlinear correlation between climate factor X and rainfall Y. Its formula is:

[0073]

[0074] Where n is the number of observation pairs, P n and Q n Represents the number of consistent pairs of X and Y and the number of inconsistent pairs of X and Y respectively. 1,2 The distribution of is close to 0 and has a variance of If the normal distribution of , then the condition for significant correlation between climate factor X and tree-ring chronology Y at the confidence level α = 5% is:

[0075]

[0076] Whether the climate factor is the main driving factor of rainfall is determined based on whether the significant correlation conditions are met.

[0077] S3. Establish time-varying cumulative distribution functions of precipitation and climate driving factor sequences at different levels;

[0078] Furthermore, the step S3 is specifically as follows:

[0079] The three-parameter gamma distribution is used to fit the rainfall, denoted as Y~Γ(α,β,ε), and its probability density function is:

[0080]

[0081] Where Γ(·) is the gamma function, α>0 is the shape parameter, β>0 is the scale parameter, and ε is the location parameter. Taking rainfall as an example, a generalized additive model is established to simulate the time-varying frequency distribution. Assuming that the sequence has n observations and its time series is t (t=1,2,...,n), then the non-uniformly varying rainfall y t The distribution function of t ,β t ,ε t ), define θ=(α T ,β T ,ε T ) represents the time-varying distribution function parameter matrix, let k = 1, 2, 3, and the values ​​represent the three parameters of shape, scale, and position respectively, then the time vector of the kth distribution function parameter is θ k =(θ 1k ,θ 2k ,...,θ nk ) T . Remember g k (·) represents θ k and the corresponding explanatory variable Xk The monotonic functional relationship between and the random effect term is expressed as follows:

[0082]

[0083] In the formula is the explanatory variable matrix; For I k The length of the regression parameter vector; h jk (·) represents the distribution parameter for the influencing variable e jk relationship function.

[0084] S4. Based on the time-varying cumulative distribution function, a deep learning fusion model is established to predict future precipitation;

[0085] Further, such as Figure 2 As shown, the step S4 is specifically as follows:

[0086] Establish a LSTM neural network based on LSTM And the convolutional neural network f CNN The variable weight fusion prediction model F t , the formula is:

[0087] F t =θ t ·f LSTM +(1-θ t )·f CNN (7)

[0088] where f LSTM for:

[0089] Forget gate: f t =σ(W f ·[h t-1 ,x t ]+b f ) (8)

[0090] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i ) (9)

[0091]

[0092] Where h t-1 is the value of the hidden layer at the previous moment, x t is the current input at time t, σ is the sigmoid function, W f and b f are the forgotten weights and bias terms, respectively, and f t is the output of the forget gate; Wi and W c is the input gate weight, b i and b c is the bias term of the input gate, tanh() is the activation function, i t is the input gate output, Candidate status.

[0093] f CNN for:

[0094]

[0095] Where b is the deviation, Z l and Z l+1 represents the convolution input and output of the l+1th layer, k is the number of channels, f, s0 and p are the convolution layer parameters, corresponding to the convolution kernel size, convolution step size and number of padding layers.

[0096] It should be noted that, for the sake of convenience, the aforementioned method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited to the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously.

[0097] Based on the same idea as the precipitation classification dynamic prediction model design method based on deep learning fusion in the above-mentioned embodiment, the present invention also provides a precipitation classification dynamic prediction model design system based on deep learning fusion, which can be used to execute the above-mentioned precipitation classification dynamic prediction model design method based on deep learning fusion. For ease of explanation, in the structural schematic diagram of the embodiment of the precipitation classification dynamic prediction model design system based on deep learning fusion, only the parts related to the embodiment of the present invention are shown. Those skilled in the art can understand that the illustrated structure does not constitute a limitation on the system, and may include more or fewer components than the illustrated structure, or combine certain components, or arrange the components differently.

[0098] See also Figure 3 In another embodiment of the present application, a precipitation classification dynamic prediction model design system 100 based on deep learning fusion is provided, the system comprising a classification module 101, a climate driving factor identification module 102, a time-varying cumulative distribution function establishment module 103 and a precipitation model establishment module 104;

[0099] The classification module 101 is used to classify precipitation according to its magnitude;

[0100] The climate driving factor identification module 102 is used to identify climate driving factors for different levels of precipitation;

[0101] The time-varying cumulative distribution function establishment module 103 is used to establish the time-varying cumulative distribution functions of precipitation and climate driving factor sequences of different levels;

[0102] The precipitation model building module 104 is used to build a deep learning fusion model based on a time-varying cumulative distribution function to predict future precipitation.

[0103] It should be noted that the precipitation classification dynamic prediction model design system based on deep learning fusion of the present invention corresponds one to one with the precipitation classification dynamic prediction model design method based on deep learning fusion of the present invention. The technical features and beneficial effects described in the above-mentioned embodiment of the precipitation classification dynamic prediction model design method based on deep learning fusion are applicable to the embodiment of the precipitation classification dynamic prediction model design method based on deep learning fusion. For specific contents, please refer to the description in the embodiment of the method of the present invention, which will not be repeated here. This is hereby declared.

[0104] See also Figure 4 In one embodiment, an electronic device for a precipitation classification dynamic prediction model design method based on deep learning fusion is provided. The electronic device 200 may include a first processor 201, a first memory 202 and a bus, and may also include a computer program stored in the first memory 202 and executable on the first processor 201, such as a precipitation classification dynamic prediction model design program 203 based on deep learning fusion.

[0105] Among them, the first memory 202 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the first memory 202 can be an internal storage unit of the electronic device 200, such as a mobile hard disk of the electronic device 200. In other embodiments, the first memory 202 can also be an external storage device of the electronic device 200, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 200. Further, the first memory 202 can also include both an internal storage unit of the electronic device 200 and an external storage device. The first memory 202 can not only be used to store application software and various types of data installed in the electronic device 200, such as the code of the precipitation classification dynamic prediction model design program 203 based on deep learning fusion, but also can be used to temporarily store data that has been output or is to be output.

[0106] In some embodiments, the first processor 201 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The first processor 201 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes various functions and processes data of the electronic device 200 by running or executing programs or modules stored in the first memory 202, and calling data stored in the first memory 202.

[0107] Figure 4 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 4 The structure shown does not constitute a limitation on the electronic device 200 , and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0108] The precipitation classification dynamic prediction model design program 203 based on deep learning fusion stored in the first memory 202 in the electronic device 200 is a combination of multiple instructions. When running in the first processor 201, it can achieve:

[0109] S1. Classify precipitation according to its magnitude;

[0110] S2, identify climate driving factors for different levels of precipitation;

[0111] S3. Establish time-varying cumulative distribution functions of precipitation and climate driving factor sequences at different levels;

[0112] S4. Based on the time-varying cumulative distribution function, a deep learning fusion model is established to predict future precipitation.

[0113] Furthermore, if the module / unit integrated in the electronic device 200 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0114] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0115] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A precipitation classification dynamic prediction model design method based on deep learning fusion, characterized in that: The steps include: S1. Classify precipitation according to its magnitude; S2, identify climate driving factors for different levels of precipitation; S3. Establish time-varying cumulative distribution functions of precipitation and climate driving factor sequences at different levels; S4. Based on the time-varying cumulative distribution function, a deep learning fusion model is established to predict future precipitation.

2. According to the method for designing a precipitation classification dynamic prediction model based on deep learning fusion according to claim 1, it is characterized in that: The step S1 comprises: The K-means++ algorithm is used to select the Euclidean distance, and cluster analysis is performed to divide the precipitation frequency levels; the Euclidean distance is selected to measure the distance of rainfall Y in different periods: Based on the Euclidean distance method, clustering is performed with k points in space as the center, and the closest objects are classified. Through an iterative method, the value of each cluster center is updated one by one. The objective function achieved is: Where c1, c2, ..., c k are the center points of k classes respectively; C(Y i ) indicates Y i is the center point of the class to which it belongs; D 2 Y i With C(Y i )The square of the distance between two points.

3. According to the method for designing a precipitation classification dynamic prediction model based on deep learning fusion according to claim 1, it is characterized in that: The step S2 comprises: The Kendall rank correlation coefficient is used to quantify the nonlinear correlation between climate factor X and rainfall Y, and the calculation formula is: Where n is the number of observation pairs, P n and Q n Represents the number of consistent pairs of X and Y and the number of inconsistent pairs of X and Y respectively; τ 1,2 The distribution of is close to 0 and has a variance of If the normal distribution of , the condition for significant correlation between climate factor X and tree-ring chronology Y at the set confidence level is: Whether the climate factor is the main driving factor of rainfall is determined based on whether the significant correlation conditions are met.

4. According to claim 3, a precipitation classification dynamic prediction model design method based on deep learning fusion is characterized in that: The confidence level α=5%.

5. According to the method for designing a precipitation classification dynamic prediction model based on deep learning fusion according to claim 1, it is characterized in that: The step S3 comprises: The three-parameter gamma distribution is used to fit the rainfall, denoted as Y~Γ(α,β,ε), and its probability density function is: Where Γ(·) is the gamma function, α>0 is the shape parameter, β>0 is the scale parameter, and ε is the location parameter. Based on the rainfall, a generalized additive model is established to simulate the time-varying frequency distribution. Assuming that the sequence has n observations and its time series is t (t=1,2,...,n), then the non-uniformly varying rainfall y t The distribution function of t ,β t ,ε t ), define θ=(α T ,β T ,ε T ) represents the time-varying distribution function parameter matrix, let k = 1, 2, 3, and the values ​​represent the three parameters of shape, scale, and position respectively, then the time vector of the kth distribution function parameter is θ k =(θ 1k ,θ 2k ,...,θ nk ) T ; Remember g k (·) represents θ k and the corresponding explanatory variable X k The monotonic functional relationship between and the random effect term is expressed as follows: In the formula is the explanatory variable matrix; For I k The length of the regression parameter vector; h jk (·) represents the distribution parameter for the influencing variable e jk relationship function.

6. According to the method for designing a precipitation classification dynamic prediction model based on deep learning fusion according to claim 1, it is characterized in that: The step S4 comprises: Establish a LSTM neural network based on LSTM And the convolutional neural network f CNN The variable weight fusion prediction model F t , the formula is: F t =θ t ·f LSTM +(1-θ t )·f CNN (7) where f LSTM for: Forget gate: f t =σ(W f ·[h t-1 ,x t ]+b f ) (8) Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i ) (9) Where h t-1 is the value of the hidden layer at the previous moment, x t is the current input at time t, σ is the sigmoid function, W f and b f are the forgotten weights and bias terms, respectively, and f t is the output of the forget gate; W i and W c is the input gate weight, b i and b c is the bias term of the input gate, tanh() is the activation function, i t is the input gate output, Candidate status; f CNN for: Where b is the deviation, Z l and Z l+1 represents the convolution input and output of the l+1th layer, k is the number of channels, f, s0 and p are the convolution layer parameters, corresponding to the convolution kernel size, convolution step size and number of padding layers.

7. A precipitation classification dynamic prediction model design system based on deep learning fusion, characterized by: A precipitation classification dynamic prediction model design method based on deep learning fusion applied to any one of claims 1-6, comprising a classification module, a climate driving factor identification module, a time-varying cumulative distribution function establishment module and a precipitation model establishment module; The classification module is used to classify precipitation according to its magnitude; The climate driving factor identification module is used to identify climate driving factors for different levels of precipitation; The time-varying cumulative distribution function establishment module is used to establish the time-varying cumulative distribution functions of precipitation and climate driving factor sequences of different levels; The precipitation model building module is used to build a deep learning fusion model based on a time-varying cumulative distribution function to predict future precipitation.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the precipitation classification dynamic prediction model design method based on deep learning fusion as described in any one of claims 1-6.

9. A computer-readable storage medium storing a program, characterized in that: When the program is executed by a processor, the precipitation classification dynamic prediction model design method based on deep learning fusion described in any one of claims 1 to 6 is implemented.

Images