A short-term heavy rainfall early warning method based on deep learning

Abstract

The present invention discloses a short-time heavy rainfall early warning method based on deep learning, comprising: obtaining combined reflectivity factor data of a weather radar network covering a target area and its surrounding areas, collecting rainfall data of the target area, and preprocessing the combined reflectivity factor data and the rainfall data; obtaining an SDHR label according to the rainfall data, extracting, selecting and classifying the combined reflectivity factor data to obtain impact data; performing an impact analysis on the rainfall data to obtain a risk coefficient, and constructing a short-time heavy rainfall early warning model using a ConvLSTM algorithm according to the impact data, the risk coefficient and the SDHR label; inputting the observation data into the short-time heavy rainfall early warning model, and outputting an early warning result. This method can not only improve the accuracy of short-time heavy rainfall early warning, but also has good interpretability, and can be directly applied to a short-time heavy rainfall early warning system.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological disaster early warning, and in particular to a short-term heavy rainfall early warning method based on deep learning. Background Art

[0002] Short Duration Heavy Rainfall (SDHR), as a meteorological phenomenon with strong suddenness, high rainfall intensity and short duration, poses a serious threat to human society and the natural environment. It may not only cause secondary disasters such as urban waterlogging, landslides, and mud-rock flows, but may also lead to traffic disruptions, infrastructure damage, and agricultural losses, causing serious impacts on people's lives and property safety and social and economic development. Traditional short-term heavy rainfall warning methods mainly rely on the experience of meteorological experts and numerical weather forecast models. These methods have limitations when processing massive meteorological data and are greatly affected by subjectivity. In addition, the accuracy of numerical weather forecast models is limited in the face of complex terrain and small and medium-scale weather systems, and it is difficult to meet the warning needs for short-term heavy rainfall, a localized and rapidly changing weather phenomenon. With the development of deep learning technology, its application in the field of meteorology has gradually attracted attention, showing great potential in short-term heavy rainfall warning, and is expected to provide more accurate and efficient warning solutions. Summary of the invention

[0003] The purpose of the present invention is to provide a short-term heavy rainfall early warning method based on deep learning.

[0004] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0005] The present invention comprises the following steps:

[0006] Acquire combined reflectivity factor data of a weather radar network covering a target area and its surrounding areas, collect rainfall data of the target area, and pre-process the combined reflectivity factor data and the rainfall data;

[0007] Obtaining SDHR labels according to the rainfall data, extracting, selecting and classifying the combined reflectivity factor data to obtain impact data;

[0008] Performing an impact analysis on the rainfall data to obtain a risk coefficient, and constructing a short-term heavy rainfall warning model using a ConvLSTM algorithm based on the impact data, the risk coefficient and the SDHR label;

[0009] The observation data is input into the short-term heavy rainfall warning model, and the warning result is output.

[0010] Furthermore, the method of obtaining the combined reflectivity factor data of the radar network of the target area and its surrounding areas and collecting the rainfall data of the target area includes:

[0011] The radar network's equipment includes S-band radar and C-band radar, which collects rainfall data from automatic weather stations within the target area.

[0012] Furthermore, the method for obtaining the SDHR tag according to the rainfall data includes:

[0013] The standard for short-term heavy rainfall is that the cumulative rainfall exceeds 20 mm within 1 hour or the cumulative rainfall exceeds 50 mm within 3 hours. When the rainfall data of 5 or more automatic weather stations in the target area meet the short-term heavy rainfall standard at the same time, the SDHR label is 1, otherwise it is 0.

[0014] Furthermore, the method of extracting, selecting and classifying the combined reflectivity factor data to obtain the influencing data comprises:

[0015] Extract the feature data of the combined reflectivity factor data, including the maximum reflectivity value, the average value of the combined reflectivity factor, the change rate, the proportion of high reflectivity areas and the proportion of low reflectivity areas, and the correlation between the feature data and the SDHR label:

[0016]

[0017] in is a feature set and feature set The mutual information between For SDHR tags, is the characteristic data, and is the index of the feature, yes and The joint probability distribution of yes The marginal probability of yes The marginal probability distribution of Based on prior knowledge and The weight function between

[0018] Remove the feature data that is less than the preset correlation threshold to obtain the influencing features, establish data points, use the silhouette coefficient method to obtain the number of clusters, and cluster the data points:

[0019]

[0020]

[0021] in, is the objective function, n is the number of data points, c is the number of clusters, is the membership of the ith data point to the jth cluster, M is the fuzzy factor, is the fuzzy penalty coefficient of the jth cluster, is the data point i, is the center point of cluster j, is the set of data points in cluster j, h is the bandwidth parameter, is the regularization parameter, is the density of cluster j, is the Dirac function, is the distance threshold for measuring the local density of data points in cluster j,

[0022] Membership function:

[0023]

[0024] in, is the Euclidean distance between the i-th data point and the cluster center of the j-th cluster, is the maximum value of the Euclidean distance between all data points and the cluster center of the jth cluster, It is the sum of the rates of change of the feature data corresponding to the data point i and the center of cluster j. is the maximum value of the rate of change of all data points and the characteristic data of the cluster center, is the adjustment weight of the change rate of the data point and the characteristic data of the cluster center,

[0025] Cluster center update formula:

[0026]

[0027] in and are the center points of cluster j after iteration s and iteration s+1 respectively. is the membership of the ith data point to the jth cluster after iteration s+1 steps, is the learning rate after s iterations,

[0028] Divide the data points into the cluster with the largest membership and evaluate the clustering results:

[0029]

[0030] in Score the clustering results, where and They are the data points in cluster m and cluster p respectively. They are iteratively updated until the clustering result score no longer increases. The influencing data include influencing features, category labels and membership to category labels.

[0031] Furthermore, the method of performing impact analysis on the rainfall data to obtain a risk coefficient includes:

[0032] A risk prediction model is established using a neural network algorithm based on rainfall data and the SDHR label of the forecast time. The neural network algorithm includes an input layer, a function layer, and an output layer. The regularized loss function of the function layer is:

[0033]

[0034] Among them, y is the SDHR label of the prediction time, a is the prediction probability of y=1 output by the output layer, J is the rainfall data, is the weight matrix of the function layer, for The elements in is the regularization coefficient,

[0035] The predicted probability of the SDHR label being 1 at the predicted time output by the risk prediction model is taken as the risk coefficient.

[0036] Furthermore, a method for constructing a short-term heavy rainfall warning model using a ConvLSTM algorithm according to the impact data, the risk coefficient and the SDHR label includes:

[0037] The category label is converted into numerical data, the impact data and risk coefficient are input data, the SDHR label of the prediction time is the label of the input data, and the formulas of the input gate, forget gate, output gate, hidden state and storage unit of the ConvLSTM algorithm include:

[0038]

[0039] in, , , , and are the input gate, forget gate, output gate, hidden state and storage unit at time step t respectively. and are the hidden state and storage unit at time step t-1, is the input data at time step t, is the sigmoid activation function, is the hyperbolic tangent activation function, , , and are the bias items of the input gate, forget gate, output gate and storage unit respectively, , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, , , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, , , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, Describing convolution, represents element-wise multiplication, is the attention weight matrix at time step t, and They are The weight matrix and bias term,

[0040] Loss function of the short-term heavy rainfall warning model:

[0041]

[0042] in, is the number of samples of input data, is the label of the i-th sample, is the probability of short-term heavy rainfall predicted by the short-term heavy rainfall warning model, indicating the prediction label The probability of is the adjustment parameter, is the k-th order statistical characteristic of rainfall data J, including skewness, kurtosis and coefficient of variation, is the statistical characteristic order,

[0043] An early warning will be issued when the predicted probability of short-term heavy rainfall is greater than 0.5.

[0044] In a second aspect, an embodiment of the present application further provides an electronic device, including:

[0045] A processor; and a memory arranged to store computer executable instructions, which when executed cause the processor to perform the method steps described in the first aspect.

[0046] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple applications, the electronic device executes the method steps described in the first aspect.

[0047] The beneficial effects of the present invention are:

[0048] The present invention is a short-term heavy rainfall early warning method based on deep learning. Compared with the prior art, the present invention has the following technical effects:

[0049] The short-term heavy rainfall early warning method based on deep learning has significant advantages. It accurately utilizes multi-source data and deeply mines feature associations to build models. It effectively improves early warning accuracy and enhances interpretability, thereby gaining critical time for disaster prevention and reducing losses, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flowchart of the steps of a short-term heavy rainfall early warning method based on deep learning of the present invention;

[0051] Figure 2 It is a schematic diagram of the structure of an electronic device in an embodiment of this specification. DETAILED DESCRIPTION

[0052] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0053] A short-term heavy rainfall early warning method based on deep learning of the present invention comprises the following steps:

[0054] like Figure 1 As shown, in this embodiment, the following steps are included:

[0055] Acquire combined reflectivity factor data of a weather radar network covering a target area and its surrounding areas, collect rainfall data of the target area, and pre-process the combined reflectivity factor data and the rainfall data;

[0056] In the actual evaluation, the combined reflectivity factor data of the weather radar network covering the 800km×800km area in City A and its surrounding areas were collected during June, July and August from 2016 to 2021. The horizontal spatial resolution of the data was 1km×1km, and the temporal resolution was 6 minutes. The radar network consists of 10 radars, including 8 S-band radars and 2 C-band radars; the 10-minute resolution rainfall data set of 556 automatic weather stations was obtained from the Meteorological Observation Center of the Meteorological Bureau of City A, covering June, July and August from 2016 to 2021. City A has a total of 5 districts, ranging in size from 1000 square kilometers to 2000 square kilometers. These 5 districts are regarded as 5 target areas; preprocessing includes: using a smoothing filter to resample the combined reflectivity factor data of 800 × 800 grid points to 64 × 64 grid points, normalizing the combined reflectivity factor data, and standardizing the rainfall data;

[0057] Obtaining SDHR labels according to the rainfall data, extracting, selecting and classifying the combined reflectivity factor data to obtain impact data;

[0058] In the actual evaluation, the SDHR labels for June, July and August from 2016 to 2021 are:

[0059]

[0060] The influencing features of the combined reflectivity factor data are the maximum reflectivity value, the average value of the combined reflectivity factor, the change rate, and the proportion of high reflectivity areas. The influencing features are divided into three categories:

[0061] The first category has high maximum reflectivity, high average reflectivity, high variation rate, and a large proportion of high reflectivity areas; the second category has medium maximum reflectivity, medium average reflectivity, medium variation rate, and high reflectivity areas; the third category has high and low maximum reflectivity, low average reflectivity, low variation rate, and a low proportion of high reflectivity areas;

[0062] Performing an impact analysis on the rainfall data to obtain a risk coefficient, and constructing a short-term heavy rainfall warning model using a ConvLSTM algorithm based on the impact data, the risk coefficient and the SDHR label;

[0063] In the actual assessment, the rainfall data of the previous hour is used to predict the risk factor every 10 minutes in the next two hours, with a total of 12 predictions. The impact data and rainfall data of the previous hour are used to predict the risk factor every 10 minutes in the next two hours, with a total of 12 predictions.

[0064] The observation data is input into the short-term heavy rainfall warning model, and the warning result is output.

[0065] In this embodiment, the method of obtaining the combined reflectivity factor data of the radar network of the target area and its surrounding areas and collecting the rainfall data of the target area includes:

[0066] The radar network includes S-band radar and C-band radar, which collect rainfall data from automatic weather stations in the target area. The time range of the collected combined reflectivity factor data and rainfall data is the same.

[0067] In this embodiment, the method for obtaining the SDHR tag according to the rainfall data includes:

[0068] The standard for short-term heavy rainfall is that the cumulative rainfall exceeds 20 mm within 1 hour or the cumulative rainfall exceeds 50 mm within 3 hours. When the rainfall data of 5 or more automatic weather stations in the target area meet the short-term heavy rainfall standard at the same time, the SDHR label is 1, otherwise it is 0.

[0069] In this embodiment, the method of extracting, selecting and classifying the combined reflectivity factor data to obtain the influence data includes:

[0070] Extract the characteristic data of the combined reflectivity factor data, including the maximum reflectivity value, the average value of the combined reflectivity factor, the rate of change, the proportion of high reflectivity areas and the proportion of low reflectivity areas. The spatial distribution feature is the characteristic vector converted from the reflectivity intensity distribution map. The correlation between the characteristic data and the SDHR label is:

[0071]

[0072] in is a feature set and feature set The mutual information between For SDHR tags, is the characteristic data, and is the index of the feature, yes and The joint probability distribution of yes The marginal probability of yes The marginal probability distribution of Based on prior knowledge and The weight function between

[0073] Remove the feature data that is less than the preset correlation threshold to obtain the influencing features, establish data points, use the silhouette coefficient method to obtain the number of clusters, and cluster the data points:

[0074]

[0075]

[0076] in, is the objective function, n is the number of data points, c is the number of clusters, is the membership of the ith data point to the jth cluster, M is the fuzzy factor, is the fuzzy penalty coefficient of the jth cluster, is the data point i, is the center point of cluster j, is the set of data points in cluster j, h is the bandwidth parameter, is the regularization parameter, is the density of cluster j, is the Dirac function, is the distance threshold for measuring the local density of data points in cluster j,

[0077] Membership function:

[0078]

[0079] in, is the Euclidean distance between the i-th data point and the cluster center of the j-th cluster, is the maximum value of the Euclidean distance between all data points and the cluster center of the jth cluster, It is the sum of the rates of change of the feature data corresponding to the data point i and the center of cluster j. is the maximum value of the rate of change of all data points and the characteristic data of the cluster center, is the adjustment weight of the change rate of the data point and the characteristic data of the cluster center,

[0080] Cluster center update formula:

[0081]

[0082] in and are the center points of cluster j after iteration s and iteration s+1 respectively. is the membership of the ith data point to the jth cluster after iteration s+1 steps, is the learning rate after s iterations,

[0083] Divide the data points into the cluster with the largest membership and evaluate the clustering results:

[0084]

[0085] in Score the clustering results, where and They are the data points in cluster m and cluster p respectively. They are iteratively updated until the clustering result score no longer increases. The influencing data include influencing features, category labels and membership to category labels.

[0086] In this embodiment, the method of performing impact analysis on the rainfall data to obtain a risk coefficient includes:

[0087] A risk prediction model is established using a neural network algorithm based on rainfall data and the SDHR label of the forecast time. The neural network algorithm includes an input layer, a function layer, and an output layer. The regularized loss function of the function layer is:

[0088]

[0089] Among them, y is the SDHR label of the prediction time, a is the prediction probability of y=1 output by the output layer, J is the rainfall data, is the weight matrix of the function layer, for The elements in is the regularization coefficient,

[0090] The predicted probability of the SDHR label being 1 at the predicted time output by the risk prediction model is taken as the risk coefficient.

[0091] In this embodiment, the method of constructing a short-term heavy rainfall warning model using the ConvLSTM algorithm according to the impact data, the risk coefficient and the SDHR label includes:

[0092] The category label is converted into numerical data, the impact data and risk coefficient are input data, the SDHR label of the prediction time is the label of the input data, and the formulas of the input gate, forget gate, output gate, hidden state and storage unit of the ConvLSTM algorithm include:

[0093]

[0094] in, , , , and are the input gate, forget gate, output gate, hidden state and storage unit at time step t respectively. and are the hidden state and storage unit at time step t-1, is the input data at time step t, is the sigmoid activation function, is the hyperbolic tangent activation function, , , and are the bias items of the input gate, forget gate, output gate and storage unit respectively, , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, , , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, , , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, Describing convolution, represents element-wise multiplication, is the attention weight matrix at time step t, and They are The weight matrix and bias term,

[0095] Loss function of the short-term heavy rainfall warning model:

[0096]

[0097] in, is the number of samples of input data, is the label of the i-th sample, is the probability of short-term heavy rainfall predicted by the short-term heavy rainfall warning model, indicating the prediction label The probability of is the adjustment parameter, is the k-th order statistical characteristic of rainfall data J, including skewness, kurtosis and coefficient of variation, is the statistical characteristic order,

[0098] An early warning will be issued when the predicted probability of short-term heavy rainfall is greater than 0.5.

[0099] Figure 2 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 2 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. The memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. Of course, the electronic device may also include hardware required for other services.

[0100] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0101] The memory is used to store the program. Specifically, the program may include a program code, and the program code includes a computer operation instruction. The memory may include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0102] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a short-term heavy rainfall warning device based on deep learning at the logical level. The processor executes the program stored in the memory and is specifically used to execute any of the above-mentioned short-term heavy rainfall warning methods based on deep learning.

[0103] The above application Figure 1 The short-term heavy rainfall warning method based on deep learning disclosed in the illustrated embodiment can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in a decoding processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0104] The electronic device may also perform Figure 1 A short-term heavy rainfall early warning method based on deep learning is proposed and implemented Figure 1 The functions of the illustrated embodiment will not be described in detail in the embodiments of the present application.

[0105] An embodiment of the present application also proposes a computer-readable storage medium, which stores one or more programs, and the one or more programs include instructions. When the instructions are executed by an electronic device including multiple applications, any one of the aforementioned short-term heavy rainfall warning methods based on deep learning is executed.

[0106] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0107] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0108] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0109] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0110] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0111] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0112] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0113] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0114] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A short-term heavy rainfall early warning method based on deep learning, characterized in that: The following steps are involved: Acquire composite reflectivity data of the radar network of the target area and its surrounding areas, collect rainfall data of the target area, and pre-process the composite reflectivity data and the rainfall data; Obtain SDHR labels according to the rainfall data, extract, select and classify the composite reflectivity data to obtain impact data; Performing an impact analysis on the rainfall data to obtain a risk coefficient, and constructing a short-term heavy rainfall warning model using a ConvLSTM algorithm based on the impact data, the risk coefficient and the SDHR label; Inputting the observation data into the short-term heavy rainfall warning model and outputting the warning result; The method for obtaining the SDHR label according to the rainfall data comprises: The standard for short-term heavy rainfall is that the cumulative rainfall exceeds 20 mm within 1 hour or exceeds 50 mm within 3 hours. When the rainfall data of 5 or more automatic weather stations in the target area meet the standard for short-term heavy rainfall at the same time, the SDHR label is 1, otherwise it is 0; The method of performing impact analysis on the rainfall data to obtain a risk coefficient includes: A risk prediction model is established using a neural network algorithm based on rainfall data and the SDHR label of the forecast time. The neural network algorithm includes an input layer, a function layer, and an output layer. The regularized loss function of the function layer is: Among them, y is the SDHR label of the prediction time, a is the prediction probability of y=1 output by the output layer, J is the rainfall data, is the weight matrix of the function layer, for The elements in is the regularization coefficient, The predicted probability of the SDHR label being 1 at the predicted time output by the risk prediction model is taken as the risk coefficient.

2. A short-term heavy rainfall early warning method based on deep learning according to claim 1, characterized in that: The method of obtaining composite reflectivity data of a radar network in a target area and its surrounding areas and collecting rainfall data in the target area comprises: The radar network's equipment includes S-band radar and C-band radar, which collects rainfall data from automatic weather stations within the target area.

3. According to the short-term heavy rainfall early warning method based on deep learning as claimed in claim 1, it is characterized in that: The method of extracting, selecting and classifying the composite reflectivity data to obtain the influencing data comprises: Extract the characteristic data of the composite reflectivity data, including the maximum reflectivity value, the average reflectivity factor, the change rate, the proportion of high reflectivity area and the proportion of low reflectivity area, and the correlation between the characteristic data and the SDHR label: in is a feature set and feature set The mutual information between For SDHR tags, is the characteristic data, and is the index of the feature, yes and The joint probability distribution of yes The marginal probability of yes The marginal probability distribution of Based on prior knowledge and The weight function between Remove the feature data that is less than the preset correlation threshold to obtain the influencing features, establish data points, use the silhouette coefficient method to obtain the number of clusters, and cluster the data points: in, is the objective function, n is the number of data points, c is the number of clusters, is the membership of the ith data point to the jth cluster, M is the fuzzy factor, is the fuzzy penalty coefficient of the jth cluster, is the data point i, is the center point of cluster j, is the set of data points in cluster j, h is the bandwidth parameter, is the regularization parameter, is the density of cluster j, is the Dirac function, is the distance threshold for measuring the local density of data points in cluster j, Membership function: in, is the Euclidean distance between the i-th data point and the cluster center of the j-th cluster, is the maximum value of the Euclidean distance between all data points and the cluster center of the jth cluster, It is the sum of the rates of change of the feature data corresponding to the data point i and the center of cluster j. is the maximum value of the rate of change of all data points and the characteristic data of the cluster center, is the adjustment weight of the change rate of the data point and the characteristic data of the cluster center, Cluster center update formula: in and are the center points of cluster j after iteration s and iteration s+1 respectively. is the membership of the ith data point to the jth cluster after iteration s+1 steps, is the learning rate after s iterations, Divide the data points into the cluster with the largest membership and evaluate the clustering results: in Score the clustering results, where and are the data points in cluster m and cluster p respectively, is the set of data points of cluster p, which is iteratively updated until the clustering result score no longer increases. The influencing data include influencing features, category labels and membership to category labels.

4. The short-term heavy rainfall early warning method based on deep learning according to claim 1 is characterized in that: The method of constructing a short-term heavy rainfall warning model using a ConvLSTM algorithm according to the impact data, the risk coefficient and the SDHR label includes: The category label is converted into numerical data, the impact data and risk coefficient are input data, the SDHR label of the prediction time is the label of the input data, and the formulas of the input gate, forget gate, output gate, hidden state and storage unit of the ConvLSTM algorithm include: in, , , , and are the input gate, forget gate, output gate, hidden state and storage unit at time step t respectively. and are the hidden state and storage unit at time step t-1, is the input data at time step t, is the sigmoid activation function, is the hyperbolic tangent activation function, , , and are the bias items of the input gate, forget gate, output gate and storage unit respectively, , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, , , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, , , and They are to the weight matrices of the input gate, forget gate, hidden state, and output gate, Describing convolution, represents element-wise multiplication, is the attention weight matrix at time step t, and They are The weight matrix and bias term, Loss function of the short-term heavy rainfall warning model: in, is the number of samples of input data, is the label of the i-th sample, is the probability of short-term heavy rainfall predicted by the short-term heavy rainfall warning model, indicating the prediction label The probability of is the adjustment parameter, is the k-th order statistical characteristic of rainfall data J, including skewness, kurtosis and coefficient of variation, is the statistical characteristic order, An early warning will be issued when the predicted probability of short-term heavy rainfall is greater than 0.

5.

5. An electronic device comprising: processor; as well as A memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the method of any one of claims 1 to 4.

6. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, enables the electronic device to execute the method according to any one of claims 1 to 4.

Images