An online intelligent assessment system for urban land value based on big data analysis

Abstract

This invention discloses an online intelligent assessment system for urban land value based on big data analysis, comprising: a data normalization module for acquiring and preprocessing multi-source heterogeneous data; an association mapping module for performing temporal registration, using data components as graph nodes and calculating association weights to construct a value association graph; a propagation solution module for locating nodes to be assessed, iteratively updating access probabilities along node connections to obtain an association propagation state sequence; an evolution and recombination module for performing multi-scale time window rearrangement, elevating to a linear evolution space and recursively recombinating to obtain the value evolution state; an inversion assessment module for variational Bayesian inversion, filtering out posterior state components and retaining posterior parameters to obtain the value assessment result; and an incremental update module for acquiring feedback data and incrementally updating the value association graph. This invention enables online intelligent assessment of urban land value, improving the accuracy of value assessment.

Description

Technical Field

[0001] This invention relates to the field of land valuation technology, and in particular to an online intelligent valuation system for urban land based on big data analysis. Background Technology

[0002] With the increasing demands for urban spatial governance, resource allocation, and digital management, land valuation has gradually shifted from traditional manual assessment methods to information-based and model-based approaches. Current technologies typically involve collecting transaction records, planning information, time-series monitoring data, and other relevant data to conduct statistical analysis, rule-based judgment, or model prediction of land values ​​to generate corresponding assessment results. Some solutions introduce big data analytics to aggregate and process multi-source data, combining machine learning models, graph models, or time-series analysis methods to output valuation results, thereby improving assessment efficiency, shortening the assessment cycle, and enabling a degree of automation in land valuation.

[0003] However, existing technologies still have significant shortcomings in processing multi-source heterogeneous data. Data from different sources differ in field structure, time granularity, recording methods, and collection frequency. If only simple alignment, splicing, or uniform encoding methods are used for processing, it is easy to cause temporal misalignment, information loss, and distorted correlations, resulting in a lack of stable data foundation for subsequent mapping, propagation, and evaluation stages. Especially in online evaluation scenarios, if the front-end data is not sufficiently organized, the construction of value correlations can easily deviate from the actual change process, making it difficult to support subsequent continuous solving and dynamic updates.

[0004] Furthermore, existing technologies also have limitations in value relationship modeling and evaluation inference. One type of approach treats each influencing factor as an independent input for regression evaluation, lacking the ability to organize the correlation weights and node connections between multiple data components, making it difficult to form a propagable and solvable value relationship graph. Another type of approach, although introducing graph structures or time series models, usually remains at the level of single propagation or static prediction, failing to effectively connect the propagation state sequence with the evolution state, making it difficult to characterize the evolution process of value states over a continuous time range. Moreover, after the evaluation results are formed, existing technologies often lack an incremental update mechanism for feedback data, making it impossible to continuously revise the value relationship graph based on actual observation results, resulting in a lag in the evaluation results and insufficient online adaptability.

[0005] Therefore, how to provide an online intelligent assessment system for urban land value based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an online intelligent assessment system for urban land value based on big data analysis. This invention fully utilizes dynamic time warping, random walk with restart, deep Koopman operator, dynamic mode decomposition, variational Bayesian inversion, and incremental Kalman correction. It details the implementation process of urban land multi-source heterogeneous data warping, value association graph construction, association propagation state solution, value evolution state reorganization, assessment result inversion, and feedback incremental update. It has the advantages of strong association modeling ability, high online assessment ability, and good dynamic correction ability.

[0007] According to an embodiment of the present invention, an online intelligent assessment system for urban land value based on big data analysis includes: The data standardization module is used to acquire multi-source heterogeneous urban land data and preprocess it to form a standard data sequence. The association graphing module is used to perform temporal registration on standard data sequences. The registered data components are used as graph nodes. The module calculates the association weights between graph nodes and adjusts the node connection relationships to construct a value association graph. The propagation solution module is used to locate the node to be evaluated in the value association graph. It uses a random walk algorithm with restart to iteratively update the access probability along the node connection relationship, and obtains the association propagation state sequence corresponding to the node to be evaluated under the constraints of restart probability and convergence threshold. The evolution and recombination module is used to rearrange the associated propagation state sequence in a multi-scale time window. The deep Koopman operator is used to elevate the rearranged state trajectory to a linear evolution space. Combined with dynamic mode decomposition to strip away the dominant artificial modes and recursively recombine them, the value evolution state is obtained. The inversion evaluation module is used to perform variational Bayesian inversion on the value evolution state, use the maximum mean difference to screen out posterior state components with excessive deviation, and use the Lagrange multiplier method to calibrate the posterior parameters corresponding to the retained components to obtain the value evaluation results. The incremental update module is used to obtain feedback data on the value assessment results and perform incremental updates on the value association graph in conjunction with incremental Kalman correction.

[0008] Optionally, the data standardization module includes: acquiring multi-source heterogeneous data; renaming and rearranging similar fields in each source data according to a unified field order; aligning and merging the renamed data records according to the same identifier field and acquisition time; imputing missing values ​​in the aligned data records using adjacent time points and correcting outliers using the sliding median; and arranging the corrected data records according to a unified field order and time order to form a standard data sequence.

[0009] Optionally, the graph weighting module includes: Extract the time series corresponding to each data component from the standard data sequence, select a reference sequence, use the dynamic time warping method to calculate the alignment path of each time series relative to the reference sequence, and adjust the time position of each time series according to the alignment path to obtain the registered data components. The data component is a sequence of values ​​for the same field at different time positions in a standard data sequence; Using the registered data components as graph nodes, calculate the mutual information value between graph nodes, and use the mutual information value as the association weight between graph nodes; Remove node connection edges with association weights lower than a preset threshold, and retain node connection edges with association weights not lower than the preset threshold to form the initial node connection relationship; The initial node connection relationships are divided using the Louvain community partitioning method, and the node connection edges within the same partition and between different partitions are adjusted based on the partitioning results to construct a value association graph.

[0010] Optionally, the construction of the value association graph includes: Each graph node in the initial node connection relationship is treated as an independent partition. The change in modularity of each graph node after being merged into the adjacent partition is calculated. The change in modularity is the difference between the increment of the weight of the node connection edge in the partition and the corresponding expected increment of the weight of the connection edge before and after the current graph node is merged into the adjacent partition. The adjacent partition with the largest change in modularity greater than zero is determined as the target partition of the current graph node, and the current graph node is assigned to the target partition. When the change in modularity is not greater than zero, keep the original partition of the current graph node unchanged, and perform the partitioning adjustment on all graph nodes in sequence to obtain the first partitioning result; Merge the graph nodes belonging to the same partition in the first partitioning result, recalculate the weights of the node connection edges between the merged graph nodes, and obtain new node connection relationships. Based on the new node connection relationships, the change in modularity of each graph node after merging into adjacent partitions is recalculated. The corresponding graph node is determined by the adjacent partition with the largest change in modularity that is greater than zero, until the partition to which each graph node belongs no longer changes. Retain the nodes within the same partition, and filter out the nodes whose association weights between different partitions are lower than a preset threshold to construct a value association graph.

[0011] Optionally, the propagation solution module includes: Locate the node to be evaluated in the value association graph, set the initial access probability of the node to be evaluated as the starting value, set the initial access probability of the other graph nodes to zero, and form an initial access probability sequence. The node to be evaluated is the graph node in the value association graph that corresponds to the data component to be evaluated. Read the node connection edge weights between each graph node, sum the weights of all node connection edges corresponding to each graph node, and divide the weight of each node connection edge by the sum of the weights of the node connection edges of the corresponding graph node to obtain the access probability ratio of each graph node along each node connection edge. When the sum of the weights of the edges connecting the nodes corresponding to a graph node is zero, the access probability of that graph node remains unchanged in its original position. Based on the proportion of the access probability propagated outward from each graph node along the node connection edge, the access probability in the previous round of access probability sequence is distributed to the adjacent graph nodes along the node connection edge, and the access probability flowing into the same graph node is accumulated to obtain the current round of propagation access probability sequence. The current round of propagation access probability sequence is scaled according to the non-restart probability, and the restart probability is written into the corresponding position of the node to be evaluated to form the current round of access probability sequence; Calculate the probability difference between the current round's visit probability sequence and the previous round's visit probability sequence at the corresponding positions of each graph node, and sum the absolute values ​​of all probability differences. Compare the summed result with the convergence threshold. When the accumulated result is greater than the convergence threshold, the current round of access probability sequence is used as the input for the next round of propagation and the access probability allocation continues. When the cumulative result is not greater than the convergence threshold, the access probability sequence of each round is arranged in the order of the propagation rounds to obtain the associated propagation state sequence corresponding to the node to be evaluated.

[0012] Optionally, the convergence threshold is obtained by sequentially inputting the initial access probability sequence into the probability transition matrix for two rounds of propagation, obtaining the first round access probability sequence and the second round access probability sequence respectively, calculating the probability difference between the first round access probability sequence and the second round access probability sequence at the corresponding positions of each graph node, summing the absolute values ​​of all probability differences, and dividing the summation by the number of graph nodes.

[0013] Optionally, the evolutionary recombination module includes: The associated propagation state sequence is divided into sliding segments according to a preset window length and window step, and the states in each time window are arranged in chronological order to form a window state sequence. The deep Koopman operator is used to perform dimensionality-upgrading mapping on the state sequence of each window to obtain the corresponding dimensionality-upgrading state, and the state transition matrix is ​​calculated based on the dimensionality-upgrading state at adjacent time steps. The upgraded states are sequentially delayed and spliced ​​to form a Hankel matrix. Dynamic mode decomposition is then performed on the Hankel matrix to obtain multiple evolution modes. The dominant evolutionary modes are selected based on the modal amplitude and continuous retention length of each evolutionary mode, and the dominant evolutionary modes are recursively reorganized according to the state transition matrix to form the value evolutionary state.

[0014] Optionally, the determination of the evolutionary mode includes: Extract the upgraded states from multiple consecutive time points in chronological order, and then concatenate the upgraded states from adjacent time points in columns to form multiple sets of delayed state blocks. Arrange multiple sets of delayed state blocks into a Hankel matrix according to the delay order, and take the column blocks in the Hankel matrix except for the last column as the preorder matrix, and take the column blocks except for the first column as the postorder matrix; Perform singular value decomposition on the preorder matrix, arrange the singular values ​​in descending order, determine the boundary position based on the maximum decay interval between adjacent singular values, and extract the singular components corresponding to the boundary position to form a reduced-order state subspace. The preorder and postorder matrices are projected onto the reduced-order state subspace, and the low-dimensional evolution matrix is ​​determined based on the projected preorder and postorder matrices. Eigenvalue decomposition is performed on the low-dimensional evolution matrix to obtain the eigenvectors corresponding to each eigenvalue, and the eigenvectors are mapped back to the up-dimensional state space to form multiple evolution modes.

[0015] Optionally, the inversion evaluation module includes: Variational Bayesian inversion is performed on the value evolution state to obtain the posterior parameters corresponding to each state component. Based on the posterior parameters, each state component in the value evolution state is separated to form multiple posterior state components. The posterior parameters are the posterior mean and posterior variance corresponding to each state component. The posterior mean is used to determine the value center of the state component, and the posterior variance is used to determine the discrete range of the value of the state component. Extract the state value sequence of each posterior state component and the state value sequence of the value evolution state respectively, and calculate the maximum mean difference between each posterior state component and the value evolution state. Sort all maximum mean differences by numerical value, calculate the difference between adjacent maximum mean differences, take the smaller maximum mean difference corresponding to the boundary position with the largest difference as the screening threshold, and screen out the posterior state components with maximum mean differences greater than the screening threshold. Establish a constraint relationship with a sum of 1 for the posterior parameters corresponding to the retained posterior state components, and use the Lagrange multiplier method to solve for the posterior parameters that satisfy the constraint relationship; The posterior parameters obtained by solving are multiplied by the corresponding retained posterior state components, and the multiplication results are summed to obtain the value assessment result.

[0016] Optionally, the incremental update module includes: Obtain the feedback data corresponding to the value assessment result, extract the feedback value from the feedback data, and calculate the difference between the feedback value and the value assessment result to form the feedback residual. The feedback data is the actual observation data of the period corresponding to the value assessment result, which is used to characterize the deviation between the actual result and the assessment result. Locate the node to be evaluated in the value association graph, extract the node connection edges connected to the node to be evaluated, read the association weights corresponding to each node connection edge, and distribute the feedback residuals to each node connection edge according to the proportion of each association weight in the total association weights, forming an edge residual sequence. The current association weight corresponding to the edge connecting each node is used as the predicted association weight, the edge residual sequence is used as the observation bias, the incremental Kalman correction is used to calculate the Kalman gain corresponding to the edge connecting each node, and the association weight correction amount is obtained based on the Kalman gain and the observation bias. The correction values ​​of each association weight are superimposed on the current association weight of the corresponding node's connecting edge, replacing the original association weight, to obtain the updated value association graph.

[0017] The beneficial effects of this invention are: First, by performing unified normalization, temporal registration, and association mapping on multi-source heterogeneous data, this invention transforms the originally scattered and misaligned data records into a computable value association graph, reducing the interference of inconsistent data granularity and temporal position offset on the evaluation results, and improving the consistency of the evaluation input basis and the ability to express associations.

[0018] Secondly, this invention uses restarted random walks, deep Koopman operators, and dynamic mode decomposition to continuously solve the propagation and evolution states in value associations, so that value assessment no longer stops at static fitting, but can reflect the state propagation and time evolution process, thus enhancing the stability and temporal adaptability of the assessment results.

[0019] Finally, this invention performs incremental correction on the feedback information after the evaluation results are formed through variational Bayesian inversion and incremental Kalman correction, so that the value association graph can be continuously adjusted with new feedback, thereby improving the real-time performance of online evaluation and the ability to dynamically update subsequent results. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This invention presents a flowchart illustrating the modular structure of an online intelligent assessment system for urban land value based on big data analysis. Figure 2This is a flowchart illustrating the correlation propagation state solution of an online intelligent assessment system for urban land value based on big data analysis proposed in this invention. Figure 3 This is a flowchart illustrating the feedback incremental update process of an online intelligent assessment system for urban land value based on big data analysis, as proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-3 A smart online urban land value assessment system based on big data analysis includes: The data standardization module is used to acquire multi-source heterogeneous urban land data and preprocess it to form a standard data sequence. The association graphing module is used to perform temporal registration on standard data sequences. The registered data components are used as graph nodes. The module calculates the association weights between graph nodes and adjusts the node connection relationships to construct a value association graph. The propagation solution module is used to locate the node to be evaluated in the value association graph. It uses a random walk algorithm with restart to iteratively update the access probability along the node connection relationship, and obtains the association propagation state sequence corresponding to the node to be evaluated under the constraints of restart probability and convergence threshold. The evolution and recombination module is used to rearrange the associated propagation state sequence in a multi-scale time window. The deep Koopman operator is used to elevate the rearranged state trajectory to a linear evolution space. Combined with dynamic mode decomposition to strip away the dominant artificial modes and recursively recombine them, the value evolution state is obtained. The inversion evaluation module is used to perform variational Bayesian inversion on the value evolution state, use the maximum mean difference to screen out posterior state components with excessive deviation, and use the Lagrange multiplier method to calibrate the posterior parameters corresponding to the retained components to obtain the value evaluation results. The incremental update module is used to obtain feedback data on the value assessment results and perform incremental updates on the value association graph in conjunction with incremental Kalman correction.

[0023] In this embodiment, the data standardization module includes: acquiring multi-source heterogeneous data; renaming and rearranging similar fields in each source data according to a unified field order; aligning and merging the renamed data records according to the same identifier field and acquisition time; imputing missing values ​​in the aligned data records using adjacent time points and correcting outliers using the sliding median; and arranging the corrected data records according to a unified field order and time order to form a standard data sequence.

[0024] In this embodiment, the graph weighting module includes: Extract the time series corresponding to each data component from the standard data sequence, select a reference sequence, use the dynamic time warping method to calculate the alignment path of each time series relative to the reference sequence, and adjust the time position of each time series according to the alignment path to obtain the registered data components. The data component is a sequence of values ​​for the same field at different time positions in a standard data sequence; Using the registered data components as graph nodes, calculate the mutual information value between graph nodes, and use the mutual information value as the association weight between graph nodes; Remove node connection edges with association weights lower than a preset threshold, and retain node connection edges with association weights not lower than the preset threshold to form the initial node connection relationship; The preset threshold is formed by arranging all associated weights in descending order to form a weight sequence, calculating the difference between adjacent associated weights in the weight sequence, and selecting the smaller associated weight in the pair of adjacent associated weights with the largest difference as the preset threshold. The initial node connection relationships are divided using the Louvain community partitioning method, and the node connection edges within the same partition and between different partitions are adjusted based on the partitioning results to construct a value association graph.

[0025] In this embodiment, the construction of the value association graph includes: Each graph node in the initial node connection relationship is treated as an independent partition. The change in modularity of each graph node after being merged into the adjacent partition is calculated. The change in modularity is the difference between the increment of the weight of the node connection edge in the partition and the corresponding expected increment of the weight of the connection edge before and after the current graph node is merged into the adjacent partition. The adjacent partition with the largest change in modularity greater than zero is determined as the target partition of the current graph node, and the current graph node is assigned to the target partition. When the change in modularity is not greater than zero, keep the original partition of the current graph node unchanged, and perform the partitioning adjustment on all graph nodes in sequence to obtain the first partitioning result; Merge the graph nodes belonging to the same partition in the first partitioning result, recalculate the weights of the node connection edges between the merged graph nodes, and obtain new node connection relationships. Based on the new node connection relationships, the change in modularity of each graph node after merging into adjacent partitions is recalculated. The corresponding graph node is determined by the adjacent partition with the largest change in modularity that is greater than zero, until the partition to which each graph node belongs no longer changes. Retain the nodes within the same partition, and filter out the nodes whose association weights between different partitions are lower than a preset threshold to construct a value association graph.

[0026] In this embodiment, the propagation solution module includes: Locate the node to be evaluated in the value association graph, set the initial access probability of the node to be evaluated as the starting value, set the initial access probability of the other graph nodes to zero, and form an initial access probability sequence. The node to be evaluated is the graph node in the value association graph that corresponds to the data component to be evaluated. Read the node connection edge weights between each graph node, sum the weights of all node connection edges corresponding to each graph node, and divide the weight of each node connection edge by the sum of the weights of the node connection edges of the corresponding graph node to obtain the access probability ratio of each graph node along each node connection edge. When the sum of the weights of the edges connecting the nodes corresponding to a graph node is zero, the access probability of that graph node remains unchanged in its original position. Based on the proportion of the access probability propagated outward from each graph node along the node connection edge, the access probability in the previous round of access probability sequence is distributed to the adjacent graph nodes along the node connection edge, and the access probability flowing into the same graph node is accumulated to obtain the current round of propagation access probability sequence. The current round of propagation access probability sequence is scaled according to the non-restart probability, and the restart probability is written into the corresponding position of the node to be evaluated to form the current round of access probability sequence; Calculate the probability difference between the current round's visit probability sequence and the previous round's visit probability sequence at the corresponding positions of each graph node, and sum the absolute values ​​of all probability differences. Compare the summed result with the convergence threshold. When the accumulated result is greater than the convergence threshold, the current round of access probability sequence is used as the input for the next round of propagation and the access probability allocation continues. When the cumulative result is not greater than the convergence threshold, the access probability sequence of each round is arranged in the order of the propagation rounds to obtain the associated propagation state sequence corresponding to the node to be evaluated.

[0027] In this embodiment, the convergence threshold is obtained by sequentially inputting the initial access probability sequence into the probability transition matrix for two rounds of propagation, obtaining the first round access probability sequence and the second round access probability sequence respectively, calculating the probability difference between the first round access probability sequence and the second round access probability sequence at the corresponding positions of each graph node, summing the absolute values ​​of all probability differences, and dividing the summation by the number of graph nodes.

[0028] In this embodiment, the evolutionary recombination module includes: The associated propagation state sequence is divided into sliding segments according to a preset window length and window step, and the states in each time window are arranged in chronological order to form a window state sequence. The deep Koopman operator is used to perform dimensionality-upgrading mapping on the state sequences of each window to obtain the corresponding dimensionality-upgrading states. The state transition matrix is ​​then calculated based on the dimensionality-upgrading states at adjacent time points. Specifically, this includes: For each state in each window state sequence, matrix transformation and nonlinear activation are performed in chronological order to map each state to a high-dimensional feature vector, thereby obtaining the corresponding up-dimensional state at each time step. The up-dimensional state is a high-dimensional state column vector formed by up-dimensional mapping of each state at each time step. In each window state sequence, the up-dimensional states at adjacent time steps are paired up, and the up-dimensional state at the previous time step is taken as the previous state vector, and the up-dimensional state at the next time step is taken as the subsequent state vector. Arrange all the previous state vectors in column order to form the previous state matrix, and arrange all the subsequent state vectors corresponding to the previous state vectors in column order to form the subsequent state matrix; Calculate the product of the previous state matrix and its transpose, and calculate the product of the subsequent state matrix and the transpose of the previous state matrix; Multiply the product of the later state matrix and the transpose of the earlier state matrix by the inverse of the product of the earlier state matrix and its transpose to obtain the state transition matrix. The upgraded states are sequentially delayed and spliced ​​to form a Hankel matrix. Dynamic mode decomposition is then performed on the Hankel matrix to obtain multiple evolution modes. The dominant evolutionary modes are selected based on the modal amplitude and continuous retention length of each evolutionary mode, and the dominant evolutionary modes are recursively reorganized according to the state transition matrix to form the value evolutionary state.

[0029] In this embodiment, the determination of the evolution mode includes: Extract the upgraded states from multiple consecutive time points in chronological order, and then concatenate the upgraded states from adjacent time points in columns to form multiple sets of delayed state blocks. Arrange multiple sets of delayed state blocks into a Hankel matrix according to the delay order, and take the column blocks in the Hankel matrix except for the last column as the preorder matrix, and take the column blocks except for the first column as the postorder matrix; Perform singular value decomposition on the preorder matrix, arrange the singular values ​​in descending order, determine the boundary position based on the maximum decay interval between adjacent singular values, and extract the singular components corresponding to the boundary position to form a reduced-order state subspace. The preorder and postorder matrices are projected onto the reduced-order state subspace, and the low-dimensional evolution matrix is ​​determined based on the projected preorder and postorder matrices. Eigenvalue decomposition is performed on the low-dimensional evolution matrix to obtain the eigenvectors corresponding to each eigenvalue, and the eigenvectors are mapped back to the up-dimensional state space to form multiple evolution modes.

[0030] In this embodiment, the inversion evaluation module includes: Variational Bayesian inversion is performed on the value evolution state to obtain the posterior parameters corresponding to each state component. Based on the posterior parameters, each state component in the value evolution state is separated to form multiple posterior state components. The posterior parameters are the posterior mean and posterior variance corresponding to each state component. The posterior mean is used to determine the value center of the state component, and the posterior variance is used to determine the discrete range of the value of the state component. Extract the state value sequence of each posterior state component and the state value sequence of the value evolution state respectively, and calculate the maximum mean difference between each posterior state component and the value evolution state. Sort all maximum mean differences by numerical value, calculate the difference between adjacent maximum mean differences, take the smaller maximum mean difference corresponding to the boundary position with the largest difference as the screening threshold, and screen out the posterior state components with maximum mean differences greater than the screening threshold. Establish a constraint relationship with a sum of 1 for the posterior parameters corresponding to the retained posterior state components, and use the Lagrange multiplier method to solve for the posterior parameters that satisfy the constraint relationship; The posterior parameters obtained by solving are multiplied by the corresponding retained posterior state components, and the multiplication results are summed to obtain the value assessment result.

[0031] In this embodiment, the incremental update module includes: Obtain the feedback data corresponding to the value assessment result, extract the feedback value from the feedback data, and calculate the difference between the feedback value and the value assessment result to form the feedback residual. The feedback data is the actual observation data of the period corresponding to the value assessment result, which is used to characterize the deviation between the actual result and the assessment result. Locate the node to be evaluated in the value association graph, extract the node connection edges connected to the node to be evaluated, read the association weights corresponding to each node connection edge, and distribute the feedback residuals to each node connection edge according to the proportion of each association weight in the total association weights, forming an edge residual sequence. The current association weight corresponding to the edge connecting each node is used as the predicted association weight, the edge residual sequence is used as the observation bias, the incremental Kalman correction is used to calculate the Kalman gain corresponding to the edge connecting each node, and the association weight correction amount is obtained based on the Kalman gain and the observation bias. The correction values ​​of each association weight are superimposed on the current association weight of the corresponding node's connecting edge, replacing the original association weight, to obtain the updated value association graph.

[0032] Example 1: To verify the feasibility of this invention in practice, it was applied to a digital land value assessment scenario in a city. In this scenario, the original assessment process mainly relied on periodic summaries and static valuation models, which were insufficient in handling field differences, time misalignments, and feedback lags in multi-source heterogeneous data. The time positions of the same assessment object in records from different sources often could not correspond, resulting in an unstable mapping foundation and a significant deviation between the assessment results and actual observations. Furthermore, when new feedback data entered, it was often necessary to re-execute the entire batch calculation, indicating insufficient online assessment capabilities.

[0033] To address the aforementioned issues, various types of business data were collected in this scenario, totaling 182,436 original records with 137 different ways of expressing field names. The data normalization module first performed unified field rearrangement, record alignment, missing data correction, and chronological ordering on the multi-source data, compressing and merging field names into 48 unified fields to obtain a standard data sequence. Then, the association graphing module performed temporal registration on the standard data sequence, using the registered data components as graph nodes. The association weights between nodes were calculated, and the node connections were adjusted to form a value association graph. After processing, the average time position offset decreased from 5.2 time slices to 0.8 time slices, significantly improving the consistency of the graphing input.

[0034] In practical applications, the propagation solution module locates the nodes to be evaluated in the value association graph, uses the weights of the nodes' connecting edges to form the basis of probability transitions, and performs a random walk with restart probability. After 17 rounds, the set convergence condition is met, resulting in the corresponding association propagation state sequence. Subsequently, the evolution and recombination module performs multi-scale time window segmentation on the association propagation state sequence, uses the deep Koopman operator to complete the dimensionality increase mapping, and obtains multiple evolutionary modes through Hankel matrix and dynamic mode decomposition. Finally, four dominantly directed modes are retained to form the value evolution state.

[0035] The inversion evaluation module then performs variational Bayesian inversion on the value evolution state, filters out posterior state components with excessive deviations by using the maximum mean difference, and solves and synthesizes the posterior parameters corresponding to the retained components to output the value evaluation results. During continuous operation, each batch of feedback data averages 126 records. The incremental update module directly uses the feedback residuals to perform incremental Kalman correction on the association weights in the value association graph, without needing to re-execute full graph construction and full propagation.

[0036] In the same scenario, the present invention was compared and verified with the conventional static evaluation process and the graph evaluation process without incremental updates. The results are shown in Table 1. The present invention performs better in terms of field alignment completeness, evaluation error, update time, and continuous evaluation stability. Specifically, the average absolute error of value evaluation was reduced to 0.046, the average relative error was reduced to 4.3%, the time for a single round of feedback update was shortened to 9 minutes, and the fluctuation range of three consecutive rounds of evaluation was controlled within 2.1%.

[0037] Table 1 Comparison of Online Intelligent Valuation Results for Urban Land

[0038] As can be seen from the data in Table 1, the present invention can maintain the computability of the value association graph and the continuity of the evaluation results under the condition of continuous input of multi-source data, and is applicable to the scenario of intelligent online value evaluation of urban land.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart online urban land value assessment system based on big data analysis, characterized in that, include: The data standardization module is used to acquire multi-source heterogeneous urban land data and preprocess it to form a standard data sequence. The association graphing module is used to perform temporal registration on standard data sequences. The registered data components are used as graph nodes. The module calculates the association weights between graph nodes and adjusts the node connection relationships to construct a value association graph. The propagation solution module is used to locate the node to be evaluated in the value association graph. It uses a random walk algorithm with restart to iteratively update the access probability along the node connection relationship, and obtains the association propagation state sequence corresponding to the node to be evaluated under the constraints of restart probability and convergence threshold. The evolution and recombination module is used to rearrange the associated propagation state sequence in a multi-scale time window. The deep Koopman operator is used to elevate the rearranged state trajectory to a linear evolution space. Combined with dynamic mode decomposition to strip away the dominant artificial modes and recursively recombine them, the value evolution state is obtained. The inversion evaluation module is used to perform variational Bayesian inversion on the value evolution state, use the maximum mean difference to screen out posterior state components with excessive deviation, and use the Lagrange multiplier method to calibrate the posterior parameters corresponding to the retained components to obtain the value evaluation results. The incremental update module is used to obtain feedback data on the value assessment results and perform incremental updates on the value association graph in conjunction with incremental Kalman correction.

2. The intelligent online land value assessment system based on big data analysis for urban areas as described in claim 1, characterized in that, The data standardization module includes: acquiring multi-source heterogeneous data; renaming and rearranging similar fields in each source data according to a unified field order; aligning and merging the renamed data records according to the same identifier field and acquisition time; imputing missing values ​​in the aligned data records using adjacent time points and correcting outliers using the sliding median; and arranging the corrected data records according to a unified field order and time order to form a standard data sequence.

3. The intelligent online urban land value assessment system based on big data analysis according to claim 1, characterized in that, The composition weighting module includes: Extract the time series corresponding to each data component from the standard data sequence, select a reference sequence, use the dynamic time warping method to calculate the alignment path of each time series relative to the reference sequence, and adjust the time position of each time series according to the alignment path to obtain the registered data components. The data component is a sequence of values ​​for the same field at different time positions in a standard data sequence; Using the registered data components as graph nodes, calculate the mutual information value between graph nodes, and use the mutual information value as the association weight between graph nodes; Remove node connection edges with association weights lower than a preset threshold, and retain node connection edges with association weights not lower than the preset threshold to form the initial node connection relationship; The initial node connection relationships are divided using the Louvain community partitioning method, and the node connection edges within the same partition and between different partitions are adjusted based on the partitioning results to construct a value association graph.

4. The intelligent online urban land value assessment system based on big data analysis according to claim 3, characterized in that, The construction of the value association graph includes: Each graph node in the initial node connection relationship is treated as an independent partition. The change in modularity of each graph node after being merged into the adjacent partition is calculated. The change in modularity is the difference between the increment of the weight of the connection edge of the node in the partition and the corresponding expected increment of the weight of the connection edge before and after the current graph node is merged into the adjacent partition. The adjacent partition with the largest change in modularity greater than zero is determined as the target partition of the current graph node, and the current graph node is assigned to the target partition. When the change in modularity is not greater than zero, keep the original partition of the current graph node unchanged, and perform the partitioning adjustment on all graph nodes in sequence to obtain the first partitioning result; Merge the graph nodes belonging to the same partition in the first partitioning result, recalculate the weights of the node connection edges between the merged graph nodes, and obtain new node connection relationships. Based on the new node connection relationships, the change in modularity of each graph node after merging into adjacent partitions is recalculated. The corresponding graph node is determined by the adjacent partition with the largest change in modularity that is greater than zero, until the partition to which each graph node belongs no longer changes. Retain the nodes within the same partition, and filter out the nodes whose association weights between different partitions are lower than a preset threshold to construct a value association graph.

5. The intelligent online land value assessment system based on big data analysis for urban areas according to claim 1, characterized in that, The propagation solution module includes: Locate the node to be evaluated in the value association graph, set the initial access probability of the node to be evaluated as the starting value, set the initial access probability of the other graph nodes to zero, and form an initial access probability sequence. The node to be evaluated is the graph node in the value association graph that corresponds to the data component to be evaluated. Read the node connection edge weights between each graph node, sum the weights of all node connection edges corresponding to each graph node, and divide the weight of each node connection edge by the sum of the weights of the node connection edges of the corresponding graph node to obtain the access probability ratio of each graph node along each node connection edge. When the sum of the weights of the edges connecting the nodes corresponding to a graph node is zero, the access probability of that graph node remains unchanged in its original position. Based on the proportion of the access probability propagated outward from each graph node along the node connection edge, the access probability in the previous round of access probability sequence is distributed to the adjacent graph nodes along the node connection edge, and the access probability flowing into the same graph node is accumulated to obtain the current round of propagation access probability sequence. The current round of propagation access probability sequence is scaled according to the non-restart probability, and the restart probability is written into the corresponding position of the node to be evaluated to form the current round of access probability sequence; Calculate the probability difference between the current round's visit probability sequence and the previous round's visit probability sequence at the corresponding positions of each graph node, and sum the absolute values ​​of all probability differences. Compare the summed result with the convergence threshold. When the accumulated result is greater than the convergence threshold, the current round of access probability sequence is used as the input for the next round of propagation and the access probability allocation continues. When the cumulative result is not greater than the convergence threshold, the access probability sequence of each round is arranged in the order of the propagation rounds to obtain the associated propagation state sequence corresponding to the node to be evaluated.

6. The intelligent online land value assessment system based on big data analysis according to claim 5, characterized in that, The convergence threshold is obtained by sequentially inputting the initial access probability sequence into the probability transition matrix for two rounds of propagation, obtaining the first round access probability sequence and the second round access probability sequence respectively, calculating the probability difference between the first round access probability sequence and the second round access probability sequence at the corresponding positions of each graph node, summing the absolute values ​​of all probability differences, and dividing the summation by the number of graph nodes.

7. The intelligent online land value assessment system based on big data analysis for urban areas according to claim 1, characterized in that, The evolutionary recombination module includes: The associated propagation state sequence is divided into sliding segments according to a preset window length and window step, and the states in each time window are arranged in chronological order to form a window state sequence. The deep Koopman operator is used to perform dimensionality-upgrading mapping on the state sequence of each window to obtain the corresponding dimensionality-upgrading state, and the state transition matrix is ​​calculated based on the dimensionality-upgrading state at adjacent time steps. The upgraded states are sequentially spliced ​​with delay to form a Hankel matrix, and dynamic mode decomposition is performed on the Hankel matrix to obtain multiple evolution modes; The dominant evolutionary modes are selected based on the modal amplitude and continuous retention length of each evolutionary mode, and the dominant evolutionary modes are recursively reorganized according to the state transition matrix to form the value evolutionary state.

8. The intelligent online land value assessment system based on big data analysis according to claim 7, characterized in that, The determination of the evolutionary mode includes: Extract the upgraded states from multiple consecutive time points in chronological order, and then concatenate the upgraded states from adjacent time points in columns to form multiple sets of delayed state blocks. Arrange multiple sets of delayed state blocks into a Hankel matrix according to the delay order, and take the column blocks in the Hankel matrix except for the last column as the preorder matrix, and take the column blocks except for the first column as the postorder matrix; Perform singular value decomposition on the preorder matrix, arrange the singular values ​​in descending order, determine the boundary position based on the maximum decay interval between adjacent singular values, and extract the singular components corresponding to the boundary position to form a reduced-order state subspace. The preorder and postorder matrices are projected onto the reduced-order state subspace, and the low-dimensional evolution matrix is ​​determined based on the projected preorder and postorder matrices. Eigenvalue decomposition is performed on the low-dimensional evolution matrix to obtain the eigenvectors corresponding to each eigenvalue, and the eigenvectors are mapped back to the up-dimensional state space to form multiple evolution modes.

9. The intelligent online land value assessment system based on big data analysis for urban areas according to claim 1, characterized in that, The inversion evaluation module includes: Variational Bayesian inversion is performed on the value evolution state to obtain the posterior parameters corresponding to each state component. Based on the posterior parameters, each state component in the value evolution state is separated to form multiple posterior state components. The posterior parameters are the posterior mean and posterior variance corresponding to each state component. The posterior mean is used to determine the value center of the state component, and the posterior variance is used to determine the discrete range of the value of the state component. Extract the state value sequence of each posterior state component and the state value sequence of the value evolution state respectively, and calculate the maximum mean difference between each posterior state component and the value evolution state. Sort all maximum mean differences by numerical value, calculate the difference between adjacent maximum mean differences, take the smaller maximum mean difference corresponding to the boundary position with the largest difference as the screening threshold, and screen out the posterior state components with maximum mean differences greater than the screening threshold. Establish a constraint relationship with a sum of 1 for the posterior parameters corresponding to the retained posterior state components, and use the Lagrange multiplier method to solve for the posterior parameters that satisfy the constraint relationship; The posterior parameters obtained by solving are multiplied by the corresponding retained posterior state components, and the multiplication results are summed to obtain the value assessment result.

10. The intelligent online land value assessment system based on big data analysis for urban areas according to claim 1, characterized in that, The incremental update module includes: Obtain the feedback data corresponding to the value assessment result, extract the feedback value from the feedback data, and calculate the difference between the feedback value and the value assessment result to form the feedback residual. The feedback data is the actual observation data of the period corresponding to the value assessment result, which is used to characterize the deviation between the actual result and the assessment result. Locate the node to be evaluated in the value association graph, extract the node connection edges connected to the node to be evaluated, read the association weights corresponding to each node connection edge, and distribute the feedback residuals to each node connection edge according to the proportion of each association weight in the total association weights, forming an edge residual sequence. The current association weight corresponding to the edge connecting each node is used as the predicted association weight, the edge residual sequence is used as the observation bias, the incremental Kalman correction is used to calculate the Kalman gain corresponding to the edge connecting each node, and the association weight correction amount is obtained based on the Kalman gain and the observation bias. The correction values ​​of each association weight are superimposed on the current association weight of the corresponding node's connecting edge, replacing the original association weight, to obtain the updated value association graph.

Images