Training method, device and equipment of lightning ignition prediction model and medium

Abstract

The invention relates to a training method, device and equipment of a lightning ignition prediction model and a medium. Comprising the steps that multiple sets of training sample sets of a target area in a historical time period are acquired, the data size difference between non-ignition case data and ignition case data contained in each set of training sample sets is small, and each set of training sample sets comprises environment data and classification labels of whether a fire occurs or not; performing iterative training on the preset model by utilizing the environment data and the classification labels in each group of training sample sets until the accuracy of the preset model under the current iteration frequency is the highest, and determining the preset model under the current iteration frequency as a candidate model; and from candidate models determined based on the multiple groups of training sample sets, obtaining a model with the highest accuracy rate as a lightning fire ignition prediction model of the target area. Thus, the problem that the number of the ignition cases and the number of the non-ignition cases are unbalanced is solved, and the method can be suitable for predicting the ignition result of the lightning fire in an area with complex environment data.

Description

Technical Field

[0001] The present disclosure relates to the technical field of lightning fire prediction, and in particular to a training method, device, equipment and medium for a lightning fire ignition prediction model. Background Art

[0002] In recent years, lightning fires have occurred frequently and burned a large area, causing great ecological damage and greenhouse gas emissions. Therefore, in order to reduce losses and promote early fire fighting, it is necessary to predict the results of lightning fire ignition.

[0003] Since lightning fire ignition involves complex environmental data, and the cases of lightning fire ignition and non-ignition are unbalanced, the research on lightning fire ignition prediction faces great difficulties. Therefore, it is urgent to provide a lightning fire ignition prediction method to accurately predict the results of lightning fire ignition based on complex environmental data and unbalanced cases. Summary of the invention

[0004] In order to solve the above technical problems, the present disclosure provides a training method, device, equipment and medium for a lightning fire ignition prediction model.

[0005] In a first aspect, the present disclosure provides a method for training a lightning fire ignition prediction model, comprising:

[0006] Acquire multiple sets of training sample sets in the target area within a historical time period, wherein the difference between the amount of non-ignition case data and the amount of ignition case data contained in each set of training sample sets is less than a first preset threshold, and the ignition case data and the non-ignition case data in each set of training sample sets both contain environmental data and a classification label of whether a fire has occurred;

[0007] Iteratively train the preset model using the environmental data and the classification label of whether a fire has occurred contained in each group of training sample sets until the accuracy of the preset model at the current iteration number is the highest, and determine the preset model at the current iteration number as the candidate model;

[0008] From the candidate models determined based on the multiple groups of training sample sets, the model with the highest accuracy is obtained as the lightning fire ignition prediction model for the target area.

[0009] In a second aspect, the present disclosure provides a training device for a lightning fire ignition prediction model, comprising:

[0010] A training sample set acquisition module, used to acquire multiple sets of training sample sets in a target area within a historical time period, wherein the difference between the amount of non-ignition case data and the amount of ignition case data contained in each set of training sample sets is less than a first preset threshold, and the ignition case data and the non-ignition case data in each set of training sample sets both contain environmental data and a classification label of whether a fire has occurred;

[0011] An iterative training module, used to iteratively train the preset model using the environmental data and the classification label of whether a fire has occurred contained in each group of training sample sets, until the accuracy of the preset model at the current iteration number is the highest, and determine the preset model at the current iteration number as the candidate model;

[0012] The model determination module is used to obtain the model with the highest accuracy from the candidate models determined based on multiple groups of training sample sets as the lightning fire ignition prediction model for the target area.

[0013] In a third aspect, an embodiment of the present disclosure further provides an electronic device, the device comprising:

[0014] one or more processors;

[0015] a storage device for storing one or more programs,

[0016] When one or more programs are executed by one or more processors, the one or more processors implement the method provided in the first aspect.

[0017] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method provided in the first aspect is implemented.

[0018] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has the following advantages:

[0019] A training method, device, equipment and medium for a lightning fire ignition prediction model according to an embodiment of the present disclosure include: obtaining multiple groups of training sample sets in a target area within a historical time period, wherein the difference in data volume between non-ignition case data and ignition case data contained in each group of training sample sets is less than a first preset threshold, and the ignition case data and non-ignition case data in each group of training sample sets both contain environmental data and classification labels for whether a fire has occurred; iteratively training a preset model using the environmental data and classification labels for whether a fire has occurred contained in each group of training sample sets until the accuracy of the preset model at the current iteration number is the highest, and determining the preset model at the current iteration number as a candidate model; and obtaining the model with the highest accuracy from the candidate models determined based on the multiple groups of training sample sets as the lightning fire ignition prediction model for the target area. In this way, by using multiple groups of training sample sets containing ignition cases and non-ignition cases with similar data amounts, the lightning fire ignition prediction model with the highest accuracy is obtained through iterative training, which overcomes the problem of imbalance in the number of ignition cases and non-ignition cases. In addition, since the model training process actually processes the complex environmental data that cause ignition of fires in multiple groups of training sample sets, in areas with complex environmental data, the trained lightning fire ignition prediction model can also accurately predict the ignition results of lightning fires. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0022] Figure 1 A flowchart of a method for training a lightning fire ignition prediction model provided by an embodiment of the present disclosure;

[0023] Figure 2 A schematic diagram of the structure of a training device for a lightning fire ignition prediction model provided by an embodiment of the present disclosure;

[0024] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0027] In order to accurately predict the ignition result of lightning fire, the following Figure 1 The training method of the lightning fire ignition prediction model provided in the embodiment of the present disclosure is described. In the embodiment of the present disclosure, the training method of the lightning fire ignition prediction model can be executed by an electronic device or a server. The electronic device may include a device with communication functions such as a tablet computer, a desktop computer, a laptop computer, and may also include a device simulated by a virtual machine or a simulator. The server may be a cloud server or a server cluster. Below, the sample data synthesis method is specifically explained with the electronic device as the execution subject.

[0028] Figure 1 A flow chart of a method for training a lightning fire ignition prediction model provided by an embodiment of the present disclosure is shown.

[0029] like Figure 1 As shown, the training method of the lightning fire ignition prediction model may include the following steps.

[0030] S110. Obtain multiple groups of training sample sets in the target area within a historical time period, wherein the difference in the amount of non-ignition case data and ignition case data contained in each group of training sample sets is less than a first preset threshold, and the ignition case data and non-ignition case data in each group of training sample sets both contain environmental data and a classification label for whether a fire has occurred.

[0031] In this embodiment, an area where fires ignited by lightning strikes frequently is taken as a target area. The electronic device obtains ignition case data and non-ignition case data in the target area within a historical time period, and divides the ignition case data and non-ignition case data into multiple groups of training sample sets. The difference between the amount of non-ignition case data and the amount of non-ignition case data contained in each group of training sample sets is less than a first preset threshold, so that the ignition cases and non-ignition cases in each group of training sample sets remain balanced.

[0032] Optionally, the first preset threshold may be a smaller value preset based on experience.

[0033] In addition, each ignition case data and non-ignition case data includes environmental data and a classification label of whether a fire has occurred. Optionally, the environmental data includes one or more combinations of the following: weather characteristics, soil moisture, lightning fire data that caused a fire, lightning fire data that did not cause a fire, no lightning fire data, vegetation characteristics, and terrain characteristics.

[0034] In specific implementation, the data volume of lightning fire data that caused fires in each group of training sample sets is, for example, 162, the data volume of lightning fire data that did not cause fires is, for example, 162, and the data volume of no lightning fire is, for example, 81. In this case, the difference in data volume between non-ignition case data and ignition case data is still less than the first preset threshold, so that the ignition cases and non-ignition cases in each group of training sample sets are balanced.

[0035] Optionally, the weather characteristics further include temperature, precipitation, relative humidity, wind speed and other characteristics. The lightning fire data that caused the fire and the lightning fire data that did not cause the fire further include peak current, polarity and lightning density and other characteristics. The vegetation characteristics further include broadleaf tree coverage and coniferous tree coverage and other characteristics; the terrain characteristics further include altitude, slope and slope aspect and other characteristics.

[0036] Optionally, the classification label can be pre-calibrated based on experience, for example, if a fire has occurred, the classification label is 1, and if a fire has not occurred, the classification label is 0.

[0037] It can be understood that, since the environmental data contained in the ignition case data and the non-ignition case data further include complex environmental features, each set of training sample sets can accommodate complex environmental features while maintaining balance.

[0038] In this embodiment, the specific implementation method of S110 includes but is not limited to the following methods: obtaining a first number of ignition case data and a second number of non-ignition case data from historical environmental data in a historical time period, wherein the second number is greater than the first number, and the difference between the second number and the first number is greater than a second preset threshold; randomly extracting a third number of non-ignition case data from the second number of non-ignition case data multiple times, wherein the difference between the second number and the third number of data is less than a first preset threshold; using the first number of ignition case data and the third number of first non-ignition case data randomly extracted each time to form a training sample set.

[0039] Specifically, in the face of a large amount of historical environmental data in a historical time period, the electronic device obtains a first number of ignition case data and a second number of non-ignition case data without considering the balance. Since the number of ignitions of lightning fires is much less than the number of non-ignitions, the second number is much larger than the first number, resulting in an imbalance between the first number of ignition case data and the second number of non-ignition case data. In order to solve the imbalance problem of ignition cases and non-ignition cases, the electronic device adopts a multiple random under-sampling method to randomly extract a third number of non-ignition case data from the second number of non-ignition case data multiple times, and then uses the first number of ignition case data and the third number of non-ignition case data randomly extracted each time to form each training sample set, so that the ignition case data and non-ignition case data in each training sample set remain balanced.

[0040] Optionally, the second preset threshold may be a maximum value preset based on experience.

[0041] Considering that highly correlated environmental data in the ignition case data and non-ignition case data will affect the accuracy and efficiency of model training. In order to solve this problem, the first number of ignition case data and the third number of non-ignition case data randomly drawn each time are used to form each group of training sample sets. The specific implementation method includes but is not limited to the following methods: feature correlation analysis is performed on the third number of non-ignition case data randomly drawn each time and the first number of ignition case data to determine the feature correlation; from the third number of non-ignition case data randomly drawn each time and the first number of ignition case data, environmental data with a feature correlation greater than a preset correlation threshold is deleted; each group of training sample sets is formed by deleting the third number of non-ignition case data and the first number of ignition case data with a feature correlation greater than the preset correlation threshold.

[0042] Specifically, for the third number of non-ignition case data and the first number of ignition case data randomly drawn each time, the electronic device can use methods such as Spearman correlation analysis to perform feature correlation analysis to determine the feature correlation between different environmental data. If the feature correlation is high, it means that the different environmental data are highly correlated. This highly correlated different environmental data will affect accuracy and efficiency. The electronic device needs to delete the environmental data whose feature correlation is greater than a preset correlation threshold from the third number of non-ignition case data and the first number of ignition case data randomly drawn each time, and then use the remaining data to construct each group of training sample sets.

[0043] In this way, by excluding highly relevant environmental data, the redundancy of each set of training samples is reduced, so that when each set of training samples is used for model training, the efficiency and accuracy of model training can be improved.

[0044] S120, using the environmental data and classification labels of whether a fire has occurred contained in each group of training sample sets, iteratively train the preset model until the accuracy of the preset model at the current iteration number is the highest, and determine the preset model at the current iteration number as the candidate model.

[0045] In this embodiment, the electronic device selects a neural network model such as an XGBoost model as a preset model. Before model training, a Bayesian optimization method can be used to search for the best hyperparameters of the preset model within a preset search range. Then, the preset model with the best hyperparameters is iteratively trained using the environmental data and classification labels of whether a fire has occurred contained in each set of training sample sets until the accuracy of the preset model at the current number of iterations is the highest, and the preset model at the current number of iterations is determined as a candidate model.

[0046] Optionally, the optimal hyperparameters of the preset model include but are not limited to one or more of the following combinations: learning rate (0.2), maximum depth (7), number of estimators (400), subsample (0.6), sampling by tree column (0.6), and the optimal hyperparameter of gamma (0.4).

[0047] Optionally, the number of training sample sets used in this embodiment may be 1000, and the number of preset models is also 1000.

[0048] Among them, the specific implementation method of S120 includes but is not limited to the following methods: based on the preset model, the environmental data contained in each group of training sample sets is classified and processed to obtain the predicted category of whether a fire occurs; according to the predicted category and classification label, the accuracy of the preset model is calculated, and the loss of the predicted category and classification label is calculated using the preset loss function to obtain the loss data of the preset model; based on the loss data, the preset model is iteratively trained until the accuracy of the preset model under the current number of iterations is the highest, and the preset model under the current number of iterations is obtained as a candidate model.

[0049] Among them, the accuracy of the preset model is determined based on the positive and negative examples in the predicted category and the positive and negative examples in the classification label; the preset loss function includes a first regularization term that penalizes the coefficients in the form of absolute values ​​and a second regularization term that penalizes the coefficients in the form of the sum of squares.

[0050] Specifically, the electronic device uses a preset model and a parallel computing method to classify the environmental data contained in each training sample set to obtain a predicted category of whether a fire occurs. Then, based on the positive and negative examples in the predicted category and the positive and negative examples in the classification label, the confusion matrix at each training time is calculated as the accuracy of the preset model at each training time. In addition, the predicted category and the classification label are input into a preset loss function containing a first regularization term and a second regularization term to calculate the loss data of the preset model at each training time, so as to use the regularization term to reduce overfitting and improve the generalization ability of the model. Then, the preset model is iteratively trained based on the loss data, and the preset model is optimized by tree pruning optimization during the iterative training process to remove branches that contribute less to the prediction accuracy to control the complexity of the preset model, until the accuracy of the preset model at the current iteration number is the highest, and the preset model at the current iteration number is used as a candidate model.

[0051] Optionally, the accuracy of the preset model can be determined as follows:

[0052] Accuracy=(TP+TN) / (TP+TN+FP+FN)

[0053] Among them, TP is a positive example in both the predicted category and the classification label, TN is a negative example in both the predicted category and the classification label, FN is a negative example in the predicted category and a positive example in the classification label, and FN is a positive example in the predicted category and a negative example in the classification label.

[0054] In this way, through iterative training, a candidate model determined by each set of training samples is obtained.

[0055] S130. From the candidate models determined based on the multiple training sample sets, obtain the model with the highest accuracy as the lightning fire ignition prediction model for the target area.

[0056] Since the electronic device determines a candidate model using each set of training samples, the electronic device continues to compare the accuracy corresponding to each candidate model, and selects the candidate model with the highest accuracy from multiple candidate models as the lightning fire ignition prediction model for the target area.

[0057] In this way, since the ignition case data and non-ignition case data in each group of training sample sets are balanced and also accommodate complex environmental characteristics, the lightning fire ignition prediction model trained based on multiple groups of training sample sets can adapt to areas with complex environmental data and can accurately predict the ignition results of lightning fires.

[0058] A training method for a lightning fire ignition prediction model of an embodiment of the present disclosure includes: obtaining multiple groups of training sample sets in a target area within a historical time period, wherein the difference in data volume between non-ignition case data and ignition case data contained in each group of training sample sets is less than a first preset threshold, and the ignition case data and non-ignition case data in each group of training sample sets both contain environmental data and classification labels for whether a fire has occurred; using the environmental data and classification labels for whether a fire has occurred contained in each group of training sample sets, iteratively training a preset model until the accuracy of the preset model at the current iteration number is the highest, and determining the preset model at the current iteration number as a candidate model; and obtaining the model with the highest accuracy from the candidate models determined based on the multiple groups of training sample sets as the lightning fire ignition prediction model for the target area. In this way, by using multiple groups of training sample sets containing ignition cases and non-ignition cases with similar data amounts, the lightning fire ignition prediction model with the highest accuracy is obtained through iterative training, which overcomes the problem of imbalance in the number of ignition cases and non-ignition cases. In addition, since the model training process actually processes the complex environmental data that cause ignition of fires in multiple groups of training sample sets, in areas with complex environmental data, the trained lightning fire ignition prediction model can also accurately predict the ignition results of lightning fires.

[0059] In some embodiments, after executing S130, the method further includes: obtaining multiple groups of test sample sets in the target area within a historical time period, wherein the difference in data volume between non-ignition case data and ignition case data contained in each group of test sample sets is less than a first preset threshold, and the ignition case data and non-ignition case data in each group of test sample sets both contain environmental data and a classification label of whether a fire has occurred; based on the lightning fire ignition prediction model and the environmental data contained in each group of test sample sets, calculating a contribution value of the environmental data to the classification label, wherein the contribution value is used to explain the degree of contribution of the environmental data to the classification label.

[0060] The acquisition method of each test sample set is the same as the acquisition method of each training sample set, which will not be described in detail here. Optionally, the number of training sample sets is greater than or equal to twice the number of test sample sets. For example, there are 162 training sample sets and 70 test sample sets.

[0061] Considering that lightning fire ignition prediction models are usually considered to be black box models, they fit parameters in hidden dimensions without providing clear relationships, which makes it difficult for researchers to understand the inherent principles of lightning fire ignition prediction models. In order to solve this problem, this embodiment uses methods such as the additional explanation (SHAP) to analyze the environmental data and lightning fire ignition prediction models contained in multiple test sample sets to calculate the contribution value of the environmental data to the classification label, so as to use the contribution value to explain the contribution degree of the environmental data to the classification label. If the contribution value is a positive value, it means that the environmental data has made a favorable contribution to the classification label. If the contribution value is a negative value, it means that the environmental data has made an unfavorable contribution to the classification label. For example, the weather characteristics and soil moisture in the environmental data have the highest contribution to the classification label of a fire, the lightning fire data in the environmental data has the second highest contribution to the classification label of a fire, the vegetation characteristics in the environmental data have the third highest contribution to the classification label of a fire, and the terrain characteristics in the environmental data have the fourth highest contribution to the classification label of a fire.

[0062] In this way, by determining the contribution value of environmental data to the classification label, it is easier for researchers to understand the inherent principles of the lightning fire ignition prediction model, which is beneficial for researchers to deepen their understanding of the lightning fire ignition prediction process.

[0063] The present disclosure also provides a training device for a lightning fire ignition prediction model for implementing the training method for the lightning fire ignition prediction model. Figure 2 In the disclosed embodiment, the training device of the lightning fire ignition prediction model can be an electronic device or a server. The electronic device can include a tablet computer, a desktop computer, a laptop computer and other devices with communication functions, and can also include a virtual machine or a device simulated by a simulator. The server can be a cloud server or a server cluster. Below, the sample data synthesis method is specifically explained with an electronic device as the execution subject.

[0064] Figure 2 A schematic structural diagram of a training device for a lightning fire ignition prediction model provided by an embodiment of the present disclosure is shown.

[0065] like Figure 2 As shown, the training device 200 of the lightning fire ignition prediction model may include:

[0066] The training sample set acquisition module 210 is used to acquire multiple sets of training sample sets in the target area within a historical time period, wherein the difference between the amount of non-ignition case data and the amount of ignition case data contained in each set of training sample sets is less than a first preset threshold, and the ignition case data and the non-ignition case data in each set of training sample sets both contain environmental data and a classification label of whether a fire has occurred;

[0067] An iterative training module 220 is used to iteratively train the preset model using the environmental data and the classification label of whether a fire has occurred contained in each group of training sample sets until the accuracy of the preset model at the current iteration number is the highest, and the preset model at the current iteration number is determined as a candidate model;

[0068] The model determination module 230 is used to obtain the model with the highest accuracy from the candidate models determined based on the multiple groups of training sample sets as the lightning fire ignition prediction model of the target area.

[0069] A training device for a lightning fire ignition prediction model according to an embodiment of the present disclosure includes: obtaining multiple groups of training sample sets in a target area within a historical time period, wherein the difference in data volume between non-ignition case data and ignition case data contained in each group of training sample sets is less than a first preset threshold, and the ignition case data and non-ignition case data in each group of training sample sets both contain environmental data and classification labels for whether a fire has occurred; iteratively training a preset model using the environmental data and classification labels for whether a fire has occurred contained in each group of training sample sets until the accuracy of the preset model at the current iteration number is the highest, and determining the preset model at the current iteration number as a candidate model; and obtaining the model with the highest accuracy from the candidate models determined based on the multiple groups of training sample sets as the lightning fire ignition prediction model for the target area. In this way, by using multiple groups of training sample sets containing ignition cases and non-ignition cases with similar data amounts, the lightning fire ignition prediction model with the highest accuracy is obtained through iterative training, which overcomes the problem of imbalance in the number of ignition cases and non-ignition cases. In addition, since the model training process actually processes the complex environmental data that cause ignition of fires in multiple groups of training sample sets, in areas with complex environmental data, the trained lightning fire ignition prediction model can also accurately predict the ignition results of lightning fires.

[0070] In some embodiments of the present disclosure, the training sample set acquisition module 210 includes:

[0071] a case data acquisition unit, configured to acquire a first quantity of ignition case data and a second quantity of non-ignition case data from the historical environmental data of the historical time period, wherein the second quantity is greater than the first quantity, and a difference between the second quantity and the first quantity is greater than a second preset threshold;

[0072] a multiple random extraction unit, configured to randomly extract a third number of non-ignition case data from the second number of non-ignition case data for multiple times, wherein a difference between the second number and the third number of data is less than the first preset threshold;

[0073] The training sample set forming unit is used to form a training sample set using the first number of ignition case data and the third number of non-ignition case data randomly selected each time.

[0074] In some embodiments of the present disclosure, the training sample set constitutes a unit, specifically used for:

[0075] Performing feature correlation analysis on the third number of non-ignition case data and the first number of ignition case data randomly selected each time to determine feature correlation;

[0076] From the third number of non-ignition case data and the first number of ignition case data randomly selected each time, delete the environmental data whose feature correlation is greater than a preset correlation threshold;

[0077] Each group of training sample sets is formed by deleting the third number of non-ignition case data and the first number of ignition case data whose feature correlation is greater than the preset correlation threshold.

[0078] In some embodiments of the present disclosure, the iterative training module 220 includes:

[0079] A classification processing unit, configured to classify the environmental data contained in each training sample set based on the preset model to obtain a prediction category of whether a fire occurs;

[0080] A calculation unit, used to calculate the accuracy of a preset model according to the predicted category and the classification label, and perform loss calculation on the predicted category and the classification label using a preset loss function to obtain loss data of the preset model;

[0081] An iterative training unit is used to iteratively train the preset model based on the loss data until the accuracy of the preset model at the current iteration number is the highest, and obtain the preset model at the current iteration number as the candidate model.

[0082] In some embodiments of the present disclosure, the accuracy of the preset model is determined based on the positive examples and negative examples in the predicted category and the positive examples and negative examples in the classification label; the preset loss function includes a first regularization term that penalizes the coefficients in the form of absolute values ​​and a second regularization term that penalizes the coefficients in the form of the sum of squares.

[0083] In some embodiments of the present disclosure, the environmental data includes one or more combinations of the following:

[0084] Weather characteristics, soil moisture, lightning fire data that caused fires, lightning fire data that did not cause fires, no lightning fire data, vegetation characteristics, and terrain characteristics.

[0085] In some embodiments of the present disclosure, the device further includes:

[0086] A test sample set acquisition module, used to acquire multiple test sample sets of a target area within a historical time period, wherein the difference between the amount of non-ignition case data and the amount of ignition case data contained in each test sample set is less than the first preset threshold, and the ignition case data and the non-ignition case data in each test sample set both contain environmental data and a classification label of whether a fire has occurred;

[0087] A contribution value calculation module is used to calculate the contribution value of the environmental data to the classification label based on the lightning fire ignition prediction model and the environmental data contained in each test sample set, wherein the contribution value is used to explain the contribution degree of the environmental data to the classification label.

[0088] It should be noted that Figure 2 The training device 200 of the lightning fire ignition prediction model shown can be executed Figure 1 The various steps in the method embodiment shown in the figure are implemented Figure 1 The various processes and effects in the method embodiment shown are not described in detail here.

[0089] Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present disclosure is shown.

[0090] like Figure 3 As shown, the electronic device may include a processor 301 and a memory 302 storing computer program instructions.

[0091] Specifically, the processor 301 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0092] The memory 302 may include a large capacity memory for information or instructions. By way of example and not limitation, the memory 302 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 302 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 302 may be inside or outside the integrated gateway device. In a particular embodiment, the memory 302 is a non-volatile solid-state memory. In a particular embodiment, the memory 302 includes a read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (Electrically Erasable Programmable ROM, EEPROM), an electrically rewritable ROM (EAROM) or a flash memory, or a combination of two or more of these.

[0093] The processor 301 reads and executes the computer program instructions stored in the memory 302 to perform the steps of the training method of the lightning fire ignition prediction model provided in the embodiment of the present disclosure.

[0094] In one example, the electronic device may further include a transceiver 303 and a bus 304. Figure 3 As shown, the processor 301, the memory 302 and the transceiver 303 are connected via a bus 304 and communicate with each other.

[0095] The bus 304 includes hardware, software, or both. For example, but not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industrial Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a Memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, the bus 304 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0096] The following is an embodiment of a computer-readable storage medium provided in an embodiment of the present disclosure. The computer-readable storage medium and the training method of the lightning fire ignition prediction model in the above-mentioned embodiments belong to the same inventive concept. For details not described in detail in the embodiment of the computer-readable storage medium, reference can be made to the embodiment of the training method of the lightning fire ignition prediction model in the above-mentioned embodiments.

[0097] This embodiment provides a storage medium containing computer executable instructions. When the computer executable instructions are executed by a computer processor, they are used to execute a training method for a lightning fire ignition prediction model, including:

[0098] Acquire multiple sets of training sample sets in the target area within a historical time period, wherein the difference between the amount of non-ignition case data and the amount of ignition case data contained in each set of training sample sets is less than a first preset threshold, and the ignition case data and the non-ignition case data in each set of training sample sets both contain environmental data and a classification label of whether a fire has occurred;

[0099] Iteratively train the preset model using the environmental data and the classification label of whether a fire has occurred contained in each group of training sample sets until the accuracy of the preset model at the current iteration number is the highest, and determine the preset model at the current iteration number as the candidate model;

[0100] From the candidate models determined based on the multiple groups of training sample sets, the model with the highest accuracy is obtained as the lightning fire ignition prediction model for the target area.

[0101] Of course, the storage medium containing computer executable instructions provided in the embodiments of the present disclosure is not limited to the above method operations, and the computer executable instructions can also execute the relevant operations in the training method of the lightning fire ignition prediction model provided in any embodiment of the present disclosure.

[0102] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present disclosure can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present disclosure can be essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer cloud platform (which can be a personal computer, a server, or a network cloud platform, etc.) to execute the training method of the lightning fire ignition prediction model provided in each embodiment of the present disclosure.

[0103] Note that the above are only preferred embodiments of the present disclosure and the technical principles used. Those skilled in the art will understand that the present disclosure is not limited to the specific embodiments herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present disclosure. Therefore, although the present disclosure is described in more detail through the above embodiments, the present disclosure is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present disclosure, and the scope of the present disclosure is determined by the scope of the appended claims.

Claims

1. A training method for a lightning fire ignition prediction model, comprising: Acquire multiple sets of training sample sets in the target area within a historical time period, wherein the difference between the amount of non-ignition case data and the amount of ignition case data contained in each set of training sample sets is less than a first preset threshold, and the ignition case data and the non-ignition case data in each set of training sample sets both contain environmental data and a classification label of whether a fire has occurred; Iteratively train the preset model using the environmental data and the classification label of whether a fire has occurred contained in each group of training sample sets until the accuracy of the preset model at the current iteration number is the highest, and determine the preset model at the current iteration number as the candidate model; From the candidate models determined based on the multiple groups of training sample sets, the model with the highest accuracy is obtained as the lightning fire ignition prediction model for the target area.

2. The method according to claim 1, characterized in that The step of obtaining multiple sets of training sample sets of the target area within a historical time period includes: Acquire a first quantity of ignition case data and a second quantity of non-ignition case data from the historical environmental data of the historical time period, wherein the second quantity is greater than the first quantity, and a difference between the second quantity and the first quantity is greater than a second preset threshold; Randomly extracting a third number of non-ignition case data from the second number of non-ignition case data for multiple times, wherein a difference between the second number and the third number of data is less than the first preset threshold; A set of training sample sets is formed by using the first number of ignition case data and the third number of non-ignition case data randomly selected each time.

3. The method according to claim 2, characterized in that The method of using the first number of ignition case data and the third number of non-ignition case data randomly selected each time to form each training sample set includes: Performing feature correlation analysis on the third number of non-ignition case data and the first number of ignition case data randomly selected each time to determine feature correlation; From the third number of non-ignition case data and the first number of ignition case data randomly selected each time, delete the environmental data whose feature correlation is greater than a preset correlation threshold; Each group of training sample sets is formed by deleting the third number of non-ignition case data and the first number of ignition case data whose feature correlation is greater than the preset correlation threshold.

4. The method according to claim 1, characterized in that The method of iteratively training the preset model by using the environmental data and the classification label of whether a fire has occurred contained in each group of training sample sets until the accuracy of the preset model at the current iteration number is the highest, and determining the preset model at the current iteration number as a candidate model includes: Based on the preset model, the environmental data contained in each group of training sample sets are classified to obtain a prediction category of whether a fire occurs; According to the predicted category and the classification label, the accuracy of the preset model is calculated, and the loss of the predicted category and the classification label is calculated using a preset loss function to obtain loss data of the preset model; The preset model is iteratively trained based on the loss data until the accuracy of the preset model at the current iteration number is the highest, and the preset model at the current iteration number is obtained as the candidate model.

5. The method according to claim 4, characterized in that The accuracy of the preset model is determined based on the positive examples and negative examples in the predicted category and the positive examples and negative examples in the classification label; the preset loss function includes a first regularization term that penalizes the coefficients in the form of absolute values ​​and a second regularization term that penalizes the coefficients in the form of the sum of squares.

6. The method according to claim 1, characterized in that The environmental data includes one or more of the following combinations: Weather characteristics, soil moisture, lightning fire data that caused fires, lightning fire data that did not cause fires, no lightning fire data, vegetation characteristics, and terrain characteristics.

7. The method according to claim 1, characterized in that Also includes: Acquire multiple groups of test sample sets in the target area within a historical time period, wherein the difference between the amount of non-ignition case data and the amount of ignition case data contained in each group of test sample sets is less than the first preset threshold, and the ignition case data and the non-ignition case data in each group of test sample sets both contain environmental data and a classification label of whether a fire has occurred; Based on the lightning fire ignition prediction model and the environmental data included in each test sample set, a contribution value of the environmental data to the classification label is calculated, wherein the contribution value is used to explain the degree of contribution of the environmental data to the classification label.

8. A training device for a lightning fire ignition prediction model, comprising: A training sample set acquisition module, used to acquire multiple sets of training sample sets in a target area within a historical time period, wherein the difference between the amount of non-ignition case data and the amount of ignition case data contained in each set of training sample sets is less than a first preset threshold, and the ignition case data and the non-ignition case data in each set of training sample sets both contain environmental data and a classification label of whether a fire has occurred; An iterative training module, used to iteratively train the preset model using the environmental data and the classification label of whether a fire has occurred contained in each group of training sample sets, until the accuracy of the preset model at the current iteration number is the highest, and determine the preset model at the current iteration number as the candidate model; The model determination module is used to obtain the model with the highest accuracy from the candidate models determined based on multiple groups of training sample sets as the lightning fire ignition prediction model for the target area.

9. An electronic device, characterized in that: include: processor; A memory for storing executable instructions; The processor is used to read the executable instructions from the memory and execute the executable instructions to implement the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the method according to any one of claims 1 to 7.

Images