Mouse harm degree prediction method

Abstract

The invention relates to the technical field of murine hazard monitoring and prevention, and discloses a murine hazard degree prediction method, which comprises the following steps: data acquisition and primary processing: a data input module acquires 13 original indexes of a target murine living area, vegetation forest age and historical hazard degree are divided into a light grade, a middle grade and a heavy grade, and the data input module outputs the 13 original indexes of the target murine living area; the harm degrees of the three levels are respectively assigned to 1-3 levels. According to the mouse hazard degree prediction method, a prediction model collects 13 kinds of data including environment, vegetation and historical hazard degree through a data acquisition module, key factors influencing the mouse hazard degree are comprehensively covered, the environment data includes 11 indexes, soil components and climate parameters reflect basic conditions of mouse survival, and the environment data includes 11 indexes including 11 indexes, and vegetation forest age and historical hazard degree data respectively reflect food resource and population dynamic information, and a multi-source data fusion mode can more comprehensively and accurately reflect the actual situation of rat hazard.

Description

Technical Field

[0001] The present invention relates to the technical field of rodent hazard monitoring and prevention, and in particular to a method for predicting the degree of rodent hazard. Background Art

[0002] The management of harmful rodents has been a major problem in grassland resource protection and desert afforestation over the past few decades. However, a lack of understanding of the reproduction and outbreak patterns of rodents has brought huge challenges to protecting desertification management results and maintaining ecological balance. Among them, the great gerbil is a harmful rodent distributed in Central Asia and is currently the most common rodent pest. Its harmfulness is mainly reflected in two aspects. On the one hand, the great gerbil feeds on desert vegetation, causing vegetation to lose water and die. The digging of burrows has led to a significant decrease in surface plant coverage, exacerbating regional desertification. On the other hand, the great gerbil is the main host of rat plague, leishmaniasis, and Karimabad virus, posing a great threat to human health. Therefore, the prediction of rodents is very important for grassland resource protection and desert afforestation. However, the existing rodent damage monitoring model has some defects, such as: At present, surveys on giant gerbils mainly rely on manual surveys, namely mousetrap or damaged plant surveys. Not only is the work efficiency extremely low, but there is also a great risk of plague infection. Further manual use of infrared cameras in monitoring is limited by high equipment costs and subsequent maintenance pressure, making it difficult to promote and apply them in large desert areas. Existing models are mostly applicable to specific desert areas, and have weak generalization capabilities for complex environments, namely mixed areas of grassland, Gobi, and desert. In addition, the sample size is insufficient, making it difficult to meet large-scale monitoring needs.

[0003] In response to the above problems, there is an urgent need to carry out innovative designs based on the original rodent damage prediction model. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for predicting the degree of damage caused by rodents, so as to solve the problem proposed in the above background technology that the investigation of giant gerbils mainly relies on manual investigation, which has extremely low work efficiency and a great risk of plague infection. In the process of further manual monitoring using infrared cameras, it is limited by the high equipment cost and subsequent maintenance pressure, and it is difficult to promote and apply it in large desert areas. The existing models are mostly applicable to specific desert areas, have weak generalization ability for complex environments, and have insufficient sample size, making it difficult to meet large-scale monitoring needs.

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a method for predicting the degree of rodent damage, comprising data collection and preliminary processing: a data input module collects 13 original indicators of the target rodent living area, among which the vegetation age and historical damage degree are divided into light, medium and heavy levels, and the three levels of damage are assigned values of 1-3 respectively; Data preprocessing: Due to the wide variety of model factors in the above steps, principal component analysis is required before building the model to reduce the collinearity of the model. At this time, the data preprocessing module will collect the data to form the original data matrix of variable sampling. ; Then standardize the original data matrix of variable sampling to obtain the standardized matrix ; Solving for the normalized matrix The correlation coefficient matrix The characteristic equation of the eigenvalue and eigenvector is obtained. To solve the characteristic equation and eigenvector, first calculate the characteristic equation ,Will The eigenvalues are arranged in descending order, that is, ; Then calculate , respectively substitute the eigenvalues Find the corresponding ; Then, the data of the above indicators are used to preliminarily estimate the number of principal components. The number of principal components selected should make the cumulative contribution rate reach 85%~95%. The number of principal components selected depends on the cumulative contribution rate of the principal components. By calculating the principal component contribution rate and cumulative contribution rate ,Finally, the principal component loadings and principal component scores are calculated; Arrange the principal component characteristic root contribution rates calculated above in descending order, select six principal components with cumulative contribution rates ≥ 85%, generate the eigenvectors after dimensionality reduction, and construct the ELM learning samples; PSO-ELM model construction and training: Initialize the PSO algorithm parameters in the PSO-ELM model construction module and input the ELM learning samples into the PSO-ELM model; The inverse of the mean square error of the training set is used as the fitness function to calculate the fitness of each particle and determine the optimal position each particle passes through. and the optimal location found by the group , by tracking and updating these two optimal positions, the particle velocity and position are updated; The updating process is repeated until the upper limit of the number of iterations is reached or the fitness meets the accuracy requirement. The optimized input weights and thresholds are used to train a single hidden layer feedforward neural network using ELM. The number of nodes in the ELM hidden layer is set to 25, and the activation function uses the sigmoid function. Model prediction and comparative evaluation: The comparative evaluation module receives the PSO-ELM model trained by the PSO-ELM model construction module, processes the input data, and outputs a predicted value of the rodent damage level of 1-3; The BP neural network is introduced as a control model, and the performance of the PSO-ELM model is verified by comparing the accuracy of the test set.

[0006] By adopting the above technical solution, by collecting 13 indicators in three categories, namely environment, vegetation, and historical damage level, we can comprehensively cover the key factors affecting the survival and reproduction of rodents, solve the limitations of traditional models that rely on a single factor, and improve the ecological fit of the prediction model.

[0007] Preferably, the original indicators in the data collection and preliminary processing include environment, vegetation and historical damage degree data, and the vegetation age and historical damage degree are converted into 1-3 level numerical variables to form original data.

[0008] By adopting the above technical solution, it is clear that the original indicators include environmental, vegetation and historical damage degree data, and the vegetation age and historical damage degree are converted into 1-3 level numerical variables, making the data more suitable for model processing, ensuring the consistency and standardization of the data, and facilitating model learning and analysis of the relationship between various factors and the degree of rodent damage.

[0009] Preferably, the original data matrix in the data preprocessing step is The formula is: ; Where: is the number of samples, is the number of indicators in the sample.

[0010] The above technical solution provides a clear mathematical structure for data organization and storage, making the data processing process clearer and more standardized, and facilitating the standardization of subsequent data.

[0011] Preferably, the correlation coefficient in the data preprocessing step is The calculation formula is: ; Where: express and The correlation coefficient between .

[0012] The above technical solution provides a calculation formula for the correlation coefficient R, which can accurately measure the correlation between various indicators, provide a basis for eliminating collinearity in subsequent principal component analysis, help screen out variables that have a greater impact on the degree of harm to rodents and are independent of each other, and improve the quality of model input data.

[0013] Preferably, the principal component contribution rate of the data preprocessing step is The formula is: ; Cumulative contribution rate The formula is .

[0014] The above technical solution ensures that the selected principal components can retain the information of the original data to the greatest extent possible, while reducing the data dimension, so that the model can reflect key information and have high computational efficiency.

[0015] Preferably, the principal component load in the data preprocessing step refers to the load coefficient corresponding to each principal component, and the calculation formula of the principal component load is: ; Substituting the original sample data or the standardized sample data into the above principal component expression can calculate the principal component score. The calculation formula is: .

[0016] Using the above technical solution, the original data is converted into more representative principal components through formulas. Principal components are new variables that integrate information from multiple original indicators, which help to extract potential features in the data and provide more effective input for subsequent model training.

[0017] Preferably, the model structure and the PSO algorithm parameters in training are: particle swarm size 20, maximum number of iterations 100, inertia weight 0.9, learning factor , speed range , position interval .

[0018] The above technical solution is used to determine the parameters of the PSO algorithm. The parameters are optimized and set, which helps the PSO algorithm achieve a better balance when searching for the optimal solution and ensures the global search capability, thereby better optimizing the ELM model and improving the prediction performance of the model.

[0019] Preferably, during the PSO algorithm optimization process in the PSO-ELM prediction model module, the particle velocity is updated by the formula: ; Update the particle position through the formula: ; Where: and are learning factors, representing individual experience and collective experience, also known as acceleration constants; and A random number between 0 and 1; is the search dimension, that is, the number of parameters that need to be optimized.

[0020] Using the above technical solution, the formulas for updating the particle velocity and position in the PSO algorithm are given, and the optimization process and mechanism of the algorithm are clarified. This enables the PSO algorithm to continuously adjust the search direction and step size based on the particle's own experience and the group's experience, gradually find a better solution, and provide more appropriate initial weights and thresholds for the ELM model.

[0021] Preferably, the PSO-ELM prediction model needs to initialize the particle group, substitute the input layer weight and hidden layer threshold corresponding to each particle into the ELM training algorithm, and the fitness function of the i-th particle is Expressed as the inverse of the mean square error; For each particle, its current fitness and In contrast, if This indicates that the current fitness is higher, and the individual's historical best position will be updated with the current position. , otherwise keep Unchanged, similarly compare the current fitness value and and ,when Update the global best position .

[0022] By adopting the above technical solution, the inverse of the mean square error is used as the fitness function, and the rules for particle fitness comparison and position update are clarified, so that the PSO algorithm can be optimized in the direction of reducing the mean square error and improving the model prediction accuracy, ensuring that the model continues to improve and enhance its performance during the training process.

[0023] Preferably, the model prediction and the prediction of the rodent hazard value in the comparative evaluation are based on the optimal input weights obtained by the PSO algorithm. and threshold Then, using the ELM training algorithm, substitute the formula , the model prediction value can be calculated.

[0024] By adopting the above technical solution, based on the optimal parameters obtained by the PSO algorithm and using the ELM training algorithm to calculate the model prediction value, it is possible to accurately output the predicted value of the degree of rodent damage of level 1-3, providing clear quantitative results for relevant management departments, facilitating the formulation of targeted prevention and control measures.

[0025] Compared with the prior art, the present invention has the following beneficial effects: the method for predicting the degree of damage caused by rodents: 1. This prediction model uses a data collection module to collect 13 types of data across three categories: environment, vegetation, and historical damage levels. These data comprehensively cover the key factors influencing rodent damage. Environmental data includes 11 indicators, ranging from soil composition to climate parameters, reflecting the basic conditions for rodent survival. Vegetation age and historical damage level data reflect food resources and population dynamics, respectively. This multi-source data fusion approach, compared to traditional prediction methods based on a single or limited number of factors, can more comprehensively and accurately reflect the actual situation of rodent damage, providing a rich data foundation for accurate prediction. 2. Due to the large amount of environmental factor data collected, principal component analysis (PCA) was used to preprocess the data. By constructing the original data matrix, standardizing the data, and solving the correlation coefficient matrix, six principal components with a cumulative contribution rate ≥ 85% were selected to generate the eigenvectors after dimensionality reduction. PCA dimensionality reduction effectively eliminates collinearity between indicators, reduces data dimensions, simplifies the model structure, and improves model training efficiency. For example, in actual data processing, after dimensionality reduction of the original 13-dimensional data, while retaining 88.01% of key information, it reduced data redundancy and computational complexity, significantly increasing model training speed and avoiding prediction bias caused by indicator correlation.

[0026] 3. This prediction model uses the PSO algorithm to optimize it to address the problem of training accuracy and time being affected by the random determination of the initial input weights and thresholds of the ELM. The PSO algorithm simulates the foraging behavior of a flock of birds to initialize the particle population, using the inverse of the mean square error of the training set as the fitness function, continuously updating the particle speed and position to find the optimal solution. Each particle represents a set of initial weights and thresholds for the ELM. Through continuous iterative optimization, the ELM model can quickly converge to a better solution during training. The final constructed PSO-ELM model combines the global search capability of PSO with the advantages of ELM's fast training speed, low computational complexity, and low overfitting probability, significantly improving prediction accuracy and stability.

[0027] 4. The comparative evaluation module in this prediction model incorporates a BP neural network as a control model. The performance of the PSO-ELM model is verified by comparing the accuracy of the test set. Under identical test conditions, a detailed comparison of the prediction results of the two models clearly demonstrates the superiority of the PSO-ELM model in predicting the extent of rodent damage. This comparative evaluation mechanism not only provides strong evidence for the reliability of the PSO-ELM model but also provides solid theoretical support for its promotion in practical applications, allowing users to more intuitively understand the advantages of the model and enhance their trust in it. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic diagram of the module flow of the rodent hazard degree prediction method of the present invention; Figure 2Schematic diagram of the PCA-PSO-ELM model process of the present invention; Figure 3 This is a schematic diagram of the distribution of giant gerbil monitoring points in northwest China according to the present invention; Figure 4 This is a schematic diagram of the principal component analysis correlation coefficient heat map and contribution rate curve of the present invention; Figure 5 Schematic diagram of the iterative convergence curve of the PSO algorithm of the present invention; Figure 6 Schematic diagram comparing the prediction results of the PSO-ELM and BP models of the present invention. DETAILED DESCRIPTION

[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0030] See also Figures 1-6 The present invention provides a technical solution: a method for predicting the degree of damage caused by rodents, comprising the following implementation operations: Data collection and preliminary processing of data input module: Multi-source indicator collection: Environmental data (11 items): Collect surface soil gravel content, altitude, surface soil silt fraction, surface soil organic carbon, pH value, annual average temperature, precipitation in the wettest month, annual precipitation, average temperature in the wettest quarter, precipitation in the highest quarter, and average temperature in the lowest quarter in the target rodent habitat; Vegetation data (1 item): Count the age of vegetation. Trees less than 5 years old are classified as young and assigned a value of 1. Trees between 5 and 10 years old are classified as middle-aged and assigned a value of 2. Trees older than 10 years old are classified as mature and assigned a value of 3. Historical damage level (1 item): The number of mice captured by mousetraps is 1 (less than 9 = 1), 2 (10-19 = 2), 3 (20 or more), and the plant damage rate is 1 (less than 30%), 2 (30%-49%), and 3 (50% or more). The more serious level is assigned a value of 1-3. Data matrix construction: forming the original data matrix , where n is the number of samples and p=13 is the number of indicators; Data standardization: Z-score standardization is performed on the original data matrix to eliminate the dimension effect and obtain a standardized matrix , the formula is: in For the The mean of the indicators, is the standard deviation; The data preprocessing module solves the correlation coefficient matrix: Calculate the correlation coefficient matrix of the standardized matrix , the formula is: ; in, express and The correlation coefficient between Eigenvalue and eigenvector calculation: Solving the characteristic equation , and get the eigenvalue , arranged in descending order and the corresponding eigenvectors ; Principal component screening: Calculate the principal component contribution and cumulative contribution rate , select the first 6 principal components with cumulative contribution rate ≥ 85% and generate a 6-dimensional feature vector; Principal component score calculation: By principal component loading and normalized data, calculating principal component scores , construct ELM learning samples; PSO-ELM model construction and training in the PSO-ELM model construction module: ELM Infrastructure: Input layer: 6-dimensional principal component features; Hidden layer: 25 nodes, activation function is sigmoid function: ; Output layer: 1-3 levels of harm, namely mild, moderate, and severe; PSO algorithm parameter initialization: The particle swarm size is 20, the maximum number of iterations is 100, the inertia weight is 0.9, and the learning factor , speed range , position interval ; Particle update and optimization: Speed update: ; Location Updates: ,in is the optimal position of an individual, is the global optimal position; Fitness function: expressed as the inverse of the mean square error of the training set, that is , update individual and global best positions in iterations; Model training in the model training and prediction module: Divide the dimensionality-reduced data into 80 training sets and 11 test sets, input the ELM model optimized by PSO, and determine the optimal input weight by minimizing MSE and hidden layer threshold ; The comparative evaluation module performs prediction and comparative evaluation on the model, and the prediction output is: Input the test set data and use ELM to calculate the output hazard level (level 1-3). The formula is: ,in is the hidden layer output matrix; Comparative experiment settings: BP neural network: hidden layer nodes 50, iterations 1000, learning rate 0.1, target accuracy 0.0001, maximum number of failures 200; Accuracy calculation: Comparing the prediction results of the two models on the test set, the accuracy of the PSO-ELM model reached 90.9%, and the BP model was 81.82%; The above model content is added to the embodiment, combined with the accompanying drawings of the specification Figure 3 As shown in the figure, the autumn damage degree prediction of giant gerbils is based on 91 sets of data from 46 monitoring points in four provinces / autonomous regions in northwestern China, namely Inner Mongolia Autonomous Region, Gansu Province, Ningxia Hui Autonomous Region, and Xinjiang Uygur Autonomous Region from 2020 to 2022. One set is missing due to the epidemic. The PSO-ELM model prediction process is implemented: Indicator definitions and data sources in the data input module: Environmental data (11 items): collected through weather stations and soil samplers: soil gravel content, altitude, soil silt content, soil organic carbon, pH value, annual average temperature, wet month precipitation, annual precipitation, wet season average temperature, warm season precipitation, and dry season average temperature; Vegetation data (1 item): Field survey of vegetation stand age: trees less than 5 years old are classified as young and assigned a value of 1; trees between 5 and 10 years old are classified as middle-aged and assigned a value of 2; trees older than 10 years old are classified as mature and assigned a value of 3, including but not limited to statistics on the age of Haloxylon ammodendron shrubs; Historical hazard level (1 item): Mousetrap method: Capture of less than 9 mice = Level 1 (mild), capture of 10-19 mice = Level 2 (moderate), capture of 20 mice or more = Level 3 (severe); Damage rate method: Damage rate of 100 plants is less than 30% = Level 1, damage rate of 30%-49% of plants = Level 2, damage rate of 50% or more of plants = Level 3; the more serious level is assigned; The data preprocessing module constructs the original data matrix: form Matrix, each row corresponds to a monitoring point, and each column corresponds to an indicator, such as the data of a point in Inner Mongolia Autonomous Region in 2020: ; Data standardization: in For the The mean of the indicators, is the standard deviation, eliminating the dimension effect; Correlation coefficient matrix calculation: Calculate the Pearson correlation coefficient matrix of the standardized matrix , PCA needs to be used to eliminate collinearity; Eigenvalue and Eigenvector Solving: Solving the Characteristic Equation , we get 13 eigenvalues, which are arranged in descending order as follows: , among which the cumulative contribution rate of the first 6 eigenvalues reached 88.01%; Principal component screening and score calculation: Select the first 6 principal components and calculate the principal component loads , is the eigenvector element, and the standardized data is substituted into the principal component score: ,Finally 91 groups of 6-dimensional feature vectors are generated as ELM input samples; ELM network structure in the PSO-ELM model building module: Input layer: 6 neurons (principal component features); Hidden layer: 25 neurons, activation function is sigmoid: ; Output layer: 3 neurons, corresponding to levels 1-3 of hazard, converted into probability output through softmax; PSO algorithm initialization: Parameter settings: particle swarm size 20, maximum number of iterations 100, inertia weight 0.9, learning factor , speed range , position interval , the search dimension is 2, input weights and hidden layer thresholds; Particle update and optimization process: Speed update: ,in is a random number between 0 and 1. is the optimal position of an individual, is the global optimal position; Location Updates: , update particle positions in iterations; Fitness calculation: the fitness of the i-th particle ,in, is the mean square error of ELM in the training set under the current parameters. By comparing and and Update the best position; The model training and prediction module implements training set division and iteration: the first 80 groups of data are used as training sets, and the last 11 groups are used as test sets; PSO iterates 100 times, converges around 60 times, and obtains the optimal input weights and hidden layer threshold , substitute into the ELM training algorithm; The comparative evaluation module performs prediction and comparative evaluation on the model. PSO-ELM prediction: Input the 6-dimensional features of the test set and calculate the hidden layer output , the hazard level prediction value is obtained through the output layer. For example, if the output of a test point is [0.1, 0.3, 0.6], it is judged as level 3 (severe); BP neural network comparison: Parameter settings: 50 hidden layer nodes, 1000 iterations, 0.1 learning rate, and 0.0001 target accuracy; Prediction results: Both high-risk test samples were predicted as level 2 (moderate), with an overall accuracy of 81.82%. PSO-ELM correctly predicted 10 / 11 groups, with an accuracy of 90.9%.

[0031] Working principle: When using this rodent damage prediction method, we first collect 13 indicators in three categories: environment, vegetation, and historical damage, covering the core factors affecting rodent survival. Then, we assign 1-3 levels to tree age and damage degree, convert qualitative data into quantitative input, adapt the machine learning model, and then construct a matrix of raw data. matrix; Z-score standardization was used to convert indicators of different dimensions into dimensionless data, avoiding model bias caused by differences in numerical ranges and ensuring the accuracy of principal component analysis. Principal component analysis was used to reduce the dimensionality of the 13-dimensional original data to 6 dimensions, retaining 88.01% of the key information, eliminating correlations between indicators and reducing model complexity. During data processing, a unified data dimension was used to avoid model bias caused by differences in indicator units and ensure the accuracy of principal component analysis. By principal component loading Construct linear combinations to convert standardized data into principal component scores , forming the input feature vector of the ELM model, reducing the complexity of the model while retaining key information, and inputting the processed data into the trained PSO-ELM model, which then outputs the predicted value of the rodent hazard level of 1-3; The BP neural network was introduced as a control model, and the performance of the PSO-ELM model was verified by comparing the accuracy of the test set. By identifying the key principal components, the model can reveal the main driving factors of rodent damage and provide a scientific basis for management departments. Among them, high-altitude and low-vegetation coverage areas need to be monitored in key areas, and ecological regulation or precise rodent control measures should be deployed in advance to balance ecological protection and rodent control.

[0032] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for predicting the degree of rodent damage, characterized in that: The following steps are involved: Data collection and preliminary processing: The data input module collects 13 raw indicators of the target rodent habitat. Among them, the vegetation age and historical damage level are divided into light, medium and severe levels, and the three levels of damage are assigned values of 1-3 respectively; Data preprocessing: Due to the wide variety of model factors in the above steps, principal component analysis is required before building the model to reduce the collinearity of the model. At this time, the data preprocessing module will collect the data to form the original data matrix of variable sampling. ; Then standardize the original data matrix of variable sampling to obtain the standardized matrix ; Solving for the normalized matrix The correlation coefficient matrix The characteristic equation of the eigenvalue and eigenvector is obtained. To solve the characteristic equation and eigenvector, first calculate the characteristic equation ,Will The eigenvalues are arranged in descending order, that is, ; Then calculate , respectively substitute the eigenvalues Find the corresponding ; Then, the data of the above indicators are used to preliminarily estimate the number of principal components. The number of principal components selected should make the cumulative contribution rate reach 85%~95%. The number of principal components selected depends on the cumulative contribution rate of the principal components. By calculating the principal component contribution rate and cumulative contribution rate ,Finally, the principal component loadings and principal component scores are calculated; Arrange the principal component characteristic root contribution rates calculated above in descending order, select six principal components with cumulative contribution rates ≥ 85%, generate the eigenvectors after dimensionality reduction, and construct the ELM learning samples; PSO-ELM model construction and training: Initialize the PSO algorithm parameters in the PSO-ELM model construction module and input the ELM learning samples into the PSO-ELM model; The inverse of the mean square error of the training set is used as the fitness function to calculate the fitness of each particle and determine the optimal position each particle passes through. and the optimal location found by the group , by tracking and updating these two optimal positions, the particle velocity and position are updated; The updating process is repeated until the upper limit of the number of iterations is reached or the fitness meets the accuracy requirement. The optimized input weights and thresholds are used to train a single hidden layer feedforward neural network using ELM. The number of nodes in the ELM hidden layer is set to 25, and the activation function uses the sigmoid function. Model prediction and comparative evaluation: The comparative evaluation module receives the PSO-ELM model trained by the PSO-ELM model construction module, processes the input data, and outputs the predicted value of the rodent damage level of 1-3; The BP neural network is introduced as a control model, and the performance of the PSO-ELM model is verified by comparing the accuracy of the test set.

2. The method for predicting the degree of rodent damage according to claim 1, wherein: The original indicators in the data collection and preliminary processing include environment, vegetation and historical damage degree data, and the vegetation age and historical damage degree are converted into 1-3 level numerical variables to form original data.

3. The method for predicting the degree of rodent damage according to claim 1, wherein: The original data matrix in the data preprocessing step The formula is: ； Where: is the number of samples, is the number of indicators in the sample.

4. The method for predicting the degree of rodent damage according to claim 1, wherein: Correlation coefficients in the data preprocessing step The calculation formula is: ； Where: express and The correlation coefficient between .

5. The method for predicting the degree of rodent damage according to claim 1, wherein: The principal component contribution rate of the data preprocessing step The formula is: ; Cumulative contribution rate The formula is .

6. The method for predicting the degree of rodent damage according to claim 1, wherein: The principal component load in the data preprocessing step refers to the load coefficient corresponding to each principal component. The calculation formula of the principal component load is: ； Substituting the original sample data or the standardized sample data into the above principal component expression can calculate the principal component score. The calculation formula is: 。 7. The method for predicting the degree of rodent damage according to claim 1, wherein: The model structure and PSO algorithm parameters in training are: particle swarm size 20, maximum number of iterations 100, inertia weight 0.9, learning factor , speed range , position interval .

8. The method for predicting the degree of rodent damage according to claim 1, wherein: During the PSO algorithm optimization process in the PSO-ELM prediction model module, the particle velocity is updated by the formula: ； Update the particle position through the formula: ； Where: and are learning factors, representing individual experience and collective experience, also known as acceleration constants; and A random number between 0 and 1; is the search dimension, that is, the number of parameters that need to be optimized.

9. The method for predicting the degree of rodent damage according to claim 8, wherein: The PSO-ELM prediction model requires initializing the particle population, substituting the input layer weight and hidden layer threshold corresponding to each particle into the ELM training algorithm, and the fitness function of the i-th particle is Expressed as the inverse of the mean square error; For each particle, its current fitness and In contrast, if This indicates that the current fitness is higher, and the individual's historical best position will be updated with the current position. , otherwise keep Unchanged, similarly compare the current fitness value and and ,when Update the global best position .

10. The method for predicting the degree of rodent damage according to claim 9, characterized in that: The model predicts the numerical value of rodent damage in the comparative evaluation, and the optimal input weights are obtained by the PSO algorithm. and threshold Then, using the ELM training algorithm, substitute the formula , the model prediction value can be calculated.

Images