Mining area ecological carbon sequestration evaluation method, system, equipment and medium

Abstract

The invention relates to a mining area ecological carbon sequestration evaluation method, system and device and a medium. The method comprises the following steps: performing structured processing on original data of a mining area to generate a preprocessed data set; inputting the preprocessed data set into a preset ecological dynamic model, outputting a dynamic model calculation result, and generating a training data set; based on the training data set, training a preset machine learning model to obtain a carbon sequestration prediction model, inputting the to-be-predicted data set into the carbon sequestration prediction model for prediction, and outputting a prediction result and a feature importance weight; and based on natural language query of the user, target data is retrieved from a calculation result and a prediction result of the kinetic model, and an interpretive response is generated in combination with the feature importance weight. According to the method, the precision, the efficiency and the result interpretability of mining area ecological carbon sequestration evaluation are improved by fusing the ecological dynamics model and the machine learning model, meanwhile, the user interaction capability is enhanced, and specific query requirements can be quickly responded.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, and in particular relates to a method, system, equipment and medium for assessing ecological carbon sequestration in mining areas. Background Technology

[0002] As the core area of ​​high-intensity human resource extraction activities, mining areas are directly constrained by problems such as surface vegetation destruction, soil structure disturbance, and carbon pool imbalance, which restrict the stable functioning of the ecosystem's carbon sequestration. Therefore, accurate assessment of the ecological carbon sequestration capacity of mining areas has become the core technical support for the formulation of ecological restoration plans, monitoring of restoration effects, and management of carbon sink resources.

[0003] Existing carbon sequestration assessment technologies in mining areas are mainly divided into three categories: The first is the traditional field survey method, which obtains carbon storage data by manually collecting vegetation and soil profile samples and combining them with laboratory analysis. However, this method requires a large investment of manpower and resources, has a long survey cycle, and limited coverage, making it difficult to meet the assessment needs of large-scale and dynamic mining areas. Furthermore, the data has poor timeliness and cannot reflect changes in carbon sequestration capacity during ecological restoration. The second is the single ecological dynamics model assessment method. This type of method is based on mechanisms such as vegetation growth and soil carbon cycling. While it can reflect the inherent laws of carbon sequestration processes, it is inadequate for assessing the heterogeneity of mining area topography and the diversity of vegetation types. The first approach suffers from several drawbacks. First, it lacks adaptability, making model parameter calibration difficult and limiting prediction accuracy in complex mining scenarios. It also fails to output the weights of feature variables on carbon sequestration. Second, it employs single machine learning model prediction methods, such as decision tree models, random forest models, or support vector machine models. While these methods can improve prediction accuracy through extensive data training, they suffer from "black box" computational defects. They cannot explain the rationality of prediction results by incorporating ecological dynamics mechanisms and lack the ability to interact with users' natural language queries, making it difficult to quickly respond to carbon sequestration data retrieval needs and influencing factor analysis needs for specific land types and years. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, system, equipment, and medium for assessing ecological carbon sequestration in mining areas to address the aforementioned technical issues. This aims to improve the accuracy and efficiency of ecological carbon sequestration assessment in mining areas, enhance the interpretability of assessment results and user interactivity, and better meet the needs of developing ecological restoration plans, monitoring restoration effects, and managing carbon sink resources in mining areas.

[0005] Firstly, this application provides a method for assessing ecological carbon sequestration in mining areas, including:

[0006] The raw data from the mining area is structured to extract parameters such as vegetation growth rate, vegetation coverage, vegetation mortality rate, soil organic carbon content, and land area, generating a preprocessed dataset. The preprocessed dataset is then input into a pre-defined ecological dynamics model, which outputs the dynamics model calculation results, including vegetation carbon storage, soil carbon storage, biodiversity index, and total carbon sequestration. Based on the vegetation carbon storage, soil carbon storage, vegetation coverage, soil organic carbon content, and total carbon sequestration, a training dataset is generated.

[0007] Based on the training dataset, a carbon sequestration prediction model is obtained by training a preset machine learning model. A dataset to be predicted with the same data structure as the preprocessed dataset is obtained. The dataset to be predicted is input into the carbon sequestration prediction model for prediction, and the prediction results and feature importance weights are output. The preset machine learning model can be any one of the following models: decision tree model, random forest model, or support vector machine model.

[0008] Based on the user's natural language query, target data is retrieved from the calculation and prediction results of the dynamic model, and an explanatory response is generated by combining feature importance weights.

[0009] In one embodiment, the preprocessed dataset is input into a preset ecological dynamics model, and the output includes the dynamics model calculation results containing vegetation carbon storage, soil carbon storage, biodiversity index, and total carbon sequestration, including:

[0010] Based on the vegetation growth rate, vegetation cover, and vegetation mortality rate in the preprocessed dataset, vegetation carbon storage is calculated using a recursive formula. The recursive formula for vegetation carbon storage is as follows:

[0011]

[0012]

[0013] in, This represents the total vegetation mortality rate in year t. This indicates the total number of vegetation types within the preset area. Indicates the first Vegetation mortality rate of similar vegetation Indicates the t-th year Vegetation carbon storage of vegetation-like structures This represents the vegetation carbon storage in year t. This represents the vegetation carbon storage in year t-1. This represents the vegetation growth rate in year t. This represents the vegetation cover rate in year t.

[0014] Based on the soil organic carbon content and land area in the preprocessed dataset, the soil carbon storage is calculated using the soil carbon storage formula:

[0015]

[0016] in, This represents the soil carbon storage in year t. This represents the soil organic carbon content in year t. This represents the total vegetation area of ​​the preset region.

[0017] Based on vegetation carbon storage and soil carbon storage, the biodiversity index is calculated using the biodiversity index formula; the biodiversity index formula is:

[0018]

[0019] in, This represents the biodiversity index for year t. This represents the weighting coefficient corresponding to vegetation carbon storage. This represents the weighting coefficient corresponding to soil carbon storage, and satisfies... + =1;

[0020] The total carbon sequestration is obtained by adding the vegetation carbon storage and the soil carbon storage.

[0021] The system integrates vegetation carbon storage, soil carbon storage, biodiversity index, and total carbon sequestration, and outputs the results of the dynamic model calculation.

[0022] In one embodiment, a carbon sequestration prediction model is trained on a pre-defined machine learning model based on a training dataset. A dataset to be predicted, with a data structure consistent with the pre-processed dataset, is then obtained. This dataset is input into the carbon sequestration prediction model for prediction, and the prediction results and feature importance weights are output, including:

[0023] Vegetation carbon storage, soil carbon storage, vegetation coverage, and soil organic carbon content were extracted as feature variables from the training dataset, and total carbon sequestration was extracted as the target variable.

[0024] The feature variables and target variables are input into a pre-defined machine learning model, and the model is trained to obtain a carbon sequestration prediction model. This carbon sequestration prediction model is trained through the following steps:

[0025] If the preset machine learning model is a decision tree model, the optimal splitting feature is obtained by calculating the information gain, gain ratio or Gini impurity of each feature variable, and the decision tree node is recursively constructed in combination with the target variable until the preset stopping condition is met. Then, post-pruning is performed to obtain the carbon sequestration prediction model.

[0026] If the preset machine learning model is a random forest model, then a subset of training samples is selected from the training dataset by sampling with replacement, a subset of features is randomly extracted from the subset of training samples, and multiple independent decision trees are constructed by combining the target variable. The independent decision trees are trained and integrated by the feature subsets to obtain the carbon sequestration prediction model.

[0027] If the preset machine learning model is a support vector machine model, then the linear separability of the sample data is determined based on the sample data composed of feature variables and target variables. If the sample data is non-linearly separable, the feature variables are mapped to a high-dimensional space through a kernel function to obtain linearly separable feature data in the high-dimensional space. Based on the linearly separable feature data and target variables, slack variables are introduced and hyperparameters are configured to construct a constrained optimization problem with the goal of minimizing the classification margin loss. The constrained optimization problem is solved to obtain the optimal parameters of the maximum margin classification hyperplane. The optimal parameters of the maximum margin classification hyperplane are used as core components to form a carbon sequestration prediction model.

[0028] Obtain the dataset to be predicted, which includes the current vegetation growth rate, current vegetation coverage, current vegetation mortality rate, current soil organic carbon content, and current land area parameters corresponding to the preprocessed dataset.

[0029] Input the dataset to be predicted into the carbon sequestration prediction model, and perform calculations through the carbon sequestration prediction model to obtain the prediction results;

[0030] Based on the carbon sequestration prediction model, the influence weights of each feature variable on the total carbon sequestration are calculated, and feature importance weights are generated.

[0031] In one embodiment, based on the user's natural language query, target data is retrieved from the results of dynamic model calculations and predictions, and an explanatory response is generated by combining feature importance weights, including:

[0032] Based on a pre-defined terminology database for ecological carbon sequestration in mining areas, semantic parsing is performed on natural language queries to extract target land type, target year, and target carbon sequestration indicators.

[0033] Based on the target land type and target year, matching target data is retrieved from the dynamic model calculation results and prediction results to obtain the retrieval results;

[0034] Determine the source of the target data in the search results. If the target data comes from the prediction results, sort the key influencing features according to their weight based on the importance weight, and generate an explanatory statement containing the degree of influence of each key influencing feature on the target carbon sequestration index. If the target data comes from the calculation results of the kinetic model, generate an explanatory statement on the calculation basis of the corresponding target carbon sequestration index.

[0035] Integrate target data, explanatory statements, or statements explaining the basis for calculations to generate and output an explanatory response.

[0036] In one embodiment, the dataset to be predicted is input into the carbon sequestration prediction model, and the prediction result is obtained by performing calculations through the carbon sequestration prediction model, including:

[0037] If the carbon sequestration prediction model is a decision tree model, then extract the feature variables to be predicted from the dataset to be predicted, recursively traverse the decision tree structure through the node judgment rules of the decision tree model, output the first predicted total carbon sequestration, and output the first predicted total carbon sequestration as the prediction result.

[0038] If the carbon sequestration prediction model is a random forest model, then the feature variables to be predicted in the dataset to be predicted are input into each independent decision tree of the random forest model to obtain the independent prediction value of each independent decision tree. The average value of each independent prediction value is taken, and the second predicted total carbon sequestration is output. The second predicted total carbon sequestration is output as the prediction result.

[0039] If the carbon sequestration prediction model is a support vector machine model, then the feature variables to be predicted in the dataset to be predicted are mapped to a high-dimensional space through a kernel function to obtain the feature data to be predicted in the high-dimensional space. The feature data to be predicted is substituted into the optimal parameters of the maximum margin classification hyperplane to calculate the third predicted total carbon sequestration, and the third predicted total carbon sequestration is output as the prediction result.

[0040] Secondly, this application also provides a mining area ecological carbon sequestration assessment system, including:

[0041] The data preprocessing and feature extraction module is used to perform structured processing on the raw data of the mining area, extract parameters such as vegetation growth rate, vegetation coverage, vegetation mortality rate, soil organic carbon content, and land area, and generate a preprocessed dataset. The preprocessed dataset is then input into a preset ecological dynamics model, which outputs the dynamics model calculation results including vegetation carbon storage, soil carbon storage, biodiversity index, and total carbon sequestration. Based on the vegetation carbon storage, soil carbon storage, vegetation coverage, soil organic carbon content, and total carbon sequestration, a training dataset is generated.

[0042] The model calculation and training data generation module is used to train a preset machine learning model based on the training dataset to obtain a carbon sequestration prediction model, obtain a dataset to be predicted with the same data structure as the preprocessed dataset, input the dataset to be predicted into the carbon sequestration prediction model for prediction, and output the prediction results and feature importance weights; wherein the preset machine learning model is any one of the decision tree model, random forest model or support vector machine model.

[0043] The carbon sequestration prediction and result interpretation module is used to retrieve target data from the calculation and prediction results of the kinetic model based on the user's natural language query, and generate an interpretive response by combining feature importance weights.

[0044] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the first aspect.

[0045] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the first aspect.

[0046] The aforementioned method, system, equipment, and medium for assessing ecological carbon sequestration in mining areas firstly involves structuring the raw data from the mining area and extracting parameters to generate a preprocessed dataset. This preprocessed dataset is then input into an ecological dynamics model to output results and generate a training dataset. This addresses the issues of inconsistent data formats and insufficient adaptability of single-mechanism models in traditional mining area carbon sequestration assessments, improving the standardization of the assessment data foundation and the mechanistic fit of the dynamic calculations. Secondly, a pre-defined machine learning model is trained based on the training dataset to obtain a carbon sequestration prediction model. This model is then input into the dataset to be predicted and outputs prediction results and feature importance weights, combining ecological dynamics mechanisms with machine learning prediction capabilities. This improves the accuracy of carbon sequestration prediction and the interpretability of feature impacts. Finally, the system retrieves target data based on user natural language queries and generates explanatory responses by combining feature importance weights. This addresses the lack of interactivity in traditional assessment methods, improving the convenience of carbon sequestration data retrieval and the understandability of the results. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart of a method for assessing ecological carbon sequestration in mining areas, provided as an exemplary embodiment of the present invention;

[0049] Figure 2 A flowchart of a method for generating an interpretive response is provided as an exemplary embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of a mining area ecological carbon sequestration assessment system provided as an exemplary embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0052] In one embodiment, such as Figure 1 As shown, a method for assessing ecological carbon sequestration in mining areas is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0053] S101: The original data of the mining area is structured and extracted to obtain parameters such as vegetation growth rate, vegetation coverage, vegetation mortality rate, soil organic carbon content and land area, generating a preprocessed dataset. The preprocessed dataset is then input into a preset ecological dynamics model, which outputs the dynamics model calculation results including vegetation carbon storage, soil carbon storage, biodiversity index and total carbon sequestration. Based on the vegetation carbon storage, soil carbon storage, vegetation coverage, soil organic carbon content and total carbon sequestration, a training dataset is generated.

[0054] Specifically, raw data from mining areas can be obtained through multiple sources, including remote sensing image interpretation, field ecological sampling, and soil physicochemical analysis. However, the data formats (text, tables, raster, etc.) and dimensions (such as vegetation cover percentage and soil organic carbon content) exhibit heterogeneity. Therefore, the raw mining area data can be structured by unifying field definitions, standardizing dimensions, and filling in missing values ​​(based on statistical averages of similar areas within the mining area). This eliminates the interference of data heterogeneity on subsequent model calculations and ensures the consistency and validity of the input data. Subsequently, parameters such as vegetation growth rate, vegetation cover, vegetation mortality rate, soil organic carbon content, and land area can be extracted from the structured data. These parameters are the core driving factors of carbon cycling in the mining area ecosystem. Among them, vegetation growth rate corresponds to the intensity of carbon absorption, vegetation cover determines the spatial scale of carbon absorption, vegetation mortality rate is the natural output of the carbon pool, soil organic carbon content is the basic quantitative indicator of the soil carbon pool, and land area is a necessary basis for converting the spatial scale of carbon storage. By integrating these parameters, a preprocessed dataset can be obtained. By inputting a preprocessed dataset into a predefined ecological dynamics model, the carbon exchange process between vegetation and soil systems can be quantified based on the mechanistic logic of the carbon cycle and the model's built-in carbon flow simulation rules, outputting the dynamics model calculation results. Among these, vegetation carbon storage and soil carbon storage represent mechanistic carbon sequestration results, the biodiversity index is a correlation indicator between carbon sequestration capacity and ecological stability, and total carbon sequestration is the comprehensive carbon sequestration level of the vegetation-soil system. A training dataset is generated based on these results and feature parameters, providing a training foundation of mechanistic labels and feature variables for subsequent machine learning models. This allows for the combination of mechanistic logic and data mining capabilities, avoiding the limitations of a single model.

[0055] S102: Based on the training dataset, train the preset machine learning model to obtain the carbon sequestration prediction model, obtain the dataset to be predicted with the same data structure as the preprocessed dataset, input the dataset to be predicted into the carbon sequestration prediction model for prediction, and output the prediction results and feature importance weights; wherein the preset machine learning model is any one of the decision tree model, random forest model or support vector machine model.

[0056] Specifically, machine learning models can learn the complex relationships between carbon sequestration and various ecological parameters through large amounts of historical data, thereby enabling predictions of unknown data. Illustratively, the preset machine learning model can be any of the following: decision tree model, random forest model, or support vector machine model. These models have different advantages when processing ecological data. Decision tree models are easy to understand and interpret, random forest models can effectively handle noise and outliers in the data, while support vector machine models have good classification and regression performance in high-dimensional data spaces. By obtaining a dataset to be predicted with the same data structure as the preprocessed dataset, this dataset is input into the carbon sequestration prediction model for prediction. This model can quickly output prediction results and feature importance weights. The feature importance weights reflect the degree of contribution of each ecological parameter to the carbon sequestration prediction, helping to further understand which factors in the ecosystem have the greatest impact on carbon sequestration capacity.

[0057] S103: Based on the user's natural language query, retrieve target data from the dynamic model calculation results and prediction results, and generate an explanatory response by combining feature importance weights.

[0058] Specifically, data retrieval based on users' natural language queries can extract target information such as target land type, target year, and target carbon sequestration index through semantic parsing logic. Then, based on preset data mapping rules, matching data can be located from the dynamic model calculation results and prediction results. Furthermore, in addition to outputting target data, the influence logic of each carbon sequestration factor on the data can be supplemented to generate explanatory responses. For example, when a user queries the carbon sequestration amount of a certain region in a certain year, the response can not only include the specific carbon sequestration amount but also explain that "vegetation coverage is the main influencing factor, with a weight of 35%", etc. This effectively solves the problem of traditional assessment methods that only output data but cannot explain the causes of the results, improving the application guidance and understandability of the assessment results.

[0059] The above method first provides a data foundation for ecological carbon sequestration assessment by structuring the raw data of the mining area and extracting key parameters, thus enhancing the reliability of the assessment. Second, by integrating ecological dynamics models and machine learning models, it overcomes the limitations of single methods, improving the accuracy and efficiency of the assessment, while also outputting feature importance weights to enhance the interpretability of the results. Finally, by incorporating natural language query functionality, it quickly responds to specific user queries, improving user interactivity and the practicality of the assessment.

[0060] In one embodiment, the preprocessed dataset is input into a predefined ecological dynamics model, and the output includes the dynamics model calculation results containing vegetation carbon storage, soil carbon storage, biodiversity index, and total carbon sequestration, including:

[0061] Based on the vegetation growth rate, vegetation coverage and vegetation mortality rate in the preprocessed dataset, the vegetation carbon storage is calculated using the vegetation carbon storage recursive formula.

[0062] Based on the soil organic carbon content and land area in the preprocessed dataset, the soil carbon storage is calculated using the soil carbon storage formula.

[0063] Based on vegetation carbon storage and soil carbon storage, the biodiversity index is calculated using the biodiversity index formula.

[0064] The total carbon sequestration is obtained by adding the vegetation carbon storage and the soil carbon storage.

[0065] The system integrates vegetation carbon storage, soil carbon storage, biodiversity index, and total carbon sequestration, and outputs the results of the dynamic model calculation.

[0066] Specifically, vegetation carbon storage can be calculated based on the core mechanisms of ecosystem carbon cycling. This means that vegetation absorbs carbon through photosynthesis to form carbon storage, while simultaneously releasing carbon through natural death. The dynamic changes in vegetation carbon storage need to be comprehensively quantified by combining the growth and death processes. Therefore, the calculation process can begin by obtaining the carbon loss for the target year using the total vegetation mortality rate formula, and then combining this with the previous year's carbon storage and the current year's carbon absorption to derive the vegetation carbon storage for the target year. For example, the total vegetation mortality rate formula can be:

[0067]

[0068] in, This represents the total vegetation mortality rate in year t, expressed in MgC / ha; This represents the total number of vegetation types within the preset area. Common vegetation types in mining areas include Cropland, Forest, Grassland, etc., which need to be determined through vegetation classification information in the preprocessed dataset. Indicates the first The vegetation mortality rate of similar vegetation can be determined based on long-term observation data of similar vegetation in the mining area or industry-standard statistical data. Indicates the t-th year Vegetation carbon storage, measured in MgC / ha, is a parameter for the initial year that can be calibrated using in-situ sampling measurements or remote sensing data. The formula for the target year's vegetation carbon storage is:

[0069]

[0070] in, This represents the vegetation carbon storage in year t, expressed in MgC / ha; The vegetation carbon storage in year t-1 is expressed in MgC / ha. If it is the initial year, the preset baseline carbon storage value is used. This represents the vegetation growth rate in year t, expressed in MgC / ha, which can be directly extracted from the preprocessed dataset. The vegetation cover in year t is expressed as a proportion, ranging from 0 to 1, and is derived from the preprocessed dataset. This recursive formula allows for precise quantification of the annual dynamic changes in vegetation carbon storage, reflecting the temporal evolution of the carbon sequestration capacity of vegetation in the mining area.

[0071] Specifically, soil carbon storage is an important component of the carbon pool in the mining area ecosystem. Its quantification hinges on extending the soil organic carbon content per unit area to the total regional scale. The calculation formula can be:

[0072]

[0073] in, This represents the soil carbon storage in year t, expressed in MgC. The soil organic carbon content in year t is expressed in MgC / ha, and the standardized value is extracted from the preprocessed dataset. The total vegetation area of ​​the preset region is expressed in ha, calculated using land area parameters and the proportion of vegetation cover in the preprocessed dataset. The formula is based on the linear correlation between soil organic carbon content and soil carbon storage. This means that a healthy soil carbon pool, with higher organic carbon content, supports a richer microbial community, indirectly affecting vegetation growth and overall carbon sequestration efficiency. This aligns with the synergistic mechanism of soil and vegetation in mining area ecosystems.

[0074] Specifically, there is a significant correlation between biodiversity and the carbon sequestration capacity of ecosystems. High vegetation carbon storage indicates good vegetation cover and ecosystem stability, while high soil carbon storage signifies excellent soil structure and sufficient nutrients. Both can jointly provide suitable habitats for organisms. Therefore, this correlation can be quantified by a linear combination formula, which is as follows:

[0075]

[0076] in, This represents the biodiversity index for year t. This represents the weighting coefficient corresponding to vegetation carbon storage. This represents the weighting coefficient corresponding to soil carbon storage, and satisfies... + =1;

[0077] in, denoted as the biodiversity index for year t, which is a dimensionless value; a higher value indicates richer biodiversity. Indicates the weighting coefficient corresponding to vegetation carbon storage; This represents the weighting coefficient corresponding to soil carbon storage, and satisfies... + =1, and this coefficient needs to be calibrated based on ecological survey data of the mining area. Weights can be assigned according to the impact of vegetation carbon storage and soil carbon storage on local biodiversity. This formula quantifies the correlation between carbon sequestration indicators and ecological stability indicators, providing support for subsequent comprehensive assessments of the ecological quality of the mining area.

[0078] Specifically, the carbon sequestration capacity of a mining area ecosystem is composed of both vegetation carbon pools and soil carbon pools. Vegetation carbon storage reflects the carbon fixation level of the plant community, while soil carbon storage reflects the soil's ability to sequester carbon. Therefore, the sum of the two is the total carbon sequestration, which comprehensively reflects the overall carbon sequestration level of the mining area ecosystem. Furthermore, vegetation carbon storage and soil carbon storage correspond to the carbon sequestration status of the two core carbon pools, respectively. The biodiversity index reflects the ecological synergy in the carbon sequestration process, and the total carbon sequestration provides a quantitative result of the overall carbon sequestration level. Therefore, the integration of these four indicators can fully meet the needs of subsequent applications. The integration process can follow a unified data format standard, associating various indicators with corresponding year and regional information to form structured dynamic model calculation results. This result provides training data with mechanistic labels for machine learning models and multi-dimensional basic data support for user queries, ensuring the smooth progress of subsequent processes.

[0079] In one embodiment, a carbon sequestration prediction model is obtained by training a pre-defined machine learning model based on a training dataset. A dataset to be predicted, with a data structure consistent with the pre-processed dataset, is then input into the carbon sequestration prediction model for prediction. The prediction result and feature importance weights are output, including:

[0080] Vegetation carbon storage, soil carbon storage, vegetation coverage, and soil organic carbon content were extracted as feature variables from the training dataset, and total carbon sequestration was extracted as the target variable.

[0081] The feature variables and target variables are input into a pre-defined machine learning model, and the model is trained to obtain a carbon sequestration prediction model. This carbon sequestration prediction model is trained through the following steps:

[0082] If the preset machine learning model is a decision tree model, the optimal splitting feature is obtained by calculating the information gain, gain ratio or Gini impurity of each feature variable, and the decision tree node is recursively constructed in combination with the target variable until the preset stopping condition is met. Then, post-pruning is performed to obtain the carbon sequestration prediction model.

[0083] If the preset machine learning model is a random forest model, then a subset of training samples is selected from the training dataset by sampling with replacement, a subset of features is randomly extracted from the subset of training samples, and multiple independent decision trees are constructed by combining the target variable. The independent decision trees are trained and integrated by the feature subsets to obtain the carbon sequestration prediction model.

[0084] If the preset machine learning model is a support vector machine model, then the linear separability of the sample data is determined based on the sample data composed of feature variables and target variables. If the sample data is non-linearly separable, the feature variables are mapped to a high-dimensional space through a kernel function to obtain linearly separable feature data in the high-dimensional space. Based on the linearly separable feature data and target variables, slack variables are introduced and hyperparameters are configured to construct a constrained optimization problem with the goal of minimizing the classification margin loss. The constrained optimization problem is solved to obtain the optimal parameters of the maximum margin classification hyperplane. The optimal parameters of the maximum margin classification hyperplane are used as core components to form a carbon sequestration prediction model.

[0085] Obtain the dataset to be predicted, which includes the current vegetation growth rate, current vegetation coverage, current vegetation mortality rate, current soil organic carbon content, and current land area parameters corresponding to the preprocessed dataset.

[0086] Input the dataset to be predicted into the carbon sequestration prediction model, and perform calculations through the carbon sequestration prediction model to obtain the prediction results;

[0087] Based on the carbon sequestration prediction model, the influence weights of each feature variable on the total carbon sequestration are calculated, and feature importance weights are generated.

[0088] Specifically, the training dataset includes core carbon sequestration indicators and basic ecological parameters calculated by the dynamic model. Based on the inherent correlation mechanism of carbon sequestration in the mining area, feature variables can be selected. Since vegetation carbon storage and soil carbon storage are direct components of carbon sequestration, vegetation cover determines the spatial range of carbon absorption, and soil organic carbon content is the core quantitative indicator of soil carbon pool, these four types of variables collectively cover the key links in the carbon cycle. Therefore, using vegetation carbon storage, soil carbon storage, vegetation cover, and soil organic carbon content as feature variables ensures that the main influencing factors of carbon sequestration are captured during subsequent model training. Furthermore, the target variable, total carbon sequestration, can be used as a comprehensive evaluation indicator, directly reflecting the overall carbon sequestration capacity of the mining area ecosystem. Its correlation with the feature variables is the core objective of model training; predicting this variable allows for accurate quantification of the carbon sequestration level in the mining area. Subsequently, the feature variables and target variables are input into a pre-set machine learning model for training, resulting in a carbon sequestration prediction model. Illustratively, if the pre-set machine learning model is a decision tree model, the core of the decision tree model is to progressively divide the data through feature splitting. Therefore, the optimal splitting feature can be selected based on the calculation results by first calculating the information gain, gain ratio, or Gini impurity of each feature variable. Information gain can be used to assess feature importance by calculating the change in the disorder of the dataset before and after feature splitting. The disorder of the dataset is represented by entropy, and the formula for calculating entropy is:

[0089]

[0090] in, Represents the dataset entropy, Represents the dataset The set of categories of the target variable. Indicates category In the dataset The proportion of samples in the sample. The formula for calculating information gain is:

[0091]

[0092] in, Representation of features For dataset Information gain Representation of features The set of all possible values, Represents the dataset Chinese characteristics Values a subset of and Representing datasets respectively and subsets The number of samples is considered. Gain ratio, in addition to information gain, can also incorporate intrinsic feature values ​​to mitigate the bias of information gain towards features with multiple values. Gini impurity can be used to evaluate the splitting effect by calculating the purity of the dataset; its core is to measure the probability that two randomly drawn samples will have inconsistent classes. After calculation, the feature with the highest information gain, highest gain ratio, or lowest Gini impurity can be selected as the optimal splitting feature for the current node, and the node subset is divided based on the range of the target variable. Subsequently, during the recursive construction of subtrees, preset stopping conditions can include, but are not limited to, the tree depth reaching a preset threshold, the number of node samples being less than the minimum sample threshold, and the purity of the target variable within a node reaching a preset purity threshold. Furthermore, even after the preset stopping conditions are met, post-pruning can be performed, that is, after the complete decision tree is built, branches that contribute little to the prediction accuracy can be deleted, ultimately forming a carbon sequestration prediction model with strong generalization ability. In addition, pre-pruning during tree construction can be performed to terminate the split early and avoid overfitting.

[0093] Specifically, if the preset machine learning model is a random forest model, which can improve the accuracy and stability of the model based on the principle of ensemble learning by collaborative prediction of multiple decision trees, then a training sample subset can be selected from the training dataset using bootstrap sampling. From this subset, a feature subset can be randomly extracted, and multiple independent decision trees can be constructed by combining them with the target variable. The carbon sequestration prediction model is then obtained by training and integrating these independent decision trees using the feature subsets. Bootstrap sampling randomly extracts samples from the original training dataset to form a new training sample subset. The number of samples in each subset is the same as in the original training dataset. During the sampling process, some samples may be extracted repeatedly, and some may not be extracted. This method ensures the diversity of training data for each decision tree. The feature subset extraction can employ a random selection strategy, where each time a node is split, a portion of the features (usually the square root of the total number of features or one-third of the total number of features) is randomly selected as the splitting candidate features for that node. This operation further enhances the independence of each decision tree and reduces the excessive influence of a single feature on the model's prediction results. The training process for each independent decision tree is consistent with the aforementioned decision tree model, involving feature selection, node splitting, recursive construction, and pruning. After multiple decision trees are trained, they can be integrated to form a random forest model. The integration logic can be based on the regression task characteristics of carbon sequestration prediction, employing a mean fusion strategy. This involves averaging the prediction results of multiple decision trees to reduce the prediction variance of a single decision tree, thereby improving the model's adaptability and prediction robustness to complex ecological data in mining areas.

[0094] Specifically, if the preset machine learning model is a support vector machine (SVM) model, the core of the SVM model is to find the optimal hyperplane that maximizes the class margin. For the regression task of predicting carbon sequestration in mining areas, kernel functions can be used to handle the nonlinear relationship between feature variables and target variables. For example, the linear correlation matrix of the feature variables can be calculated first. Based on the sample data composed of feature variables and target variables, the linear separability of the sample data can be determined. If the rank of the matrix equals the feature dimension, it is considered linearly separable; otherwise, it is nonlinearly separable. The kernel function can be a radial basis function (RBF), a polynomial kernel function, or a linear kernel function. The radial basis function is preferred because it is more adaptable to nonlinear data. Its formula is:

[0095]

[0096] in, Indicates sample and samples The inner product after kernel function mapping The bandwidth parameter represents the kernel function and is used to control the range of influence of the kernel function. Indicates sample and samples The Euclidean distance between them.

[0097] After mapping feature variables to a high-dimensional space using kernel functions, the originally non-linearly separable sample data is transformed into linearly separable feature data. At this point, the objective of constructing the maximum margin classification hyperplane can be transformed into solving a constrained optimization problem. Illustratively, slack variables can be introduced during the construction process. To allow for a small number of misclassified samples, and to configure the hyperparameter C to balance the classification margin size with the number of misclassified samples, the objective function of the constrained optimization problem is:

[0098]

[0099] The constraints are:

[0100]

[0101] in, Describes the normal vector of the hyperplane. The bias term represents the hyperplane. Indicates hyperparameters, Indicates the first Slack variables for each sample, Indicates the number of samples. Indicates the first The target variable values ​​for each sample Indicates the first The high-dimensional feature vector obtained by mapping the feature variables of each sample using a kernel function. This represents the inner product of the normal vector and the high-dimensional eigenvector. The optimal parameters of the hyperplane are obtained by solving this constrained optimization problem using the Lagrange multiplier method. and Using this optimal parameter as a core component, a support vector machine carbon sequestration prediction model can be formed that can accurately fit the relationship between carbon sequestration and feature variables.

[0102] Subsequently, the dataset to be predicted can be obtained by following the principle of data structure consistency, ensuring that its field definitions and unit standards are completely consistent with those of the preprocessed dataset, thus avoiding model prediction errors caused by differences in input data formats. The current vegetation growth rate, current vegetation coverage, and current vegetation mortality rate are used to indirectly reflect the dynamics of vegetation carbon storage during the target period, while the current soil organic carbon content and current land area are used to indirectly reflect the soil carbon storage level during the target period. These parameters can be processed using the same structured processing procedure as the preprocessed dataset (unifying fields, standardizing units, and filling in missing values) to form a dataset to be predicted that meets the model input requirements, providing a reliable data foundation for subsequent prediction calculations. By inputting the dataset to be predicted into the carbon sequestration prediction model, the model can quantitatively predict the total carbon sequestration in the mining area during the target period based on the association rules between the feature variables and the target variable learned during the training phase, obtaining prediction results and providing direct data support for mining area carbon sequestration assessment. Furthermore, based on the carbon sequestration prediction model, the influence weights of each feature variable on the total carbon sequestration can be calculated, quantifying the contribution of each feature variable to the prediction results and generating feature importance weights. This provides a quantitative basis for targeted optimization of ecological restoration in mining areas. For decision tree models, the feature contribution can be calculated as the influence weight by statistically analyzing the number of splits of each feature variable in the tree structure and the increase in purity of split nodes. For random forest models, the feature importance weights can be obtained by summing the total reduction in node impurity of each feature variable across all independent decision trees and normalizing the result. For support vector machine models, the influence weights of each feature variable can be quantified by calculating the absolute value of the weight coefficients corresponding to the feature variables and combining this with the feature contribution after kernel function mapping. The generated feature importance weights are presented numerically; a higher weight value indicates a greater impact of the feature on carbon sequestration, which can directly guide the prioritization of ecological restoration measures in mining areas.

[0103] In one embodiment, the dataset to be predicted is input into a carbon sequestration prediction model, and the prediction results are obtained through calculations performed by the carbon sequestration prediction model, including:

[0104] If the carbon sequestration prediction model is a decision tree model, then the feature variables to be predicted in the dataset to be predicted are extracted, and the decision tree structure is recursively traversed according to the node judgment rules of the decision tree model to output the first predicted total carbon sequestration. The first predicted total carbon sequestration is then output as the prediction result.

[0105] Specifically, the feature variables to be tested can be extracted from the dataset to be predicted, and their types are completely consistent with those in the training phase, including vegetation carbon storage, soil carbon storage, vegetation cover, and soil organic carbon content. Then, based on the optimal splitting features and splitting thresholds determined in the training phase, the decision tree model traverses from the root node, selecting corresponding child nodes according to the relationship between the value of the feature variable to be tested and the node splitting threshold, continuing the traversal until a leaf node is reached. The mean or mode of the target variable corresponding to the leaf node is the first predicted total carbon sequestration amount for that sample. This process follows the classification and regression logic of the decision tree, ensuring that the prediction results are consistent with the association rules learned during model training, thereby achieving accurate quantification of carbon sequestration amount for the target time period.

[0106] If the carbon sequestration prediction model is a random forest model, then the feature variables to be predicted in the dataset to be predicted are input into each independent decision tree of the random forest model to obtain the independent prediction value of each independent decision tree. The average value of each independent prediction value is taken, and the second predicted total carbon sequestration is output. The second predicted total carbon sequestration is then output as the prediction result.

[0107] Specifically, the prediction process of the random forest model embodies the collaborative characteristics of ensemble learning. The feature variables to be tested can be input into each independent decision tree. Each decision tree, based on its own node splitting rules and structure, independently outputs its corresponding carbon sequestration prediction value (i.e., independent prediction value). Due to differences in the training data and feature subsets of each decision tree, the independent prediction values ​​may fluctuate slightly. Subsequently, by taking the arithmetic mean of all independent prediction values, the prediction bias of a single decision tree can be effectively offset, reducing the prediction variance of the model. This mean is the second predicted total carbon sequestration, which integrates the prediction information of multiple decision trees. Compared with a single decision tree, it has higher prediction accuracy and stability, and is suitable for carbon sequestration prediction in complex ecological environments of mining areas.

[0108] If the carbon sequestration prediction model is a support vector machine model, then the feature variables to be predicted in the dataset to be predicted are mapped to a high-dimensional space through a kernel function to obtain the feature data to be predicted in the high-dimensional space. The feature data to be predicted is substituted into the optimal parameters of the maximum margin classification hyperplane to calculate the third predicted total carbon sequestration, and the third predicted total carbon sequestration is output as the prediction result.

[0109] Specifically, the prediction process of the support vector machine model needs to be consistent with the kernel function used in the training phase to ensure the continuity of feature mapping. For example, the feature variables to be tested can first be mapped to a high-dimensional space using a pre-defined kernel function, transforming them into feature data to be predicted in a high-dimensional space. This data resides in the same feature space as the feature data mapped during the training phase, satisfying the applicability conditions of the optimal hyperplane. The optimal parameters of the maximum margin classification hyperplane obtained through training are then substituted into the feature data to be predicted. and It can be done through formula Calculate the predicted value, where This indicates the third predicted total carbon sequestration. This represents the high-dimensional feature vector after mapping the feature variables to be measured. This represents the inner product of the optimal normal vector and the high-dimensional eigenvector. This represents the optimal bias term. The calculation process follows the classification and regression principle of support vector machines, and can achieve accurate prediction of the original nonlinear carbon sequestration data through linear partitioning in high-dimensional space. The output third predicted total carbon sequestration amount can then be directly used for the assessment of the carbon sequestration capacity of the mining area.

[0110] In one embodiment, such as Figure 2 As shown, based on the user's natural language query, target data is retrieved from the calculation and prediction results of the dynamic model, and an explanatory response is generated by combining feature importance weights, including:

[0111] S201: Based on a pre-defined terminology database for ecological carbon sequestration in mining areas, semantic parsing is performed on natural language queries to extract target land type, target year, and target carbon sequestration indicators;

[0112] S202: Based on the target land type and target year, retrieve matching target data from the dynamic model calculation results and prediction results to obtain the retrieval results;

[0113] S203: Determine the source of the target data in the search results. When the target data comes from the prediction results, based on the feature importance weight, sort the key influencing features according to their weight ratio and generate an explanatory statement containing the degree of influence of each key influencing feature on the target carbon sequestration index. If the target data comes from the calculation results of the kinetic model, generate an explanatory statement on the calculation basis of the corresponding target carbon sequestration index.

[0114] S204: Integrate target data, explanatory statements, or statements explaining the basis for calculations to generate and output an explanatory response.

[0115] Specifically, a pre-defined terminology database for ecological carbon sequestration in mining areas is the core support for semantic parsing. Its construction needs to cover core concepts and variations related to ecological carbon sequestration in mining areas, including land type terms (such as Cropland, Forest, Grassland, Wetland, etc.), carbon sequestration indicator terms (such as vegetation carbon storage, soil carbon storage, biodiversity index, total carbon sequestration, etc.), time-related terms (such as specific year, year range), and related logical terms (such as impact, calculation basis, predicted value, etc.). This terminology database can also be maintained through a combination of manual annotation and machine learning expansion to ensure coverage of commonly used industry expressions and users' possible natural language expression habits. The semantic parsing process can adopt a four-step workflow: word segmentation, terminology matching, entity recognition, and logical extraction. First, the natural language query is broken down into its smallest semantic units using a word segmentation algorithm. Then, precise and fuzzy matching (based on edit distance algorithms) is performed with the terms in the terminology database to filter out potential key entities. Subsequently, named entity recognition models, such as those based on CRF or BERT, are used to annotate the word segmentation results, which can determine the semantic units corresponding to land type, year, and carbon sequestration indicators. Finally, by extracting the relationship logic between the three through the rule engine, the core of the user query can be clarified. For example, to obtain data and related explanations of specific carbon sequestration indicators for a certain land type within a certain time period, the final output is a structured target land type, target year, and target carbon sequestration indicator.

[0116] Specifically, both the dynamic model calculation results and prediction results can be stored in a structured format. The data storage structure uses land type-year-carbon sequestration index-value as the core index. Land type corresponds one-to-one with land type terms in the terminology database. The year is stored in a four-digit format. The carbon sequestration index includes four core indicators: vegetation carbon storage, soil carbon storage, biodiversity index, and total carbon sequestration. The value field stores the corresponding calculation or prediction result. Therefore, based on the target land type and target year, the retrieval process from the dynamic model calculation and prediction results can be based on a multi-condition joint query logic. First, the land type field in the data is matched according to the target land type. Then, the year field is matched according to the target year (or time range). Finally, the value corresponding to the target carbon sequestration index is located. If the target year is a specific year, the year field is matched precisely. If it is a time range, the corresponding data for all years within that range is retrieved and sorted by year. If the year is not specified, the target data for the latest year is returned by default. During the search process, a data source identification field can be set to distinguish whether the data comes from the results of dynamic model calculations (identified as "mechanism calculation") or prediction results (identified as "model prediction"). This identification field is entered synchronously when the data is generated, which can ensure that the search results can clearly trace the data source and provide a basis for subsequent differentiated interpretation.

[0117] Specifically, when determining the source of target data in search results, this can be done through the source identifier field. For example, when the data comes from prediction results, the feature importance weights have already been generated during the model prediction stage, including weight values ​​corresponding to four types of feature variables: vegetation carbon storage, soil carbon storage, vegetation coverage, and soil organic carbon content. The weight values ​​can be sorted from largest to smallest, and the top 3-5 features with the highest weight percentages can be selected as key influencing features (with a total weight percentage of no less than 70%). Then, personalized content can be generated based on a preset explanatory statement template, such as "The main influencing feature of the target carbon sequestration indicator is vegetation coverage, with a weight percentage of 16%, among which the influence of vegetation coverage is the most significant." The weight percentage in the statement can be calculated using "feature weight value / total weight value of all features × 100%", ensuring the accuracy of the quantitative expression. When the data comes from the calculation results of a kinetic model, the explanatory statement needs to be generated based on the calculation formula and core parameters of that carbon sequestration indicator, and the calculation basis explanation logic differs for different carbon sequestration indicators. The explanation of vegetation carbon storage should link the total vegetation mortality rate formula to the carbon storage recursive formula, clarifying that its calculation depends on the previous year's carbon storage, the current year's vegetation growth rate, vegetation cover, and the mortality rates of various vegetation types. The explanation of soil carbon storage should clarify that its calculation is based on the product of soil organic carbon content and total vegetation area. The explanation of the biodiversity index should state that it is a linear combination of vegetation carbon storage and soil carbon storage, depending on their weighting coefficients. The explanation of total carbon sequestration should clarify that it is the arithmetic sum of vegetation carbon storage and soil carbon storage, ensuring that the explanation accurately reflects the core logic of the calculation mechanism and allows users to understand the basis for data generation.

[0118] Finally, the target data, explanatory statements, or calculation basis statements can be integrated. The integration process follows the principles of "data priority, supplementary explanation, and logical coherence." The retrieved target data can first be presented in a clear quantitative form, including the name of the carbon sequestration index, its corresponding value, and unit. Then, corresponding explanatory statements are attached according to the data source. If it is predictive data, key influencing features and their degree of influence are added; if it is mechanistic calculation data, the calculation basis is added. Finally, a Natural Language Generation (NLG) algorithm is used to merge the data and explanatory statements into a fluent response text, avoiding mechanical splicing. The response text format can be standardized as "The carbon sequestration index of a certain land type in a certain year is XX (unit). This data comes from XX (model prediction / mechanistic calculation), where: XX (explanatory statement content)", ensuring that users can quickly obtain the core data and understand the underlying logic. Output methods support direct text display, file export (such as PDF format), and result visualization charts (such as feature importance weight charts and carbon sequestration change trend charts) to adapt to the needs of different user scenarios, while ensuring a balance between the professionalism and readability of the response content, satisfying the in-depth needs of technical personnel and facilitating understanding by non-professionals.

[0119] Based on the same inventive concept, this application also provides a mining area ecological carbon sequestration assessment system for implementing the above-mentioned mining area ecological carbon sequestration assessment method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the mining area ecological carbon sequestration assessment system provided below can be found in the limitations of the mining area ecological carbon sequestration assessment method described above, and will not be repeated here.

[0120] In one exemplary embodiment, such as Figure 3 As shown, a mining area ecological carbon sequestration assessment system 300 is provided, comprising:

[0121] The data preprocessing and feature extraction module 301 is used to perform structured processing on the original data of the mining area, extract parameters such as vegetation growth rate, vegetation coverage, vegetation mortality rate, soil organic carbon content, and land area, and generate a preprocessed dataset. The preprocessed dataset is input into a preset ecological dynamics model, and the output includes the dynamics model calculation results containing vegetation carbon storage, soil carbon storage, biodiversity index, and total carbon sequestration. Based on the vegetation carbon storage, soil carbon storage, vegetation coverage, soil organic carbon content, and total carbon sequestration, a training dataset is generated.

[0122] The model calculation and training data generation module 302 is used to train a preset machine learning model based on the training dataset to obtain a carbon sequestration prediction model, obtain a dataset to be predicted with the same data structure as the preprocessed dataset, input the dataset to be predicted into the carbon sequestration prediction model for prediction, and output the prediction results and feature importance weights; wherein the preset machine learning model is any one of the decision tree model, random forest model or support vector machine model.

[0123] The carbon sequestration prediction and result interpretation module 303 is used to retrieve target data from the kinetic model calculation results and prediction results based on the user's natural language query, and generate an interpretive response by combining feature importance weights.

[0124] In one exemplary embodiment, the present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the mining area ecological carbon sequestration assessment method of this application. A multi-core processor is preferred to improve the system's parallel processing capability. The memory provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate large amounts of data and computational tasks.

[0125] In one exemplary embodiment, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of a mining area ecological carbon sequestration assessment method according to this application. The computer-readable storage medium may include: a read-only memory, a random access memory (RAM), a solid-state drive (SSD), or an optical disc, etc.

[0126] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for assessing ecological carbon sequestration in a mining area, characterized by, The method comprises: Structural processing of the original data of the mining area, extracting vegetation growth rate, vegetation coverage, vegetation mortality rate, soil organic carbon content and land area parameters to generate a pre-processing data set; inputting the pre-processing data set into a preset ecological dynamics model to output the dynamics model calculation results containing vegetation carbon storage, soil carbon storage, biodiversity index and total carbon sequestration, and generating a training data set based on the vegetation carbon storage, soil carbon storage, vegetation coverage, soil organic carbon content and total carbon sequestration; Based on the training data set, a carbon sequestration prediction model is obtained by training a preset machine learning model, a to-be-predicted data set consistent with the data structure of the pre-processing data set is obtained, the to-be-predicted data set is input into the carbon sequestration prediction model for prediction, and a prediction result and a feature importance weight are output; wherein the preset machine learning model is any one of a decision tree model, a random forest model or a support vector machine model; Based on the natural language query of the user, target data is retrieved from the dynamics model calculation results and the prediction results, and an explanatory response is generated in combination with the feature importance weight.

2. The method of claim 1, wherein, The inputting of the pre-processing data set into the preset ecological dynamics model to output the dynamics model calculation results containing vegetation carbon storage, soil carbon storage, biodiversity index and total carbon sequestration comprises: Based on the vegetation growth rate, vegetation coverage and vegetation mortality rate in the pre-processing data set, the vegetation carbon storage is calculated by the vegetation carbon storage recursive formula; the vegetation carbon storage recursive formula is: wherein, represents a total vegetation mortality rate in the t-th year, represents a total number of types of vegetation within a preset area, represents the vegetation mortality rate of the vegetation of the t-th type in the t-th year, represents the vegetation mortality rate of the vegetation of the t-th type in the t-th year, represents a vegetation carbon storage of the vegetation of the t-th type in the t-th year, represents the vegetation carbon storage of the vegetation of the t-th type in the t-th year, represents the vegetation carbon storage of the vegetation of the t-th type in the t-th year, represents the vegetation carbon storage of the vegetation of the t-th type in the t-1-th year, represents the vegetation growth rate of the vegetation of the t-th type in the t-th year, represents the vegetation coverage of the vegetation of the t-th type in the t-th year; Based on the soil organic carbon content and land area in the pre-processing data set, the soil carbon storage is calculated by the soil carbon storage formula; the soil carbon storage formula is: wherein, denotes the soil carbon stock of the year t, denotes the soil organic carbon content of the year t, denotes the total vegetation area of the predetermined area; Based on the vegetation carbon storage and the soil carbon storage, the biodiversity index is calculated by the biodiversity index formula; the biodiversity index formula is: wherein, represents the biodiversity index of the t-th year, represents a weight coefficient corresponding to the vegetation carbon storage, represents a weight coefficient corresponding to the soil carbon storage, and satisfies + = 1. The vegetation carbon storage and the soil carbon storage are added to obtain the total carbon sequestration; The vegetation carbon storage, the soil carbon storage, the biodiversity index and the total carbon sequestration are integrated to output the dynamics model calculation results.

3. The method of claim 1, wherein, The training of the preset machine learning model based on the training data set to obtain the carbon sequestration prediction model, the acquisition of the to-be-predicted data set consistent with the data structure of the pre-processing data set, the inputting of the to-be-predicted data set into the carbon sequestration prediction model for prediction, and the outputting of the prediction result and the feature importance weight comprise: The vegetation carbon storage, the soil carbon storage, the vegetation coverage and the soil organic carbon content are extracted from the training data set as feature variables, and the total carbon sequestration is extracted as a target variable; The feature variables and the target variable are input into the preset machine learning model, the preset machine learning model is trained, and the carbon sequestration prediction model is obtained; wherein the carbon sequestration prediction model is trained by the following steps: If the preset machine learning model is the decision tree model, information gain, gain ratio or Gini impurity of each feature variable is calculated, an optimal split feature is selected based on the calculation result, a decision tree node is constructed recursively in combination with the target variable, and after-pruning processing is performed after a preset stop condition is met, to obtain the carbon fixation amount prediction model; If the preset machine learning model is the random forest model, a training sample subset is selected from the training data set through sampling with replacement, a feature subset is randomly selected from the training sample subset, a plurality of independent decision trees are constructed in combination with the target variable, and each independent decision tree is trained and integrated through the feature subset, to obtain the carbon fixation amount prediction model; If the preset machine learning model is the support vector machine model, the linear separability of sample data composed of the feature variables and the target variable is judged; if the sample data is non-linearly separable, the feature variables are mapped to a high-dimensional space through a kernel function to obtain linearly separable feature data in the high-dimensional space; based on the linearly separable feature data and the target variable, a slack variable is introduced and a hyperparameter is configured to construct a constrained optimization problem with the minimum classification interval loss as the target; the optimal parameters of the maximum interval classification hyperplane are obtained by solving the constrained optimization problem; and the optimal parameters of the maximum interval classification hyperplane are used as core components to form the carbon fixation amount prediction model; The to-be-predicted data set is obtained, which includes current vegetation growth rate, current vegetation coverage, current vegetation mortality rate, current soil organic carbon content and current land area parameters corresponding to the preprocessed data set; The to-be-predicted data set is input into the carbon fixation amount prediction model, and the carbon fixation amount prediction model is operated to obtain the prediction result; Based on the carbon fixation amount prediction model, the influence weight of each feature variable on the total carbon fixation amount is calculated to generate the feature importance weight.

4. The method of claim 3, wherein, The target data is retrieved from the kinetic model calculation result and the prediction result based on the user's natural language query, and an explanatory response is generated in combination with the feature importance weight, including: Based on a preset mining area ecological carbon fixation terminology library, the natural language query is semantically analyzed to extract the target land type, target year and target carbon fixation index; Based on the target land type and the target year, matching target data is retrieved from the kinetic model calculation result and the prediction result to obtain a retrieval result; The source of the target data in the retrieval result is judged, and when the target data is derived from the prediction result, the key influence features are sorted by weight proportion based on the feature importance weight to generate an explanation sentence containing the influence degree of each key influence feature on the target carbon fixation index; if the target data is derived from the kinetic model calculation result, a calculation basis explanation sentence corresponding to the target carbon fixation index is generated; The target data, the explanation sentence or the calculation basis explanation sentence are integrated to generate and output the explanatory response.

5. The method of claim 3, wherein, The method comprises the following steps: If the carbon fixation amount prediction model is the decision tree model, the measured feature variables of the to-be-predicted data set are extracted, the node judgment rules of the decision tree model are used to recursively traverse the decision tree structure, a first predicted total carbon fixation amount is output, and the first predicted total carbon fixation amount is taken as the prediction result and output; If the carbon fixation amount prediction model is the random forest model, the measured feature variables of the to-be-predicted data set are input into each independent decision tree of the random forest model respectively, independent prediction values of each independent decision tree are obtained, a second predicted total carbon fixation amount is output by taking the mean of the independent prediction values, and the second predicted total carbon fixation amount is taken as the prediction result and output; If the carbon fixation amount prediction model is the support vector machine model, the measured feature variables of the to-be-predicted data set are mapped to the high-dimensional space through the kernel function to obtain to-be-predicted feature data in the high-dimensional space, the to-be-predicted feature data is substituted into the optimal parameters of the maximum interval classification hyperplane, a third predicted total carbon fixation amount is calculated, and the third predicted total carbon fixation amount is taken as the prediction result and output.

6. A mine site ecological carbon sequestration assessment system, characterized by, The system comprises: The data preprocessing and feature extraction module is configured to perform structural processing on the original data of the mining area, extract the vegetation growth rate, vegetation coverage, vegetation mortality rate, soil organic carbon content and land area parameters, and generate a preprocessed data set; input the preprocessed data set into a preset ecological dynamics model, output the calculation results of the dynamics model containing the vegetation carbon storage, soil carbon storage, biodiversity index and total carbon fixation amount, and generate a training data set based on the vegetation carbon storage, soil carbon storage, vegetation coverage, soil organic carbon content and total carbon fixation amount; The model calculation and training data generation module is configured to train a preset machine learning model based on the training data set to obtain a carbon fixation amount prediction model, obtain a to-be-predicted data set consistent with the data structure of the preprocessed data set, input the to-be-predicted data set into the carbon fixation amount prediction model for prediction, and output the prediction result and feature importance weight; wherein the preset machine learning model is any one of a decision tree model, a random forest model or a support vector machine model; The carbon fixation amount prediction and result interpretation module is configured to retrieve target data from the calculation results of the dynamics model and the prediction result based on a natural language query of a user, and generate an explanatory response in combination with the feature importance weight. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 5.

Images