Karst carbon sink change prediction method and system based on machine learning

Abstract

This invention provides a machine learning-based method and system for predicting changes in karst carbon sinks. It acquires environmental baseline data and measured carbon sink data for karst areas, and constructs a dynamic coupling body for karst carbon sinks, including elements such as real-time correlation links. A pre-trained carbon sink evolution prediction model is initiated to perform bidirectional interactive modeling with this coupling body. The carbon sink evolution prediction model generates feature iteration instructions to drive feature updates in the coupling body, and the coupling body provides feedback on feature change signals to correct model parameters, obtaining intermediate results from multiple rounds of interactive modeling. Based on these intermediate results, operations such as correlation link strengthening are performed on the coupling body to obtain the evolved coupling body. The carbon sink evolution prediction model is then used to perform cross-dimensional feature decoding and correlation trend inference on the evolved coupling body to generate karst carbon sink change prediction results. This invention considers the relationship between environmental factors and carbon sinks, improving prediction accuracy and adaptability through dynamic interaction and optimization.

Description

Technical Field

[0001] This invention relates to the field of karst ecology research, and more specifically, to a method and system for predicting changes in karst carbon sinks based on machine learning. Background Technology

[0002] In the field of ecological research, karst carbon sinks, as an important component of the carbon cycle, are crucial for addressing global climate change and maintaining ecosystem balance. Currently, most existing methods for predicting karst carbon sinks are based on simple statistical models or empirical formulas. These methods often consider only a limited number of environmental factors and do not delve deeply enough into the complex relationships between environmental factors and carbon sink data. For example, some methods focus only on the impact of a few environmental variables such as temperature and precipitation on karst carbon sinks, ignoring the combined effects of many other potential influencing factors. Furthermore, existing methods lack effective modeling of the dynamic interaction between environmental and carbon sink data, making it difficult to capture the complex correlation patterns between the two over time. In addition, during the prediction process, existing methods typically use static model parameters, which cannot be dynamically adjusted and optimized based on feedback from actual data, significantly limiting the accuracy and reliability of the prediction results. Therefore, developing a more advanced and comprehensive method for predicting karst carbon sink changes is of significant practical importance. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for predicting changes in karst carbon sinks based on machine learning, the method comprising: The environmental baseline data and measured carbon sink data of the karst area were obtained. The environmental baseline data included various raw environmental data that affect the carbon sink of karst, and the measured carbon sink data included the actual carbon sink data continuously recorded in the karst area. A dynamic coupling body for karst carbon sinks is constructed based on environmental baseline data and measured carbon sink data. The dynamic coupling body for karst carbon sinks includes real-time correlation links between environmental data and carbon sink data, feature interaction channels, and coupling strength identifiers. Initiate bidirectional interactive modeling between the pre-trained carbon sink evolution prediction model and the dynamic coupling of karst carbon sinks. Drive the feature update of the dynamic coupling of karst carbon sinks by generating feature iteration instructions through the carbon sink evolution prediction model. Correct the parameters of the carbon sink evolution prediction model by the feedback feature change signals from the dynamic coupling of karst carbon sinks, and generate intermediate results of multiple rounds of interactive modeling. Based on the intermediate results of multi-round interactive modeling, the dynamic coupling body of karst carbon sink is strengthened by association link enhancement, feature channel expansion and coupling strength improvement, and the evolved dynamic coupling body of karst carbon sink is obtained. By using a carbon sink evolution prediction model, cross-dimensional feature decoding and correlation trend inference are performed on the evolved dynamic coupling of karst carbon sinks to generate karst carbon sink change prediction results.

[0004] In another aspect, embodiments of the present invention also provide a karst carbon sink change prediction system based on machine learning, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0005] Based on the above, this embodiment of the invention acquires comprehensive environmental baseline data and continuous measured carbon sink data from karst areas to construct a dynamic coupling system for karst carbon sinks. This system deeply correlates environmental and carbon sink data, forming a complex system including real-time correlation links, feature interaction channels, and coupling strength indicators. This system can characterize the dynamic interaction between environmental factors and carbon sinks. It initiates bidirectional interactive modeling between a pre-trained carbon sink evolution prediction model and the dynamic coupling system for karst carbon sinks, achieving bidirectional information flow and dynamic optimization between the model and the coupling system. The model generates feature iteration instructions to drive feature updates in the coupling system, while the coupling system provides feedback signals of feature changes to correct model parameters. After multiple rounds of interaction, intermediate results are generated, enabling the model to continuously adapt to changes in actual data, improving prediction accuracy and adaptability. Based on the intermediate results of multi-round interactive modeling, the dynamic coupling system for karst carbon sinks is optimized, strengthening correlation links, expanding feature channels, and increasing coupling strength, further enhancing the coupling system's ability to express the relationship between the environment and carbon sinks. Finally, by using a carbon sink evolution prediction model, cross-dimensional feature decoding and correlation trend inference are performed on the evolved coupled body. This allows for in-depth analysis of the potential correlation between the environment and carbon sink characteristics from multiple dimensions, accurate inference of the changing trend of karst carbon sink, and generation of prediction results with high reliability and high accuracy. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the execution flow of the machine learning-based method for predicting changes in karst carbon sinks provided in an embodiment of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of the machine learning-based karst carbon sink change prediction system provided in an embodiment of the present invention. Detailed Implementation

[0008] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a machine learning-based method for predicting changes in karst carbon sinks, as provided in one embodiment of the present invention. The following is a detailed description of this machine learning-based method for predicting changes in karst carbon sinks.

[0009] Step S110: Obtain environmental baseline data and measured carbon sink data for the karst area. The environmental baseline data includes various raw environmental data that affect the carbon sink in the karst area, and the measured carbon sink data includes the actual carbon sink data continuously recorded in the karst area.

[0010] In this embodiment, a representative karst region was selected as the study area, which possesses a long-term environmental monitoring foundation and a complete carbon sink observation system. Environmental baseline data was acquired through multiple environmental monitoring stations set up within this region. These stations are distributed across different geomorphic units and vegetation types to ensure data representativeness. The environmental baseline data encompasses multiple aspects of raw environmental data, including atmospheric, hydrological, soil, and biological data. Atmospheric data includes air temperature, relative humidity, gas composition concentration, precipitation, and airflow conditions, which are continuously collected at fixed time intervals using meteorological monitoring equipment at the stations. Hydrological data includes the flow rate, water temperature, pH, and dissolved substance content of major water bodies within the region, collected periodically through hydrological monitoring stations. Soil data includes the organic matter content, pH, physical structure parameters, and microbial activity of soils at different depths, obtained through soil sampling and laboratory analysis. Biological data includes the coverage, growth, and physiological activity indicators of major vegetation within the region, as well as the types and quantities of animals in the soil, obtained through regular field surveys and sampling analysis. Carbon sequestration data is continuously recorded through carbon sequestration observation plots established in the region, using a combination of observation techniques. This includes indicators such as ecosystem carbon uptake, carbon release, and net carbon exchange. The data recording time intervals are matched with the environmental baseline data to ensure consistency across time. During data collection, sensitive information such as monitoring station locations is anonymized by converting specific location information into regional codes. Furthermore, encryption technology is used for data transmission and storage to ensure data security and privacy, preventing unauthorized access and data leakage.

[0011] Step S120: Construct a dynamic coupling body for karst carbon sinks based on environmental baseline data and measured carbon sink data. The dynamic coupling body for karst carbon sinks includes real-time correlation links between environmental data and carbon sink data, feature interaction channels, and coupling strength identifiers.

[0012] In this embodiment, after acquiring environmental baseline data and measured carbon sink data, the construction of a dynamic coupling body for karst carbon sinks begins. The construction process mainly includes preprocessing of environmental baseline data and measured carbon sink data, feature extraction, association binding, and construction of the coupling body structure, ultimately forming a coupling body that reflects the dynamic relationship between environmental data and carbon sink data.

[0013] Step S121: Perform category splitting on the environmental baseline data to obtain multiple environmental data subsets. Each environmental data subset corresponds to an environmental factor that affects carbon sinks, and each environmental data subset contains continuous record data of the corresponding category factor.

[0014] Next, the acquired environmental baseline data was categorized. Based on the different aspects of the impact of environmental factors on karst carbon sinks, the environmental baseline data was divided into multiple environmental data subsets. For example, atmospheric data was integrated into an atmospheric environmental data subset, which includes continuous records of air temperature, relative humidity, gas composition concentration, precipitation, and air flow; hydrological data was integrated into a hydrological environmental data subset, which includes continuous records of water flow, water temperature, pH, and dissolved substance content; soil data was integrated into a soil environmental data subset, which includes continuous records of organic matter content, pH, physical structure parameters, and microbial activity at different soil depths; and biological data was integrated into a biological environmental data subset, which includes continuous records of vegetation cover, growth, physiological activity indicators, and soil animal species and quantities. Each environmental data subset has a clear category identifier, and the time range of the data is consistent with the time range of the measured carbon sink data.

[0015] Step S122: Divide the measured carbon sink data into time axis segments to obtain multiple carbon sink data subsets. Each carbon sink data subset corresponds to a continuous recording period and contains continuous carbon sink data within the corresponding period.

[0016] Next, the measured carbon sink data is divided into time-series segments. Considering the influence of seasonal and other periodic factors on karst carbon sink processes, the measured carbon sink data is divided into multiple consecutive recording periods in chronological order. For example, the data can be divided into multiple periods according to seasonal cycles, with each period corresponding to a season. In this way, each subset of carbon sink data contains continuous records of carbon sink data within that season, such as continuous records of indicators like ecosystem carbon uptake, carbon release, and net carbon exchange within that season. This time-series segmentation allows each subset of carbon sink data to correspond to a relatively independent and characteristic time unit.

[0017] Step S123: Extract the temporal variation features of each environmental data subset. The temporal variation features reflect the change trajectory, change pattern and change correlation of the corresponding environmental elements over the recording period.

[0018] Next, the temporal variation characteristics of each environmental data subset were extracted. Taking the atmospheric environmental data subset as an example, this subset contains long-term series data such as air temperature and relative humidity. By analyzing these data, the characteristics such as their variation trajectory, variation pattern, and variation correlation with the recording period were extracted. These characteristics can reflect the changes of atmospheric environmental elements at different time stages and their potential impact on carbon sequestration processes.

[0019] Step S1231: Perform continuous segmentation processing on the continuous recorded data of the environmental data subset to obtain multiple data segments. Each data segment obtained by segmentation corresponds to a recording sub-period. The time length of the data segment is consistent with the granularity of the recording sub-period of the carbon sink data subset.

[0020] First, the continuously recorded data of the atmospheric environment data subset is segmented. Based on the granularity of the recording period of the carbon sink data subset, the continuously recorded data of the atmospheric environment data subset is divided into multiple data segments. For example, if the recording period of the carbon sink data subset is one season, it is subdivided into three months as recording sub-periods. Then, the continuously recorded data of the atmospheric environment data subset is also correspondingly divided into multiple data segments according to the three-month time length, with each data segment corresponding to the atmospheric environment data of one recording sub-period. This processing ensures that the segmentation of the environmental data is consistent with the time granularity of the carbon sink data, facilitating subsequent feature correlation analysis within the same sub-period.

[0021] Step S1232: Analyze the change trajectory of environmental data in each data segment, determine the starting point, node and end point of change of environmental data within the data segment, and extract the data fluctuation characteristics during the change process.

[0022] Next, the trajectory of environmental data changes within each data segment is analyzed. Taking air temperature data from one data segment (such as the first month of a quarter) as an example, the starting point of the change is determined by analyzing the continuously recorded temperature data for that month, with the temperature value at the beginning of the month as the starting point and the temperature value at the end of the month as the ending point. Within the trajectory, the points where significant temperature changes are observed are identified as change nodes, such as moments of significant temperature increases or decreases. Then, the data fluctuation characteristics during the change process are extracted, including the daily and weekly temperature variations within the month, as well as the temperature trends and rates of change before and after the change nodes. These fluctuation characteristics reflect the dynamic changes of environmental data within that data segment.

[0023] Step S1233: Calculate the frequency of change of environmental data in each data segment. The frequency of change reflects the number of times the environmental data changes significantly within the data segment and the pattern of the change interval.

[0024] Next, the frequency of environmental data changes in each data segment is statistically analyzed. For the aforementioned air temperature data segment, a standard for a significant change is defined; for example, a significant change is considered to occur when the temperature change exceeds a certain threshold within a certain time period. By scanning the temperature data throughout the entire data segment, the number of significant changes meeting this standard is counted. Simultaneously, the time intervals between these significant changes are analyzed to identify patterns in the intervals, such as whether they exhibit a uniform or concentrated distribution, and the average interval time. This information reflects the activity level and periodicity of environmental data changes.

[0025] Step S1234: Analyze the characteristics of environmental data change and connection between adjacent data segments, and determine the correlation and transition rules between the end point of change of the preceding data segment and the starting point of change of the subsequent data segment.

[0026] Next, the characteristics of environmental data transitions between adjacent data segments are analyzed. Taking two adjacent data segments (such as the first and second months of a quarter) as an example, the transition of environmental data between these two data segments is studied.

[0027] For example, step S1234-1: Extract the endpoint value and trend direction of environmental data changes in the prior data segment. The trend direction reflects the change status of the data at the endpoint.

[0028] First, extract the endpoint values ​​of environmental data (such as air temperature) from the prior data segment (the first month), that is, the temperature value at the last moment of that month. At the same time, by analyzing the temperature data changes in the last period of that month (such as the last week), determine the direction of the trend, such as whether it shows an upward trend, a downward trend, or tends to stabilize.

[0029] Step S1234-2: Extract the starting point value and trend direction of environmental data changes in the continuation data segment, so that the starting time and the ending time of the previous data segment form a continuous connection.

[0030] Next, extract the starting point value of environmental data changes in the subsequent data segment (the second month), that is, the temperature value at the beginning of the month. This time is continuous with the end time of the previous data segment, ensuring the temporal continuity of the data. Similarly, analyze the temperature data changes in the initial period of the month (such as the first week) to determine the direction of the change trend.

[0031] Step S1234-3: Calculate the difference between the end value of the preceding data segment and the beginning value of the subsequent data segment, and analyze the change and connection characteristics reflected by the magnitude and sign of the difference.

[0032] Calculate the difference between the end value of the preceding data segment and the beginning value of the following data segment. The magnitude of the difference can indicate the range of change between the two data segments at the junction. The sign of the difference (positive or negative) can indicate whether the beginning value of the following data segment is higher or lower than the end value of the preceding data segment, thereby analyzing the characteristics of the transition, such as whether it is a smooth transition or a jump change.

[0033] Steps S1234-4: Compare the changing trends of the preceding data segment and the subsequent data segment to determine whether the changing trends of the two data segments are consistent, opposite, or whether there is a transitional change.

[0034] By comparing the changing trends of earlier and subsequent data segments, we can determine whether their trends are consistent (e.g., both showing an upward trend), completely opposite (e.g., one rising and the other falling), or involve a transitional change (e.g., a shift from an upward trend to a stable trend). This trend comparison can reflect the continuity and consistency of environmental data changes between adjacent time periods.

[0035] Steps S1234-5: Integrate numerical differences, trend direction comparison results, and change connection characteristics to form change connection features between adjacent data segments. Change connection features reflect the continuity and transition patterns of environmental data changes within consecutive sub-periods.

[0036] By integrating the numerical differences, trend direction comparisons, and change continuity characteristics obtained above, a comprehensive change continuity feature between adjacent data segments is formed. This change continuity feature can comprehensively reflect the consistency of environmental data changes within consecutive sub-periods, including the magnitude of changes, the continuity of trends, and the smoothness of transitions.

[0037] Step S124: Extract the temporal variation features of each carbon sink data subset. The temporal variation features reflect the change trajectory, change pattern and connection characteristics of the carbon sink data in the corresponding recording period.

[0038] In this embodiment, after extracting the temporal variation characteristics of environmental data subsets, temporal variation characteristics are extracted for each carbon sink data subset. Taking a certain carbon sink data subset (corresponding to carbon sink data for a certain season) as an example, this carbon sink data subset contains continuous records of indicators such as ecosystem carbon uptake, carbon release, and net carbon exchange during that season. By analyzing these data, their change trajectories are extracted to determine the starting value, important change nodes, and ending value of the carbon sink data within that recording period; the change patterns are analyzed, such as whether the carbon sink data shows an increasing, decreasing, or periodic fluctuation pattern within that period; simultaneously, the connection characteristics between this carbon sink data subset and adjacent carbon sink data subsets are studied, including numerical transitions and trend continuity. These temporal variation characteristics can reflect the dynamic changes of the carbon sink process in different recording periods.

[0039] Step S125: Associate and bind the temporal change characteristics of the environmental data subset corresponding to the same recording period with the temporal change characteristics of the carbon sink data subset, assign a coupling strength identifier to each binding result, and connect all the initial coupling units formed by binding through real-time association links and open feature interaction channels to form a dynamic coupling body for karst carbon sink.

[0040] Then, the temporal variation characteristics of environmental data subsets corresponding to the same recording period are associated and bound with the temporal variation characteristics of carbon sink data subsets. For example, the temporal variation characteristics of atmospheric environmental data subsets, hydrological environmental data subsets, soil environmental data subsets, and biological environmental data subsets corresponding to a certain quarter are associated with the temporal variation characteristics of carbon sink data subsets for that quarter, forming an initial coupling unit. A coupling strength identifier is assigned to each such binding result. This coupling strength identifier is determined based on the tightness of the association between the temporal variation characteristics of environmental data and the temporal variation characteristics of carbon sink data; the tighter the association, the higher the value of the coupling strength identifier. All initial coupling units formed in this way are connected in chronological order through a real-time association link, and a feature interaction channel is opened simultaneously, enabling the transmission and interaction of feature information between the initial coupling units, ultimately forming a dynamic coupling body for karst carbon sinks.

[0041] Step S130: Initiate bidirectional interactive modeling between the pre-trained carbon sink evolution prediction model and the dynamic coupling of karst carbon sinks. Drive the feature update of the dynamic coupling of karst carbon sinks by generating feature iteration instructions through the carbon sink evolution prediction model. Correct the parameters of the carbon sink evolution prediction model by the feature change signals fed back from the dynamic coupling of karst carbon sinks, and generate intermediate results of multiple rounds of interactive modeling.

[0042] In this embodiment, after constructing the dynamic coupler of the karst carbon sink, a bidirectional interactive modeling process is initiated between the pre-trained carbon sink evolution prediction model and the dynamic coupler. The pre-trained carbon sink evolution prediction model is trained based on a large amount of historical environmental and carbon sink data and has the ability to predict carbon sink change trends. During the bidirectional interactive modeling process, the carbon sink evolution prediction model and the dynamic coupler of the karst carbon sink interact with each other. The model drives the feature update of the coupler by generating feature iteration instructions, while the coupler corrects the model's parameters by feeding back feature change signals. After multiple rounds of such interaction, intermediate results of multiple rounds of interactive modeling are generated.

[0043] Step S131: Input all initial coupling units of the dynamic coupling body of karst carbon sink and their corresponding coupling strength identifiers into the feature receiving module of the carbon sink evolution prediction model, and generate the initial input features of the model through feature adaptation processing inside the feature receiving module.

[0044] First, all initial coupling units in the dynamic coupling body of the karst carbon sink, along with the coupling strength identifier corresponding to each initial coupling unit, are input into the feature receiving module of the carbon sink evolution prediction model. The feature receiving module performs feature adaptation processing on the input initial coupling units and coupling strength identifiers, including format unification, dimensional adjustment, and data standardization for different types of features, so that these features can meet the requirements of subsequent model processing, ultimately generating the initial input features for the model.

[0045] Step S132: The initial input features of the carbon sink evolution prediction model are transformed, correlated, and key information extracted through the hierarchical modeling module. This generates feature iteration instructions that drive the feature update of the dynamic coupling body of karst carbon sink. The feature iteration instructions include the feature optimization direction, optimization range, and optimization objective.

[0046] Next, the hierarchical modeling module of the carbon sink evolution prediction model processes the initial input features. This module comprises multiple processing levels. First, feature transformation converts the initial input features into a more suitable feature representation for model analysis. Then, association mining analyzes the intrinsic relationships between different features to identify feature combinations that significantly impact carbon sink prediction. Following this, key information extraction extracts information reflecting key patterns of carbon sink changes from the feature combinations obtained through association mining. Based on these processing results, feature iteration instructions are generated. These instructions include the feature optimization direction (e.g., enhancing the expression of certain environmental features), the optimization scope (e.g., features specific to a particular time period), and the optimization objective (e.g., improving the correlation between features and carbon sink data).

[0047] Step S1321: Input the initial input features of the model into the first feature processing layer of the hierarchical modeling module, and use the feature mapping algorithm of the first feature processing layer to transform the input features into a standard feature vector of uniform dimension, thereby generating standard dimension features.

[0048] After receiving the initial input features from the model, the first feature processing layer of the hierarchical modeling module processes these features using a feature mapping algorithm. This algorithm maps initial input features of different types and dimensions to a unified feature space, transforming them into standard feature vectors with the same dimension, thus generating standard-dimensional features. This process allows subsequent feature processing to be performed in a unified dimensional space, facilitating comparison and computation between different features.

[0049] Step S1322: Input the standard dimensional features into the second feature processing layer of the hierarchical modeling module, and analyze the correlation between different feature items in the standard dimensional features through the association mining algorithm of the second feature processing layer to select feature combinations with effective correlation.

[0050] The generated standard dimensional features are input into the second feature processing layer, which uses an association mining algorithm to analyze different feature items in the standard dimensional features. By calculating the correlation coefficient, mutual information, and other indicators between feature items, the correlation between them is determined. Then, based on a set threshold, feature combinations with effective correlations are selected. These feature combinations can more comprehensively reflect the complex relationship between environmental factors and carbon sinks.

[0051] Step S1323: Perform feature fusion processing on the selected feature combinations, integrate the information of each feature item in the feature combination through weighted fusion, highlight the influence weight of key features, and generate a set of related features.

[0052] The selected feature combinations undergo feature fusion processing using a weighted fusion method. Different weights are assigned to each feature item based on its importance within the feature combination, and then the information from each feature item is integrated according to its weight. The weights are determined based on the degree of influence of each feature item on carbon sink changes; the greater the influence, the higher the weight, thus highlighting the impact of key features and ultimately generating a set of associated features.

[0053] Step S1324: Input the associated feature set into the third feature processing layer of the hierarchical modeling module, and enhance the expression intensity of key feature components in the associated feature set through the feature enhancement algorithm of the third feature processing layer, suppress the interference of irrelevant features, and generate enhanced associated features.

[0054] The set of associated features is input into the third feature processing layer, which uses a feature enhancement algorithm to process the set of associated features. This feature enhancement algorithm identifies key feature components by analyzing the feature components and enhances their expression intensity, while suppressing irrelevant feature components that are not closely related to carbon sink changes, reducing their interference with model analysis, and generating enhanced associated features.

[0055] Step S1325: The instructions of the enhanced correlation feature input hierarchical modeling module are used to generate a sub-layer. Based on the optimization objectives preset by the carbon sink evolution prediction model and the feature requirements of the karst carbon sink dynamic coupling body, a preliminary feature update instruction is generated. The preliminary instruction is then processed for format adaptation to obtain the final feature iteration instruction.

[0056] The enhanced correlation features are input into the instruction generation sublayer. Based on the optimization objectives preset by the carbon sink evolution prediction model (such as improving the accuracy of carbon sink prediction) and the feature requirements of the karst carbon sink dynamic coupler (such as feature completeness and timeliness), the instruction generation sublayer generates preliminary feature update instructions. Then, the preliminary instructions undergo format adaptation processing to conform to the instruction format that the karst carbon sink dynamic coupler can recognize and execute, resulting in the final feature iteration instructions.

[0057] Step S133: Transmit the feature iteration command to the karst carbon sink dynamic coupler, trigger the temporal change feature optimization of the initial coupler unit, update the feature details and correlation of the initial coupler unit, and generate the optimized initial coupler unit.

[0058] The generated feature iteration instructions are transmitted to the karst carbon sink dynamic coupler. Upon receiving the instructions, the coupler optimizes the temporal variation characteristics of the initial coupler units according to the feature optimization direction, optimization range, and optimization objective contained in the instructions. For example, for the temporal variation characteristics of a certain subset of environmental data specified in the instructions, its feature details are adjusted, such as the smoothness of the change trajectory and the recognition accuracy of change nodes. At the same time, the correlation between the initial coupler units is updated to make the correlation tighter and more reasonable, and finally, the optimized initial coupler units are generated.

[0059] Step S134: Extract the feature change signal of the optimized initial coupling unit. The feature change signal includes the difference information before and after feature update, the change information of coupling strength identifier and the transmission status information of associated link. Feed the feature change signal back to the parameter correction module of the carbon sink evolution prediction model.

[0060] After the initial coupling unit is optimized, the characteristic change signal of the optimized initial coupling unit is extracted. This characteristic change signal contains several aspects of information. First, there is the difference information before and after the feature update, that is, the differences in the change trajectory and change pattern between the optimized time-series change features and those before optimization. Second, there is the change information of the coupling strength identifier, that is, whether the coupling strength identifier of the optimized initial coupling unit has changed and the degree of change. Third, there is the transmission status information of the associated links, that is, the efficiency and delay of feature transmission in the associated links. These characteristic change signals are fed back to the parameter correction module of the carbon sink evolution prediction model.

[0061] Step S135: The parameter correction module adjusts the internal parameters of the hierarchical modeling module based on the feature change signal. The adjustment range covers the feature processing weight, association mining rules and key information extraction threshold, and generates parameter adjustment results. After multiple iterations, the feature iteration instructions of each round are integrated with the parameter adjustment results to form a multi-round interactive modeling intermediate result.

[0062] After receiving feature change signals, the parameter correction module adjusts the internal parameters of the hierarchical modeling module based on these signals. The adjustments include: feature processing weights (adjusting the processing weights of different features by each feature processing layer in the hierarchical modeling module to adapt to feature changes); association mining rules (modifying the rules for judging the correlation between feature items in the association mining algorithm to improve the accuracy of association mining); and key information extraction thresholds (adjusting the threshold for extracting key information from associated features to ensure that the extracted information is more critical and effective). After these parameter adjustments, parameter adjustment results are generated. Multiple iterations are performed according to steps S131 to S135, with each iteration generating corresponding feature iteration instructions and parameter adjustment results. Finally, the feature iteration instructions and parameter adjustment results from each round are integrated to form the intermediate results of multi-round interactive modeling.

[0063] Step S140: Based on the intermediate results of multi-round interactive modeling, perform correlation link strengthening, feature channel expansion and coupling strength enhancement on the dynamic coupler of karst carbon sink to obtain the evolved dynamic coupler of karst carbon sink.

[0064] In this embodiment, after obtaining intermediate results from multiple rounds of interactive modeling, the dynamic coupling body of karst carbon sink is optimized based on these intermediate results, including operations such as strengthening the association links, expanding the feature channels, and improving the coupling strength, so that the performance of the coupling body is further improved, and finally the evolved dynamic coupling body of karst carbon sink is obtained.

[0065] Step S141: Extract the feature optimization direction information contained in the feature iteration instructions of each round from the intermediate results of multi-round interactive modeling, and integrate all optimization direction information to form a comprehensive optimization requirement.

[0066] The feature optimization direction information contained in the feature iteration instructions of each round is extracted from the intermediate results of multi-round interactive modeling. This information reflects the focus and direction of feature optimization of the coupled body in different rounds. All these optimization direction information are integrated, and the common needs and main trends are analyzed to form a comprehensive optimization requirement. This comprehensive optimization requirement clarifies the goals and directions that the linkage strengthening needs to achieve.

[0067] Step S142: Based on the comprehensive optimization requirements analysis, analyze the transmission status of the associated links between the initial coupling units in the dynamic coupling body of karst carbon sink, and identify the weak links where the transmission efficiency does not meet the optimization requirements.

[0068] Based on the comprehensive optimization requirements, a comprehensive analysis of the transmission status of the interconnected links between the initial coupling units in the dynamic coupling body of karst carbon sinks is conducted. The transmission status includes indicators such as the speed, accuracy, and stability of characteristic transmissions. By comparing these indicators with the standards set in the comprehensive optimization requirements, the interconnected link sections whose transmission efficiency does not meet the requirements are identified, and these sections are the weak links that need to be strengthened.

[0069] Step S143: Adjust the link structure of weak links in the associated links, increase the number of feature transmission channels, optimize channel transmission paths, and improve the feature transmission efficiency of the links.

[0070] For the identified weak links in the associated links, the link structure is adjusted. Specific measures include increasing the number of feature transmission channels to improve the total capacity of feature transmission through parallel transmission; optimizing the transmission path of the channels to reduce the number of nodes and path length during transmission, thereby reducing transmission latency and improving the feature transmission efficiency of the link.

[0071] Step S1431: Obtain the current structural information of the weak link in the associated link, including the number of transmission channels, channel transmission path, channel bandwidth and channel delay parameters.

[0072] First, obtain the current structural information of the weak link in the associated link, and understand in detail the number of existing transmission channels, the transmission path of each channel, the bandwidth of the channel, and the transmission delay parameters.

[0073] Step S1432: Based on the characteristic transmission volume requirements and transmission efficiency standards in the comprehensive optimization requirements, calculate the target number of channels, target transmission paths, target bandwidth, and target delay parameters required for link structure adjustment.

[0074] Based on the requirements for feature transmission volume (such as the amount of feature data to be transmitted per unit time) and transmission efficiency standards (such as the upper limit of transmission delay) in the comprehensive optimization requirements, and combined with the current structural information, calculate the target number of channels (how many channels need to be added), target transmission path (what the optimized path should look like), target bandwidth (the bandwidth that each channel needs to achieve), and target delay parameters (to what extent the transmission delay needs to be reduced) required for link structure adjustment.

[0075] Step S1433: Increase the number of transmission channels for weak links according to the target parameters. The new channels and the original channels form a parallel transmission architecture to realize the redundancy backup function of feature transmission.

[0076] Based on the calculated target number of channels, additional transmission channels are added to the weakest links in the associated links. The newly added channels, together with the existing channels, form a parallel transmission architecture. This architecture not only increases the total capacity of feature transmission but also enables redundant backup of feature transmission. When one channel fails, other channels can continue to transmit feature data, ensuring transmission reliability.

[0077] Step S1434: Optimize the transmission path of the transmission channel, eliminate redundant transmission nodes in the path, shorten the feature transmission distance, and reduce transmission delay.

[0078] The transmission path of the transmission channel is optimized by analyzing the role and necessity of each transmission node in the path, eliminating redundant transmission nodes, and simplifying the transmission path. At the same time, the path is replanned to shorten the transmission distance from the sender to the receiver, thereby reducing latency and improving transmission speed.

[0079] Step S1435: Adjust the bandwidth parameters of each transmission channel so that the channel bandwidth can meet the transmission volume requirements of the corresponding characteristics, and make the transmission delay of all channels reach the transmission standard specified by the target delay parameter, thus completing the link structure adjustment.

[0080] Based on the target bandwidth and target delay parameters, adjust the bandwidth parameters of each transmission channel. For channels that need to transmit a large amount of feature data, appropriately increase their bandwidth; for channels that transmit a smaller amount of data, maintain or appropriately reduce the bandwidth to achieve reasonable resource allocation. Through adjustment, ensure that the bandwidth of each channel can meet the transmission volume requirements of the corresponding feature, and that the transmission delay of all channels meets the transmission standard specified by the target delay parameters, thereby completing the link structure adjustment.

[0081] Step S144: Extract the model adaptation information contained in the parameter adjustment results of each round from the intermediate results of multi-round interactive modeling, and integrate all adaptation information to form a model parameter adaptation standard.

[0082] Model adaptation information is extracted from the intermediate results of multi-round interactive modeling, including the parameter adjustment results of each round. This information reflects the adaptation of the carbon sink evolution prediction model to the characteristics of the coupled body under different parameter adjustments. All this adaptation information is integrated to analyze the optimal fit between model parameters and coupled body characteristics, forming a model parameter adaptation standard.

[0083] Step S145: Adjust the transmission rules of the associated links based on the model parameter adaptation standard, including the characteristic transmission format, transmission frequency and transmission priority, so that the characteristics of the associated link transmission are adapted to the parameter requirements of the carbon sink evolution prediction model, and complete the association link strengthening.

[0084] Based on model parameter adaptation standards, the transmission rules of the linkage links are adjusted. These rules include: feature transmission format (ensuring the transmitted feature data format matches the model parameter requirements); transmission frequency (adjusting the feature transmission interval based on the model's feature processing speed and requirements); and transmission priority (setting higher transmission priority for features that significantly impact model prediction results to ensure their priority transmission). These adjustments ensure good compatibility between the features transmitted through the linkage links and the parameter requirements of the carbon sink evolution prediction model, thus strengthening the linkage links.

[0085] Step S150: The evolved dynamic coupling of karst carbon sinks is subjected to cross-dimensional feature decoding and correlation trend inference through the carbon sink evolution prediction model to generate karst carbon sink change prediction results.

[0086] In this embodiment, after obtaining the evolved dynamic coupling body of the karst carbon sink, a carbon sink evolution prediction model is used to perform cross-dimensional feature decoding and correlation trend inference to generate karst carbon sink change prediction results. Cross-dimensional feature decoding aims to extract and analyze feature information from different dimensions of the coupling body, while correlation trend inference is based on these feature information to analyze the trend of carbon sink change, thereby achieving prediction of future carbon sink changes.

[0087] Step S151: Input all the initial coupling units, the enhanced correlation links, and the expanded feature interaction channels of the evolved karst carbon sink dynamic coupler into the deep decoding module of the carbon sink evolution prediction model.

[0088] First, all the initial coupling units, enhanced correlation links, and expanded feature interaction channels contained in the evolved karst carbon sink dynamic coupler are transmitted as input to the deep decoding module of the carbon sink evolution prediction model. These inputs contain rich environmental and carbon sink correlation feature information, which forms the basis for cross-dimensional feature decoding and correlation trend inference.

[0089] Step S152: The feature integration submodule of the deep decoding module summarizes, classifies and unifies the temporal change features of all initial coupling units to generate a global feature set, which contains full-dimensional correlation features between the environment and carbon sink.

[0090] After receiving the input initial coupling units, the feature integration submodule of the deep decoding module summarizes the temporal variation features of each initial coupling unit, consolidating the feature information scattered across various coupling units. Then, these features are categorized according to environmental element type, time period, etc., making the features more organized. Next, dimensionality unification processing is performed, transforming features of different categories and dimensions into a unified dimensional space to eliminate the impact of dimensional differences, ultimately generating a global feature set. This global feature set contains full-dimensional correlation features between the environment and carbon sinks, comprehensively reflecting the complex relationship between the two.

[0091] Step S153: The feature parsing submodule of the deep decoding module performs hierarchical decomposition of the global feature set, and obtains feature components of different dimensions according to environmental element category, time period and correlation strength. Each feature component obtained by decomposition retains complete change trajectory information.

[0092] The feature parsing submodule performs hierarchical decomposition of the generated global feature set. First, it splits the global feature set into different feature groups according to environmental element categories (such as atmosphere, hydrology, soil, and biology). Then, within each feature group, it further splits it according to time periods (such as different seasons and months). Finally, based on the different correlation strengths, the features of each time period are split into different feature components. Each feature component obtained from the decomposition retains its complete change trajectory information within the corresponding environmental element category and time period, including the start point, change nodes, change endpoints, and fluctuations during the change process.

[0093] Step S154: Through the association mining submodule of the deep decoding module, cross-dimensional association analysis is performed on the feature components of different dimensions to mine the potential association between environmental element features and carbon sink features, the indirect association between different environmental element features, and the inheritance association of carbon sink features in different time periods, and generate feature association map.

[0094] The association mining submodule performs cross-dimensional association analysis on feature components of different dimensions.

[0095] Step S1541: Extract the core attribute information of each dimension feature component. The core attribute information reflects the essential characteristics, change patterns and influence range of the feature component.

[0096] First, for each dimension's feature components, extract their core attribute information. This core attribute information includes the essential characteristics reflected by the feature components (e.g., atmospheric temperature feature components reflect temperature variation characteristics), their variation patterns (e.g., periodic changes, increasing changes), and their scope of influence (e.g., the degree and extent of their impact on carbon sequestration processes). This information forms the basis for correlation analysis.

[0097] Step S1542: Based on the type and attribute characteristics of the feature components, establish cross-dimensional association analysis rules. The association analysis rules specify the association judgment criteria, association strength calculation methods, and association result classification basis between different types of feature components.

[0098] Based on the type of feature components (such as atmospheric, hydrological, etc.) and their attribute characteristics, cross-dimensional correlation analysis rules are established. These rules clearly define the criteria for judging the correlation between different types of feature components (such as considering a correlation when the similarity of the changing trends of two feature components reaches a certain level), the method for calculating the correlation strength (such as measuring the correlation strength by calculating the correlation coefficient), and the classification basis for the correlation results (such as classifying the correlation into strong, medium, and weak correlations based on the correlation strength).

[0099] Step S1543: According to the cross-dimensional association analysis rules, compare the core attribute information of any two feature components of different dimensions to determine whether there is a direct or indirect association between the two feature components.

[0100] According to the established cross-dimensional correlation analysis rules, the core attribute information of any two feature components from different dimensions is compared. For example, the atmospheric temperature feature component is compared with the carbon sink feature component to analyze whether their variation patterns are similar and whether their influence ranges overlap, thereby determining whether there is a direct correlation between them; or a third feature component (such as the vegetation growth feature component) is used to determine whether there is an indirect correlation between two feature components.

[0101] Step S1544: For a combination of related feature components, calculate the correlation strength of the combination of feature components, classify the correlation level according to the correlation strength, and record the correlation type, correlation level and correlation influence range.

[0102] For combinations of feature components that are determined to be correlated, their correlation strength is calculated according to the correlation strength calculation method specified in the correlation analysis rules. Then, the correlation level is classified according to the magnitude of the correlation strength. For example, correlation strength greater than a certain threshold is classified as strong correlation, correlation strength between two thresholds is classified as medium correlation, and correlation strength less than a certain threshold is classified as weak correlation. At the same time, the correlation type (direct or indirect correlation), correlation level, and correlation impact range (such as the degree of impact on carbon sink changes) of the feature component combination are recorded.

[0103] Step S1545: Integrate the correlation analysis results of all feature component combinations to construct a feature correlation map that includes direct correlation, indirect correlation and correlation level. The feature correlation map fully presents the correlation relationship between all dimension feature components.

[0104] By integrating the correlation analysis results of all feature component combinations, information such as direct and indirect correlations and their corresponding correlation levels is presented in the form of a graph, constructing a feature correlation graph. This feature correlation graph can intuitively and completely display the correlation relationships between feature components of all dimensions.

[0105] Step S155: The trend inference submodule of the deep decoding module analyzes the changing patterns of carbon sink characteristics based on the feature correlation map, infers the changing direction and magnitude of carbon sink characteristics in successive periods, transforms the inference results into a quantitative description, and generates karst carbon sink change prediction results.

[0106] The trend projection submodule analyzes the changing patterns of carbon sink characteristics based on feature correlation maps. By observing the correlation between carbon sink characteristics and other environmental factors, as well as the changes in carbon sink characteristics themselves over different time periods, it summarizes the changing patterns of carbon sink characteristics, such as the periodic changes or increasing / decreasing patterns exhibited by carbon sinks under the combined influence of certain environmental factors. Then, based on these patterns, it projects the direction (e.g., increase, decrease, or stability) and magnitude (e.g., approximate range of change) of carbon sink characteristics in the next period (i.e., a future time period). Finally, it converts the projected direction and magnitude of change into quantitative descriptions, such as using specific textual descriptions of the expected changes in carbon sinks in the future, generating karst carbon sink change prediction results.

[0107] Figure 2 The illustration shows exemplary hardware and software components of a machine learning-based karst carbon sink change prediction system 100 that can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the machine learning-based karst carbon sink change prediction system 100 and to perform the functions in this application.

[0108] The machine learning-based karst carbon sink change prediction system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the machine learning-based karst carbon sink change prediction method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0109] For example, the machine learning-based karst carbon sink change prediction system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the machine learning-based karst carbon sink change prediction system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The machine learning-based karst carbon sink change prediction system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0110] For ease of explanation, only one processor is described in the machine learning-based karst carbon sink change prediction system 100. However, it should be noted that the machine learning-based karst carbon sink change prediction system 100 of this application may also include multiple processors, and therefore the steps performed by one processor described in this application may also be performed jointly by multiple processors or individually. For example, if the processor of the machine learning-based karst carbon sink change prediction system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0111] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned machine learning-based karst carbon sink change prediction method is implemented.

[0112] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1.A method for predicting karst carbon sink change based on machine learning, characterized in that, The method includes: The environmental baseline data and measured carbon sink data of the karst area were obtained. The environmental baseline data included various raw environmental data that affect the carbon sink of karst, and the measured carbon sink data included the actual carbon sink data continuously recorded in the karst area. A dynamic coupling body for karst carbon sinks is constructed based on environmental baseline data and measured carbon sink data. The dynamic coupling body for karst carbon sinks includes real-time correlation links between environmental data and carbon sink data, feature interaction channels, and coupling strength identifiers. Initiate bidirectional interactive modeling between the pre-trained carbon sink evolution prediction model and the dynamic coupling of karst carbon sinks. Drive the feature update of the dynamic coupling of karst carbon sinks by generating feature iteration instructions through the carbon sink evolution prediction model. Correct the parameters of the carbon sink evolution prediction model by the feedback feature change signals from the dynamic coupling of karst carbon sinks, and generate intermediate results of multiple rounds of interactive modeling. Based on the intermediate results of multi-round interactive modeling, the dynamic coupling body of karst carbon sink is strengthened by association link enhancement, feature channel expansion and coupling strength improvement, and the evolved dynamic coupling body of karst carbon sink is obtained. By using a carbon sink evolution prediction model, cross-dimensional feature decoding and correlation trend inference are performed on the evolved dynamic coupling of karst carbon sink to generate karst carbon sink change prediction results. The construction of the dynamic coupling body of karst carbon sink based on environmental baseline data and measured carbon sink data includes: The environmental baseline data is categorized and split to obtain multiple environmental data subsets. Each environmental data subset corresponds to an environmental factor that affects carbon sinks, and each environmental data subset contains continuous record data of the corresponding category factor. The measured carbon sink data is divided into multiple carbon sink data subsets by time axis. Each carbon sink data subset corresponds to a continuous recording period and contains continuous carbon sink data within the corresponding period. Extract the temporal variation characteristics of each environmental data subset. The temporal variation characteristics reflect the change trajectory, change pattern and change correlation of the corresponding environmental elements over the recording period. Extract the temporal variation features of each carbon sink data subset. The temporal variation features reflect the change trajectory, change pattern and connection characteristics of the carbon sink data in the corresponding recording period. The temporal variation characteristics of the environmental data subset corresponding to the same recording period are associated and bound with the temporal variation characteristics of the carbon sink data subset. A coupling strength identifier is assigned to each binding result. All the initial coupling units formed by the binding are connected through real-time association links and feature interaction channels are opened to form a dynamic coupling body for karst carbon sink. 2.The method of claim 1, wherein, The bidirectional interactive modeling of the pre-trained carbon sink evolution prediction model and the dynamic coupling of karst carbon sinks includes: All initial coupling units and corresponding coupling strength identifiers of the dynamic coupling body of karst carbon sink are input into the feature receiving module of the carbon sink evolution prediction model. The initial input features of the model are generated through feature adaptation processing inside the feature receiving module. The hierarchical modeling module of the carbon sink evolution prediction model performs feature transformation, correlation mining and key information extraction on the initial input features of the model, and generates feature iteration instructions that drive the feature update of the dynamic coupling body of karst carbon sink. The feature iteration instructions include feature optimization direction, optimization range and optimization objective. The feature iteration command is transmitted to the dynamic coupler of the karst carbon sink, which triggers the optimization of the temporal variation features of the initial coupler unit, updates the feature details and correlations of the initial coupler unit, and generates the optimized initial coupler unit. The feature change signal of the optimized initial coupling unit is extracted. The feature change signal contains the difference information before and after feature update, the change information of coupling strength identifier and the transmission status information of associated link. The feature change signal is fed back to the parameter correction module of the carbon sink evolution prediction model. The parameter correction module adjusts the internal parameters of the hierarchical modeling module based on the feature change signal. The adjustment range covers the feature processing weight, association mining rules and key information extraction threshold, and generates parameter adjustment results. After multiple iterations, the feature iteration instructions of each round are integrated with the parameter adjustment results to form a multi-round interactive modeling intermediate result. 3.The method of claim 1, wherein, The process of strengthening the correlation links of the dynamic coupling body of karst carbon sink based on intermediate results of multi-round interactive modeling includes: Extract the feature optimization direction information contained in the feature iteration instructions of each round from the intermediate results of multi-round interactive modeling, and integrate all optimization direction information to form a comprehensive optimization requirement; Based on the comprehensive optimization requirements analysis, the transmission status of the associated links between the initial coupling units in the dynamic coupling body of karst carbon sink is analyzed, and weak links in which the transmission efficiency does not meet the optimization requirements are identified. Adjust the link structure of weak links in the associated links, increase the number of feature transmission channels, optimize channel transmission paths, and improve the feature transmission efficiency of the links; Extract model adaptation information from the intermediate results of multi-round interactive modeling, including the parameter adjustment results of each round, and integrate all adaptation information to form a model parameter adaptation standard; Based on the model parameter adaptation standard, the transmission rules of the associated links are adjusted, including the characteristic transmission format, transmission frequency and transmission priority, so that the characteristics of the associated link transmission are adapted to the parameter requirements of the carbon sink evolution prediction model, thus completing the strengthening of the associated links. 4.The method of claim 1, wherein, The method of performing cross-dimensional feature decoding and correlation trend inference on the evolved dynamic coupling of karst carbon sinks using a carbon sink evolution prediction model includes: All initial coupling units, enhanced correlation links, and expanded feature interaction channels of the evolved karst carbon sink dynamic coupler are input into the deep decoding module of the carbon sink evolution prediction model. The feature integration submodule of the deep decoding module summarizes, classifies, and unifies the temporal variation features of all initial coupled units to generate a global feature set, which contains full-dimensional correlation features between the environment and carbon sink. The feature parsing submodule of the deep decoding module performs hierarchical decomposition of the global feature set, and obtains feature components of different dimensions according to environmental element category, time period and correlation strength. Each feature component obtained by the decomposition retains complete change trajectory information. The association mining submodule of the deep decoding module performs cross-dimensional association analysis on feature components of different dimensions, explores the potential association between environmental element features and carbon sink features, the indirect association between different environmental element features and the inheritance association of carbon sink features in different time periods, and generates feature association map. The trend extrapolation submodule of the deep decoding module analyzes the changing patterns of carbon sink characteristics based on feature correlation maps, extrapolates the changing direction and magnitude of carbon sink characteristics in successive periods, transforms the extrapolation results into quantitative descriptions, and generates karst carbon sink change prediction results. 5.The method of claim 1, wherein, The extraction of temporal variation features for each subset of environmental data includes: The continuous recording data of the environmental data subset is continuously segmented to obtain multiple data segments. Each data segment corresponds to a recording sub-period. The time length of the data segment is consistent with the granularity of the recording sub-period of the carbon sink data subset. Analyze the trajectory of environmental data changes in each data segment to determine the starting point, nodes, and endpoints of environmental data changes within the data segment, and extract the data fluctuation characteristics during the change process. The frequency of environmental data changes in each data segment is statistically analyzed. The frequency of change reflects the number of times and the pattern of the intervals between significant changes in environmental data within the data segment. Analyze the characteristics of environmental data change transitions between adjacent data segments to determine the correlation and transition patterns between the endpoint of the change in the preceding data segment and the starting point of the change in the following data segment. The change trajectory, fluctuation characteristics, change frequency and connection characteristics of all data segments corresponding to the environmental data subset are integrated to form a feature sequence. The feature sequence is then subjected to dimensional normalization to obtain the temporal change characteristics of the environmental data subset. 6.The method of claim 2, wherein, The hierarchical modeling module of the carbon sink evolution prediction model performs feature transformation on the initial input features of the model, including: The initial input features of the model are input into the first feature processing layer of the hierarchical modeling module. The feature mapping algorithm of the first feature processing layer transforms the input features into a standard feature vector of uniform dimension, generating standard dimension features. The standard dimensional features are input into the second feature processing layer of the hierarchical modeling module. The correlation mining algorithm of the second feature processing layer analyzes the correlation between different feature items in the standard dimensional features and filters out feature combinations with effective correlation. The selected feature combinations are subjected to feature fusion processing. The information of each feature item in the feature combination is integrated through weighted fusion, highlighting the influence weight of key features, and generating a set of related features. The associated feature set is input into the third feature processing layer of the hierarchical modeling module. The feature enhancement algorithm of the third feature processing layer enhances the expression intensity of key feature components in the associated feature set, suppresses the interference of irrelevant features, and generates enhanced associated features. The instruction generation sub-layer of the enhanced correlation feature input hierarchical modeling module generates preliminary feature update instructions based on the preset optimization objectives of the carbon sink evolution prediction model and the feature requirements of the dynamic coupling body of karst carbon sink. The preliminary instructions are then processed for format adaptation to obtain the final feature iteration instructions. 7.The method of claim 3, wherein, The link structure adjustment for weak links in the associated links includes: Obtain the current structural information of the weak link in the associated link, including the number of transmission channels, channel transmission path, channel bandwidth, and channel delay parameters; Based on the characteristic transmission volume requirements and transmission efficiency standards in the comprehensive optimization needs, the target number of channels, target transmission paths, target bandwidth, and target delay parameters required for link structure adjustment are calculated. Increase the number of transmission channels for weak links according to the target parameters, and form a parallel transmission architecture with the existing channels to realize the redundancy backup function of feature transmission. Optimize the transmission path of the transmission channel, eliminate redundant transmission nodes in the path, shorten the feature transmission distance, and reduce transmission latency; Adjust the bandwidth parameters of each transmission channel to ensure that the channel bandwidth meets the transmission volume requirements of the corresponding characteristics, and ensure that the transmission delay of all channels meets the transmission standard specified by the target delay parameter, thus completing the link structure adjustment. 8.The method of claim 4, wherein, The cross-dimensional correlation analysis of feature components of different dimensions through the correlation mining submodule of the deep decoding module includes: Extract the core attribute information of each dimension's feature component. The core attribute information reflects the essential characteristics, variation patterns, and scope of influence of the feature component. Based on the type and attribute characteristics of feature components, cross-dimensional association analysis rules are established. The association analysis rules specify the association judgment criteria, association strength calculation method and association result classification basis between different types of feature components. According to the cross-dimensional correlation analysis rules, the core attribute information of any two feature components of different dimensions is compared to determine whether there is a direct or indirect correlation between the two feature components. For a combination of related feature components, calculate the correlation strength of the combination, classify the correlation level according to the correlation strength, and record the correlation type, correlation level and correlation influence range. By integrating the correlation analysis results of all feature component combinations, a feature correlation map is constructed that includes direct correlation, indirect correlation, and correlation level. The feature correlation map fully presents the correlation relationships between all dimensional feature components. 9.A machine learning-based karst carbon sink change prediction system, characterized in that, The machine learning-based karst carbon sink change prediction system includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the machine learning-based karst carbon sink change prediction method according to any one of claims 1-8.

Images