CN-122022277-A - Convergence and migration full-flow digital management and control method and system based on multi-source data fusion

CN122022277ACN 122022277 ACN122022277 ACN 122022277ACN-122022277-A

Abstract

The invention provides a collection and migration full-flow digital management and control method and system based on multi-source data fusion, comprising the steps of constructing project genetic map feature space and generating a typical project genotype knowledge base through DBSCAN clustering; the method comprises the steps of matching a new project with an optimal genotype through Wasserstein distance, activating a transducer-Time 2Seq model to predict fund use trend, triggering a context-aware gating mechanism to dynamically adjust the attention weight and the elastic coefficient of a cost sub-project when the deviation between prediction and actual expenditure exceeds a threshold value, generating a multi-scenario budget adjustment suggestion packet based on a correction result, outputting a confidence ordering strategy through semantic matching, and storing all key data through a blockchain intelligent contract to ensure non-falsification and traceability.

Inventors

CHEN HANG
WU XIAOJIAN
HU HESONG
SHEN HAO
LIN TAO
LI XIN
LIU YADONG

Assignees

广州市建筑科学研究院集团有限公司
广州市市政集团有限公司房地产置业分公司
广州建设工程质量安全检测中心有限公司

Dates

Publication Date: 20260512
Application Date: 20260109

Claims (10)

1. The collection and migration full-flow digital management and control method based on multi-source data fusion is characterized by comprising the following steps of: S1, constructing a project genetic map feature vector space based on regional development indexes, house density gradients, property complexity scores, arrangement mode preference distribution, historical subscription rate curve forms and compensation standard floating interval parameters of collection and migration projects, and mapping historical project data to the space to form a genetic feature matrix; s2, executing a DBSCAN clustering algorithm on the gene feature matrix, identifying typical project genotypes with similar capital expenditure rhythms, and generating a genotype knowledge base; s3, inputting new project item data into the genetic map feature vector space, calculating the Wasserstein distance between the new project item data and each genotype, and activating a corresponding transform-Time 2Seq reference prediction model after matching with an optimal reference genotype; S4, based on the deviation amount of the fund usage trend predicted value and the actual expenditure curve output by the reference prediction model, triggering a situation awareness gating mechanism when the deviation exceeds a dynamic threshold value, and generating an elasticity coefficient correction parameter by adjusting the attention weight distribution of each cost sub-item in a transducer decoder; s5, constructing a multi-scenario budget adjustment suggestion packet according to the corrected prediction result, and performing semantic matching on the project stage target priority label and the applicable condition label in the suggestion packet by using a rule reasoning engine to generate a budget adjustment strategy recommendation set; and S6, writing the genotype matching record, the model parameter update log and the budget adjustment strategy recommendation set into a distributed account book through a blockchain intelligent contract to generate a tamper-proof record.
2. The method for controlling the collection and migration whole-flow digitization based on multi-source data fusion according to claim 1, wherein the step S1 specifically comprises: Carrying out standardization treatment on the regional development index of the collection and removal project to obtain a regional development index standardization value; Based on house density gradient data, executing a spatial interpolation algorithm, generating a house density map under continuous spatial distribution, and outputting a house density gradient feature vector; Weighting calculation is carried out on the property types, the amounts of rights and interests disputes and the definition of property attributions related to the items according to the property complexity scoring rule, and a property complexity scoring index is generated; classifying and coding the proportion of the arrangement modes selected by residents in the historical items by using an arrangement mode preference distribution statistical model, and extracting an arrangement mode preference distribution feature vector; Based on the historical subscription rate curve form data, carrying out normalization processing on the curve, and extracting standardized historical subscription rate form features; according to the compensation standard floating interval parameters, combining the policy file and market fluctuation data to construct a compensation standard floating coefficient matrix; Splicing the region development index standardization value, the house density gradient feature vector, the property complexity grading index, the arrangement mode preference distribution feature vector, the historical subscription rate morphological feature and the compensation standard floating coefficient matrix to construct a project genetic map feature vector space; And mapping the data sample of the history collection and removal project to the project genetic map feature vector space to form a genetic feature matrix.
3. The method for digitized control of collection and migration complete process based on multi-source data fusion according to claim 2, wherein step S1 further comprises adopting missing value mean filling, outlier triple standard deviation checking, Z-score standardization and Min-Max standardization to the regional development index, and using Spearman correlation coefficient to evaluate correlation score with fund mode, and selecting correlation score index with correlation threshold value greater than 0.5.
4. The method for controlling the collection and migration whole-flow digitization based on multi-source data fusion according to claim 1, wherein the step S2 specifically comprises: Constructing a genetic feature matrix based on historical project data in a project genetic map feature vector space; executing a DBSCAN clustering algorithm on the gene feature matrix, setting a neighborhood radius and a minimum sample number parameter, calculating similarity between feature vectors based on Euclidean distance measurement, identifying data clusters with connected densities, and dividing historical item groups with similar capital expenditure rhythms; feature induction and label naming are carried out on each clustering result, and core feature combinations of items in each clustering result are extracted based on feature vector mean values of clustering centers to generate corresponding genotype labels; calculating the cost of the item group corresponding to each genotype label to form a rule statistical parameter; And structuring and organizing the genotype labels and the corresponding cost formation rule statistical parameters, constructing a genotype knowledge base, and storing by adopting a relational database table structure.
5. The method for controlling the collection and migration whole-flow digitization based on multi-source data fusion according to claim 1, wherein the step S3 specifically comprises: generating an initial feature vector of a new project based on the regional development index, the house density gradient, the property complexity score, the arrangement mode preference distribution, the historical subscription rate curve form and the compensation standard floating interval parameter provided in the collection and removal project setting stage; performing normalization processing on the initial feature vector to obtain a normalized feature vector; calculating the distribution difference between the standardized feature vector and each historical genotype central point by using a Wasserstein distance measurement algorithm based on the constructed gene feature matrix to obtain a matching similarity score; according to the matching similarity score, selecting a historical genotype with the highest score as an optimal reference genotype, and extracting a corresponding transducer-Time 2Seq reference prediction model and initial parameter configuration thereof from a genotype knowledge base; And binding an input interface of the reference prediction model with the multisource characteristic tensor of the current stage of the new project to form a complete model input channel.
6. The method for collecting, removing and digitally controlling the whole flow based on multi-source data fusion according to claim 5, wherein the Wasserstein distance measurement algorithm is used for measuring the similarity of heterogeneous multi-dimensional feature distribution, the single-dimensional distance integrates the score according to an empirical distribution function, and feature importance weighted aggregation is adopted under the multi-dimension, so that the standardized feature vector is converted into scoring data of the similarity degree of a new project and each historical genotype structure.
7. The method for controlling the collection and migration whole-flow digitization based on multi-source data fusion according to claim 1, wherein the step S4 specifically comprises: Calculating cycle-by-cycle deviation of funds output by the reference prediction model by using a trend predicted value and an actual funds expenditure curve of the project, and performing dynamic threshold modeling on a deviation sequence based on a sliding window mechanism to obtain a deviation judging signal; Based on the result that the deviation judgment signal exceeds a preset dynamic threshold, generating a situation awareness gating mechanism activation instruction, and loading an external driving factor feature vector related to the current project stage; dynamically adjusting the attention weight distribution of each cost sub-item in the transducer decoder based on the external driving factor feature vector by using an attention weight adjusting module in a gating mechanism to generate an attention weight matrix containing an elastic coefficient correction factor; Updating parameters of the cost increase elastic coefficient in the reference prediction model based on the attention weight matrix to generate an elastic coefficient correction parameter set adapting to the current project characteristic; And feeding back the elastic coefficient correction parameter set to the reference prediction model, and re-executing fund expenditure trend prediction calculation to generate a corrected fund usage trend curve.
8. The method for digitized control of collection and migration complete process based on multi-source data fusion of claim 7 wherein said external driving factor feature vector comprises policy adjustment event codes, staged capital expenditure sequence volatility, and sign-up progress deviation metrics.
9. The method for controlling the collection and migration whole-flow digitization based on multi-source data fusion according to claim 1, wherein the step S5 specifically comprises: modeling different budget adjustment scenes based on the corrected fund use trend prediction result, and generating budget adjustment suggestion packages of multiple types of budget adjustment scenes; Carrying out semantic coding on a target priority label of a current stage of the item, and converting the target priority label into a structured semantic vector representation by using a natural language processing technology; Based on the applicable condition label set in the budget adjustment suggestion package, carrying out item-by-item comparison on semantic vectors of project stage target priorities and semantic vectors of all applicable condition labels by adopting a semantic similarity calculation algorithm, and identifying a most matched budget adjustment scene; constructing a matching degree scoring matrix of a budget adjustment strategy according to the semantic matching result; And carrying out weighted aggregation calculation on the matching degree scoring matrix to generate comprehensive confidence degree scores of each budget adjustment strategy, and sequencing a budget adjustment strategy recommendation set according to the scores from high to low to identify an optimal adjustment scheme.
10. A collection and migration full-flow digital control system based on multi-source data fusion is characterized in that the collection and migration full-flow digital control method based on multi-source data fusion is adopted for collection and migration full-flow digital control.

Description

Convergence and migration full-flow digital management and control method and system based on multi-source data fusion Technical Field The invention relates to the technical field of intelligent monitoring and anomaly detection of urban rainwater pipe networks, in particular to a collecting and removing full-flow digital management and control method and system based on multi-source data fusion. Background The mainstream technical scheme in the field of dynamic prediction of collected removed funds and intelligent budget adjustment mainly adopts a traditional linear regression method based on time sequence analysis and business rule modeling to estimate the trend of the removed funds and budget. These techniques generally consider the project as a homogeneous object, and directly apply a unified mathematical model or simple parameter configuration, failing to fully consider the significant differences of the removed project in terms of regional attributes, property structures, policy guidance, etc., resulting in insufficient model generalization capability. With the advancement of digital management and control of smart cities and government affairs, industry technology has gradually introduced multisource data fusion, an end-to-end neural network and a certain degree of online feedback mechanism, in an effort to improve dynamic responsiveness of fund prediction, but obvious short plates still exist in project type adaptation surfaces; At present, the main problems existing in the prior art are as follows: (1) The model has poor adaptability, the prior art carries out fund trend prediction on all the removed projects by using a fixed parameter or unified modeling method, the dynamic adaptation capability aiming at different project types is lacked, the model generalization is weak, and the prediction misalignment is easy to cause when facing project heterogeneity or policy environment mutation; (2) The heterogeneous factors are absent, the mainstream technology is usually focused on the traditional progress, indexes and historical average values, the core characteristics affecting the fund expenditure such as regional development indexes, property complexity scores, arrangement mode preference and the like cannot be structurally modeled, and the fund prediction logic is single; (3) The existing method mainly processes prediction errors by manual intervention or post-correction, and is difficult to deal with real-time change of fund use modes in the process of removing due to lack of intelligent feedback and closed-loop optimization mechanisms and lack of automatic sensing abnormality and model parameter self-adaptive adjustment means; (4) The budget adjustment suggestion is not intelligent enough, and a plurality of methods can only provide budget adjustment suggestion based on fixed ratio or historical experience, and lack high-confidence, multi-scenario and intelligent budget adjustment suggestions for complex scenes such as actual migration progress, subscription rate fluctuation, policy event influence and the like; (5) The process traceability is poor, and the current technology lacks an efficient traceability evidence-preserving mechanism in the model parameter change, fund prediction and budget adjustment strategy execution links, so that the process audit and business compliance requirements cannot be effectively supported. Disclosure of Invention The invention aims to solve the technical problems and provides a collection and migration full-flow digital management and control method based on multi-source data fusion. The technical scheme of the invention is realized by a collection and migration full-flow digital management and control method based on multi-source data fusion, which comprises the following steps: S1, constructing a project genetic map feature vector space based on regional development indexes, house density gradients, property complexity scores, arrangement mode preference distribution, historical subscription rate curve forms and compensation standard floating interval parameters of collection and migration projects, and mapping historical project data to the space to form a genetic feature matrix; S2, executing a DBSCAN clustering algorithm on the gene feature matrix, identifying typical project genotypes with similar capital expenditure rhythms, and generating a genotype knowledge base containing genotype labels and corresponding cost formation rules; S3, inputting new project item data into the genetic map feature vector space, calculating the Wasserstein distance between the new project item data and each genotype, and activating a corresponding transform-Time 2Seq reference prediction model after matching with the optimal reference genotype, wherein the model inputs a multisource feature tensor comprising project progress indexes, staged capital expenditure sequences and policy adjustment event codes; S4, based on the deviation amount of the fund usage trend predicted value a