Intelligent environmental impact assessment method and system

Abstract

The invention relates to the technical field of environmental impact assessment, in particular to an intelligent environmental impact assessment method and system, and the method comprises the following steps: collecting environmental data including pollutant discharge, water resource consumption, air quality change and biodiversity, building an impact factor data set, and segmenting the data set to obtain an impact factor data set; and calling a kernel function type to carry out numerical prediction on pollutant emission. According to the method, by integrating data processing and machine learning technologies, the accuracy and efficiency of environmental impact assessment are improved, numerical prediction is performed on environmental data by using a kernel function and a support vector regression model, the adaptability and precision of a prediction model are enhanced, model parameters are optimized through automatic population generation and crossover mutation operation, and the accuracy and efficiency of environmental impact assessment are improved. The model can better learn and adapt in a complex data environment, sensitivity and response speed to environmental changes are further improved through dynamic clustering analysis and time sequence adjustment, and scientific support is provided for formulating timely and effective environmental protection measures.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental impact assessment, and in particular to an intelligent environmental impact assessment method and system. Background Art

[0002] The field of environmental impact assessment technology mainly involves systematic analysis and data-driven assessment of certain environmental impacts. By predicting, analyzing, monitoring and managing the impact of these activities on the natural environment, environmental impact assessment technology provides a basis for scientific decision-making, helps formulate measures to mitigate negative impacts, and ensures the stability of ecosystems and sustainable development of mankind. This technical field combines environmental science, engineering technology, and policies and regulations to comprehensively assess the impact of air, water, soil and ecosystems, aiming to protect environmental quality and ensure that economic activities are coordinated with ecological protection.

[0003] Among them, the intelligent environmental impact assessment method refers to the application of artificial intelligence, data processing and automated analysis technology to improve the efficiency and accuracy of the environmental impact assessment process. Its main uses include the use of machine learning algorithms, big data analysis and intelligent prediction models to efficiently process various environmental data, thereby assisting decision makers to identify potential environmental risks in a timely and accurate manner, optimize response measures, and promote a balance between economic activities and environmental protection.

[0004] Existing technologies in environmental impact assessment mostly rely on static data analysis and lack the ability to dynamically adapt to environmental changes. This limits the real-time and accuracy of the assessment, especially when dealing with ecosystem data involving complex interactions. Traditional methods fail to fully utilize automated and data-driven technologies, resulting in inefficiency in processing large-scale environmental data. In addition, the lack of effective parameter optimization and model update mechanisms results in insufficient predictive capabilities in a changing environment, making it difficult to provide immediate decision support for environmental management, thus affecting the implementation of environmental protection measures and the protection of environmental quality. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent environmental impact assessment method and system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an intelligent environmental impact assessment method, comprising the following steps: S1: Collect environmental data including pollutant emissions, water resource consumption, air quality changes and biodiversity to establish an impact factor data set, segment the data set, call the kernel function type to make numerical predictions on pollutant emissions, calculate kernel function parameters, encode the calculation results and randomly sample, call the sampling results to generate populations, and generate initial population individual data; S2: Based on the individual data of the initial population, select individuals from the population based on the matching degree, cross-validate the mean square error of the set, select individuals for crossover, call the perturbation mutation operation, perturb and mutate multiple parameters, and generate optimized individuals after crossover mutation; S3: Based on the optimized individuals after the cross-mutation, multiple individuals are trained on air quality change data, the validation set error is calculated, the hyperparameter combination with the best matching degree is selected, the training data set is called to retrain the support vector regression model, and the optimal support vector regression model is generated; S4: Based on the optimal support vector regression model, group prediction is performed on the environmental data of biodiversity impact, dimensionality reduction is performed on the original biodiversity data, potential eigenvalues ​​are extracted, and the eigenvalues ​​are combined with the predicted values ​​to generate low-dimensional embedded predicted eigenvalues; S5: Based on the low-dimensional embedded prediction feature value, the contrast loss of the air quality change samples in the sliding window is calculated, the samples are selected as the initial clustering centers based on the fit, the positions and numbers of the clustering centers are adjusted by calling the time series, and the dynamic clustering results of the environmental impact factors are obtained.

[0007] The influencing factor data set specifically includes pollutant emissions, water resource consumption, air quality changes, and biodiversity. The initial population individual data includes population individuals, cross-validation set mean square error, individual cross-data, and disturbance variation data. The optimal support vector regression model includes a training data set, a hyperparameter combination, and a validation set error. The low-dimensional embedded prediction eigenvalues ​​specifically include potential eigenvalues ​​and predicted values ​​of the original biodiversity data. The dynamic clustering results of environmental impact factors specifically refer to the initial cluster centers, the locations and numbers of cluster centers, and time series adjustment results.

[0008] As a further solution of the present invention, the step of obtaining the initial population individual data is specifically as follows: S111: Collect and clean environmental data on pollutant emissions, water resource consumption, air quality changes and biodiversity, and establish an environmental impact factor data set based on the environmental data; S112: segmenting the environmental impact factor data set, using function types to predict pollutant emissions on the segmented data, calculating parameters of each kernel function, encoding the prediction results, and generating a set of encoding results; S113: Perform random sampling on the encoding result set, and initialize the population using a heuristic algorithm based on the sampling results, using the formula: ; Calculate the individual data of the initial population and generate the initial population through weighted averaging and discreteness adjustment; in, Representative The critical weight of the encoding result, Represents the adjustment coefficient, which is used to control the influence of random sampling results on weight calculation. Representative Sample values, represents the average value of the sampled values, Represents the total number of encoding results, Represents the number of sample values.

[0009] As a further solution of the present invention, the step of obtaining the optimized individuals after the crossover mutation is specifically as follows: S211: extracting target performance indicators from the individual data of the initial population, evaluating the matching degree of each individual, screening high matching degree individuals by comparing the matching degree threshold, and generating a high matching degree population set; S212: cross-validating the individuals in the high-matching population set, calculating the mean square error of the individuals after the cross-validation, selecting the individuals with the best performance according to the error minimization principle, and generating a set of preferred individuals; S213: Perform parameter perturbation and mutation operations on the preferred individual set, simulate gene mutation through mathematical operations, and use the formula: ; Calculate the individual parameter values ​​after mutation, and generate optimized individuals through linear combination and normalized perturbation; in, Represents the adjustment coefficient of the original individual parameter weight, represents the adjustment coefficient of the difference compensation term, represents the adjustment coefficient of the degree of penalty for the difference term, represents the weight coefficient of the normalized perturbation, Represents the first Individual parameter values, Representatives and The external parameter values ​​associated with each individual, represents the adjustment parameter value of the external influence term, represents the individual parameter value involved in the difference calculation, Represents the reference parameter value used for difference comparison, Represents the standard deviation of the difference in the variation parameter.

[0010] As a further solution of the present invention, the step of obtaining the optimal support vector regression model is specifically as follows: S311: Based on the optimized individuals after the crossover mutation, each individual is called as an independent parameter to train the air quality change data, and the verification set errors corresponding to the multiple individuals are calculated. By comparing the error of each individual with the specified error threshold, some individuals with the smallest error are screened to obtain a set of individuals with optimal matching degree; S312: Performing hyperparameter combination analysis on the individuals in the preferred matching degree individual set, calling the training parameters of multiple individuals, combining the model performance evaluation index, and using the formula: ; Calculate the matching degree of the hyperparameter combination and select the optimal parameters through weighted square root and logarithmic transformation; in, Representative The weight coefficient of each individual, Representative The bias adjustment coefficient for each individual, represents the mean of all individual performance evaluation values, represents a small parameter to avoid zero denominator, Representative The performance evaluation value of each individual, represents the number of individuals; S313: Call the optimal hyperparameter combination to retrain the support vector regression model, verify and optimize the training results, and obtain the optimal support vector regression model based on the error performance of the model on the verification set.

[0011] As a further solution of the present invention, the step of obtaining the low-dimensional embedding prediction feature value is specifically: S411: calling the optimal support vector regression model to classify and predict the environmental data on biodiversity impact, analyzing the data characteristics of the differentiated groups, outputting the prediction results through the model, and obtaining a prediction output set for each data group; S412: taking the predicted output set of each data group as input, using the compressor technology to perform dimensionality reduction processing on the original biodiversity data, extracting key potential eigenvalues ​​by calculating and analyzing the performance of each data point in the current feature space, and generating a potential eigenvalue set; S413: Processing the potential feature value set, combining the data in the predicted output set of each data group, using the formula: ; Calculate low-dimensional embedding prediction feature values ​​and extract latent features through standardization and logarithmic transformation; in, Representative The weight parameter of each feature, Representative The bias adjustment coefficient of the feature, Representative The logarithmic smoothing adjustment parameter of the features, Representative The average value of the data for each feature, Representative The standard deviation of the feature, Representative The original data points, represents the number of features, Represents the index of the low-dimensional embedded feature.

[0012] As a further solution of the present invention, the steps for obtaining the dynamic clustering results of the environmental impact factors are specifically as follows: S511: Based on the low-dimensional embedded prediction feature value, the air quality change samples in the sliding window are compared for loss, the change loss value of each sample in the sliding window is calculated, the loss values ​​are compared, and a sample set with a large difference is obtained through comparison; S512: Based on the sample set with large differences, a fit calculation is performed on each sample, and a sample is selected using the fit result. The sample with the highest fit is used as the initial cluster center, using the formula: ; Calculate the cluster fitness of the sample and select the initial cluster center through deviation and exponential decay; in, Representative The weight parameter of each sample is Representative The distribution width adjustment coefficient of samples, Representative The exponential weight parameter of samples, Representative The decay rate adjustment coefficient of samples, Representative The mean of the samples, Representative The standard deviation of the samples, Representative The value of the samples, represents the number of samples; S513: calling the initial cluster center set, adjusting the position of each cluster center in combination with the time series information, and adjusting the number and position of cluster centers by calculating the distribution characteristics of adjacent time intervals of multiple centers to obtain the dynamic clustering result of environmental impact factors.

[0013] An intelligent environmental impact assessment system, the intelligent environmental impact assessment system is used to execute the above intelligent environmental impact assessment method, the system comprises: The factor data processing module extracts numerical features based on environmental data on pollutant emissions, water resource consumption, air quality changes, and biodiversity, calculates the mean and change rate by time segment, compares the data features of differentiated segments, and generates basic data on environmental factors; The population optimization module, based on the basic data of environmental factors, pairs the pollutant emission values ​​with the water resource consumption, generates an initial population, calculates the distance and offset of pollutant changes within the population, adjusts the population structure through random disturbance, and generates dynamic optimization results; A support vector modeling module, based on the dynamic optimization results, groups the air quality change data, extracts the change rates and distribution characteristics of multiple groups, calculates and compares the cumulative errors, selects the grouped data with the smallest error, jointly calculates the influencing features and reduces the dimensionality, and generates low-dimensional embedded feature values; The dynamic clustering analysis module calculates the air quality change comparison difference within the sliding window based on the low-dimensional embedded feature value, selects the sample with the smallest difference as the cluster center, adjusts the position and number of the cluster center, calculates the dynamic offset trend, and generates the dynamic clustering result of the environmental factor.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, the accuracy and efficiency of environmental impact assessment are significantly improved by integrating data processing and machine learning techniques. Numerical prediction of environmental data is performed using kernel functions and support vector regression models, which enhances the adaptability and accuracy of the prediction model. Automated population generation and cross-mutation operations optimize model parameters, enabling the model to better learn and adapt in a complex data environment. Dynamic cluster analysis and time series adjustment further enhance the sensitivity and response speed to environmental changes, providing scientific support for the formulation of timely and effective environmental protection measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 This is a flow chart of the steps for obtaining individual data of the initial population of the present invention; Figure 3 This is a flow chart of the steps for obtaining optimized individuals after cross-mutation of the present invention; Figure 4 A flowchart of the steps for obtaining the optimal support vector regression model of the present invention; Figure 5 A flowchart of the steps for obtaining low-dimensional embedding prediction feature values ​​of the present invention; Figure 6 The present invention is a flowchart of the steps for obtaining the dynamic clustering results of environmental impact factors. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the 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 orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0018] Example 1: Please refer to Figure 1 The present invention provides a technical solution: an intelligent environmental impact assessment method, comprising the following steps: S1: Collect environmental data including pollutant emissions, water resource consumption, air quality changes and biodiversity to establish an impact factor data set, segment the data set, call the kernel function type to make numerical predictions on pollutant emissions, calculate kernel function parameters, encode the calculation results and randomly sample, call the sampling results to generate populations, and generate initial population individual data; S2: Based on the individual data of the initial population, select individuals from the population based on the matching degree, cross-validate the mean square error of the set, select individuals for crossover, call the perturbation mutation operation, perturb and mutate multiple parameters, and generate optimized individuals after crossover mutation; S3: Based on the optimized individuals after cross-mutation, call multiple individuals to train the air quality change data, calculate the error of the validation set, select the hyperparameter combination with the best matching degree, call the training data set to retrain the support vector regression model, and generate the optimal support vector regression model; S4: Based on the optimal support vector regression model, the environmental data on biodiversity impacts are grouped and predicted, the original biodiversity data is reduced in dimension, potential eigenvalues ​​are extracted, and the eigenvalues ​​are combined with the predicted values ​​to generate low-dimensional embedded predicted eigenvalues; S5: Based on the low-dimensional embedding prediction feature value, the contrast loss of air quality change samples in the sliding window is calculated, and the samples are selected as the initial clustering centers based on the fit. The time series is called to adjust the position and number of clustering centers to obtain the dynamic clustering results of environmental impact factors.

[0019] The influencing factor data sets specifically include pollutant emissions, water resource consumption, air quality changes, and biodiversity. The initial population individual data include population individuals, cross-validation set mean square error, individual cross-data, and perturbation variation data. The optimal support vector regression model includes training data sets, hyperparameter combinations, and validation set errors. The low-dimensional embedded prediction eigenvalues ​​specifically include the potential eigenvalues ​​and predicted values ​​of the original biodiversity data. The dynamic clustering results of environmental impact factors specifically refer to the initial cluster centers, the locations and numbers of cluster centers, and the time series adjustment results.

[0020] See also Figure 2 , the specific steps for obtaining the individual data of the initial population are: S111: Collect and clean environmental data on pollutant emissions, water resource consumption, air quality changes and biodiversity, and establish an environmental impact factor data set based on the environmental data; When conducting environmental monitoring, a large amount of data on pollutant emissions, water resource consumption, air quality changes and biodiversity was collected. These data sources included ground monitoring stations, satellite remote sensing and online sensors. The data obtained through these channels were preliminarily cleaned and verified to remove obvious errors and outliers. At the same time, according to the data integrity and reliability standards, the quality of the data set was ensured to meet the requirements of subsequent analysis. The processed data were classified and organized to form data sets of environmental impact factors of different categories.

[0021] S112: segmenting the environmental impact factor data set, using the function type to predict pollutant emissions on the segmented data, calculating the parameters of each kernel function, encoding the prediction results, and generating a set of encoding results; First, the environmental impact factor dataset was divided into time and space: in the time dimension, the training unit was divided into sliding windows of 24 consecutive months, and the last three months of data in each window were retained as the validation subset, and the total proportion of the validation set was set to 20%; in the spatial dimension, stratified sampling was carried out according to the three major climate zones (humid zone, semi-arid zone, and plateau zone) to which the monitoring stations belonged, and 15% of the station data in each climate zone were randomly retained as cross-regional test samples. Subsequently, the linear kernel function, radial basis kernel function, and third-order polynomial kernel function were used to construct parallel prediction models, where the bias constant of the linear kernel function was set to 0.001, the width parameter search range of the radial basis kernel function was 0.5 to 8.0, and the regularization coefficient optimization interval of the polynomial kernel function was 1.0 to 50.0. For each kernel function, the Bayesian optimization algorithm was used to tune the hyperparameters, the maximum number of iterations was set to 100, and the time series cross-validation method was used to divide the data into 5 folds, with the mean square error less than 2.5 μg / m3 as the convergence condition. After completing the parameter optimization, the prediction results of the three kernel models are dynamically weighted and encoded: according to the average absolute percentage error of each model on the validation set (8.2% for linear kernel, 6.7% for radial basis kernel, and 11.3% for polynomial kernel), the weight coefficients are assigned as 0.42, 0.53, and 0.05 according to the inverse proportion of the error. The weighted prediction values ​​are converted into 8-bit binary code strings and concatenated with the key parameters of each kernel function (linear kernel bias constant, radial basis kernel width, and polynomial kernel order). Finally, an initial population set containing 1,200 groups of codes is generated. The length of each group of codes is fixed to 32 bits, of which the first 12 bits represent the kernel function combination mode, the middle 15 bits store the parameter optimization value, and the last 5 bits record the environmental constraint threshold (such as the industrial waste gas sulfur dioxide concentration limit of 150 mg / m3 is encoded as 10010). According to actual measurements, this step improves the prediction stability of the model under complex meteorological conditions by 18.6% through collaborative prediction of multi-kernel functions, and the cross-validation error fluctuation range is reduced from ±1.8 micrograms / cubic meter to ±0.7 micrograms / cubic meter.

[0022] S113: Perform random sampling on the encoding result set, and use a heuristic algorithm to initialize the population based on the sampling results, using the formula: ; Calculate the individual data of the initial population and generate the initial population through weighted averaging and discreteness adjustment; in, Representative The critical weight of the encoding result, Represents the adjustment coefficient, which is used to control the influence of random sampling results on weight calculation. Representative Sample values, represents the average value of the sampled values, Represents the total number of encoding results, Represents the number of sample values.

[0023] This formula is used to generate an initial score that represents the overall state of the current environment. , as a starting point for subsequent simulation or evaluation. It integrates multiple sources ( arrive ) Basic environmental indicators (These indicators have different original units and need to be converted into a unified environmental impact score before calculation), and weights are assigned according to the quality of data from each source. At the same time, the random sampling test results of the environmental state are considered (which also needs to be converted into a score) by adjusting the coefficient To balance the stability and representativeness of the score, ensuring that the initial score reflects the core situation while also including a certain amount of natural fluctuations.

[0024] Parameter acquisition, setting and dimension conversion: The system needs to integrate data from three monitoring stations ( ) data to assess the initial environmental state.

[0025] Basic environmental indicators : Monitoring Station 1( ) mainly monitors air quality, with a PM2.5 reading of 45μg / m³; Monitoring Station 2( ) monitors water quality, with a turbidity reading of 15 NTU; Monitoring Station 3( ) monitors vegetation coverage, with a reading of 75%; Score conversion rules: In order to integrate these data in different units, a set of conversion rules is set to map them to an environmental impact score of [0,100], where a higher score indicates a better status. The rules are based on the recognized standards of each indicator or the ideal interval defined by experts.

[0026] PM2.5: The good range is [0,35], 0μg / m³ corresponds to a score of 100, and more than 75μg / m³ corresponds to a score of 0. Using linear interpolation, 45μg / m³ is in the (35,75] interval, and converted proportionally: .

[0027] Turbidity: Excellent range [0,5], 0 NTU corresponds to 100 points, more than 25 NTU corresponds to 0 points. Linear conversion: . (Correction: The previous calculation was incorrect and should be based on this rule) The actual value should be: If the ideal value is low, the score decreases as the value increases.

[0028] set up .

[0029] Vegetation coverage: ideal range is [70,100], with 0 points for less than 70% and 100% for 100 points. Linear conversion: (Correction: Re-understand the rules) Set the rules: The higher the coverage, the better, set 70% as the benchmark score of 50, 100% as 100 points, and linear interpolation.

[0030] ;

[0031] The converted base score is: , , .

[0032] Weight :The weight reflects the relative reliability of the data from each monitoring station, calculated based on the accuracy level of its equipment and recent maintenance records. The accuracy levels are divided into I, II, and III, corresponding to basic scores of 10, 8, and 6 respectively. Good maintenance records add 2 points, average add 0 points, and poor records subtract 2 points.

[0033] Weight calculation: Station 1: Class II accuracy (8 points), well maintained (+2 points), total score 10; Station 2: Level I accuracy (10 points), general maintenance (+0 points), total score 10; Station 3: Class II accuracy (8 points), poor maintenance record (-2 points), total score 6; Weight setting: Use scores directly as weights . , , .

[0034] On-site sampling and testing :Randomly select 4 points in the evaluation area ( ) to conduct a quick comprehensive assessment and produce a score.

[0035] Sampling rating: , , , . These ratings have been given on a consistent basis.

[0036] Sampling mean : Calculates the arithmetic mean of the sample scores.

[0037] calculate: .

[0038] Adjustment factor :This coefficient is used to adjust the impact of the discreteness of the sampling data on the final score. According to the system goal setting, if the emphasis is on score stability, The value is too small; if the volatility of the environment needs to be reflected, the value is too large. Definition criteria: Low impact (stability first): ; Medium Impact (Balanced): ; High impact (reflecting volatility): ; Setting calculation: Considering the need to reflect certain actual environmental fluctuations, but not wanting to be overly affected by random sampling, choose the middle value of the medium impact interval. .

[0039] Calculation derivation: 1. Calculate the sum of the weighted scores of each source (numerator): ; 2. Calculate the sum of the weights (denominator): ; 3. Calculate the weighted average of the base scores: ; 4. Calculate the sum of squared differences between the sample scores and the mean: ; 5. Calculate the standard deviation adjustment: ; 6. Calculate the final initial environmental status score : ; The generated initial environmental state score is 46.232 (0-100 points), which will serve as the basis for subsequent individual evolution calculations.

[0040] See also Figure 3 , the specific steps for obtaining the optimized individuals after crossover mutation are: S211: extracting target performance indicators from the individual data of the initial population, evaluating the matching degree of each individual, screening high matching degree individuals by comparing the matching degree threshold, and generating a high matching degree population set; When evaluating the fitness of individuals in a population, we focus on data analysis of individual performance, whereby the fitness of each individual is defined by collecting actual observation data through specific measurements of environmental fitness indicators, such as survival rate and reproduction rate. Individuals with high fitness exhibit higher survival ability and reproduction success rate under given environmental conditions. After screening and processing, these data form a set of population individual data with high fitness. These data sets are obtained through real-time monitoring and statistical analysis of multiple ecological parameters, including individual growth rate, immunity performance, and ability to quickly adapt to environmental changes.

[0041] S212: Perform cross validation on the individuals in the high matching population set, calculate the mean square error of the individuals after cross validation, select the individuals with the best performance according to the error minimization principle, and generate a set of preferred individuals; In the process of population cross-validation, we focus on the calculation of mean square error, which is carried out by comparing the difference between actual value and model predicted value. Data collection involves individual performance in environmental simulation experiments, such as efficiency in finding food and ability to escape predators. These data are obtained through precise ecological simulation equipment, and then processed through mathematical modeling and statistical analysis to calculate the mean square error of each individual. Individuals with the smallest mean square error are selected for further genetic operations, including gene crossover and mutation, in order to generate offspring with higher survival adaptability. The optimal individual set is obtained from these operations and represents potential genetic advantages.

[0042] S213: Perform parameter perturbation and mutation operations on the selected individual set, simulate gene mutation through mathematical operations, and use the formula: ; Calculate the individual parameter values ​​after mutation, and generate optimized individuals through linear combination and normalized perturbation; in, Represents the adjustment coefficient of the original individual parameter weight, represents the adjustment coefficient of the difference compensation term, represents the adjustment coefficient of the degree of penalty for the difference term, represents the weight coefficient of the normalized perturbation, Represents the first Individual parameter values, Representatives and The external parameter values ​​associated with each individual, represents the adjustment parameter value of the external influence term, represents the individual parameter value involved in the difference calculation, Represents the reference parameter value used for difference comparison, Represents the standard deviation of the difference in the variation parameter.

[0043] This formula simulates the environmental status score ( ) evolves under the influence of external pressure and internal variation, generating new scores It takes into account the inheritance of the original state (by the coefficient control), external environment forecast ( ) and interventions ( )'s combined impact (coefficient Adjust the overall effect, to mediate the negative or counteracting effects of interventions), and to interact with other individuals in the population ( )Comparison of random disturbances (coefficients Control intensity, The purpose is to simulate the adaptive adjustment of environmental systems to changes.

[0044] Parameter acquisition, setting and data association: generated using S113 Perform mutation calculation as the first individual.

[0045] Original individual score : Data source: Directly use the calculation results of the previous step ; External environment forecast Impact of interventions : How to obtain: Predictions from authoritative climate models for the next cycle, converted into scores. is the estimated short-term negative economic impact based on currently implemented environmental policies (such as factory production restrictions) (also converted into a score, with a low score indicating a small impact); Scoring: Climate Model Prediction Scoring Policy Negative Impact Score . (The scores are all on a scale of 0-100); Comparing individuals within a population : How to obtain: Randomly select two other individuals from the current simulated population (assuming that several initial individuals have been generated) that are different from Individual ratings of Select rating: and .

[0046] Population score standard deviation : Calculation method: Calculate the standard deviation of the scores of all individuals in the current population. Assuming that the current population contains 10 individuals, the standard deviation of their scores is calculated; Calculation results: .

[0047] Adjustment coefficient :These coefficients are set according to the simulation goal (conservative adaptation or aggressive exploration) and the judgment of the importance of each factor; Set up references and calculations: : Controls the degree of inheritance of the original state. It is hoped that individuals can gradually adjust based on the current state, so it is set to a higher value. Value range definition: [0,0.3) low inheritance, [0.3,0.7) medium inheritance, [0.7,1.0] high inheritance. Select high inheritance: ; :Adjust the intensity of the combined impact of the external environment and intervention. External forecasts are an important reference and are set at medium to high. Value range: [0,0.4) low impact, [0.4,0.8) medium impact, [0.8,1.2] high impact. Select medium impact: ; : Adjust for the negative impacts of interventions. The negative economic impacts of policies need to be considered, but should not completely offset their environmental benefits and should be set to a smaller value. Value range: [0,0.2) low suppression, [0.2,0.5) medium suppression, [0.5,1.0] high suppression. Select low suppression: ; : Controls the intensity of disturbance introduced by differences within the population. To promote diversity, allow some random exploration and set to a moderate value. Value range: [0,0.3) low disturbance, [0.3,0.7) medium disturbance, [0.7,1.0] high disturbance. Choose medium disturbance: ; Calculation derivation: 1. Calculation inheritance part : ; 2. Calculate the external impact part : ; 3. Calculate the internal disturbance part : ; 4. Calculate the final score after mutation : ; Quantification of variation: Calculation of relative magnitude of variation: ; The range of variation is as follows: slight variation: (0%, 10%] moderate variation: (10%, 30%] significant variation: (30%, 50%] severe variation: >50%. The variation is 39.84%, which is a "significant variation", indicating that the environmental status score of this individual has undergone a significant adaptive change under the influence of simulated internal and external factors, reaching 64.64685 points.

[0048] See also Figure 4 , the steps to obtain the optimal support vector regression model are as follows: S311: Based on the optimized individuals after crossover mutation, each individual is called as an independent parameter to train the air quality change data, and the verification set errors corresponding to the multiple individuals are calculated. By comparing the error of each individual with the specified error threshold, some individuals with the smallest error are screened to obtain a set of individuals with optimal matching degree; When training based on the optimized individuals after cross-mutation, it is first necessary to collect air quality data. These data cover multiple dimensions such as pollutant concentration and meteorological conditions. The data collection process involves the use of various sensors for real-time monitoring. After being collected, these monitoring data need to be preprocessed to remove noise and outliers. The processing methods include technical means such as data smoothing, anomaly detection and replacement of missing values. The preprocessed data ensures the quality of the training set and makes the model training more accurate. Subsequently, the system will calculate the performance of each individual model on the validation set. This step is completed by calculating the error between the predicted value and the actual value. Commonly used error calculation methods include mean square error (MSE) and mean absolute error (MAE). The error calculation result will directly affect the individual selection process. According to the set error threshold, the system selects the individuals with the best performance to enter the next round of training or for practical application. In this process, the size of the error directly determines whether the individual can be selected, thereby ensuring the optimization direction and effect of the model.

[0049] S312: Perform hyperparameter combination analysis on individuals in the matching optimal individual set, call the training parameters of multiple individuals, and combine the model performance evaluation index to adopt the formula: ; Calculate the matching degree of the hyperparameter combination and select the optimal parameters through weighted square root and logarithmic transformation; in, Representative The weight coefficient of each individual, Representative The bias adjustment coefficient for each individual, represents the mean of all individual performance evaluation values, represents a small parameter to avoid the denominator being zero, Representative The performance evaluation value of each individual, represents the number of individuals; This formula is used to evaluate different hyperparameter combinations ( arrive ) under which the performance matching degree of the environmental prediction model Performance is measured by the model’s prediction error rate The matching degree takes into account the error rate Relative to the average error rate of all combinations The degree of deviation (volatility, determined by metric) and the stability of the error rate itself (low error tendency, due to measure, Small positive number to avoid division by zero). Weight and It reflects the different emphasis on volatility and stability. The lower the value, the better the overall match of the set of hyperparameters.

[0050] Parameter acquisition, setting and dimension conversion: evaluation The accuracy of the prediction of the environmental status score for the next week by a combination of hyperparameters.

[0051] Model prediction error rate : Obtaining method: Use each set of hyperparameters ( ) respectively train the environmental prediction models and test them on an independent validation dataset, calculating the mean absolute percentage error (MAPE) between the predicted scores and the actual scores.

[0052] Get the results (MAPE%): , , , . (The error rate is a percentage, with uniform units); Average error rate : calculate: ; Small positive number : Set reference: Make sure Much greater than 0. Error rate The minimum is 6.2%; Settings: .

[0053] Weight coefficient : Set weights to balance the need for forecasts close to the average and low absolute forecast errors.

[0054] Setting reference and calculation: Assume that the absolute value of the forecast error is low (stability), so The weight should be relative Higher. Set a benchmark. If a certain hyperparameter combination training time ( ) is shorter, then its volatility is allowed to be slightly larger ( slightly larger); if the standard deviation of the error on the validation set ( ) is smaller, the stability is considered to be better ( larger).

[0055] Let the basic weight , .

[0056] Calculate the adjustment factor: , .

[0057] Assume that the training time and error standard deviation data are as follows: k=1:T=2h,SD=1.5%->Avg(T)=2.5h,Avg(SD)=1.25%; k=2:T=3h,SD=1.0%; k=3:T=2.5h,SD=1.2%; k=4:T=2.5h,SD=1.3%; Calculate the weight (for simplicity, directly set the final value to reflect the calculation idea): (fast training but high variance); (slow training but low variance); (average level); (average training time, slightly larger variance); This example has unified settings, focusing more on low error: , For all .

[0058] Calculation derivation: Calculate each hyperparameter group Performance items: 1. k=1: ; 2. k=2: ; 3. k=3: ; 4.k=4: ; Calculate the sum and find the average matching degree H: ; ; Matching metric: The lower the value, the better the match. High match: Medium match: Low match: Calculated , which is a "medium match". Among them, the third group of hyperparameters (H term = 0.795) performed the best, and the first group (H term = 1.674) performed the worst. Based on this evaluation, the third or second group of hyperparameters can be selected for subsequent environmental prediction tasks.

[0059] S313: Call the optimal hyperparameter combination to retrain the support vector regression model, verify and optimize the training results, and obtain the optimal support vector regression model based on the error performance of the model on the validation set.

[0060] After obtaining the optimal hyperparameter combination, the support vector regression model is retrained using these parameters. During the retraining process, the system adjusts each hyperparameter to adapt to the training data set, including the selection of the kernel function, the adjustment of the regularization parameter, and the optimization of the kernel function parameters. For example, when using the radial basis function as the kernel, parameter optimization involves adjusting its bandwidth. This process is achieved through cross-validation to ensure that the model not only performs well on the current data set, but also has good generalization capabilities. Through this method, the error on the validation set is gradually reduced during the training process of the support vector machine. The generated model can accurately predict changes in air quality and has important application value for environmental monitoring and management.

[0061] See also Figure 5 , the specific steps for obtaining the low-dimensional embedding prediction feature value are: S411: calling the optimal support vector regression model to classify and predict the environmental data on biodiversity impacts, analyzing the data characteristics of the differentiated groups, outputting the prediction results through the model, and obtaining a prediction output set for each data group; Based on the optimal support vector regression model, the biodiversity factors in the environmental impact data will be predicted. First, the data will be classified and summarized to ensure that each group represents an independent biodiversity indicator. According to the different characteristics of biodiversity, such as population density and distribution range, appropriate statistical methods will be selected to extract data characteristics. The algorithm principle of support vector machine will be used to perform predictive analysis on the data of each group. By comparing the predicted values ​​with the actual observed values, the accuracy and reliability of the model will be evaluated. This process not only involves the adjustment and optimization of the model, but also needs to take into account the nonlinear characteristics of the data and potential outliers. The model parameters are iteratively optimized to improve the accuracy of the prediction. This method can effectively predict the future change trends of different biological communities and provide a scientific basis for biodiversity conservation and management.

[0062] S412: taking the predicted output set of each data group as input, using the compressor technology to perform dimensionality reduction processing on the original biodiversity data, extracting key potential eigenvalues ​​by calculating and analyzing the performance of each data point in the current feature space, and generating a potential eigenvalue set; Compressors are used to reduce the dimensionality of raw biodiversity data. Through dimensionality reduction of multi-dimensional data, large amounts of complex environmental data can be processed and analyzed more effectively. First, the data is preprocessed, including cleaning, standardization, and outlier processing. Then, appropriate dimensionality reduction techniques are selected, such as principal component analysis or linear discriminant analysis. These techniques can help identify the most important features in the data, reduce data redundancy, and increase processing speed. Through the reduced dimensionality data, the data structure and pattern can be seen more clearly, which facilitates subsequent data analysis and feature extraction. In this process, the algorithm parameters need to be precisely adjusted to ensure that the key information of the data is not lost during the dimensionality reduction process, so that the compressed data can still effectively reflect the key characteristics of biodiversity.

[0063] S413: Process the potential feature value set, combine the data in the predicted output set of each data group, and use the formula: ; Calculate low-dimensional embedding prediction feature values ​​and extract latent features through standardization and logarithmic transformation; in, Representative The weight parameter of each feature ( is the index, ), Representative The bias adjustment coefficient of the feature, Representative The logarithmic smoothing adjustment parameter of the features, Representative The average value of the data for each feature, Representative The standard deviation of the feature, Representative The original data points, represents the number of features, Represents the index of the low-dimensional embedded feature.

[0064] This formula is used to convert high-dimensional raw environmental monitoring data (Include features) into low-dimensional embedding features The transformation process takes into account each original feature With its mean The degree of deviation (through the standardization item ), as well as the logarithmic scale information of the eigenvalues ​​themselves (via the logarithmic term ). are the weights and tuning parameters of the feature embedding model, is the standard deviation of the feature. The goal is to extract key low-dimensional information that can effectively represent the structure of the original data .

[0065] Parameter acquisition, setting and dimension conversion: Processing a The data points of the original environment features are embedded into a one-dimensional space ( ).

[0066] Original data points : Acquisition method: Extract a record from the environmental monitoring database at a specific time point, containing multiple indicators.

[0067] Extract data: Feature 1 (air temperature, ): °C; Feature 2 (Relative humidity, ): %; Feature 3 (wind speed, ): m / s; Feature mean With standard deviation : Calculation method: The mean and standard deviation of each feature are calculated based on a large amount of historical monitoring data including the current data point.

[0068] Calculation results: °C, °C; %, %; m / s, m / s; Embedding model parameters : How to obtain: These parameters are usually learned on a large amount of data by training a dimensionality reduction model (such as some form of autoencoder or a specially designed embedding algorithm). The goal is to make the reduced dimensionality features best reconstruct the original data or perform well in downstream tasks (such as classification, clustering).

[0069] Setting value (obtained through model training): ; ; ; Handling different dimensions: The formula implicitly uses Deviation The logarithmic terms The original value is also used directly, and the logarithmic transformation itself can handle data of different magnitudes. The units are different, but the formula design has taken compatibility into consideration. The calculation results is a dimensionless embedding value.

[0070] Calculation derivation: Calculate each feature Contributions: 1.m=1(temperature): ; ; ; ; 2.m=2(humidity): ; ; ; ; 3.m=3(wind speed): ; ; ; ; Calculate the final low-dimensional embedding feature value (because ): ; The low-dimensional embedding feature value of this data point is 3.4203. This value can be used for subsequent clustering, classification or visualization tasks, which compresses the main information of the original three features.

[0071] See also Figure 6 , the specific steps for obtaining the dynamic clustering results of environmental impact factors are: S511: Based on the low-dimensional embedding prediction feature value, the air quality change samples in the sliding window are compared for loss, the change loss value of each sample in the sliding window is calculated, the loss values ​​are compared, and a sample set with a large difference is obtained through comparison; In the process of air quality prediction based on the optimal support vector regression model, it is first necessary to perform feature extraction and data preprocessing on each sample in the environmental data set, which includes normalizing the air quality data, deleting invalid or abnormal data, and ensuring the data quality of the input model. The prediction model applies a variety of kernel functions to handle different types of data distributions when processing input data. The selection of each kernel function is based on the characteristics and distribution of the data. Through these steps, the model can more accurately predict future trends in air quality changes and ensure the accuracy and reliability of the prediction results. Analyzing the output set of prediction results can help environmental scientists and policymakers better understand the patterns and trends of air quality changes.

[0072] S512: Based on the sample set with large differences, the fit is calculated for each sample, and the sample is selected using the fit result. The sample with the highest fit is used as the initial cluster center, using the formula: ; Calculate the cluster fitness of the sample and select the initial cluster center through deviation and exponential decay; in, Representative The weight parameter of each sample is Representative The distribution width adjustment coefficient of samples, Representative The exponential weight parameter of samples, Representative The decay rate adjustment coefficient of samples, Representative The mean of the samples, Representative The standard deviation of the samples, Representative The value of the samples, represents the number of samples; This formula is used to evaluate a specific sample of environmental data. (Include The first of the indicators index value) becomes the fitness of a cluster center The fitness takes into account the sample index value and its potential clustering ( Represents the statistical characteristics of a cluster center (mean , standard deviation ) of the deviation (by measure, as an adjustment parameter), and the exponential decay effect of the index value itself on its fitness (given by measure, is the adjustment parameter). The higher the value, the more suitable the sample point is for clustering. center.

[0073] Parameter acquisition, setting and dimension conversion: Evaluate a Key indicators (such as the low-dimensional embedding features output by S413 )’s sample points have the potential to become cluster centers.

[0074] Sample index value : Data source: low-dimensional embedding feature values ​​calculated in the previous step S413. (n=1). (This value is dimensionless) Latent clustering statistics : Acquisition method: These values ​​represent samples The characteristic statistics of the cluster to which the data belongs. They can obtain the center and standard deviation of each cluster by performing preliminary exploratory clustering (such as K-Means) on the data set, or preset the characteristic range of typical environmental patterns based on domain knowledge.

[0075] Set value (assumed to be obtained through preliminary clustering): Sample The cluster that it belongs to most has a low-dimensional embedding feature with a mean of , the standard deviation is .

[0076] Adjustment parameters : Acquisition method: These parameters control the relative importance and specific form of the deviation term and attenuation term in the fitness calculation. They are usually determined by experiments or optimization algorithms (such as adjusting parameters to optimize a certain evaluation indicator of the final clustering result) based on the specific requirements of the clustering task and data characteristics.

[0077] Set up references and calculations: : Adjust the overall weight of the deviation term. The larger the value, the greater the penalty for samples that deviate from the center. Range: [0.5, 2.0]. Setting ; : Smooth term, and They work together on the denominator and affect the sensitivity to the standard deviation. The bigger, yes The lower the sensitivity of the change. Range: [0.1,1.0]. Setting ; : Adjust the base weight of the exponential decay term. Range: [0.1,1.0]. Setting ; : Controls the rate of exponential decay. The larger the value, the faster the decay. The contribution to fitness decreases rapidly. Interval: [0.01,0.5]. Setting ; Calculation derivation: Due to , we only need to calculate the term where n=1.

[0078] 1. Calculate the deviation term: ; 2. Calculate the exponential decay term: ; 3. Calculate the final fitness (m represents the potential cluster being evaluated): ; Adaptation quantification: The value is used to compare the potential of different sample points to become cluster centers. The higher the value, the better the fit. Low fitness: Medium fit: High adaptability: The calculated fitness , which belongs to "medium fitness". This means that this sample point (low-dimensional eigenvalue is 3.4203) has a certain potential to become the representative center of the environmental pattern of its type, but it is not the best choice. It needs to be compared with the fitness of other sample points to finally determine the initial cluster center.

[0079] S513: Call the initial cluster center set, adjust the position of each cluster center in combination with the time series information, and adjust the number and position of cluster centers by calculating the distribution characteristics of multiple centers in adjacent time intervals to obtain the dynamic clustering results of environmental impact factors.

[0080] When adjusting the position and number of cluster centers in a time series, the key is to consider the periodicity and trend characteristics of the time series data. This process first requires an in-depth analysis of the time series data to identify key time nodes and changing trends, including data smoothing, trend decomposition, and periodicity analysis. Statistical methods such as moving average or exponential smoothing are used to predict future data points. These prediction results will directly affect the initial setting and subsequent adjustments of the cluster centers. Through this method, the positions of the cluster centers can be dynamically adjusted to adapt to the rapid changes in environmental factors. The number of cluster centers is also adjusted based on the deviation between the predicted results and the actual observed values. A dynamic threshold is used to determine the increase or decrease of the cluster center. This method not only improves the flexibility of clustering, but also enhances the model's ability to adapt to environmental changes. The dynamic clustering results of environmental influencing factors obtained will provide more accurate and real-time support for decision-making.

[0081] An intelligent environmental impact assessment system, which is used to execute the above intelligent environmental impact assessment method, comprises: The factor data processing module extracts numerical features based on environmental data on pollutant emissions, water resource consumption, air quality changes, and biodiversity, calculates the mean and change rate by time segment, compares the data features of differentiated segments, and generates basic data on environmental factors; The population optimization module, based on the basic data of environmental factors, pairs the pollutant emission values ​​with the water resource consumption, generates the initial population, calculates the distance and offset of pollutant changes within the population, adjusts the population structure through random disturbance, and generates dynamic optimization results; The support vector modeling module groups the air quality change data based on the dynamic optimization results, extracts the change rate and distribution characteristics of multiple groups, calculates and compares the cumulative error, selects the grouped data with the smallest error, jointly calculates the influencing features and reduces the dimension, and generates low-dimensional embedded feature values; The dynamic clustering analysis module calculates the contrast difference of air quality changes within the sliding window based on low-dimensional embedded eigenvalues, selects the sample with the smallest difference as the cluster center, adjusts the position and number of cluster centers, calculates the dynamic offset trend, and generates dynamic clustering results of environmental factors.

[0082] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession can use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent environmental impact assessment method, characterized in that: The following steps are involved: Collect environmental data including pollutant emissions, water resource consumption, air quality changes and biodiversity to establish an impact factor data set, segment the data set, call the kernel function type to make numerical predictions on pollutant emissions, calculate the kernel function parameters, encode the calculation results and randomly sample, call the sampling results to generate populations, and generate initial population individual data; Based on the individual data of the initial population, select individuals of the population based on the matching degree, cross-validate the mean square error of the set, select individuals for crossover, call the perturbation mutation operation, perturb and mutate multiple parameters, and generate optimized individuals after crossover mutation; Based on the optimized individuals after the crossover mutation, multiple individuals are trained on air quality change data, the verification set error is calculated, the hyperparameter combination with the best matching degree is selected, the training data set is called to retrain the support vector regression model, and the optimal support vector regression model is generated; Based on the optimal support vector regression model, the environmental data on biodiversity impacts are grouped and predicted, the original biodiversity data is subjected to dimensionality reduction processing, potential eigenvalues ​​are extracted, and the eigenvalues ​​are combined with the predicted values ​​to generate low-dimensional embedded predicted eigenvalues; Based on the low-dimensional embedded prediction feature value, the contrast loss of air quality change samples in the sliding window is calculated, samples are selected as initial clustering centers based on the degree of fit, and the positions and numbers of clustering centers are adjusted by calling the time series to obtain the dynamic clustering results of environmental impact factors.

2. The intelligent environmental impact assessment method according to claim 1, characterized in that: The influencing factor data set specifically includes pollutant emissions, water resource consumption, air quality changes, and biodiversity. The initial population individual data includes population individuals, cross-validation set mean square error, individual cross-data, and disturbance variation data. The optimal support vector regression model includes a training data set, a hyperparameter combination, and a validation set error. The low-dimensional embedded prediction eigenvalues ​​specifically include potential eigenvalues ​​and predicted values ​​of the original biodiversity data. The dynamic clustering results of environmental impact factors specifically refer to the initial cluster centers, the locations and numbers of cluster centers, and time series adjustment results.

3. The intelligent environmental impact assessment method according to claim 2, characterized in that: The steps for obtaining the individual data of the initial population are specifically as follows: Collect and clean environmental data on pollutant emissions, water resource consumption, air quality changes and biodiversity, and build a data set of environmental impact factors based on environmental data; Segmenting the environmental impact factor data set, using function types to predict pollutant emissions on the segmented data, calculating parameters of each kernel function, encoding the prediction results, and generating a set of encoding results; Random sampling is performed on the encoding result set, and a heuristic algorithm is used to initialize the population based on the sampling results, using the formula: ； Calculate the individual data of the initial population and generate the initial population through weighted averaging and discreteness adjustment; in, Representative The critical weight of the encoding result, Represents the adjustment coefficient, which is used to control the influence of random sampling results on weight calculation. Representative Sample values, represents the average value of the sampled values, Represents the total number of encoding results, Represents the number of sample values.

4. The intelligent environmental impact assessment method according to claim 3 is characterized in that: The steps for obtaining the optimized individuals after the crossover mutation are specifically as follows: Extracting target performance indicators from the individual data of the initial population, evaluating the matching degree of each individual, screening high matching degree individuals by comparing the matching degree threshold, and generating a high matching degree population set; Performing cross-validation on the individuals in the high-matching population set, calculating the mean square error of the individuals after the cross-validation, selecting the individuals with the best performance according to the error minimization principle, and generating a set of preferred individuals; The parameter perturbation mutation operation is performed on the preferred individual set, and the gene mutation is simulated by mathematical operation, using the formula: ； Calculate the individual parameter values ​​after mutation, and generate optimized individuals through linear combination and normalized perturbation; in, Represents the adjustment coefficient of the original individual parameter weight, represents the adjustment coefficient of the difference compensation term, represents the adjustment coefficient of the degree of penalty for the difference term, represents the weight coefficient of the normalized perturbation, Represents the first Individual parameter values, Representatives and The external parameter values ​​associated with each individual, represents the adjustment parameter value of the external influence term, represents the individual parameter value involved in the difference calculation, Represents the reference parameter value used for difference comparison, Represents the standard deviation of the difference in the variation parameter.

5. The intelligent environmental impact assessment method according to claim 4, characterized in that: The steps for obtaining the optimal support vector regression model are specifically as follows: Based on the optimized individuals after the crossover mutation, each individual is called as an independent parameter to train the air quality change data, and the verification set errors corresponding to the multiple individuals are calculated. By comparing the error of each individual with the specified error threshold, some individuals with the smallest error are screened to obtain a set of individuals with optimal matching degree; Perform hyperparameter combination analysis on the individuals in the preferred matching degree individual set, call the training parameters of multiple individuals, and combine the model performance evaluation index to adopt the formula: ； Calculate the matching degree of the hyperparameter combination and select the optimal parameters through weighted square root and logarithmic transformation; in, Representative The weight coefficient of each individual, Representative The bias adjustment coefficient for each individual, represents the mean of all individual performance evaluation values, represents a small parameter to avoid the denominator being zero, Representative The performance evaluation value of each individual, represents the number of individuals; The optimal hyperparameter combination is called to retrain the support vector regression model, the training results are verified and optimized, and the optimal support vector regression model is obtained in combination with the error performance of the model on the verification set.

6. The intelligent environmental impact assessment method according to claim 5, characterized in that: The steps for obtaining the low-dimensional embedding prediction feature value are specifically as follows: Calling the optimal support vector regression model to classify and predict environmental data on biodiversity impacts, analyzing data characteristics of differentiated groups, outputting prediction results through the model, and obtaining a prediction output set for each data group; Taking the predicted output set of each data group as input, using the compressor technology to perform dimensionality reduction processing on the original biodiversity data, extracting key potential eigenvalues ​​by calculating and analyzing the performance of each data point in the current feature space, and generating a potential eigenvalue set; The potential feature value set is processed, combined with the data in the predicted output set of each data group, using the formula: ； Calculate low-dimensional embedding prediction feature values ​​and extract latent features through standardization and logarithmic transformation; in, Representative The weight parameter of each feature, Representative The bias adjustment coefficient of the feature, Representative The logarithmic smoothing adjustment parameter of the features, Representative The average value of the data for each feature, Representative The standard deviation of the feature, Representative The original data points, represents the number of features, Represents the index of the low-dimensional embedded feature.

7. The intelligent environmental impact assessment method according to claim 6, characterized in that: The steps for obtaining the dynamic clustering results of the environmental impact factors are specifically as follows: Based on the low-dimensional embedded prediction feature value, the air quality change samples in the sliding window are compared for loss, the change loss value of each sample in the sliding window is calculated, the size of the loss value is compared, and a sample set with a large difference is obtained through comparison; Based on the sample set with large differences, the fit is calculated for each sample, and the sample is selected using the fit result. The sample with the highest fit is used as the initial cluster center, using the formula: ； Calculate the cluster fitness of the sample and select the initial cluster center through deviation and exponential decay; in, Representative The weight parameter of each sample is Representative The distribution width adjustment coefficient of samples, Representative The exponential weight parameter of samples, Representative The decay rate adjustment coefficient of samples, Representative The mean of the samples, Representative The standard deviation of the samples, Representative The value of the samples, represents the number of samples; The initial cluster center set is called, and the position of each cluster center is adjusted in combination with the time series information. The number and position of cluster centers are adjusted by calculating the distribution characteristics of adjacent time intervals of multiple centers to obtain the dynamic clustering result of environmental impact factors.

8. An intelligent environmental impact assessment system, characterized in that: According to any one of claims 1 to 7, the intelligent environmental impact assessment method comprises: The factor data processing module extracts numerical features based on environmental data on pollutant emissions, water resource consumption, air quality changes, and biodiversity, calculates the mean and change rate by time segment, compares the data features of differentiated segments, and generates basic data on environmental factors; The population optimization module, based on the basic data of environmental factors, pairs the pollutant emission values ​​with the water resource consumption, generates an initial population, calculates the distance and offset of pollutant changes within the population, adjusts the population structure through random disturbance, and generates dynamic optimization results; A support vector modeling module, based on the dynamic optimization results, groups the air quality change data, extracts the change rates and distribution characteristics of multiple groups, calculates and compares the cumulative errors, selects the grouped data with the smallest error, jointly calculates the influencing features and reduces the dimensionality, and generates low-dimensional embedded feature values; The dynamic clustering analysis module calculates the air quality change comparison difference within the sliding window based on the low-dimensional embedded feature value, selects the sample with the smallest difference as the cluster center, adjusts the position and number of the cluster center, calculates the dynamic offset trend, and generates the dynamic clustering result of the environmental factor.

Images