Land resource information data intelligent management system and method

Abstract

The invention discloses a land resource information data intelligent management system and method, and relates to the technical field of geographic information, and the method comprises the steps: S1, constructing a fusion feature vector, and improving the modeling integrity of multi-dimensional land information; s2, through a linkage mechanism of lateral deviation analysis and trend prediction, prospective identification of resource state changes is realized, and a basis is provided for accurate clustering of resource fluctuation sensitive areas; and meanwhile, a dimension clustering weight Q is constructed, so that the interference of redundant information dimensions on a clustering result can be effectively suppressed. And S3, a label credibility level set is constructed based on the matching value PS, semantic alignment of the clustering result and the existing land class label is realized, the management landing capability of the classification result is enhanced, and a basis is provided for setting a fidelity threshold for subsequent compression. And S4, dynamically adjusting the compression strength by combining the credibility level of the label through difference compression and fidelity evaluation of the resource vector of each partition, and ensuring the reduction precision of the high-credibility semantic region.

Description

Technical Field

[0001] This invention relates to the field of geographic information technology, specifically to an intelligent management system and method for land resource information data. Background Technology

[0002] With the rapid development of geographic information technology, remote sensing data processing, and big data platforms, land resource management is gradually entering a digital and intelligent stage. However, most existing management methods remain at the level of static retrieval of structured data, lacking the ability to integrate multi-source data and comprehensively model multi-dimensional heterogeneous elements. In the intelligent management of land resource information data, especially in tasks such as regional coordination, land use planning, and value assessment, there is an urgent need to have the ability to analyze the horizontal differences between different land zones in order to identify the location, potential advantages, and risk factors of the target area within the overall regional system. However, existing methods lack the extraction and utilization of "horizontal deviation characteristics," resulting in regional zones being treated as isolated samples in the intelligent management system, making it difficult to support truly comparative management decisions.

[0003] The root cause of these deficiencies lies in the current land resource management system's lack of an intelligent analysis framework that integrates multimodal information fusion, structural difference perception, trend dynamic driving, and semantic compression control. The failure to establish a cross-temporal, cross-spatial, and cross-dimensional resource state vector system results in a lack of visual evidence of real-world differences between zones, leading to delayed regulatory responses and inaccurate resource allocation. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent management system and method for land resource information data, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent management method for land resource information data, comprising the following steps:

[0006] S1: Divide the land area to be managed into zones, obtain multi-source land resource information for each zone from the cadastral management system, and construct feature vectors through information fusion.

[0007] S2: Based on feature vectors Analyze the deviation of indicators between zones and predict the development trends of each dimension within each zone to estimate the resource value at the next future time. By combining the real-time data of each dimension within each partition, a dimensional clustering weight Q is set for each dimension within each partition;

[0008] S3: Based on the index deviation between partitions analyzed in S2, preset the initial cluster center vector. The new cluster center vector is generated by updating using a weighted average method. The new cluster center vector Compared with standard label vectors Perform similarity matching to obtain the matching value PS, and construct a set of tag credibility levels based on the matching value PS;

[0009] S4: Compress multi-source land resource information within each partition, and after compression and restoration, determine the fidelity Fd of each partition. Combined with the label credibility level set in S3, dynamically trigger the compression intensity correction operation.

[0010] Preferably, step S1 specifically includes:

[0011] S11: Based on the geographical requirements of intelligent management of land resource information data, predetermine the land area to be managed, and divide the land area to be managed into spatial grids to generate several sub-zones;

[0012] S12: Call the platform interface of the land registration management system and natural resource database to obtain multi-source land resource information for each partition. The multi-source land resource information includes numerical information, text information and time attribute information; and convert the text information into a numerical vector using TF-IDF technology.

[0013] S13: The multi-source land resource information in S12 is standardized using Z-score standardization, and the numerical, textual, and temporal attribute information are concatenated to construct the feature vectors of each partition.

[0014] Preferably, step S2 specifically includes:

[0015] S21: Based on the feature vectors of each partition obtained in S13 Used to analyze the deviation of indicators between different partitions, specifically:

[0016]

[0017] In the formula, XD i L represents the lateral deviation index of the i-th partition. i，j This represents the resource value of the i-th partition in the j-th dimension. J represents the average resource value of all partitions except the i-th partition in the j-th dimension, where J represents the total number of resource dimensions, j represents the resource dimension number, and i represents the partition number.

[0018] Preferably, step S2 further includes:

[0019] S22: Predicting the development trends of each dimension within each partition includes:

[0020] Based on the feature vectors of each partition obtained in S13 Extract the time attribute information of each partition;

[0021] By using the time attribute information of each partition, the resource value at the next future moment can be predicted. Specifically:

[0022]

[0023] In the formula, L represents the predicted resource value of the i-th partition in the j-th dimension at the next future time step; i，j (t) represents the resource value of the current i-th partition in the j-th dimension, L i，j (t-1) represents the resource value of the i-th partition in the j-th dimension at the previous time step; δ i，j represents the trend adjustment coefficient of the j-th dimension within the i-th partition; t represents the current time point.

[0024] Preferably, step S2 further includes:

[0025] S23: Calculate the clustering weight Q for each dimension within each partition, specifically:

[0026]

[0027] In the formula, ΔL i，j (x) represents the time-series variation term. This represents the predicted resource value of the i-th partition in the j-th dimension at time x+1.

[0028]

[0029] In the formula, Q i，j R represents the dimensional clustering weight set for the i-th partition on the j-th dimension, where j and R both represent resource dimension numbers, and R≠j.

[0030] Preferably, step S3 specifically includes:

[0031] S31: Pre-determine the number of clusters K;

[0032] S32: Based on the lateral deviation index XD of the i-th partition in S21 i The method for obtaining the lateral deviation index XD of each partition is as follows: The lateral deviation indices XD of each partition are obtained, and then sorted in ascending order. The top K partitions with the largest lateral deviation indices XD after sorting are selected as the K initial cluster center vectors.

[0033] S33: By setting the clustering weight Q and resource value L for each dimension within each partition, and combining the Euclidean distance algorithm, the distance of each partition from the K initial cluster center vectors is calculated respectively. The clustering distance D is specifically:

[0034]

[0035] In the formula, D iq Represents the distance from the i-th partition to the q-th initial cluster center vector. Cluster distance, L represents the mean of the dimensional clustering weights of the i-th and j-th partitions along the dimension of variable R. i，R L represents the resource value of the i-th partition along dimension R. q，R Represents the initial cluster center vector of the qth cluster. Resource values ​​in dimension R of the internal variable, where q represents the initial cluster center vector. The number.

[0036] Preferably, step S3 further includes:

[0037] S34: For each partition, iterate through all initial cluster center vectors. Calculate the cluster distance D from each partition to each initial cluster center, and assign the i-th partition to the initial cluster center vector with the smallest distance. Within, to generate new cluster center vectors

[0038] S35: Repeat steps S33 and S4 until a new cluster center vector is obtained. No further changes, stop repetitive operations, and add new cluster center vectors. The corresponding partitions are determined within the internal space;

[0039] S36: Based on multiple sets of standard label vectors built into the land registration management system and natural resources database. And respectively with each new cluster center vector Perform similarity matching and obtain the matching value PS, specifically:

[0040]

[0041] S37: Obtain the vectors of each new cluster center based on the method of obtaining the matching value PS. Compared with standard label vectors The matching value PS is selected, and the corresponding new cluster center vector is chosen. The standard label vector with the largest matching PS value As the new cluster center vector The standard label, that is, the vector of each new cluster center With each standard label vector Perform one-to-one matching and count the vectors of each new cluster center. The maximum value of the corresponding matching value PS is used, and the vectors of each new cluster center are calculated. The maximum value of the corresponding matching PS value is compared with a pre-set matching threshold to construct a set of tag trust levels, specifically:

[0042] If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS is greater than the matching threshold, it indicates that the current new cluster center vector is... If the credibility of the labels in each partition is in a qualified state, then a first fidelity threshold value B1 will be set.

[0043] If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS equals the matching threshold, it indicates the current new cluster center vector. If the credibility of the labels in each partition is in a qualified state, then the second fidelity threshold value B2 will be set.

[0044] If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS is less than the matching threshold, it indicates that the current new cluster center vector is... If the credibility of the labels in each partition is unqualified, then the third fidelity threshold value B3 will be set.

[0045] Based on the first fidelity threshold value B1, the second fidelity threshold value B2, and the third fidelity threshold value B3, a set of label credibility levels is generated.

[0046] Preferably, step S4 specifically includes:

[0047] S41: For each new cluster center vector The data is compressed and then restored to obtain the vectors of the new cluster centers. compressed and restored vector

[0048] S42: Calculate the vectors of each new cluster center. The fidelity Fd of each partition is as follows:

[0049]

[0050] In the formula, f represents the number of the new cluster center vector. Let f represent the new cluster center vector. Fd represents the compressed and decompressed vector after the f-th new cluster center vector has been compressed and decompressed. f This represents the fidelity of each partition within the f-th new cluster center vector.

[0051] Preferably, step S4 further includes:

[0052] S43: Based on the corresponding new cluster center vector in S37 The set fidelity threshold is used to determine the vectors of each new cluster center. Whether the compression results of each partition are satisfactory is as follows:

[0053] If the corresponding new cluster center vector When the fidelity Fd of each partition exceeds its corresponding set fidelity threshold, it means that the compression result of the corresponding partition is qualified and the compressed data is retained.

[0054] If the corresponding new cluster center vector If the fidelity Fd of each partition does not exceed its corresponding set fidelity threshold, it indicates that the compression identification of the corresponding partition has failed, and a fallback mechanism is entered to dynamically correct the compression intensity, specifically as follows:

[0055]

[0056] In the formula, CR represents the compression ratio corresponding to the modified f-th new cluster center vector. old This represents the system's uniform compression ratio before the correction, β represents the adjustment coefficient, and B... f This represents the set fidelity threshold value corresponding to the f-th new cluster center vector.

[0057] A land resource information data intelligent management system and method, including a feature module, a weight module, a clustering module and a correction module;

[0058] The feature module is used to divide the land area to be managed into zones and obtain multi-source land resource information for each zone from the cadastral management system. After information fusion, feature vectors are constructed.

[0059] The weight module is used for feature vector-based weighting. Analyze the deviation of indicators between zones and predict the development trends of each dimension within each zone to estimate the resource value at the next future time. By combining the real-time data of each dimension within each partition, a dimensional clustering weight Q is set for each dimension within each partition;

[0060] The clustering module is used to preset the initial cluster center vector based on the index deviation between partitions. The new cluster center vector is generated by updating using a weighted average method. The new cluster center vector Compared with standard label vectors Perform similarity matching to obtain the matching value PS, and construct a set of tag credibility levels based on the matching value PS;

[0061] The correction module is used to compress multi-source land resource information within each partition, and after compression and restoration, determine the fidelity Fd of each partition, and dynamically trigger the compression intensity correction operation in combination with the tag credibility level set.

[0062] This invention provides an intelligent management system and method for land resource information data, which has the following beneficial effects:

[0063] (1) This method is mainly reflected in the improvement of the intelligent linkage management mechanism for multi-source data fusion, horizontal difference identification, trend prediction, semantic clustering, and compression fidelity control. By constructing the fusion feature vector through step S1, the system can uniformly process numerical information, textual information, and time attribute information from the cadastral management system, thereby improving the modeling completeness of multi-dimensional land information. Step S2 achieves the forward-looking identification of resource status changes through the linkage mechanism of horizontal deviation analysis and trend prediction, providing a basis for accurate clustering of resource fluctuation sensitive areas; at the same time, the construction of dimensional clustering weight Q can effectively suppress the interference of redundant information dimensions on the clustering results. Step S3 constructs a label credibility level set based on the matching value PS, realizing the semantic alignment between the clustering results and existing land category labels, enhancing the management and implementation capability of the classification results, and providing a basis for setting fidelity thresholds for subsequent compression. Step S4, through differential compression and fidelity assessment of resource vectors in each zone, and dynamic adjustment of compression intensity based on label credibility level, ensures the restoration accuracy of highly credible semantic regions and further avoids distortion of important land parcel information. In summary, the overall solution constructs a closed-loop intelligent management process of data fusion, trend analysis, cluster modeling, and compression feedback, which significantly improves the intelligence, refinement, and effectiveness of land information data in management, regulation, and planning decisions.

[0064] (2) By introducing a lateral deviation index and a dynamic trend prediction mechanism, the ability to identify resource differences between zones and judge future evolution trends is effectively improved. S21 By constructing a lateral deviation index, the system can quantitatively identify key zones that significantly deviate from the regional mean in each resource dimension, enabling priority monitoring of potential anomalies or strategic plots. S22 Further introduces a resource trend prediction mechanism based on time-series data. Using the historical resource sequences of each zone, the variance of resource changes is calculated and a trend adjustment coefficient is set. Based on the intensity of data fluctuations, the reliability of the prediction results is dynamically adjusted, thereby improving the accuracy of predicting future resource evolution directions.

[0065] (3) By introducing a dimensional clustering weight Q construction mechanism, the feature sensitivity and redundancy suppression capability of clustering modeling are further enhanced. Specifically, the system first calculates the temporal variation term of each partition in each dimension based on the resource value prediction results to identify the activity level of the current resource dimension. Subsequently, combined with the structural coupling degree between each resource dimension, a clustering weight Q with both trend-driven and association-suppressing characteristics is constructed, so that in the subsequent clustering process, highly active and structurally independent dimensions have a greater impact on the clustering results, while highly redundant or highly coupled dimensions are dynamically compressed in their influence.

[0066] (4) S35 adopts the weighted Euclidean distance algorithm, combined with the dimensional clustering weight Q set for each partition, to accurately measure the multidimensional resource differences between partitions, dynamically update the clustering structure, and ensure the effectiveness of classification and the expression of real differences. S36-S37 further obtains the matching value PS by performing similarity matching with the preset standard label vector in the cadastral management system, and constructs a label credibility level set accordingly, realizing the semantic empowerment and credibility assessment of the classification results, laying the foundation for subsequent compression. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of the intelligent management method for land resource information data according to the present invention;

[0068] Figure 2 This is a logic diagram of an intelligent management method for land resource information data according to the present invention;

[0069] Figure 3 This is a block diagram of an intelligent management system for land resource information data according to the present invention. Detailed Implementation

[0070] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Example 1

[0072] Please see Figure 1 and Figure 2 This invention provides an intelligent management method for land resource information data, comprising the following steps:

[0073] S1: Divide the land area to be managed into zones, obtain multi-source land resource information for each zone from the cadastral management system, and construct feature vectors through information fusion.

[0074] S2: Based on feature vectors Analyze the deviation of indicators between zones and predict the development trends of each dimension within each zone to estimate the resource value at the next future time. By combining the real-time data of each dimension within each partition, a dimensional clustering weight Q is set for each dimension within each partition;

[0075] S3: Based on the index deviation between partitions analyzed in S2, preset the initial cluster center vector. The new cluster center vector is generated by updating using a weighted average method. The new cluster center vector Compared with standard label vectors Perform similarity matching to obtain the matching value PS, and construct a set of tag credibility levels based on the matching value PS;

[0076] S4: Compress multi-source land resource information within each partition, and after compression and restoration, determine the fidelity Fd of each partition. Combined with the label credibility level set in S3, dynamically trigger the compression intensity correction operation.

[0077] In this embodiment, the method can further realize an integrated management process of multi-source land resource information fusion expression, horizontal comparative analysis, trend-driven clustering, and compressed fidelity control. Through the fusion modeling of structured and unstructured data of each partition in S1, the system can construct a unified vector expression with spatial, attribute, and temporal dimensions; S2, through the combination of deviation analysis and trend prediction, can not only identify significantly different blocks, but also predict potential change trends, and adaptively construct dimensional clustering weights, so that the clustering mechanism has dynamic adaptability.

[0078] S3 further introduces a standard label semantic matching mechanism, enabling the clustering results to have semantic recognition capabilities and generating a credibility level, effectively supporting subsequent differentiated processing strategies; S4 sets a fidelity threshold through the label credibility level during the information compression stage, realizing dynamic control of the compression intensity of important partitions and avoiding information loss. This mechanism not only improves the intelligence and targeting of the system's data processing, but also ensures the availability and semantic integrity of the compressed data, ultimately achieving the dual goals of data accuracy assurance and computing power optimization for the land resource intelligent management system.

[0079] Example 2

[0080] Please refer to Figure 1 Specifically, the steps in S1 include:

[0081] S11: Based on the geographical requirements of intelligent management of land resource information data, predetermine the land area to be managed, and divide the land area to be managed into spatial grids to generate several sub-zones;

[0082] S12: Call the platform interface of the land registration management system and natural resource database to obtain multi-source land resource information for each partition. The multi-source land resource information includes numerical information, text information and time attribute information. Among them, the numerical information includes, but is not limited to, area, population density and environmental indicators; the text information includes, but is not limited to, land use and ownership; the time attribute includes all dimensions represented in time series form; and convert the text information into a numerical vector using TF-IDF technology.

[0083] S13: The multi-source land resource information in S12 is standardized using Z-score standardization to construct continuous vector segments at the same numerical scale, addressing the issue of different dimensions and ranges dominating the overall vector. Numerical, textual, and temporal attribute information are then concatenated to construct feature vectors for each partition.

[0084] In this embodiment, step S1 of the present invention realizes standardized modeling and vector fusion of multi-source land resource information, which has significant practical effects and intelligent foundation value. Through spatial gridding in S11, a large area of ​​land is accurately divided into several standard partitions with spatial boundaries, which facilitates data collection and subsequent unified management. In step S12, the land registration management system and natural resource database are connected. The system can automatically capture the numerical information, text attributes and time series changes corresponding to each partition, and transform text labels such as land use and ownership type into dense numerical vectors through the TF-IDF method, which enhances the structural expression ability of semantic attributes. Step S13 introduces Z-score standardization technology to unify the scale of multi-dimensional indicators, effectively solving the problem of vector dominance bias caused by inconsistent dimensions and large ranges among different indicators, and splicing the information of various types to form a unified feature vector, providing a unified modeling foundation for subsequent trend analysis, cluster judgment and fidelity compression.

[0085] Taking a certain urban outskirts area as an example, plot A has a small area but high pollution levels and is designated for industrial use; plot B has a large area but is designated for warehousing. Traditional systems often ignore the differences in their attributes, while this method can transform area, pollution, and use into a unified vector structure, clearly revealing the structural deviations and semantic differences between plots, and improving the understanding ability and data utilization efficiency of the management system.

[0086] Example 3

[0087] Please refer to Figure 1 Specifically, the steps in S2 include:

[0088] S21: Based on the feature vectors of each partition obtained in S13 Used to analyze the deviation of indicators between different partitions, specifically:

[0089]

[0090] In the formula, XD i L represents the lateral deviation index of the i-th partition. i，j This represents the resource value (numerical or metric value) of the i-th partition in the j-th dimension. J represents the average resource value of all partitions except the i-th partition in the j-th dimension, where J represents the total number of resource dimensions, j represents the resource dimension number, and i represents the partition number.

[0091] When the lateral deviation index XD of the i-th partition i The larger the value, the greater the difference between the corresponding zoning indicator and the entire land area to be managed, making it a priority for analysis or regulation.

[0092] By analyzing the index deviation between each partition, we construct a "horizontal feature perspective" of the data to distinguish the vertical trend (time change). The horizontal deviation index provides a data foundation for modeling cross-regional differences at the same time. At the same time, it can be used to sort plots and prioritize them in subsequent clustering, compression and other stages.

[0093] The specific steps in S2 also include:

[0094] S22: Predicting the development trends of each dimension within each partition includes:

[0095] Based on the feature vectors of each partition obtained in S13 Extract the time attribute information of each partition;

[0096] By using the time attribute information of each partition, the resource value at the next future moment can be predicted. Specifically:

[0097]

[0098] In the formula, L represents the predicted resource value of the i-th partition in the j-th dimension at the next future time step; i，j (t) represents the resource value of the current i-th partition in the j-th dimension, L i，j (t-1) represents the resource value of the i-th partition in the j-th dimension at the previous time step; δ i，j This represents the trend adjustment coefficient for the j-th dimension within the i-th partition, used to control the forecast increment; t represents the current time point.

[0099] Predicting resource values ​​in the next future moment Used to dynamically update clustering input vectors to identify potential partitions that tend to cluster towards areas of difference;

[0100] Since land resources are not static but in a continuous evolutionary process, traditional land clustering methods, based on the feature vectors at the current moment, ignore future development trends. This may lead to the easy clustering of plots that are about to change into the same category as plots that are stable. Furthermore, the clustering and labeling results of traditional land clustering methods are only applicable to the current moment and lack sustainable planning capabilities. Therefore, when constructing land feature vectors, the system needs to introduce future trend predictions to give the model a dynamic evolutionary perspective.

[0101] Wherein, the trend adjustment coefficient δ in the j-th dimension within the i-th partition. i，j The influence of the change in the magnitude of the change between the current and previous time points on the prediction is determined as follows:

[0102] For the historical change sequence L of dimension j within the i-th partition i，j (tk), ..., L i，j (t+1), calculate the following volatility index: where L i，j (tk) represents the resource value of the j-th dimension of the i-th partition at time point tk;

[0103]

[0104] In the formula, Var i，j L represents the volatility (variance) of the i-th partition over the last k time points in the j-th dimension; x represents the time number, L i，j (x) represents the resource value of the i-th partition in the j-th dimension at time point x. Let represent the mean (moving average) of the j-th dimension within the i-th partition at these k time points; since the starting point is k-1 steps backward from t+1, the starting point is t+1-(k-1)=tk;

[0105] Based on the volatility (variance) Var of the i-th partition in the j-th dimension over the most recent k time points. i，j Set the trend adjustment coefficient δ for the j-th dimension within the i-th partition. i，j :

[0106]

[0107] In the formula, r represents the index range of all dimensions in the corresponding partition, that is, the index variables of all feature dimensions in the current partition i; Var i，r Let represent the volatility of the i-th partition over the last k time points along the dimension r.

[0108] In this embodiment, by constructing a lateral deviation index and trend prediction model, the identification of resource differences between regions and the dynamic grasp of future evolution trends are realized, significantly improving the foresight and responsiveness of resource status assessment. Specifically, S21 calculates the degree of deviation of each partition from the overall region based on feature vector comparison. The larger the lateral deviation index, the more significant the heterogeneity of the partition in a specific dimension, and the more likely it is to be prioritized for analysis and regulation.

[0109] S22 further extracts time attribute information to construct a future resource prediction model. It introduces a trend adjustment coefficient to dynamically weight the predicted increments of different zones based on historical fluctuations, making the prediction results more closely match actual evolutionary trends. For example, if a region B shows an increasing trend in its "population density" data over the past three periods, with a high trend adjustment coefficient, the system predicts it will continue to rise significantly in the next timeframe. Simultaneously, this region also exhibits a large lateral deviation index in population density, indicating it is far higher than other zones. Based on this, the system identifies this zone as a potential over-density risk area, subsequently triggering land use structure optimization suggestions. This mechanism realizes a shift from a "static comparison" to a "dynamic evolution" management approach, providing scientific support for the refined and proactive regulation of land resources.

[0110] Example 4

[0111] Please refer to Figure 1 Specifically, the S2 steps also include:

[0112] S23: Calculate the clustering weight Q for each dimension within each partition, specifically:

[0113]

[0114] In the formula, ΔL i，j (x) represents the time-series variation term. This represents the predicted resource value of the i-th partition in the j-th dimension at time x+1.

[0115] Timing variation term ΔL i，j (x) is used to judge the degree of drastic change in the dimension in the future. The larger the value, the stronger the trend, and the more likely it is to affect the clustering. The weight tends to increase.

[0116] Specifically, by incorporating the prediction results into the cluster analysis process, it helps to make the clustering consider future state trends, achieve evolutionary clustering, and improve the long-term validity and explanatory power of the results.

[0117]

[0118] In the formula, Q i，j R represents the dimensional clustering weight set for the i-th partition on the j-th dimension, where j and R both represent resource dimension numbers, and R≠j;

[0119] It is a suppressive term whose function is to reduce the weight of the current dimension in the cluster if the trend of change in the current dimension is actually "pulled" or "co-driven" by other dimensions. This is reflected in the summation term within the current dimension. It iterates through all other dimensions R that are not equal to j. This represents the change value of the Rth dimension at this moment. In summary, the larger this suppression term is, the more the trend is mainly driven by other dimensions. Since the current dimension is not independent, its weight will be weakened.

[0120] Using the exponential decay function exp(- * The purpose is to form a smooth nonlinear suppression factor. When the coupling is strong and other dimensions change drastically, the exponent value will approach 0, and the weight will be drastically suppressed. When the coupling is weak or other dimensions change little, the exponent is close to 1 and does not suppress the current dimension.

[0121] This product value represents the strength of the independent trend contribution in the corresponding dimension at the current time point x. If the trend is large and independent, a high value is retained. If the trend is large but depends on other dimensions, it is exponentially decayed and ultimately contributes little.

[0122] The calculation process of the dimensional clustering weight Q is intended to express the weighted accumulation of the change values ​​of the corresponding dimension over the past k time points. Among them, the dimension with large change and small influence of other dimensions will be assigned a higher clustering weight.

[0123] Specifically, each dimension within each partition is assigned a clustering weight Q to control the impact of different dimensions on distance in clustering and to suppress redundant dimensions. The core idea of ​​this formula is that if a certain dimension changes drastically but it is not strongly coupled with the changes of other dimensions (i.e., relatively independent), then it is given a higher clustering weight; if a certain dimension fluctuates significantly but is highly coupled with multiple dimensions at the same time, then its changes are diluted to prevent noise from affecting the clustering.

[0124] Xt jR This indicates the structural coupling degree between the j-th dimension and the R-th dimension (the larger the value, the stronger the relationship between the dimensions);

[0125] Wherein, the structural coupling degree Xt between the j-th dimension and the R-th dimension jR Obtain it using the following formula:

[0126]

[0127] In the formula, This represents the rate of change of the i-th partition along the j-th dimension. Y represents the rate of change of the i-th partition along the R-th dimension, Y represents the total number of partitions, and d represents the calculus symbol. The expression is a derivative expression;

[0128] The formula for obtaining the structural coupling degree is based on simulating the degree of synergy when two dimensions change simultaneously. If both increase or decrease simultaneously, the product is positive, representing positive coupling. If one increases and the other decreases, the product is negative, representing negative coupling. If one of them changes very little, it indicates weak coupling or independence. By summing up the synergistic changes of all partitions and normalizing the results, a coupling strength that is effective for the global structure can be obtained.

[0129] The structural coupling degree indirectly illustrates that the higher the prominence of multidimensional differences, the more worthy the structural characteristic fluctuations it carries are to be included in coupling degree modeling, so as to avoid ignoring key structures.

[0130] In this embodiment, the introduction of a dimensional clustering weight Q construction mechanism in step S2 enables precise control over the influence of multidimensional land resource characteristics on clustering analysis, demonstrating significant technical advantages. Specifically, step S23 calculates the predicted changes in each partition at future times to form a time-series variation term, and then uses exponential adjustment based on the structural coupling degree between dimensions to dynamically construct a dimensional weight Q reflecting "trend independence." This weight controls the participation of each dimension in distance calculation during clustering, suppressing interference from highly redundant or collinear dimensions on the clustering results. For example, if a region C exhibits highly consistent trends in both environmental indicators and population density, the system automatically reduces the clustering weight of one dimension by assessing its coupling degree, thus avoiding clustering bias due to redundant information. Conversely, if a region shows a significant trend in the land price dimension and has no strong coupling relationship with other dimensions, it is given a higher clustering weight, strengthening its dominant role in classification. Through this mechanism, the system's clustering structure becomes more representative and discriminative, effectively improving the ability of spatial partitioning results to reflect real resource state differences and providing a more reliable structural foundation for subsequent label matching and semantic classification.

[0131] Example 5

[0132] Please refer to Figure 1 Specifically, the S3 steps include:

[0133] S31: Pre-determine the number of clusters K;

[0134] S32: Based on the lateral deviation index XD of the i-th partition in S21 i The method for obtaining the lateral deviation index XD of each partition is as follows: The lateral deviation indices XD of each partition are obtained, and then sorted in ascending order. The top K partitions with the largest (more representative, greater diversity) lateral deviation indices XD after sorting are selected as the K initial cluster center vectors.

[0135] It should be noted that the top K partitions with the largest lateral deviation index XD are selected as the K initial cluster center vectors. The purpose is to avoid prematurely falling into marginal plots, and also to avoid classification bias or local optima caused by traditional random initialization; that is, to avoid highly dissimilar samples being misclassified into homogeneous clusters, thereby improving the class interpretation ability.

[0136] S33: By setting the clustering weight Q and resource value L for each dimension within each partition, and combining the Euclidean distance algorithm, the distance of each partition from the K initial cluster center vectors is calculated respectively. The clustering distance D is specifically:

[0137]

[0138] In the formula, D iq Represents the distance from the i-th partition to the q-th initial cluster center vector. Cluster distance, L represents the mean of the dimensional clustering weights of the i-th and j-th partitions along the dimension of variable R. i，R L represents the resource value of the i-th partition along dimension R. q，R Represents the initial cluster center vector of the qth cluster. Resource values ​​in dimension R of the internal variable, where q represents the initial cluster center vector. The number.

[0139] It should be noted that the clustering results obtained by the clustering distance D will not be obscured by redundant dimensions, ensuring the clustering's responsiveness to the dimensions that dominate future trends, while providing a reliable grouping basis for semantic label matching and regional strategy recommendation.

[0140] The specific steps in S3 also include:

[0141] S34: For each partition, iterate through all initial cluster center vectors. Calculate the cluster distance D from each partition to each initial cluster center, and assign the i-th partition to the initial cluster center vector with the smallest distance. Within, to generate new cluster center vectors

[0142] S35: Repeat steps S33 and S4 until a new cluster center vector is obtained. No further changes, stop repetitive operations, and add new cluster center vectors. The corresponding partitions are determined within the internal space;

[0143] Specifically, by jointly modeling the magnitude of changes in the land resource dimension sequence and the correlation between dimensions, a trend-oriented dimension clustering weight model was established. This effectively alleviates the problem of multidimensional data redundancy and interference, making subsequent clustering results more stable and forward-looking, and achieving adaptive adjustment capability that aligns the clustering dimension weights with the trend evolution direction.

[0144] S36: Based on multiple sets of standard label vectors built into the land registration management system and natural resources database. And respectively with each new cluster center vector Perform similarity matching and obtain the matching value PS, specifically:

[0145]

[0146] S37: Obtain the vectors of each new cluster center based on the method of obtaining the matching value PS. Compared with standard label vectors The matching value PS is selected, and the corresponding new cluster center vector is chosen. The standard label vector with the largest matching PS value As the new cluster center vector The standard label, that is, the vector of each new cluster center With each standard label vector Perform one-to-one matching and count the vectors of each new cluster center. The maximum value of the corresponding matching value PS is used, and the vectors of each new cluster center are calculated. The maximum value of the corresponding matching PS value is compared with a pre-set matching threshold to construct a set of tag trust levels, specifically:

[0147] If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS is greater than the matching threshold, it indicates that the current new cluster center vector is... If the credibility of the labels in each partition is in a qualified state, that is, the credibility is high, then a first fidelity threshold value B1 will be set.

[0148] If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS equals the matching threshold, it indicates the current new cluster center vector. If the credibility of the labels in each partition is in a qualified state, that is, the credibility is medium, then the second fidelity threshold value B2 will be set.

[0149] If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS is less than the matching threshold, it indicates that the current new cluster center vector is... If the credibility of the labels in each partition is unqualified, i.e., the credibility is low, then the third fidelity threshold value B3 will be set.

[0150] Based on the first fidelity threshold B1, the second fidelity threshold B2, and the third fidelity threshold B3, a set of label credibility levels is generated, wherein the first fidelity threshold B1 < the second fidelity threshold B2 < the third fidelity threshold B3.

[0151] Specifically, the aforementioned confidence levels (high, medium, and low) are discrete hierarchical representations of the semantic similarity confidence score (matching value PS) output by S36. This reflects the system's judgment on the reliability of the corresponding partition clustering results. The lower the confidence level, the higher the fidelity threshold requirement. Higher confidence levels indicate greater system trust in the labels, clearer semantics, and less impact from information compression. Conversely, lower confidence levels suggest less trust or uncertainty in the labels, indicating a risk of semantic drift for the corresponding plots. Further information compression in this case would further lose original data, potentially compressing important features that were already difficult to identify, leading to an amplification effect of errors. Therefore, to avoid decision-making errors caused by information loss, fidelity needs to be improved. In other words, losing important prediction trends or parameters during compression would severely impact subsequent management.

[0152] The matching value PS is used to reflect the standard label vector. With the new cluster center vector semantic similarity;

[0153] It should be noted that all the parameters mentioned above have been normalized to ensure that the clustering weights of all dimensions or parameters are on the same numerical scale.

[0154] In this embodiment, step S3 significantly improves the stability, accuracy, and interpretability of land zoning clustering by constructing an initialization mechanism based on the lateral deviation index, a weighted distance-driven iterative clustering process, and a semantic matching method with standard label vectors. Specifically, step S32 introduces the lateral deviation index XD of the zoning and sorts it in ascending order to select initial cluster centers from zonings whose resource status is closer to the regional average, thus avoiding structural imbalance in the early stage of clustering due to deviation from extreme value samples.

[0155] S35 introduces dimensional clustering weight Q to participate in weighted Euclidean distance calculation, enabling adaptive control of cluster affiliation based on resource dimension differences, and generating stable new cluster centers through iteration. S36-S37 perform similarity matching with cadastral label vectors to obtain the matching value PS for each cluster center, assigning it a corresponding standard label and credibility level, forming a label credibility level set, providing a basis for differentiated fidelity control in subsequent compression modules. For example, if the semantic matching degree of a new cluster center vector is high (e.g., a PS value of 0.95 with the "permanent basic farmland" label), the system can automatically set the fidelity threshold for that cluster to B1, ensuring the information integrity of important plots within that cluster in subsequent compression. Through this step, the system realizes a linkage mechanism from structural classification to semantic cognition, enhancing the business application value and data expression capabilities of land clustering results.

[0156] Example 6

[0157] Please refer to Figure 1 Specifically, the S4 steps include:

[0158] S41: For each new cluster center vector The data is compressed and then restored to obtain the vectors of the new cluster centers. compressed and restored vector

[0159] S42: Calculate the vectors of each new cluster center. The fidelity Fd of each partition is as follows:

[0160]

[0161] In the formula, f represents the number of the new cluster center vector. Let f represent the new cluster center vector. Fd represents the compressed and decompressed vector after the f-th new cluster center vector has been compressed and decompressed. f This represents the fidelity of each partition within the f-th new cluster center vector;

[0162] The calculation of fidelity Fd also takes into account future data. The purpose is that if the current state may appear stable, but the future trend may change drastically, if compression is based only on the current value, key future early warning information will be lost. Therefore, the predicted future resource state should be taken into account in the compression evaluation process to ensure that those partitions with high future risk or high evolution can obtain higher fidelity.

[0163] Fidelity Fd reflects the information retention rate; a value close to 1 indicates good restoration performance.

[0164] The specific steps in S4 also include:

[0165] S43: Based on the corresponding new cluster center vector in S37 The set fidelity threshold is used to determine the vectors of each new cluster center. Whether the compression results of each partition are satisfactory is as follows:

[0166] If the corresponding new cluster center vector When the fidelity Fd of each partition exceeds its corresponding set fidelity threshold, it means that the compression result of the corresponding partition is qualified and the compressed data is retained.

[0167] If the corresponding new cluster center vector If the fidelity Fd of each partition does not exceed its corresponding set fidelity threshold, it indicates that the compression identification of the corresponding partition has failed, and a fallback mechanism is entered to dynamically correct the compression intensity, specifically as follows:

[0168]

[0169] In the formula, CR represents the compression ratio corresponding to the modified f-th new cluster center vector. old This indicates the system's unified compression ratio before the correction, β represents the adjustment coefficient (value from 0.5 to 1.5), and B... f This represents the set fidelity threshold value corresponding to the f-th new cluster center vector.

[0170] The lower the credibility level, the higher the fidelity threshold B, and the more stringent the compression. Once the fidelity Fd does not exceed its corresponding set fidelity threshold, the compression ratio will be automatically reduced to perform more conservative compression.

[0171] In this embodiment, the method effectively achieves a dynamic balance between data compression efficiency and information retention accuracy by introducing a label credibility-driven compression fidelity control mechanism in step S4. S41 first performs vector compression and restoration operations on the new cluster center vectors to construct a compressed and restored vector structure. S42 calculates the fidelity Fd based on the difference between the vectors before and after compression, used to quantify the information retention rate after compression; the closer the value is to 1, the better the restoration effect. S43 introduces a fidelity threshold B set in the label credibility level set as a control standard to automatically determine whether the compression of each partition is qualified. When the Fd of a partition is lower than its corresponding threshold B, the system will automatically reduce the compression ratio through a compression ratio correction formula to execute a more conservative compression strategy, ensuring that key semantic information is not lost. This mechanism guarantees the data restoration capability of high-value or low-matching partitions, effectively preventing data distortion or semantic shift caused by excessive compression, and improving the data reliability and adaptability of the system in resource planning and management.

[0172] Example 7

[0173] Please refer to Figure 3 Specifically: a land resource information data intelligent management system and method, including a feature module, a weight module, a clustering module and a correction module;

[0174] The feature module is used to divide the land area to be managed into zones and obtain multi-source land resource information for each zone from the cadastral management system. After information fusion, feature vectors are constructed.

[0175] The weight module is used for feature vector-based weighting. Analyze the deviation of indicators between zones and predict the development trends of each dimension within each zone to estimate the resource value at the next future time. By combining the real-time data of each dimension within each partition, a dimensional clustering weight Q is set for each dimension within each partition;

[0176] The clustering module is used to preset the initial cluster center vector based on the index deviation between partitions. The new cluster center vector is generated by updating using a weighted average method. The new cluster center vector Compared with standard label vectors Perform similarity matching to obtain the matching value PS, and construct a set of tag credibility levels based on the matching value PS;

[0177] The correction module is used to compress multi-source land resource information within each partition, and after compression and restoration, determine the fidelity Fd of each partition, and dynamically trigger the compression intensity correction operation in combination with the tag credibility level set.

[0178] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent management of land resource information data, characterized in that: Includes the following steps, S1: Divide the land area to be managed into zones, obtain multi-source land resource information for each zone from the cadastral management system, and construct feature vectors through information fusion. S2: Based on feature vectors Analyze the deviation of indicators between zones and predict the development trends of each dimension within each zone to estimate the resource value at the next future time. By combining the real-time data of each dimension within each partition, a dimensional clustering weight Q is set for each dimension within each partition; S3: Based on the index deviation between partitions analyzed in S2, preset the initial cluster center vector. The new cluster center vector is generated by updating using a weighted average method. The new cluster center vector Compared with standard label vectors Perform similarity matching to obtain the matching value PS, and construct a set of tag credibility levels based on the matching value PS; S4: Compress multi-source land resource information within each partition, and after compression and restoration, determine the fidelity Fd of each partition. Combined with the label credibility level set in S3, dynamically trigger the compression intensity correction operation.

2. The intelligent management method for land resource information data according to claim 1, characterized in that: The specific steps in S1 include: S11: Based on the geographical requirements of intelligent management of land resource information data, predetermine the land area to be managed, and divide the land area to be managed into spatial grids to generate several sub-zones; S12: Call the platform interface of the land registration management system and natural resource database to obtain multi-source land resource information for each partition. The multi-source land resource information includes numerical information, text information and time attribute information; and convert the text information into a numerical vector using TF-IDF technology. S13: The multi-source land resource information in S12 is standardized using Z-score standardization, and the numerical, textual, and temporal attribute information are concatenated to construct the feature vectors of each partition.

3. The intelligent management method for land resource information data according to claim 2, characterized in that: The specific steps in S2 include: S21: Based on the feature vectors of each partition obtained in S13 Used to analyze the deviation of indicators between different partitions, specifically: In the formula, XD i L represents the lateral deviation index of the i-th partition. i，j This represents the resource value of the i-th partition in the j-th dimension. J represents the average resource value of all partitions except the i-th partition in the j-th dimension, where J represents the total number of resource dimensions, j represents the resource dimension number, and i represents the partition number.

4. The intelligent management method for land resource information data according to claim 3, characterized in that: S2 also includes the following specific steps: S22: Predicting the development trends of each dimension within each partition includes: Based on the feature vectors of each partition obtained in S13 Extract the time attribute information of each partition; By using the time attribute information of each partition, the resource value at the next future moment can be predicted. Specifically: In the formula, L represents the predicted resource value of the i-th partition in the j-th dimension at the next future time step; i，j (t) represents the resource value of the current i-th partition in the j-th dimension, L i，j (t-1) represents the resource value of the i-th partition in the j-th dimension at the previous time step; δ i，j represents the trend adjustment coefficient of the j-th dimension within the i-th partition; t represents the current time point.

5. The intelligent management method for land resource information data according to claim 4, characterized in that: S2 also includes the following specific steps: S23: Calculate the clustering weight Q for each dimension within each partition, specifically: In the formula, ΔL i，j (x) represents the time-series variation term. L represents the predicted resource value of the i-th partition in the j-th dimension at time x+1; i，j (x) represents the resource value of the i-th partition in the j-th dimension at time x; In the formula, Q i，j Let represent the dimensional clustering weight set for the i-th partition on the j-th dimension, where j and R both represent resource dimension numbers, and R≠j, and x represents the time number.

6. The intelligent management method for land resource information data according to claim 5, characterized in that: The specific steps in S3 include: S31: Pre-determine the number of clusters K; S32: Based on the lateral deviation index XD of the i-th partition in S21 i The method for obtaining the lateral deviation index XD of each partition is as follows: The lateral deviation indices XD of each partition are obtained, and then sorted in ascending order. The top K partitions with the largest lateral deviation indices XD after sorting are selected as the K initial cluster center vectors. S33: By setting the clustering weight Q and resource value L for each dimension within each partition, and combining the Euclidean distance algorithm, the distance of each partition from the K initial cluster center vectors is calculated respectively. The clustering distance D is specifically: In the formula, D iq Represents the distance from the i-th partition to the q-th initial cluster center vector. Cluster distance, L represents the mean of the dimensional clustering weights of the i-th and j-th partitions along the dimension of variable R. i，R L represents the resource value of the i-th partition along dimension R. q，R Represents the initial cluster center vector of the qth cluster. Resource values ​​in dimension R of the internal variable, where q represents the initial cluster center vector. The number.

7. The intelligent management method for land resource information data according to claim 6, characterized in that: The specific steps in S3 also include: S34: For each partition, iterate through all initial cluster center vectors. Calculate the cluster distance D from each partition to each initial cluster center, and assign the i-th partition to the initial cluster center vector with the smallest distance. Within, to generate new cluster center vectors S35: Repeat steps S33 and S4 until a new cluster center vector is obtained. No further changes, stop repetitive operations, and add new cluster center vectors. The corresponding partitions are determined within the internal space; S36: Based on multiple sets of standard label vectors built into the land registration management system and natural resources database. And respectively with each new cluster center vector Perform similarity matching and obtain the matching value PS, specifically: S37: Obtain the vectors of each new cluster center based on the method of obtaining the matching value PS. Compared with standard label vectors The matching value PS is selected, and the corresponding new cluster center vector is chosen. The standard label vector with the highest matching PS value As the center vector of this new cluster The standard label, that is, the vector of each new cluster center With each standard label vector Perform one-to-one matching and count the vectors of each new cluster center. The maximum value of the corresponding matching value PS is used, and the vectors of each new cluster center are calculated. The maximum value of the corresponding matching PS value is compared with a pre-set matching threshold to construct a set of tag trust levels, specifically: If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS is greater than the matching threshold, it indicates that the current new cluster center vector is... If the credibility of the labels in each partition is in a qualified state, then a first fidelity threshold value B1 will be set. If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS equals the matching threshold, it indicates the current new cluster center vector. If the credibility of the labels in each partition is in a qualified state, then the second fidelity threshold value B2 will be set. If the corresponding new cluster center vector When the maximum value of the corresponding matching value PS is less than the matching threshold, it indicates that the current new cluster center vector is... If the credibility of the labels in each partition is unqualified, then the third fidelity threshold value B3 will be set. Based on the first fidelity threshold value B1, the second fidelity threshold value B2, and the third fidelity threshold value B3, a set of label credibility levels is generated.

8. The intelligent management method for land resource information data according to claim 7, characterized in that: The specific steps of S4 include: S41: For each new cluster center vector The data is compressed and then restored to obtain the vectors of the new cluster centers. compressed and restored vector S42: Calculate the vectors of each new cluster center. The fidelity Fd of each partition is as follows: In the formula, f represents the number of the new cluster center vector. Let f represent the new cluster center vector. Fd represents the compressed and decompressed vector after the f-th new cluster center vector has been compressed and decompressed. f This represents the fidelity of each partition within the f-th new cluster center vector.

9. The intelligent management method for land resource information data according to claim 8, characterized in that: The specific steps in S4 also include: S43: Based on the corresponding new cluster center vector in S37 The set fidelity threshold is used to determine the vectors of each new cluster center. Whether the compression results of each partition are satisfactory is as follows: If the corresponding new cluster center vector When the fidelity Fd of each partition exceeds its corresponding set fidelity threshold, it means that the compression result of the corresponding partition is qualified and the compressed data is retained. If the corresponding new cluster center vector If the fidelity Fd of each partition does not exceed its corresponding set fidelity threshold, it indicates that the compression identification of the corresponding partition has failed, and a fallback mechanism is entered to dynamically correct the compression intensity, specifically as follows: In the formula, CR represents the compression ratio corresponding to the modified f-th new cluster center vector. old This represents the system's uniform compression ratio before the correction, β represents the adjustment coefficient, and B... f This represents the set fidelity threshold value corresponding to the f-th new cluster center vector.

10. A land resource information data intelligent management system and method, used to implement the land resource information data intelligent management method according to any one of claims 1 to 9, characterized in that: It includes a feature module, a weight module, a clustering module, and a correction module; The feature module is used to divide the land area to be managed into zones and obtain multi-source land resource information for each zone from the cadastral management system. After information fusion, feature vectors are constructed. The weight module is used for feature vector-based weighting. Analyze the deviation of indicators between zones and predict the development trends of each dimension within each zone to estimate the resource value at the next future time. By combining the real-time data of each dimension within each partition, a dimensional clustering weight Q is set for each dimension within each partition; The clustering module is used to preset the initial cluster center vector based on the index deviation between partitions. The new cluster center vector is generated by updating using a weighted average method. The new cluster center vector Compared with standard label vectors Perform similarity matching to obtain the matching value PS, and construct a set of tag credibility levels based on the matching value PS; The correction module is used to compress multi-source land resource information within each partition, and after compression and restoration, determine the fidelity Fd of each partition, and dynamically trigger the compression intensity correction operation in combination with the tag credibility level set.

Images