Lightning early warning method, device and equipment based on weight integrated neural network

Abstract

The invention discloses a lightning early warning method, device and equipment based on a weight integrated neural network, and relates to the technical field of meteorological observation. According to the method, firstly, a meteorological parameter combination composed of all meteorological parameters which can be observed in an observation area is acquired, and then respective independent lightning early warning neural network models are established for different meteorological parameters in the meteorological parameter combination; integrating lightning forecast results of the models in a weighted summation mode, automatically learning optimal integrated weight coefficient groups corresponding to different sub-meteorological parameter combinations through a genetic algorithm in the process, and respectively learning and storing the weight coefficient groups for parameter permutation combinations; in the real-time early-warning stage, the corresponding lightning early-warning neural network model and the weight group are called for any available meteorological parameters in the current time period to give out a lightning forecast result, and the problems that an existing neural network forecasting method is poor in model stability and generalization and large in modeling time consumption are solved.

Description

Technical Field

[0001] The present application relates to the field of meteorological observation technology, and in particular to a lightning warning method, device and equipment based on a weighted integrated neural network. Background Art

[0002] The current mainstream technology for lightning warning is to identify thunderstorms through meteorological parameter observation data such as lightning location data and radar observation data, and to extrapolate their moving direction to obtain the size of future thunderstorms and their possible locations, and to obtain the probability of lightning in the region through the relationship between radar and other data and the threshold value of lightning occurrence; or to establish a threshold relationship between meteorological parameters and future lightning occurrence or a multivariate linear equation to predict the probability of future lightning occurrence. For example, by establishing a linear relationship between the frequency of future lightning occurrence in the target area and the radar observation data in the surrounding area, the frequency and probability of lightning occurrence can be predicted. Since the occurrence mechanism of lightning has not yet been clarified, the relationship between its occurrence and other meteorological parameters is very complex and may not obey a linear relationship. Therefore, the forecast accuracy of this type of method is not high; this type of method requires manual adjustment of weights and thresholds in different regions and seasons, so the generalization is also poor.

[0003] With the development of science and technology, with the help of the update of modern equipment such as radars and satellites, a large amount of relevant atmospheric observation data has been accumulated, and the quality has been greatly improved, which has improved the effect of short-term lightning forecasting methods and also bred the emergence of neural network lightning warning methods. Compared with the traditional method, the lightning warning neural network model established using the neural network method relies on a large number of learnable parameters and can more accurately identify the complex nonlinear relationship between lightning and other meteorological parameters. Therefore, with the large amount of accumulated data and high quality today, its warning effect has gradually surpassed the mainstream linear method, making it one of the important methods of lightning warning. However, the current neural network model used for lightning warning can only accept fixed-form data as input, that is, the temporal and spatial resolution of the input data and the type of meteorological parameters must be fixed, which leads to the following problems.

[0004] 1. Poor model stability. In the real-time warning process, the meteorological parameter observations used by the warning model are likely to be missing for a short period of time or the quality inspection of some observation data is not up to standard, resulting in some parameters being unavailable. At this time, the warning model cannot continue to make forecasts, and this situation will become more frequent as the number of parameters used increases.

[0005] 2. Poor generalization of the model. The available meteorological parameters in different regions are different, which will result in the lightning warning neural network model established in the current region being unable to be generalized to other regions with different available meteorological parameters.

[0006] 3. It takes a lot of time to select parameters when building the model. At present, the method of selecting meteorological parameters for building lightning warning neural network models is generally to build the model by arranging and combining various parameters and selecting the parameter combination with the best modeling effect. The time consumption of this method increases exponentially with the increase in the number of parameters. Summary of the invention

[0007] The purpose of this application is to provide a lightning warning method, device and equipment based on a weighted integrated neural network to overcome the problems of poor model stability, poor generalization and long modeling time in existing neural network prediction methods.

[0008] To achieve the above objectives, this application provides the following solutions.

[0009] In a first aspect, the present application provides a lightning warning method based on a weighted integrated neural network, comprising the following steps.

[0010] Obtain a historical observation data sequence of each meteorological parameter in a meteorological parameter combination; the meteorological parameter combination is a combination of meteorological parameters that can be observed in the observation area.

[0011] Using the sliding window method, sample sets are established based on the historical observation data series of each meteorological parameter to obtain the sample sets corresponding to each meteorological parameter.

[0012] The neural network models are trained respectively using the sample sets corresponding to each meteorological parameter to obtain the lightning warning neural network model corresponding to each meteorological parameter.

[0013] A genetic algorithm is used to determine the weight of the lightning forecast result output by the lightning warning neural network model corresponding to each meteorological parameter in each sub-meteorological parameter combination as the weight corresponding to each meteorological parameter in each sub-meteorological parameter combination; each sub-meteorological parameter combination is obtained by permuting and combining the meteorological parameters in the meteorological parameter combination.

[0014] The current observation data sequence of each meteorological parameter in the current meteorological parameter combination is obtained; the current meteorological parameter combination is a combination of meteorological parameters of the observation area that can be currently observed.

[0015] The current observation data sequence of each meteorological parameter in the current meteorological parameter combination is respectively input into the lightning warning neural network model corresponding to each meteorological parameter in the current meteorological parameter combination to obtain the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination.

[0016] The weight of each meteorological parameter in the sub-meteorological parameter combination that is consistent with the type of meteorological parameters included in the current meteorological parameter combination is determined as the weight of each meteorological parameter in the current meteorological parameter combination.

[0017] The final prediction result is calculated based on the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination and the weight corresponding to each meteorological parameter in the current meteorological parameter combination.

[0018] In a second aspect, the present application provides a lightning warning device based on a weighted integrated neural network, wherein the lightning warning device based on a weighted integrated neural network applies the above-mentioned lightning warning method based on a weighted integrated neural network, and the lightning warning device based on a weighted integrated neural network includes the following modules.

[0019] The historical observation data acquisition module is used to obtain the historical observation data sequence of each meteorological parameter in the meteorological parameter combination; the meteorological parameter combination is a combination of meteorological parameters that can be observed in the observation area.

[0020] The sample set building module is used to use the sliding window method to build sample sets based on the historical observation data sequence of each meteorological parameter, so as to obtain the sample set corresponding to each meteorological parameter.

[0021] The model training module is used to train the neural network model using the sample set corresponding to each meteorological parameter to obtain the lightning warning neural network model corresponding to each meteorological parameter.

[0022] The weight determination module is used to use a genetic algorithm to determine the weight of the lightning forecast result output by the lightning warning neural network model corresponding to each meteorological parameter in each sub-meteorological parameter combination as the weight corresponding to each meteorological parameter in each sub-meteorological parameter combination; each sub-meteorological parameter combination is obtained by arranging and combining each meteorological parameter in the meteorological parameter combination.

[0023] The current observation data acquisition module is used to obtain the current observation data sequence of each meteorological parameter in the current meteorological parameter combination; the current meteorological parameter combination is a combination of meteorological parameters of the observation area that can be observed currently.

[0024] The prediction module is used to input the current observation data sequence of each meteorological parameter in the current meteorological parameter combination into the lightning warning neural network model corresponding to each meteorological parameter in the current meteorological parameter combination, and obtain the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination.

[0025] The weight selection module is used to determine the weight of each meteorological parameter in the sub-meteorological parameter combination that is consistent with the type of meteorological parameters included in the current meteorological parameter combination as the weight of each meteorological parameter in the current meteorological parameter combination.

[0026] The comprehensive calculation module is used to calculate the final prediction result according to the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination and the weight corresponding to each meteorological parameter in the current meteorological parameter combination.

[0027] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned lightning warning method based on weighted integrated neural network.

[0028] According to the specific embodiments provided in this application, this application has the following technical effects.

[0029] The present application provides a lightning warning method, device and equipment based on a weighted integrated neural network. The present application first obtains a meteorological parameter combination consisting of all meteorological parameters that can be observed in the observation area, and then establishes independent lightning warning neural network models for each different meteorological parameter in the meteorological parameter combination; the lightning forecast results of each model are integrated by weighted summation, and in this process, the optimal integrated weight coefficient group corresponding to different sub-meteorological parameter combinations is automatically learned by a genetic algorithm, and the weight coefficient group is learned and saved for parameter permutations and combinations (i.e., sub-meteorological parameter combinations); in the real-time warning stage, for any available meteorological parameter in the current period, the corresponding model and weight coefficient group are called to give the lightning forecast result, thereby overcoming the problems of poor model stability, poor generalization and long modeling time in the existing neural network prediction method. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in 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 paying creative work.

[0031] Figure 1 A schematic flow chart of a lightning warning method based on a weighted integrated neural network provided in one embodiment of the present application.

[0032] Figure 2 A schematic diagram of the principle of a lightning warning method based on a weighted integrated neural network provided in one embodiment of the present application.

[0033] Figure 3 A schematic diagram of the overall structure of the AutoFitNet neural network model provided in one embodiment of the present application.

[0034] Figure 4 A schematic diagram of the structure of a Multi-Scale Conv layer provided in one embodiment of the present application.

[0035] Figure 5 A schematic diagram of a flow chart of a lightning forecast process provided in accordance with an embodiment of the present application.

[0036] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0038] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0039] In an exemplary embodiment, Figure 1 and Figure 2 As shown, a lightning warning method based on a weighted integrated neural network is provided, including the following steps 101 to 108.

[0040] Step 101, obtaining a historical observation data sequence of each meteorological parameter in a meteorological parameter combination; the meteorological parameter combination is a combination of meteorological parameters that can be observed in an observation area.

[0041] Step 102: Using a sliding window method, a sample set is established based on the historical observation data sequence of each meteorological parameter to obtain a sample set corresponding to each meteorological parameter.

[0042] Step 103: Use the sample sets corresponding to each meteorological parameter to train the neural network model respectively, and obtain the lightning warning neural network model corresponding to each meteorological parameter.

[0043] Step 104, using a genetic algorithm to determine the weight of the lightning forecast result output by the lightning warning neural network model corresponding to each meteorological parameter in each sub-meteorological parameter combination as the weight corresponding to each meteorological parameter in each sub-meteorological parameter combination; each sub-meteorological parameter combination is obtained by arranging and combining each meteorological parameter in the meteorological parameter combination.

[0044] Step 105, obtaining the current observation data sequence of each meteorological parameter in the current meteorological parameter combination; the current meteorological parameter combination is a combination of meteorological parameters of the observation area that can be currently observed; wherein the current observation data sequence is a sequence of observation data within a sliding window containing the current observation data arranged in chronological order, and the sliding window is a sliding window used in the sliding window method.

[0045] Step 106, respectively inputting the current observation data sequence of each meteorological parameter in the current meteorological parameter combination into the lightning warning neural network model corresponding to each meteorological parameter in the current meteorological parameter combination to obtain the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination.

[0046] Step 107: Determine the weight of each meteorological parameter in the sub-meteorological parameter combination that is consistent with the type of meteorological parameters included in the current meteorological parameter combination as the weight of each meteorological parameter in the current meteorological parameter combination.

[0047] Step 108, calculating the final prediction result according to the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination and the weight corresponding to each meteorological parameter in the current meteorological parameter combination.

[0048] Implementing the above steps 101 to 108, based on the weighted integrated neural network, by means of model separation and weighted integration, allows the input of the model to change, thereby enabling the model to continue to operate when some data is missing, so that the model can output forecast results more stably, and at the same time enable the model to be extended to different regions where there are differences in available meteorological parameters. The weighted integration mechanism of the neural network will automatically give higher (lower) weights to meteorological parameters with high (low) correlation with the probability of future lightning occurrence, without the need to perform permutation and combination modeling of parameters for screening, so its modeling time is linearly related to the number of parameters, and its modeling time is greatly reduced compared to traditional neural networks.

[0049] In the above steps 101 to 108, the probability of lightning in the target area within the next 0-1 hour (hereinafter referred to as lightning probability) can be predicted by using multi-source observation parameters in the observation area (including but not limited to lightning location, radar reflectivity, meteorological station observation, satellite observation). The method includes: selecting an observation area, determining the available observation parameters in the area, preprocessing the parameter data, and integrating them into a data set that can be used for neural network training and testing; using neural networks to establish independent lightning warning neural network models for different parameters; integrating the forecast results of each lightning warning neural network model by weighted summation, automatically learning the optimal integrated weight coefficient group through genetic algorithms, and learning and saving weight coefficient groups for different parameter permutations and combinations; in the real-time warning stage, inputting any available parameters in the current period into the model, calling the corresponding lightning warning neural network model and weight group to give the lightning forecast result.

[0050] After the lightning warning neural network model is established, it can still use any combination of meteorological parameters as input for lightning warning. This enables the model to continue to provide lightning warnings when the observation of a certain meteorological parameter in the real-time forecast is interrupted. Therefore, compared with the existing technology, this model can provide real-time lightning warnings more stably while ensuring accuracy.

[0051] In another exemplary embodiment of the present application, step 102 is a process of quality detection and preprocessing of observation data. The staff determines the forecast area and available meteorological parameter observation data, performs quality control and preprocessing on them, and constructs a historical data set (i.e., sample set) for neural network model training. The specific implementation method is: quality detection of the selected meteorological parameter historical observation data, removal of unqualified data, interpolation of the smaller missing parts of the historical observation data of each parameter (using linear interpolation, Gaussian diffusion, etc., according to the data characteristics selection method), to obtain the historical observation data sequence of each meteorological parameter, and use the sliding window method to establish a sample set for each meteorological parameter. For example, the length of the sliding window and the prediction window are set to 1 hour, and the observation data within one hour is used as the sample (the time resolution of each meteorological parameter is different), and the probability of lightning in the next hour is used as the annotation (the time resolution is 10 minutes, and each frame contains the lightning that occurs within 10 minutes from the time point). According to the time order, the training set, the validation set, and the test set are divided in a ratio of 8:1:1. The purpose of dividing the sample set in time order is to prevent data leakage from causing an over-evaluation of the model effect.

[0052] Exemplarily, the above step 102 specifically includes the following steps 201-202.

[0053] Step 201, setting the length of the sliding window of the kth meteorological parameter; k=1, 2, ..., K; K is the number of meteorological parameter types in the meteorological parameter combination.

[0054] Step 202, performing a sliding window operation on the historical observation data sequence of the kth meteorological parameter according to the length of the sliding window of the kth meteorological parameter, determining each historical observation data of the kth meteorological parameter within the sliding window as a sample input, and determining the lightning probability of each time step within the prediction window as a sample label, and constructing a sample set corresponding to the kth meteorological parameter; the prediction window is located after the sliding window.

[0055] In another exemplary embodiment of the present application, the above steps 103 and 104 use the sample set to construct a lightning warning neural network model to obtain a lightning warning neural network model corresponding to each meteorological parameter and a weight group corresponding to each parameter combination.

[0056] In another exemplary embodiment of the present application, the neural network model in the above step 103 is an AutoFitNet (automatic adaptive network) neural network model, and the AutoFitNet neural network model is improved by the ConvLSTM (Convolution Long Short-Term Memory convolution long short-term memory network) model, such as Figure 3 and Figure 4 As shown in the figure, the improvement method is to add a fully connected reduction layer and two multi-scale convolution layers at the input end of the ConvLSTM model, and add two multi-scale convolution layers and a fully connected magnification layer at the output end of the ConvLSTM model.

[0057] like Figure 3 As shown in the figure, the AutoFitNet neural network model of this application adds a fully connected reduction layer (Multi-Layer Perceptron Down, MLP Down), a fully connected amplification layer (Multi-Layer Perceptron Up, MLP Up) and a multi-scale convolution layer (Multi-Scale Conv) to the ConvLSTM model. The fully connected reduction layer enables the model to adapt to inputs of any size without changing the structure; the multi-scale convolution layer can extract spatial features of multiple resolutions in the input.

[0058] Let the input be [X1,X2,...X Tn ], the number of observation frames in the sliding window (i.e. 1 hour) is Tn, and the Tn value of each parameter is inconsistent; the output is recorded as [Y1, Y2, ... Y Tm ], the number of frames in one hour is Tm, and Tm can be adjusted according to demand. The specific working process of the AutoFitNet neural network model is as follows: Step 1: Use the fully connected downscaling layer (MLP Down) to compress the input features. The formula for using the fully connected downscaling layer is shown in formula (1).

[0059]

[0060] in, is the feature map output by the fully connected reduction layer, is the first feature map in the fully connected reduction layer input Line The observed values ​​of the meteorological parameters of the characteristic points of the column, that is, for No. OK The observed values ​​of meteorological parameters of the characteristic points of the column, is the first feature map in the fully connected reduction layer input Line The weight parameter of the feature points of the column, and are the number of rows and columns of feature points in the feature map input to the fully connected reduction layer, respectively.

[0061] The fully connected reduction layer inputs data of any resolution and outputs a feature map with 128 channels and a resolution of 32×32.

[0062] Step 2: After two layers of multi-scale convolution (Multi-Scale Conv) layers (its structure is as follows Figure 4 As shown in the figure, the network is stacked to extract the spatial features of the input and the shortcut (shortest path) technology is used to speed up the model convergence.

[0063] Step 3: Use the ConvLSTM layer in the ConvLSTM model to learn the temporal relationship between input and output. When the ConvLSTM layer is running, it needs to Perform the following calculations: Input Gate: (2) Forget Gate: (3) Candidate cell states: (4) Cell status update: (5) Output Gate: (6) Hide status updates: (7) in, is the time step The output state of the input gate, is the convolution kernel of the input state of the input gate, is the time step The input state, is the convolution kernel of the hidden state of the input gate, is the time step The hidden state of is the bias of the input gate, is the Sigmod activation function, Represents a convolution operation. is the time step The output state of the forget gate is is the convolution kernel of the input state of the forget gate, is the convolution kernel of the hidden state of the forget gate, is the bias of the forget gate. is the time step The output state of the output gate, is the convolution kernel of the input state of the output gate, is the convolution kernel of the hidden state of the output gate, is the bias of the output gate. is the time step The hidden state of means that the matrix elements are multiplied one by one, is the time step The cell state, is the time step The cell state, is the time step Candidate cell states, and are the convolution kernels of the input state and hidden state of the hidden layer, is the bias of the hidden layer, is the tanh activation function.

[0064] The ConvLSTM model of this application is divided into an encoding module and a decoding module. The encoding module is used to learn the input time features, and the decoding module is used to learn the output time features. The encoding module and the decoding module both contain two ConvLSTM layers. In the encoding module, the input of each time step of the ConvLSTM layer is , the input of cell state and hidden state is time step The cell state and hidden state , the output is the time step The hidden state , the initial input cell state and hidden state are zero tensors; in the decoding module, the input of a time step of the ConvLSTM layer is the time step The hidden state , the input of cell state and hidden state is time step The cell state and hidden state , the output is the hidden state [H1,H2,...,H m ], the initial input cell state and hidden state are the output of the last time step of the encoding module (such as Figure 3 as shown).

[0065] The hidden states [H1,H2,...,H m ] After a multi-scale convolutional layer and a fully connected magnification layer, it is mapped into a label [Y1,Y2,...,Y m].

[0066] In the above step 103, the following loss function is used to train the model.

[0067] (8) in, is the loss function, is the length of the prediction window, is the time step in the prediction window The error weight is used to adjust the prediction time point that the model focuses on. is the total number of feature points, , and are the number of rows and columns of feature points in the target feature map, respectively. To adjust the weights so that the model tends to learn the features of small samples with higher lightning content, is the adjustment coefficient, and the target feature map is the marked time step The feature map or the time step of the neural network model output The feature map of The time step of the label The total number of lightning flashes in the feature graph of The time step of the label The feature map of Line The lightning probability of the characteristic points of the column, The time step output by the neural network model The feature map of Line The lightning probability of the characteristic points of the column.

[0068] After training, the lightning warning neural network model corresponding to each meteorological parameter is obtained.

[0069] In another exemplary embodiment, in the above step 104, the sample is input into the established model output lightning forecast result, and the error is calculated with the annotation (using formula (8)), and the optimization problem is as follows.

[0070] (9) in, is the weight vector, , , , , For the first, second, Type The weights of the lightning forecast results output by the lightning warning neural network model corresponding to the meteorological parameters, for dimensional set of real numbers, For the The first The meteorological parameters The loss function value of samples is For the The number of meteorological parameter types in each meteorological parameter combination. is the number of samples for each meteorological parameter. For each parameter combination, the genetic algorithm is used to calculate the optimal weight group.

[0071] Output the lightning warning neural network model corresponding to each of the above parameters, as well as the weight group corresponding to the parameter permutation and combination.

[0072] In another exemplary embodiment, the above steps 105 to 108 are a process of performing a real-time lightning warning. Figure 5 As shown, specifically: automatically obtain the observation data of the selected meteorological parameters within 1 hour before the current time (that is, the current observation data sequence of each meteorological parameter in the current meteorological parameter combination), and filter out the parameters that have observations and meet the standards after automated data quality inspection and preprocessing, and call the model corresponding to the parameter and the weight group corresponding to the current meteorological parameter combination to obtain the final prediction result. This process does not require manual operation and is fully automated.

[0073] Based on the same inventive concept, the embodiment of the present application also provides a lightning warning device based on a weighted integrated neural network for implementing the lightning warning method based on a weighted integrated neural network involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of one or more lightning warning devices based on a weighted integrated neural network provided below can refer to the limitations of the lightning warning method based on a weighted integrated neural network above, and will not be repeated here.

[0074] In an exemplary embodiment, a lightning warning device based on a weighted integrated neural network is provided, comprising the following modules.

[0075] The historical observation data acquisition module is used to obtain the historical observation data sequence of each meteorological parameter in the meteorological parameter combination; the meteorological parameter combination is a combination of meteorological parameters that can be observed in the observation area.

[0076] The sample set building module is used to use the sliding window method to establish sample sets based on the historical observation data sequence of each meteorological parameter, so as to obtain the sample set corresponding to each meteorological parameter.

[0077] The model training module is used to train the neural network model using the sample set corresponding to each meteorological parameter to obtain the lightning warning neural network model corresponding to each meteorological parameter.

[0078] The weight determination module is used to use a genetic algorithm to determine the weight of the lightning forecast result output by the lightning warning neural network model corresponding to each meteorological parameter in each sub-meteorological parameter combination as the weight corresponding to each meteorological parameter in each sub-meteorological parameter combination; each sub-meteorological parameter combination is obtained by arranging and combining each meteorological parameter in the meteorological parameter combination.

[0079] The current observation data acquisition module is used to obtain the current observation data sequence of each meteorological parameter in the current meteorological parameter combination; the current meteorological parameter combination is a combination of meteorological parameters of the observation area that can be currently observed; wherein the current observation data sequence is a sequence of observation data in a sliding window containing the current observation data arranged in chronological order, and the sliding window is a sliding window used in the sliding window method.

[0080] The prediction module is used to input the current observation data sequence of each meteorological parameter in the current meteorological parameter combination into the lightning warning neural network model corresponding to each meteorological parameter in the current meteorological parameter combination, and obtain the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination.

[0081] The weight selection module is used to determine the weight of each meteorological parameter in the sub-meteorological parameter combination that is consistent with the type of meteorological parameters included in the current meteorological parameter combination as the weight of each meteorological parameter in the current meteorological parameter combination.

[0082] The comprehensive calculation module is used to calculate the final prediction result according to the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination and the weight corresponding to each meteorological parameter in the current meteorological parameter combination.

[0083] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a lightning warning method based on a weighted integrated neural network is implemented.

[0084] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0085] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0086] 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 computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0087] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0088] 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.

[0089] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A lightning warning method based on weighted integrated neural network, characterized in that: include: Acquire a historical observation data sequence of each meteorological parameter in a meteorological parameter combination; the meteorological parameter combination is a combination of meteorological parameters that can be observed in the observation area; Using the sliding window method, sample sets are established based on the historical observation data series of each meteorological parameter to obtain the sample sets corresponding to each meteorological parameter; The neural network model is trained respectively using the sample set corresponding to each meteorological parameter to obtain the lightning warning neural network model corresponding to each meteorological parameter; A genetic algorithm is used to determine the weight of the lightning forecast result output by the lightning warning neural network model corresponding to each meteorological parameter in each sub-meteorological parameter combination as the weight corresponding to each meteorological parameter in each sub-meteorological parameter combination; each sub-meteorological parameter combination is obtained by arranging and combining each meteorological parameter in the meteorological parameter combination; Obtaining a current observation data sequence of each meteorological parameter in a current meteorological parameter combination; the current meteorological parameter combination is a combination of meteorological parameters of an observation area that can be currently observed; Inputting the current observation data sequence of each meteorological parameter in the current meteorological parameter combination into the lightning warning neural network model corresponding to each meteorological parameter in the current meteorological parameter combination, respectively, to obtain the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination; Determine the weight of each meteorological parameter in the sub-meteorological parameter combination that is consistent with the type of meteorological parameters included in the current meteorological parameter combination as the weight of each meteorological parameter in the current meteorological parameter combination; The final prediction result is calculated based on the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination and the weight corresponding to each meteorological parameter in the current meteorological parameter combination.

2. The lightning warning method based on weighted integrated neural network according to claim 1 is characterized in that: Using the sliding window method, sample sets are established based on the historical observation data series of each meteorological parameter, and the sample sets corresponding to each meteorological parameter are obtained, including: Set the length of the sliding window of the kth meteorological parameter; k=1, 2, ..., K; K is the number of meteorological parameters in the meteorological parameter combination; A sliding window operation is performed on the historical observation data sequence of the kth meteorological parameter according to the length of the sliding window of the kth meteorological parameter, each historical observation data of the kth meteorological parameter in the sliding window is determined as a sample input, and the lightning probability of each time step in the prediction window is determined as a sample label to construct a sample set corresponding to the kth meteorological parameter; the prediction window is located after the sliding window.

3. The lightning warning method based on weighted integrated neural network according to claim 1 is characterized in that: The neural network model is obtained by improving the ConvLSTM model; The improvement method is to add a fully connected reduction layer and two multi-scale convolution layers at the input end of the ConvLSTM model, and add two multi-scale convolution layers and a fully connected magnification layer at the output end of the ConvLSTM model.

4. The lightning warning method based on weighted integrated neural network according to claim 3 is characterized in that: The fully connected reduction layer is used to compress the size of the input feature using the following formula; ； in, is the feature map output by the fully connected reduction layer, is the first feature map in the fully connected reduction layer input Line The observed values ​​of meteorological parameters of the characteristic points of the column, is the first feature map in the fully connected reduction layer input Line The weight parameter of the feature points of the column, and are the number of rows and columns of feature points in the feature graph input to the fully connected reduction layer, respectively. is the ReLU activation function.

5. The lightning warning method based on weighted integrated neural network according to claim 3 is characterized in that: The ConvLSTM model includes an encoding module and a decoding module; the encoding module and the decoding module each include two ConvLSTM layers.

6. The lightning warning method based on weighted integrated neural network according to claim 5 is characterized in that: The ConvLSTM layer includes: an input gate, a forget gate and an output gate; The formula of the input gate is: ； in, is the time step The output state of the input gate, is the convolution kernel of the input state of the input gate, is the time step The input state, is the convolution kernel of the hidden state of the input gate, is the time step The hidden state of is the bias of the input gate, is the Sigmod activation function, Represents the convolution operation; The formula of the forget gate is: ； in, is the time step The output state of the forget gate is is the convolution kernel of the input state of the forget gate, is the convolution kernel of the hidden state of the forget gate, is the bias of the forget gate; The formula for the output gate is: ； in, is the time step The output state of the output gate, is the convolution kernel of the input state of the output gate, is the convolution kernel of the hidden state of the output gate, is the bias of the output gate; The update formula for the hidden state of the ConvLSTM layer is: ； ； ； in, is the time step The hidden state of means that the matrix elements are multiplied one by one, is the time step The cell state, is the time step The cell state, is the time step Candidate cell states, and are the convolution kernels of the input state and hidden state of the hidden layer, is the bias of the hidden layer, is the tanh activation function.

7. The lightning warning method based on weighted integrated neural network according to claim 1 is characterized in that: The loss function for training the neural network model is: ； in, is the loss function, is the length of the prediction window, is the time step in the prediction window The error weight, is the total number of feature points, , and are the number of rows and columns of feature points in the target feature map, respectively. is the adjustment coefficient, and the target feature map is the marked time step The feature map or the time step of the neural network model output The feature map of The time step of the label The total number of lightning flashes in the feature graph of The time step of the label The feature map of Line The lightning probability of the characteristic points of the column, The time step output by the neural network model The feature map of Line The lightning probability of the characteristic points of the column.

8. The lightning warning method based on weighted integrated neural network according to claim 1 is characterized in that: The optimization goal in the process of using genetic algorithm to determine the weight of the lightning forecast result output by the lightning warning neural network model corresponding to each meteorological parameter in each sub-meteorological parameter combination is: ； in, is the weight vector, , , , , For the first, second, Type The weights of the lightning forecast results output by the lightning warning neural network model corresponding to the meteorological parameters, for dimensional set of real numbers, For the The first The meteorological parameters The loss function value of samples is For the The number of meteorological parameter types in each meteorological parameter combination. is the number of samples for each meteorological parameter.

9. A lightning warning device based on a weighted integrated neural network, characterized in that: The lightning warning device based on the weighted integrated neural network applies the lightning warning method based on the weighted integrated neural network according to any one of claims 1 to 8, and the lightning warning device based on the weighted integrated neural network comprises: A historical observation data acquisition module is used to acquire a historical observation data sequence of each meteorological parameter in a meteorological parameter combination; the meteorological parameter combination is a combination of meteorological parameters that can be observed in the observation area; A sample set building module is used to use a sliding window method to build sample sets based on the historical observation data sequence of each meteorological parameter, so as to obtain a sample set corresponding to each meteorological parameter; A model training module is used to train a neural network model using a sample set corresponding to each meteorological parameter to obtain a lightning warning neural network model corresponding to each meteorological parameter; A weight determination module is used to determine the weight of the lightning forecast result output by the lightning warning neural network model corresponding to each meteorological parameter in each sub-meteorological parameter combination by using a genetic algorithm as the weight corresponding to each meteorological parameter in each sub-meteorological parameter combination; each sub-meteorological parameter combination is obtained by arranging and combining each meteorological parameter in the meteorological parameter combination; A current observation data acquisition module is used to acquire a current observation data sequence of each meteorological parameter in a current meteorological parameter combination; the current meteorological parameter combination is a combination of meteorological parameters of an observation area that can be currently observed; A prediction module is used to input the current observation data sequence of each meteorological parameter in the current meteorological parameter combination into the lightning warning neural network model corresponding to each meteorological parameter in the current meteorological parameter combination, and obtain the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination; A weight selection module is used to determine the weight of each meteorological parameter in the sub-meteorological parameter combination that is consistent with the type of meteorological parameters included in the current meteorological parameter combination as the weight of each meteorological parameter in the current meteorological parameter combination; The comprehensive calculation module is used to calculate the final prediction result according to the lightning forecast result corresponding to each meteorological parameter in the current meteorological parameter combination and the weight corresponding to each meteorological parameter in the current meteorological parameter combination.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the lightning warning method based on a weighted integrated neural network according to any one of claims 1 to 8.

Images