A data analysis method based on land resource comprehensive management

Abstract

The application discloses a data analysis method based on land resource comprehensive management; comprising: joint coding time tags, spatial coordinates and attribute tags of multi-source land data into three-dimensional unified expression data set, so that direct distance operation can be carried out on any two data units in time, space and attribute dimensions; reverse tracing abnormal data units along data blood relationship diagram to generate causal chain tracing results containing source, propagation path and range; parallel deducing multiple sets of repair schemes in digital twin model and executing the optimal one; dividing subspaces according to constraint types to independently search candidate schemes, generating decision scheme after merging and conflict resolution; feeding back execution effect to constraint condition and dynamically adjusting constraint strength. The application realizes cross-dimension joint operation, abnormal source diagnosis, scheme virtual deduction and constraint continuous evolution, and forms a whole-process closed loop from data perception to decision optimization.

Description

Technical Field

[0001] This invention relates to the field of land resource management technology, and specifically to a data analysis method based on comprehensive land resource management. Background Technology

[0002] Land resources are fundamental to human survival and development. Scientific and efficient land resource management is crucial for ensuring food security, promoting ecological civilization, and driving sustainable economic and social development. With the rapid development of remote sensing, IoT sensing, and geographic information systems, the field of land resource management has accumulated massive amounts of multi-source heterogeneous data. How to fully mine and utilize this data, and realize the transformation of land resource management from static monitoring to dynamic perception, and from experience-driven to data-driven approaches, has become a critical issue that urgently needs to be addressed. Currently, some patented technologies involve data analysis and management of land resources, such as processing spatial information of land reserves based on multi-source data, constructing intelligent analysis systems for intensive land use, and conducting spatiotemporal change analysis using geographic vector plots as the smallest spatial unit. These technologies have explored data fusion and visualization to some extent.

[0003] However, existing technologies still have the following significant shortcomings: First, the processing of time-labeled data, spatial coordinate data, and attribute feature data is usually done in a "serial" manner, processing them separately and then associating and combining them. The three types of data are always in a fragmented state and cannot be jointly calculated and analyzed across dimensions within the same framework. Second, the processing of outlier data is limited to imputation and replacement based on similarity, focusing only on "how to fill out outliers" without the ability to trace the causal relationship of the source, propagation path, and scope of impact of the outlier. The repair remains at the "symptom-treating" level and cannot achieve the "root cause" solution. Third, the application of digital twin technology is limited to the level of three-dimensional visualization and has not yet realized its functional potential in parallel simulation of multiple schemes and the verification of the best effect. Fourth, land use optimization decisions are mostly one-time static calculations, lacking an adaptive adjustment mechanism that feeds back the implementation effect to the constraints and drives the dynamic evolution of the optimization target.

[0004] To address the shortcomings of existing technologies, this invention provides a data analysis method based on integrated land resource management. This method constructs a unified three-dimensional representation framework of time, space, and attributes; establishes a reverse tracing mechanism for the causal chain of abnormal data; performs multi-scheme simulation and optimization in a digital twin; and dynamically feeds back execution feedback to constraints for continuous evolution. This forms a closed-loop process from data perception, anomaly diagnosis, intelligent repair to decision optimization and feedback evolution, aiming to improve the intelligence level and scientific nature of integrated land resource management. Summary of the Invention

[0005] To address the shortcomings of existing land resource management data analysis technologies, such as the inability to jointly process fragmented multi-source data in three dimensions, the lack of source causal tracing in abnormal data processing (remaining only at the level of interpolation and replacement), the limitation of digital twin applications to visualization and the inability to perform parallel simulation and verification of multiple schemes, and the lack of an adaptive adjustment mechanism driven by execution feedback in land use optimization decisions, this invention provides a data analysis method based on comprehensive land resource management. The aim is to achieve closed-loop management of multi-source land data throughout the entire process, from unified three-dimensional representation, anomaly causal tracing, digital twin simulation and repair to execution feedback-driven processes, thereby improving the intelligence level and scientific nature of comprehensive land resource management.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a data analysis method based on comprehensive land resource management, comprising the following steps:

[0007] Step S1: Construct a unified three-dimensional representation dataset of time, space, and attributes. Acquire multi-source land data for the target area, which includes time labels, spatial coordinates, and attribute labels. Jointly encode the time labels, spatial coordinates, and attribute labels to generate a unified three-dimensional representation dataset. Specifically, construct a three-dimensional data space with time axis, spatial coordinate axis, and attribute axis as base coordinates. Map each multi-source land data point to a data point in the three-dimensional data space, with each data point carrying time dimension coordinate values, spatial dimension coordinate values, and attribute dimension vector values. Define the comprehensive distance between any two data points in the three-dimensional data space as the weighted sum of the time dimension distance, spatial dimension distance, and attribute dimension distance. Use this comprehensive distance as the result of direct distance calculations between any two data units in the time, spatial, and attribute dimensions. Output the set of all mapped data points as the unified three-dimensional representation dataset. Through this unified three-dimensional representation framework, heterogeneous data of time, space, and attributes are unified in the same geometric representation system. Direct distance calculations can be performed between any two data points in all three dimensions, providing a data foundation for subsequent cross-dimensional joint analysis.

[0008] Step S2: Reverse tracing of abnormal causal chains based on the data lineage graph. Anomaly detection is performed on the three-dimensional unified representation dataset to identify abnormal data units that deviate from the normal fluctuation range. For each abnormal data unit, reverse tracing is performed along the data propagation path to locate the source data unit, identify the propagation path nodes, and delineate the affected data units, generating an abnormal causal chain tracing result. Specifically, a data lineage graph for the target area is pre-constructed. This graph records the source, processing and transformation process, and output destination of each data unit. The source includes the data acquisition device identifier, acquisition timestamp, and acquisition location. The processing and transformation process includes at least one operation among data cleaning, data interpolation, data fusion, and data conversion. The output destination includes the downstream data processing module identifier and data storage location. When an abnormal data unit is detected, the data lineage graph is traversed backwards from the abnormal data unit, marking each predecessor node sequentially until the original acquisition node without predecessor nodes is reached. The original acquisition node is then identified as the source data unit. All nodes and edges traversed during the reverse traversal are recorded as the propagation path node sequence. Data units directly or indirectly affected by all nodes in the propagation path node sequence are marked as the affected data unit range. Finally, the source data unit, the propagation path node sequence, and the affected data unit range are combined to form the anomaly causal chain tracing result. This reverse tracing mechanism based on the data lineage graph transforms anomaly handling from traditional "imputation and replacement" to a full-chain causal diagnosis of "source location—path tracing—range delineation," providing precise targeting for subsequent radical remediation.

[0009] Step S3: Parallel Simulation and Optimal Repair Based on Digital Twin. Based on the anomaly causal chain tracing results, a virtual land model corresponding to the physical land of the target area is constructed in the digital twin space. In the virtual land model, at least two repair schemes are generated for the source data unit. Simulations are performed on each repair scheme, and a selected repair scheme is chosen based on the simulation results. This chosen repair scheme is applied to all data units covered by the anomaly causal chain tracing results, generating a repaired three-dimensional unified representation dataset. Specifically, based on the anomaly type of the source data unit, at least two different repair operation directions are selected from the repair strategy library. These different strategy directions include root cause elimination strategies and impact compensation strategies. For each strategy direction, a complete repair operation sequence is generated, and each repair operation sequence is output as a repair scheme. Each set of repair operation sequences is executed in the virtual land model, simulating the data changes caused by the repair operation sequences to all data units covered by the abnormal causal chain tracing results. The comprehensive state evaluation value of the virtual land model after the execution of each set of repair operation sequences is recorded as the simulation result corresponding to that set of repair schemes. The comprehensive state evaluation value is calculated by weighting multiple evaluation indicators, including the data completeness, data consistency, and data prediction accuracy after repair. The higher the comprehensive state evaluation value, the better the repair effect. The comprehensive state evaluation values ​​corresponding to each set of repair schemes are compared, and the set of repair operation sequences with the highest comprehensive state evaluation value is selected as the selected repair scheme. The selected repair scheme is applied to all data units covered by the abnormal causal chain tracing results to generate a three-dimensional unified representation dataset after repair. The above method upgrades digital twins from traditional "visualization tools" to "multi-scheme parallel simulation and optimal verification platform," enabling repair schemes to complete effect pre-evaluation in virtual space before optimal execution, greatly improving the scientificity and reliability of the repair.

[0010] Step S4: Intelligent Land Use Decision-Making Based on Constraint Subspace Division and Conflict Resolution. Obtain a multi-dimensional constraint set, including soil physicochemical property constraints, topographic constraints, climate constraints, locational economic constraints, and policy planning constraints. Associate and map the repaired 3D unified representation data set with the multi-dimensional constraint set to generate a constraint-tagged 3D data set. Divide the constraint-tagged 3D data set into multiple constraint subspaces according to the constraint tags. Search for candidate land use schemes that satisfy the corresponding constraints in each constraint subspace to obtain a candidate scheme set for each constraint subspace. Merge and resolve conflicts in the candidate scheme sets of each constraint subspace to generate an output decision scheme that satisfies all multi-dimensional constraints and has the optimal comprehensive benefit index. Specifically, based on the constraint type identifier contained in the constraint tags, data units with the same constraint type identifier are grouped into the same constraint subspace, with each constraint subspace corresponding to one constraint type. For each constrained subspace, a local search strategy corresponding to that constrained type is adopted to generate at least one candidate land use scheme that satisfies all constraints within that constrained subspace. The generated set of candidate land use schemes is taken as the candidate scheme set for that constrained subspace. The candidate scheme sets of all constrained subspaces are combined using a Cartesian product to form multiple cross-subspace combination schemes. For each cross-subspace combination scheme, constraint conflicts between different constrained subspaces are detected. These conflicting constraints are adjusted according to a preset conflict resolution rule to generate conflict-free global candidate schemes. The preset conflict resolution rule includes: when two candidate schemes from different constrained subspaces generate different land use assignments on the same data unit, the matching degree of the attribute feature vector of that data unit in the three-dimensional unified representation dataset with the constraint conditions corresponding to each constrained subspace is compared, and the assignment with the highest matching degree is taken as the final assignment after conflict resolution. From all conflict-free global candidate schemes, the scheme that satisfies all multi-dimensional constraints and has the optimal comprehensive benefit index is selected as the output decision scheme. The aforementioned "divide and conquer-integrate" optimization strategy differs from the conventional paradigm of constructing a global objective function and solving it all at once in existing technologies. By dividing the constraint subspace, solving locally independently, and resolving conflicts across subspaces in a step-by-step process, it effectively reduces the solution complexity of high-dimensional constraint optimization problems and improves the efficiency and quality of decision generation.

[0011] Step S5: Perform feedback-driven dynamic adjustment and continuous evolution of constraint strength. Output the output decision scheme to the execution end and collect the actual execution effect data generated during the execution process. Feedback the actual execution effect data to the multidimensional constraint set, adjust the constraint strength of each constraint in the multidimensional constraint set according to the actual execution effect data, and generate an updated multidimensional constraint set for the next data analysis cycle. Specifically, calculate the deviation between the actual execution effect data and the predicted effect data of the output decision scheme before execution. The deviation value includes positive and negative deviations. When the deviation value is positive, it indicates that the actual execution effect is better than the predicted effect. In this case, constraints that contribute positively are positively adjusted to strengthen their constraint strength, and constraints that contribute negatively are negatively adjusted to weaken their constraint strength. When the deviation value is negative, constraints that contribute positively are negatively adjusted to weaken their constraint strength, and constraints that contribute negatively are positively adjusted to strengthen their constraint strength. Based on the magnitude and direction of the deviation value, an adjustment coefficient is determined for each constraint condition, which can be a positive or negative adjustment value. The current constraint strength of each constraint condition is multiplied by the corresponding adjustment coefficient to obtain the updated constraint strength. All updated constraint strengths are then recombine to generate an updated multidimensional constraint set. This method ensures that the optimization objective and constraints are no longer preset and fixed, but continuously evolve with each execution feedback—the system "learns" in each loop which constraints are hard and which objectives are adjustable, allowing the starting point of the next optimization to be better than the previous one, thus achieving adaptive evolution of the optimization framework.

[0012] The five steps described above are interconnected and progressive: the three-dimensional unified representation data set generated in step S1 provides the data foundation for cross-dimensional joint analysis in the reverse tracing of step S2; the abnormal causal chain tracing results in step S2 provide a precise targeting range for the digital twin repair in step S3; the repaired data set in step S3 provides high-quality data input for the application optimization in step S4; the output decision scheme in step S4 drives the self-adjustment of constraint strength in the execution feedback in step S5; and the updated constraint set in step S5 is fed back to step S4, forming a complete continuous evolution closed loop.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] This invention constructs a three-dimensional data space with time, spatial, and attribute axes as base coordinates. Each data point is mapped to a data point carrying three-dimensional coordinates, and a comprehensive distance weighted sum is defined as the direct distance calculation result. This fundamentally breaks through the existing serial processing paradigm of "processing three types of data separately and then associating and combining them," making cross-dimensional joint querying and analysis a native operation. Simultaneously, this invention pre-constructs a data lineage diagram recording the source, processing transformation process, and output destination of each data unit. When an anomaly occurs, it traverses backward from the anomaly node to accurately locate the source, completely identify the propagation path, and scientifically delineate the scope of impact, forming a complete causal chain tracing result. This transforms anomaly handling from "filling in the blanks according to similar samples" to "rooting out the problem according to the causal chain." Based on this, this invention upgrades digital twins from traditional visualization tools to a multi-scheme parallel simulation and verification platform. By generating at least two repair schemes with different strategy directions and simulating their execution effects in the twin before selecting the optimal one, it achieves a closed-loop repair process of "virtual preliminary verification, real-world optimal execution."

[0015] This invention employs a "divide and conquer" strategy in multi-constraint land use optimization decision-making. It divides the space into subspaces based on constraint type and performs independent local searches within each subspace. A global solution is then generated through Cartesian product combination and conflict resolution based on attribute feature vector matching. This effectively breaks through the conventional paradigm of constructing a global objective function and solving it in a single step, significantly reducing the complexity of solving high-dimensional constraint optimization problems. Furthermore, this invention feeds back the actual effect data after each decision is implemented to the constraints themselves. By calculating the deviation between the actual and predicted effects, the constraint strength of each constraint is dynamically adjusted differentially—constraints contributing positively are strengthened, and those contributing negatively are weakened—allowing the optimization objective and constraints to continuously evolve with execution feedback, achieving a leap from one-time static optimization to continuous adaptive evolution. These five steps are interconnected and progressive, forming a complete logical loop through the constraint update mechanism driven by execution feedback. This achieves fully automated closed-loop management of the entire process, from data perception, anomaly diagnosis, intelligent repair, decision optimization to continuous evolution. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of a data analysis method based on comprehensive land resource management according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, the present invention provides a data analysis method based on comprehensive land resource management, which includes the following steps S1 to S5.

[0019] Step S1: Obtain multi-source land data for the target area, and jointly encode the time labels, spatial coordinates, and attribute labels to generate a three-dimensional unified representation data set.

[0020] In this embodiment, multi-source land data for the target area is first acquired. This multi-source land data includes time-series monitoring data with time tags, spatial distribution data with spatial coordinates, and attribute feature data with attribute tags. Specifically, the time-series monitoring data may include multi-period remote sensing imagery, monitoring data continuously collected by ground sensors, and historical land use records. Each set of time-series monitoring data is accompanied by a clear time tag to identify the data's acquisition time or time interval. The spatial distribution data may include geographic coordinate data, topographic data, administrative boundary data, and land parcel vector data. Each set of spatial distribution data is accompanied by clear spatial coordinates to identify the data's specific location in geographic space. The attribute feature data may include soil physicochemical property data, hydrological condition data, climate factor data, and land cover type data. Each set of attribute feature data is accompanied by a clear attribute tag to identify the type of land resource characteristic described by the data.

[0021] After acquiring the aforementioned multi-source land data, the time labels, spatial coordinates, and attribute labels are jointly encoded to generate a unified three-dimensional representation dataset. The specific implementation method is as follows:

[0022] A three-dimensional data space is constructed using a time axis, spatial coordinate axes, and attribute axes as base coordinates. The time axis represents the temporal dimension information of the data, the spatial coordinate axes represent the spatial location information of the data, and the attribute axes represent the attribute feature information of the data. Each multi-source land data point is mapped to a data point in the three-dimensional data space. The mapped data point carries time dimension coordinate values, spatial dimension coordinate values, and attribute dimension vector values. The time dimension coordinate values ​​come from the original time label carried by the data point, the spatial dimension coordinate values ​​come from the original spatial coordinates carried by the data point, and the attribute dimension vector values ​​come from the specific attribute values ​​corresponding to the original attribute labels carried by the data point.

[0023] The comprehensive distance between any two data points in a 3D data space is defined as the weighted sum of the time dimension distance, spatial dimension distance, and attribute dimension distance. This comprehensive distance is used as the result of direct distance calculations between the two data units in the time, spatial, and attribute dimensions. Specifically, the time dimension distance is the absolute value of the coordinate difference between the two data points on the time axis; the spatial dimension distance is the Euclidean distance between the two data points on the spatial coordinate axes; and the attribute dimension distance is the magnitude of the vector difference between the two data points on the attribute axes. The comprehensive distance is obtained by multiplying the time dimension distance, spatial dimension distance, and attribute dimension distance by their respective weighting coefficients and then summing them.

[0024] In one specific embodiment, the weighting coefficient for the time dimension distance is set to 0.3, the weighting coefficient for the spatial dimension distance is set to 0.4, and the weighting coefficient for the attribute dimension distance is set to 0.3, with the sum of the three weighting coefficients being 1. In application scenarios that focus on the spatiotemporal evolution patterns of land resources, the weighting coefficients for the time dimension distance can be adjusted to 0.4, the spatial dimension distance to 0.4, and the attribute dimension distance to 0.2. In application scenarios that focus on the differences in the attribute characteristics of land resources, the weighting coefficients for the attribute dimension distance can be adjusted to 0.5, the time dimension distance to 0.2, and the spatial dimension distance to 0.3. The specific values ​​of the weighting coefficients can be flexibly configured according to the actual application scenario and management needs.

[0025] All mapped data points are output as a unified three-dimensional representation data set. Through this unified three-dimensional representation framework, heterogeneous data of time, space, and attributes are unified in the same geometric representation system. Direct distance calculations can be performed between any two data points in all three dimensions, providing a unified data foundation for subsequent anomaly detection, cross-dimensional joint analysis, and intelligent decision-making.

[0026] Step S2: Perform anomaly detection on the three-dimensional unified representation dataset to identify anomalous data units; perform reverse tracing along the data propagation path to generate anomaly causal chain tracing results.

[0027] In this embodiment, anomaly detection is first performed on the 3D unified representation dataset generated in step S1. Anomaly detection can be performed using a statistical threshold-based method. In one specific embodiment, the 3σ principle is used for anomaly detection: for each data unit in the 3D unified representation dataset, the comprehensive distance distribution between the data unit and its neighboring data units is calculated. Using the average of all comprehensive distances as the center and three times the standard deviation as the threshold range, when the comprehensive distance between a data unit and its neighboring data units exceeds this threshold range, the data unit is marked as an anomaly. The selection range of neighboring data units can be set according to the data density; for example, the eight data points closest to the data point on the spatial coordinate axis can be used as neighboring data units. In another embodiment, anomaly detection can also use a machine learning method based on isolated forests. Multiple isolated trees are constructed to score data units for anomalies, and those scoring above a preset threshold are marked as anomaly data units.

[0028] After identifying abnormal data units, reverse tracing is performed along the data propagation path for each abnormal data unit. To achieve reverse tracing, a data lineage graph for the target area needs to be pre-constructed. A data lineage graph is a directed graph structure where each node represents a data unit, and each directed edge represents the source, processing, or output relationship between data units. The data lineage graph records the source, processing, and output of each data unit. The source includes the data acquisition device identifier, acquisition timestamp, and acquisition location, indicating which device acquired the data unit at what time and place. The processing includes at least one operation among data cleaning, data interpolation, data fusion, and data transformation, indicating which data processing steps the data unit underwent. The output includes the downstream data processing module identifier and data storage location, indicating which downstream modules the data unit was transmitted to or where it was stored.

[0029] When an abnormal data unit is detected, a reverse traversal is performed starting from the abnormal data unit in the data lineage graph. The reverse traversal direction is along the opposite direction of the directed edges, i.e., tracing back from the current node to its predecessor node. Each predecessor node is marked sequentially until the original acquisition node without a predecessor node is reached. The original acquisition node is a node without a predecessor node in the data lineage graph, representing the original source of the data—that is, the raw data directly collected by the data acquisition device, not derived data generated after any processing or transformation. The original acquisition node is positioned as the source data unit of the abnormal data unit.

[0030] All nodes and directed edges traversed during the reverse traversal are recorded as a propagation path node sequence. This sequence fully records the complete path information of the anomalous data unit, from its originating data unit, through intermediate processing steps and data transformations, ultimately reaching the current anomalous data unit. Data units directly or indirectly affected by all nodes in the propagation path node sequence are marked as the affected data unit range. This range includes not only the nodes themselves but also all downstream nodes pointed to by these nodes via directed edges—in other words, all data units affected by the anomalous source disturbance during data propagation.

[0031] Finally, the source data unit, the propagation path node sequence, and the affected data unit range are combined to form the anomaly causal chain tracing result. The anomaly causal chain tracing result fully answers three core questions: Where did the anomaly come from (source data unit), where did the anomaly pass through (propagation path node sequence), and who was affected by the anomaly (affected data unit range).

[0032] Step S3: Based on the results of the abnormal causal chain tracing, construct a virtual land model in the digital twin space; generate at least two repair schemes in the virtual land model and conduct simulations on them respectively; select the optimal repair scheme based on the simulation results and execute it to generate a set of three-dimensional unified representation data after repair.

[0033] In this embodiment, firstly, based on the abnormal causal chain tracing results generated in step S2, a virtual land model corresponding to the physical land of the target area is constructed in the digital twin space. The virtual land model is a high-precision mapping of the physical land of the target area in the digital space, containing information such as geographic coordinates, topography, soil properties, and current land use that are consistent with the physical land. The construction of the virtual land model can employ three-dimensional geographic information system technology to reconstruct the multi-source land data of the target area in three dimensions in the digital space, forming an interactive and simulateable virtual mirror.

[0034] After the virtual land model is constructed, at least two remediation plans are generated for each source data unit. The specific implementation method is as follows:

[0035] Based on the anomaly type of the source data unit, at least two different repair strategies are selected from the repair strategy library. The repair strategy library is a pre-established knowledge base containing various anomaly types and their corresponding repair strategies. The method for constructing the repair strategy library is as follows: for common data anomaly types in the field of land resource management—including missing data anomalies, jump data anomalies, drift data anomalies, and logical data conflict anomalies—at least one corresponding repair operation template is configured for each anomaly type. The repair strategy for missing data anomalies is an interpolation operation template based on the spatiotemporal neighborhood mean; the repair strategy for jump data anomalies is a filtering operation template based on sliding window smoothing; the repair strategy for drift data anomalies is a drift reduction operation template based on the reference baseline; and the repair strategy for logical data conflict anomalies is a consistency correction operation template based on association rules.

[0036] Different strategies include root cause elimination strategies and impact compensation strategies. The core idea of ​​root cause elimination strategies is to directly address the source of the anomaly—the source data unit—by correcting acquisition errors, filling in missing values, or rectifying erroneous values ​​in the source data, thus eliminating the cause of the anomaly at its root. The core idea of ​​impact compensation strategies is not to directly modify the source data, but rather to offset the impact of anomaly propagation by applying compensation operations at various nodes along the anomaly propagation path—for example, adding correction factors in the data processing stage, adjusting fusion weights in the data fusion stage, or correcting deviations in the data output stage.

[0037] For each strategy direction, a complete repair operation sequence is generated. A repair operation sequence is a collection of data repair operations executed in a specific order; each repair operation includes an operation type, operation parameters, and operation object. Each repair operation sequence is output as a repair plan.

[0038] After generating at least two remediation plans, simulations are performed on each plan. The simulations are executed within a virtual land model, and the specific implementation is as follows:

[0039] Each set of remediation operation sequences is executed in the virtual land model, simulating the data changes caused by the remediation operation sequence to all data units covered by the anomaly causal chain tracing results. Each data unit in the virtual land model corresponds one-to-one with an actual data unit in the physical land; therefore, executing the remediation operation sequence in the virtual land model is equivalent to simulating the effect of the remediation plan in the physical world within a virtual environment. The comprehensive state evaluation value of the virtual land model after executing each set of remediation operation sequences is recorded as the simulation result corresponding to that remediation plan.

[0040] The comprehensive status assessment value is calculated by weighting multiple assessment indicators. These indicators include the completeness of the restored data, data consistency, and data prediction accuracy. Data completeness measures the degree to which missing values ​​are filled in the restored dataset—fewer missing values ​​and more complete data indicate higher data completeness. Data consistency measures the logical consistency between different data units in the restored dataset across time, space, and attribute dimensions—fewer contradictions and greater logical self-consistency indicate higher data consistency. Data prediction accuracy measures the accuracy of land resource status predictions based on the restored dataset—the closer the prediction results are to the actual observation results, the higher the prediction accuracy. In one specific embodiment, the weighting coefficients for data completeness, data consistency, and data prediction accuracy are set to 0.3, 0.3, and 0.4, respectively, and the comprehensive status assessment value is the weighted sum of these three. A higher comprehensive status assessment value indicates a better restoration effect.

[0041] After obtaining the comprehensive state assessment values ​​corresponding to each set of repair schemes, the comprehensive state assessment values ​​of each set of repair schemes are compared, and the repair operation sequence with the highest comprehensive state assessment value is selected as the selected repair scheme. The selected repair scheme is applied to all data units covered by the abnormal causal chain tracing results—that is, it is applied to all data units in the source data unit, all nodes in the propagation path node sequence, and all data units in the affected data unit range simultaneously—to generate a three-dimensional unified representation data set after repair.

[0042] Through the above methods, digital twins have been upgraded from traditional "visualization tools" to "multi-scheme parallel simulation and optimization verification platforms". The repair schemes are pre-evaluated in virtual space before being optimized for execution, avoiding the trial-and-error costs and data risks that may be caused by directly executing the repair schemes in the physical world.

[0043] Step S4: Obtain the multidimensional constraint set, associate and map the repaired three-dimensional unified expression data set with the multidimensional constraint set to generate a three-dimensional data set with constraint labels; divide the data into multiple constraint subspaces according to the constraint labels, search for candidate land use schemes in each constraint subspace and merge and resolve conflicts to generate an output decision scheme.

[0044] In this embodiment, a multi-dimensional constraint set for the target area is first obtained. This set includes constraints on soil physicochemical properties, topography, climate, location and economics, and policy and planning. Soil physicochemical property constraints include limitations such as soil pH range, soil organic matter content threshold, and soil texture type, defining the basic requirements for soil physicochemical properties for a given land use. Topography constraints include limitations such as slope range, elevation range, and aspect conditions, defining the basic requirements for topography for a given land use. Climate constraints include limitations such as average annual precipitation range, average annual temperature range, and frost-free period length, defining the basic requirements for climate conditions for a given land use. Location and economic constraints include limitations such as distance from major transportation routes, distance from the city center, and land economic value range, defining the locational and economic feasibility of a given land use. Policy and planning constraints include limitations such as land use zoning determined by the overall land use plan, ecological protection red line boundaries, and basic farmland protection zones, ensuring that the land use plan complies with national and local policies and regulations.

[0045] The repaired 3D unified representation dataset generated in step S3 is then associated with the multidimensional constraint set. The specific method of association mapping is as follows: for each data unit in the repaired 3D unified representation dataset, each constraint in the multidimensional constraint set is traversed, and it is determined whether the data unit satisfies the constraint. If it does, a constraint marker corresponding to the constraint is added to the data unit; otherwise, no constraint marker is added. After association mapping, a 3D dataset with constraint markers is generated.

[0046] After generating a 3D dataset with constraint labels, the dataset is divided into multiple constraint subspaces according to the constraint labels. The specific implementation is as follows:

[0047] Based on the constraint type identifier contained in the constraint tags, data units with the same constraint type identifier are grouped into the same constraint subspace. The constraint type identifier is used to distinguish different categories of constraints, including "Soil," "Topography," "Climate," "Economy," and "Policy" constraints. Each constraint subspace corresponds to one constraint type. All data units carrying the "Soil" constraint tag are grouped into the soil constraint subspace, all data units carrying the "Topography" constraint tag are grouped into the topography constraint subspace, all data units carrying the "Climate" constraint tag are grouped into the climate constraint subspace, all data units carrying the "Economy" constraint tag are grouped into the economy constraint subspace, and all data units carrying the "Policy" constraint tag are grouped into the policy constraint subspace.

[0048] After dividing the constraint subspace, candidate land use schemes that satisfy the corresponding constraints in each constraint subspace are searched. The specific implementation is as follows:

[0049] For each constraint subspace, a local search strategy corresponding to that constraint type is employed to generate at least one candidate land use scheme that satisfies all constraints within that subspace. The local search strategies differ for different constraint types. For the soil constraint subspace, the local search strategy is a rule-matching method based on soil suitability evaluation—selecting the top three candidate uses based on the suitability matching relationship between soil physicochemical properties and different land uses. For the topography constraint subspace, the local search strategy is a spatial analysis method based on topography analysis—selecting candidate uses that meet the topographic requirements based on the suitability matching relationship between topographic factors such as slope and elevation and different land uses. For the policy constraint subspace, the local search strategy is a compliance check method based on a rule engine—selecting compliant candidate uses based on the land use control rules in policy planning documents. The generated set of candidate land use schemes is then used as the candidate scheme set for that constraint subspace.

[0050] After obtaining the candidate solution sets for each constraint subspace, the candidate solution sets for each constraint subspace are merged and conflict resolved. The specific implementation is as follows:

[0051] The candidate solutions from all constrained subspaces are combined using Cartesian products. A Cartesian product combination involves selecting one candidate solution from each constrained subspace and combining them into a cross-subspace combination solution. When the soil constrained subspace has 3 candidate solutions, the topography constrained subspace has 2 candidate solutions, and the climate constrained subspace has 2 candidate solutions, the Cartesian product combination results in 3 × 2 × 2 = 12 cross-subspace combination solutions.

[0052] For each cross-subspace combination scheme, constraint conflicts are detected between different constraint subspaces. A constraint conflict refers to a situation where candidate schemes from different constraint subspaces assign different land use values ​​to the same data unit. A constraint conflict occurs when a candidate scheme in the soil constraint subspace marks a plot as "arable land," while a candidate scheme in the policy constraint subspace marks the same plot as "ecological land."

[0053] Constraint conflict terms are adjusted according to preset conflict resolution rules to generate conflict-free global candidate solutions. The preset conflict resolution rules include: when candidate solutions from two constraint subspaces generate different land use assignments on the same data unit, the matching degree between the attribute feature vector of that data unit in the three-dimensional unified representation dataset and the corresponding constraint conditions of each constraint subspace is compared, and the assignment with the highest matching degree is taken as the final assignment after conflict resolution. The attribute feature vector is the set of all attribute values ​​of that data unit in the attribute dimension, and the matching degree is a quantitative measure of the degree of fit between the attribute feature vector and a certain constraint condition. When the soil attribute feature vector of a certain plot has a matching degree of 0.85 with the soil conditions for "arable land" use, but a matching degree of 0.60 with the soil conditions for "ecological land" use, "arable land" is taken as the final assignment after conflict resolution.

[0054] The optimal solution, satisfying all multidimensional constraints and possessing the best comprehensive benefit index, is selected from all conflict-free global candidate solutions and designated as the output decision solution. The comprehensive benefit index is a weighted sum of economic, ecological, and social benefit indices. In one specific embodiment, the economic benefit index is the output value per unit area of ​​land, the ecological benefit index is the ecosystem service value, and the social benefit index is the employment-generating effect. The weighting coefficients for these three indices are set to 0.4, 0.3, and 0.3, respectively.

[0055] By employing the aforementioned "divide and conquer-integrate" optimization strategy, the solution complexity of high-dimensional constrained optimization problems is effectively reduced, and the efficiency and quality of decision generation are improved.

[0056] Step S5: Output the decision scheme to the execution end and collect the actual execution effect data; feed the actual execution effect data back to the multi-dimensional constraint set, adjust the constraint strength of each constraint, and generate an updated multi-dimensional constraint set for the next data analysis cycle.

[0057] In this embodiment, the output decision scheme generated in step S4 is first output to the land resource management execution terminal. The execution terminal is the information system of the land resource management department, the land use planning implementation platform, or the construction management system of land consolidation projects. The output decision scheme is provided to the execution terminal in the form of a visual map, data report, or operation instructions to guide actual land resource management actions.

[0058] During the implementation of the output decision-making plan, data on the actual implementation effects are continuously collected. This data includes actual land output data, actual ecological indicator data, and actual social feedback data. Actual land output data refers to actual crop yields or actual income from economic forests; actual ecological indicator data includes changes in soil quality, vegetation cover, or biodiversity; and actual social feedback data includes community satisfaction scores or the number of jobs created. This data is obtained through a variety of methods, including ground monitoring, remote sensing observation, and socio-economic surveys.

[0059] The actual execution results data is fed back to the multi-dimensional constraint set, and the constraint strength of each constraint in the multi-dimensional constraint set is adjusted based on the actual execution results data. The specific implementation method is as follows:

[0060] Calculate the deviation between the actual execution effect data and the predicted effect data of the output decision scheme before execution. The predicted effect data is the expected effect data derived from historical data and model predictions when generating the output decision scheme in step S4. The deviation value equals the actual execution effect data minus the predicted effect data. The deviation value includes positive and negative deviations. When the actual execution effect data is greater than the predicted effect data, the deviation value is positive, indicating that the actual execution effect is better than the predicted effect; when the actual execution effect data is less than the predicted effect data, the deviation value is negative, indicating that the actual execution effect is worse than the predicted effect.

[0061] Based on the magnitude and direction of the deviation value, an adjustment coefficient is determined for each constraint. The adjustment coefficient is either a positive or negative adjustment value; a positive adjustment value is greater than 1, and a negative adjustment value is a positive number less than 1. The rule for determining the adjustment coefficient is: Adjustment Coefficient = 1 + Deviation Rate × Preset Coefficient, where the deviation rate is the ratio of the deviation value to the predicted data, and the preset coefficient is a positive number. When the deviation rate is positive, the adjustment coefficient is a positive adjustment value; when the deviation rate is negative, the adjustment coefficient is a negative adjustment value. The preset coefficient ranges from 0.1 to 0.5, and the specific value is determined based on the confidence level of the constraint; the higher the confidence level, the smaller the preset coefficient.

[0062] When the deviation is positive, constraints that contribute positively are adjusted to strengthen their binding force, while constraints that contribute negatively are adjusted to weaken their binding force. Constraints that contribute positively are those that perform better than expected during the implementation of the decision-making scheme and positively promote good implementation results; constraints that contribute negatively are those that perform worse than expected during the implementation of the decision-making scheme and hinder good implementation results. When the deviation is negative, constraints that contribute positively are adjusted negatively to weaken their binding force, while constraints that contribute negatively are adjusted positively to strengthen their binding force.

[0063] The updated constraint strength is obtained by multiplying the current constraint strength of each constraint by the corresponding adjustment factor. When the current constraint strength of a certain constraint is 0.8 and the corresponding adjustment factor is 1.1, the updated constraint strength is 0.88.

[0064] All updated constraint strengths are recombined to generate an updated set of multidimensional constraints for the next data analysis cycle. This updated set of multidimensional constraints will then serve as input for step S4, participating in the next round of land use optimization decisions.

[0065] Through the aforementioned dynamic adjustment mechanism of constraint strength driven by execution feedback, the optimization objectives and constraints are no longer preset and fixed, but evolve continuously with each execution feedback. In each loop, the system "learns" which constraints are hard and which objectives are adjustable, and the starting point of the next optimization is better than the previous one, thus realizing the adaptive evolution of the optimization framework.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data analysis method based on comprehensive land resource management, characterized in that, Includes the following steps: Step S1: Obtain multi-source land data of the target area, the multi-source land data including time labels, spatial coordinates and attribute labels; jointly encode the time labels, spatial coordinates and attribute labels to generate a three-dimensional unified representation data set, in which any two data units in the three-dimensional unified representation data set can perform direct distance calculations in the time dimension, spatial dimension and attribute dimension; Step S2: Perform anomaly detection on the three-dimensional unified representation dataset to identify anomalous data units; for each anomalous data unit, perform reverse tracing along the data propagation path to locate the source data unit of the anomalous data unit, identify the propagation path nodes of the anomalous data unit, and delineate the range of affected data units influenced by the anomalous data unit, generating anomaly causal chain tracing results; Step S3: Based on the abnormal causal chain tracing results, construct a virtual land model in the digital twin space; in the virtual land model, generate at least two repair schemes for the source data unit, simulate and deduce each repair scheme, select a repair scheme as the selected repair scheme based on the simulation and deduction results, and apply the selected repair scheme to all data units covered by the abnormal causal chain tracing results to generate a repaired three-dimensional unified expression data set; Step S4: Obtain the multidimensional constraint set, and perform an association mapping between the repaired three-dimensional unified expression data set and the multidimensional constraint set to generate a three-dimensional data set with constraint labels; The three-dimensional data set with constraint labels is divided into multiple constraint subspaces according to the constraint labels. In each constraint subspace, candidate land use schemes that satisfy the corresponding constraint conditions of the constraint subspace are searched to obtain a set of candidate schemes for each constraint subspace. The sets of candidate schemes for each constraint subspace are merged and conflict resolved to generate an output decision scheme that satisfies all multidimensional constraints and has the best comprehensive benefit index. Step S5: Output the output decision scheme to the execution end and collect the actual execution effect data generated by the output decision scheme during the execution process; feed the actual execution effect data back to the multidimensional constraint set, adjust the constraint strength of each constraint in the multidimensional constraint set according to the actual execution effect data, and generate an updated multidimensional constraint set for the next data analysis cycle.

2. The data analysis method based on comprehensive land resource management according to claim 1, characterized in that, The step S1, which involves jointly encoding the time label, the spatial coordinates, and the attribute label to generate a three-dimensional unified representation data set, specifically includes: Construct a three-dimensional data space with the time axis, spatial coordinate axis, and attribute axis as the base coordinates; Each multi-source land data point is mapped to a data point in the three-dimensional data space, and the data point carries time dimension coordinates, spatial dimension coordinates, and attribute dimension vectors. The comprehensive distance between any two data points in the three-dimensional data space is defined as the weighted sum of the time dimension distance, spatial dimension distance, and attribute dimension distance, and the comprehensive distance is used as the direct distance calculation result. The set of all mapped data points is output as the three-dimensional unified representation data set.

3. The data analysis method based on comprehensive land resource management according to claim 1, characterized in that, The step of performing reverse tracing along the data propagation path in step S2 specifically includes: A data lineage diagram of the target region is pre-constructed, which records the source of each data unit, the processing and transformation process, and the output destination. When an abnormal data unit is detected, the data lineage graph is traversed in reverse from the abnormal data unit, and each predecessor node is marked in turn until the original acquisition node without a predecessor node is reached. The original acquisition node is then located as the source data unit. Record all nodes and edges traversed during the reverse traversal as the propagation path node sequence; The data units directly or indirectly affected by all nodes in the propagation path node sequence are marked as the range of the affected data units; The source data unit, the propagation path node sequence, and the affected data unit range are combined to form the abnormal causal chain tracing result.

4. The data analysis method based on comprehensive land resource management according to claim 1, characterized in that, Step S3, which involves generating at least two repair schemes for the source data unit within the virtual land model, specifically includes: Based on the anomaly type of the source data unit, at least two different strategy directions for repair operations are selected from the repair strategy library. The different strategy directions include root cause elimination strategy and impact compensation strategy. For each strategy direction, a complete set of repair operation sequences is generated, and each set of repair operation sequences is output as a repair plan. The specific steps for simulating and deducing each repair scheme include: Each set of remediation operation sequences is executed in the virtual land model to simulate the data changes generated by the remediation operation sequences on all data units covered by the abnormal causal chain tracing results. The comprehensive state evaluation value of the virtual land model after the execution of each set of remediation operation sequences is recorded as the simulation and deduction result corresponding to the remediation scheme. The step of selecting a repair scheme based on the simulation results specifically includes: Compare the comprehensive status evaluation values ​​corresponding to each set of repair schemes, and select the repair operation sequence with the highest comprehensive status evaluation value as the selected repair scheme; The selected repair scheme is applied to all data units covered by the abnormal causal chain tracing results to generate a repaired three-dimensional unified representation data set.

5. The data analysis method based on comprehensive land resource management according to claim 1, characterized in that, The step S4, which involves dividing the constrained 3D data set into multiple constrained subspaces according to the constraint markers, specifically includes: Based on the constraint type identifier contained in the constraint tag, data units with the same constraint type identifier are grouped into the same constraint subspace, and each constraint subspace corresponds to a constraint type. The step of searching for candidate land use schemes that satisfy the corresponding constraints in each constraint subspace specifically includes: For each constraint subspace, a local search strategy corresponding to the constraint type is adopted to generate at least one candidate land use scheme that satisfies all constraints in the constraint subspace, and the generated set of candidate land use schemes is taken as the set of candidate schemes for the constraint subspace. The steps of merging and resolving conflicts in the candidate solution sets of each constraint subspace specifically include: Combine the candidate schemes of all constrained subspaces by Cartesian product to form multiple cross-subspace combination schemes; For each cross-subspace combination scheme, the constraint conflict items of the combination scheme in different constraint subspaces are detected, and the constraint conflict items are adjusted according to the preset conflict resolution rules to generate a conflict-free global candidate scheme. The solution that satisfies all multidimensional constraints and has the best comprehensive benefit index is selected from all conflict-free global candidate solutions and is taken as the output decision solution.

6. The data analysis method based on comprehensive land resource management according to claim 1, characterized in that, The step S5, which involves adjusting the constraint strength of each constraint in the multidimensional constraint set based on the actual execution effect data, specifically includes: Calculate the deviation between the actual execution effect data and the predicted effect data of the output decision scheme before execution; Based on the magnitude and direction of the deviation value, an adjustment coefficient is determined for each constraint condition, wherein the adjustment coefficient is a positive adjustment value or a negative adjustment value; Multiply the current constraint strength of each constraint condition by the corresponding adjustment factor to obtain the updated constraint strength; All updated constraint strengths are recombined to generate an updated set of multidimensional constraints.

7. The data analysis method based on comprehensive land resource management according to claim 3, characterized in that, In the steps of constructing the data lineage diagram, the source of generation includes the data acquisition device identifier, acquisition timestamp, and acquisition location; the processing and transformation process includes at least one operation among data cleaning, data interpolation, data fusion, and data conversion; and the output destination includes the downstream data processing module identifier and data storage location.

8. The data analysis method based on comprehensive land resource management according to claim 4, characterized in that, The comprehensive status assessment value is calculated by weighting multiple assessment indicators, including the integrity of the repaired data, data consistency, and data prediction accuracy. The higher the comprehensive status assessment value, the better the repair effect.

9. A data analysis method based on comprehensive land resource management according to claim 5, characterized in that, The preset conflict resolution rules include: when two candidate schemes in two constraint subspaces are detected to generate different land use assignments on the same data unit, the matching degree of the attribute feature vector of the data unit in the three-dimensional unified expression data set with the constraint conditions corresponding to each constraint subspace is compared, and the assignment with the highest matching degree is taken as the final assignment after conflict resolution.

10. A data analysis method based on comprehensive land resource management according to claim 6, characterized in that, The deviation value includes positive deviation and negative deviation. When the deviation value is positive, it means that the actual performance is better than the predicted performance. In this case, the constraints that make positive contributions are positively adjusted to enhance the constraint strength, and the constraints that make negative contributions are negatively adjusted to weaken the constraint strength. When the deviation value is negative, the constraint condition that makes a positive contribution is negatively adjusted to weaken the constraint strength, and the constraint condition that makes a negative contribution is positively adjusted to strengthen the constraint strength.

Images