Bursaphelenchus xylophilus disease transmission risk prediction method based on data model hybrid drive

Abstract

The invention discloses a data model hybrid driven pine wood nematode disease transmission risk prediction method, and belongs to the technical field of pine wood nematode disease monitoring. The objective of the invention is to improve pine wood nematode disease transmission risk prediction precision. The method comprises the following steps: constructing a multi-source factor data matrix; according to the propagation law of the pine wood nematode disease, a cellular automaton model is constructed, and the spatial diffusion process of the pine wood nematode disease in each county-level administrative division geographic unit in the forest is calculated and simulated; constructing a TabNet neural network model, performing feature learning on the multi-source factor data matrix, and identifying key driving factors of pine wood nematode disease transmission; a PINN physical information neural network framework is constructed, a cellular automaton model is used as a physical constraint module, a TabNet neural network model is used as a data driving module, and the physical constraint module and the data driving module are linked through a composite loss function; and using the trained PINN physical information neural network framework to predict the propagation and diffusion trend of the pine wood nematode disease in the future and realize visualization.

Description

Technical Field

[0001] This invention belongs to the field of pine wilt disease monitoring technology, specifically involving a data model-driven method for predicting the spread risk of pine wilt disease. Background Technology

[0002] Pine wilt disease (PWD) is a devastating forestry quarantine disease caused by the pine wilt nematode (Bursaphelenchus xylophilus), characterized by rapid spread, high mortality, and wide-ranging damage. It is transmitted between pine trees by vector insects carrying the nematodes and can spread over long distances during timber transport, logging, and distribution. Since its introduction to my country, the disease has been spreading continuously in several provinces, posing a serious threat to ecological security and pine forest resources. Therefore, a highly accurate method for predicting the spread risk of pine wilt disease is urgently needed. Summary of the Invention

[0003] The problem this invention aims to solve is to improve the accuracy of predicting the spread risk of pine wilt disease, and proposes a data model-driven method for predicting the spread risk of pine wilt disease.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A data model-driven method for predicting the transmission risk of pine wilt disease includes the following steps:

[0006] S1. Collect baseline data and influencing factor data on pine wilt disease occurrence under the geographical units of county-level administrative divisions across the country, and construct a multi-source factor data matrix;

[0007] S2. Based on the transmission pattern of pine wilt disease, a cellular automata model was constructed to calculate and simulate the spatial diffusion process of pine wilt disease in the forest of each county-level administrative geographical unit;

[0008] S3. Construct a TabNet neural network model, perform feature learning on the multi-source factor data matrix obtained in step S1, and identify the key driving factors for the spread of pine wilt disease;

[0009] S4. Construct the PINN physical information neural network framework, using the cellular automata model as the physical constraint module and the TabNet neural network model as the data driving module. Link the physical constraint module and the data driving module through a composite loss function. Train the model based on the multi-source factor data matrix obtained in step S1 to obtain the trained PINN physical information neural network framework.

[0010] S5. Use the trained PINN physical information neural network framework to predict and visualize the future spread of pine wilt disease.

[0011] Furthermore, the specific implementation method of step S1 includes the following steps:

[0012] S1.1. Collect historical baseline data on pine wilt disease, including the disease occurrence range and dominant tree species information over the years, and encode them according to county-level administrative regions, setting the label for annual occurrence as 1 and the label for no occurrence as 0;

[0013] S1.2. Collect multi-source heterogeneous influencing factor data related to pine wilt disease, including natural environmental factors and anthropogenic factors. The natural environmental factors include NDVI, annual maximum temperature, annual average wind speed, annual average precipitation, DEM, and solar radiation. The anthropogenic factor is road density.

[0014] Then construct the multi-source factor data matrix, denoted as... :

[0015]

[0016] in, Let represent the j-th influence factor of the i-th county unit, d be the total number of features of the influence factor, and N be the total number of samples;

[0017] Then, the multi-source factor data matrix is ​​batch normalized to obtain the expression:

[0018]

[0019] in, and These are the mean and standard deviation, respectively. This is the multi-source factor data matrix after batch normalization.

[0020] Furthermore, the specific implementation method of step S2 includes the following steps:

[0021] S2.1. State definition: Set each county-level unit cell as Ci, which has a discrete state Si(t)∈{0,1,2} at any time t, corresponding to three types: healthy county, neighboring county, and epidemic county, respectively;

[0022] The initial state of the system is infected cells, that is, the epidemic county is determined based on historical data, where the initial state of Ci is determined by the historical baseline survey data of pine wilt disease in step S1;

[0023] S2.2. Defining the transmission rule: For neighboring counties adjacent to the affected county, the transmission rate is determined by susceptibility, and the formula for calculating the transmission probability is:

[0024]

[0025] in, Let represent the probability that the i-th county-level unit, currently in a susceptible state, changes from a healthy state to an infected state during the model evolution at time t. The spatial explanatory power of the normalized main influencing factors obtained from the single-factor detection function in the geographic detector model. denoted as the influencing factor, where n is the number of infected county-level units adjacent to susceptible county-level unit i and potentially having a transmission impact on it;

[0026] S2.3. State Update and Iteration: Update the state of each county-level unit cell Ci according to the calculation results of step S2.2, and use it as the initial state for the next iteration. Repeat step S2.2 to substitute the updated state into the next round of simulation until the preset simulation time is reached.

[0027] Furthermore, the TabNet neural network model constructed in step S3 adopts a three-layer TabNet structure, where each layer includes a feature transformer and an attention transformer.

[0028] Feature transformer 1 has an input dimension of 7 and an output dimension of 64; attention transformer 1 has an input dimension of 64 and an output dimension of 7.

[0029] Feature transformer 2 has an input dimension of 64 and an output dimension of 64; attention transformer 2 has an input dimension of 64 and an output dimension of 7.

[0030] The input dimension of feature transformer 3 is 64 and the output dimension is 32; the input dimension of attention transformer 3 is 32 and the output dimension is 7.

[0031] Furthermore, the specific implementation method of step S4 includes the following steps:

[0032] S4.1. Construct the PINN physical information neural network framework, using a cellular automata model as the physical constraint module and a TabNet neural network model as the data-driven module. Joint optimization is performed using a composite loss function. The overall mapping relationship of the model is expressed as:

[0033]

[0034] in, For physical propagation rule parameters, For the nonlinear mapping function of the neural network, The predicted risk probability output by the model;

[0035] S4.2. At the physical level, the spatial propagation of pine wilt disease is represented by cellular automata rules, resulting in:

[0036]

[0037] in, This represents the infection status of county-level unit i at time t. For the neighborhood set of county-level units, This represents the propagation influence weight of county-level unit i on county-level unit j. Indicates environmental factors, Indicates human factors; The propagation function defines the physical laws governing the spread of pine wilt disease from the infected area to adjacent areas and then to the uninfected area.

[0038] S4.3. Configure the TabNet neural network model to automatically learn the nonlinear relationships between multidimensional input features. The forward propagation expression is:

[0039]

[0040] in, For the parameterization mapping of the TabNet network, A mask matrix is ​​selected for the features of the k-th layer to dynamically determine the key features. For activation function, and These are the first learnable parameter and the second learnable parameter, respectively.

[0041] S4.4. Set up end-to-end optimization using a composite loss function, defined as follows:

[0042]

[0043] in, This represents the error term between the model predictions and the observed data. For physical consistency constraints, For parameter regularization terms, and These are the balance coefficients of the physical consistency constraint term and the balance coefficients of the parameter regularization term, respectively.

[0044]

[0045]

[0046] in, To represent the real observational data used for training and validating the model, i.e., actual observations of the occurrence status of pine wilt disease collected from surveys, To represent the county-level unit state variables predicted by the model for the next time point t+1, This represents the initial state variable of the county-level unit at the current time.

[0047] By minimizing Parameter optimization is achieved so that the prediction results can reflect the statistical regularity of the observed data and follow the physical constraints of pine wilt disease transmission.

[0048] Furthermore, in step S4, the counties with outbreaks are used as positive samples, and the remaining healthy counties are used as negative samples. The dataset is divided into training set, validation set and test set in a ratio of 8:1:1.

[0049] Furthermore, step S5 uses the trained PINN physical information neural network framework to predict the occurrence of pine wilt disease at the county-level across the country in future years based on historical occurrences. In ArcGIS software, the county codes are matched and connected to generate a visual map, including the future spread range and risk level classification of pine wilt disease.

[0050] The beneficial effects of this invention are:

[0051] This invention presents a data model-driven method for predicting the spread risk of pine wilt disease. For the first time, it integrates a cellular automata (CA) propagation mechanism model with a TabNet neural network structure within the PINN (Physics-Informed Neural Network) framework. By introducing propagation mechanism constraints during training, it achieves a shift from traditional "data fitting" to "data learning under mechanism constraints," enhancing the model's physical consistency and biological plausibility.

[0052] This invention presents a data model-driven method for predicting the spread risk of pine wilt disease. This method jointly optimizes data loss and physical loss, enabling the model to accurately fit historical disease incidence data while conforming to the natural spread patterns of pine wilt disease. This mechanism effectively avoids the overfitting and logical deviation problems that occur with pure data models, improving prediction stability and generalization ability.

[0053] This invention presents a data model-driven method for predicting the spread of pine wilt disease. By leveraging the attention structure in the TabNet model, it assesses the feature importance of input multi-source heterogeneous influencing factor data, enabling automatic feature selection and visual interpretation. This mechanism not only improves the model's learning efficiency but also reveals key driving factors influencing the spread of pine wilt disease, providing a scientific basis for prevention and control strategies.

[0054] This invention discloses a data model-driven method for predicting the spread risk of pine wilt disease. By embedding spatial neighborhood update rules from cellular automata, this invention can simulate the spread of the disease in the temporal dimension, generating a spatial diffusion sequence of pine wilt disease. This process can be used to analyze the potential spread paths and speeds of the disease under different environmental conditions or management measures, demonstrating strong dynamic prediction capabilities.

[0055] This invention presents a data model-driven method for predicting the spread risk of pine wilt disease, enabling visualization of the predicted spread trend. Based on the model output, spatial visualization is performed using the ArcGIS platform, matching county-level administrative divisions to create maps of future spread range, risk level classification, and dominant spread paths. This assists managers in implementing targeted prevention and control strategies, demonstrating significant practical value and potential for widespread application. Attached Figure Description

[0056] Figure 1 This is a flowchart of a data model-driven method for predicting the spread risk of pine wilt disease, as described in this invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described specific embodiments are merely some embodiments of the invention, and not all of them. The components of the specific embodiments of the invention described and shown in the accompanying drawings can typically be arranged and designed in various different configurations, and the invention may also have other embodiments.

[0058] Therefore, the following detailed description of specific embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected specific embodiments of the invention. All other specific embodiments obtained by those skilled in the art based on these specific embodiments without inventive effort are within the scope of protection of this invention.

[0059] To further understand the invention's content, features, and effects, the following specific embodiments are provided, along with accompanying drawings. Figure 1 Detailed explanation is as follows:

[0060] Example 1:

[0061] A data model-driven method for predicting the transmission risk of pine wilt disease includes the following steps:

[0062] S1. Collect baseline data and influencing factor data on pine wilt disease occurrence under the geographical units of county-level administrative divisions across the country, and construct a multi-source factor data matrix;

[0063] Furthermore, the specific implementation method of step S1 includes the following steps:

[0064] S1.1. Collect historical baseline data on pine wilt disease, including the disease occurrence range and dominant tree species information over the years, and encode them according to county-level administrative regions, setting the label for annual occurrence as 1 and the label for no occurrence as 0;

[0065] S1.2. Collect multi-source heterogeneous influencing factor data related to pine wilt disease, including natural environmental factors and anthropogenic factors. The natural environmental factors include NDVI, annual maximum temperature, annual average wind speed, annual average precipitation, DEM, and solar radiation. The anthropogenic factor is road density.

[0066] Then construct the multi-source factor data matrix, denoted as... :

[0067]

[0068] in, Let represent the j-th influence factor of the i-th county unit, d be the total number of features of the influence factor, and N be the total number of samples;

[0069] Then, the multi-source factor data matrix is ​​batch normalized to obtain the expression:

[0070]

[0071] in, and These are the mean and standard deviation, respectively. This is the multi-source factor data matrix after batch normalization.

[0072] Furthermore, the data collection spanned from 2000 to 2023.

[0073] S2. Based on the transmission pattern of pine wilt disease, a cellular automata model was constructed to calculate and simulate the spatial diffusion process of pine wilt disease in the forest of each county-level administrative geographical unit;

[0074] Based on the transmission characteristics of pine wilt disease and the spatial pattern of the study area, a cellular automata model based on vector units was constructed. The model uses county-level administrative regions as irregular polygonal cells and years as discrete time steps to simulate the spatiotemporal transmission process of pine wilt disease within the area.

[0075] Furthermore, the specific implementation method of step S2 includes the following steps:

[0076] S2.1. State definition: Set each county-level unit cell as Ci, which has a discrete state Si(t)∈{0,1,2} at any time t, corresponding to three types: healthy county, neighboring county, and epidemic county, respectively;

[0077] The initial state of the system is infected cells, that is, the epidemic county is determined based on historical data, where the initial state of Ci is determined by the historical baseline survey data of pine wilt disease in step S1;

[0078] S2.2. Defining the transmission rule: For neighboring counties adjacent to the affected county, the transmission rate is determined by susceptibility, and the formula for calculating the transmission probability is:

[0079]

[0080] in, Let represent the probability that the i-th county-level unit, currently in a susceptible state, changes from a healthy state to an infected state during the model evolution at time t. The spatial explanatory power of the normalized main influencing factors obtained from the single-factor detection function in the geographic detector model. denoted as the influencing factor, where n is the number of infected county-level units adjacent to susceptible county-level unit i and potentially having a transmission impact on it;

[0081] S2.3. State Update and Iteration: Update the state of each county-level unit cell Ci according to the calculation results of step S2.2, and use it as the initial state for the next iteration. Repeat step S2.2 to substitute the updated state into the next round of simulation until the preset simulation time is reached.

[0082] S3. Construct a TabNet neural network model, perform feature learning on the multi-source factor data matrix obtained in step S1, and identify the key driving factors for the spread of pine wilt disease;

[0083] Furthermore, the TabNet neural network model constructed in step S3 adopts a three-layer TabNet structure, where each layer includes a feature transformer and an attention transformer.

[0084] Feature transformer 1 has an input dimension of 7 and an output dimension of 64; attention transformer 1 has an input dimension of 64 and an output dimension of 7.

[0085] Feature transformer 2 has an input dimension of 64 and an output dimension of 64; attention transformer 2 has an input dimension of 64 and an output dimension of 7.

[0086] The input dimension of feature transformer 3 is 64 and the output dimension is 32; the input dimension of attention transformer 3 is 32 and the output dimension is 7.

[0087] Furthermore, the TabNet model is used for feature learning of multi-source influencing factors. The model assigns weights to different features through a sequential attention mechanism, automatically selecting key driving factors and performing deep embedding representations. The specific technical solution of the TabNet model is as follows:

[0088] TabNet uses an attention mask at each decision step t. Feature selection is performed, and the mask is generated by the attention converter using the following expression:

[0089]

[0090] in, Use probabilities for the features from the previous step. This is the result of the feature transformation of the previous layer. For trainable weight matrix, Representing element-wise multiplication, Sparsemax is a sparse normalization function that sets some feature weights to zero to achieve automatic feature selection.

[0091] The expression for extracting high-level semantic features from the selected features using a feature transformer is as follows:

[0092]

[0093] in, and Here are the weights and biases for this layer, and ReLU is the activation function.

[0094] Decision characteristics of each step's output The convergence forms the final prediction as follows:

[0095]

[0096] in, Here, T represents the learnable step weights, and T represents the number of decision steps.

[0097] The final propagation risk prediction result is obtained through a fully connected layer and a sigmoid activation function, expressed as:

[0098]

[0099] in, The sigmoid function is used to output the propagation probability. , representing the infection risk of the i-th county in the next year.

[0100] We use weighted binary cross-entropy loss to adjust for the imbalance between positive and negative samples, as expressed in the following expression:

[0101]

[0102] in, To improve the ability to identify disease-affected samples (label=1).

[0103] The TabNet model outputs the selection probability of each feature at each step through an attention mask matrix, enabling interpretable analysis of transmission risk. The global feature importance can be obtained by summarizing these probabilities, thereby identifying key driving factors in the spread of pine wilt disease.

[0104] A three-layer TabNet structure is adopted, and the specific process is shown in step S3. Each layer contains a feature transformer and an attention transformer. The input and output dimensions and key parameters of each layer are shown in Table 1.

[0105] Table 1

[0106]

[0107] S4. Construct the PINN physical information neural network framework, using the cellular automata model as the physical constraint module and the TabNet neural network model as the data driving module. Link the physical constraint module and the data driving module through a composite loss function. Train the model based on the multi-source factor data matrix obtained in step S1 to obtain the trained PINN physical information neural network framework.

[0108] Furthermore, the specific implementation method of step S4 includes the following steps:

[0109] S4.1. Construct the PINN physical information neural network framework, using a cellular automata model as the physical constraint module and a TabNet neural network model as the data-driven module. Joint optimization is performed using a composite loss function. The overall mapping relationship of the model is expressed as:

[0110]

[0111] in, For physical propagation rule parameters, For the nonlinear mapping function of the neural network, The predicted risk probability output by the model;

[0112] S4.2. At the physical level, the spatial propagation of pine wilt disease is represented by cellular automata rules, resulting in:

[0113]

[0114] in, This represents the infection status of county-level unit i at time t. For the neighborhood set of county-level units, This represents the propagation influence weight of county-level unit i on county-level unit j. Indicates environmental factors, Indicates human factors; The propagation function defines the physical laws governing the spread of pine wilt disease from the infected area to adjacent areas and then to the uninfected area.

[0115] S4.3. Configure the TabNet neural network model to automatically learn the nonlinear relationships between multidimensional input features. The forward propagation expression is:

[0116]

[0117] in, For the parameterization mapping of the TabNet network, A mask matrix is ​​selected for the features of the k-th layer to dynamically determine the key features. For activation function, and These are the first learnable parameter and the second learnable parameter, respectively.

[0118] S4.4. Set up end-to-end optimization using a composite loss function, defined as follows:

[0119]

[0120] in, This represents the error term between the model predictions and the observed data. For physical consistency constraints, For parameter regularization terms, and These are the balance coefficients of the physical consistency constraint term and the balance coefficients of the parameter regularization term, respectively.

[0121]

[0122]

[0123] in, To represent the real observational data used for training and validating the model, i.e., actual observations of the occurrence status of pine wilt disease collected from surveys, To represent the county-level unit state variables predicted by the model for the next time point t+1, This represents the initial state variable of the county-level unit at the current time.

[0124] By minimizing Parameter optimization is achieved so that the prediction results can reflect the statistical regularity of the observed data and follow the physical constraints of pine wilt disease transmission.

[0125] Furthermore, in step S4, counties with outbreaks are used as positive samples, and other healthy counties are used as negative samples. The dataset is divided into training, validation, and test sets in an 8:1:1 ratio. The overall model structure consists of four parts, including the input layer, data-driven module, physical constraint module, and loss function. Input layer: The input data uses the county-level scale as the basic prediction unit and includes natural environmental factors (NDVI, annual maximum temperature, annual average wind speed, annual average precipitation, DEM, solar radiation) and anthropogenic factors (road density), totaling seven features.

[0126] S5. Use the trained PINN physical information neural network framework to predict and visualize the future spread of pine wilt disease.

[0127] Furthermore, step S5 uses the trained PINN physical information neural network framework to predict the occurrence of pine wilt disease at the county-level across the country in future years based on historical occurrences. In ArcGIS software, the county codes are matched and connected to generate a visual map, including the future spread range and risk level classification of pine wilt disease.

[0128] This embodiment overcomes the problems of strong parameter dependence of mechanistic models, insufficient interpretability of data models, and loose coupling of hybrid models in existing methods for predicting the spread of pine wilt disease. It proposes a data model-driven hybrid approach for predicting the risk of pine wilt disease transmission. Under a unified framework of physical constraints and data-driven approaches, this method achieves high-precision prediction and interpretable analysis of the pine wilt disease transmission process. The core idea of ​​this embodiment is to organically integrate the cellular automata (CA) propagation mechanism model with the TabNet neural network model, and achieve collaborative optimization of the model through the Physical Information Neural Network (PINN) framework. This allows the learning of propagation laws and multi-source features to complement each other, thereby achieving a balance between theoretical interpretability and prediction accuracy.

[0129] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0130] Although this application has been described above with reference to specific embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of this application. In particular, as long as there is no structural conflict, the features in the specific embodiments disclosed in this application can be combined with each other in any way. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, this application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A data model hybrid-driven method for predicting the risk of spread of pine wilt disease, characterized by, Comprise the following steps: S1. Collect the background data and influence factor data of pine wood nematode disease under the county-level administrative division geographical unit, and construct a multi-source factor data matrix; S2. According to the transmission rule of pine wood nematode disease, a cellular automaton model is constructed to calculate and simulate the spatial diffusion process of pine wood nematode disease in each county-level administrative division geographical unit in the forest; S3. Construct a TabNet neural network model to learn the characteristics of the multi-source factor data matrix obtained in step S1, and identify the key driving factors of the spread of pine wood nematode disease; S4. Construct a PINN physical information neural network framework, use the cellular automaton model as the physical constraint module, and use the TabNet neural network model as the data driven module. Link the physical constraint module and the data driven module through the composite loss function, train the model based on the multi-source factor data matrix obtained in step S1, and obtain the trained PINN physical information neural network framework; S5. Use the trained PINN physical information neural network framework to predict the future spread and diffusion trend of pine wood nematode disease and realize visualization.

2. The data model hybrid-driven method for predicting the spread of pine wilt disease according to claim 1, wherein, The specific implementation method of step S1 comprises the following steps: S1.

1. Collect the historical background survey data of pine wood nematode disease, including the range of disease occurrence in previous years, the information of dominant tree species, encode according to the county-level administrative division, set the label of annual occurrence as 1, and the label of non-occurrence as 0; S1.

2. Collect multi-source heterogeneous influence factor data related to pine wood nematode disease, including natural environment factors and human factors. The natural environment factors include NDVI, annual maximum temperature, annual average wind speed, annual average precipitation, DEM, and solar radiation. The human factor is road density; Then construct the multi-source factor data matrix denoted as : ； wherein, denotes the jth influencing factor of the ith county unit, d is the total number of characteristics of the influencing factor, and N is the total number of samples; Then the multi-source factor data matrix is batch normalized to obtain the expression: ； wherein, and are the mean and standard deviation, respectively, is the batch-normalized multi-source factor data matrix.

3. The data model hybrid-driven method of predicting the risk of spread of pine wilt disease according to claim 2, wherein, The specific implementation method of step S2 comprises the following steps: S2.

1. State definition: set each county unit cell as Ci, which has a discrete state Si(t)∈{0,1,2} at any time t, corresponding to healthy county, neighborhood county and epidemic county respectively; The initial state of the system infected cell, that is, the epidemic county is determined according to the historical data, wherein the initial state of Ci is determined by the historical background survey data of pine wood nematode disease in step S1; S2.

2. Define the transmission rule: for the neighborhood county adjacent to the epidemic county, the transmission rate is determined by the susceptibility, and the calculation formula of the transmission probability is: ； wherein, Pi(t) is the probability of the i-th county unit currently in the susceptible state being transferred from the healthy state to the infected state in the model evolution process at time t, is the spatial explanatory power of the normalized main influencing factor obtained by the single-factor detection function in the geographical detector model, represents the influencing factor, and n is the number of infected county units adjacent to the susceptible county unit i and possibly having a propagation impact on the susceptible county unit i; S2.

3. State update and iteration: update the state of each county unit cell Ci according to the calculation result of step S2.2, and use it as the initial state of the next iteration. Repeat step S2.2 to substitute the updated state into the next round of simulation until the preset simulation time is reached.

4. The data model hybrid-driven method of predicting the risk of spread of pine wilt disease according to claim 3, wherein, The TabNet neural network model constructed in step S3 adopts a three-layer TabNet structure, wherein each layer includes a feature transformer and an attention transformer; The input dimension of feature transformer 1 is 7, and the output dimension is 64; the input dimension of attention transformer 1 is 64, and the output dimension is 7; The input dimension of feature transformer 2 is 64, and the output dimension is 64; the input dimension of attention transformer 2 is 64, and the output dimension is 7; The input dimension of the feature transformer 3 is 64, and the output dimension is 32; the input dimension of the attention transformer 3 is 32, and the output dimension is 7.

5. The data model hybrid-driven method of predicting the risk of spread of pine wilt disease according to claim 4, wherein, The specific implementation method of step S4 includes the following steps: S4.

1. Constructing a PINN physical information neural network framework, taking a cellular automaton model as a physical constraint module and a TabNet neural network model as a data-driven module, and performing joint optimization through a composite loss function, the overall mapping relationship of the model is represented as: ； wherein, is a physical propagation rule parameter, is a non-linear mapping function of a neural network, is a predicted risk probability of the model output; S4.

2. Set at the physical level, the spatial propagation of pine wilt disease is represented by the cellular automaton rule, and obtain: ； wherein, denotes the infection status of county unit i at time t, is a neighborhood set of county units, denotes the propagation influence weight of county unit i on county unit j, denotes the environmental factor, denotes the human factor; is a propagation function, which defines the physical law of the spread of pine wilt disease from the infected area to the adjacent area to the uninfected area; S4.

3. Set the TabNet neural network model to automatically learn the nonlinear relationship between multi-dimensional input features, and the forward propagation expression is: ； wherein, is a parameterized mapping of a TabNet network, is a feature selection mask matrix of the kth layer for dynamically determining key features, is an activation function, and are first and second learnable parameters, respectively. S4.

4. Set to optimize end-to-end through a composite loss function, and the composite loss function is defined as: ； wherein, represents an error term between the model prediction and the observation data, is a physical consistency constraint term, is a parameter regularization term, and are balancing coefficients of the physical consistency constraint term and the parameter regularization term, respectively. ； ； wherein, is to represent the true observation data for training and validating the model, i.e. the actual observation value of the pine wilt disease occurrence status collected from the survey, is to represent the model-predicted next-time t+1-year county-level unit state variable, is to represent the initialization state variable of the county-level unit at the current time t; by minimizing Parameter optimization is implemented so that the prediction results can reflect both the statistical regularity of the observation data and the physical constraints of the spread of pine wilt disease.

6. The data model hybrid-driven method of predicting the risk of spread of pine wilt disease according to claim 5, wherein, In step S4, the epidemic counties are taken as positive samples, and the remaining healthy counties are taken as negative samples. The data set is divided into a training set, a validation set and a test set in a ratio of 8:1:

1.

7. The data model hybrid-driven method of predicting the risk of spread of pine wilt disease according to claim 6, wherein, Step S5 uses the trained PINN physical information neural network framework to predict the occurrence of pine wilt disease at the county level nationwide in future years based on historical occurrences. In the Arcgis software, the county codes are matched and connected to generate a visual map, including the future propagation range of pine wilt disease and the risk level division.

Images