Digital management and control method and system for collection and removal whole process based on multi-source data fusion

Abstract

The invention provides a multi-source data fusion-based collection and demolition whole-process digital management and control method and system, and the method comprises the steps: constructing a project genetic map feature space, and generating a typical project genotype knowledge base through DBSCAN clustering; after the optimal genotype of the new project is matched through the Wasserstein distance, a Transformer-Time2Seq model is activated to predict the fund use trend; when the deviation between the predicted expenditure and the actual expenditure exceeds a threshold value, triggering a context-aware gating mechanism to dynamically adjust the attention weight and the elastic coefficient of the cost subitem; generating a multi-scene budget adjustment suggestion package based on the correction result, and outputting a confidence ranking strategy through semantic matching; and all key data are stored through a block chain smart contract, so that tampering resistance and traceability are ensured.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and anomaly detection technology for urban stormwater pipe networks, and in particular to a digital management and control method and system for the entire process of land acquisition and demolition based on multi-source data fusion. Background Technology

[0002] Currently, the mainstream technical solutions in the field of dynamic forecasting and intelligent budget adjustment of land acquisition and demolition funds mainly adopt time series analysis, business rule modeling, and traditional linear regression methods to predict the trend of land acquisition and demolition fund usage and prepare budgets. These technologies usually treat projects as homogeneous objects, directly applying a uniform mathematical model or simple parameter configuration, failing to fully consider the significant differences in land acquisition and demolition projects in terms of regional attributes, property rights structure, policy orientation, etc., resulting in insufficient model generalization ability. With the advancement of smart cities and digital management and control of government affairs, industry technologies have gradually introduced multi-source data fusion, end-to-end neural networks, and a certain degree of online feedback mechanisms to improve the dynamic responsiveness of fund forecasting, but there are still obvious shortcomings in terms of project type adaptability. The main problems with existing technologies are as follows: (1) Poor model adaptability. Existing technologies mostly use fixed parameters or unified modeling methods to predict the capital trend of all demolition projects. They lack the ability to dynamically adapt to different project types and have weak model generalization. When faced with project heterogeneity or sudden changes in the policy environment, they are prone to inaccurate predictions. (2) Lack of heterogeneous factors. Mainstream technologies often focus only on traditional progress, indicators and historical averages, and fail to structurally model core features that affect capital expenditure, such as regional development index, property rights complexity score and resettlement method preference. The logic of capital prediction is simplistic. (3) Lack of intelligent feedback and closed-loop optimization mechanism. Existing methods mainly rely on manual intervention or post-event correction to deal with prediction errors. They lack automatic detection of anomalies and adaptive adjustment of model parameters, making it difficult to cope with the real-time changes in the use of funds during the demolition process. (4) The budget adjustment suggestions are not intelligent enough. Most methods can only provide budget increase suggestions based on fixed ratios or historical experience. For complex scenarios such as actual demolition progress, fluctuations in signing rate, and the impact of policy events, there is a lack of high-confidence, multi-scenario, and intelligent budget adjustment recommendations. (5) Poor process traceability: Current technology lacks an efficient traceability and evidence preservation mechanism in the process of model parameter changes, fund forecasting and budget adjustment strategy execution, and cannot effectively support process auditing and business compliance requirements. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, this invention provides a digital management and control method for the entire process of land acquisition and demolition based on multi-source data fusion.

[0004] The technical solution of this invention is implemented as follows: A method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion, comprising: S1: Based on the regional development index, housing density gradient, property rights complexity score, resettlement method preference distribution, historical signing rate curve shape and compensation standard fluctuation range parameters of the expropriation and demolition project, construct the project gene map feature vector space, and map historical project data to this space to form a gene feature matrix. S2: Execute the DBSCAN clustering algorithm on the gene feature matrix to identify typical project genotypes with similar fund expenditure rhythms and generate a genotype knowledge base containing genotype labels and corresponding cost composition rules. S3: Input the new project establishment data into the gene map feature vector space, calculate its Wasserstein distance with each genotype, and activate the corresponding Transformer-Time2Seq benchmark prediction model after matching the optimal reference genotype. The input of this model includes a multi-source feature tensor containing project progress indicators, phased fund expenditure sequences and policy adjustment event codes. S4: Based on the deviation between the predicted value of fund usage trend output by the benchmark prediction model and the actual expenditure curve, when the deviation exceeds the dynamic threshold, the context-aware gating mechanism is triggered. By adjusting the attention weight distribution of each cost sub-item in the Transformer decoder, an elastic coefficient correction parameter adapted to the characteristics of the current project is generated. S5: Construct a multi-scenario budget adjustment suggestion package based on the corrected prediction results, and use a rule reasoning engine to semantically match the project stage goal priority labels with the applicable condition labels in the suggestion package to generate a budget adjustment strategy recommendation set with confidence ranking; S6: Write the genotype matching record, model parameter update log, and budget adjustment strategy recommendation set into the distributed ledger through a blockchain smart contract to generate an immutable evidence record containing timestamps and operation traceability information.

[0005] The present invention also provides a digital management and control system for the entire process of land acquisition and demolition based on multi-source data fusion, which adopts the above-mentioned digital management and control method for the entire process of land acquisition and demolition based on multi-source data fusion for digital management and control of land acquisition and demolition.

[0006] The present invention provides a digital management and control method and system for the entire process of land acquisition and demolition based on multi-source data fusion, which has the following beneficial effects: (1) By introducing the concept of “project gene” and constructing a quantifiable feature vector system, this invention abstracts the core attributes of expropriation and demolition projects into a structured representation of multi-dimensional indicators such as regional development index, property rights complexity score, and resettlement preference distribution. This effectively overcomes the shortcomings of traditional cost prediction methods in identifying heterogeneity among projects. On this basis, clustering algorithms are used to extract typical “genotypes” from historical projects to form a project template library with regular expenditure patterns and cost composition characteristics. This allows new projects to quickly locate the closest reference model based on their preliminary data through similarity matching at the initial stage of their launch, significantly improving the accuracy and rationality of the prediction starting point. Compared with the general time series model or static regression method commonly used in existing technologies, this invention realizes the transformation from “generalization” to “tailored approach”, greatly enhancing the adaptability of the prediction model to the evolution path of projects in different regions, policies and property rights environments. (2) This invention adopts an improved Transformer-Time2Seq hybrid architecture as the core prediction engine, which integrates the ability to model long-term time series dependencies with the nonlinear response mechanism of external driving factors. It can simultaneously capture the time dynamics of endogenous variables of projects such as signing rate and assessment progress, as well as the impact of exogenous shocks such as policy adjustment events and macroeconomic fluctuations, further improving the prediction accuracy and timeliness. It also designs a context-aware gating mechanism, which automatically triggers the local parameter reweighting module when a significant deviation between actual expenditure and predicted value is detected, dynamically adjusting the growth elasticity coefficient of key cost sub-items such as compensation fee and relocation fee, realizing online adaptive optimization of the model, effectively alleviating the prediction instability problem of traditional fixed parameter models when facing sudden situations or execution deviations, and generating multi-scenario adjustment suggestion packages in combination with budget prediction results. It also intelligently matches the project stage target priority with the rule reasoning engine, realizing closed-loop linkage from prediction output to decision support, significantly improving the foresight and strategy flexibility of fund allocation. (3) By constructing a full-chain technical framework of "gene extraction - matching activation - dynamic prediction - budget recommendation - on-chain evidence storage", a highly traceable and trustworthy intelligent budget assistance system has been formed. All gene matching results, model update logs, and budget change trajectories are stored on the blockchain to ensure that the entire process is traceable and tamper-proof, meeting the strict requirements of transparency and compliance for major livelihood projects. This system not only achieves accurate prediction and adaptive response at the technical level, but also establishes a standardized and reusable knowledge accumulation mechanism at the system level, supporting cross-regional and cross-cycle experience transfer and continuous model evolution. Attached Figure Description

[0007] Figure 1 This is a flowchart of a digital management and control method for the entire process of land acquisition and demolition based on multi-source data fusion, according to the present invention. Figure 2This is a sub-flowchart of a digital management and control method for the entire process of land acquisition and demolition based on multi-source data fusion, according to the present invention. Figure 3 This is another sub-flowchart of the present invention, which is a digital management and control method for the entire process of land acquisition and demolition based on multi-source data fusion. Detailed Implementation

[0008] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0009] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0010] like Figure 1 As shown, this invention provides a method and system for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion, specifically including: S1: Based on the regional development index, housing density gradient, property rights complexity score, resettlement method preference distribution, historical signing rate curve shape and compensation standard fluctuation range parameters of the expropriation and demolition project, construct the project gene map feature vector space, and map historical project data to this space to form a gene feature matrix. S2: Execute the DBSCAN clustering algorithm on the gene feature matrix to identify typical project genotypes with similar fund expenditure rhythms and generate a genotype knowledge base containing genotype labels and corresponding cost composition rules. S3: Input the new project establishment data into the gene map feature vector space, calculate its Wasserstein distance with each genotype, and activate the corresponding Transformer-Time2Seq benchmark prediction model after matching the optimal reference genotype. The input of this model includes a multi-source feature tensor containing project progress indicators, phased fund expenditure sequences and policy adjustment event codes. S4: Based on the deviation between the predicted value of fund usage trend output by the benchmark prediction model and the actual expenditure curve, when the deviation exceeds the dynamic threshold, the context-aware gating mechanism is triggered. By adjusting the attention weight distribution of each cost sub-item in the Transformer decoder, an elastic coefficient correction parameter adapted to the characteristics of the current project is generated. S5: Construct a multi-scenario budget adjustment suggestion package based on the corrected prediction results, and use a rule reasoning engine to semantically match the project stage goal priority labels with the applicable condition labels in the suggestion package to generate a budget adjustment strategy recommendation set with confidence ranking; S6: Write the genotype matching record, model parameter update log, and budget adjustment strategy recommendation set into the distributed ledger through a blockchain smart contract to generate an immutable evidence record containing timestamps and operation traceability information.

[0011] Step S1: Based on the regional development index, housing density gradient, property rights complexity score, resettlement method preference distribution, historical signing rate curve shape, and compensation standard fluctuation range parameters of the expropriation and demolition project, a project gene map feature vector space is constructed, and historical project data is mapped to this space to form a gene feature matrix. Specifically, this includes: S1.1: Standardize the regional development index of the expropriation and demolition project to eliminate the interference of economic differences in different regions on subsequent modeling and obtain the standardized value of the regional development index. Data cleaning methods (parameters: missing value filling rule = mean filling, outlier detection threshold = 3 times standard deviation) are used to clean the regional development index data of expropriation and demolition projects to achieve completeness and consistency of the original regional development-related indicators. Furthermore, through a standardization processing algorithm (parameter: method = Z-score standardization), the numerical transformation of regional development indices in different regions is achieved on a unified scale, and a standardized index dataset with the influence of units removed is obtained. Furthermore, the standardized value of the regional development index is calculated using the following formula: in, For the first The region's standardized regional development index For the first The region's original regional development index. The mean of the regional development index of the sample. The standard deviation of the regional development index for the sample; Furthermore, using a normalization algorithm (parameters: method = Min-Max normalization, range setting = [0,1]), the standardized regional development index value is mapped to the interval [0,1], and a normalized regional development score is generated so that it can be concatenated with other feature vectors in the feature space; Furthermore, the correlation between the regional development index and the fund usage pattern is quantified through correlation analysis (parameter: Spearman correlation coefficient calculation). A correlation scoring index with a correlation threshold greater than 0.5 is selected for subsequent feature importance assessment. Through the above standardization and normalization processing methods, the original regional development index results output in the previous step are transformed into standardized values ​​that can be directly used for the construction of the project gene map feature vector space, thereby achieving the technical effect of eliminating the differences in economic levels among different regions. For example, considering land acquisition and demolition projects in three districts of a city, the original regional development indices were 87.5, 65.2, and 102.8, respectively. Missing values ​​were filled in using the mean to obtain the complete dataset. The standardized value for each region was calculated using the Z-score standardization method, where the overall mean μ = 85.17 and the standard deviation σ = 15.42. The standardized value for the first district was calculated using the formula. ≈0.151, the standardized value for the second zone is 0.151. ≈ The standardized value of the third zone is ≈1.146. Then perform Min-Max normalization, setting the minimum value to -1.295 and the maximum value to 1.146. The normalized value for the first region is... ≈0.543. According to Spearman correlation coefficient calculation, the correlation score between this normalized value and the specific fund usage pattern is 0.72, which is significantly higher than the association threshold of 0.5, thus meeting the conditions for inclusion in the project's genomic map feature space; S1.2: Based on the house density gradient data, perform a spatial interpolation algorithm to generate a house density map under continuous spatial distribution and output the house density gradient feature vector; S1.3: Based on the property rights complexity scoring rules, the property rights types, number of ownership disputes, and clarity of property rights ownership involved in the project are weighted and calculated to quantify property rights complexity and generate property rights complexity scoring indicators. S1.4: Using a statistical model of resettlement method preference distribution, the proportion of resettlement methods chosen by residents in historical projects is classified and coded to extract the feature vector of resettlement method preference distribution; S1.5: Based on historical signing rate curve shape data, the Dynamic Time Warping (DTW) algorithm is used to normalize the curve in order to extract standardized historical signing rate shape features. S1.6: Based on the parameters of the compensation standard floating range, and in conjunction with policy documents and market fluctuation data, construct a compensation standard floating coefficient matrix to quantify the dynamic range of changes in the compensation standard; S1.7: The standardized values ​​of the regional development index, the feature vector of housing density gradient, the property rights complexity scoring index, the feature vector of resettlement method preference distribution, the morphological characteristics of historical signing rate, and the compensation standard floating coefficient matrix are spliced ​​together to construct the feature vector space of the project gene map. The input data includes standardized values ​​of regional development index, housing density gradient feature vector, property rights complexity scoring index, resettlement method preference distribution feature vector, historical contract signing rate morphological characteristics, and compensation standard floating coefficient matrix obtained through steps S1.1 to S1.6, all of which are in structured numerical or vector form. A feature vector concatenation algorithm (parameters: setting the concatenation order according to the feature dimension consistency principle and keeping the feature index position stable) is used to concatenate various heterogeneous features dimension by dimension according to a predetermined sequence and form a unified high-dimensional feature vector. Furthermore, by using a vector dimension mapping method (parameter: mapping using the feature domain label index table), the ordered arrangement of features from different sources in the corresponding dimensions of the concatenated vector is achieved, and a comprehensive vector structure containing all feature components is obtained. Furthermore, through the feature domain normalization fusion process (parameter: Z-score normalization, performing a standardization transformation with a mean of 0 and a variance of 1 on each dimension), the numerical scale of each feature is unified, providing numerical stability for distance calculation in subsequent cluster analysis; Furthermore, the correlation within the fused feature vector is tested using a covariance consistency test algorithm (parameter: Pearson correlation coefficient between features calculated based on the feature covariance matrix), and a covariance consistency judgment index is generated. By using the feature space modeling method (parameter: Euclidean space embedding model), the fused feature vectors from the previous step are mapped to a continuous high-dimensional feature vector space to form the project gene map feature vector space, thereby enabling the comparable representation of the structural features of different projects. For example, in a certain practical application scenario, the standardized value of the regional development index is... The house density gradient eigenvector has a length of The sequence of numbers, the property rights complexity scoring index is The feature vector of the resettlement method preference distribution includes Similar to proportional data, the historical signing rate morphology is characterized by a normalized length. The sequence of numbers, the compensation standard fluctuation coefficient matrix is ​​as follows × A matrix is ​​formed by sequentially concatenating the aforementioned features using an eigenvector concatenation algorithm to create a matrix of length. The high-dimensional vector, after Z-score normalization, has its numerical values ​​for each dimension stabilized within [-]. , Between ] . The average correlation coefficient was calculated using the covariance consistency test algorithm. It meets the preset correlation threshold upper limit. The requirements are met. Finally, the normalized high-dimensional vector is embedded into the gene map feature vector space to form spatial points that can be used for subsequent clustering, achieving high distinguishability of the fused features in spatial structure; S1.8: Map the data samples of historical expropriation and demolition projects to the project's gene map feature vector space to form a gene feature matrix, providing input for subsequent cluster analysis.

[0012] Step S2: Perform the DBSCAN clustering algorithm on the gene feature matrix to identify typical project genotypes with similar funding expenditure rhythms, generating a genotype knowledge base containing genotype labels and corresponding cost composition patterns. Specifically, this includes: S2.1: Based on historical project data in the project gene map feature vector space, construct a gene feature matrix, where each row corresponds to the feature vector of a historical project, including regional development index, housing density gradient, property rights complexity score, resettlement method preference distribution, historical signing rate curve shape and compensation standard floating range parameters, to form structured data input; Based on historical project data in the project gene map feature vector space, a structured data mapping method (parameters: standardized value of regional development index, housing density gradient feature vector, property rights complexity scoring index, resettlement method preference distribution feature vector, historical signing rate morphological characteristics, compensation standard floating coefficient matrix) is adopted to achieve unified vectorized encoding of multidimensional features of historical projects. Furthermore, by using a feature dimension alignment algorithm (parameters: longest feature dimension benchmark, missing feature filling rule), the feature vectors of different historical projects are made completely consistent in dimension and position, and a multi-project feature set with strong comparability is obtained. Furthermore, a hybrid data standardization and normalization method (parameters: Z-score standardization, Min-Max normalization) is adopted to achieve uniform scaling of the full-dimensional feature vector in terms of dimensions and generate a standardized vector set with cluster analysis input conditions; Furthermore, through data cleaning and outlier removal algorithms (parameters: three standard deviation threshold detection, quantile method), outlier samples in the feature vector set are identified and removed, and historical project feature data with high stability is obtained. By using matrix construction, the standardized feature vectors of the above historical projects are arranged into rows to form a gene feature matrix, thus realizing the structured data preparation for the subsequent DBSCAN algorithm input and the identification of similar fund expenditure rhythms. For example, in processing 50 historical expropriation and demolition projects, the regional development index, after Z-score standardization, ranges from -2.1 to 1.8; the housing density gradient feature vector, after Min-Max normalization, ranges from 0 to 1; the property rights complexity score, calculated according to rules and normalized, ranges from 0.15 to 0.93; the resettlement method preference distribution feature vector, after proportional encoding and normalization, ranges from 0.05 to 0.95; the historical signing rate curve morphology feature, after DTW normalization, ranges from 0.03 to 0.97; and the compensation standard fluctuation coefficient matrix, after normalization, ranges from 0.02 to 0.88. A missing feature imputation rule is used to fill in any missing parameters with the mean of their corresponding dimensions, ensuring that all feature vectors have consistent dimensions. After outlier removal, the remaining 47 sets of project feature vectors constitute a 47×6 gene feature matrix, where each column represents a feature dimension and each row corresponds to a project feature set. This matrix, as input for DBSCAN clustering, significantly improves the accuracy and stability of genotype identification for subsequent similar projects; S2.2: Perform DBSCAN clustering algorithm on the gene feature matrix, set the neighborhood radius ε and minimum sample number MinPts parameters, calculate the similarity between feature vectors based on Euclidean distance metric, identify densely connected data clusters, and divide historical project groups with similar fund expenditure rhythms; S2.3: For each clustering result, feature summarization and label naming are performed. Based on the mean of the feature vector of the cluster center, the core feature combination of the project is extracted to generate the corresponding genotype label, such as 'high density low compensation type', 'low signing rate high volatility type', etc., so as to form an interpretable project genotype classification system. S2.4: For each genotype tag corresponding to the project group, calculate its cost composition pattern statistical parameters, including the average proportion, standard deviation, and expenditure rhythm curve morphology parameters of each cost sub-item (such as compensation fee, relocation fee, and bonus), in order to form structured cost composition pattern description data; For the project group fund expenditure data sample corresponding to each genotype label obtained from cluster analysis, the sub-item statistical method (parameter: list of cost sub-item types) is used to extract the fund proportion vector of each cost sub-item; Furthermore, by using the mean calculation method (parameter: total sample size N), the expected value of the proportion of each cost sub-item is estimated, and the average proportion index data is obtained; the mean calculation formula is: in Let i be the cost percentage of the i-th sample. The total number of samples; Furthermore, by using the sample standard deviation calculation method (parameter: average proportion μ), the dispersion of the proportion distribution of each cost sub-item is quantified, and a standard deviation index is generated; the standard deviation calculation formula is as follows: Furthermore, by using time series morphological analysis methods (parameter: set of time indexes for fund expenditures), the morphological parameters of the expenditure rhythm curves of each cost sub-item are extracted, and morphological statistics such as periodic indices and volatility coefficients are obtained. Furthermore, through feature combination and structured encapsulation processing, the average proportion, standard deviation, and expenditure rhythm pattern parameters are transformed into structured cost composition pattern description data, thereby realizing the persistence and callability of cost pattern features. Through the above-mentioned itemized statistics and morphological analysis algorithms, the project group expenditure data results of the previous step are transformed into structured cost composition pattern indicators with statistical characteristics and time series pattern compatibility, so as to achieve the basic data guarantee required for the subsequent construction of the genotype knowledge base. For example, in a project group whose genotype label corresponding to the clustering result is "high-density low-compensation type", the cost sub-items include three categories: compensation fees, relocation fees, and bonuses. The sample size is configured as N=50, and the sample mean of the proportion of compensation fees is calculated. The standard deviation is calculated using the formula. The sample mean of relocation cost as a percentage of total cost was The standard deviation is The average percentage of bonuses is The standard deviation is Analysis of expenditure patterns shows that compensation payments are concentrated in the early stages of the cycle, with a volatility coefficient of [missing value]. The cyclical index is Relocation costs are paid in the medium term, with a volatility coefficient of [missing value]. Bonuses are distributed in a concentrated manner at the end of the period, with a volatility coefficient of [missing value]. The final output of structured cost composition data includes the average proportion, standard deviation, and rhythm parameter of the three cost sub-items. Each parameter significantly improved the matching accuracy of similar items in the knowledge base retrieval test. S2.5: The genotype tags and corresponding cost composition statistical parameters are structured and organized to build a genotype knowledge base, which is stored in a relational database table structure to support rapid retrieval and retrieval when the Wasserstein distance matching and prediction model is activated.

[0013] like Figure 2As shown, step S3 involves inputting the new project initiation data into the gene map feature vector space, calculating its Wasserstein distance to each genotype, matching the optimal reference genotype, and then activating the corresponding Transformer-Time2Seq baseline prediction model. This model is input to a multi-source feature tensor containing project progress indicators, phased funding expenditure sequences, and policy adjustment event codes. Specifically, this includes: S3.1: Based on parameters such as regional development index, housing density gradient, property rights complexity score, resettlement method preference distribution, historical signing rate curve shape and compensation standard fluctuation range provided during the project establishment stage of land acquisition and demolition, generate the initial feature vector of the new project as the input sample in the project gene map feature space. Based on the original parameter data obtained during the project approval stage of land acquisition and demolition, such as regional development index, housing density gradient, property rights complexity score, distribution of resettlement method preferences, historical signing rate curve shape, and compensation standard fluctuation range, a multi-source parameter acquisition interface module (configured with HTTP / JSON transmission protocol and sampling frequency of once a day) is used to achieve synchronous aggregation of different types of survey data. The regional development index parsing algorithm (parameter mapping method: economic level classification table) converts the original index values ​​from administrative statistics to the system's standard dimensions and generates regional development index input units. Furthermore, through the housing density gradient calculation program (input: GIS plot coordinate set and housing unit area statistics), the bilinear spatial interpolation method is used to construct the housing density gradient curve of continuous spatial distribution and obtain the housing density gradient vector set; Furthermore, using a property rights complexity weighted scoring function, the property rights type category coding matrix is ​​weighted and calculated with the number of ownership disputes and the clarity of ownership index. The calculation formula is as follows: in, This is a comprehensive score for the complexity of property rights. For various property rights weighting coefficients, The corresponding complexity score, This represents the total number of property rights types. Through the above calculations, a property rights complexity scoring index vector is generated. Furthermore, based on the statistical procedure for the distribution of resettlement mode preferences, multinomial category coding is used to transform the historical distribution of residents' resettlement options into vector coding, generating a feature vector of resettlement mode preferences; Furthermore, the Dynamic Time Warping (DTW) algorithm (with Euclidean distance as the distance metric) is applied to perform normalization matching on the historical signing rate curve shape data to generate a standardized signing rate shape feature vector. Furthermore, based on the quantitative model of the floating range of compensation standards (input parameters: policy fluctuation rate, market volatility index), a floating coefficient matrix of compensation standards is constructed to characterize the dynamic adjustment space of compensation parameters; By using a feature space concatenation algorithm (method: vector-level concatenation), the regional development index input unit, housing density gradient vector group, property rights complexity scoring index vector, resettlement method preference feature vector, standardized contract signing rate morphology feature vector, and compensation standard floating coefficient matrix are merged in a predefined dimension order to generate the initial feature vector of the new project. This vector serves as the input sample in the project genotype feature space, thus preparing the raw data required for similarity matching with historical genotypes in the main step S3. For example, during the project approval stage of an urban renewal and demolition project in a certain city, the original value of the regional development index was provided by the local statistics bureau, which was 3.45. After standardization based on economic level, the input unit for the regional development index was generated as follows: The building density gradient input is a 100-dimensional density gradient vector generated using bilinear interpolation based on GIS coordinates and building area statistics. The minimum value is... maximum value In the calculation of property rights complexity, there are 5 types of property rights, and the weight coefficients for each type are as follows: , , , , The corresponding complexity score is , , , , Substitute the values ​​into the formula to obtain the comprehensive score. The resettlement method preference vector is encoded as [0.55, 0.35, 0.10], representing the proportional distribution of choices for monetary compensation, in-kind resettlement, and other methods. The historical contract signing rate morphological vector, after DTW standardization, is a 60-dimensional vector, with a fluctuation range of... to The compensation standard fluctuation coefficient matrix is ​​based on the policy fluctuation rate. Market volatility index The generated matrix, with a dimension of 3×3, is used to characterize the interval changes under different compensation types. All the above feature units are concatenated in sequence to generate a new initial feature vector with a comprehensive dimension of 239, which serves as the input sample for the subsequent S3.2 normalization processing, effectively improving the accuracy and robustness of genotype matching; S3.2: Perform normalization processing on the initial feature vector to eliminate the dimensional differences between features of different dimensions, obtain a standardized feature vector, and provide a unified numerical basis for subsequent similarity matching with historical genotypes; S3.3: Based on the constructed gene feature matrix, the Wasserstein distance metric algorithm is used to calculate the distribution difference between the standardized feature vector of the new project and the centroid of each historical genotype, and a set of matching similarity scores is obtained to reflect the structural similarity between the new project and the genotypes of various typical projects. Based on the constructed gene feature matrix, the Wasserstein distance metric algorithm (input parameters: standardized feature vector of the new project, set of feature vectors of historical genotype centroids) is used to realize the quantitative representation of the differences between the feature distribution of the new project and each historical genotype. Furthermore, by using a probability distribution mapping method (parameters: number of feature dimensions, number of samples), the standardized feature vectors of the new project are converted into the form of an empirical cumulative distribution function, thus establishing a basis for comparison in the same distribution space for subsequent distance calculations; Furthermore, a one-dimensional distribution decomposition technique (parameters: the range of values ​​and distribution density of each feature dimension) is adopted to decompose the Wasserstein distance in the multi-dimensional feature space into a weighted sum of distances in each dimension, generating a multi-dimensional distance vector, which is used to analyze the contribution of a single feature to the overall distance. Furthermore, the one-dimensional Wasserstein distance is calculated using the following formula. : in, The cumulative distribution function of the new project characteristics. Let be the cumulative distribution function of the historical genotype centroid characteristics. It is a distribution variable; Furthermore, based on the aforementioned single-dimensional distance vector, a weighted aggregation algorithm (weighting based on feature importance evaluation index) is used to calculate the overall multidimensional Wasserstein distance, forming a set of distance scalars that reflect the degree of similarity between the new project and each historical genotype structure; Furthermore, through a standardization process (parameters: maximum distance, minimum distance), the overall distance scalar set is transformed into a matching similarity score set, so that the scores are distributed in the [0,1] interval, which facilitates subsequent optimal genotype matching and ranking. By using the Wasserstein distance metric algorithm and weighted aggregation processing, the standardized feature vectors from the previous step are transformed into scoring data on the structural similarity between the new project and each historical genotype, thus achieving the precise screening effect of the optimal reference genotype for the new project. For example, in a land acquisition and demolition project in a city center, the standardized feature vector of the new project has a length of 6, corresponding to a regional development index of 0.82, a housing density gradient of 0.65, a property rights complexity score of 0.74, a resettlement method preference distribution of 0.58, a historical signing rate pattern of 0.63, and a compensation standard fluctuation range of 0.49. The empirical cumulative distribution function is fitted with a histogram distribution of 200 samples, and the Wasserstein distances for each dimension are calculated to be 0.05, 0.08, 0.06, 0.07, 0.09, and 0.04, respectively. The feature importance weight vector is set to [0.2, 0.15, 0.25, 0.1, 0.2, 0.1], and the weighted aggregation yields the overall distance. The result was 0.064. Further standardization of the score set resulted in a similarity score of 0.936 for this distance. In the genotype knowledge base, this project achieved the highest similarity score of 0.936 with the "high-density, low-compensation" genotype, successfully matching the optimal reference genotype and providing accurate input for subsequent prediction model activation. S3.4: Based on the matching similarity score, select the historical genotype with the highest score as the optimal reference genotype, and extract the corresponding Transformer-Time2Seq benchmark prediction model and its initial parameter configuration from the genotype knowledge base as the basic model for predicting the trend of new project fund usage. S3.5: Bind the input interface of the benchmark prediction model with multi-source feature tensors such as the project progress indicators, phased fund expenditure sequence and policy adjustment event codes of the current stage of the new project to form a complete model input channel to support subsequent dynamic prediction and context-aware adjustment; Based on the optimal reference genotype selected in S3.4 and its corresponding Transformer-Time2Seq benchmark prediction model, the multi-source feature data of the current stage of the new project is used as the input binding object. The feature mapping interface binding method (parameters: model input layer identifier mapping table, feature tensor dimension rules) is adopted to achieve accurate mapping between the project progress indicator tensor and the prediction model input layer, ensuring that the time series dimension and feature dimension are consistent within the model. Furthermore, by constructing a serialized data channel (parameters: definition of the phased expenditure sequence format, timestamp index rules), the step size of the expenditure sequence in the time dimension is aligned, and a phased expenditure sequence tensor in the same batch as the progress indicators is obtained. Furthermore, an event coding fusion algorithm (parameters: policy adjustment event coding table, trigger time window Δt) is adopted to achieve alignment and fusion of the policy adjustment event coding vector with the above progress indicators and expenditure sequences on the time axis, and to generate an extended feature tensor containing event semantics; Furthermore, a feature normalization algorithm (parameters: historical estimates of mean μ and standard deviation σ) is applied to perform normalization transformations on the schedule index tensor, expenditure sequence tensor, and event encoding vector, respectively. The calculation formulas are as follows: in, These are the original eigenvalues. This represents the mean of the corresponding feature in the historical samples. The standard deviation of the corresponding feature; Furthermore, a tensor concatenation operation is used (parameter: concatenation dimension is set to feature dimension) to concatenate the normalized progress indicators, phased fund expenditures and policy event codes in time series structure to generate a multi-source feature fusion tensor. By using a multi-source feature binding process, the results of the previous step are transformed into a fusion feature tensor that is fully mapped to the input layer of the Transformer-Time2Seq prediction model, thereby achieving a unified input data structure that supports dynamic prediction and context-aware adjustment. For example, in the second phase of a land acquisition and demolition project, the progress indicator tensor includes a signing rate of 0.65%, an assessment completion rate of 0.78%, and a demolition completion rate of 0.32%. The phased expenditure sequence is [3.2 million, 4.5 million, 5.1 million] yuan, and the policy adjustment event encoding vector is [1, 0, 0], indicating that the first policy adjustment has occurred. A feature mapping interface binding method is used to bind the signing rate, assessment completion rate, and demolition completion rate to the model input layers p1, p2, and p3, respectively. The expenditure sequence is aligned to the batches t1, t2, and t3 of the progress indicators through timestamp indexes. The policy event encoding vector is fused with the progress and expenditure data of the corresponding batches within a time window of Δt = 7 days. In the normalization process, assuming the mean signing rate μ = 0.5 and σ = 0.1, the normalized result is... =1.5; the expenditure amount was normalized using the historical average μ=4 million and σ=500,000, and the normalized value of the first period's 3.2 million was 1.5. =-1.6. After normalization, the three types of features are concatenated according to the feature dimensions to form a fusion tensor input model with shape [3,8], which realizes the function of predicting capital trends in this stage and subsequent context-aware adjustment.

[0014] like Figure 3 As shown, step S4 involves determining the deviation between the predicted fund usage trend output by the benchmark prediction model and the actual expenditure curve. When the deviation exceeds a dynamic threshold, a context-aware gating mechanism is triggered. This mechanism adjusts the attention weight distribution of each cost item in the Transformer decoder to generate elasticity coefficient correction parameters adapted to the characteristics of the current project. Specifically, this includes: S4.1: Calculate the periodic deviation between the predicted value of fund usage trend output by the benchmark prediction model and the actual fund expenditure curve of the project. Perform dynamic threshold modeling on the deviation sequence based on the sliding window mechanism to obtain the deviation judgment signal used to trigger the context-aware gating mechanism. For the time series samples of the predicted value of fund use trend output by the benchmark prediction model and the actual fund expenditure curve of the project, the period deviation calculation method (parameters: period length Δt, sampling interval δt) is used to extract the period-by-period difference sequence of the two curves at the same time anchor point. Furthermore, by using the mean square deviation calculation method (parameters: sliding window length w, window step size s), the stability of the periodic deviation sequence within each sliding window interval is evaluated, and the window mean square deviation value sequence is obtained, where the mean square deviation formula is: in, This represents the deviation value for a single period within the window. This is the mean of all deviation values ​​within the window. This represents the number of sampling points within the window. Furthermore, a dynamic threshold modeling algorithm is employed (parameter: historical deviation sequence statistical characteristics). , Sensitivity factor This enables real-time threshold calculation based on historical statistical features and generates a threshold function. Serialized representation; Furthermore, a threshold comparison operation is performed (parameter: the current window mean square deviation value sequence). Dynamic threshold function This allows for the determination of deviation exceeding the limit at each time point and the generation of a deviation state vector, where each element takes a value of 0 or 1, representing the state of not exceeding the threshold and the state of exceeding the threshold, respectively. By using the edge detection algorithm of the deviation state vector (parameter: minimum duration period m of state change), the result of the previous step is transformed into a deviation judgment signal, thereby realizing the generation of the trigger condition for the context-aware gating mechanism. For example, in the monitoring of funding trends for a certain land acquisition and demolition project, a period length Δt of 7 days and a sampling interval δt of 1 day were selected. The deviation between the predicted curve output by the benchmark prediction model and the actual expenditure curve was calculated periodically to obtain a deviation sequence of length 90. With a sliding window length w = 14 days and a window step size s = 1 day, the mean squared deviation formula was applied for calculation, where n = 14. The maximum value of the mean squared deviation in the sample sequence reached 1200, significantly exceeding the historical average. =300 and standard deviation The dynamic threshold T(t) composed of 200 is 300 + 2·200 = 700. Based on the threshold comparison result, a deviation state vector of length 77 is generated, where three consecutive periods have a value of 1. This value is identified as a valid trigger segment by the edge detection algorithm, outputting a deviation judgment signal. This signal is then used by the context-aware gating mechanism to load external driving factors for attention weight adjustment. In this implementation, the correct triggering of the deviation judgment signal significantly improves the model's adaptability and prediction accuracy during periods of capital fluctuation. S4.2: Based on the result that the deviation judgment signal exceeds the preset dynamic threshold, generate the context-aware gating mechanism activation command, and load the feature vector of external driving factors related to the current project stage, including policy adjustment event codes, volatility of phased capital expenditure sequence and contract progress deviation indicators, as the input basis for the gating mechanism. Based on the result that the deviation judgment signal output by S4.1 exceeds the preset dynamic threshold, the event-driven signal parsing method (parameters: deviation judgment signal bit width 16 bits, threshold comparator resolution 0.01) is used to generate the activation command of the context-aware gating mechanism. Furthermore, by using an external driving factor loading algorithm (parameters: data channel bandwidth 10Mbps, buffer length 256 records), the parsing and input of policy adjustment event codes related to the current project stage are realized, and the event code vector matrix is ​​obtained, ensuring the sensitivity of the gating mechanism to policy changes when the strategy is adjusted; Furthermore, using the time-series volatility calculation method (with a window length set to 7 days, the volatility measurement formula is as follows): in This represents the daily expenditure amount. The sample mean. (For the number of sample days), the volatility feature vector of the phased capital expenditure sequence is extracted, and a normalized volatility index is generated to ensure the comparability of features under different time windows; Furthermore, a method for calculating the deviation from the contract signing schedule is adopted (parameters: deviation tolerance ratio 0.05, deviation calculation formula). in Based on the current signing rate, (Referring to the historical signing rate of the reference genotype), the signing progress deviation index is calculated, and the signing progress feature vector is input into the gating mechanism; By using the aforementioned external driving factor loading algorithm and feature calculation method, the deviation judgment signal from the previous step is transformed into a structured contextual input vector, enabling the contextual awareness gating mechanism to comprehensively perceive the project policy background, fund expenditure fluctuations, and contract execution deviations before parameter correction. For example, during the implementation of a land acquisition and demolition project, when the mean of the predicted deviation sequence of fund usage calculated by the sliding window is 0.15 and exceeds the set dynamic threshold of 0.1, an activation signal is output by the deviation comparator. The external driving factor loading module parses two policy adjustment event codes from the policy database. The coding format is 8-bit binary, representing the increase in compensation standards and the delay in the construction of resettlement housing, respectively. The fund expenditure volatility calculation module collects the daily expenditure sequence within a 7-day window. The volatility index value is calculated using the formula for 10,000 yuan. 10,000 yuan; the contract signing progress analysis module records the current contract signing rate. Historical reference genotype concurrent contract signing rate The calculated deviation is = This indicates that the signing progress is behind schedule. The above three types of feature vectors, after being uniformly normalized, form the context input vector. The system implements an entry control mechanism to adaptively adjust and trigger adjustments based on current funding forecast deviations. In subsequent S4.3, the system dynamically allocates attention weights for each cost sub-item in a targeted manner. S4.3: Utilize the attention weight adjustment module in the gating mechanism to dynamically adjust the attention weight distribution of each cost sub-item (such as compensation fee, relocation fee, and bonus) in the Transformer decoder based on the feature vector of the external driving factor, and generate an attention weight matrix containing the elasticity coefficient correction factor to reflect the sensitivity of the current project characteristics to the growth elasticity of various expenditure items. Based on the input of the feature vector of external driving factors, a feature channel including policy adjustment event coding, volatility of phased fund expenditure sequence and deviation of signing progress indicators is set as the weight adjustment trigger signal source of the gating mechanism; A multi-head attention weight adjustment algorithm (parameters: number of attention heads h=8, weight initialization method is mean normalization) is adopted to perform initial mapping adjustment on the weight distribution of each cost item in the Transformer decoder, forming a dynamic weight distribution base table; Furthermore, by using the feature sensitivity analysis method (parameter: sensitivity calculation period T = 5 prediction period windows), the influence intensity between external driving factors and each cost sub-item is quantitatively evaluated, and a sensitivity vector is obtained for the re-correction of weight distribution; Furthermore, based on the normalized weighted update formula: in, For the updated attention weight components, These are the original attention weight components. The sensitivity value is used to redistribute the weights of each attention weight component according to its influence intensity while keeping the sum of the values ​​equal to 1. By using the above formula, the original weight vector is weighted by sensitivity to obtain the elasticity coefficient correction weight matrix, so as to realize the adaptive adjustment of the growth elasticity of different cost sub-items in the prediction of capital trends. For example, for a land acquisition and demolition project where the signing rate is significantly lower than the predicted value, the policy adjustment event code value in the external driving factor feature vector is set to 2 (representing a temporary reduction in subsidy policy), the volatility of the phased capital expenditure sequence is 0.35, and the deviation index of the signing progress is 0.15. The multi-head attention weight adjustment algorithm initializes the original weights of each cost sub-item (compensation fee, relocation fee, and bonus) to 0.4, 0.35, and 0.25, respectively. The sensitivity vector generated by feature sensitivity analysis is [0.6, 0.2, 0.2]. In the normalized weighted update formula, the compensation fee weight is corrected as follows: The new weight values ​​are calculated, and the other two items are updated in the same way to form a corrected weight matrix [0.63, 0.20, 0.17]. This matrix significantly improves the influence of compensation fee-related input variables on the model in subsequent trend prediction, making the prediction curve closer to the actual expenditure trajectory, and the output elasticity coefficient correction parameter has higher adaptability and accuracy. S4.4: Update the cost growth elasticity coefficient in the baseline prediction model based on the attention weight matrix to generate a set of elasticity coefficient correction parameters adapted to the characteristics of the current project. Each correction parameter corresponds to the growth response sensitivity of a type of cost sub-item to improve the model's adaptability to project heterogeneity. S4.5: Feed the set of elasticity coefficient correction parameters back to the benchmark prediction model, re-execute the fund expenditure trend prediction calculation, and generate the corrected fund usage trend curve to form a closed-loop feedback mechanism and achieve continuous optimization of the model prediction accuracy. For the set of elastic coefficient correction parameters adapted to the characteristics of the current project, a parameter injection method (parameters: cost sub-item identifier, correction coefficient value, injection position index) is used to replace and update the weights of each sequence feature within the benchmark prediction model. Furthermore, by using a cyclic iterative optimization algorithm (parameters: learning rate set to 0.001, iteration steps to 50), the correction coefficients are gradually converged within the model time step range, and the updated multi-source feature weight tensor is obtained. Furthermore, a sliding window recalculation method (with the window width set to 5 prediction periods) is adopted to realize the batch re-evaluation of the capital expenditure trend sequence and generate a corrected prediction vector. Furthermore, by using a trend smoothing filter algorithm (parameter: smoothing factor set to 0.15), the high-frequency fluctuation component in the corrected prediction curve is suppressed, and the smoothed capital usage trend curve data is obtained. By using a closed-loop error feedback mechanism (parameter: error threshold set to 0.03), the latest prediction curve is compared with the actual expenditure curve. If the error is lower than the set threshold, the current combination of correction coefficients is maintained; otherwise, the elasticity coefficient correction process is re-triggered to achieve continuous optimization of the model prediction accuracy. For example, in a land acquisition and demolition project, the input set of elasticity coefficient correction parameters includes three types of cost sub-items: a correction coefficient of 0.92 for compensation fees, a correction coefficient of 1.05 for relocation fees, and a correction coefficient of 0.88 for bonuses. Using a parameter injection method, the compensation fee coefficient is injected into the relevant column of the weight matrix in the second layer of the model decoder, the relocation fee coefficient into the third layer, and the bonus coefficient into the first layer. A cyclic iterative optimization algorithm is used to perform 50 steps of parameter fine-tuning, with a learning rate set to 0.001, so that the predicted weights corresponding to each cost sub-item stabilize within the new coefficient value range during time series modeling. A sliding window recalculation method is used to batch re-predict the fund expenditure sequence for the next 5 months, resulting in the corrected predicted sequence {105.6, 108.2, 110.4, 112.1, 113.0} (unit: ten thousand yuan). A smoothing filter algorithm with a smoothing factor of 0.15 is used to process the forecast sequence, reducing high-frequency fluctuations between monthly forecasts, and outputting a smoothed curve {105.8, 108.0, 110.2, 112.0, 113.1}. The corrected curve and actual expenditures are evaluated using the following mean squared error calculation formula: in, For the i-th predicted value, Let i be the actual expenditure value. The sample size is given. The calculated mean squared error is 21,400 yuan, which is lower than the amount range corresponding to the error threshold of 0.03, indicating that the corrected model has significantly improved prediction accuracy and can stably support the generation process of budget adjustment strategies in the long term.

[0015] Step S5: Based on the corrected prediction results, a multi-scenario budget adjustment suggestion package is constructed. A rule-based reasoning engine is applied to semantically match the project stage goal priority tags with the applicable condition tags in the suggestion package, generating a budget adjustment strategy recommendation set with confidence ranking. The rule-based reasoning engine is the core module for intelligent matching of project stage goals and budget adjustment suggestions. It generates a matching score matrix by semantically matching the project stage goal priority tags with the applicable condition tags in the budget adjustment suggestion package. Based on the funding risk response rules and stage goal matching rules, it generates budget adjustment strategy templates for typical scenarios. It dynamically outputs the weight coefficients of the matching dimensions according to project stage characteristics to perform priority weighting calculations on the score matrix. Finally, it performs semantic tagging and confidence ranking on the budget adjustment strategy recommendation set, ultimately achieving intelligent transformation from prediction results to decision support. Specifically, this includes: S5.1: Based on the revised forecast results of fund usage trends, model different budget adjustment scenarios and generate budget adjustment suggestion packages for multiple budget adjustment scenarios, including 'high liquidity demand' and 'low contract conversion period', so as to form a variety of optional budget adjustment strategies to meet the fund management needs under different project stages and changes in the external environment; Based on the revised data set of fund usage trend prediction curves, it is used as the core input for budget adjustment scenario modeling to ensure that the multidimensional parametric features of the prediction results are fully preserved in subsequent scenario generation. A scenario classification algorithm (parameters: based on project progress phase division, external policy event coding, and liquidity demand index) is used to achieve preliminary classification of funding adjustment scenarios. Furthermore, by using a multivariate regression analysis method (parameters: historical budget change records of similar projects, current cost composition weights, and predicted capital volatility), we can quantitatively model the budget change magnitude under each classification scenario and obtain a scenario parameter matrix that includes the budget adjustment magnitude and adjustment direction. Furthermore, a clustering validity test algorithm (parameters: Davies-Bouldin index, Silhouette coefficient) is used to evaluate the quality of the scenario clustering results, and to select scenario categories with high stability and significant discriminative power to generate an optimized set of scenario classification labels. Furthermore, based on the scenario tag set, a rule template generation engine (parameters: various funding risk response rules, project stage target priority matching rules) is used to generate strategy templates for typical budget adjustment scenarios such as "high liquidity needs" and "low contract conversion period", and form a scenario-strategy mapping relationship table. By using scenario template generation and forecast data fusion processing, the forecast curve from the previous step is transformed into a multi-scenario budget adjustment suggestion package, enabling diversified coverage and targeted optimization of budget strategies at different project stages. For example, in a certain land acquisition and demolition project, the revised fund usage trend forecast curve shows that fund expenditures will exhibit cyclical high fluctuations over the next three months, with a liquidity demand index of 0.85 and a contract conversion rate forecast of 0.35. Based on the project progress stage division, the current stage is the mid-term demolition progress, and the policy event code indicates an expected increase in compensation standards in the near future. Using a scenario classification algorithm, the project is categorized as a high liquidity demand type, and multiple regression analysis calculates a positive adjustment of 0.15 billion yuan in budget demand, with the adjustment direction leaning towards upfront payments. The formula is: in, Let i be the predicted funding requirement for period i. For the target budget value, The forecast period is indicated by a Silhouette coefficient of 0.78, suggesting high stability in scenario classification. The rule template generation engine automatically generates budget strategies for the "high liquidity demand" scenario, including measures such as releasing some compensation funds in advance and increasing emergency payment reserves. These strategies are then mapped to scenario labels to form a final budget recommendation package. Applying this strategy package results in a significantly reduced risk of funding shortages and a significantly improved flexibility in fund management. S5.2: Semantically encode the target priority tags of the current stage of the project, and use natural language processing technology to transform them into structured semantic vector representations so as to semantically match them with the applicable condition tags in the budget adjustment suggestion package later. S5.3: Based on the set of applicable condition labels in the budget adjustment suggestion package, a semantic similarity calculation algorithm (such as cosine similarity or BERT semantic matching model) is used to compare the semantic vector of the project stage target priority with the semantic vector of each applicable condition label to identify the most matching budget adjustment scenario. S5.4: Based on the semantic matching results, construct a matching score matrix for budget adjustment strategies, where each row corresponds to a budget adjustment suggestion, each column corresponds to a matching dimension, and the score reflects the degree of fit between the suggestion and the current project stage goals; Based on the semantic matching results of the project phase target priority tags and the applicable condition tags of the budget adjustment suggestion package, a scoring matrix construction algorithm (parameter settings: matrix dimension is m×n, where m is the number of budget adjustment suggestions and n is the number of matching dimensions) is adopted to realize the structured matching results into a two-dimensional scoring matrix data structure. Furthermore, by using a matching metric model (with a cosine similarity threshold set at 0.85 and semantic weight coefficients ranging from 0.1 to 0.9), the scoring values ​​of each budget adjustment suggestion in each matching dimension are quantified, and a preliminary scoring matrix is ​​obtained. Furthermore, a normalization algorithm (method: Min-Max Scaling, range [0,1]) is adopted to normalize the values ​​of each column of the initial scoring matrix and generate a normalized scoring matrix to eliminate the influence of the dimensions of scores of different matching dimensions. Furthermore, a weighting algorithm (method: dimensional importance weighting, with weight vectors output by the rule-based inference engine based on project stage characteristics) is used to perform dimensional priority weighting operations on the normalized scoring matrix, generating a weighted scoring matrix. The weighting formula is: in, For the weighted scoring matrix, the first... line, number The column's rating value, This represents the value at the corresponding position in the normalized scoring matrix. For the first Weight coefficients for each matching dimension; Through matrix construction and weighted operation processing, the semantic matching results of the previous step are transformed into a quantifiable and comparable matching score matrix, so as to realize the structured expression of the fit between each budget adjustment suggestion and the project stage goal; For example, in the scenario of fund management for a land acquisition and demolition project, the budget adjustment suggestion package contains 5 suggestions, each scored on 3 matching dimensions (fund liquidity guarantee, contract signing progress matching degree, and policy adaptability). The preliminary scoring matrix obtained in the semantic matching stage is normalized using the Min-Max normalization method, mapping each column score to the [0,1] interval to obtain a normalized matrix. The rule reasoning engine analyzes the target priority in the analysis stage and outputs a weight vector. In the weighted calculation, the above formula is applied to calculate the weighted value of each element. For example, the first recommendation, in terms of the weighted value for liquidity protection, is divided into... After weighting, a complete weighted scoring matrix is ​​obtained. The results show that this matrix significantly improves the comparability of budget adjustment recommendations across different dimensions, supports subsequent confidence level calculations and rankings, and provides decision-makers with a reference for the optimal adjustment strategy for the current stage. S5.5: Perform weighted aggregation calculation on the matching score matrix to generate a comprehensive confidence score for each budget adjustment strategy, and sort the budget adjustment strategy recommendation set from high to low according to the score to support decision-makers in quickly identifying the optimal adjustment plan; Based on the constructed budget adjustment strategy matching score matrix, a weighted aggregation calculation method is used (parameter: matching dimension weight coefficient vector). (Determined by the statistical results of the importance of historical decisions), achieving a comprehensive fusion of multi-dimensional matching scores; Furthermore, by using a normalization method (parameter: minimum-maximum normalization interval set to [0,1]), the scores of different matching dimensions are converted within a unified numerical range, and a normalized score matrix is ​​obtained, providing standardized input data for weighted aggregation; Furthermore, a vectorized dot product operation method is adopted (parameter: weight coefficient vector). With each row vector of the normalized score matrix This implements the calculation of a single-strategy comprehensive score and generates a comprehensive confidence score for each budget adjustment strategy, where the dot product formula is defined as: in For comprehensive scoring, Let i be the weight coefficient of the i-th dimension. The normalized score value for the i-th dimension; Furthermore, using a stable sorting algorithm (parameter: sorting key is the comprehensive confidence score, stability is used to maintain the original order of the same scoring strategies), the budget adjustment strategy recommendation set is sorted in descending order of comprehensive confidence score, and the sorted strategy sequence is obtained; By using weighted aggregation calculation and stable sorting processing, the matching degree score matrix result of the previous step is transformed into comprehensive confidence score data, thereby achieving the priority intelligent sorting technology effect of the budget adjustment strategy recommendation set. For example, in a large-scale real estate expropriation and demolition project, the matching score matrix includes five budget adjustment strategies. Each strategy has scores of [0.82, 0.75, 0.90], [0.65, 0.88, 0.77], [0.93, 0.82, 0.85], [0.78, 0.80, 0.80], and [0.70, 0.68, 0.72] respectively across the three matching dimensions: stage goal alignment, liquidity satisfaction, and policy adaptability. The weight coefficients for each dimension are determined by the expert group based on historical project statistics and are set to [0.4, 0.35, 0.25]. After min-max normalization, the scores are normalized to [0,1] within the original range. Using vectorized dot product operations, the comprehensive score for the first strategy is calculated as follows: The initial score was 0.8175. The comprehensive scores for all strategies were calculated sequentially, yielding [0.8175, 0.76855, 0.86975, 0.792, 0.701]. After processing with a stable ranking algorithm, the strategy with the highest comprehensive score (the 3rd strategy) was ranked first in the recommendation set. Its flexibility in fund utilization and policy adaptation were significantly better than other strategies, resulting in the strategy sequence [3,1,4,2,5]. This effectively supports decision-makers in quickly selecting the optimal solution.

[0016] Step S6: The genotype matching record, model parameter update log, and budget adjustment strategy recommendation set are written into the distributed ledger via a blockchain smart contract, generating an immutable evidence record containing timestamps and operation traceability information. Specifically, this includes: S6.1: Perform serialization encoding on the genotype matching records, and encapsulate the genotype tags, Wasserstein distance matching results and matching timestamps into a structured byte stream based on the Protocol Buffers data structure to form persistent matching log entries; S6.2: Perform hash digest generation on the model parameter update log. Use the SHA-256 algorithm to calculate the digest of the attention weight distribution adjustment record in the Transformer-Time2Seq model to obtain the parameter update fingerprint for integrity verification. S6.3: Perform semantic labeling on the budget adjustment strategy recommendation set. Based on the confidence ranking results output by the rule reasoning engine, add a stage target priority label and an applicable condition label to each budget adjustment suggestion to generate a structured recommendation strategy set. S6.4: Input the genotype matching log entries, model parameter update fingerprints, and structured budget recommendation strategy set into the blockchain smart contract, and execute the transaction proposal construction and endorsement process based on the chaincode interface of the Hyperledger Fabric platform to generate a block transaction data package to be submitted; The genotype matching log entries processed by serialization and encoding, the model parameter update fingerprint generated by SHA-256 hash digest, and the structured budget recommendation strategy set with stage target priority and applicable condition labels are loaded as input objects into the calling interface of the blockchain smart contract, and bound to the input channel constructed by the transaction proposal to realize the mapping relationship between data and contract state variables. The chaincode interface calling method (parameters: Hyperledger Fabric platform version number, chaincode function name, and call parameter set) is used to locate and encapsulate each input object in the chaincode function, forming the payload structure required for the transaction proposal. Furthermore, by using a transaction payload signature algorithm (parameters: ECDSA elliptic curve type, private key length, digest algorithm type), the caller digitally signs the transaction proposal payload, generating a signature packet that can be verified by endorsing nodes, ensuring the verifiability and non-repudiation of the transaction proposal source; Furthermore, through the transaction endorsement strategy execution module (parameters: endorsement node list, strategy type, node weight factor), chaincode functions are called in parallel among multiple endorsement nodes to simulate transaction execution, and the execution results and signatures are returned to form an endorsement response set; Furthermore, a transaction data packet synthesis algorithm (parameters: payload structure, signature packet, endorsement response set) is adopted to achieve structured merging of transaction proposal data, digital signatures and endorsement responses, generating a block transaction data packet containing transaction metadata, payload data, signature list and endorsement list; By combining transaction data packets with chaincode interface binding, the input data from the previous step is transformed into block transaction data that can be submitted to the blockchain network, achieving the expected technical effects of pre-transaction integrity verification and traceability assurance. For example, the following data is input into the chaincode interface: a genotype matching log entry (256 bytes, encoded using Protocol Buffers), a model parameter update fingerprint (64 bytes, calculated using SHA-256), and a budget adjustment strategy set (containing 10 suggestions, with a total semantic label length of 512 bytes) for a land acquisition and demolition project. The chaincode function is named "commitProjectRecords," and the platform version is set to 2.2. The transaction payload signature algorithm is called, using the secp256r1 curve, a 256-bit private key, and the SHA-256 digest algorithm, generating a 72-byte signature packet. The endorsement strategy specifies three nodes, with the strategy type being "AND" mode and each node having a weight of 1. The endorsement response set contains the execution result hash and signature of each node. After the transaction data packet is synthesized, a data structure of approximately 1100 bytes in total length is generated, containing transaction metadata (project ID, timestamp), payload data (log entry, fingerprint, strategy set), a signature list, and an endorsement list. In the subsequent PBFT consensus phase, the transaction was successfully written into the distributed ledger through multi-node collaborative verification, ensuring that the immutability and traceability of the data throughout the entire process were significantly improved. S6.5: Perform consensus verification and distributed ledger writing operations on the block transaction data packet. After completing multi-node collaborative verification based on the PBFT consensus mechanism, write the transaction data into the distributed ledger and generate an immutable evidence record containing timestamps, operation traceability identifiers and block height.

[0017] The present invention also provides a digital management and control system for the entire process of land acquisition and demolition based on multi-source data fusion, which adopts the above-mentioned digital management and control method for the entire process of land acquisition and demolition based on multi-source data fusion for digital management and control of land acquisition and demolition.

[0018] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0019] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A digital management and control method for the entire process of land acquisition and demolition based on multi-source data fusion, characterized in that, Includes the following steps: S1: Based on the regional development index, housing density gradient, property rights complexity score, resettlement method preference distribution, historical signing rate curve shape, and compensation standard fluctuation range parameters of the expropriation and demolition projects, construct the project gene map feature vector space, and map historical project data to this space to form a gene feature matrix. S2: Execute the DBSCAN clustering algorithm on the gene feature matrix to identify typical project genotypes with similar fund expenditure rhythms and generate a genotype knowledge base; S3: Input the new project establishment data into the gene map feature vector space, calculate its Wasserstein distance with each genotype, and activate the corresponding Transformer-Time2Seq benchmark prediction model after matching the optimal reference genotype. S4: Based on the deviation between the predicted value of fund usage trend output by the benchmark prediction model and the actual expenditure curve, when the deviation exceeds the dynamic threshold, the context-aware gating mechanism is triggered, and the elasticity coefficient correction parameter is generated by adjusting the attention weight distribution of each cost item in the Transformer decoder. S5: Construct a multi-scenario budget adjustment suggestion package based on the corrected prediction results, and use a rule reasoning engine to semantically match the project stage goal priority labels with the applicable condition labels in the suggestion package to generate a budget adjustment strategy recommendation set; S6: The genotype matching record, model parameter update log, and the budget adjustment strategy recommendation set are written into the distributed ledger through a blockchain smart contract to generate an immutable evidence record.

2. The method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion as described in claim 1, characterized in that, Step S1 specifically includes: The regional development index of the expropriation and demolition project is standardized to obtain the standardized value of the regional development index. Based on the building density gradient data, a spatial interpolation algorithm is executed to generate a building density map under continuous spatial distribution and output the building density gradient feature vector. Based on the property rights complexity scoring rules, the property rights types involved in the project, the number of ownership disputes, and the clarity of property rights ownership are weighted and calculated to generate a property rights complexity scoring index. Using a statistical model of resettlement method preference distribution, the proportion of resettlement methods chosen by residents in historical projects is classified and coded, and the feature vector of resettlement method preference distribution is extracted. Based on historical signing rate curve shape data, the curve is normalized and standardized historical signing rate shape features are extracted. Based on the parameters of the compensation standard floating range, and combined with policy documents and market fluctuation data, a compensation standard floating coefficient matrix is ​​constructed. The standardized value of the regional development index, the housing density gradient feature vector, the property rights complexity scoring index, the resettlement method preference distribution feature vector, the historical signing rate morphological characteristics, and the compensation standard floating coefficient matrix are concatenated to construct the project gene map feature vector space. Data samples from historical land acquisition and demolition projects are mapped to the project's gene map feature vector space to form a gene feature matrix.

3. The method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion as described in claim 2, characterized in that, Step S1 further includes filling missing values ​​with the mean, screening outliers by three times the standard deviation, Z-score standardization and Min-Max normalization for the regional development index, and using Spearman correlation coefficient to evaluate the correlation score with the funding pattern, selecting a correlation score index with a correlation threshold greater than 0.

5.

4. The method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion as described in claim 1, characterized in that, Step S2 specifically includes: A gene feature matrix is ​​constructed based on historical project data in the project gene map feature vector space. The DBSCAN clustering algorithm is applied to the gene feature matrix. The neighborhood radius and minimum number of samples are set. The similarity between feature vectors is calculated based on the Euclidean distance metric. Density-connected data clusters are identified, and historical project groups with similar fund expenditure rhythms are divided. For each clustering result, feature summarization and label naming are performed. Based on the mean feature vector of the cluster center, the core feature combination of the items in each clustering result is extracted to generate the corresponding genotype label. For each genotype tag corresponding to a project group, calculate the statistical parameters of its cost composition pattern; The genotype tags and corresponding cost composition statistical parameters are structured and organized to construct a genotype knowledge base, which is then stored using a relational database table structure.

5. The method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion as described in claim 1, characterized in that, Step S3 specifically includes: Based on the regional development index, housing density gradient, property rights complexity score, resettlement method preference distribution, historical signing rate curve shape, and compensation standard fluctuation range parameters provided during the project approval stage, an initial feature vector for the new project is generated. The initial feature vector is normalized to obtain a standardized feature vector; Based on the constructed gene feature matrix, the Wasserstein distance metric algorithm is used to calculate the distribution difference between the standardized feature vector and the centroid of each historical genotype to obtain the matching similarity score. Based on the matching similarity score, the historical genotype with the highest score is selected as the optimal reference genotype, and the corresponding Transformer-Time2Seq benchmark prediction model and its initial parameter configuration are extracted from the genotype knowledge base. The input interface of the benchmark prediction model is bound to the multi-source feature tensor of the current stage of the new project to form a complete model input channel.

6. The method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion as described in claim 5, characterized in that, The Wasserstein distance metric algorithm is used to measure the similarity of heterogeneous multidimensional feature distributions. The single-dimensional distance is scored by integral of the empirical distribution function, while the multidimensional distance is weighted by feature importance and aggregated to transform the standardized feature vector into scoring data on the similarity between the new project and each historical genotype structure.

7. The method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion as described in claim 1, characterized in that, Step S4 specifically includes: The deviation between the predicted value of fund usage trend output by the benchmark prediction model and the actual fund expenditure curve of the project is calculated on a periodic basis. Dynamic threshold modeling is performed on the deviation sequence based on the sliding window mechanism to obtain the deviation judgment signal. Based on the result that the deviation judgment signal exceeds the preset dynamic threshold, a context-aware gating mechanism activation command is generated, and the feature vector of external driving factors related to the current project stage is loaded. By utilizing the attention weight adjustment module in the gating mechanism, the attention weight distribution of each cost item in the Transformer decoder is dynamically adjusted based on the feature vector of the external driving factor, generating an attention weight matrix that includes an elasticity coefficient correction factor. Based on the attention weight matrix, the cost growth elasticity coefficient in the benchmark prediction model is updated to generate a set of elasticity coefficient correction parameters adapted to the characteristics of the current project. The set of elasticity coefficient correction parameters is fed back to the benchmark prediction model, and the fund expenditure trend prediction calculation is re-executed to generate the corrected fund usage trend curve.

8. The method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion as described in claim 7, characterized in that, The feature vector of the external driving factors includes policy adjustment event codes, volatility of phased fund expenditure sequences, and deviation indicators of contract signing progress.

9. The method for digital management and control of the entire process of land acquisition and demolition based on multi-source data fusion as described in claim 1, characterized in that, Step S5 specifically includes: Based on the revised forecast results of fund usage trends, models are built for different budget adjustment scenarios, and budget adjustment suggestion packages for multiple budget adjustment scenarios are generated; Semantically encode the target priority labels of the current stage of the project, and use natural language processing technology to transform them into structured semantic vector representations; Based on the set of applicable condition tags in the budget adjustment suggestion package, a semantic similarity calculation algorithm is used to compare the semantic vector of the project stage target priority with the semantic vector of each applicable condition tag to identify the most matching budget adjustment scenario. Based on the semantic matching results, construct a matching score matrix for the budget adjustment strategy; The matching score matrix is ​​weighted and aggregated to generate a comprehensive confidence score for each budget adjustment strategy. The budget adjustment strategy recommendation set is then sorted from high to low based on the scores to identify the optimal adjustment scheme.

10. A digital management and control system for the entire process of land acquisition and demolition based on multi-source data fusion, characterized in that: The digital management and control method for the entire process of land acquisition and demolition based on multi-source data fusion, as described in any one of claims 1-9, is used for the digital management and control of the entire land acquisition and demolition process.

Images