A sparrow optimization lightning early warning method and system based on modal decomposition

Abstract

This invention provides a sparrow-optimized lightning early warning method and system based on mode decomposition, relating to the field of lightning early warning technology. The sparrow-optimized lightning early warning method based on mode decomposition includes: decomposing the atmospheric electric field time series to obtain IMF components and residual terms; calculating the sample entropy value of each IMF component; and classifying and reconstructing based on the sample entropy value of each IMF component to obtain high-frequency signals, low-frequency signals, and trend signals; optimizing the hyperparameters of a preset LSTM prediction model to obtain the optimal parameter set; dividing the atmospheric electric field time series into training and testing sets; and inputting the high-frequency, low-frequency, and trend signals into the trained LSTM prediction model for prediction processing. This method addresses the problem that atmospheric electric fields during thunderstorms often differ across regions, leading to varying threshold selections and resulting in poor universality, low accuracy, and weak effectiveness in early warning methods.

Description

Technical Field

[0001] This invention relates to the field of lightning warning technology, specifically to a method and system for optimal sparrow lightning warning based on mode decomposition. Background Technology

[0002] Lightning disasters are listed by the United Nations as one of the ten most serious natural disasters, and the damage they cause is commonplace in our lives. The flight of aircraft and missiles is often affected by thunderstorms; electromagnetic pulses generated during thunderstorms can interfere with wireless communication; buildings and overhead power lines are frequently struck by lightning; lightning strikes on trees cause forest fires, leading to increased air pollution, global warming, and a series of other problems. Therefore, providing early warnings of lightning to minimize its damage and losses has become a societal imperative. Atmospheric electric field strength is a fundamental parameter in atmospheric physics and atmospheric electricity. The time series of atmospheric electric fields reflects the entire process of charge formation, accumulation, and discharge in thunderstorm clouds. Thunderstorm activity often causes changes in the ground electric field. Analyzing the intrinsic characteristics of the electric field under different weather conditions is crucial for lightning early warning. Utilizing atmospheric electric field strength for lightning early warning has become a method; therefore, studying atmospheric electric field strength and issuing lightning warnings before thunderstorms arrive is of great practical significance for social life and production safety.

[0003] Currently, most lightning warnings both domestically and internationally are based on data collected by Doppler weather radar, lightning locators, and atmospheric electric field meters. Some researchers have also combined these types of data to improve the effectiveness of lightning warnings. The traditional threshold method has been used in the field of lightning warnings for a long time, but the atmospheric electric field during thunderstorms often varies in different regions, so the selection of thresholds often differs, resulting in problems such as poor universality, low accuracy, and weak effectiveness of the warning method. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for sparrow optimization-based lightning early warning, which solves the problem that atmospheric electric fields during thunderstorms often differ in different regions, resulting in varying threshold selections and thus causing early warning methods to suffer from poor universality, low accuracy, and weak effectiveness.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] Firstly, a sparrow optimization-based lightning early warning method based on mode decomposition is provided, including:

[0009] The atmospheric electric field time series is decomposed to obtain IMF components and residual terms. The sample entropy value of each IMF component is calculated, and the sample entropy value of each IMF component is classified and reconstructed to obtain high-frequency signals, low-frequency signals and trend signals.

[0010] The hyperparameters of the preset LSTM prediction model are optimized to obtain the optimal parameter set;

[0011] The atmospheric electric field time series is divided into a training set and a test set. The samples in the training set are input into the preset LSTM prediction model for model training, and the trained LSTM prediction model is verified through the test set.

[0012] High-frequency signals, low-frequency signals, and trend signals are input into the trained LSTM prediction model for prediction processing. The prediction results are summed to obtain the predicted time series of atmospheric electric field time series. Lightning warnings are then issued based on the predicted time series.

[0013] Preferably, the atmospheric electric field time series is obtained in the following way:

[0014] Receive test data from the target area;

[0015] The atmospheric electric field time series of the target area was determined based on the test data.

[0016] Preferably, the test data includes atmospheric electric field strength data and lightning location data.

[0017] Preferably, the atmospheric electric field time series x i (t) is decomposed, specifically including the following steps:

[0018] S13: Construct a new noisy sequence, using the following formula:

[0019] x i (t)=x(t)+w i (t)

[0020] In the formula, x(t) is the time series of the atmospheric electric field, and w i (t) represents the i-th white noise signal that follows a standard normal distribution;

[0021] S13: Perform EMD decomposition to obtain the first intrinsic mode component:

[0022]

[0023] S13: Calculate the first surplus:

[0024]

[0025] S14: Calculate the k-th residual and the (k+1)-th eigenmode component:

[0026]

[0027]

[0028] σ1 is the initial difference of white noise;

[0029] S15: Repeat step S14 until the remaining amount cannot be further decomposed, then terminate. The final residual output is:

[0030]

[0031] S16: The final expression for x(t) is:

[0032]

[0033] Preferably, the step of calculating the sample entropy value of each IMF component and classifying and reconstructing the sample entropy value of each IMF component to obtain high-frequency signals, low-frequency signals, and trend signals specifically includes:

[0034] Calculate the sample entropy of each IMF component and trend signal, and use this as the basis for classification and reconstruction:

[0035] SampEn(m, r, N) = -ln[A m (r) / B m (r)]}

[0036] The sample entropy values ​​of each IMF component and trend signal of the atmospheric electric field time series are obtained, and then classified and reconstructed into high-frequency signals, low-frequency signals and trend signals.

[0037] Preferably, the step of optimizing the hyperparameters of the preset LSTM prediction model to obtain the optimal parameter set includes the following steps:

[0038] S21: Use the sparrow optimization algorithm to optimize the hyperparameter set of the preset LSTM prediction model. Set the sparrow population to consist of discoverers and joiners. Assume the number of sparrows in the population is n, and the dimension of the variable to be optimized is d. The population representation is as follows:

[0039]

[0040] S22: The discoverer's responsibility is to provide foraging locations for the population. The discoverer's location update strategy is as follows:

[0041]

[0042] In the formula, t is the current iteration number, and item maxX represents the maximum number of iterations; the position information of the i-th sparrow in the j-th dimension is represented as X. i,j The size of R2 and ST is compared to update the discoverer's location information.

[0043] S23: The positions of newcomers have been updated as follows:

[0044]

[0045] The globally optimal position X occupied by the discoverer p And the worst position X in the world wrost This determines the changes in the location information of the participants.

[0046] S24: In addition, 10% to 20% of sparrows in the population have danger recognition capabilities. Their initial positions are randomly generated, and the update strategy is as follows:

[0047]

[0048] S25: Calculate the population fitness and determine whether the globally optimal sparrow has been found. If not, return to step S22 and continue iterating.

[0049] S26: The pre-set LSTM prediction model is first generated by the forget gate f t To filter out information to be forgotten, input the current time as x. t The state h at the previous moment t-1 Substitute into the sigmoid activation function σ:

[0050] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0051] S27: Then enter door i t The sigmoid layer updates the information again and multiplies it with the tanh layer to perform joint forgetting and retention of information; the calculation rules are as follows:

[0052] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0053]

[0054] S28: Then, the information state of the previous layer is multiplied point by point with the forgetting vector, and the resulting value is added point by point to the output value of the input gate. At this point, the update of the information state is completed; the calculation rules are as follows:

[0055]

[0056] S29: Finally, output gate o t The new information state sequence {h0, h1, ..., h} is... t-1 Pass it to the next time step:

[0057] o t =σ(W o [h t-1 ,x t ]+b o )

[0058] h t =o t *tanh(C t )

[0059] In the formula, W f W i W o The weights of the forget gate, input gate, and output gate, respectively, b f b i b o The bias vectors for the forget gate, input gate, and output gate, respectively.

[0060] Preferably, the validation of the trained LSTM prediction model using a test set includes:

[0061] The mean absolute percentage error and root mean square error on the test set are calculated to verify the performance of the trained LSTM prediction model.

[0062] Secondly, a lightning early warning system based on mode decomposition is provided, the system comprising the following modules:

[0063] The preprocessing module is used to decompose the atmospheric electric field time series to obtain IMF components and residual terms, calculate the sample entropy value of each IMF component, and classify and reconstruct the sample entropy value of each IMF component to obtain high-frequency signals, low-frequency signals and trend signals.

[0064] The processing module optimizes the hyperparameters of the preset LSTM prediction model to obtain the optimal parameter set.

[0065] The training and testing module divides the atmospheric electric field time series into a training set and a test set. The samples in the training set are input into the preset LSTM prediction model for model training, and the trained LSTM prediction model is verified through the test set.

[0066] The early warning module is used to input high-frequency signals, low-frequency signals and trend signals into the trained LSTM prediction model for prediction processing, sum the prediction results to obtain the predicted time series of atmospheric electric field, and issue lightning warnings based on the predicted time series.

[0067] Thirdly, a computer-readable storage medium is provided for storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any of the methods described.

[0068] Fourthly, a computing device is provided, comprising:

[0069] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described.

[0070] (III) Beneficial Effects

[0071] This invention presents a sparrow-optimized lightning early warning method and system based on mode decomposition. It preprocesses the original atmospheric electric field time series to effectively extract the intrinsic mode components of the time series, while also achieving a certain degree of denoising. The entropy of the decomposed subsequence samples is calculated, and the reconstructed signal based on this calculation effectively represents the characteristics of the original sequence, while reducing the complexity of LSTM modeling. Using LSTM as the prediction model, the hyperparameters of LSTM are optimized through the SSA optimization algorithm, effectively improving the model's prediction accuracy and enhancing network detection performance. Experimental results show that, based on the mean absolute percentage error (MAPE), the CEEMDAN-SSA-LSTM model has the lowest result (0.1246), which is 48.9% lower than using only EEMD-LSTM modeling. The root mean square error (RMSE) is reduced by 61.8%, and the mean absolute error (MAE) is reduced by 62%. Furthermore, the determination coefficient (R²) of the model constructed in this paper is 0.9883, closest to 1, indicating a good model fit. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of the sparrow optimization lightning early warning method based on mode decomposition of the present invention;

[0073] Figure 2 This is a diagram showing the CEEMDAN decomposition results of the atmospheric electric field time series in an embodiment of the present invention.

[0074] Figure 3 This is a graph showing the high-frequency signal, low-frequency signal, and trend signal after classification and reconstruction in an embodiment of the present invention;

[0075] Figure 4 This is a graph showing the high-frequency sequence prediction results in an embodiment of the present invention;

[0076] Figure 5 This is a graph showing the prediction results of low-frequency sequences in an embodiment of the present invention;

[0077] Figure 6 This is a graph showing the trend prediction results in an embodiment of the present invention;

[0078] Figure 7 This is a comparison chart of prediction results from different models in an embodiment of the present invention. Detailed Implementation

[0079] The technical solutions in the embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0080] Example

[0081] like Figure 1 As shown, one embodiment of the present invention provides a sparrow optimization lightning early warning method based on mode decomposition, including:

[0082] The atmospheric electric field time series is decomposed to obtain IMF components and residual terms. The sample entropy value of each IMF component is calculated, and the sample entropy value of each IMF component is classified and reconstructed to obtain high-frequency signals, low-frequency signals and trend signals.

[0083] The hyperparameters of the preset LSTM prediction model are optimized to obtain the optimal parameter set. The optimal parameter set is defined according to the loss function, that is, the loss function curve converges after a certain number of iterations and the value decreases to infinitely close to zero.

[0084] The atmospheric electric field time series is divided into a training set and a test set. The samples in the training set are input into the preset LSTM prediction model for model training, and the trained LSTM prediction model is verified through the test set.

[0085] High-frequency signals, low-frequency signals, and trend signals are input into the trained LSTM prediction model for prediction processing. The prediction results are summed to obtain the predicted time series of atmospheric electric field time series. Lightning warnings are then issued based on the predicted time series.

[0086] Furthermore, the atmospheric electric field time series is obtained as follows:

[0087] Receive test data from the target area;

[0088] The atmospheric electric field time series of the target area was determined based on the test data.

[0089] Furthermore, the test data includes atmospheric electric field strength data and lightning location data.

[0090] Furthermore, the atmospheric electric field time series x i (t) is decomposed, specifically including the following steps:

[0091] S13: Construct a new noisy sequence, using the following formula:

[0092] x i (t)=x(t)+w i (t)

[0093] In the formula, x(t) is the time series of the atmospheric electric field, and w i (t) represents the i-th white noise signal that follows a standard normal distribution;

[0094] S13: Perform EMD decomposition to obtain the first intrinsic mode component:

[0095]

[0096] S13: Calculate the first surplus:

[0097]

[0098] S14: Calculate the k-th residual and the (k+1)-th eigenmode component:

[0099]

[0100]

[0101] σ1 is the initial difference of white noise;

[0102] S15: Repeat step S14. The process terminates when the residual cannot be further decomposed. The stopping condition is whether the amplitude range of the residual is within a predetermined threshold range. That is, if it is less than the threshold, the amplitude change is very small and sufficiently stable, so there is no need to continue decomposing. The final residual output is:

[0103]

[0104] S16: The final expression for x(t) is:

[0105]

[0106] Furthermore, the sample entropy value of each IMF component is calculated, and the sample entropy value of each IMF component is classified and reconstructed to obtain high-frequency signals, low-frequency signals, and trend signals, specifically including:

[0107] Calculate the sample entropy of each IMF component and trend signal, and use this as the basis for classification and reconstruction:

[0108] SampEn(m, r, N) = -ln[A m (r) / B m (r)]}

[0109] The sample entropy values ​​of each IMF component and trend signal of the atmospheric electric field time series are obtained, and then classified and reconstructed into high-frequency signals, low-frequency signals and trend signals.

[0110] Furthermore, the hyperparameters of the preset LSTM prediction model are optimized to obtain the optimal parameter set, including the following steps:

[0111] S21: Use the sparrow optimization algorithm to optimize the hyperparameter set of the preset LSTM prediction model. Set the sparrow population to consist of discoverers and joiners. Assume the number of sparrows in the population is n, and the dimension of the variable to be optimized is d. The population representation is as follows:

[0112]

[0113] S22: The discoverer's responsibility is to provide foraging locations for the population. The discoverer's location update strategy is as follows:

[0114]

[0115] In the formula, t is the current iteration number, and item max X represents the maximum number of iterations; the position information of the i-th sparrow in the j-th dimension is represented as X. i,j The size of R2 and ST is compared to update the discoverer's location information.

[0116] S23: The positions of newcomers have been updated as follows:

[0117]

[0118] The globally optimal position X occupied by the discoverer p And the worst position X in the world wrost This determines the changes in the location information of the participants.

[0119] S24: In addition, 10% to 20% of sparrows in the population have danger recognition capabilities. Their initial positions are randomly generated, and the update strategy is as follows:

[0120]

[0121] S25: Calculate the population fitness and determine whether the globally optimal sparrow has been found. If not, return to step S22 and continue iterating.

[0122] S26: The pre-set LSTM prediction model is first generated by the forget gate f t To filter out information to be forgotten, input the current time as x. t The state h at the previous moment t-1 Substitute into the sigmoid activation function σ:

[0123] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0124] S27: Then enter door i t The sigmoid layer updates the information again and multiplies it with the tanh layer to perform joint forgetting and retention of information; the calculation rules are as follows:

[0125] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0126]

[0127] S28: Then, the information state of the previous layer is multiplied point by point with the forgetting vector, and the resulting value is added point by point to the output value of the input gate. At this point, the update of the information state is completed; the calculation rules are as follows:

[0128]

[0129] S29: Finally, output gate o t The new information state sequence {h0, h1, ..., h} is... t-1 Pass it to the next time step:

[0130] o t =σ(W o [h t-1 ,x t ]+b o )

[0131] h t =o t *tanh(C t )

[0132] In the formula, W f W i W o The weights of the forget gate, input gate, and output gate, respectively, b f b i b oThe bias vectors for the forget gate, input gate, and output gate, respectively.

[0133] Furthermore, the trained LSTM prediction model is validated using a test set, including:

[0134] The mean absolute percentage error and root mean square error on the test set are calculated to verify the performance of the trained LSTM prediction model.

[0135] Wherein, IMF stands for Intrinsic Mode Function and LSTM stands for Long Short-Term Memory Network.

[0136] This paper uses atmospheric electric field data and corresponding lightning location data. The predicted data is based on electric field data recorded by an atmospheric electric field meter in a specific region from June to November 2009, with time units accurate to the second. The effective lightning recording monitoring radius is 20 km. For thunderstorm weather samples, the lifecycle of a thunderstorm includes an approach phase, an arrival phase, and a dissipation phase, lasting approximately 20 to 40 minutes. Therefore, the duration of the thunderstorm weather sample sequence was chosen to be one hour. For non-thunderstorm weather samples, electric field data of equal length were randomly selected from the sample library. If no thunderstorm occurred within one hour before or after the sample, it was classified as a clear weather sample. Three thunderstorm weather samples and three clear weather samples of equal length were randomly selected, forming a total of 6 × 3600 weather sample libraries.

[0137] The CEEMDAN algorithm is used to decompose the electric field time series, which can effectively obtain the main characteristic terms and trend terms of the electric field time series, and can effectively suppress noise interference. The results after decomposition by the CEEMDAN algorithm are as follows: Figure 2 As shown, IMF1 to IMF11 represent the signal distribution from high frequency to low frequency, and their variation characteristics represent detailed information about the formation, discharge, and dissipation of a thunderstorm cloud. Res is the trend term, which continuously increases from -2.13 to 2.1. The fluctuation amplitude of the electric field time series samples also shows a continuously decreasing trend, thus verifying the characterization role of the trend term. The reconstruction dimension m is defined as 2, the threshold r as 0.2, and sampling is not repeated. The sample entropy value of each IMF other than the trend term is calculated. The reconstructed sequence is as follows: Figure 3 .

[0138] Figure 4-6 The image shows the prediction results for each reconstructed sequence. It can be seen that the prediction results are good for low-frequency and trend terms. The prediction curve is obtained by summing the subsequences.

[0139] Figure 7A comparison of prediction results from different models is shown in the figure. The solid line represents the true value curve, and other curves have corresponding captions. It can be seen that the prediction performance of the MLP-based model is poor, with a significant deviation between the predicted and true value curves. The EEMD-LSTM model without SSA hyperparameter optimization also fails to achieve a good fit. Both models suffer from prediction lag and underfitting. In contrast, the model presented in this paper demonstrates superior prediction performance, better fit, and higher prediction accuracy. Compared to using only the EEMD-LSTM model, the MAPE is reduced by 48.9%, the RMSE by 61.8%, and the MAE by 62%. Furthermore, the coefficient of determination R of the model presented in this paper is significantly lower. 2 The value is 0.9883, which is closest to 1, indicating that the model fits well.

[0140] Another embodiment of the present invention provides a lightning early warning system based on mode decomposition, the system comprising the following modules:

[0141] The preprocessing module is used to decompose the atmospheric electric field time series to obtain IMF components and residual terms, calculate the sample entropy value of each IMF component, and classify and reconstruct the sample entropy value of each IMF component to obtain high-frequency signals, low-frequency signals and trend signals.

[0142] The processing module optimizes the hyperparameters of the preset LSTM prediction model to obtain the optimal parameter set.

[0143] The training and testing module divides the atmospheric electric field time series into a training set and a test set. The samples in the training set are input into the preset LSTM prediction model for model training, and the trained LSTM prediction model is verified through the test set.

[0144] The early warning module is used to input high-frequency signals, low-frequency signals and trend signals into the trained LSTM prediction model for prediction processing, sum the prediction results to obtain the predicted time series of atmospheric electric field, and issue lightning warnings based on the predicted time series.

[0145] Embodiments of this application may be provided as methods or computer program products. Therefore, this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this 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 code. The solutions in the embodiments of this application may be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0146] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0147] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0148] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0149] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A sparrow optimization-based lightning early warning method based on mode decomposition, characterized in that, include: The atmospheric electric field time series is decomposed to obtain IMF components and residual terms. The sample entropy value of each IMF component is calculated, and the sample entropy value of each IMF component is classified and reconstructed to obtain high-frequency signals, low-frequency signals and trend signals. The hyperparameters of the preset LSTM prediction model are optimized to obtain the optimal parameter set; The atmospheric electric field time series is divided into a training set and a test set. The samples in the training set are input into the preset LSTM prediction model for model training, and the trained LSTM prediction model is verified through the test set. High-frequency signals, low-frequency signals, and trend signals are input into the trained LSTM prediction model for prediction processing. The prediction results are summed to obtain the predicted time series of atmospheric electric field time series. Lightning warnings are then issued based on the predicted time series. The process of optimizing the hyperparameters of the preset LSTM prediction model to obtain the optimal parameter set includes the following steps: S21: Use the sparrow optimization algorithm to optimize the hyperparameter set of the preset LSTM prediction model. Set the sparrow population to consist of discoverers and joiners. Assume the number of sparrows in the population is n, and the dimension of the variable to be optimized is d. The population representation is as follows: S22: The discoverer's responsibility is to provide foraging locations for the population. The discoverer's location update strategy is as follows: In the formula, t is the current iteration number, and item max X represents the maximum number of iterations; the position information of the i-th sparrow in the j-th dimension is represented as X. i,j The size of R2 and ST is compared to update the discoverer's location information. S23: The positions of newcomers have been updated as follows: The globally optimal position X occupied by the discoverer p And the worst position X in the world wrost This determines the changes in the location information of the participants; S24: In addition, 10% to 20% of sparrows in the population have danger recognition capabilities. Their initial positions are randomly generated, and the update strategy is as follows: S25: Calculate the population fitness and determine whether the globally optimal sparrow has been found. If not, return to step S22 and continue iterating. S26: The pre-set LSTM prediction model is first generated by the forget gate f t To filter out information to be forgotten, input the current time as x. t The state h at the previous moment t-1 Substitute into the sigmoid activation function σ: S27: Then enter door i t The sigmoid layer updates the information again and multiplies it with the tanh layer to perform joint forgetting and retention of information; the calculation rules are as follows: i t =σ(W i ·[h t-1 ,x t ]+b i ) S28: Then, the information state of the previous layer is multiplied point by point with the forgetting vector, and the resulting value is added point by point to the output value of the input gate. At this point, the update of the information state is completed; the calculation rules are as follows: S29: Finally, output gate o t The new information state sequence {h0, h1, ..., h} is... t-1 Pass it to the next time step: the t =σ(W o [h t-1 ,x t ]+b o ) h t =o t *fishy(C) t ) In the formula, W f W i W o The weights of the forget gate, input gate, and output gate, respectively, b f b i b o The bias vectors for the forget gate, input gate, and output gate, respectively.

2. The sparrow optimization lightning early warning method based on mode decomposition according to claim 1, characterized in that: The atmospheric electric field time series was obtained as follows: Receive test data from the target area; The atmospheric electric field time series of the target area was determined based on the test data.

3. The sparrow optimization lightning early warning method based on mode decomposition according to claim 2, characterized in that: The test data includes atmospheric electric field strength data and lightning location data.

4. The sparrow optimization lightning early warning method based on mode decomposition according to claim 3, characterized in that: The atmospheric electric field time series x i (t) is decomposed, specifically including the following steps: S13: Construct a new noisy sequence, using the following formula: x i (t)=x(t)+w i (t) In the formula, x(t) is the time series of the atmospheric electric field, and w i (t) represents the i-th white noise signal that follows a standard normal distribution; S13: Perform EMD decomposition to obtain the first intrinsic mode component: S13: Calculate the first surplus: S14: Calculate the k-th residual and the (k+1)-th eigenmode component: σ1 is the initial difference of white noise; S15: Repeat step S14 until the remaining amount cannot be further decomposed, then terminate. The final residual output is: S16: The final expression for x(t) is:

5. The sparrow optimization lightning early warning method based on mode decomposition according to claim 4, characterized in that: The calculation of the sample entropy value of each IMF component, and the classification and reconstruction of the sample entropy value of each IMF component to obtain high-frequency signals, low-frequency signals, and trend signals, specifically includes: Calculate the sample entropy of each IMF component and trend signal, and use this as the basis for classification and reconstruction: SampEn(m,r,N)z-ln[A m (r) / B m (r)]} The sample entropy values ​​of each IMF component and trend signal of the atmospheric electric field time series are obtained, and then classified and reconstructed into high-frequency signals, low-frequency signals and trend signals.

6. The sparrow optimization lightning early warning method based on mode decomposition according to claim 5, characterized in that: The validation of the trained LSTM prediction model using a test set includes: The mean absolute percentage error and root mean square error on the test set are calculated to verify the performance of the trained LSTM prediction model.

7. A system for executing the sparrow optimization lightning early warning method based on mode decomposition as described in any one of claims 1-6, characterized in that, The system includes the following modules: The preprocessing module is used to decompose the atmospheric electric field time series to obtain IMF components and residual terms, calculate the sample entropy value of each IMF component, and classify and reconstruct the sample entropy value of each IMF component to obtain high-frequency signals, low-frequency signals and trend signals. The processing module optimizes the hyperparameters of the preset LSTM prediction model to obtain the optimal parameter set. The training and testing module divides the atmospheric electric field time series into a training set and a test set. The samples in the training set are input into the preset LSTM prediction model for model training, and the trained LSTM prediction model is verified through the test set. The early warning module is used to input high-frequency signals, low-frequency signals and trend signals into the trained LSTM prediction model for prediction processing, sum the prediction results to obtain the predicted time series of atmospheric electric field, and issue lightning warnings based on the predicted time series.

8. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods according to claims 1-6.

9. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods according to claims 1-6.

Images