An ecological security early warning method based on multi-scale fusion and grey prediction

Abstract

This invention provides an ecological security early warning method based on multi-scale fusion and grey prediction, relating to the field of ecological security assessment and early warning technology. The invention divides the target area into multiple micro-grid units, obtains standardized feature data using multi-source monitoring data, and constructs a multi-dimensional feature data matrix through spatial alignment mapping. Based on this matrix, a historical time series is constructed, and objective entropy weighting is used to determine objective weights and perform time-series weighted fusion to construct an initial index sequence. Single-period feature vectors are input into a preset machine learning model for training, quantifying contribution weights to dynamically compensate the initial index and constructing a corrected index sequence. The corrected index sequence is input into a GM(1,1) grey prediction model for time-series evolution deduction, outputting a security early warning value. The security early warning values ​​are aggregated to construct an early warning distribution matrix, extracting regional early warning features and comparing them with preset regional security thresholds to determine the comprehensive ecological security early warning level, effectively improving the dynamic early warning capability for ecological security.

Description

Technical Field

[0001] This invention relates to the field of ecological security assessment and early warning technology, specifically to an ecological security early warning method based on multi-scale fusion and grey prediction. Background Technology

[0002] In recent years, with rapid urbanization, regional ecosystems have faced unprecedented pressure, making accurate ecological security early warning a core prerequisite for preventing environmental risks. Traditional ecological assessments often rely on single macroscopic remote sensing or static statistical data, providing only a rough superficial description of surface landscape patterns. However, true ecological degradation often originates from the hidden deterioration of microscopic natural units before spreading to the macroscopic space. Current technologies lack mechanisms for deep spatial coupling between microscopic environmental elements and macroscopic landscape characteristics, leading to complete failure to detect early microscopic damage and easily resulting in the underestimation of potential microscopic ecological damage.

[0003] Existing ecological security assessment methods typically collect basic statistical data of the target area and perform standardized processing to construct a routine risk assessment indicator system. However, the shortcomings of existing technologies lie in their over-reliance on static constants set by expert experience for simple linear superposition calculations during weight allocation. Furthermore, due to a lack of ability to capture the dynamic evolution of multidimensional characteristics, they often fail to objectively depict the true evolutionary patterns of ecosystems at different spatiotemporal points. Overall, these traditional security assessment methods mainly rely on the conventional approach of fixed weight settings and shallow data fusion to achieve ecological risk rating, resulting in a lack of dynamic adaptability in the final calculated security index. The assessment results often exhibit significant subjective experience bias and quantitative distortion.

[0004] In complex real-world ecological environments, the superposition of natural conditions and the intensity of human activities often leads to intricate interweaving and nonlinear distortions of multiple ecological elements. However, existing prediction techniques often directly input historical surface data into conventional models for simple time-series extrapolation, making it difficult to decouple the nonlinear mapping relationship between ecological phenomena and underlying driving mechanisms, resulting in ambiguous ecological significance of early warning results. Furthermore, localized random fluctuations or short-term data noise often exist in regional ecological environments. Existing shallow prediction methods are prone to misinterpreting such isolated local disturbances as macroscopic, continuous ecological degradation trends, leading to drastic jumps in prediction results. This makes it difficult to objectively and stably reflect the true evolutionary patterns and spatial early warning levels of regional ecosystems using conventional models.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide an ecological security early warning method based on multi-scale fusion and grey prediction to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: An ecological security early warning method based on multi-scale fusion and grey prediction, comprising the following steps: Step 1: Divide the target area into multiple micro-grid units, obtain multi-source monitoring data for each time node within the historical time window, perform hierarchical extraction and normalization on the multi-source monitoring data to obtain standardized feature data, and perform spatial alignment mapping based on micro-grid units to construct a multi-dimensional feature data matrix containing macro and micro dimensions. Step 2: Construct the historical time series corresponding to each micro-grid unit based on the multi-dimensional feature data matrix. Use the objective entropy weight method to extract the feature information of the standardized feature data in the historical time series to determine the objective weight. Then, perform time series weighted fusion based on the objective weight to construct the initial index sequence of each micro-grid unit. Step 3: For each time node in the historical time series, extract the corresponding standardized feature data to construct a single-period feature vector. Use the initial index of the corresponding time node in the initial index sequence as the supervision label. Input the single-period feature vector into the preset machine learning model for regression fitting training. Use the trained model to quantify the contribution weight of each feature to the change in ecological security, and use the contribution weight to dynamically compensate the initial index. Construct a modified index sequence according to the time series combination. Step 4: Input the correction index sequence of each micro-grid unit into the GM(1,1) grey prediction model and perform first-order accumulation generation and temporal evolution deduction, and output the safety warning value of the corresponding micro-grid unit within the preset prediction period. Step 5: Aggregate the safety warning values ​​of all micro-grid units to construct a warning distribution matrix for the target area, extract the regional warning features of the warning distribution matrix, and compare them with the preset regional safety threshold to determine the comprehensive ecological safety warning level of the target area.

[0008] Furthermore, the multi-source monitoring data includes natural environment data, activity load data, and landscape pattern data; The landscape pattern data specifically includes the Shannon diversity index, the Simpson diversity index, patch spread and patch cohesion; the natural environment data specifically includes vegetation cover and soil erosion; and the activity load data specifically includes industrial emissions and population density.

[0009] Furthermore, the specific steps for obtaining the standardized feature data are as follows: The global maximum and global minimum values ​​of each specific indicator in the landscape pattern data within the historical time window are obtained. The range standardization method is used to perform positive normalization on the original landscape data of each micro-grid unit to obtain the corresponding normalized Shannon diversity index, normalized Simpson diversity index, normalized patch spread and normalized patch assemblage. For the landscape pattern data of each micro-grid unit at each time node after positive normalization, the normalized Shannon diversity index and the normalized Simpson diversity index are multiplied by the value 0.2 to construct the weighted Shannon diversity index and the weighted Simpson diversity index, respectively. The normalized patch spread and the normalized patch assemblage are multiplied by the value 0.3 to construct the weighted patch spread and the weighted patch assemblage. The weighted Shannon diversity index, the weighted Simpson diversity index, the weighted patch spread and the weighted patch assemblage are summed and superimposed with a preset translation constant to generate the comprehensive landscape index of each micro-grid unit at the corresponding time node. Obtain the global maximum and global minimum values ​​of each specific indicator in the natural environment data and activity load data within a historical time window; For vegetation cover as a positive indicator, the original value corresponding to each time node is subtracted from its corresponding global minimum value, and the difference is divided by the range between the global maximum and global minimum values ​​corresponding to the vegetation cover. The quotient is then summed with a preset translation constant to generate the normalized value of the positive indicator at the corresponding time node. For soil erosion, industrial emissions and population density, which are negative indicators, the global maximum value of each negative indicator is subtracted from its original value at each time point. The difference is then divided by the range between the global maximum and the global minimum value. The quotient is then summed with a preset translation constant to generate the normalized value of each negative indicator at the corresponding time point. Standardized feature data are constructed by using the comprehensive landscape index, the normalized values ​​of positive indicators corresponding to positive indicators, and the normalized values ​​of negative indicators corresponding to various negative indicators.

[0010] Furthermore, the specific steps for constructing the multidimensional feature data matrix are as follows: The normalized values ​​of the positive indicators of vegetation cover and the negative indicators of soil erosion, which belong to the natural environment data, are combined in a multidimensional sequence according to time nodes to construct a normalized natural environment data matrix. Similarly, the normalized values ​​of the negative indicators of industrial emissions and population density, which belong to the activity load data, are combined in a multidimensional sequence according to time nodes to construct a normalized activity load data matrix. The values ​​of the landscape comprehensive index, the normalized natural environment data matrix, and the normalized activity load data matrix are all uniformly mapped to a specific non-zero interval. A multidimensional feature data matrix is ​​constructed using the spatial coordinates of each micro-grid unit and its corresponding time node as a spatiotemporal alignment reference. The multidimensional feature data matrix consists of two data layers. The landscape comprehensive index is used as a feature channel and is spliced ​​with the normalized natural environment data matrix to jointly construct the first data layer of the multidimensional feature data matrix to represent the macro-dimensional dimension. The second data layer of the multidimensional feature data matrix is ​​constructed using the normalized activity load data matrix to represent the micro-dimensional dimension.

[0011] Furthermore, the specific steps for determining the objective weights are as follows: In the multidimensional feature data matrix, the standardized feature data of each micro-grid unit at each time node within the historical time window, distributed in the macro and micro dimensions, are extracted along the time dimension. The standardized feature data is then arranged in rows and columns according to the corresponding time nodes and indicator categories to construct the historical time series sequence corresponding to each micro-grid unit. The standardized feature data in the historical time series are traversed, and the ratio of the value of each feature at each time node to the sum of its values ​​in the entire historical time window is calculated to obtain the feature weight at the corresponding time node. Calculate the product of the feature weight and its natural logarithm at each time point, and sum the product values ​​along the time dimension. Multiply the sum by a smoothing adjustment coefficient to calculate the feature information dispersion of each standardized feature data. The smoothing adjustment coefficient is the ratio between the negative one value and the natural logarithm of the total number of time points within the historical time window. The difference between the numerical value and the feature information dispersion of each standardized feature data is calculated to obtain the corresponding feature difference coefficient. The ratio of the feature difference coefficient of a single standardized feature data to the sum of the feature difference coefficients of all standardized feature data is then calculated to determine the objective weight of each standardized feature data.

[0012] Furthermore, the specific steps for constructing the initial exponential sequence are as follows: The standardized feature data of each item in the historical time series is multiplied by its corresponding objective weight at the corresponding time node to generate the corresponding weighted value. The micro-grid unit and the corresponding time node are used as the joint dimension. The weighted values ​​belonging to the macro-dimensional and micro-dimensional dimensions are reduced and summed to construct the initial index sequence corresponding to each micro-grid unit.

[0013] Furthermore, the preset machine learning model is constructed based on a combination of the XGBoost regression architecture and the SHAP analysis framework; The specific quantification steps for the contribution weight are as follows: For each micro-grid unit, the standardized feature data of each time node in its historical time series is extracted, and the single-period feature vector of the corresponding time node is constructed. The single-period feature vector of each time node is then rearranged with the time node as the sample dimension and the feature category contained in the single-period feature vector as the data feature dimension to construct the input feature matrix. Using the input feature matrix as input, the initial exponents in the initial exponent sequence of the corresponding micro-grid units are used as supervision labels and simultaneously input into the XGBoost regression architecture for regression fitting training to construct a nonlinear mapping relationship model. The SHAP analysis framework is invoked to calculate the marginal contribution of each standardized feature data in the input feature matrix to the output of the nonlinear mapping relationship model, and the marginal contribution is mapped to the quantified influence value of the corresponding standardized feature data. For each time point, the absolute value of the quantitative impact value corresponding to each standardized feature data is extracted, and the ratio of the absolute value of the quantitative impact value of a single standardized feature data to the sum of the absolute values ​​of the quantitative impact values ​​of all standardized feature data is calculated to determine the contribution weight of each standardized feature data at the current time point.

[0014] Furthermore, the specific construction steps of the modified exponential sequence are as follows: The contribution weights corresponding to each standardized feature data distributed in the micro dimension of the multidimensional feature data matrix are aggregated to generate the negative pressure coefficient at the current time point, and the contribution weights corresponding to each standardized feature data distributed in the macro dimension of the multidimensional feature data matrix are aggregated to generate the ecological resilience coefficient at the current time point. The difference between the ecological resilience coefficient and the negative pressure coefficient is calculated, and the opposite of its squared term is extracted to obtain the net ecological driving coefficient at the corresponding time node. The numerical value is summed with the net ecological driving coefficient to construct a dynamic compensation multiplier. The initial index at the corresponding time node is multiplied with the dynamic compensation multiplier to generate the correction index at the current time node. The correction indices at each time node are cascaded in chronological order to construct the correction index sequence for the corresponding micro-grid unit.

[0015] Furthermore, the specific steps for outputting the safety warning value are as follows: For each micro-grid unit, the values ​​in its correction exponent sequence are processed by first-order accumulation generation in chronological order to construct a smoothed accumulation generation sequence. The values ​​in the modified exponent sequence are extracted to form a data observation vector, and the mean data of adjacent terms in the accumulated generation sequence are extracted to construct a feature column. The feature column is then concatenated with a preset constant column to construct a background value matrix. Based on the GM(1,1) grey prediction model architecture, the least squares fitting algorithm is used to fit the data observation vector and the background value matrix to extract the development coefficient and grey action of the corresponding micro grid unit. Based on the development coefficient and the gray action quantity, a time-series evolution mapping relationship is constructed, and the time step corresponding to the preset prediction period is substituted into the time-series evolution mapping relationship to perform evolution deduction calculation, so as to obtain the cumulative prediction value corresponding to the preset prediction period. The accumulated predicted values ​​are subjected to first-order cumulative reduction and restoration processing to generate the corresponding micro-grid unit's safety warning value within the preset prediction period.

[0016] Furthermore, the spatial coordinates of all micro-grid units within the target area are extracted, and the safety warning values ​​of each micro-grid unit within the preset prediction period are spatially topologically stitched together according to their corresponding spatial coordinates to construct a warning distribution matrix. The maximum and minimum values ​​of the safety warning values ​​in the warning distribution matrix are obtained to determine the numerical range. The numerical range is then divided into four consecutive numerical intervals at equal intervals, and mapped sequentially in ascending order of numerical values ​​to low-risk threshold interval, medium-risk threshold interval, relatively high-risk threshold interval, and high-risk threshold interval. Traverse the warning distribution matrix and count the number of micro-grid units whose safety warning values ​​fall within the high-risk threshold range. Calculate the proportion of the number of micro-grid units to the total number of micro-grid units in the target area, and use this proportion as the regional warning feature of the target area. The regional early warning characteristics are compared with the preset regional safety threshold, and the comprehensive ecological security early warning level of the entire target area is determined based on the comparison results.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention extracts standardized feature data from multi-source monitoring data into hierarchical layers based on micro-grid units, constructs a multi-dimensional feature data matrix and historical time series containing both macro and micro dimensions, and uses the objective entropy weighting method to extract feature information and determine objective weights to construct an initial index sequence. Using the initial index as a supervision label, a pre-set machine learning model is used to quantify contribution weights and perform dynamic compensation to construct a corrected index sequence. This invention overcomes the shortcomings of traditional evaluation methods, which heavily rely on subjective static weighting and have a single scale. By closely combining deep feature fusion with a low-level dynamic compensation mechanism, it more accurately reflects the evolutionary state of ecological security. This invention also inputs the modified index sequence of each micro-grid unit into the GM(1,1) grey prediction model for first-order cumulative generation and temporal evolution deduction to output a safety warning value. It aggregates the safety warning values ​​of all micro-grid units to construct a warning distribution matrix, then extracts regional warning features and compares them with a preset regional safety threshold to determine the comprehensive ecological safety warning level. This mechanism effectively eliminates short-term data noise interference caused by local random fluctuations in the ecological environment, preventing the misjudgment of isolated surface numerical jumps as macroscopic continuous ecological degradation trends. Finally, by shielding the underlying environmental background noise and short-term data anomalies, it ensures that the output comprehensive warning level can stably and intuitively reflect the long-term ecological degradation risk of the target area. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is a multi-dimensional comparative evolution diagram of the initial index, the revised index, and the safety warning value. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0020] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0021] Example: Please see Figures 1-2 The present invention provides a technical solution: An ecological security early warning method based on multi-scale fusion and grey prediction, comprising the following steps: Step 1: Divide the target area into multiple micro-grid units, obtain multi-source monitoring data for each time node within the historical time window, perform hierarchical extraction and normalization on the multi-source monitoring data to obtain standardized feature data, and perform spatial alignment mapping based on the micro-grid units to construct a multi-dimensional feature data matrix containing macro and micro dimensions.

[0022] In this embodiment, the target area is first divided into spatially continuous raster evaluation units. To match the basic resolution of conventional multispectral remote sensing imagery and high-precision digital elevation models, and considering the regional spatial scale requirements, the physical size of the micro-raster unit is set to 30 meters × 30 meters. For irregular areas at the edge of the target area, spatial mask extraction is performed based on the true vector boundary of the target area. For edge raster units separated by boundary lines, the actual overlapping area ratio within the target area is calculated. If this area ratio is not less than 50%, it is retained as a valid micro-raster unit; otherwise, it is discarded. This division mechanism standardizes and discretizes irregular macroscopic natural geographic space, effectively avoiding the omission or redundancy of edge area data, thereby accurately capturing the local degradation characteristics of the micro-environment. Simultaneously, since a single data source cannot fully reflect the complex physical processes of ecosystem evolution, this embodiment extracts multi-source monitoring data from each of the past 10 complete calendar years. The multi-source monitoring data includes natural environment data, activity load data, and landscape pattern data. The landscape pattern data specifically includes the Shannon diversity index, Simpson diversity index, patch spread and patch cohesion, used to quantify the connectivity and fragmentation characteristics of macroscopic spatial structure; the natural environment data specifically includes vegetation cover and soil erosion, used to reflect the evolution of the physical properties of the ecological baseline; the activity load data specifically includes industrial emissions and population density, used to characterize the external stress pressures caused by human activities on the natural environment.

[0023] To eliminate dimensional differences between different data types and avoid excessive disparities in value ranges, this embodiment performs dimensionality reduction and weighting processing on the landscape pattern data. First, the global maximum and global minimum values ​​of the Shannon diversity index, Simpson diversity index, patch spread, and patch assemblage within the historical time window are obtained from the landscape pattern data. Then, the range standardization method is used to perform forward normalization on the above four original landscape data for each micro-grid unit, so that their values ​​are uniformly mapped to... The corresponding normalized Shannon diversity index, normalized Simpson diversity index, normalized patch spread, and normalized patch assemblage are obtained within a given time interval. Subsequently, for the forward-normalized landscape pattern data of each micro-grid unit at each time node, the normalized Shannon diversity index and normalized Simpson diversity index are multiplied by 0.2 to construct weighted Shannon diversity index and weighted Simpson diversity index, respectively. The normalized patch spread and normalized patch assemblage are multiplied by 0.3 to construct weighted patch spread and weighted patch assemblage, respectively. The weighted Shannon diversity index, weighted Simpson diversity index, weighted patch spread, and weighted patch assemblage are then summed and superimposed with a preset translation constant to generate a comprehensive landscape index for each micro-grid unit at the corresponding time node, thereby compressing multidimensional landscape features into a single comprehensive quantitative representation. To avoid absolute zero values ​​during feature extraction and subsequent logarithmic calculations, the preset translation constant is typically set to be an order of magnitude smaller than the smallest non-zero value of the original index. Therefore, in this embodiment, the preset translation constant is set to 0.001. The specific calculation formula for the comprehensive landscape index is as follows: In the formula, As a comprehensive landscape index, To normalize the Shannon diversity index, To normalize the Simpson diversity index, To normalize plaque spread, To normalize plaque binding degree, This is a preset translation constant.

[0024] The landscape comprehensive index characterizes the overall connectivity and complexity of the macroscopic spatial structure of an ecosystem under specific temporal and spatial conditions. Its value increases positively with the improvement of landscape diversity and spatial aggregation. This index employs a linear combination of differentiated weights and a constant superposition. Since connectivity has a stronger weakening effect on ecological resistance than species richness alone, assigning higher weights to spread and integration not only aligns with the physical aggregation patterns of ecological patches but also objectively quantifies changes in resilience caused by habitat fragmentation. A larger index indicates a rich variety of patch types within the region with close physical connections, and that ecological corridors are not fragmented; a smaller index indicates highly fragmented habitats and a fragile ecological physical framework. Therefore, this index can objectively quantify and characterize the comprehensive resilience level of regional ecological space.

[0025] Further, the global maximum and global minimum values ​​of each specific indicator in the natural environment data and activity load data within the historical time window are obtained. For vegetation cover, a positive indicator, the original value at each time node is subtracted from its corresponding global minimum value. The difference is then divided by the range between the global maximum and global minimum values ​​for that vegetation cover. The quotient is then summed with a preset translation constant to generate the normalized value for the positive indicator at the corresponding time node. For soil erosion, industrial emissions, and population density, negative indicators, the global maximum value for each negative indicator is subtracted from its original value at each time node. The difference is then divided by the range between the global maximum and global minimum values ​​for that negative indicator. The quotient is then summed with a preset translation constant to generate the normalized value for each negative indicator at the corresponding time node. A positive / negative differentiation normalization method is used, ensuring that larger processed values ​​generally represent a better (or worse) state. By overlaying a preset translation constant, all values ​​can be mapped to a specific non-zero interval. Standardized feature data is constructed using the comprehensive landscape index, the normalized values ​​of positive indicators, and the normalized values ​​of negative indicators. The specific calculation formula for the normalized values ​​of the positive indicators is as follows: The specific formula for calculating the normalized value of the negative index is as follows: In the formula, For specific micro-grid cells within the target area, in the first... At the first point in time, regarding the first Normalized values ​​of positive or negative indicators from multi-source monitoring data. For this specific micro-grid unit in the first The first time point The original values ​​of the indicators, For all micro-grid cells of the target region within the entire historical time window, with respect to the first... The original set of values ​​composed of the indicators, This is a preset translation constant.

[0026] The normalized values ​​obtained after processing each indicator represent the purely relative characteristic values ​​after stripping away the original physical dimensions and magnitudes. The magnitude of these values ​​reflects whether the individual ecological state of the micro-grid at the current time point is trending towards improvement or deterioration. Historical global maxima and minima jointly define the absolute physical carrying capacity boundary of specific ecological indicators during long-term evolution, while the original values ​​objectively reflect the actual state at the current time. For positive indicators such as vegetation cover, a larger original value indicates a better natural background, and the difference from the minimum value increases accordingly, thus driving a positive increase in the dependent variable. For negative indicators such as pollution emissions, negative normalization is performed by subtracting the maxima, so that a larger original value results in a smaller calculated normalized value. By uniformly mapping and stretching distinct engineering measurement units such as area proportion and emission quality to a specific dimensionless relative comparison interval, it is ensured that any increase in any value in the multidimensional feature matrix represents a reduction in real-world ecological risk.

[0027] Furthermore, to eliminate potential mathematical boundary vulnerabilities induced by specific spatial and temporal constraints when executing the aforementioned range standardization calculation logic, this embodiment introduces a division-by-zero anomaly protection mechanism. After obtaining the global maximum and global minimum values ​​of each specific indicator, the system performs a range validity check: if it is determined that the global maximum and global minimum values ​​of a certain indicator are equal throughout the entire historical time window, to avoid a systemic collapse due to a zero denominator in the difference division operation, this embodiment directly bypasses the conventional normalization formula, forcibly equating the normalized value of the single indicator at each time node to the preset translation constant. This further addresses algorithmic boundary vulnerabilities caused by extreme dead-zone data.

[0028] To address the misalignment and interference between data from different sources in terms of spatial resolution and temporal frequency, this embodiment uses the spatial coordinates of each micro-grid unit and its corresponding time node as a spatiotemporal alignment benchmark, ensuring that the same spatial coordinates have consistent feature dimensions on the same time profile. Normalized values ​​of positive vegetation cover and negative soil erosion are combined in a multidimensional sequence according to time nodes to construct a normalized natural environment data matrix; normalized values ​​of negative industrial emissions and population density are combined in a multidimensional sequence according to time nodes to construct a normalized activity load data matrix. This results in a multidimensional feature data matrix consisting of two data layers: the landscape comprehensive index is used as a feature channel and concatenated with the normalized natural environment data matrix to construct the first data layer of the multidimensional feature data matrix, representing the macroscopic dimension; the normalized activity load data matrix is ​​used to construct the second data layer of the multidimensional feature data matrix, representing the microscopic dimension.

[0029] Step 2: Construct the historical time series corresponding to each micro-grid unit based on the multi-dimensional feature data matrix. Use the objective entropy weight method to extract the feature information of the standardized feature data in the historical time series to determine the objective weight. Then, perform time-series weighted fusion based on the objective weight to construct the initial index sequence of each micro-grid unit.

[0030] In this embodiment, extracting the evolution patterns of feature data along the time axis is the foundation for dynamic prediction. Therefore, standardized feature data of each micro-grid unit, distributed across the macro and micro dimensions, is extracted from the multi-dimensional feature data matrix along the time dimension at each time node within the historical time window. This standardized feature data is then arranged in rows and columns according to the corresponding time nodes and indicator categories to construct the historical time series sequence corresponding to each micro-grid unit. The static spatial data, originally scattered across various time points, is then concatenated chronologically to characterize the dynamic evolution patterns of various ecological elements within each micro-grid over time.

[0031] In multi-dimensional feature evaluation, human experience bias can easily lead to distorted evaluation results. To ensure the pure objectivity of the quantification process, this embodiment uses the objective entropy weight method to determine the objective weights of each standardized feature data. The standardized feature data in the historical time series are traversed, and the ratio of its value at each time node to the sum of its values ​​over the entire historical time window is calculated to obtain the feature weight at the corresponding time node. The specific calculation formula is as follows: In the formula, For specific micro-grid cells within the target area, in the first... At the first point in time, regarding the first The proportion of standardized feature data in the features, For specific micro-grid cells within the target area, in the first... At the first point in time, regarding the first Normalized values ​​of positive or negative indicators from multi-source monitoring data. This represents the total number of time points included within the historical time window.

[0032] The characteristic weight is used to characterize the relative temporal contribution and evolutionary proportion of a single standardized indicator at a specific time point relative to the entire historical observation period. A larger value indicates that the state of the ecological element at that moment is more dramatic or prominent compared to its overall historical performance; a smaller value indicates that the element is currently in a relatively stable state. This characteristic weight is jointly determined by the standardized characteristic value at the current moment and the cumulative total value of the characteristic over the entire historical period. The current value acts as a direct driving force, and its increase positively impacts the weight; the cumulative total value over the entire historical period serves as an environmental background baseline, playing a role in global normalization constraints.

[0033] Calculate the product of the feature weight and its natural logarithm at each time point, and sum the products along the time dimension. Multiply this sum by a smoothing adjustment coefficient to calculate the feature information dispersion of each standardized feature data. The smoothing adjustment coefficient is the ratio of a value of negative one to the natural logarithm of the total number of time points within the historical time window. The specific formula for calculating the feature information dispersion is as follows: In the formula, For a specific micro-grid cell within the target area, regarding the first... The dispersion of feature information in standardized feature data. This represents the total number of time nodes contained within the historical time window. For specific micro-grid cells within the target area, in the first... At the first point in time, regarding the first The proportion of standardized feature data in a feature set.

[0034] The feature information dispersion is used to characterize the drastic evolution and information content of a specific ecological indicator within the entire historical time window. A smaller value indicates greater differences in the indicator's performance at different time points, implying a wealth of ecological evolution information; conversely, a larger value, approaching a uniform extreme value, indicates that the indicator changes smoothly and evenly over time, approximating a static background. This dispersion dependent variable is driven by the feature weight of each time point and the total number of time points. The feature weight reflects the relative contribution of a single point to the global picture. When a sudden ecological change occurs at a certain moment, leading to an abnormally high weight at that point, the negative nonlinear amplification of the logarithmic operator forces a sharp decrease in the final dispersion. The total number of time points serves as a smoothing adjustment benchmark, forcibly normalizing the result to eliminate the dimensional influence of the observation period. This logic reduces the weight of environmental factors with smaller fluctuations in the time series, while assigning higher information dispersion to elements with larger numerical fluctuations, highlighting their driving influence on changes in the regional ecological state.

[0035] The difference between the numerical value and the feature information dispersion of each standardized feature data is calculated to obtain the corresponding feature difference coefficient. The ratio of the feature difference coefficient of a single standardized feature data to the sum of the feature difference coefficients of all standardized feature data is then calculated to determine the objective weight of each standardized feature data. The greater the numerical difference and the more drastic the fluctuation of a certain indicator over time, the more effective information it contains in the current ecological evolution process, and the stronger its driving effect on the comprehensive evaluation results. Through logarithmic operations and dispersion extraction, static and ineffective background interference is effectively removed. The specific calculation formula for the objective weight is as follows: In the formula, For the first Objective weights of standardized feature data For a specific micro-grid cell within the target area, regarding the first... The dispersion of feature information in standardized feature data. This represents the total number of standardized feature data items belonging to both macro and micro dimensions.

[0036] The objective weight is used to characterize the true relative contribution of a single standardized ecological indicator in the comprehensive evaluation system. A larger value indicates a stronger decisive influence of that ecological element on the overall early warning result; a smaller value indicates a weaker influence. This objective weight dependent variable is jointly determined by the characteristic difference coefficient of the individual indicator and the sum of characteristic differences across the entire domain. The smaller the dispersion of the indicator's characteristic information, the more drastic its numerical fluctuations over time, and the larger the characteristic difference coefficient obtained by subtracting it from a constant; the sum of characteristic differences across the entire domain plays a global normalization constraint role. The objective weight determined by the above method can reduce the proportion of static background elements in the evaluation and correspondingly increase the proportion of dynamic disturbance elements, thereby accurately mapping the actual contribution of each indicator to the comprehensive early warning result.

[0037] The standardized feature data of each item in the historical time series is multiplied by its corresponding objective weight at each time node to generate a weighted value. Using the micro-grid unit and the corresponding time node as a joint dimension, the weighted values ​​belonging to the macro-dimension and micro-dimension are summed after dimensionality reduction to construct the initial index sequence for each micro-grid unit. The specific calculation formula for the initial index of the initial index sequence is as follows: In the formula, For specific micro-grid cells within the target area, in the first... The initial index at each time point This represents the total number of standardized feature data items categorized into macro and micro dimensions. For the first Objective weights of standardized feature data For specific micro-grid cells within the target area, in the first... At the first point in time, regarding the first Normalized values ​​of positive or negative indicators of multi-source monitoring data.

[0038] The initial index characterizes the initial baseline state of comprehensive ecological security for a specific micro-grid unit at a given historical time point. A higher value indicates a healthier and more stable overall ecological environment in that region at that time; a lower value directly reflects the severe ecological degradation or external stress risks facing the region. The dependent variable of this initial index is driven by the standardized characteristic data of each evaluation indicator and its corresponding objective weights. The standardized characteristic data objectively reflects the quality level of individual ecological elements, and an increase in their values ​​directly and positively drives the growth of the comprehensive index. The objective weights reflect the contribution ratio of each element; when the weight of a certain ecological element is high, its numerical change has a more significant impact on the comprehensive index result.

[0039] Step 3: For each time node in the historical time series, extract the corresponding standardized feature data to construct a single-period feature vector. Use the initial index of the corresponding time node in the initial index sequence as the supervision label. Input the single-period feature vector into the preset machine learning model for regression fitting training. Use the trained model to quantify the contribution weight of each feature to the change in ecological security, and use the contribution weight to dynamically compensate the initial index. Construct a corrected index sequence according to the time series combination.

[0040] In this embodiment, natural environmental factors and human activity loads exhibit a highly complex nonlinear interweaving in spatiotemporal evolution. Traditional static linear superposition methods easily mask localized sudden environmental stresses or hidden ecological degradation. Therefore, this embodiment extracts standardized feature data from historical time-series sequences of specific micro-grid units within the target area, constructing single-period feature vectors for each time node to establish an absolute comprehensive state snapshot of each spatial discrete grid on a single specific historical cross-section. Subsequently, the single-period feature vectors of each time node are rearranged using the time node as the sample dimension and the feature categories contained in the single-period feature vectors as the data feature dimension to construct an input feature matrix. An initial index is used as a supervision label representing the historical comprehensive environmental benchmark, and the input feature matrix is ​​used as input, synchronously input into an XGBoost regression architecture for regression fitting training. This decouples the nonlinear feedback effects of mutual constraints and promotion among multidimensional environmental factors, constructing a nonlinear mapping relationship model.

[0041] The SHAP analysis framework is used to calculate the marginal contribution of each standardized feature data in the input feature matrix to the output of the nonlinear mapping model, and this marginal contribution is mapped to the quantified impact value of the corresponding standardized feature data. For each time point, the absolute value of the quantified impact value corresponding to each standardized feature data is extracted, and the ratio of the absolute value of the quantified impact value of a single standardized feature data to the sum of the absolute values ​​of the quantified impact values ​​of all standardized feature data is calculated to determine the contribution weight of each standardized feature data at the current time point, thereby quantifying the actual driving weight of each environmental indicator on ecosystem evolution at the corresponding time point.

[0042] The contribution weights corresponding to each standardized feature data distributed in the micro-dimension of the multi-dimensional feature data matrix are aggregated to generate the negative pressure coefficient at the current time point. The specific calculation formula is as follows: In the formula, For specific micro-grid cells within the target area, in the first... The negative pressure coefficient at each time point This represents the total number of standardized feature data items in the micro-dimensional dimension. For the first Micro-standardized feature data in the first Contribution weights at each time point For the first Micro-standardized feature data in the first The normalized values ​​of negative indicators at each time point are derived from the multidimensional feature data matrix.

[0043] The negative pressure coefficient characterizes the combined state of a micro-level spatial unit under environmental stress and human interference at a specific historical juncture. Because the underlying data undergoes inverse normalization, a larger value indicates a weaker actual external stress on the current environment, suggesting the region is in a low-pressure, safe state; conversely, a smaller value indicates the system is experiencing strong external pollution or destructive interference, with a sharp increase in ecological risk. This dependent variable is driven by the aggregation of normalized values ​​of various micro-level negative characteristics and dynamic contribution weights. The normalized values ​​record the physical control level of individual stress factors, and their decay pulls down the overall coefficient; the dynamic contribution weights can adaptively adjust according to the dominance of the current stress source, making changes in key stress factors more significantly reflected in the negative pressure coefficient.

[0044] The contribution weights corresponding to each standardized feature data distributed in the macroscopic dimension of the multidimensional feature data matrix are aggregated to generate the ecological resilience coefficient at the current time point. The specific calculation formula is as follows: In the formula, For specific micro-grid cells within the target area, in the first... Ecological resilience coefficient at each time point This represents the total number of standardized feature data items in the macro dimension. For the first The macro-standardized characteristic data in the first Contribution weights at each time point For the first The macro-standardized characteristic data in the first The normalized values ​​of the positive indicators at each time point are derived from the multidimensional feature data matrix.

[0045] The ecological resilience coefficient characterizes the comprehensive buffering capacity of micro-spatial units to resist external disturbances and maintain system stability at specific historical junctures. A higher value indicates stronger self-repair and resilience of the current ecosystem, signifying a healthy region; conversely, a lower value indicates a fragile ecological structure, with the physical framework for resisting external stresses easily collapsing. This dependent variable is driven by the aggregation of normalized values ​​of macroscopic positive characteristics and dynamic contribution weights. Normalized values ​​objectively record the actual levels of individual elements such as vegetation cover or landscape connectivity; their increase directly boosts overall resilience. Dynamic contribution weights act as non-linear regulators; when a macroscopic element plays a dominant supporting role in maintaining the current balance, its high weight significantly amplifies even minor improvements in that factor, leading to a surge in the dominant value.

[0046] The difference between the ecological resilience coefficient and the negative pressure coefficient is calculated, and the negative of its squared term is extracted to obtain the net ecological driving coefficient at the corresponding time point. The net ecological driving coefficient is summed with the initial value to construct a dynamic compensation multiplier. The initial index at the corresponding time point is then multiplied by the dynamic compensation multiplier to generate the correction index at the current time point. The correction indices at each time point are then concatenated in chronological order to construct a correction index sequence for the corresponding micro-grid unit. The specific calculation formula for the correction index is as follows: In the formula, For specific micro-grid cells within the target area, in the first... The correction index at each time point For specific micro-grid cells within the target area, in the first... The initial index at each time point It represents the difference between the ecological resilience coefficient and the negative pressure coefficient.

[0047] The correction index characterizes the true ecological security baseline state of micro-spatial units after experiencing the nonlinear counterbalancing of external pressure and internal resilience. A higher value indicates a favorable historical baseline and minimal current environmental disturbance, suggesting the system is in a highly stable and healthy state. Conversely, a lower value reveals intense ecological tug-of-war and extreme structural fragility. This dependent variable is jointly driven by the initial environmental baseline and the dynamic compensation multiplier. The initial environmental baseline defines the fundamental framework formed by the region's long-term evolution, determining the physical upper limit of the correction result. The dynamic compensation multiplier, constructed based on the net difference between ecological resilience and negative pressure, dynamically corrects the initial index. When the system tilts significantly towards external stress or internal resilience, the compensation multiplier decreases, thus characterizing a decline in the ecosystem's stability against fluctuations.

[0048] Furthermore, since the mathematical modeling premise of the GM(1,1) grey prediction model strictly requires that the input sequence must be a non-negative sequence, in order to prevent the time series extrapolation algorithm from failing due to data exceeding the limits in extreme ecological environment change scenarios, this embodiment calculates the net ecological driving coefficient and dynamic compensation multiplier at the corresponding time node. If the absolute difference is caused by an extreme imbalance between ecological resilience and negative pressure, the calculation will be performed accordingly. This leads to the dynamic compensation multiplier calculated in conventional methods. In this embodiment, the dynamic compensation multiplier at that time point is forcibly reset to a preset small positive protection value (set to 0.01 in this embodiment). Through the above-mentioned adaptive threshold hard truncation mechanism, it can be ensured that the output modified exponent sequence is always a strictly non-negative sequence, avoiding the risk of program underlying logic collapse.

[0049] Step 4: Input the correction index sequence of each micro-grid unit into the GM(1,1) grey prediction model and perform first-order accumulation generation and temporal evolution deduction to output the safety warning value of the corresponding micro-grid unit within the preset prediction period.

[0050] In this embodiment, the evolution of the natural ecological environment is influenced by multiple complex factors, and actual monitoring data often exhibits characteristics of limited sample size and random fluctuation noise. Therefore, this embodiment performs first-order cumulative generation processing on the values ​​in the modified exponential sequence of each micro-grid unit in chronological order to construct a smoothed cumulative generation sequence. The calculation formula is as follows: In the formula, This is a set of data observations extracted from the corrected index sequences of the corresponding micro-grid cells. This represents the total number of historical time points contained in the data observation set. For specific micro-grid cells within the target area, in the first... The cumulative generated value under each evolution step size For this specific micro-grid unit in the first The correction index at each time point This is used for the cumulative generation and time step index for temporal evolution deduction.

[0051] The accumulated value is used to characterize the cumulative ecological evolution state of a specific microgrid after smoothing. A larger value indicates a higher total amount of ecological security baseline accumulated within a set time span. This dependent variable is driven by the correction index at each historical time point and the time step. This calculation logic closely aligns with the evolutionary nature of complex ecological environments: real ecological monitoring data often involves a large amount of high-frequency random oscillations caused by short-term extreme weather or sudden human interference, making direct extrapolation prone to trend misjudgment.

[0052] The numerical values ​​in the modified exponent sequence are extracted to form a data observation vector, and the mean data of adjacent terms in the accumulated generation sequence are extracted to construct a feature column. This feature column is then concatenated with a preset constant column where all elements are always 1 to construct a background value matrix. The calculation formula used is as follows: In the formula, A background value matrix constructed for specific micro-grid cells within the target region. The data observation vector constructed for the corresponding micro-grid unit. This represents the total number of historical time points contained in the data observation set. For this specific micro-grid unit in the first The original correction index at each time point For this specific micro-grid unit in the first The cumulative generated value obtained under each evolution step size.

[0053] The background value matrix and data observation vector are used to characterize the numerical mapping boundary between the intrinsic evolutionary dynamics of the micro-ecological grid and external environmental stresses in the discrete time domain. The numerical magnitude of the background value matrix defines the historical damping pattern in the ecological evolution process, while the numerical magnitude of the data observation vector directly reflects the actual environmental increment of the ecological state in a single period. The dependent variables of this matrix and vector are driven by the cumulative generated values ​​of adjacent time nodes and the original correction index. The adjacent cumulative generated values ​​are averaged using the trapezoidal integral rule to approximate the true continuous environmental evolution background; while the original correction index, as the actual driving increment, determines the vertical numerical composition of the observation vector with each fluctuation of its input.

[0054] Based on the GM(1,1) grey prediction model architecture, a least-squares fitting algorithm is used to fit the data observation vector and the background value matrix, extracting the development coefficient and grey effect of the corresponding micro-grid units. Extracting the mean of adjacent terms to construct the background value can further weaken the jump interference caused by extreme anomalies, while the development coefficient reflects the inertial development rate of the ecological state's own evolution, and the grey effect characterizes the comprehensive driving force of complex external environmental conditions on the predicted trend. The calculation formulas are as follows: In the formula, To accumulate the continuous derivatives of the generated sequence with respect to time, To correspond to the development coefficient of the micro-grid unit, For the continuous variable representation of the accumulated generated sequence, To correspond to the gray action amount of the micro-grid unit, The parameter vector is composed of the development coefficient and the gray action quantity. The background value matrix, This is the transpose of the background value matrix. This is the data observation vector.

[0055] The parameter vector encompassing the development coefficient and the gray action quantity is used to characterize the quantitative decomposition of the intrinsic evolutionary dynamics and external environmental stresses of the micro-ecological grid. The positive and negative attributes of the development coefficient are inversely correlated with the evolutionary trend; the larger the absolute value, the more severe the momentum of inertial decline or recovery. The magnitude of the gray action quantity is directly positively correlated with the intensity of external environmental stress. This dependent variable is driven by the background value matrix and the data observation vector through matrix transpose and inversion operations. The background matrix delineates the historical boundary and damping pattern of the ecological evolution process, while the data observation vector provides the actual physical increments of the evolution process. Both are used to find the parameter combination with the minimum global fitting error through the least squares principle, and the discrete observation increments are forcibly transformed into continuous feature parameters using algebraic mapping relationships.

[0056] Based on the development coefficient and the grey effect, a time-series evolution mapping relationship is constructed. The time step corresponding to the preset prediction period is then substituted into the time-series evolution mapping relationship for evolution extrapolation calculation, yielding the cumulative prediction value corresponding to the preset prediction period. The GM(1,1) grey prediction model has extremely high computational accuracy for small samples and short-to-medium-term evolution trends. In this embodiment, the preset prediction period is set to 5 years based on the current ecological environment protection planning cycle of the target area, further ensuring the objective confidence of the extrapolation data and avoiding prediction distortion caused by excessive extrapolation to the future. Finally, the cumulative prediction value is subjected to first-order cumulative reduction and restoration processing to generate the corresponding safety warning value for the micro-grid unit within the preset prediction period. The calculation formula is as follows: In the formula, For specific micro-grid cells within the target area, in the first... The cumulative predicted value obtained at each evolution step size This is the correction exponent for that specific micro-grid cell at the first historical time point. To correspond to the development coefficient of the micro-grid unit, To correspond to the gray action amount of the micro-grid unit, To accumulate and generate time step indexes for temporal evolution deduction, For specific micro-grid cells within the target area, in the first... Safety warning values ​​under each evolution step size.

[0057] The cumulative predicted value is used to characterize the smooth cumulative predicted state of a specific micro-ecological grid over a future evolutionary step. Its magnitude signifies the predicted cumulative height of the total ecological state at a specific future development stage. This dependent variable is driven by the initial state anchor point, development coefficient, grey effect, and time step. The cumulative predicted value is jointly determined by the initial state, development coefficient, grey effect, and time step. The development coefficient determines the evolutionary trend of the prediction curve; a negative value represents a continuous growth trend with time step, while a positive value represents a decaying trend of the curve converging towards an extreme value. Conversely, if the development coefficient is positive, the natural exponential term decays sharply with time step, forcing the prediction curve to converge towards an extreme constant.

[0058] The safety warning value is used to characterize the absolute state of future ecological security after being restored to its original physical dimensions. Its magnitude directly maps to the ecological degradation risk or degree of natural recovery that the microgrid will face within a specific planning period. A larger value indicates a healthier future environment, while a smaller value foreshadows an imminent risk of ecological collapse. The cumulative predicted value at the previous moment represents all past historical accumulation, while the cumulative predicted value at the next moment includes state changes within the newly added evolutionary step. By subtracting the value at the previous moment from the value at the next moment, the difference in slope and the jump magnitude determine the absolute size of the predicted increment.

[0059] Step 5: Aggregate the safety warning values ​​of all micro-grid units to construct a warning distribution matrix for the target area, extract the regional warning features of the warning distribution matrix, and compare them with the preset regional safety threshold to determine the comprehensive ecological safety warning level of the target area.

[0060] In this embodiment, the prediction results of a single micro-grid unit can only reflect the evolution of local spatial points. Therefore, this embodiment extracts the spatial coordinates of all micro-grid units within the target area, and performs spatial topological stitching of the safety warning values ​​of each micro-grid unit within a preset prediction period according to the corresponding spatial coordinates to construct a warning distribution matrix. The discrete one-dimensional prediction values ​​are then remapped back to the real two-dimensional geographic space, restoring the geometric adjacency relationship between grids, thereby intuitively displaying the contiguous expansion trajectory, edge penetration direction, and spatial clustering pattern of ecological risks.

[0061] To determine the numerical range, the maximum and minimum values ​​of the safety warning values ​​in the warning distribution matrix are obtained. In this step, when the maximum and minimum values ​​coincide in the extreme case, it indicates that the ecological evolution of all micro-grids in the target area is in an absolutely homogeneous state. At this time, the interval equal division step is skipped directly, the entire area is regarded as a single threshold interval, and the indicator is extracted for macro-level classification. Under normal numerical range, the numerical range is equally divided into four consecutive numerical intervals, and mapped in ascending order of numerical value as low-risk threshold interval, medium-risk threshold interval, relatively high-risk threshold interval, and high-risk threshold interval.

[0062] Traverse the warning distribution matrix and count the number of micro-grid units whose safety warning values ​​fall within the high-risk threshold range. Calculate the proportion of the number of such micro-grid units to the total number of micro-grid units in the target area, and use this proportion as the regional warning feature of the target area.

[0063] The regional early warning characteristics are compared with the preset regional safety threshold. The overall ecological security early warning level of the entire target area is determined based on the comparison results. If the regional early warning characteristics are greater than the preset regional safety threshold, the overall ecological security early warning level of the target area is determined to be a severe early warning level. If the regional early warning characteristics are less than or equal to the preset regional safety threshold, the comprehensive ecological security early warning level of the target area will be determined as the normal monitoring level.

[0064] Because the overall ecological environment has a certain self-buffering and dilution capacity for local risks, a global ecological cascade collapse will only be triggered when the spatial proportion of extremely high-risk areas exceeds a specific physical critical point. In this embodiment, referring to the current ecological protection red line baseline ratio of natural resource management departments, the preset area safety threshold is set to 15%. This value represents the critical value of the spatial area ratio of the number of micro-grid units in the high-risk threshold range to the total number of micro-grid units in the target area, thereby effectively filtering false alarms caused by minor local deterioration.

[0065] Table 1 is an example table of comprehensive data on the multi-source monitoring characteristics and dynamic evolution of ecological security of specific micro-grid units within the target area at 25 consecutive time points.

[0066] Table 1: Comprehensive Data Table of Multi-scale Assessment and Dynamic Evolution of Ecological Security Table 1 shows the data demonstrating the method's ability to accurately quantify the actual evolutionary trend of a region and strip away short-term anomalous noise in complex multi-source ecological monitoring scenarios. During time nodes 1 to 3, the micro-grid natural indicators are at a disadvantage, with the initial and corrected indices exhibiting stable evolutionary characteristics, reflecting that the ecosystem is in a low-level homogeneous and conventional succession stage. In node 4, although soil erosion continues to decline naturally, the vegetation cover value stagnates at 25.5, resulting in weak ecological resilience recovery. At this point, thanks to the nonlinear compensation constraint of the net ecological driving coefficient, the system accurately corrects the assessment score downward, objectively quantifying the potential system vulnerability without assessment distortion. When the evolution reaches node 12, the original monitoring data encounters short-term extreme climate deterioration, and the initial index drops from 0.476 to 0.401. Traditional static assessment is prone to triggering false alarms of systemic degradation, but the ecological security warning value finally output by this embodiment, under the smooth convergence of the background value matrix and the first-order cumulative generation, does not follow the hypersensitive jump, but maintains the macro-recovery inertia and steadily climbs to 0.531. The table demonstrates that the traditional linear superposition method is prone to confusing local sporadic environmental fluctuations with the true evolutionary patterns, while this scheme can accurately filter out false interference from the background of the surface environment.

[0067] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0068] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0069] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0070] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An ecological safety early warning method based on multi-scale fusion and grey prediction, characterized in that, The specific steps include: The target area is divided into multiple micro-grid units, and multi-source monitoring data of each time node within the historical time window are obtained. The multi-source monitoring data is extracted and normalized in layers to obtain standardized feature data. Spatial alignment mapping is performed based on the micro-grid units to construct a multi-dimensional feature data matrix containing macro and micro dimensions. Based on the multidimensional feature data matrix, the historical time series corresponding to each micro-grid unit is constructed. The objective entropy weight method is used to extract the feature information of the standardized feature data in the historical time series to determine the objective weight. Based on the objective weight, time series weighted fusion is performed to construct the initial index sequence of each micro-grid unit. For each time node in the historical time series, the corresponding standardized feature data is extracted to construct a single-period feature vector. The initial index of the corresponding time node in the initial index sequence is used as the supervision label. The single-period feature vector is input into the preset machine learning model for regression fitting training. The trained model is used to quantify the contribution weight of each feature to the change in ecological security, and the contribution weight is used to dynamically compensate the initial index. The modified index sequence is constructed according to the time series combination. The modified index sequence of each micro-grid unit is input into the GM(1,1) grey prediction model and first-order accumulation generation and time-series evolution deduction are performed to output the safety warning value of the corresponding micro-grid unit within the preset prediction period. The safety warning values ​​of all micro-grid units are aggregated to construct a warning distribution matrix for the target area. The regional warning features of this warning distribution matrix are extracted and compared with the preset regional safety threshold to determine the comprehensive ecological security warning level of the target area. 2.The ecological security early warning method based on multi-scale fusion and grey prediction according to claim 1, characterized in that: The multi-source monitoring data includes natural environment data, activity load data, and landscape pattern data; The landscape pattern data specifically includes the Shannon diversity index, the Simpson diversity index, patch spread and patch cohesion; the natural environment data specifically includes vegetation cover and soil erosion; and the activity load data specifically includes industrial emissions and population density.

3. The ecological safety early warning method based on multi-scale fusion and grey prediction according to claim 2, characterized in that: The specific steps for obtaining the standardized feature data are as follows: The global maximum and global minimum values ​​of each specific indicator in the landscape pattern data within the historical time window are obtained. The range standardization method is used to perform positive normalization on the original landscape data of each micro-grid unit to obtain the corresponding normalized Shannon diversity index, normalized Simpson diversity index, normalized patch spread and normalized patch assemblage. For the landscape pattern data of each micro-grid unit at each time node after positive normalization, the normalized Shannon diversity index and the normalized Simpson diversity index are multiplied by the value 0.2 to construct the weighted Shannon diversity index and the weighted Simpson diversity index, respectively. The normalized patch spread and the normalized patch assemblage are multiplied by the value 0.3 to construct the weighted patch spread and the weighted patch assemblage. The weighted Shannon diversity index, the weighted Simpson diversity index, the weighted patch spread and the weighted patch assemblage are summed and superimposed with a preset translation constant to generate the comprehensive landscape index of each micro-grid unit at the corresponding time node. Obtain the global maximum and global minimum values ​​of each specific indicator in the natural environment data and activity load data within a historical time window; For vegetation cover as a positive indicator, the original value corresponding to each time node is subtracted from its corresponding global minimum value, and the difference is divided by the range between the global maximum and global minimum values ​​corresponding to the vegetation cover. The quotient is then summed with a preset translation constant to generate the normalized value of the positive indicator at the corresponding time node. For soil erosion, industrial emissions and population density, which are negative indicators, the global maximum value of each negative indicator is subtracted from its original value at each time point. The difference is then divided by the range between the global maximum and the global minimum value. The quotient is then summed with a preset translation constant to generate the normalized value of each negative indicator at the corresponding time point. Standardized feature data are constructed by using the comprehensive landscape index, the normalized values ​​of positive indicators corresponding to positive indicators, and the normalized values ​​of negative indicators corresponding to various negative indicators.

4. The ecological security early warning method based on multi-scale fusion and grey prediction according to claim 3, characterized in that: The specific steps for constructing the multidimensional feature data matrix are as follows: The normalized values ​​of the positive indicators of vegetation cover and the negative indicators of soil erosion, which belong to the natural environment data, are combined in a multidimensional sequence according to time nodes to construct a normalized natural environment data matrix. Similarly, the normalized values ​​of the negative indicators of industrial emissions and population density, which belong to the activity load data, are combined in a multidimensional sequence according to time nodes to construct a normalized activity load data matrix. The values ​​of the landscape comprehensive index, the normalized natural environment data matrix, and the normalized activity load data matrix are all uniformly mapped to a specific non-zero interval. A multidimensional feature data matrix is ​​constructed using the spatial coordinates of each micro-grid unit and its corresponding time node as a spatiotemporal alignment reference. The multidimensional feature data matrix consists of two data layers. The landscape comprehensive index is used as a feature channel and is spliced ​​with the normalized natural environment data matrix to jointly construct the first data layer of the multidimensional feature data matrix to represent the macro-dimensional dimension. The second data layer of the multidimensional feature data matrix is ​​constructed using the normalized activity load data matrix to represent the micro-dimensional dimension.

5. The ecological security early warning method based on multi-scale fusion and grey prediction according to claim 4, characterized in that: The specific steps for determining the objective weights are as follows: In the multidimensional feature data matrix, the standardized feature data of each micro-grid unit at each time node within the historical time window, distributed in the macro and micro dimensions, are extracted along the time dimension. The standardized feature data is then arranged in rows and columns according to the corresponding time nodes and indicator categories to construct the historical time series sequence corresponding to each micro-grid unit. The standardized feature data in the historical time series are traversed, and the ratio of the value of each feature at each time node to the sum of its values ​​in the entire historical time window is calculated to obtain the feature weight at the corresponding time node. Calculate the product of the feature weight and its natural logarithm at each time point, and sum the product values ​​along the time dimension. Multiply the sum by a smoothing adjustment coefficient to calculate the feature information dispersion of each standardized feature data. The smoothing adjustment coefficient is the ratio between the negative one value and the natural logarithm of the total number of time points within the historical time window. The difference between the numerical value and the feature information dispersion of each standardized feature data is calculated to obtain the corresponding feature difference coefficient. The ratio of the feature difference coefficient of a single standardized feature data to the sum of the feature difference coefficients of all standardized feature data is then calculated to determine the objective weight of each standardized feature data.

6. The ecological security early warning method based on multi-scale fusion and grey prediction according to claim 5, characterized in that: The specific steps for constructing the initial exponential sequence are as follows: The standardized feature data of each item in the historical time series is multiplied by its corresponding objective weight at the corresponding time node to generate the corresponding weighted value. The micro-grid unit and the corresponding time node are used as the joint dimension. The weighted values ​​belonging to the macro-dimensional and micro-dimensional dimensions are reduced and summed to construct the initial index sequence corresponding to each micro-grid unit.

7. The ecological security early warning method based on multi-scale fusion and grey prediction according to claim 1, characterized in that: The preset machine learning model is constructed based on a combination of the XGBoost regression architecture and the SHAP analysis framework. The specific quantification steps for the contribution weight are as follows: For each micro-grid unit, the standardized feature data of each time node in its historical time series is extracted, and the single-period feature vector of the corresponding time node is constructed. The single-period feature vector of each time node is then rearranged with the time node as the sample dimension and the feature category contained in the single-period feature vector as the data feature dimension to construct the input feature matrix. Using the input feature matrix as input, the initial exponents in the initial exponent sequence of the corresponding micro-grid units are used as supervision labels and simultaneously input into the XGBoost regression architecture for regression fitting training to construct a nonlinear mapping relationship model. The SHAP analysis framework is invoked to calculate the marginal contribution of each standardized feature data in the input feature matrix to the output of the nonlinear mapping relationship model, and the marginal contribution is mapped to the quantified influence value of the corresponding standardized feature data. For each time point, the absolute value of the quantitative impact value corresponding to each standardized feature data is extracted, and the ratio of the absolute value of the quantitative impact value of a single standardized feature data to the sum of the absolute values ​​of the quantitative impact values ​​of all standardized feature data is calculated to determine the contribution weight of each standardized feature data at the current time point.

8. The ecological security early warning method based on multi-scale fusion and grey prediction according to claim 7, characterized in that: The specific steps for constructing the modified exponential sequence are as follows: The contribution weights corresponding to each standardized feature data distributed in the micro dimension of the multidimensional feature data matrix are aggregated to generate the negative pressure coefficient at the current time point, and the contribution weights corresponding to each standardized feature data distributed in the macro dimension of the multidimensional feature data matrix are aggregated to generate the ecological resilience coefficient at the current time point. The difference between the ecological resilience coefficient and the negative pressure coefficient is calculated, and the opposite of its squared term is extracted to obtain the net ecological driving coefficient at the corresponding time node. The numerical value is summed with the net ecological driving coefficient to construct a dynamic compensation multiplier. The initial index at the corresponding time node is multiplied with the dynamic compensation multiplier to generate the correction index at the current time node. The correction indices at each time node are cascaded in chronological order to construct the correction index sequence for the corresponding micro-grid unit.

9. The ecological security early warning method based on multi-scale fusion and grey prediction according to claim 1, characterized in that: The specific steps for outputting the security warning value are as follows: For each micro-grid unit, the values ​​in its correction exponent sequence are processed by first-order accumulation generation in chronological order to construct a smoothed accumulation generation sequence. The values ​​in the modified exponent sequence are extracted to form a data observation vector, and the mean data of adjacent terms in the accumulated generation sequence are extracted to construct a feature column. The feature column is then concatenated with a preset constant column to construct a background value matrix. Based on the GM(1,1) grey prediction model architecture, the least squares fitting algorithm is used to fit the data observation vector and the background value matrix to extract the development coefficient and grey action of the corresponding micro grid unit. Based on the development coefficient and the gray action quantity, a time-series evolution mapping relationship is constructed, and the time step corresponding to the preset prediction period is substituted into the time-series evolution mapping relationship to perform evolution deduction calculation, so as to obtain the cumulative prediction value corresponding to the preset prediction period. The accumulated predicted values ​​are subjected to first-order cumulative reduction and restoration processing to generate the corresponding micro-grid unit's safety warning value within the preset prediction period.

10. The ecological security early warning method based on multi-scale fusion and grey prediction according to claim 1, characterized in that: Extract the spatial coordinates of all micro-grid units within the target area, and then perform spatial topology stitching of the safety warning values ​​of each micro-grid unit within the preset prediction period according to the corresponding spatial coordinates to construct a warning distribution matrix. The maximum and minimum values ​​of the safety warning values ​​in the warning distribution matrix are obtained to determine the numerical range. The numerical range is then divided into four consecutive numerical intervals at equal intervals, and mapped sequentially in ascending order of numerical values ​​to low-risk threshold interval, medium-risk threshold interval, relatively high-risk threshold interval, and high-risk threshold interval. Traverse the warning distribution matrix and count the number of micro-grid units whose safety warning values ​​fall within the high-risk threshold range. Calculate the proportion of the number of micro-grid units to the total number of micro-grid units in the target area, and use this proportion as the regional warning feature of the target area. The regional early warning characteristics are compared with the preset regional safety threshold, and the comprehensive ecological security early warning level of the entire target area is determined based on the comparison results.

Images