A method for predicting the effects of mine ecological restoration

Abstract

This invention discloses a method for predicting the effectiveness of mine ecological restoration. The method includes constructing a current ecological state vector and a reference ecosystem database; matching the reference ecosystem to extract the target ecological state vector; establishing an inverse ecological trajectory equation to calculate the expected measure intensity, the expected ecological state vector, and the inverse trajectory deviation; adjusting the expected measure intensity based on the inverse trajectory deviation; determining an initial restoration plan and dividing the mine restoration area into different patches; determining the functional groups of the patches based on the reference ecosystem; constructing a natural invasion model; calculating the comprehensive natural resilience index and network-level natural invasion gain; optimizing the target to obtain supplementary restoration plans; constructing a hybrid prediction model; calculating the positive prediction deviation and the monitoring information gain of each patch; determining the final restoration plan; and acquiring real-time monitoring data to calculate the relative deviation and perform dynamic correction. This method can significantly improve prediction accuracy and plan feasibility, and is of great significance for improving restoration efficiency and reducing restoration costs. It provides a refined decision-making basis for national ecological civilization construction and sustainable mine development.

Description

Technical Field

[0001] This invention belongs to the field of mine ecological restoration technology, specifically relating to a method for predicting the effects of mine ecological restoration. Background Technology

[0002] With the deepening of ecological civilization construction, mine ecological restoration has become a core task for the transformation of resource-based cities and the ecological protection and restoration of national land space. In recent years, the concept of ecosystem reference has been introduced into mine restoration practice. By comparing the ecological parameters of stable zones in natural succession, quantifiable restoration targets are set for damaged mining areas. Mine ecological restoration can not only effectively restore the ecosystem function of mining areas and improve the quality of the regional ecological environment, but also directly affect the quality of life of residents around mining areas and the protection of biodiversity. With the continuous improvement of national requirements for ecological and environmental protection, establishing a precise, dynamic, and quantifiable prediction system for the effects of mine ecological restoration has become an urgent need for the industry.

[0003] Existing methods for predicting the effects of mine ecological restoration suffer from the following shortcomings: First, most methods rely on forward extrapolation based on predetermined restoration measures, resulting in weak guidance for optimizing the proposed solutions. Second, existing technologies often treat the restoration area as a homogeneous whole, neglecting the network-level recovery gain brought about by the self-organization process between internal patches. Finally, existing methods lack a dynamic correction mechanism based on real-time monitoring information gain, causing restoration measures to fail to respond to actual changes, making it difficult to guarantee long-term effectiveness and cost-effectiveness. Therefore, this invention proposes a method for predicting the effects of mine ecological restoration. It establishes a goal-oriented method for calculating the intensity of restoration measures through an inverse ecological trajectory equation, quantifies the synergistic gain between natural resilience and human intervention through patch network analysis, and achieves precise control of the restoration process through a hybrid prediction model and dynamic correction. This significantly improves prediction accuracy and feasibility, providing scientific decision support for mine ecological restoration. It is of great significance for promoting technological progress in the industry, improving restoration efficiency, and reducing restoration costs, while also providing technical support for national ecological civilization construction and sustainable mine development. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the effects of mine ecological restoration.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Acquire multi-source ecological monitoring data and historical restoration data, construct a current ecological state vector, construct a reference ecosystem database to match the reference ecosystem, and extract the target ecological state vector; Establish an inverse ecological trajectory equation, calculate the expected measure intensity, expected ecological state vector, and inverse trajectory deviation, and adjust the expected measure intensity based on the inverse trajectory deviation to determine the initial restoration plan; The mine restoration area is divided into different patches and the functional groups of each patch are determined. A natural invasion model is constructed, the comprehensive index of natural resilience and the network-level natural invasion gain are calculated, and target optimization is carried out to obtain supplementary restoration schemes. Determine a complete restoration plan and calculate intervention indicators. Combine the current ecological state vector and the comprehensive index of natural resilience to construct a hybrid prediction model, and output the state vector of the restoration process, the time-varying confidence matrix of each indicator and the prediction interval. Calculate the positive prediction bias and the monitoring information gain of each patch, determine the objective function of the remediation measures in combination with the remediation cost, determine the final remediation plan, obtain real-time monitoring data to calculate the relative bias, and perform dynamic correction. The ecological monitoring data includes space-based data, air-based data, and ground-based data; The historical repair data includes repair measures, measure intensity, repair cycle, and repair cost; The ecological state vector includes core parameters for ecological restoration; these core parameters include vegetation coverage, soil organic matter content, biodiversity index, soil retention capacity, water conservation capacity, comprehensive heavy metal pollution index, and geological disaster hazard elimination rate. The reverse ecological trajectory equation includes natural ecological process terms, intervention measures terms, environmental stress terms, and reverse goal-driven terms.

[0006] Furthermore, the matching reference ecosystem method includes: Historical restoration data from surrounding natural succession stable areas, near-natural restoration demonstration areas, and historically successful restoration areas were collected to construct a reference ecosystem database. Meteorological metadata vectors, geographic metadata vectors, soil metadata distribution, ecological structure data matrix, and disturbance history feature sets were extracted and standardized. Unique thermal coding was used to eliminate differences in dimensions and types. The disturbance intensity index of the current mine restoration area is calculated, and the similarity calculation priority is determined according to the disturbance intensity. The screening threshold is adaptively calibrated using the quantile method based on the reference ecosystem database. Calculate the Jaccard similarity coefficient for the set of historical features of interference, calculate the Mahalanobis distance for the meteorological metadata vector, calculate the Euclidean distance for the geographic metadata vector, and take the union to obtain the rigid constraint candidate set; Within the rigidly constrained candidate set, the Earth's movement distance is calculated for the distribution of soil metadata, and the structural similarity index is calculated for the ecological structure data matrix. The combined constraint conditions are then used to obtain a comprehensive weighted candidate set. The linear coefficient is dynamically adjusted based on the completeness of the data in the mine restoration area. The linear combination method is used to integrate entropy weight and analytic hierarchy process weight to calculate the comprehensive similarity index of each reference sample point in the comprehensive weighted candidate set. A preset number of reference sample points are selected in descending order as the reference ecosystem. The core parameters of ecological restoration, climate parameters, geographical parameters, soil parent material parameters and time parameters of each reference sample point are extracted and weighted averaged to obtain the target ecological state vector.

[0007] Furthermore, the method for determining the initial repair plan includes: Based on multi-source ecological monitoring data and historical restoration data, a differential equation for the forward ecological restoration process is constructed. By inverse time transformation and the introduction of an inverse target traction term, an inverse ecological trajectory equation is established. Set the inverse integration step size, take the target ecological state vector as the initial condition, use the fourth-order Runge-Kutta numerical integration to obtain the inverse trajectory and calculate the state change rate at each time step; By rearranging terms of the inverse ecological trajectory equation and substituting them into the expression for the intervention term, the expected intensity of the intervention is derived from the response function of the intervention. This response function is then mapped to the intensity of the intervention in the forward time. Substituting this into the differential equation of the forward ecological restoration process, the expected ecological state vector is obtained by integrating the equation with the current ecological state vector as the initial condition. The reverse trajectory deviation is calculated based on the expected ecological state vector and the target ecological state vector. Based on the reverse trajectory deviation and the preset threshold, the intervention index parameters are adjusted to determine the initial restoration plan. The initial restoration plan includes the intensity of the measures, the restoration period, the target state, the lower limit of the structure, the patch configuration, and the timing of the initial measures.

[0008] Furthermore, the method for constructing the natural invasion model includes: Based on the structural lower limit and initial measure sequence in the initial remediation plan, and combined with differences in topography, soil parent material and microclimate, the mine remediation area is divided into different patches and patch configuration vectors are extracted; the patch configuration vectors include patch area, distance to the nearest natural source, corridor connectivity, edge effect ratio, and microhabitat heterogeneity index; Based on the community composition, species niche characteristics, and interspecific functional complementarity of the reference ecosystem, the configuration vector of each patch is matched with the functional group suitability matrix of the reference ecosystem to determine the functional group configuration of each patch; the patch functional groups include nitrogen-fixing plant functional groups, deep-rooted soil-fixing plant functional groups, pollination network key species functional groups, surface crust functional groups, and litter decomposer functional groups. By applying the reverse principle of island biogeography, a dynamic equation for the number of naturally settled species in functional groups within a patch is established, and a natural invasion model is constructed. By introducing the time-lag coupling effect between InSAR surface deformation monitoring data and vegetation index, an InSAR deformation-vegetation index time-lag coupling correction factor is constructed and embedded into the natural invasion rate model to correct the prediction results.

[0009] Furthermore, the method for obtaining the supplementary repair solution includes: The steady-state number of naturally settled species is determined based on the steady-state conditions of the dynamic equation, and the comprehensive index of natural resilience for each patch is calculated. The network-level natural intrusion gain is determined based on the spatial distribution of the comprehensive index of natural resilience of each patch and the network effect of corridor connectivity. With the goal of maximizing the network-level natural intrusion gain, the optimization algorithm is used to solve the problem in combination with constraints. The supplementary repair scheme is determined based on the mapping relationship of measures. The supplementary repair scheme includes the addition of measures in time sequence, optimization of spatial configuration, adjustment of the natural-human intervention ratio, and a confidence matrix.

[0010] Furthermore, the method for constructing the hybrid prediction model includes: The initial repair plan and the supplementary repair plan are integrated to obtain a complete repair plan, and the intervention indicators of the complete repair plan are calculated. A hybrid prediction model was constructed, which extracted the initial ecological state vector, complete measures and plans, environmental factors and the whole process observation trajectory of completed mine restoration cases from historical restoration data to form a comprehensive set. The comprehensive set was randomly divided into a training set and a validation set in a ratio of 7:3. The hybrid prediction model includes an ecological process mechanism layer, an adverse target traction layer, a temporal attention residual correction layer, and an uncertainty quantification layer. It also uses a physical information neural network to embed physical laws and ecological principles into the loss function in the form of soft constraints. The ecological process mechanism layer introduces a comprehensive natural resilience index to correct the natural ecological process terms based on the differential equation of the positive ecological restoration process, and outputs mechanism prediction values; The reverse target traction layer takes the target ecological state index as input and calculates the target traction correction term for the corresponding reverse time. The temporal attention residual correction layer extracts historical observation residuals from historical repair data and combines them with the current state query vector to construct a residual correction term based on the temporal attention mechanism. The uncertainty quantification layer performs positive verification to obtain the repair process state vector, and then performs... Each Bootstrap resampling generates [a certain number of] samples. A Monte Carlo perturbation sample was obtained × For each set of prediction samples, calculate the prediction distribution information entropy and the time-varying confidence matrix of each indicator, and determine the prediction interval based on the Bootstrap sample quantiles.

[0011] Furthermore, the method for dynamic correction includes: The positive prediction deviation between the state vector of the restoration process and the target ecological state vector and the information gain of each patch are calculated. The measures are optimized with the objectives of maximizing the monitoring information gain, minimizing the positive prediction deviation and minimizing the restoration cost, so as to obtain the final restoration plan. During the final repair plan implementation, real-time monitoring data is continuously acquired and the relative deviation from the state vector of the repair process is calculated for dynamic correction. Specifically: Applying hard constraints of physical and ecological laws to the real-time ecological state vector for feasibility verification, the expression is as follows: ; in, As an indicator The theoretical maximum recovery rate is determined by the upper limit of the local potential annual productivity of natural vegetation, the geological timescale of soil development, and the physical limit of natural species dispersal. As an indicator The theoretical upper limit under local environmental conditions; Calculate the loss due to constraint violation ; When the relative deviation exceeds the threshold or the loss from the violation of constraints exceeds the preset value, it is determined that the ecological restoration trajectory deviates from the expectation or the physical constraints exceed the limit. Dynamic correction is then carried out. The shadow price of each target is calculated using the Lagrange multiplier method to identify overly aggressive targets. The corresponding components of the target ecological state vector are adjusted, and the additional measures required to break through the constraints and the additional time required to reach the target under the existing measures intensity are calculated. The target ecological state vector, intervention indicators, and restoration cycle are updated, and the measures restoration plan for the next cycle is determined.

[0012] The beneficial effects of this invention are: This invention provides a method for predicting the effects of mine ecological restoration. Compared with existing technologies, this invention has the following technical advantages: This invention introduces a reverse target traction term and establishes a reverse ecological trajectory equation. It solves the expected measure intensity and restoration path from the target state in reverse, which significantly improves the scientificity and accuracy of the initial restoration plan, achieves the unity of restoration goals and implementation paths, and avoids the impact of excessive intervention on the natural restoration process. This invention divides the restoration area into different patches and matches them with functional groups, constructs a natural invasion model, quantifies the comprehensive index of natural resilience of each patch and network-level natural invasion gain, and generates supplementary restoration schemes through target optimization. It makes full use of natural processes such as species dispersal and niche construction, significantly reduces the scale of investment in high-intensity artificial measures and long-term maintenance costs while ensuring restoration effects, and enhances the self-sustaining capacity of the restored ecosystem. This invention constructs a hybrid prediction model comprising an ecological process mechanism layer, a reverse target traction layer, a temporal attention residual correction layer, and an uncertainty quantification layer. It also embeds the soft constraints of the physical information neural network into the loss function to achieve adaptive fusion of mechanism-driven and data-driven approaches, and outputs a time-varying credibility matrix and prediction interval, thus solving the problems of model interpretability and credibility quantification. This invention calculates the monitoring information gain of each patch, constructs an objective function based on the repair cost, dynamically adjusts and generates the final repair plan, and uses real-time monitoring data to calculate the relative deviation for continuous correction. This allows the repair plan to be dynamically adjusted according to the actual recovery process, effectively avoiding over-repair or under-repair caused by static plans, maximizing the benefits of repair investment, and ensuring stable long-term repair results. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the steps of a method for predicting the effects of mine ecological restoration according to the present invention. Detailed Implementation

[0014] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0015] The present invention provides a method for predicting the effects of mine ecological restoration, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: Acquire multi-source ecological monitoring data and historical restoration data, construct a current ecological state vector, construct a reference ecosystem database to match the reference ecosystem, and extract the target ecological state vector; Establish an inverse ecological trajectory equation, calculate the expected measure intensity, expected ecological state vector, and inverse trajectory deviation, and adjust the expected measure intensity based on the inverse trajectory deviation to determine the initial restoration plan; The mine restoration area is divided into different patches and the functional groups of each patch are determined. A natural invasion model is constructed, the comprehensive index of natural resilience and the network-level natural invasion gain are calculated, and target optimization is carried out to obtain supplementary restoration schemes. Determine a complete restoration plan and calculate intervention indicators. Combine the current ecological state vector and the comprehensive natural resilience index to construct a hybrid prediction model, and output the state vector of the restoration process, the time-varying confidence matrix of each indicator, and the prediction interval. Calculate the positive prediction bias and the monitoring information gain of each patch, determine the objective function of the remediation measures in combination with the remediation cost, determine the final remediation plan, obtain real-time monitoring data to calculate the relative bias, and perform dynamic correction. The ecological monitoring data includes space-based data, air-based data, and ground-based data; The historical repair data includes repair measures, measure intensity, repair cycle, and repair cost; The ecological state vector includes core parameters for ecological restoration; these core parameters include vegetation coverage, soil organic matter content, biodiversity index, soil retention capacity, water conservation capacity, comprehensive heavy metal pollution index, and geological disaster hazard elimination rate. The reverse ecological trajectory equation includes natural ecological process terms, intervention measures terms, environmental stress terms, and reverse goal-driven terms. In actual assessments, taking an abandoned copper mine in Jiangnan as an example, the mining history was from 1978 to 2018, with a 6-year closure, and a total mining area of ​​500 hectares. 2 .

[0016] In this embodiment, the method for matching reference ecosystems includes: Historical restoration data from surrounding natural succession stable areas, near-natural restoration demonstration areas, and historically successful restoration areas were collected to construct a reference ecosystem database. Meteorological metadata vectors, geographic metadata vectors, soil metadata distribution, ecological structure data matrix, and disturbance history feature sets were extracted and standardized. Unique thermal coding was used to eliminate differences in dimensions and types. The disturbance intensity index of the current mine restoration area is calculated, and the similarity calculation priority is determined according to the disturbance intensity. The screening threshold is adaptively calibrated using the quantile method based on the reference ecosystem database. Calculate the Jaccard similarity coefficient for the set of historical features of interference, calculate the Mahalanobis distance for the meteorological metadata vector, calculate the Euclidean distance for the geographic metadata vector, and take the union to obtain the rigid constraint candidate set; Within the rigidly constrained candidate set, the Earth's movement distance is calculated for the distribution of soil metadata, and the structural similarity index is calculated for the ecological structure data matrix. The combined constraint conditions are then used to obtain a comprehensive weighted candidate set. The linear coefficient is dynamically adjusted based on the completeness of the data in the mine restoration area. The linear combination method is used to integrate entropy weight and analytic hierarchy process weight. The comprehensive similarity index of each reference sample point in the comprehensive weighted candidate set is calculated. A preset number of reference sample points are selected in descending order as the reference ecosystem. The core ecological restoration parameters, climate parameters, geographical parameters, soil parent material parameters and time parameters of each reference sample point are extracted and weighted averaged to obtain the target ecological state vector. In actual assessments, the expression for the disturbance intensity index of the current mine restoration area is: ; in, The interference intensity index. This represents the percentage of the area disturbed by mining. For the maximum mining depth, This represents the number of historical disturbances. The comprehensive index of heavy metal pollution. These are the weighting coefficients; Through an integrated space-air-ground monitoring network, imagery data from the Sentinel-2 satellite in April 2025, vegetation elevation data from ICESat-2, aerial survey data from UAV LiDAR in March 2025 (resolution 0.5m), monitoring data from 52 fixed quadrats, and soil profile sampling data were acquired. Historical remediation data of the mine's trial remediation blocks from 2019 to 2024 were also extracted. Construct the current ecological state vector ; Historical restoration data of three types of reference areas within a 250km radius were collected to construct a reference ecosystem database. The database includes 23 reference sampling points. Meteorological metadata vectors, geographic metadata vectors, soil parent material distribution, ecological structure data matrix, and disturbance history feature set were standardized. The measured data from the mine showed that the mining disturbance area accounted for 0.65%, the maximum mining depth was 150m, the number of historical disturbances was 12, and the comprehensive heavy metal pollution index was 8.5. The preset weight coefficients for each item were 0.30, 0.25, 0.20, and 0.25. The calculated mine disturbance intensity index was 5.974, which is greater than the high disturbance threshold of 5.0, indicating a high disturbance intensity level. A similarity analysis of the historical disturbance feature set was determined, with the priority of similarity calculation being the historical disturbance feature set, meteorological metadata, geographic metadata, soil metadata, and ecological structure data. The quantile method was used to adaptively calibrate the screening threshold, with the 75th quantile taken as the rigid constraint threshold. The screening thresholds were 0.40, 3.50, 2.80, 1.50, and 0.65, respectively. Dimensionally, Jaccard coefficients, Mahalanobis distances, and Euclidean distances were calculated for 23 reference points. Taking the key points' Jaccard coefficients, Mahalanobis distances, and Euclidean distances as examples, the coefficients were: Tongguanshan natural forest (plot number REF-03): 0.15, 2.10, 1.85; Dexing copper mine demonstration area (plot number REF-07): 0.52, 1.80, 2.20; Tongling restoration project area (plot number REF-12): 0.48, 2.40, 2.55; Anqing iron mine restoration area (plot number REF-15): 0.35, 3.20, 2.10; Chizhou copper mine restoration area (plot number REF-18): 0.42, 2.90, 2.75; and Huangshan natural forest (plot number REF-21): 0.12, 1.50, 1.20. This process yielded a candidate set of rigid constraints. ; Within the rigidly constrained candidate set, the soil and ecological structure dimensions are further calculated, with the Earth's distance of movement and structural similarity for each sampling point as follows: The values ​​are 1.20 and 0.72. The values ​​are 0.85 and 0.68. The values ​​are 1.26 and 0.63. The values ​​are 1.35 and 0.71. With values ​​of 1.53 and 0.77, a comprehensive weighted candidate set is obtained. ; The meteorological data completeness in the mine restoration area is 95%, geographical data is 100%, soil data is 85%, ecological structure data is 80%, and historical disturbance data is 90%. The overall data completeness is 0.88, and the linear coefficient is set as entropy weight. Hierarchical Analysis Weights The comprehensive weighting of each dimension was obtained as follows: historical interference 0.330, meteorological 0.180, geographical 0.150, soil 0.188, and ecological structure 0.152. The comprehensive similarity of the candidate set samples was calculated to be 0.883, 0.855, and 0.785 respectively, thus obtaining the target ecological state vector. .

[0017] In this embodiment, the method for determining the initial repair plan includes: Based on multi-source ecological monitoring data and historical restoration data, a differential equation for the forward ecological restoration process is constructed. By inverse time transformation and the introduction of an inverse target traction term, an inverse ecological trajectory equation is established. Set the inverse integration step size, take the target ecological state vector as the initial condition, use the fourth-order Runge-Kutta numerical integration to obtain the inverse trajectory and calculate the state change rate at each time step; By rearranging terms of the inverse ecological trajectory equation and substituting them into the expression for the intervention term, the expected intensity of the intervention is derived from the response function of the intervention. This response function is then mapped to the intensity of the intervention in the forward time. Substituting this into the differential equation of the forward ecological restoration process, the expected ecological state vector is obtained by integrating the equation with the current ecological state vector as the initial condition. The reverse trajectory deviation is calculated based on the expected ecological state vector and the target ecological state vector. Based on the reverse trajectory deviation and a preset threshold, the intervention index parameters are adjusted to determine the initial restoration plan. The initial restoration plan includes the intensity of the measures, the restoration period, the target state, the lower limit of the structure, the patch configuration, and the timing of the initial measures. In practical assessments, a differential equation for the positive ecological restoration process is constructed based on multi-source ecological monitoring data and historical restoration data. The expression is as follows: ; Where t is the forward time. For natural ecological processes, This is an ecological state vector, and the remaining vector is a 7-dimensional sub-vector composed of core parameters for ecological restoration. For ecological parameter set, For intervention measures, Let the measure strength vector be... For the set of measures effectiveness parameters, For environmental stress, An environmental factor vector. For the corresponding parameter set of coercion; Based on the differential equation of the forward ecological restoration process, and through reverse time transformation and the introduction of a reverse objective traction term, a reverse ecological trajectory equation is established, expressed as: ; in, As a reverse time variable, For the repair cycle, As a reverse target traction item, The target ecological state vector; The natural ecological process terms are determined by the ecological state vector and the set of ecological parameters, and are expressed using an extended Logistic growth model, as follows: ; in, , The intrinsic growth rate is a diagonal matrix. This is the upper limit vector of environmental theory, determined by the corresponding components of the target ecological state vector. Let be the extinction threshold vector for each core parameter, when When this term is negative, it simulates population decline below the extinction threshold. The interaction matrix between parameters determined for the ecological parameter set. Indicates the component index of the core parameter of ecological state. Represents the Hadamard product. Represents matrix multiplication; The intervention item is determined by the measure intensity vector and the measure effectiveness parameter set, and is expressed as follows: ; in, Let the measure strength vector be... For the set of measures effectiveness parameters, For the measure effectiveness matrix, For measure dimension indexing, Let the vector be the state-dependent performance function. , For performance sensitivity coefficient, The half-effect threshold; The environmental stress term is determined by the environmental factor vector and the stress response parameter set, and its expression is: ; in, The stress response parameter matrix, Indexed by environmental factors. Represented as an ecological state vector A diagonal matrix with elements on the main diagonal; the negative sign indicates the inhibitory effect of environmental stress on ecological parameters. The reverse target traction term is determined by the target ecological state vector and the ecological state vector, and its expression is: ; in, The target traction strength coefficient, This is the distance attenuation coefficient. It is the Euclidean norm. The modulation coefficients are for each stage; the exponential term controls the attenuation of the traction force amplitude with distance, and the vector term controls the direction pointing towards the target. The stage modulation coefficient adopts a piecewise function, and its expression is: ; in, , And satisfy These correspond to the rapid traction stage, the stable approach stage, and the fine adjustment stage, respectively. By rearranging terms in the inverse ecological trajectory equation and substituting them into the expression for the intervention term, we obtain the intervention response function. From this, we can deduce the required intervention intensity at each time point, expressed as: ; in, For the measure effectiveness matrix Moore-Penrose generalized inverse, This represents Hadamard division. for The environmental factor vector corresponding to the given time; The reverse trajectory deviation is calculated using the following expression: ; in, This is a deviation in the reverse trajectory. For the first The weighting coefficients of each core parameter for ecological restoration satisfy the following requirements. , Or add a minimum value Prevent division by zero; The restoration cycle is determined based on the recommended cycle for copper mines in the "Technical Specifications for Ecological Restoration of Mines". Year, preset If the step is 1, then the step size of the inverse integration is 1. In 2010, integration was performed to obtain the inverse trajectory nodes, with For example, The rate of change of state term is The components of the natural process term are: The components of the environmental stress term are: The components of the target traction item are: By reverse-engineering the measure strength vector, the expected measure strength is obtained. Each component corresponds to the intensity of vegetation reconstruction, soil improvement, slope treatment, and pollution remediation, respectively, and is mapped to the intensity of positive time-dependent measures. , to obtain the positive time-series measure intensity sequence; Using the current ecological state vector as initial conditions, the expected ecological state vector is obtained by substituting it into the differential equation of the forward ecological restoration process and integrating. The calculated reverse trajectory deviation is 11.87%, with weighting coefficients of 0.20, 0.15, 0.15, 0.15, 0.15, 0.10, and 0.10. Since 11.87% is greater than the preset threshold of 5%, the intervention indicator parameters need to be adjusted. The first adjustment involves increasing the target traction strength coefficient, decreasing the distance attenuation coefficient, adjusting the stage modulation coefficient, and optimizing the measure effectiveness parameters. The result is obtained by inverse integration. , Still not satisfied; The second adjustment extended the repair period, further optimized the effectiveness of the measures, adjusted the phase modulation coefficient, and recalculated the results. , Make fine adjustments; The third fine-tuning only increases the target traction strength coefficient, and the calculation yields... , The conditions are met; Based on the intervention indicators, the initial repair plan is determined to have a repair cycle of 22 years, with the intensity of measures being high-intensity for the first 8 years, medium-intensity for the middle 8 years, and low-intensity maintenance for the last 6 years. The target state is... The lower limit of the structure is that each parameter is not lower than the extinction threshold. The patch configuration is to divide the mine into 5 spatially heterogeneous patches. The initial measures sequence is as follows: the first 1-3 years are mainly for slope treatment and pollution remediation; the 4th-8 years are for soil improvement and vegetation reconstruction; the 9th-15th years are for vegetation reconstruction and natural assistance; and the 16th-22nd years are for key monitoring, maintenance and fine-tuning of measures.

[0018] In this embodiment, the method for constructing a natural invasion model includes: Based on the structural lower limit and initial measure sequence in the initial remediation plan, and combined with differences in topography, soil parent material and microclimate, the mine remediation area is divided into different patches and patch configuration vectors are extracted; the patch configuration vectors include patch area, distance to the nearest natural source, corridor connectivity, edge effect ratio, and microhabitat heterogeneity index; Based on the community composition, species niche characteristics, and interspecific functional complementarity of the reference ecosystem, the configuration vector of each patch is matched with the functional group suitability matrix of the reference ecosystem to determine the functional group configuration of each patch; the patch functional groups include nitrogen-fixing plant functional groups, deep-rooted soil-fixing plant functional groups, pollination network key species functional groups, surface crust functional groups, and litter decomposer functional groups. By applying the reverse principle of island biogeography, a dynamic equation for the number of naturally settled species in functional groups within a patch is established, and a natural invasion model is constructed. By introducing the time-lag coupling effect between InSAR surface deformation monitoring data and vegetation index, an InSAR deformation-vegetation index time-lag coupling correction factor is constructed and embedded into the natural invasion rate model to correct the prediction results. In practical assessments, the patch configuration vector expression is as follows: ; in, plaque The area of ​​the patch plaque The distance to the nearest natural source plaque The corridor connectivity, by patches The ecological corridor width, length, and continuity index between the nearest natural source are weighted and determined. plaque The edge effect ratio, plaque circumference, plaque Microhabitat heterogeneity index plaque The area percentage of the qth type of microhabitat. This represents the total number of microhabitat types. The expression for the natural invasion model is: ; in, For the first Functional groups in plaques The natural immigration rate For the first Functional groups in plaques Local extinction rate For function group type index; The expression for the natural immigration rate is: ; in, For the first The baseline migration rate for each functional group To reference the standard patch area in the ecosystem, For the first The area attenuation coefficient of each functional group, As the distance attenuation reference coefficient, For the first The connectivity regulation index of each functional group, For the first Environmental suitability sensitivity coefficient of each functional group plaque For the The environmental suitability index for each functional group is determined by a weighted average of soil organic matter content, soil moisture, light intensity, and matrix stability. For the first Microhabitat heterogeneity gain coefficients for each functional group This represents the maximum value of the microhabitat heterogeneity index. For the first Seasonal time-varying modulation functions for each functional group, For seasonal fluctuations, For seasonal cycles, The phase angle is given, and the constraints are satisfied. This ensures that the seasonal time-varying modulation function remains positive and avoids non-positive values ​​in the migration rate multiplier. The expression for the local extinction rate is: ; in, For the first The baseline extinction rate of each functional group For the first Habitat heterogeneity-area ratio moderating coefficients for each functional group For the first Environmental stress sensitivity coefficients of each functional group plaque The comprehensive environmental stress index is determined by a weighted average of the comprehensive heavy metal pollution index, soil erosion intensity, and geological hazard risk. plaque Inner The environmental carrying capacity of each functional group For the first The carrying capacity coefficient of each functional group; Based on the structural lower limit and initial measure sequence in the initial remediation plan, and combined with topography, soil parent material distribution maps, and microclimate differences, the mining area was divided into... The spatially heterogeneous patches are specifically designated as P1 (bottom platform of the open-pit mine), P2 (gentle slope zone on the west edge of the mine), P3 (north slope of the spoil heap), P4 (residual area of ​​industrial site), and P5 (tailings dam cover area). Construct patch configuration vectors, respectively , , , , Taking patch P1 as an example, the calculated edge effect ratio is 13.63. Due to the irregularity of the pit, it is corrected to 1.85. Microhabitat type Specifically, the areas of bare rock, gravelly soil, compacted platforms, and waterlogged depressions were 0.15, 0.30, 0.40, and 0.15, respectively. The calculated microhabitat heterogeneity index was 0.705. After a detailed on-site microhabitat survey and correction, the waterlogged depressions were seasonally dry, and the adjusted index was 0.62. Based on the community composition, species niche characteristics, and interspecific functional complementarity of the reference ecosystem, a functional group suitability matrix is ​​constructed to determine the configuration of functional groups for each patch. The specific configuration is based on the environmental suitability index determined by weighting soil organic matter, water, light, and matrix stability for each patch, combined with the matching of functional group composition ratios of the reference ecosystem. Taking patch P1 as an example, the nitrogen-fixing plant functional group is suitable, the deep-rooted soil-fixing plant functional group is relatively suitable, the pollination network key species functional group is unsuitable, the surface crust functional group is suitable, and the litter decomposer functional group is relatively suitable. Based on island biogeography theory and local species bank surveys, the area of ​​standard patches was determined. hm 2 Distance attenuation reference coefficient m, the maximum value of microhabitat heterogeneity Construct a natural invasion model; Taking patch P3 as an example, with soil organic matter of 1.8%, soil moisture of 18%, light intensity of 85%, and matrix stability of 0.65, the normalized environmental suitability index is calculated to be 0.72. In 2018, the natural immigration rate of functional group 1 within P3 was calculated to be 10.1 species / year. Assuming... The local extinction rate of functional group 1 within P3 was calculated to be 0.0023 species / year. The Sentinel-1 InSAR surface deformation monitoring data (2020-2025) and the NDVI time lag coupling effect were introduced. Taking patch P2 as an example, the deformation data corresponding to the time lag of 2023.5 in 2025 corresponds to the deformation rate anomaly of -8 mm / year in the P2 region, which is -0.8 after standardization.

[0019] In this embodiment, the method for obtaining the supplementary repair solution includes: The steady-state number of naturally settled species is determined based on the steady-state conditions of the dynamic equation, and the comprehensive index of natural resilience for each patch is calculated. Based on the spatial distribution of the comprehensive index of natural resilience of each patch and the network effect of corridor connectivity, the network-level natural intrusion gain is determined. With the goal of maximizing the network-level natural intrusion gain, an optimization algorithm is used to solve the problem in combination with constraints. Supplementary remediation schemes are determined based on the measure mapping relationship. The supplementary remediation schemes include increasing the timing of measures, optimizing spatial configuration, adjusting the ratio of natural to artificial intervention, and a confidence matrix. In actual assessment, plaques are calculated. The comprehensive index of natural resilience is expressed as: ; in, For the first The resilience weights of each functional group satisfy the following conditions: ; The expression for network-level natural intrusion gain is as follows: ; in, plaque The overall natural immigration rate, Functional group frequency / area ratio This is the InSAR-corrected coupling migration rate. plaque With plaque The distance between them, with a distance decay exponent of 2. plaque With plaque Connectivity between them For network coordination coefficient, plaque The area; The constraints are determined with the objective of maximizing network-level natural intrusion gain, and the expression is as follows: ; in, This represents the upper limit of the total remediable area of ​​the mine restoration zone. This is the minimum permissible distance between a patch and the nearest natural source, ensuring effective ecological isolation between patches and between patches and natural sources. For directed connectivity, if but for connectivity, When summing, only take , The construction cost per unit length of corridor The total budget for the corridor construction, This is the upper limit of the edge effect ratio, controlling the degree of patch fragmentation and reducing negative edge effects. Cost per unit area for constructing microhabitats Total budget for microhabitat construction; Taking functional group 1 of patch P3 as an example, the steady-state number of naturally resident species is calculated based on the steady-state conditions of the dynamic equation. ; The preset function group recovery weights are 0.25, 0.25, 0.15, 0.15, and 0.20. The calculations yield... , , , , ; Inter-patch connectivity is calculated based on the inter-patch distance matrix. , , , , , , , , , Calculate the InSAR-corrected composite natural immigration rate Seeds / year Seeds / year Seeds / year Seeds / year With a target number of species per year and a preset network coordination coefficient of 0.08, the network-level natural intrusion gain was calculated. Species / Year; The constraint parameters are preset as follows: hm 2 , m, Ten thousand yuan / km, Ten thousand yuan, , 10,000 yuan / hm 2 , The total area that can be restored in the mine restoration zone is [amount in yuan]. Verification constraints are applied. The actual minimum distance of the patch from the nearest natural source Corridor costs The actual maximum edge effect ratio Microhabitat cost All satisfy the constraints; The particle swarm optimization algorithm, with a population size of 50 and 200 iterations, is used to solve the problem. The optimization objective is to maximize the network-level natural intrusion gain, and the optimization result is the optimal patch area adjustment. Optimal corridor connectivity improvement Enhanced heterogeneity of optimal microhabitat After mapping according to the measures mapping relationship, the optimized measures are: adjusting the amount of terrain reshaping project, increasing the width of the ecological corridor from 5m to 8m, planting connected tree belts, and increasing micro-topographic undulations through pit-shaped land preparation and stone ridge construction. The supplementary remediation plan has been determined, specifically: the timing of measures has been increased, with a special measure for microhabitat substrate construction added in the first year, implemented before slope treatment; spatial allocation has been optimized, expanding the area of ​​patch P1 by 5 hm². 2 In addition, waterlogged microhabitats were added, and rocky ridge microhabitats were added to patch P4 to improve the microhabitat heterogeneity index. The corridor width between patches P2 and P3, and between P3 and P5, was increased to 8m. The ratio of natural to artificial intervention was adjusted. Patches P3 and P5 have high natural resilience, so the artificial planting density was reduced by 30% and the natural invasion waiting period was increased. Patches P1 have low natural resilience, so the proportion of artificial assisted planting was increased to 70%. A credibility matrix was output.

[0020] In this embodiment, the method for constructing a hybrid prediction model includes: The initial repair plan and the supplementary repair plan are integrated to obtain a complete repair plan, and the intervention indicators of the complete repair plan are calculated. A hybrid prediction model was constructed, which extracted the initial ecological state vector, complete measures and plans, environmental factors and the whole process observation trajectory of completed mine restoration cases from historical restoration data to form a comprehensive set. The comprehensive set was randomly divided into a training set and a validation set in a ratio of 7:3. The hybrid prediction model includes an ecological process mechanism layer, an adverse target traction layer, a temporal attention residual correction layer, and an uncertainty quantification layer. It also uses a physical information neural network to embed physical laws and ecological principles into the loss function in the form of soft constraints. The ecological process mechanism layer introduces a comprehensive natural resilience index to correct the natural ecological process terms based on the differential equation of the positive ecological restoration process, and outputs mechanism prediction values; The reverse target traction layer takes the target ecological state index as input and calculates the target traction correction term for the corresponding reverse time. The temporal attention residual correction layer extracts historical observation residuals from historical repair data and combines them with the current state query vector to construct a residual correction term based on the temporal attention mechanism. The uncertainty quantification layer performs positive verification to obtain the repair process state vector, and then performs... Each Bootstrap resampling generates [a certain number of] samples. A Monte Carlo perturbation sample was obtained × For each set of prediction samples, calculate the prediction distribution information entropy and the time-varying confidence matrix of each indicator, and determine the prediction interval based on the Bootstrap sample quantiles. In practical assessments, the expression for the ecological process mechanism layer is as follows: ; in, This is a mechanistic prediction value. This is a comprehensive index of regional natural resilience. The measure strength vector of the complete repair plan; The expression for the temporal attention residual correction layer is: ; in, This is the residual correction term. The number of historical residual key vectors and the attention weights. It is determined by the similarity between the query vector and the historical residual key vector. For the first A historical case at the same time The residual mode; The loss function expression is: ; in, For data fitting loss, The output state of the neural network. As a soft constraint of physical laws, the neural network is required to predict trajectories that satisfy ecological dynamics equations. The measure strength vector is the output of the neural network. The environmental factor vector is input to the neural network. Attracting soft constraints to the target To constrain violations and penalties, As an indicator The theoretical upper limit under local environmental conditions, These are the weighting coefficients for each loss term; The state vector expression for the repair process is: ; in, For the repair process state vector, This is a mechanistic prediction value. For target-driven correction terms, This is the residual correction term; By integrating the initial restoration plan and supplementary restoration plans, a complete restoration plan is obtained. The intervention index is calculated as the vegetation restoration intensity from year 1 to year 5. Soil amendment intensity Slope treatment intensity Pollution remediation intensity Years 6 to 12 , , , Years 13 to 18 , , , Years 19 to 22 , , , ; Thirty-five completed copper mine restoration cases were extracted from the historical restoration database and randomly divided in a 7:3 ratio, with 24 cases in the training set and 11 cases in the validation set. A hybrid prediction model was constructed and trained, with a temporal attention residual correction layer and historical residual key vectors. Uncertainty quantification layer Second-rate, Second-rate; by Taking this as an example, we calculate the comprehensive index of natural resilience for each patch. , , , , ,calculate Based on vegetation coverage For example, hour ,set up In a year, points can be obtained. ; Calculating the reverse time of the reverse target traction layer In 1990, according to Year, The year is judged to be in a stable approaching stage; the target-driven correction term is calculated. ; The temporal attention residual correction layer calculates the similarity between the query vector and 24 historical residual key vectors. The attention weights are normalized using Softmax. It is assumed that the query vector has the highest similarity to the historical case REF-H07 Dexing Copper Mine. (Residual pattern) The vegetation cover was generally underestimated by 0.035 by the mechanistic model in the 10th year, with weights... The weighted residual sum of the remaining 23 cases is 0.012, from which the residual correction term is calculated. ; Calculate the state vector of the repair process Exceeding the theoretical upper limit After being truncated by the constraint layer This triggers a constraint violation penalty, and after constraint truncation, the state vector of the complete repair process in year 10 is obtained. ; Bootstrap 500 times × Monte Carlo 1000 times to generate 500,000 sets of prediction samples and output the time-varying confidence matrix of each indicator; Taking the third observation in the validation set in the 10th year as an example, the observed values... Neural network output Calculate the data fitting loss Soft constraints of physical laws Target attracts soft constraints Constraints on penalties for violations Total loss .

[0021] In this embodiment, the method for dynamic correction includes: The positive prediction deviation between the state vector of the restoration process and the target ecological state vector and the information gain of each patch are calculated. The measures are optimized with the objectives of maximizing the monitoring information gain, minimizing the positive prediction deviation and minimizing the restoration cost, so as to obtain the final restoration plan. During the final repair plan implementation, real-time monitoring data is continuously acquired and the relative deviation from the state vector of the repair process is calculated for dynamic correction. Specifically: Applying hard constraints of physical and ecological laws to the real-time ecological state vector for feasibility verification, the expression is as follows: ; in, As an indicator The theoretical maximum recovery rate is determined by the upper limit of the local potential annual productivity of natural vegetation, the geological timescale of soil development, and the physical limit of natural species dispersal. As an indicator The theoretical upper limit under local environmental conditions; Calculate the loss due to constraint violation ; When the relative deviation is greater than the threshold or the loss from the violation of the constraint is greater than the preset value, it is determined that the ecological restoration trajectory deviates from the expectation or the physical constraint exceeds the limit. Dynamic correction is carried out. The shadow price of each target is calculated by the Lagrange multiplier method, overly aggressive targets are identified, the corresponding components of the target ecological state vector are adjusted, and the additional measures required to break through the constraints and the additional time required to reach the target under the existing measures intensity are calculated. The target ecological state vector, intervention indicators and restoration cycle are updated, and the measures restoration plan for the next cycle is determined. In practical assessments, the positive prediction deviation between the restoration process state vector and the target ecological state vector is calculated. , , , , , ; With the goals of maximizing information gain, minimizing positive prediction bias, and minimizing repair costs, information gain is the amount of uncertainty reduction after monitoring. It is assumed that the percentage reduction in post-monitoring entropy after deploying monitoring equipment in each patch is positively correlated with monitoring accuracy. The Pareto front optimal solution, obtained using the NSGA-II multi-objective optimization algorithm, corresponds to a final solution with a repair period of 22 years and a target state. The plaque configuration is based on plaque P1 as an example. The measures are scheduled as follows: in the first year, microhabitat base and slope treatment will be carried out; in the second to fourth years, soil improvement and pollution remediation will be carried out; in the fifth to tenth years, vegetation reconstruction will be carried out with a natural-artificial ratio of 4:6; in the eleventh to sixteenth years, natural-assisted restoration will be carried out with a ratio of 7:3; and in the seventeenth to twenty-second years, monitoring, maintenance and adaptive management will be carried out. The monitoring plan is to set up automatic monitoring stations for patches P1, P3 and P5, conduct quarterly manual surveys for patches P2 and P4, use drones for annual remote sensing of the entire mining area, and conduct a comprehensive baseline survey every five years. The total budgeted restoration cost is 48.5 million yuan, of which 32 million yuan is for engineering measures and 16.5 million yuan is for monitoring and maintenance. The final restoration plan has been implemented for 8 years, continuously acquiring real-time monitoring data to construct a real-time ecological state vector. Calculate the relative deviation , , , , , , The threshold for each indicator is 0.10, and the results are obtained through screening. Three items exceeded the limit; Hard constraints are imposed on the real-time ecological state vector, and the state in year 8 is verified. All indicators are lower than the theoretical upper limit. The loss due to constraint violation is calculated. Unrestrained violation; Although The physical constraints were not exceeded, but If the relative deviation is greater than the threshold of 0.10, it is determined that the ecological restoration trajectory has deviated from the expectation, and correction measures are taken. The shadow price of biodiversity bias was calculated using the Lagrange multiplier method. The index of 10,000 yuan / 0.01 increased, and the shadow price of soil retention deviation decreased. 10,000 yuan / t / (hm) 2 a) Shadow price of water conservation capacity deviation 10,000 yuan / m 3 / hm 2 ; In the target ecological state vector The biodiversity of patches P1 and P2 is only 0.35 and 0.38 respectively, and their shadow prices are the highest. Therefore, the corresponding targets are identified as overly aggressive targets. Will The value was adjusted from 0.817 to 0.78, a decrease of 4.5%. The corresponding local targets for patches P1 and P2 were adjusted to 0.70, while the values ​​for other targets remained unchanged. Under the current measures, it would take an additional 1.5 years to achieve the adjustment target. If the original cycle is maintained for 22 years, an additional investment of 3.9 million yuan is required, including 1.8 million yuan for biodiversity replanting, 1.2 million yuan for soil conservation engineering reinforcement, and 0.9 million yuan for water source conservation micro-habitat transformation. Update the target ecological state vector Update intervention indicators and increase biodiversity replanting intensity in years 9 to 11. The value increased from 0.55 to 0.70, indicating an increase in soil conservation projects. The value increased from 0.30 to 0.45, with an update and repair cycle of 23 years. The 9th to 23rd years will be remediated with updated goals, measures and cycles, and dynamic corrections will be continuously monitored and prepared for the next cycle.

[0022] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the effect of mine ecological restoration, characterized in that, Includes the following steps: S1. Obtain multi-source ecological monitoring data and historical restoration data, construct the current ecological state vector, construct a reference ecosystem database to match the reference ecosystem, and extract the target ecological state vector; S2. Establish the reverse ecological trajectory equation, calculate the expected measure intensity, expected ecological state vector and reverse trajectory deviation, and adjust the expected measure intensity according to the reverse trajectory deviation to determine the initial restoration plan; S3. Divide the mine restoration area into different patches and determine the functional groups of each patch. Construct a natural invasion model, calculate the comprehensive index of natural resilience and network-level natural invasion gain, and perform target optimization to obtain supplementary restoration schemes. S4. Determine the complete restoration plan and calculate the intervention indicators. Combine the current ecological state vector and the comprehensive index of natural resilience to construct a hybrid prediction model. Output the state vector of the restoration process, the time-varying confidence matrix of each indicator and the prediction interval. S5. Calculate the positive prediction bias and the monitoring information gain of each patch, determine the objective function of the remediation measures in combination with the remediation cost, determine the final remediation plan, obtain real-time monitoring data to calculate the relative bias, and perform dynamic correction. The ecological monitoring data includes space-based data, air-based data, and ground-based data; The historical repair data includes repair measures, measure intensity, repair cycle, and repair cost; The ecological state vector includes core parameters for ecological restoration; these core parameters include vegetation coverage, soil organic matter content, biodiversity index, soil retention capacity, water conservation capacity, comprehensive heavy metal pollution index, and geological disaster hazard elimination rate. The reverse ecological trajectory equation includes natural ecological process terms, intervention measures terms, environmental stress terms, and reverse goal-driven terms.

2. The method for predicting the effect of mine ecological restoration according to claim 1, characterized in that, The method for matching reference ecosystems includes: Historical restoration data from surrounding natural succession stable areas, near-natural restoration demonstration areas, and historically successful restoration areas were collected to construct a reference ecosystem database. Meteorological metadata vectors, geographic metadata vectors, soil metadata distribution, ecological structure data matrix, and disturbance history feature sets were extracted and standardized. Unique thermal coding was used to eliminate differences in dimensions and types. The disturbance intensity index of the current mine restoration area is calculated, and the similarity calculation priority is determined according to the disturbance intensity. The screening threshold is adaptively calibrated using the quantile method based on the reference ecosystem database. Calculate the Jaccard similarity coefficient for the set of historical features of interference, calculate the Mahalanobis distance for the meteorological metadata vector, calculate the Euclidean distance for the geographic metadata vector, and take the union to obtain the rigid constraint candidate set; Within the rigidly constrained candidate set, the Earth's movement distance is calculated for the distribution of soil metadata, and the structural similarity index is calculated for the ecological structure data matrix. The combined constraint conditions are then used to obtain a comprehensive weighted candidate set. The linear coefficient is dynamically adjusted based on the completeness of the data in the mine restoration area. The linear combination method is used to integrate entropy weight and analytic hierarchy process weight to calculate the comprehensive similarity index of each reference sample point in the comprehensive weighted candidate set. A preset number of reference sample points are selected in descending order as the reference ecosystem. The core parameters of ecological restoration, climate parameters, geographical parameters, soil parent material parameters and time parameters of each reference sample point are extracted and weighted averaged to obtain the target ecological state vector.

3. The method for predicting the effect of mine ecological restoration according to claim 1, characterized in that, The method for determining the initial repair plan includes: Based on multi-source ecological monitoring data and historical restoration data, a differential equation for the forward ecological restoration process is constructed. By inverse time transformation and the introduction of an inverse target traction term, an inverse ecological trajectory equation is established. Set the inverse integration step size, take the target ecological state vector as the initial condition, use the fourth-order Runge-Kutta numerical integration to obtain the inverse trajectory and calculate the state change rate at each time step; By rearranging terms of the inverse ecological trajectory equation and substituting them into the expression for the intervention term, the expected intensity of the intervention is derived from the response function of the intervention. This response function is then mapped to the intensity of the intervention in the forward time. Substituting this into the differential equation of the forward ecological restoration process, the expected ecological state vector is obtained by integrating the equation with the current ecological state vector as the initial condition. The reverse trajectory deviation is calculated based on the expected ecological state vector and the target ecological state vector. Based on the reverse trajectory deviation and the preset threshold, the intervention index parameters are adjusted to determine the initial restoration plan. The initial restoration plan includes the intensity of the measures, the restoration period, the target state, the lower limit of the structure, the patch configuration, and the timing of the initial measures.

4. The method for predicting the effect of mine ecological restoration according to claim 1, characterized in that, The method for constructing a natural invasion model includes: Based on the structural lower limit and initial measure sequence in the initial remediation plan, and combined with differences in topography, soil parent material and microclimate, the mine remediation area is divided into different patches and patch configuration vectors are extracted; the patch configuration vectors include patch area, distance to the nearest natural source, corridor connectivity, edge effect ratio, and microhabitat heterogeneity index; Based on the community composition, species niche characteristics, and interspecific functional complementarity of the reference ecosystem, the configuration vector of each patch is matched with the functional group suitability matrix of the reference ecosystem to determine the functional group configuration of each patch; the patch functional groups include nitrogen-fixing plant functional groups, deep-rooted soil-fixing plant functional groups, pollination network key species functional groups, surface crust functional groups, and litter decomposer functional groups. By applying the reverse principle of island biogeography, a dynamic equation for the number of naturally settled species in functional groups within a patch is established, and a natural invasion model is constructed. By introducing the time-lag coupling effect between InSAR surface deformation monitoring data and vegetation index, an InSAR deformation-vegetation index time-lag coupling correction factor is constructed and embedded into the natural invasion rate model to correct the prediction results.

5. The method for predicting the effect of mine ecological restoration according to claim 1, characterized in that, The method for obtaining the supplementary repair solution includes: The steady-state number of naturally settled species is determined based on the steady-state conditions of the dynamic equation, and the comprehensive index of natural resilience for each patch is calculated. The network-level natural intrusion gain is determined based on the spatial distribution of the comprehensive index of natural resilience of each patch and the network effect of corridor connectivity. With the goal of maximizing the network-level natural intrusion gain, the optimization algorithm is used to solve the problem in combination with constraints. The supplementary repair scheme is determined based on the mapping relationship of measures. The supplementary repair scheme includes the addition of measures in time sequence, optimization of spatial configuration, adjustment of the natural-human intervention ratio, and a confidence matrix.

6. The method for predicting the effect of mine ecological restoration according to claim 1, characterized in that, The method for constructing a hybrid prediction model includes: The initial repair plan and the supplementary repair plan are integrated to obtain a complete repair plan, and the intervention indicators of the complete repair plan are calculated. A hybrid prediction model was constructed, which extracted the initial ecological state vector, complete measures and plans, environmental factors and the whole process observation trajectory of completed mine restoration cases from historical restoration data to form a comprehensive set. The comprehensive set was randomly divided into a training set and a validation set in a ratio of 7:

3. The hybrid prediction model includes an ecological process mechanism layer, an adverse target traction layer, a temporal attention residual correction layer, and an uncertainty quantification layer. It also uses a physical information neural network to embed physical laws and ecological principles into the loss function in the form of soft constraints. The ecological process mechanism layer introduces a comprehensive natural resilience index to correct the natural ecological process terms based on the differential equation of the positive ecological restoration process, and outputs mechanism prediction values; The reverse target traction layer takes the target ecological state index as input and calculates the target traction correction term for the corresponding reverse time. The temporal attention residual correction layer extracts historical observation residuals from historical repair data and combines them with the current state query vector to construct a residual correction term based on the temporal attention mechanism. The uncertainty quantification layer performs positive verification to obtain the repair process state vector, and then performs... Each Bootstrap resampling generates [a certain number of] samples. A Monte Carlo perturbation sample was obtained × For each set of prediction samples, calculate the prediction distribution information entropy and the time-varying confidence matrix of each indicator, and determine the prediction interval based on the Bootstrap sample quantiles.

7. The method for predicting the effect of mine ecological restoration according to claim 1, characterized in that, The method for dynamic correction includes: The positive prediction deviation between the state vector of the restoration process and the target ecological state vector and the information gain of each patch are calculated. The measures are optimized with the objectives of maximizing the monitoring information gain, minimizing the positive prediction deviation and minimizing the restoration cost, so as to obtain the final restoration plan. During the final repair plan implementation, real-time monitoring data is continuously acquired and the relative deviation from the state vector of the repair process is calculated for dynamic correction. Specifically: Applying hard constraints of physical and ecological laws to the real-time ecological state vector for feasibility verification, the expression is as follows: ； in, As an indicator The theoretical maximum recovery rate is determined by the upper limit of the local potential annual productivity of natural vegetation, the geological timescale of soil development, and the physical limit of natural species dispersal. As an indicator The theoretical upper limit under local environmental conditions; Calculate the loss due to constraint violation ; When the relative deviation exceeds the threshold or the loss from the violation of constraints exceeds the preset value, it is determined that the ecological restoration trajectory deviates from the expectation or the physical constraints exceed the limit. Dynamic correction is then carried out. The shadow price of each target is calculated using the Lagrange multiplier method to identify overly aggressive targets. The corresponding components of the target ecological state vector are adjusted, and the additional measures required to break through the constraints and the additional time required to reach the target under the existing measures intensity are calculated. The target ecological state vector, intervention indicators, and restoration cycle are updated, and the measures restoration plan for the next cycle is determined.

Images