A forest fire prediction method and system based on artificial intelligence

Abstract

The present invention discloses a forest fire prediction method and system based on artificial intelligence, which belongs to the technical field of intelligent prevention and control of forest fires. The method includes data set construction, feature set construction, forest fire sensitivity prediction, forest fire location prediction and forest fire prediction report generation. The present invention adopts an improved Xception model to construct a forest image feature set, which can better capture complex features, has stronger generalization ability, and helps to improve the accuracy and stability of subsequent predictions; adopts a long-term and short-term regularized random forest model to predict forest fire sensitivity, effectively captures time series features, and predicts based on a comprehensive environmental feature set and a forest fire feature set, thereby improving the model accuracy and generalization ability; constructs a forest fire location prediction model based on a support vector machine, and adopts an improved square hinge loss function for model training, which can more accurately predict the forest fire location, and helps to improve the timeliness of fire emergency measures.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent prevention and control of forest fires, and specifically refers to a forest fire prediction method and system based on artificial intelligence. Background Art

[0002] Forest fire prediction uses artificial intelligence technology to analyze multiple data sources to predict the probability and location of forest fires, thereby providing early fire warning information, assisting relevant departments to take preventive measures in a timely manner, reducing the frequency and scale of fires, reducing damage to forests, and maintaining the health and stability of forest ecosystems. However, in the existing forest fire prediction process, there are technical problems such as complex environmental factors at forest fire locations, rich feature information, high requirements for feature extraction performance, and insufficient generalization ability of existing feature extraction methods in practical applications; forest fires are affected by multiple environmental factors, and there is a lack of a forest fire sensitivity prediction method that comprehensively considers multiple environmental factors; there is a technical problem that fire prevention emergency measures require timely discovery of fire sources, and a method for predicting forest fire locations is urgently needed. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a forest fire prediction method and system based on artificial intelligence. In view of the technical problems that in the existing forest fire prediction process, the environmental factors of the forest fire location are complex, the feature information is rich, the feature extraction performance requirements are high, and the generalization ability of the existing feature extraction method is insufficient in practical applications, this solution creatively adopts an improved Xception model to construct a forest image feature set, which can better capture complex features, adapt to image data of different scenes, has stronger generalization ability, helps to improve the accuracy and stability of subsequent predictions, and improves the calculation efficiency of the model; in view of the fact that in the existing forest fire prediction process, forest fires are affected by a variety of environmental factors, and there is a lack of a Taking into account the technical issues of forest fire sensitivity prediction methods that take into account multiple environmental factors, this scheme creatively adopts the long- and short-term regularized random forest model to predict forest fire sensitivity, effectively captures time series characteristics, and makes predictions based on the comprehensive environmental feature set and forest fire feature set, thereby improving the model accuracy and generalization ability and reducing the risk of overfitting. In view of the technical problem that in the existing forest fire prediction process, fire prevention emergency measures require timely detection of fire sources, and a method for predicting forest fire locations is urgently needed, this scheme creatively constructs a forest fire location prediction model based on support vector machines, and uses an improved square hinge loss function for model training, which can more accurately predict forest fire locations, help improve the timeliness of fire prevention emergency measures, and enhance the reliability of the overall method.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a forest fire prediction method based on artificial intelligence, the method comprising the following steps:

[0005] Step S1: Dataset construction;

[0006] Step S2: feature set construction;

[0007] Step S3: forest fire sensitivity prediction;

[0008] Step S4: forest fire location prediction;

[0009] Step S5: Generate forest fire prediction report.

[0010] Furthermore, in step S1, the data set is constructed by collecting meteorological data, terrain data, historical fire data and forest remote sensing image data, and preprocessing them through missing value filling, noise removal, data enhancement and normalization to obtain an initial forest fire data set.

[0011] Further, in step S2, the feature set construction is used to extract features from the forest fire initial data set and construct a feature set, specifically by constructing an environment feature set, a forest image feature set and a forest fire feature set, and the feature set construction includes the following steps:

[0012] Step S21: constructing an environmental feature set, including the following steps:

[0013] Step S211: using dynamic time window technology to extract dynamic features from the meteorological data in the initial forest fire data set to obtain meteorological dynamic features;

[0014] Step S212: manually extracting terrain features from the terrain data in the initial forest fire data set, and merging the meteorological dynamic features and the terrain features to obtain an environmental feature set;

[0015] Step S22: constructing a forest image feature set, specifically using an improved Xception model to construct a forest image feature set to obtain a forest image feature set, including the following steps:

[0016] Step S221: construct a traditional Xception model architecture;

[0017] Step S222: improving the traditional Xception model architecture to obtain an improved Xception model, specifically, selecting a Swish activation function to replace the ReLU activation function in the traditional Xception model, and using a one-dimensional convolutional layer to replace the global average pooling layer in the traditional Xception model to prevent channel overlap and improve the ability to capture subtle features;

[0018] The calculation formula of the Swish activation function is:

[0019] ;

[0020] Where Swish(·) is the Swish activation function, z is the input parameter, specifically the value input to the Swish activation function, and Sig(·) is the Sigmoid function;

[0021] Step S223: extracting features from the forest remote sensing image data in the forest fire initial data set through the improved Xception model to obtain a forest image feature set;

[0022] Step S23: constructing a forest fire feature set, specifically, selecting important features from the meteorological data, terrain data and historical fire data in the forest fire initial data set through the information gain feature selection method to obtain the forest fire feature set, including the following steps:

[0023] Step S231: merging the meteorological data, terrain data and historical fire data in the initial forest fire data set to obtain a forest fire numerical data set;

[0024] Step S232: calculating information entropy, specifically calculating information entropy of the target variable;

[0025] Step S233: calculating the conditional entropy, specifically calculating the conditional entropy of the target variable;

[0026] Step S234: Calculate information gain, which is used to measure the importance of characteristic factors. Specifically, the information gain of each characteristic factor in the forest fire numerical data set is calculated. The calculation formula is:

[0027] ;

[0028] In the formula, IG(Y, X) is the information gain of the characteristic factor, Y is the target variable, specifically whether a fire occurs or not, X is the characteristic factor, H(Y) is the information entropy of the target variable, and H(Y|X) is the conditional entropy of the target variable, which is used to measure the contribution of the characteristic factor to the target variable;

[0029] Step S235: Select important features, specifically, based on information gain, merge the first k feature factors with the highest information gain to obtain a forest fire feature set.

[0030] Further, in step S3, the forest fire sensitivity prediction is specifically to use a long-term and short-term regularized random forest model to predict the forest fire sensitivity based on the environmental feature set and the forest fire feature set;

[0031] The long-short-term regularized random forest model specifically includes a long-short-term memory network block and a regularized random forest block;

[0032] The long short-term memory network block is used to extract time series features and generate a forest fire meteorological index, and specifically includes an input sublayer, a long short-term memory network layer, a random inactivation layer, a fully connected layer, and an output sublayer;

[0033] The regularized random forest block is used to make predictions based on the forest fire meteorological index and the forest fire feature set;

[0034] The forest fire sensitivity prediction comprises the following steps:

[0035] Step S31: constructing a long short-term memory network block, specifically inputting the environmental feature set into the long short-term memory network block, performing time series data modeling, and generating a forest fire meteorological index, including the following steps:

[0036] Step S311: construct an input sublayer for inputting an environmental feature set;

[0037] Step S312: constructing a long short-term memory network layer for capturing the long-term dependency of time series data, wherein the long short-term memory network layer specifically includes a first long short-term network sublayer, a first random inactivation layer, a second long short-term network sublayer and a second random inactivation layer;

[0038] Step S313: construct a random dropout layer to prevent overfitting;

[0039] Step S314: construct a fully connected layer to integrate feature information;

[0040] Step S315: constructing an output sublayer for generating a forest fire meteorological index;

[0041] Step S32: constructing a regularized random forest block, specifically inputting the forest fire meteorological index and the forest fire feature set into the regularized random forest block, performing prediction, and generating a prediction result, including the following steps:

[0042] Step S321: constructing an initial random forest, specifically, taking the forest fire meteorological index and the forest fire feature set as sensitivity prediction input samples, and generating Tr initial decision trees by randomly sampling the sensitivity prediction input samples to obtain an initial random forest, wherein the initial random forest includes Tr initial decision trees;

[0043] Step S322: Calculate the average depth of each initial decision tree, add the average depth of each initial decision tree to obtain the total average depth, and construct a regularization term. The calculation formula is:

[0044] ;

[0045] In the formula, Rt is the regularization term, α is the regularization parameter, and ADepth total is the overall average depth;

[0046] Step S323: Pruning the initial decision tree using a regularization method to obtain a regularized decision tree, specifically, comparing the average depth of the initial decision tree with the regularization term. If the average depth of the initial decision tree is greater than the regularization term, pruning the initial decision tree;

[0047] Step S324: generating a prediction result by integrating the outputs of all regularized decision trees;

[0048] Step S33: constructing a forest fire sensitivity prediction model, specifically, performing model training based on a long-term and short-term regularized random forest model to obtain a forest fire sensitivity prediction model;

[0049] Step S34: Perform forest fire sensitivity prediction using a forest fire sensitivity prediction model to obtain a forest fire sensitivity level.

[0050] Further, in step S4, the forest fire location prediction is specifically to construct a forest fire location prediction model based on a support vector machine according to a forest image feature set, perform forest fire location prediction, and obtain forest fire location prediction information, including the following steps:

[0051] Step S41: constructing a support vector machine, specifically, inputting the forest image feature set into the support vector machine as a location prediction input sample, selecting a Gaussian radial basis function as a kernel function of the support vector machine, and constructing the support vector machine;

[0052] Step S42: constructing a forest fire location prediction model, specifically using an improved square hinge loss function to train a support vector machine to obtain a forest fire location prediction model, including the following steps:

[0053] Step S421: define an improved square hinge loss function, the calculation formula is:

[0054] ;

[0055] Where loss(·) is the improved square hinge loss function, c is the sample residual, which is used to represent the error between the model's prediction result of the location prediction input sample and the true label;

[0056] Step S422: construct an objective function, the calculation formula is:

[0057] ;

[0058] Where f(w,b) is the objective function, w is the model weight vector, b is the model bias term, β is the regularization parameter, i is the location prediction input sample index, n is the number of location prediction input samples, and y is i is the true label of the input sample predicted at the i-th location, x i is the feature vector of the predicted input sample at the i-th location, is the model weight vector w and the feature vector x of the input sample predicted at the i-th location i The inner product of

[0059] Step S423: Based on the objective function, the model parameters are updated using the gradient descent method, the optimal model parameters are calculated, and a forest fire location prediction model is obtained;

[0060] Step S43: Predicting the location of a forest fire using a forest fire location prediction model to obtain forest fire location prediction information.

[0061] Furthermore, in step S5, the forest fire prediction report is generated, specifically by integrating the forest fire sensitivity level and forest fire location prediction information to generate a forest fire prediction report and provide forest fire prevention and control suggestions.

[0062] The present invention provides an artificial intelligence-based forest fire prediction system, comprising: a data set construction module, a feature set construction module, a forest fire sensitivity prediction module, a forest fire location prediction module and a forest fire prediction report generation module;

[0063] The data set construction module is used for data set construction, and obtains the forest fire initial data set through data set construction, and sends the forest fire initial data set to the feature set construction module;

[0064] The feature set construction module is used for feature set construction, and obtains an environmental feature set, a forest image feature set and a forest fire feature set through feature set construction, and sends the environmental feature set and the forest fire feature set to the forest fire sensitivity prediction module, and sends the forest image feature set to the forest fire location prediction module;

[0065] The forest fire sensitivity prediction module is used for forest fire sensitivity prediction, obtains the forest fire sensitivity level through forest fire sensitivity prediction, and sends the forest fire sensitivity level to the forest fire prediction report generation module;

[0066] The forest fire location prediction module is used for forest fire location prediction, obtains forest fire location prediction information through forest fire location prediction, and sends the forest fire location prediction information to the forest fire prediction report generation module;

[0067] The forest fire prediction report generation module is used for forest fire prediction report generation. Through the forest fire prediction report generation, a forest fire prediction report is obtained and forest fire prevention and control suggestions are provided.

[0068] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0069] (1) In view of the technical problems in the existing forest fire prediction process, such as the complex environmental factors of forest fire locations, rich feature information, high requirements for feature extraction performance, and insufficient generalization ability of existing feature extraction methods in practical applications, this solution creatively adopts the improved Xception model to construct forest image feature sets, which can better capture complex features, adapt to image data of different scenes, and has stronger generalization ability, which helps to improve the accuracy and stability of subsequent predictions and improve the calculation efficiency of the model;

[0070] (2) In view of the technical problem that forest fires are affected by multiple environmental factors in the existing forest fire prediction process and there is a lack of a forest fire sensitivity prediction method that comprehensively considers multiple environmental factors, this scheme creatively adopts the long-term and short-term regularized random forest model to predict forest fire sensitivity, effectively captures time series characteristics, and makes predictions based on the environmental feature set and forest fire feature set, thereby improving the model accuracy and generalization ability and reducing the risk of overfitting.

[0071] (3) In view of the technical problem that in the existing forest fire prediction process, fire prevention emergency measures need to detect the fire source in time, and a method for predicting the location of forest fires is urgently needed, this scheme creatively constructs a forest fire location prediction model based on support vector machine and uses an improved square hinge loss function for model training. It can more accurately predict the location of forest fires, help improve the timeliness of fire prevention emergency measures, and enhance the reliability of the overall method. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic diagram of a flow chart of a forest fire prediction method based on artificial intelligence provided by the present invention;

[0073] Figure 2 A schematic diagram of a forest fire prediction system based on artificial intelligence provided by the present invention;

[0074] Figure 3 is a schematic flow chart of step S2;

[0075] Figure 4 is a schematic flow chart of step S3;

[0076] Figure 5 It is a schematic diagram of the process of step S4.

[0077] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0079] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0080] Example 1, see Figure 1 The present invention provides a forest fire prediction method based on artificial intelligence, which comprises the following steps:

[0081] Step S1: Dataset construction;

[0082] Step S2: feature set construction;

[0083] Step S3: forest fire sensitivity prediction;

[0084] Step S4: forest fire location prediction;

[0085] Step S5: Generate forest fire prediction report.

[0086] Example 2, see Figure 1 ,This embodiment is based on the above embodiment. In step S1, the data set is constructed, specifically by collecting meteorological data, terrain data, historical fire data and forest remote sensing image data, and preprocessing them through missing value filling, noise removal, data enhancement and normalization to obtain an initial forest fire data set;

[0087] The meteorological data specifically include forest area temperature, relative humidity, 24-hour rainfall, wind speed and wind direction;

[0088] The terrain data specifically includes altitude, slope and orientation;

[0089] The historical fire data includes fire type, location, date and size;

[0090] The forest remote sensing image data specifically includes forest fire remote sensing image data and forest non-fire remote sensing image data.

[0091] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the feature set construction is used to extract features from the forest fire initial data set and construct a feature set. Specifically, the feature set construction is performed through environmental feature set construction, forest image feature set construction and forest fire feature set construction, including the following steps:

[0092] Step S21: constructing an environmental feature set, including the following steps:

[0093] Step S211: using dynamic time window technology to extract dynamic features from the meteorological data in the initial forest fire data set to obtain meteorological dynamic features;

[0094] Step S212: manually extracting terrain features from the terrain data in the initial forest fire data set, and merging the meteorological dynamic features and the terrain features to obtain an environmental feature set;

[0095] Step S22: constructing a forest image feature set, specifically using an improved Xception model to construct a forest image feature set to obtain a forest image feature set, including the following steps:

[0096] Step S221: construct a traditional Xception model architecture;

[0097] Step S222: improving the traditional Xception model architecture to obtain an improved Xception model, specifically, selecting a Swish activation function to replace the ReLU activation function in the traditional Xception model, and using a one-dimensional convolutional layer to replace the global average pooling layer in the traditional Xception model to prevent channel overlap and improve the ability to capture subtle features;

[0098] The calculation formula of the Swish activation function is:

[0099] ;

[0100] Where Swish(·) is the Swish activation function, z is the input parameter, specifically the value input to the Swish activation function, and Sig(·) is the Sigmoid function;

[0101] Step S223: extracting features from the forest remote sensing image data in the forest fire initial data set through the improved Xception model to obtain a forest image feature set;

[0102] Step S23: constructing a forest fire feature set, specifically, selecting important features from the meteorological data, terrain data and historical fire data in the forest fire initial data set through the information gain feature selection method to obtain the forest fire feature set, including the following steps:

[0103] Step S231: merging the meteorological data, terrain data and historical fire data in the initial forest fire data set to obtain a forest fire numerical data set;

[0104] Step S232: calculating information entropy, specifically calculating information entropy of the target variable;

[0105] Step S233: calculating the conditional entropy, specifically calculating the conditional entropy of the target variable;

[0106] Step S234: Calculate information gain, which is used to measure the importance of characteristic factors. Specifically, the information gain of each characteristic factor in the forest fire numerical data set is calculated. The calculation formula is:

[0107] ;

[0108] In the formula, IG(Y, X) is the information gain of the characteristic factor, Y is the target variable, specifically whether a fire occurs or not, X is the characteristic factor, H(Y) is the information entropy of the target variable, and H(Y|X) is the conditional entropy of the target variable, which is used to measure the contribution of the characteristic factor to the target variable;

[0109] Step S235: selecting important features, specifically, merging the first k feature factors with the highest information gain according to the information gain to obtain a forest fire feature set;

[0110] By performing the above operations, in order to address the technical problems in the existing forest fire prediction process, such as the complex environmental factors of forest fire locations, rich feature information, high requirements for feature extraction performance, and insufficient generalization ability of existing feature extraction methods in practical applications, this solution creatively adopts an improved Xception model to construct a forest image feature set, which can better capture complex features, adapt to image data of different scenes, has stronger generalization ability, helps to improve the accuracy and stability of subsequent predictions, and improves the calculation efficiency of the model.

[0111] Example 4, see Figure 1 and Figure 4 , this embodiment is based on the above embodiment, in step S3, the forest fire sensitivity prediction is specifically to use the long-term and short-term regularized random forest model to predict the forest fire sensitivity based on the environmental feature set and the forest fire feature set;

[0112] The long-short-term regularized random forest model specifically includes a long-short-term memory network block and a regularized random forest block;

[0113] The long short-term memory network block is used to extract time series features and generate a forest fire meteorological index, and specifically includes an input sublayer, a long short-term memory network layer, a random inactivation layer, a fully connected layer, and an output sublayer;

[0114] The regularized random forest block is used to make predictions based on the forest fire meteorological index and the forest fire feature set;

[0115] The forest fire sensitivity prediction comprises the following steps:

[0116] Step S31: constructing a long short-term memory network block, specifically inputting the environmental feature set into the long short-term memory network block, performing time series data modeling, and generating a forest fire meteorological index, including the following steps:

[0117] Step S311: construct an input sublayer for inputting an environmental feature set;

[0118] Step S312: constructing a long short-term memory network layer for capturing the long-term dependency of time series data, wherein the long short-term memory network layer specifically includes a first long short-term network sublayer, a first random inactivation layer, a second long short-term network sublayer and a second random inactivation layer;

[0119] Step S313: construct a random dropout layer to prevent overfitting;

[0120] Step S314: construct a fully connected layer to integrate feature information;

[0121] Step S315: constructing an output sublayer for generating a forest fire meteorological index;

[0122] Step S32: constructing a regularized random forest block, specifically inputting the forest fire meteorological index and the forest fire feature set into the regularized random forest block, performing prediction, and generating a prediction result, including the following steps:

[0123] Step S321: constructing an initial random forest, specifically, taking the forest fire meteorological index and the forest fire feature set as sensitivity prediction input samples, and generating Tr initial decision trees by randomly sampling the sensitivity prediction input samples to obtain an initial random forest, wherein the initial random forest includes Tr initial decision trees;

[0124] Step S322: Calculate the average depth of each initial decision tree, add the average depth of each initial decision tree to obtain the total average depth, and construct a regularization term. The calculation formula is:

[0125] ;

[0126] In the formula, Rt is the regularization term, α is the regularization parameter, and ADepth total is the overall average depth;

[0127] Step S323: Pruning the initial decision tree using a regularization method to obtain a regularized decision tree, specifically, comparing the average depth of the initial decision tree with the regularization term. If the average depth of the initial decision tree is greater than the regularization term, pruning the initial decision tree;

[0128] Step S324: Generate a prediction result by integrating the outputs of all regularized decision trees;

[0129] Step S33: constructing a forest fire sensitivity prediction model, specifically, performing model training based on a long-term and short-term regularized random forest model to obtain a forest fire sensitivity prediction model;

[0130] Step S34: performing forest fire sensitivity prediction using a forest fire sensitivity prediction model to obtain a forest fire sensitivity level, wherein the forest fire sensitivity level includes extremely low sensitivity, low sensitivity, medium sensitivity, high sensitivity and extremely high sensitivity, and a higher forest fire sensitivity indicates a higher probability of forest fire occurrence;

[0131] By performing the above operations, in order to address the technical problem that forest fires are affected by multiple environmental factors in the existing forest fire prediction process and there is a lack of a forest fire sensitivity prediction method that comprehensively considers multiple environmental factors, this scheme creatively adopts the long- and short-term regularized random forest model to predict forest fire sensitivity, effectively captures time series characteristics, and makes predictions based on the comprehensive environmental feature set and forest fire feature set, thereby improving the model accuracy and generalization ability and reducing the risk of overfitting.

[0132] Example 5, see Figure 1 and Figure 5 This embodiment is based on the above embodiment. In step S4, the forest fire location prediction is specifically based on the forest image feature set, based on the support vector machine to build a forest fire location prediction model, perform forest fire location prediction, and obtain forest fire location prediction information, including the following steps:

[0133] Step S41: constructing a support vector machine, specifically, inputting the forest image feature set into the support vector machine as a location prediction input sample, selecting a Gaussian radial basis function as a kernel function of the support vector machine, and constructing the support vector machine;

[0134] The calculation formula of the radial basis kernel function is:

[0135] ;

[0136] Where RBF(·) is the radial basis kernel function, a j1 is the j1th input sample, a j2 is the j2th input sample, the input sample specifically refers to the location prediction input sample, j1 is the first index of the input sample, j2 is the second index of the input sample, the second index of the input sample is not equal to the first index of the input sample, c is the width parameter of the Gaussian function, and exp(·) is the natural exponential function;

[0137] Step S42: constructing a forest fire location prediction model, specifically using an improved square hinge loss function to train a support vector machine to obtain a forest fire location prediction model, including the following steps:

[0138] Step S421: define an improved square hinge loss function, the calculation formula is:

[0139] ;

[0140] Where loss(·) is the improved square hinge loss function, c is the sample residual, which is used to represent the error between the model's prediction result of the location prediction input sample and the true label;

[0141] Step S422: construct an objective function, the calculation formula is:

[0142] ;

[0143] Where f(w,b) is the objective function, w is the model weight vector, b is the model bias term, β is the regularization parameter, i is the location prediction input sample index, n is the number of location prediction input samples, and y is i is the true label of the input sample predicted at the i-th location, x i is the feature vector of the predicted input sample at the i-th location, is the model weight vector w and the feature vector x of the input sample predicted at the i-th location i The inner product of

[0144] Step S423: Based on the objective function, the model parameters are updated using the gradient descent method, the optimal model parameters are calculated, and a forest fire location prediction model is obtained;

[0145] Step S43: predicting the location of a forest fire using a forest fire location prediction model to obtain forest fire location prediction information;

[0146] By performing the above operations, in order to solve the technical problem that in the existing forest fire prediction process, fire prevention emergency measures need to discover the fire source in time, and a method for predicting the location of forest fires is urgently needed, this scheme creatively constructs a forest fire location prediction model based on support vector machine, and uses an improved square hinge loss function for model training. It can more accurately predict the location of forest fires, help improve the timeliness of fire prevention emergency measures, and enhance the reliability of the overall method.

[0147] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the forest fire prediction report is generated, specifically, the forest fire sensitivity level and forest fire location prediction information are integrated to generate a forest fire prediction report and provide forest fire prevention and control suggestions.

[0148] Embodiment 7, see Figure 2 , this embodiment is based on the above embodiment, and the present invention provides an artificial intelligence-based forest fire prediction system, including: a data set construction module, a feature set construction module, a forest fire sensitivity prediction module, a forest fire location prediction module and a forest fire prediction report generation module;

[0149] The data set construction module is used for data set construction, and obtains the forest fire initial data set through data set construction, and sends the forest fire initial data set to the feature set construction module;

[0150] The feature set construction module is used for feature set construction, and obtains an environmental feature set, a forest image feature set and a forest fire feature set through feature set construction, and sends the environmental feature set and the forest fire feature set to the forest fire sensitivity prediction module, and sends the forest image feature set to the forest fire location prediction module;

[0151] The forest fire sensitivity prediction module is used for forest fire sensitivity prediction, obtains the forest fire sensitivity level through forest fire sensitivity prediction, and sends the forest fire sensitivity level to the forest fire prediction report generation module;

[0152] The forest fire location prediction module is used for forest fire location prediction, obtains forest fire location prediction information through forest fire location prediction, and sends the forest fire location prediction information to the forest fire prediction report generation module;

[0153] The forest fire prediction report generation module is used for forest fire prediction report generation. Through the forest fire prediction report generation, a forest fire prediction report is obtained and forest fire prevention and control suggestions are provided.

[0154] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0155] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0156] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A forest fire prediction method based on artificial intelligence, characterized in that: The method comprises the following steps: Step S1: Dataset construction; Step S2: feature set construction, specifically, constructing a feature set through environmental feature set construction, forest image feature set construction and forest fire feature set construction; In step S2, the feature set construction is used to extract features from the forest fire initial data set and construct a feature set, including the following steps: Step S21: constructing an environmental feature set, including the following steps: Step S211: using dynamic time window technology to extract dynamic features from the meteorological data in the initial forest fire data set to obtain meteorological dynamic features; Step S212: manually extracting terrain features from the terrain data in the initial forest fire data set, and merging the meteorological dynamic features and the terrain features to obtain an environmental feature set; Step S22: constructing a forest image feature set, specifically using an improved Xception model to construct a forest image feature set to obtain a forest image feature set, including the following steps: Step S221: construct a traditional Xception model architecture; Step S222: improving the traditional Xception model architecture to obtain an improved Xception model, specifically, selecting a Swish activation function to replace the ReLU activation function in the traditional Xception model, and using a one-dimensional convolutional layer to replace the global average pooling layer in the traditional Xception model to prevent channel overlap and improve the ability to capture subtle features; The calculation formula of the Swish activation function is: ； Where Swish(·) is the Swish activation function, z is the input parameter, specifically the value input to the Swish activation function, and Sig(·) is the Sigmoid function; Step S223: extracting features from the forest remote sensing image data in the forest fire initial data set through the improved Xception model to obtain a forest image feature set; Step S23: constructing a forest fire feature set, specifically, selecting important features from the meteorological data, terrain data and historical fire data in the forest fire initial data set through the information gain feature selection method to obtain the forest fire feature set, including the following steps: Step S231: merging the meteorological data, terrain data and historical fire data in the initial forest fire data set to obtain a forest fire numerical data set; Step S232: calculating information entropy, specifically calculating information entropy of the target variable; Step S233: calculating the conditional entropy, specifically calculating the conditional entropy of the target variable; Step S234: Calculate information gain, which is used to measure the importance of characteristic factors. Specifically, the information gain of each characteristic factor in the forest fire numerical data set is calculated. The calculation formula is: ； In the formula, IG(Y, X) is the information gain of the characteristic factor, Y is the target variable, specifically whether a fire occurs or not, X is the characteristic factor, H(Y) is the information entropy of the target variable, and H(Y|X) is the conditional entropy of the target variable, which is used to measure the contribution of the characteristic factor to the target variable; Step S235: selecting important features, specifically, merging the first k feature factors with the highest information gain according to the information gain to obtain a forest fire feature set; Step S3: forest fire sensitivity prediction, specifically, using a long-term and short-term regularized random forest model to predict forest fire sensitivity based on the environmental feature set and the forest fire feature set; The long-short-term regularized random forest model specifically includes a long-short-term memory network block and a regularized random forest block. The long-short-term memory network block is used to extract time series features and generate a forest fire meteorological index. It specifically includes an input sublayer, a long-short-term memory network layer, a random inactivation layer, a fully connected layer, and an output sublayer. The regularized random forest block is used to combine the forest fire meteorological index and the forest fire feature set for prediction; In step S3, the forest fire sensitivity prediction includes the following steps: Step S31: constructing a long short-term memory network block, specifically inputting the environmental feature set into the long short-term memory network block, performing time series data modeling, and generating a forest fire meteorological index, including the following steps: Step S311: construct an input sublayer for inputting an environmental feature set; Step S312: constructing a long short-term memory network layer for capturing the long-term dependency of time series data, wherein the long short-term memory network layer specifically includes a first long short-term network sublayer, a first random inactivation layer, a second long short-term network sublayer and a second random inactivation layer; Step S313: construct a random dropout layer to prevent overfitting; Step S314: construct a fully connected layer to integrate feature information; Step S315: constructing an output sublayer for generating a forest fire meteorological index; Step S32: constructing a regularized random forest block, specifically inputting the forest fire meteorological index and the forest fire feature set into the regularized random forest block, performing prediction, and generating a prediction result, including the following steps: Step S321: constructing an initial random forest, specifically, taking the forest fire meteorological index and the forest fire feature set as sensitivity prediction input samples, and generating Tr initial decision trees by randomly sampling the sensitivity prediction input samples to obtain an initial random forest, wherein the initial random forest includes Tr initial decision trees; Step S322: Calculate the average depth of each initial decision tree, add the average depth of each initial decision tree to obtain the total average depth, and construct a regularization term. The calculation formula is: ； In the formula, Rt is the regularization term, α is the regularization parameter, and ADepth total is the overall average depth; Step S323: Pruning the initial decision tree using a regularization method to obtain a regularized decision tree, specifically, comparing the average depth of the initial decision tree with the regularization term. If the average depth of the initial decision tree is greater than the regularization term, pruning the initial decision tree; Step S324: Generate a prediction result by integrating the outputs of all regularized decision trees; Step S33: constructing a forest fire sensitivity prediction model, specifically, performing model training based on a long-term and short-term regularized random forest model to obtain a forest fire sensitivity prediction model; Step S34: performing forest fire sensitivity prediction using a forest fire sensitivity prediction model to obtain a forest fire sensitivity level; Step S4: predicting the location of forest fires, specifically, building a forest fire location prediction model based on a support vector machine according to a forest image feature set, predicting the location of forest fires, and obtaining forest fire location prediction information; In step S4, the forest fire location prediction includes the following steps: Step S41: constructing a support vector machine, specifically, inputting the forest image feature set into the support vector machine as a location prediction input sample, selecting a Gaussian radial basis function as a kernel function of the support vector machine, and constructing the support vector machine; Step S42: constructing a forest fire location prediction model, specifically using an improved square hinge loss function to train a support vector machine to obtain a forest fire location prediction model, including the following steps: Step S421: define an improved square hinge loss function, the calculation formula is: ； Where loss(·) is the improved square hinge loss function, c is the sample residual, which is used to represent the error between the model's prediction result of the location prediction input sample and the true label; Step S422: construct an objective function, the calculation formula is: ； Where f(w,b) is the objective function, w is the model weight vector, b is the model bias term, β is the regularization parameter, i is the location prediction input sample index, n is the number of location prediction input samples, and y is i is the true label of the input sample predicted at the i-th location, x i is the feature vector of the predicted input sample at the i-th location, is the model weight vector w and the feature vector x of the input sample predicted at the i-th location i The inner product of Step S423: Based on the objective function, the model parameters are updated using the gradient descent method, the optimal model parameters are calculated, and a forest fire location prediction model is obtained; Step S43: predicting the location of a forest fire using a forest fire location prediction model to obtain forest fire location prediction information; Step S5: Generate forest fire prediction report.

2. The method for predicting forest fires based on artificial intelligence according to claim 1, characterized in that: In step S5, the forest fire prediction report is generated, specifically by integrating the forest fire sensitivity level and forest fire location prediction information to generate a forest fire prediction report and provide forest fire prevention and control suggestions.

3. The method for predicting forest fires based on artificial intelligence according to claim 2, characterized in that: In step S1, the data set is constructed by collecting meteorological data, terrain data, historical fire data and forest remote sensing image data, and preprocessing them through missing value filling, noise removal, data enhancement and normalization to obtain an initial forest fire data set.

4. A forest fire prediction system based on artificial intelligence, used to implement a forest fire prediction method based on artificial intelligence as claimed in any one of claims 1 to 3, characterized in that: It includes a data set construction module, a feature set construction module, a forest fire sensitivity prediction module, a forest fire location prediction module and a forest fire prediction report generation module; The data set construction module is used for data set construction, and obtains the forest fire initial data set through data set construction, and sends the forest fire initial data set to the feature set construction module; The feature set construction module is used for feature set construction, and obtains an environmental feature set, a forest image feature set and a forest fire feature set through feature set construction, and sends the environmental feature set and the forest fire feature set to the forest fire sensitivity prediction module, and sends the forest image feature set to the forest fire location prediction module; The forest fire sensitivity prediction module is used for forest fire sensitivity prediction, obtains the forest fire sensitivity level through forest fire sensitivity prediction, and sends the forest fire sensitivity level to the forest fire prediction report generation module; The forest fire location prediction module is used for forest fire location prediction, obtains forest fire location prediction information through forest fire location prediction, and sends the forest fire location prediction information to the forest fire prediction report generation module; The forest fire prediction report generation module is used for forest fire prediction report generation. Through the forest fire prediction report generation, a forest fire prediction report is obtained and forest fire prevention and control suggestions are provided.

Images