Soil heavy metal distribution prediction method and system based on machine learning

Abstract

The invention discloses a soil heavy metal distribution prediction method and system based on machine learning, and the method comprises the steps: obtaining topographic factor land use industrial activity data, carrying out the fusion remote sensing information processing, and determining a multi-source data set; performing standardization processing according to the multi-source data set, and performing space-time registration if the scale difference after standardization processing exceeds a preset threshold value to obtain data in a unified format; key features are extracted according to the unified format data, and a dimension reduction feature set is obtained through principal component analysis; a random forest model is constructed according to the dimension reduction feature set, parameters are optimized, and a heavy metal content prediction model is determined; inputting the sampling finite point location data into a heavy metal content prediction model, judging an industrial activity influence area, and obtaining a preliminary distribution estimation result; based on the preliminary distribution estimation result, high-resolution grid data are obtained by fusing topographic factors for the space complex region; and generating a pollution distribution diagram according to the high-resolution grid data, judging a low-prediction-precision region, and obtaining a final optimized layer.

Description

Technical Field

[0001] This invention belongs to the field of soil heavy metal distribution prediction technology, and in particular relates to a method and system for predicting soil heavy metal distribution based on machine learning. Background Technology

[0002] Soil pollution is a major global concern, directly impacting ecological security, agricultural product quality, and human health. Accurately understanding the distribution patterns of heavy metals in soil is not only fundamental to assessing environmental risks but also crucial for developing precise remediation strategies. However, the spatial distribution of heavy metals in soil is highly complex, influenced by a combination of factors including topography, land use, and industrial activities. Traditional monitoring methods struggle to meet the demands for large-scale, high-precision, and dynamic data, making the efficient and comprehensive revelation of the spatial characteristics of soil pollution a pressing challenge. Traditional methods have significant limitations in predicting the distribution of heavy metals in soil. While conventional field sampling and laboratory analysis provide accurate point data, their coverage is limited and cannot reflect the overall pollution status of a region. Statistical interpolation methods, although extending point data to area data, often rely on simple mathematical assumptions and are ill-suited to the complex relationships between environmental factors. Furthermore, advancements in remote sensing technology and geographic information systems have enriched environmental data, including surface features and land use patterns; however, existing methods struggle to fully utilize this multi-source information, limiting the accuracy of prediction results. At the technical level, the core challenge in predicting heavy metal distribution in soil lies in effectively integrating multi-source environmental data and capturing the complex relationship between them and heavy metal content. The diversity of environmental data is the first key factor. These data come from different sources, such as satellite imagery, topographic features, and human activity information, with varying formats and scales. Integrating this data requires solving the challenges of spatiotemporal registration and feature extraction. The complexity of data integration further leads to another key factor: how to extract deep patterns related to heavy metal distribution from this high-dimensional, heterogeneous data. Traditional methods are insufficient in this regard, often failing to accurately reflect the nonlinear correlation between environmental variables and heavy metal content. Therefore, how to extract key features from complex and diverse environmental data and construct high-precision spatial distribution predictions based on these features has become a crucial issue in soil pollution research. For example, in an area affected by industrial activities, topography, vegetation cover, and factory distribution all contribute to heavy metal diffusion, but existing technologies struggle to integrate these factors into a unified prediction framework, resulting in the inability to accurately identify pollution hotspots or predict potential risk points. This not only limits the targeting of environmental management but also increases remediation costs and resource waste. Summary of the Invention

[0003] To address the aforementioned technical challenges, this invention proposes a machine learning-based method and system for predicting the distribution of heavy metals in soil. Through multi-source data fusion and nonlinear analysis, it significantly improves the accuracy and spatial resolution of pollution distribution prediction, providing efficient technical support for the accurate identification and environmental management of areas affected by industrial activities.

[0004] To achieve the above objectives, this invention provides a method for predicting soil heavy metal distribution based on machine learning, comprising: Acquire topographic factors, land use, and industrial activity data, perform fusion remote sensing information processing, and determine multi-source datasets; The multi-source dataset is standardized. If the scale difference after standardization exceeds a preset threshold, spatiotemporal registration is performed to obtain data in a unified format. Key features are extracted from the unified format data, and principal component analysis is used to obtain a dimensionality-reduced feature set. A random forest model is constructed based on the dimensionality-reduced feature set, and the parameters are optimized to determine the heavy metal content prediction model; The data from a limited number of sampling points are input into the heavy metal content prediction model to determine the area affected by industrial activities and obtain preliminary distribution estimation results. Based on the preliminary distribution estimation results, high-resolution grid data is obtained by incorporating terrain factors for spatially complex areas. A pollution distribution map is generated based on the high-resolution grid data, areas with low prediction accuracy are identified, and the final optimized layer is obtained.

[0005] Optionally, determining the multi-source dataset includes: Data on topographic factors, land use, and industrial activities are acquired, high-resolution image data is extracted, sensor bias is eliminated, and a corrected first image dataset is obtained. Based on the first image dataset, topographic factors are extracted, and land use types are segmented by spectral features to determine the first classification result of topography and land use; If the spectral characteristics of the industrial activity area in the first classification result do not match the preset threshold, then the first image dataset and geographic information data are integrated to obtain the fused second classification result. Based on the second classification result, the industrial activity area is further validated, and the classification boundary is optimized by combining spatial resolution to determine the multi-source dataset.

[0006] Optionally, obtaining data in a uniform format includes: The original data is obtained from the multi-source dataset, and the format of the original data is unified and the numerical range is adjusted to obtain a standardized dataset. Based on the standardized dataset, it is determined whether the scale difference exceeds a preset threshold. If it does, the time and spatial coordinates are aligned to obtain data in a unified format.

[0007] Optionally, determining the heavy metal content prediction model includes: Based on the dimensionality reduction feature set, spatiotemporal data and business classification information are obtained, feature weights are calculated, and a fused feature set is obtained. A random forest model is constructed based on the fused feature set, and the training error is obtained by calculating the difference between the predicted value and the actual value. If the training error is lower than a preset threshold, the tree depth and number of trees in the random forest model are adjusted to obtain an optimized random forest model. The fused feature set is input into the optimized random forest model for training to determine the heavy metal content prediction model.

[0008] Optionally, obtaining preliminary distribution estimates includes: Limited data is obtained from sampling points. Using a pre-established geographic information database, point information containing latitude, longitude and heavy metal content is extracted, outliers are removed, and a cleaned point dataset is obtained. Based on the cleaned point dataset, the heavy metal content in the unsampled area is calculated, and a spatial distribution grid containing latitude, longitude and predicted content is generated to obtain preliminary spatial distribution data. By analyzing the spatial distribution data and combining it with industrial activity records, the locations of pollution sources were determined, and a pollution source distribution map was obtained. Based on the pollution source distribution map, the spatial distribution data is segmented to obtain preliminary distribution estimation results.

[0009] Optionally, obtaining high-resolution grid data includes: Based on the preliminary distribution estimation results, the heavy metal content of the unsampled area is calculated, and the terrain height data is fused to generate a first spatial distribution grid containing latitude, longitude and predicted content, thus obtaining preliminary grid data. If the spatial resolution of the preliminary grid data is lower than a preset threshold, the first spatial distribution grid is analyzed, and the weights are adjusted in combination with terrain factors to generate a second spatial distribution grid that includes latitude and longitude and optimization content, thereby determining high-resolution grid data.

[0010] Optionally, obtaining the final optimized layer includes: Based on the high-resolution grid data, outliers are removed to obtain a cleaned grid dataset; Based on the cleaned grid dataset, the spatial distribution characteristics of pollution source data are analyzed, and the weights are adjusted in combination with terrain factors to generate a first pollution distribution map containing latitude and longitude and predicted pollution content. If the prediction accuracy of the first pollution distribution map is lower than a preset threshold, then the first pollution distribution map is analyzed to determine the areas with low prediction accuracy, and a dataset of low-accuracy areas is obtained. For the low-precision regional dataset, spatially complex regions are segmented, and pollution distribution weights are adjusted by combining terrain factors and pollution source data to generate a second pollution distribution map that includes latitude and longitude and optimized pollution content, thus obtaining the final optimized layer.

[0011] On the other hand, to achieve the above objectives, the present invention also provides a soil heavy metal distribution prediction system based on machine learning, comprising: The multi-source data acquisition module is used to acquire topographic factors, land use, and industrial activity data, perform fusion remote sensing information processing, and determine the multi-source dataset; The data standardization processing module is used to perform standardization processing on the multi-source dataset. If the scale difference after standardization exceeds a preset threshold, spatiotemporal registration is performed to obtain data in a unified format. The feature extraction and dimensionality reduction module is used to extract key features based on the unified format data and obtain a dimensionality-reduced feature set by using principal component analysis. The prediction model building module is used to construct a random forest model based on the dimensionality reduction feature set, optimize parameters, and determine the heavy metal content prediction model. The preliminary distribution estimation module is used to input the data from a limited number of sampling points into the heavy metal content prediction model, determine the area affected by industrial activities, and obtain preliminary distribution estimation results. The spatial interpolation optimization module is used to obtain high-resolution grid data by integrating terrain factors for spatially complex areas based on the preliminary distribution estimation results. The pollution distribution generation module is used to generate a pollution distribution map based on the high-resolution grid data, identify areas with low prediction accuracy, and obtain the final optimized layer.

[0012] Technical Effects of this Invention: This invention discloses a machine learning-based method and system for predicting soil heavy metal distribution. Addressing the operational challenges of identifying areas affected by industrial activities and generating high-precision pollution distribution maps, it integrates satellite imagery, topographic factors, and land use data to overcome the difficulties of heterogeneous multi-source data and differences in spatiotemporal scales. First, this invention generates data in a unified format through standardization and spatiotemporal registration, ensuring consistency across multiple data sources. Then, principal component analysis is used to extract key features and reduce dimensionality, constructing a random forest model to optimize the prediction of heavy metal content. Finally, kriging interpolation is combined with topographic factors to generate high-resolution grid data, ultimately forming a pollution distribution map and optimizing low-precision areas. This invention significantly improves the accuracy and spatial resolution of pollution distribution prediction through multi-source data fusion and nonlinear analysis, providing efficient technical support for the accurate identification and environmental management of areas affected by industrial activities. Attached Figure Description

[0013] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating a method for predicting the distribution of heavy metals in soil based on machine learning, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a soil heavy metal distribution prediction system based on machine learning, according to an embodiment of the present invention. Detailed Implementation

[0014] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0016] like Figure 1 As shown, this embodiment provides a method for predicting the distribution of heavy metals in soil based on machine learning, including: Acquire topographic factors, land use, and industrial activity data, perform fusion remote sensing information processing, and determine multi-source datasets; The multi-source dataset is standardized. If the scale difference after standardization exceeds a preset threshold, spatiotemporal registration is performed to obtain data in a unified format. Key features are extracted from the unified format data, and principal component analysis is used to obtain a dimensionality-reduced feature set. A random forest model is constructed based on the dimensionality-reduced feature set, and the parameters are optimized to determine the heavy metal content prediction model; The data from a limited number of sampling points are input into the heavy metal content prediction model to determine the area affected by industrial activities and obtain preliminary distribution estimation results. Based on the preliminary distribution estimation results, high-resolution grid data is obtained by incorporating terrain factors for spatially complex areas. A pollution distribution map is generated based on the high-resolution grid data, areas with low prediction accuracy are identified, and the final optimized layer is obtained.

[0017] Furthermore, identifying multi-source datasets includes: Data on topographic factors, land use, and industrial activities are acquired, high-resolution image data is extracted, sensor bias is eliminated, and a corrected first image dataset is obtained. Based on the first image dataset, topographic factors are extracted, and land use types are segmented by spectral features to determine the first classification result of topography and land use; If the spectral characteristics of the industrial activity area in the first classification result do not match the preset threshold, then the first image dataset and geographic information data are integrated to obtain the fused second classification result. Based on the second classification result, the industrial activity area is further validated, and the classification boundary is optimized by combining spatial resolution to determine the multi-source dataset.

[0018] Specifically, one implementation of this embodiment is as follows: Satellite imagery is fused with publicly available Geographic Information System (GIS) data (such as road networks and building distribution) to generate a secondary classification result. After fusion, the boundaries of industrial areas are clearer, the misclassification rate is reduced by approximately 20%, and the reliability of the classification is improved. For the secondary classification result, a Support Vector Machine (SVM) algorithm is used to perform secondary validation of the industrial activity areas. The classification boundaries are optimized by combining spatial resolution, refining the boundaries through high-resolution imagery, eliminating the influence of mixed pixels, and ultimately generating an accurate dataset of land use and industrial activity distribution. This significantly improves classification accuracy and boundary clarity. The accuracy of industrial activity distribution data can provide a reliable basis for urban planning, reduce land use conflicts, and provide data support for environmental monitoring, optimizing resource allocation efficiency.

[0019] Furthermore, obtaining data in a unified format includes: The original data is obtained from the multi-source dataset, and the format of the original data is unified and the numerical range is adjusted to obtain a standardized dataset. Based on the standardized dataset, it is determined whether the scale difference exceeds a preset threshold. If it does, the time and spatial coordinates are aligned to obtain data in a unified format.

[0020] Specifically, one implementation of this embodiment is as follows: Standardization can be achieved by adjusting the numerical range using mean normalization. The spectral values ​​of Sentinel-2 imagery and the land attribute data from ground monitoring stations are uniformly mapped to a range of 0 to 1. The brightness values ​​of the imagery range from 0 to 10000; after mean normalization, all values ​​are distributed between 0 and 1, facilitating subsequent algorithm processing. Category labels in the ground monitoring data, such as "agriculture" or "industry," can be converted into numerical codes to unify the data format. This method ensures that data from different sources have a consistent scale, improving fusion efficiency. To determine if the scale difference exceeds a preset threshold, a spatial resolution difference threshold of 5 meters and a time interval threshold of 1 day can be set. If the Sentinel-2 data resolution is 10 meters, while the ground monitoring station data corresponds to a 15-meter resolution, exceeding the threshold, spatiotemporal registration is required.

[0021] Furthermore, the heavy metal content prediction model was determined to include: Based on the dimensionality reduction feature set, spatiotemporal data and business classification information are obtained, feature weights are calculated, and a fused feature set is obtained. A random forest model is constructed based on the fused feature set, and the training error is obtained by calculating the difference between the predicted value and the actual value. If the training error is lower than a preset threshold, the tree depth and number of trees in the random forest model are adjusted to obtain an optimized random forest model. The fused feature set is input into the optimized random forest model for training to determine the heavy metal content prediction model.

[0022] Specifically, one implementation of this embodiment is as follows: When constructing a random forest model based on a fused feature set, data can be classified or regressed using random sampling and decision tree ensembles. For predicting heavy metal content in coastal city soils, the random forest model can utilize spectral and topographic information from the fused feature set to predict cadmium or lead concentrations in the soil. The model uses multiple decision trees to vote and comprehensively evaluate the predicted values. Areas with higher near-infrared values ​​and lower slopes are likely to be predicted as agricultural land with lower heavy metal content, while areas with stable spectral values ​​and flatter slopes are likely to be predicted as industrial land. The training error is calculated by comparing the difference between the model's predicted values ​​and the actual values ​​to evaluate model performance.

[0023] Assuming the actual soil cadmium content is 2.5 mg / kg, while the model predicts 2.7 mg / kg, the training error is 0.2 mg / kg. If the preset error threshold is 0.3 mg / kg, the model performance meets the requirements. If the error exceeds the threshold, the model parameters can be optimized using a grid search algorithm. Grid search can test various combinations of tree depth from 5 to 15 and the number of trees from 50 to 200 to find the parameter configuration with the smallest error, such as a tree depth of 10 and a number of trees of 100. The optimized random forest model can predict the heavy metal content of the fused feature set. Assuming the fused feature set contains data points with a spring vegetation index of 0.7 and a slope of 5 degrees, the model may predict that the cadmium content in this area is 1.8 mg / kg. This prediction result can provide a basis for soil pollution monitoring. Areas with higher predicted values ​​can be prioritized for field sampling verification, while areas with lower predicted values ​​can be used as a reference for agricultural land use planning. This method, through multi-source data fusion and model optimization, ensures that the prediction results are closer to reality, providing reliable support for environmental management and land planning.

[0024] Furthermore, the preliminary distribution estimation results include: Limited data is obtained from sampling points. Using a pre-established geographic information database, point information containing latitude, longitude and heavy metal content is extracted, outliers are removed, and a cleaned point dataset is obtained. Based on the cleaned point dataset, the heavy metal content in the unsampled area is calculated, and a spatial distribution grid containing latitude, longitude and predicted content is generated to obtain preliminary spatial distribution data. By analyzing the spatial distribution data and combining it with industrial activity records, the locations of pollution sources were determined, and a pollution source distribution map was obtained. Based on the pollution source distribution map, the spatial distribution data is segmented to obtain preliminary distribution estimation results.

[0025] Specifically, one implementation of this embodiment is as follows: In monitoring heavy metal pollution in coastal city soils, obtaining limited data from sampling sites is a common scenario. Sampling sites typically include latitude and longitude coordinates and heavy metal content information, such as the concentration of cadmium in the soil. Geographic information databases can store this data; for example, a site with latitude and longitude of 120.5°E, 30.2°N and a cadmium content of 2.3 mg / kg. Data cleaning algorithms can identify outliers. For instance, if a site has an abnormally high cadmium content of 10 mg / kg, far exceeding the normal range of 0.5-3.0 mg / kg, statistical methods such as the three-standard-deviation rule can be used to remove outliers, ensuring the reliability of the dataset. The cleaned dataset provides an accurate foundation for subsequent analysis.

[0026] Spatial interpolation algorithms can be used to estimate heavy metal content in unsampled areas. Kriging interpolation is a commonly used method that estimates the content in unknown areas based on the spatial correlation of known points. In an unsampled area located between two sampling points, where cadmium levels were measured at 2.0 and 2.6 mg / kg respectively, Kriging interpolation, considering both distance and weight, predicts a cadmium content of 2.3 mg / kg for this area. The resulting spatial distribution grid includes latitude, longitude, and the predicted content, forming preliminary spatial distribution data. This method effectively fills in data gaps, providing a comprehensive view of pollution distribution.

[0027] Furthermore, obtaining high-resolution grid data includes: Based on the preliminary distribution estimation results, the heavy metal content of the unsampled area is calculated, and the terrain height data is fused to generate a first spatial distribution grid containing latitude, longitude and predicted content, thus obtaining preliminary grid data. If the spatial resolution of the preliminary grid data is lower than a preset threshold, the first spatial distribution grid is analyzed, and the weights are adjusted in combination with terrain factors to generate a second spatial distribution grid that includes latitude and longitude and optimization content, thereby determining high-resolution grid data.

[0028] Specifically, one implementation of this embodiment is as follows: The terrain in a certain area exhibits significant elevation variations, with lower lead concentrations at higher elevations and higher concentrations at lower elevations due to depositional effects. Kernel density estimation, using a weighted topographic factor, generates a second spatial distribution grid, increasing the resolution to 200 meters and optimizing the predicted lead content to 85 mg / kg. This high-resolution grid more accurately reflects the pollution distribution. A regional segmentation algorithm divides the second spatial distribution grid and, combined with pollution source location data, generates the boundary of the affected area. A chemical plant is located at 120.4°E, 30.3°N, and kernel density analysis shows peak lead concentrations in its vicinity. Based on a density clustering algorithm, the high-pollution zone boundary can be defined as 120.3°E to 120.5°E and 30.2°N to 30.4°N, while the low-pollution zone is the surrounding area. This segmentation clearly defines the pollution range. Integrating wind direction data further optimizes pollution source location. Lead emitted from the chemical plant diffuses northeastward due to southwest winds; kernel density estimation can adjust the weights based on wind direction to accurately pinpoint the pollution diffusion path. This multi-source data fusion improves the accuracy of distribution estimation and provides a reliable basis for environmental governance.

[0029] Furthermore, obtaining the final optimized layer includes: Based on the high-resolution grid data, outliers are removed to obtain a cleaned grid dataset; Based on the cleaned grid dataset, the spatial distribution characteristics of pollution source data are analyzed, and the weights are adjusted in combination with terrain factors to generate a first pollution distribution map containing latitude and longitude and predicted pollution content. If the prediction accuracy of the first pollution distribution map is lower than a preset threshold, then the first pollution distribution map is analyzed to determine the areas with low prediction accuracy, and a dataset of low-accuracy areas is obtained. For the low-precision regional dataset, spatially complex regions are segmented, and pollution distribution weights are adjusted by combining terrain factors and pollution source data to generate a second pollution distribution map that includes latitude and longitude and optimized pollution content, thus obtaining the final optimized layer.

[0030] Specifically, one implementation of this embodiment is as follows: In the field of soil heavy metal pollution monitoring, geographic information databases provide crucial support for pollution distribution analysis. These databases typically contain information such as latitude and longitude, pollution source location, terrain elevation, and pollutant concentration. In an industrial park, the database records a location with latitude and longitude of 121.0°E, 30.7°N, terrain elevation of 40 meters, and a soil cadmium content of 2.5 mg / kg. This data provides the foundation for subsequent analysis. Data cleaning algorithms are used to ensure data reliability. If a location has an abnormally high cadmium content of 10 mg / kg, far exceeding the normal range of 1-3 mg / kg, it can be removed using anomaly detection methods based on standard deviation, retaining reliable data, such as 121.1°E, 30.8°N, with a cadmium content of 2.3 mg / kg, providing accurate input for subsequent analysis. Kernel density estimation algorithms are used to analyze the spatial distribution characteristics of pollution sources, adjusting weights based on terrain factors. In this industrial park, the pollution source is an electroplating plant located at 121.2°E, 30.6°N. Kernel density estimation generates a primary pollution distribution map by analyzing cadmium content at sampling points and considering topographical differences. Higher elevations have lower cadmium content due to weaker deposition effects, while lower elevations are found in lower-lying areas. The generated distribution map displays a latitude and longitude grid and predicted cadmium content; for example, a grid point at 121.15°E, 30.65°N has a predicted cadmium content of 2.4 mg / kg. If the accuracy of the primary pollution distribution map is insufficient, such as a deviation of more than 20% between predicted and actual values, further optimization is required. A spatial autocorrelation algorithm is used to identify areas with low prediction accuracy.

[0031] like Figure 2 As shown, this embodiment provides a soil heavy metal distribution prediction system based on machine learning, including: The multi-source data acquisition module is used to acquire topographic factors, land use, and industrial activity data, perform fusion remote sensing information processing, and determine the multi-source dataset; The data standardization processing module is used to perform standardization processing on the multi-source dataset. If the scale difference after standardization exceeds a preset threshold, spatiotemporal registration is performed to obtain data in a unified format. The feature extraction and dimensionality reduction module is used to extract key features based on the unified format data and obtain a dimensionality-reduced feature set by using principal component analysis. The prediction model building module is used to construct a random forest model based on the dimensionality reduction feature set, optimize parameters, and determine the heavy metal content prediction model. The preliminary distribution estimation module is used to input the data from a limited number of sampling points into the heavy metal content prediction model, determine the area affected by industrial activities, and obtain preliminary distribution estimation results. The spatial interpolation optimization module is used to obtain high-resolution grid data by integrating terrain factors for spatially complex areas based on the preliminary distribution estimation results. The pollution distribution generation module is used to generate a pollution distribution map based on the high-resolution grid data, identify areas with low prediction accuracy, and obtain the final optimized layer.

[0032] This invention discloses a machine learning-based method and system for predicting soil heavy metal distribution. Addressing the challenges of identifying industrial activity impact areas and generating high-precision pollution distribution maps, it integrates satellite imagery, topographic factors, and land use data to overcome the difficulties of heterogeneous multi-source data and differences in spatiotemporal scales. First, it generates data in a unified format through standardization and spatiotemporal registration to ensure consistency across multiple data sources. Then, it uses principal component analysis to extract key features and reduce dimensionality, constructing a random forest model to optimize heavy metal content prediction. Finally, it combines kriging interpolation with topographic factors to generate high-resolution grid data, ultimately forming a pollution distribution map and optimizing low-precision areas. Through multi-source data fusion and nonlinear analysis, this invention significantly improves the accuracy and spatial resolution of pollution distribution prediction, providing efficient technical support for the accurate identification and environmental management of industrial activity impact areas.

[0033] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the distribution of heavy metals in soil based on machine learning, characterized in that, include: Acquire topographic factors, land use, and industrial activity data, perform fusion remote sensing information processing, and determine multi-source datasets; The multi-source dataset is standardized. If the scale difference after standardization exceeds a preset threshold, spatiotemporal registration is performed to obtain data in a unified format. Key features are extracted from the unified format data, and principal component analysis is used to obtain a dimensionality-reduced feature set. A random forest model is constructed based on the dimensionality-reduced feature set, and the parameters are optimized to determine the heavy metal content prediction model; The data from a limited number of sampling points are input into the heavy metal content prediction model to determine the area affected by industrial activities and obtain preliminary distribution estimation results. Based on the preliminary distribution estimation results, high-resolution grid data is obtained by incorporating terrain factors for spatially complex areas. A pollution distribution map is generated based on the high-resolution grid data, areas with low prediction accuracy are identified, and the final optimized layer is obtained.

2. The method for predicting soil heavy metal distribution based on machine learning as described in claim 1, characterized in that, Determining multi-source datasets includes: Data on topographic factors, land use, and industrial activities are acquired, high-resolution image data is extracted, sensor bias is eliminated, and a corrected first image dataset is obtained. Based on the first image dataset, topographic factors are extracted, and land use types are segmented by spectral features to determine the first classification result of topography and land use; If the spectral characteristics of the industrial activity area in the first classification result do not match the preset threshold, then the first image dataset and geographic information data are integrated to obtain the fused second classification result. Based on the second classification result, the industrial activity area is further validated, and the classification boundary is optimized by combining spatial resolution to determine the multi-source dataset.

3. The method for predicting soil heavy metal distribution based on machine learning as described in claim 1, characterized in that, Obtaining data in a uniform format includes: The original data is obtained from the multi-source dataset, and the format of the original data is unified and the numerical range is adjusted to obtain a standardized dataset. Based on the standardized dataset, it is determined whether the scale difference exceeds a preset threshold. If it does, the time and spatial coordinates are aligned to obtain data in a unified format.

4. The method for predicting soil heavy metal distribution based on machine learning as described in claim 1, characterized in that, The models for predicting heavy metal content include: Based on the dimensionality reduction feature set, spatiotemporal data and business classification information are obtained, feature weights are calculated, and a fused feature set is obtained. A random forest model is constructed based on the fused feature set, and the training error is obtained by calculating the difference between the predicted value and the actual value. If the training error is lower than a preset threshold, the tree depth and number of trees in the random forest model are adjusted to obtain an optimized random forest model. The fused feature set is input into the optimized random forest model for training to determine the heavy metal content prediction model.

5. The method for predicting soil heavy metal distribution based on machine learning as described in claim 1, characterized in that, The preliminary distribution estimation results include: Limited data is obtained from sampling points. Using a pre-established geographic information database, point information containing latitude, longitude and heavy metal content is extracted, outliers are removed, and a cleaned point dataset is obtained. Based on the cleaned point dataset, the heavy metal content in the unsampled area is calculated, and a spatial distribution grid containing latitude, longitude and predicted content is generated to obtain preliminary spatial distribution data. By analyzing the spatial distribution data and combining it with industrial activity records, the locations of pollution sources were determined, and a pollution source distribution map was obtained. Based on the pollution source distribution map, the spatial distribution data is segmented to obtain preliminary distribution estimation results.

6. The method for predicting soil heavy metal distribution based on machine learning as described in claim 1, characterized in that, Obtaining high-resolution grid data includes: Based on the preliminary distribution estimation results, the heavy metal content of the unsampled area is calculated, and the terrain height data is fused to generate a first spatial distribution grid containing latitude, longitude and predicted content, thus obtaining preliminary grid data. If the spatial resolution of the preliminary grid data is lower than a preset threshold, the first spatial distribution grid is analyzed, and the weights are adjusted in combination with terrain factors to generate a second spatial distribution grid that includes latitude and longitude and optimization content, thereby determining high-resolution grid data.

7. The method for predicting soil heavy metal distribution based on machine learning as described in claim 1, characterized in that, The final optimized layer includes: Based on the high-resolution grid data, outliers are removed to obtain a cleaned grid dataset; Based on the cleaned grid dataset, the spatial distribution characteristics of pollution source data are analyzed, and the weights are adjusted in combination with terrain factors to generate a first pollution distribution map containing latitude and longitude and predicted pollution content. If the prediction accuracy of the first pollution distribution map is lower than a preset threshold, then the first pollution distribution map is analyzed to determine the areas with low prediction accuracy, and a dataset of low-accuracy areas is obtained. For the low-precision regional dataset, spatially complex regions are segmented, and pollution distribution weights are adjusted by combining terrain factors and pollution source data to generate a second pollution distribution map that includes latitude and longitude and optimized pollution content, thus obtaining the final optimized layer.

8. A soil heavy metal distribution prediction system based on machine learning, characterized in that, For implementing the machine learning-based soil heavy metal distribution prediction method as described in any one of claims 1-7, the system includes: a multi-source data acquisition module, used to acquire topographic factors, land use, and industrial activity data, perform fusion remote sensing information processing, and determine the multi-source dataset; The data standardization processing module is used to perform standardization processing on the multi-source dataset. If the scale difference after standardization exceeds a preset threshold, spatiotemporal registration is performed to obtain data in a unified format. The feature extraction and dimensionality reduction module is used to extract key features based on the unified format data and obtain a dimensionality-reduced feature set by using principal component analysis. The prediction model building module is used to construct a random forest model based on the dimensionality reduction feature set, optimize parameters, and determine the heavy metal content prediction model. The preliminary distribution estimation module is used to input the data from a limited number of sampling points into the heavy metal content prediction model, determine the area affected by industrial activities, and obtain preliminary distribution estimation results. The spatial interpolation optimization module is used to obtain high-resolution grid data by integrating terrain factors for spatially complex areas based on the preliminary distribution estimation results. The pollution distribution generation module is used to generate a pollution distribution map based on the high-resolution grid data, identify areas with low prediction accuracy, and obtain the final optimized layer.

Images