A process execution monitoring system and method suitable for engineering cost consulting projects

Abstract

The application relates to the technical field of engineering cost supervision, and discloses a process execution supervision system and method suitable for engineering cost consultation projects. The method acquires a project whole-cycle data set through a distributed data acquisition interface, and the data set at least contains contract files and progress payment vouchers; the data set is subjected to structured feature analysis to generate a multi-dimensional project feature matrix containing time nodes, cost indexes and resource allocation amounts; a pre-constructed engineering cost benchmark parameter library is called, real-time deviation degrees of the multi-dimensional project feature matrix and the benchmark parameters are calculated based on a dynamic time warping algorithm; when the real-time deviation degrees exceed a preset threshold range, an abnormal procedure identification engine is automatically triggered to extract procedure feature vectors including a construction period delay rate, a cost overrun rate and a material loss rate from deviated procedures; the procedure feature vectors are input into a pre-trained compliance verification model to output an abnormal diagnosis report containing a violation type, a risk level and a responsible subject.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engineering cost supervision, in particular to a process execution supervision system and method suitable for engineering cost consulting projects. BACKGROUND

[0002] In the current engineering cost consulting industry, the supervision of project execution process faces many challenges. Engineering cost consulting projects often have long cycles, multiple participants, and involve complex files and documents, including contract documents, design change notices, engineering quantity lists, progress payment vouchers, and other types of key data. These data are usually scattered in different business systems, departments, or among participants, forming "data silos" and making it difficult to achieve efficient integration and real-time sharing. Traditional supervision methods rely heavily on manual sorting and analysis, which not only consumes a lot of manpower and time, but also is prone to data omissions or biases due to human errors.

[0003] During project execution, due to the lack of dynamic supervision methods, project managers have difficulty in real-time monitoring of project progress, cost consumption, and resource allocation. When problems such as project delay, cost overrun, or resource mismatch occur, they often cannot be detected in a timely manner, leading to the accumulation and expansion of problems and ultimately affecting the overall efficiency of the project. At the same time, existing supervision methods analyze project characteristics from a single dimension, making it difficult to build a multi-dimensional and comprehensive project characteristic system, and unable to accurately reflect the real state and potential risks of the project.

[0004] In terms of anomaly identification and diagnosis, traditional methods rely mainly on experience-based judgment, lacking standardized analysis models and scientific evaluation methods. When anomalies occur, it is difficult to quickly locate the problem, identify the type of violation and the responsible party, leading to delayed problem resolution and lack of targeted corrective measures. This passive supervision mode not only fails to effectively prevent project risks, but also may lead to contract disputes, economic losses, and other negative consequences due to inadequate supervision, severely restricting the standardization and efficiency of the engineering cost consulting industry.

[0005] With the rapid development of the construction industry, the complexity and scale of engineering cost consulting projects are continuously increasing, placing higher demands on the accuracy, real-time nature, and intelligence of supervision. The existing supervision methods that rely on manual, decentralized, and lagging methods cannot meet the needs of industry development, and there is an urgent need for a process execution supervision method that can achieve whole-cycle data integration, dynamic characteristic analysis, real-time deviation monitoring, and intelligent anomaly diagnosis to improve the supervision level and management efficiency of engineering cost consulting projects. SUMMARY

[0006] The present application aims to provide a process execution supervision method suitable for engineering cost consulting projects to solve the problems raised in the background.

[0007] To achieve the above object, the application provides a process execution supervision method suitable for engineering cost consulting projects, which comprises the following steps:

[0008] Obtaining an engineering cost consulting project full-cycle data set through a distributed data acquisition interface, wherein the data set at least contains contract files, design change notice sheets, bill of quantities and progress payment vouchers;

[0009] Structurally analyzing the data set to generate a multi-dimensional project feature matrix containing time nodes, cost indicators and resource allocation amounts;

[0010] Calling a pre-constructed engineering cost benchmark parameter library and calculating the real-time deviation degree of the multi-dimensional project feature matrix from the benchmark parameters based on a dynamic time warping algorithm;

[0011] When detecting that the real-time deviation degree exceeds a preset threshold range, automatically triggering a process abnormality recognition engine to extract a process feature vector including a delay rate, a cost overrun rate and a material loss rate from the deviated process;

[0012] Inputting the process feature vector into a pre-trained compliance verification model to output an abnormality diagnosis report containing a violation type, a risk level and a responsible subject.

[0013] Preferably, the method further comprises:

[0014] Generating a visual warning signal according to the abnormality diagnosis report and storing the warning signal in association with the original process data in a blockchain storage module;

[0015] Based on a historical abnormality diagnosis report set, performing a classification analysis on abnormality patterns by using a spectral clustering algorithm to identify a high-frequency violation pattern set;

[0016] Optimizing the decision boundary of the compliance verification model by using the high-frequency violation pattern set to generate a model parameter update instruction;

[0017] According to the matching result of the real-time deviation degree and the high-frequency violation pattern set, dynamically adjusting the monitoring frequency threshold of the subsequent process;

[0018] Integrating the monitoring results of all processes to generate a process optimization scheme containing corrective measures, responsibility tracing paths and prevention strategies.

[0019] Preferably, the structural feature analysis of the data set comprises:

[0020] Establishing an engineering cost entity relationship model based on a knowledge graph to extract entity attributes such as contract subjects, pricing clauses and change bases from unstructured texts;

[0021] The extracted entity attributes are aligned in time sequence by using a multi-head attention mechanism to form a timestamped project feature sequence.

[0022] A multi-dimensional project feature matrix representing the correlation of the process is generated by a feature cross-fusion algorithm.

[0023] Preferably, the real-time deviation degree of the multi-dimensional project feature matrix from the benchmark parameter calculated based on the dynamic time warping algorithm comprises:

[0024] Load the standard duration curve, cost control curve, and resource allocation curve in the benchmark parameter library.

[0025] Calculate the morphological difference degree of the actual process feature sequence from each benchmark curve by using the dynamic time warping distance algorithm.

[0026] The multiple morphological difference degrees are integrated into a real-time deviation degree by a weighted fusion algorithm.

[0027] Preferably, the process anomaly recognition engine comprises:

[0028] An abnormal feature space containing the duration compression coefficient, cost fluctuation coefficient, and resource conflict coefficient is constructed.

[0029] An isolated forest algorithm is used to identify the deviated process cluster in the abnormal feature space.

[0030] A feature combination of the maximum deviation dimension is extracted from the deviated process cluster to form a process feature vector.

[0031] Preferably, the compliance verification model comprises:

[0032] The process feature vector is mapped to a pre-defined violation mode feature space.

[0033] The matching probability of the vector with each type of violation mode is calculated by a multi-layer perception network.

[0034] When the highest matching probability exceeds the confidence threshold, the corresponding violation type code and risk level label are output.

[0035] Preferably, the generation of visual warning signals comprises:

[0036] The warning color coding of the corresponding level is activated according to the risk level label.

[0037] A three-dimensional heat map containing the deviated process number, abnormal time interval, and violation details is generated.

[0038] The heat map is linked to the original credentials in the blockchain storage module to establish a verifiable data link.

[0039] Preferably, the classification analysis of the abnormal mode by using the spectral clustering algorithm comprises:

[0040] extracting the violation type code chain and the time sequence characteristics in the historical anomaly diagnosis report;

[0041] constructing a violation mode propagation relation graph through a graph convolution network;

[0042] adopting a spectral clustering algorithm to identify a high-frequency violation mode subgraph with a stable topological structure.

[0043] Preferably, the decision boundary of the compliance verification model optimized by the high-frequency violation mode set comprises:

[0044] converting the high-frequency violation mode subgraph into a decision hyperplane in a feature space;

[0045] calculating the geometric distance between the existing decision boundary of the compliance verification model and the hyperplane;

[0046] generating a neuron weight adjustment instruction when the geometric distance exceeds a tolerance threshold.

[0047] Preferably, the dynamic adjustment of the monitoring frequency threshold of the subsequent process comprises:

[0048] establishing an association rule table of the high-frequency violation mode set and the real-time deviation degree;

[0049] when the new process feature matches a specific violation mode, activating the enhanced monitoring instruction in the association rule;

[0050] amplifying the monitoring sampling frequency of the subsequent process according to the instruction intensity index.

[0051] Preferably, the present application further comprises a process execution supervision system suitable for an engineering cost consulting project, which is used to realize the above-mentioned process execution supervision method suitable for an engineering cost consulting project, comprising:

[0052] a distributed data acquisition module, which is used to acquire a whole-cycle data set of the engineering cost consulting project through a distributed data acquisition interface, wherein the data set at least contains a contract file, a design change notice, a bill of quantities and a progress payment voucher;

[0053] a structured feature analysis module, which is connected to the distributed data acquisition module and is used to analyze the structured features of the data set to generate a multi-dimensional project feature matrix containing time nodes, cost indicators and resource allocation amounts;

[0054] a deviation degree calculation module, which is connected to the structured feature analysis module and is internally provided with a pre-constructed engineering cost benchmark parameter library, and is used to call the benchmark parameter library and calculate the real-time deviation degree of the multi-dimensional project feature matrix and the benchmark parameters based on a dynamic time warping algorithm;

[0055] The process exception identification module is connected with the deviation degree calculation module, and is used for triggering a process exception identification engine automatically when it is detected that the real-time deviation degree exceeds a preset threshold range, and extracting a process feature vector including a time delay rate, a cost overrun rate and a material loss rate from the deviated process;

[0056] The compliance verification and report generation module is connected with the process exception identification module, and is internally provided with a pre-trained compliance verification model, which is used for inputting the process feature vector into the compliance verification model and outputting an exception diagnosis report containing a violation type, a risk level and a responsible subject.

[0057] Compared with the prior art, the beneficial effects of the present application are:

[0058] The process execution supervision method suitable for engineering cost consulting projects breaks the "island" state of scattered data in traditional supervision by acquiring a project full-cycle data set through a distributed data acquisition interface, and realizes the comprehensive integration of key data such as contract files, design change notice sheets, engineering quantity lists and progress payment vouchers. The centralized collection of full-cycle data enables all parties involved in the project to carry out supervision work based on a unified data basis, avoids information bias caused by inconsistent data sources, and enables supervisors to fully understand the actual situation of each stage of the project.

[0059] The structured feature analysis of the data set generates a multi-dimensional project feature matrix, which converts the originally scattered project data into structured information containing time nodes, cost indicators and resource allocation amounts. This structured processing method breaks the dependence on single data dimensions in traditional manual analysis, and through the correlation analysis of multi-dimensional features, the running state of the project can be more accurately described. Supervisors can examine the project progress from multiple angles such as time, cost and resources, find potential correlations and contradictions between different dimensions, and provide more comprehensive analysis perspectives for project supervision.

[0060] The pre-built engineering cost benchmark parameter library is called, and the real-time deviation degree is calculated by combining the dynamic time warping algorithm, which realizes the dynamic comparison between the actual state of the project and the benchmark parameters. The dynamic time warping algorithm can effectively handle the nonlinear changes of project data in time series, so that even if there is a certain fluctuation in project progress or cost consumption, the deviation degree of the actual features and the benchmark parameters can be accurately calculated. This real-time deviation monitoring mechanism changes the passive situation of traditional post-analysis supervision, enabling supervisors to timely detect abnormal trends in project operation.

[0061] When the real-time deviation exceeds the preset threshold range, the process exception recognition engine is automatically triggered and the process feature vector is extracted, realizing accurate positioning of the abnormal problem. The process feature vector covers key indicators such as time delay rate, cost overrun rate, and material loss rate, focusing on the core features of the deviated process and avoiding irrelevant information interference. Through analysis of these feature vectors, the specific process where the problem occurs can be quickly locked, and the key dimensions of abnormal performance are clear, providing a clear analysis object for subsequent problem diagnosis.

[0062] The process feature vector is input into the pre-trained compliance verification model and an abnormal diagnosis report is output, providing a systematic basis for the processing of abnormal problems. The diagnosis report includes key information such as violation type, risk level, and responsible subject, enabling regulatory personnel to fully understand the nature, severity, and related responsibility of abnormal problems. This standardized diagnosis result eliminates the subjectivity of traditional experience judgment, making abnormal processing more targeted and scientific, which helps to promote the standardization and standardization of project supervision work and promote the efficient operation of engineering cost consulting projects on the compliance track. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 The working principle diagram of the process execution supervision method for engineering cost consulting projects according to the present application;

[0064] Figure 2 The flow chart of the process execution supervision optimization and early warning method for engineering cost consulting projects;

[0065] Figure 3 The flow chart of the engineering cost data set structured feature analysis method;

[0066] Figure 4 The flow chart of the visual early warning signal generation method. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0068] Please refer to Figure 1 The present application provides a process execution supervision system and method for engineering cost consulting projects, which comprises:

[0069] The distributed data acquisition interface is used to obtain the whole-cycle data set of the engineering cost consulting project, which includes structured and unstructured data such as contract files, design change notices, bill of quantities, and progress payment vouchers. The data acquisition process uses an asynchronous message queue to realize real-time synchronization of multi-source heterogeneous data, ensuring data integrity. After the data set is obtained, it is cleaned by the data cleaning module to remove duplicate records and invalid fields, and then enters the feature analysis stage. The structured feature analysis process uses entity relationship extraction technology based on knowledge graph to convert the original data into a standardized feature vector containing time nodes, cost indicators, and resource allocation quantities, and constructs a multi-dimensional project feature matrix through tensor operation. The engineering cost benchmark parameter library stores industry standard parameter sets, including standard duration, cost control line, and resource allocation scheme for different project types. The dynamic time warping algorithm calculates the shape similarity between the actual feature matrix and the benchmark parameters, and outputs the real-time deviation index of each dimension. When the deviation exceeds the preset threshold, the anomaly recognition engine starts the feature extraction process to separate key indicators such as delay rate and cost overrun rate from the deviation data to form a feature vector. The compliance verification model uses a deep neural network architecture to calculate the matching degree of input features and predefined violation patterns through forward propagation, and finally generates an anomaly diagnosis report containing risk level and responsible subject identification.

[0070] Embodiment 1: refer to Figure 2 , which involves the whole-process processing mechanism after the generation of the anomaly diagnosis report, including visual warning, blockchain notarization, historical pattern analysis, and dynamic optimization. The system immediately starts the warning signal generation process after generating the anomaly diagnosis report, which uses multi-dimensional parameter mapping technology to convert the report content into visual elements. The color coding of the warning signal is based on the HSL color model, which distinguishes different risk types by adjusting the hue parameter, reflects the severity of the problem by the saturation parameter, and identifies the processing urgency by the lightness parameter. The three-dimensional heat map generation module uses the WebGL graphics library to realize real-time rendering on the browser side, maps the process number to the X-axis coordinate, projects the abnormal time period to the Y-axis, and distributes the importance indicators of the violation details in the Z-axis. The size of the heat map node is proportional to the deviation, and the color depth increases with the risk level, forming a three-dimensional visual presentation effect.

[0071] The blockchain storage module is built with a consortium chain architecture, containing multiple accounting nodes that jointly maintain a distributed ledger. The early warning signals and raw data are hashed to generate digital fingerprints, which are bound to timestamps by smart contracts and written to the blockchain. The storage process uses asymmetric encryption technology to ensure data transmission security, and each block contains the hash value of the previous block, forming an unalterable chain structure. The historical anomaly report set is processed using time window slicing technology, dividing the analysis period into weeks, and the violation events within each period are sorted by occurrence time to form an event sequence. In the violation pattern analysis stage, an event correlation graph is first constructed, with nodes representing different types of violation behaviors and edge weights representing the temporal correlation strength of event occurrence.

[0072] The spectral clustering algorithm performs a multi-stage computation process in violation pattern recognition. First, a similarity matrix of violation events is constructed, quantifying the distance between event feature vectors using a Gaussian kernel function. Then, the Laplacian matrix is calculated and decomposed, and the top K largest eigenvalues are selected to form a new feature space. In the reduced space, a clustering algorithm is used to divide the violation events into several categories, identifying high-frequency violation patterns that repeatedly occur. These patterns are converted into decision rules in the feature space, which are used to optimize the judgment logic of the compliance verification model.

[0073] The model optimization process uses an incremental learning strategy, incorporating newly discovered violation patterns while preserving existing knowledge. Decision boundary adjustment is achieved by calculating the geometric relationship between the existing classification hyperplane and the new pattern cluster. When a new pattern cluster is detected with a significant distance from the original boundary, the system generates neuron weight adjustment instructions. These instructions update the neural network parameters through the backpropagation algorithm, causing the decision boundary to expand towards the new pattern cluster while maintaining the ability to distinguish other categories. The dynamic adjustment mechanism of monitoring frequency is based on the Bayesian probability framework, where the system continuously evaluates the matching degree of the current deviation features with historical patterns. The matching degree calculation considers the joint probability distribution of multiple dimensions, and when the matching probability of a specific violation pattern exceeds a certain threshold, an enhanced monitoring instruction is triggered.

[0074] The generation of process optimization schemes uses a case-based reasoning approach. The system maintains a case library containing historical handling measures, and finds the most similar historical case to the current situation through similarity retrieval. The rectification measure suggestion integrates successful experiences from the case library and industry standard specifications to form actionable implementation steps. The construction of responsibility tracing paths uses a graph traversal algorithm, starting from the process node where the anomaly occurred and searching for associated responsible parties along the upstream and downstream directions of the project flow. The development of prevention strategies combines feature analysis of high-frequency violation patterns to propose targeted suggestions for system improvement and technical improvement schemes.

[0075] The association verification of the early warning signal and the original data is implemented by using a Merkle tree structure. The system calculates hash values of the early warning metadata and the corresponding voucher files of the process respectively, aggregates them layer by layer upwards according to the binary tree structure, and finally generates a unified root hash value. When verifying, the integrity of the entire set of data can be confirmed by comparing the root hash value, without the need to check the original files one by one. This mechanism not only guarantees the verifiability of the data, but also improves the verification efficiency. The storage of the historical abnormal report adopts a time series database architecture, and multiple indexes are established according to the project stage and the occurrence time, supporting historical query and statistical analysis under complex conditions.

[0076] The update and maintenance of the high-frequency violation mode adopts a version control mechanism. Whenever a new violation mode is found, the system will create a new mode version, and retain the decision rules of the historical version. This mechanism allows rollback to the previous mode set when necessary, ensuring the stability of the system during the update process. The adjustment of the monitoring frequency adopts a gradual strategy, dynamically adjusting the sampling interval according to the stability of the recent monitoring results. When no abnormality is detected for consecutive multiple periods, the system will gradually relax the monitoring frequency; otherwise, it will immediately increase the monitoring intensity. This flexible mechanism optimizes the system resource consumption while ensuring the monitoring effect.

[0077] The interactive function of the visualization interface supports multi-angle data analysis. Users can observe the data distribution in different dimensions by rotating the three-dimensional heat map, and view detailed diagnostic reports by clicking specific nodes. The system also provides a time axis scaling function, allowing users to focus on analysis of a specific time period. The blockchain storage information provides query services through a special verification interface, and users can input transaction hash values to verify the storage records of specific early warning signals. The historical mode analysis results are displayed in the form of a graph, clearly presenting the association between various types of violation behaviors and the development law.

[0078] The dynamic optimization process establishes a closed-loop mechanism for continuous improvement. After each new violation mode is found, the system evaluates the effectiveness of the existing monitoring strategy and automatically adjusts the relevant parameters. The optimization decision considers the comprehensive influence of multiple factors, including the frequency of the mode, the degree of loss caused, the difficulty of processing, etc. The system regularly generates an optimization effect evaluation report, records the implementation and actual effect of each adjustment measure, and provides a reference basis for subsequent optimization. The entire processing flow forms a complete closed loop from abnormality detection to measure improvement, realizing the continuous improvement of the project supervision capability.

[0079] Embodiment 2: see Figure 3The structured features of project data in engineering cost consulting are analyzed and the real-time deviation is calculated to realize accurate monitoring of project execution status through multi-stage processing. The original information obtained in the data collection stage includes contract texts, design change notices, and engineering quantity lists in various formats. These materials first enter the preprocessing link. The document parsing engine identifies the structural features of different file types. The contract file uses a clause segmentation algorithm to divide the text into logical units such as responsibility clauses, pricing clauses, and breach of contract clauses. The design change notice is extracted through layout analysis to extract the core elements such as change content, approval signature, etc. The unstructured text processing uses a model architecture combining bidirectional long short-term memory network and conditional random field. This model is pre-trained on the corpus in the field of engineering cost and can accurately identify professional terms and key data in the text.

[0080] The time alignment process establishes a unified time reference system for scattered data generated throughout the project cycle. The multi-head attention mechanism analyzes the context of time expressions in different documents, addressing issues such as inconsistent date formats and inconsistent time reference points. The system maintains a time calibration table, recording the date expression habits of each participant, and automatically performs time zone conversion and format standardization when processing new data. The feature extraction process not only identifies explicit time markers but also infers implicit time information in documents through semantic analysis. Relative time expressions such as "within 30 days of commencement" are converted to absolute time coordinates. Data that have completed time alignment are marked with precise timestamps, forming a sequence of features with temporal relationships.

[0081] The feature cross-fusion stage analyzes the internal relationships between different data sources. The pricing method in the contract clauses is mapped and matched with the project division in the engineering quantity list, and the content of the design change is verified with the approval records of the progress payment vouchers. The system uses tensor operation methods to construct a high-dimensional feature space, converting discrete data points into continuous feature representations. The feature vector of each process node includes time dimension, cost dimension, and resource dimension, among other quantitative indicators. These indicators are transformed into composite features with engineering significance through nonlinear transformation. The construction of the feature matrix considers the dependencies between processes, capturing long-range dependency patterns in the process through the introduction of graph neural networks.

[0082] The construction of the benchmark parameter library is based on statistical analysis results of a large number of historical projects. The standard duration curve is modeled using quantile regression methods, reflecting the duration distribution of different types of processes under normal circumstances. The cost control curve considers market factors such as material price fluctuations and labor rate changes, establishing a dynamically adjusted baseline. The resource allocation curve is generated through a constrained optimization algorithm, providing theoretical allocation schemes for various resources at different process stages. The benchmark parameters are stored in multiple dimensions such as project type, size level, and regional characteristics. When matching the current project, the nearest neighbor search algorithm is used to find the most relevant parameter set.

[0083] The implementation of the dynamic time warping algorithm includes two main stages: path search and distance calculation. The algorithm first establishes a correspondence between the baseline curve and the actual feature sequence, allowing elastic matching on the time axis to eliminate deviations caused by normal fluctuations. Path constraints limit the range of matching stretching to avoid distorted alignment that does not conform to engineering logic. The morphological difference degree calculation uses a segmented comparison strategy, dividing the entire project period into several feature intervals, and independently calculating the deviation of the actual sequence from the baseline in each interval. The difference degree index includes shape difference, amplitude difference, and trend difference, which reflect the rhythm change, size deviation, and development direction deviation of the execution process.

[0084] The weighted fusion process uses an adaptive weight distribution strategy. The system dynamically adjusts the weight coefficients of each dimension index according to the characteristics of the current stage of the project, paying more attention to schedule deviation at the beginning of the project, focusing on resource matching degree in the middle stage, and emphasizing cost control in the later stage. The weight coefficients are determined through a fuzzy reasoning mechanism, considering factors such as project type, environmental factors, and management focus. The calculation result of real-time deviation degree is the weighted combination of each dimension difference degree, which is normalized to form a standard score between 0 and 100. This score is visually displayed through color coding, helping managers quickly identify the severity of the problem.

[0085] The generation of the standard duration curve uses a statistical learning method. After cleaning and standardizing the historical project data, clustering analysis is performed according to the process type, and the standard duration of each type of process is extracted. The curve modeling considers the logical relationship between processes, and the processes on the critical path use a stricter tolerance range. The establishment of the cost control curve introduces external economic indicators as covariates, eliminating the impact of market fluctuations on the baseline value through regression analysis. The curve updating mechanism periodically absorbs data from newly completed projects to maintain the timeliness of the baseline parameters. The optimization goal of the resource allocation curve is to minimize resource idle rate while meeting the duration requirements, and a heuristic algorithm is used to solve the optimal configuration scheme under multiple constraints.

[0086] The deviation monitoring system uses an event-driven architecture design. When new process data arrives, the system automatically triggers the feature analysis and deviation calculation process. The monitoring results are updated in real time to the project dashboard, and abnormal situations immediately trigger the warning mechanism. The system supports multi-granularity monitoring views, allowing managers to grasp the overall project execution status or analyze the detailed deviation of specific processes. Historical deviation records are stored in chronological order, supporting trend analysis and periodic pattern mining. Managers can compare deviation patterns at different stages to identify systematic problems and improvement opportunities in project management.

[0087] Data storage and access adopt a hierarchical architecture to achieve efficient management. Raw documents are stored in a distributed file system, and structured feature data is saved in a time-series database for fast querying and analysis. The system interface supports multiple data access modes, including batch export, real-time push, and conditional query. The security mechanism uses role-based access control, and managers at different levels can only view data and reports within their permission scope. Audit logs record all key operations to support operation tracing and responsibility identification.

[0088] The entire implementation process emphasizes the coordinated operation of each link. The quality of feature analysis directly affects the accuracy of subsequent analysis, so the system sets strict data verification rules. The applicability of benchmark parameters is ensured through similarity evaluation, avoiding the application of irrelevant historical standards to the current project. The deviation calculation considers the reasonable variation range in engineering practice, avoiding overreaction to normal fluctuations. The experience data accumulated during system operation is used to improve the feature analysis model and optimize the benchmark parameters, forming a continuous improvement positive cycle. This closed-loop design enables the system to adapt to the characteristics of different types of engineering projects and provide precise process supervision services.

[0089] Example 3: Focuses on the specific implementation process of process anomaly identification and compliance verification, and realizes the precise identification of project execution risks through the construction of a multi-level anomaly detection and classification system. The construction of the anomaly feature space starts from the original monitoring data, and the principal component analysis method is used to reduce the dimension of high-dimensional features. This method converts possibly correlated variables into linearly uncorrelated principal components through orthogonal transformation, forming a new feature coordinate system. In the field of engineering cost, the duration compression coefficient is obtained by calculating the ratio of actual duration to planned duration, reflecting the abnormal situation of progress management; the cost fluctuation coefficient uses the moving standard deviation method to measure the intensity of expenditure changes; the resource conflict coefficient quantifies the competition intensity between different processes. These features are standardized to form a new anomaly detection space, which has significantly lower dimension than the original feature space but retains most of the key information.

[0090] The implementation of the Isolation Forest algorithm in the anomaly feature space adopts a recursive partitioning strategy. The algorithm constructs multiple randomly generated isolation trees, each tree divides the data space into nested hyper-rectangular regions by randomly selecting features and splitting values. Normal data points are usually located in deeper nodes, while abnormal points can be isolated by shallower splitting surfaces. In the context of engineering cost supervision, the isolation process of this algorithm considers the engineering significance of process features, avoiding the generation of division methods that violate industry common sense. The path length calculation phase records the number of edges required for each data point to reach the isolation node, forming the basis for anomaly score evaluation. The detection of deviating process clusters uses a density-based clustering method, which can identify areas with uneven density distribution in the feature space and identify process sets with similar anomaly patterns.

[0091] The extraction process of the procedure feature vector implements a feature selection strategy to identify the most discriminative feature combination from the deviated procedure cluster. The system evaluates the importance of features using the maximum-relevance minimum-redundancy criterion, which requires the selected features to be highly relevant to the abnormal class while maintaining low redundancy among features. For the schedule delay abnormality, the system may select features such as schedule compression factor, critical path float time, etc.; for the cost overrun abnormality, it focuses on indicators such as cost fluctuation factor, material price differential rate, etc. The final procedure feature vector contains both the original monitoring data and the derived variables generated through feature engineering, comprehensively representing the abnormal characteristics of the procedure.

[0092] The architecture design of the compliance verification model adopts a deep neural network. The input layer receives the procedure feature vector, with dimensions consistent with the feature selection results. Three hidden layers are configured with different numbers of neurons, gradually abstracting higher-level violation pattern features. The hidden layer activation function uses the Leaky ReLU, which preserves a small gradient in the negative input region to avoid neuron death. The construction of the violation pattern feature space uses dimensionality reduction visualization technology to project high-dimensional features onto a two-dimensional plane for manual analysis. Each region in this space corresponds to a predefined violation type, and the region boundary is determined by the feature distribution of historical violation cases.

[0093] The matching probability calculation process implements normalization processing to make the output results have a probability interpretation. Given the procedure feature vector , the probability that it belongs to the th violation pattern is calculated by the Softmax function:

[0094]

[0095] where represents the original activation value of the th output node, and is the total number of violation types. This formula ensures that the sum of the probabilities of all categories is 1, facilitating the comparison of the relative likelihood of different violation types. The generation of the risk level label is based on the pre-configured mapping rules, which convert the probability output into discrete risk levels. For example, probability values in the 0-0.3 interval are mapped to low risk, 0.3-0.7 to medium risk, and 0.7 and above to high risk. The responsibility subject identification process combines contract clause analysis and procedure participant records to extract relevant responsible party information through named entity recognition technology.

[0096] The model training process adopts a phased strategy. In the initial stage, the historical violation case dataset is used for pre-training to establish basic violation identification capabilities. In the subsequent stage, online learning is used, and newly confirmed violation cases are continuously added to the training set to gradually improve the model's identification range. The training algorithm uses the adaptive moment estimation optimizer to automatically adjust the learning rate of each parameter. To prevent overfitting, the model introduces a dropout layer and an L2 regularization term, which randomly masks some neuron connections during the training process and penalizes large weight parameters.

[0097] The operation of the anomaly detection and classification system uses an event-driven architecture. When new process data arrives, the system automatically triggers the feature extraction and anomaly scoring process. Processes with anomaly scores exceeding the threshold enter the detailed analysis stage, generating a diagnostic report containing multiple candidate violation types. The system supports a manual review mechanism, allowing managers to confirm or correct the automatic diagnosis results, which are used to improve subsequent analysis models. The historical violation case library is continuously updated, recording the feature vector, final identified type, and handling measures of each violation event, forming a constantly enriched knowledge base.

[0098] The visualization analysis interface provides multi-angle data exploration functions. Users can observe the distribution of processes in the anomaly feature space through scatter plots and analyze the correlation between different features using heat maps. The projection map of the violation pattern feature space displays the clustering areas of various violation cases, helping to understand the model decision basis. The diagnostic report display interface uses a hierarchical presentation method, first displaying the highest-risk violation type, and supports expanding to view detailed feature analysis and similar historical cases.

[0099] During system implementation, special attention is paid to the particularity of engineering practice. Anomaly detection not only considers statistical deviation but also combines business rules in the field of engineering cost. For example, cost increases caused by certain design changes are considered as compliant behavior, and the system needs to make a comprehensive judgment combined with change approval records. The handling logic of the compliance verification model remains transparent, and important decisions can be traced back to specific feature influencing factors. System performance optimization is targeted at the characteristics of engineering cost data, including the ability to handle a large number of numerical features and the efficiency of parsing semi-structured engineering documents.

[0100] The model update mechanism realizes the continuous accumulation of knowledge. Whenever a new violation pattern is discovered or an existing pattern changes significantly, the system will start the model retraining process. The update process uses a rolling time window strategy, prioritizing the impact of recent cases while appropriately retaining representative historical cases. The version control system records the content and effect of each model update, supporting version rollback if necessary. This mechanism enables the system to adapt to changes in industry policies and the emergence of new violation patterns, maintaining long-term effectiveness.

[0101] The whole implementation process emphasizes the coordination and balance of each link. The sensitivity and false positive rate of anomaly detection need to be balanced to avoid excessive interference with normal engineering management activities. Compliance judgment considers the specific project background. The same numerical characteristics may have different compliance implications under different contract clauses. The system provides rich contextual information to assist decision-making, including benchmark data from similar projects and historical violation case handling results. The analysis log generated during operation is used to periodically evaluate system performance and identify areas for improvement. This closed-loop design enables continuous optimization of the anomaly identification and compliance verification system, providing reliable process monitoring support for engineering cost consulting projects.

[0102] Example 4: Refer to Figure 4 , focusing on visual early warning and violation pattern analysis, the system operation mechanism is illustrated through specific engineering cases. In a commercial complex cost consulting project, the system detects cost abnormal fluctuations in the curtain wall engineering process, triggering the early warning generation process. The risk level assessment module labels this anomaly as an "orange" level warning according to the deviation degree, corresponding to the visual parameters of hue 30 degrees, saturation 80%, and lightness 90% in the HSL color model. The three-dimensional heat map generation engine extracts the key attributes of the abnormal process: process number C-203 (X-axis coordinate value 203), abnormal time period March 15-25, 2023 (Y-axis timestamp interval value), and cost overrun rate 28% (Z-axis height value 28), generating an interactive three-dimensional data point.

[0103] When the blockchain storage module handles this early warning event, the system automatically packages the following original credentials: curtain wall engineering subcontract clause 5.2 (providing material price difference adjustment mechanism), material price confirmation sheet on March 18 (recording a 23% increase in aluminum profile market price), and progress payment application (showing actual usage exceeding budget by 12%). After generating digital fingerprints using the SHA-3 algorithm, these files are written to the distributed ledger based on Hyperledger Fabric, and the transaction record contains the information shown in Table 1.

[0104] Table 1: Blockchain storage transaction record.

[0105]

[0106] The historical anomaly analysis module scans the violation records of similar projects in the past six months and finds three similar cases. When the spectral clustering algorithm processes these historical violation events, it first constructs a violation propagation graph containing 12 nodes, with an edge weight of 0.78 between the "material price difference dispute" node and the "progress payment delay" node, indicating strong correlation. When analyzing this topology using a graph convolution network, it identifies a violation pattern subgraph with "design change-material replacement-price dispute" as the typical path. This subgraph contains 5 tightly connected nodes, reflecting the complete transmission chain from design adjustment to final cost overrun.

[0107] When the visualization interface presents the analysis results, the force-directed graph layout algorithm is used to display the propagation relationship of violation patterns. The size of the node represents the frequency of occurrence of this type of violation, and the thickness of the edge reflects the strength of the temporal association between events. When the user clicks on the "material price difference dispute" node, the system pops up a list of associated cases, showing that the average processing period for similar events in the past six months is 22 working days, and the fluctuation range of the amount involved is between 15% and 35%. This kind of visual analysis helps project managers understand the positioning of the current anomaly in the historical context.

[0108] The decision boundary optimization module converts the newly identified violation patterns into geometric constraints in the feature space. The system detects that the existing compliance verification model has too loose a judgment boundary for the "design change triggered material price difference" type of violation, resulting in early risk signals being ignored. The optimization process calculates the distance between the newly added pattern cluster center and the original decision hyperplane, and when this distance exceeds the pre-set tolerance threshold, it generates a neural network adjustment instruction. The specific operations include: increasing the number of neurons in the second hidden layer to 128, adjusting the weight initialization strategy of the fully connected layer, and adding a reinforcement penalty term for the material price difference feature in the loss function.

[0109] The verification link between the three-dimensional heat map and the blockchain data is realized through the Merkle tree structure. Taking the curtain wall engineering early warning as an example, the system constructs a hash tree containing the following nodes: the leaf nodes are the hash values of each original certificate, the intermediate nodes contain the early warning metadata hash, and the root node integrates all the verification information. When the auditor needs to verify the authenticity of a certain early warning, the system only needs to provide the path hash sequence from the leaf node to the root node, and the integrity proof can be completed without the need to retrieve all the original files. This mechanism ensures the credibility of the data while significantly reducing the complexity of the verification work.

[0110] The dynamic update of high-frequency violation patterns uses a version control strategy. When the "material price difference dispute" type of violation repeatedly occurs in three consecutive projects, the system automatically creates a new pattern version v3.2, which strengthens the check on the time association between design change notifications and material valuation documents. The version iteration record contains the pattern feature change explanation, the effective time and the impact range evaluation, and the manager can view the difference comparison between versions at any time. This mechanism ensures that the pattern update process is transparent and controllable, avoiding the introduction of unpredictable judgment bias due to automatic adjustment.

[0111] The interactive function of the visualization early warning system has shown its value in the case of curtain wall engineering. The project director found that the cost overrun anomaly was related in time to the curtain wall design change made last month by rotating the three-dimensional heat map. After clicking on the abnormal data point, the system displayed a detailed diagnostic report, pointing out that the failure to revalue in a timely manner after the design change was the root cause. The blockchain storage verification function quickly confirmed the authenticity and integrity of the relevant documents. These pieces of information helped the team reach a supplementary agreement with the design and construction units within 48 hours, controlling the cost overrun impact within a reasonable range.

[0112] The construction of the violation mode knowledge base adopts a multi-source data fusion approach. In addition to the abnormal events automatically analyzed by the system, it also integrates structured data such as typical violation cases published by industry regulators and historical project audit reports of enterprises. The knowledge base maintains the complete life cycle information of violation modes, including the time of first discovery, the latest case, and the effectiveness statistics of treatment plans. When a new project has an abnormal signal, the system not only matches the technical characteristics but also refers to the treatment experience of similar events to generate more operational improvement suggestions.

[0113] The entire implementation process embodies the idea of closed-loop management. After the curtain wall engineering early warning event is handled, the system automatically generates a case summary, updates the judgment logic for material price difference violations, and adjusts the monitoring frequency of related processes. Three months later, the project review shows that similar design changes that cause cost anomalies are detected earlier, and the processing period is shortened to an average of 15 working days. This closed-loop mechanism from anomaly detection to system improvement enables the process monitoring and management system to have continuous evolution capabilities, gradually forming a risk prevention and control system that adapts to the specific project management characteristics of the enterprise.

[0114] Example 5: Dynamic adjustment mechanism for engineering cost consulting project monitoring frequency, precise supervision is achieved by establishing an intelligent association between violation modes and monitoring strategies. At the initial stage of system operation, a basic monitoring rule library is established, including the standard sampling frequency of each process under different project types. These frequency settings are based on industry standard requirements and combined with enterprise historical project data statistical analysis to determine the initial value. When a new project starts, the system automatically matches the initial monitoring plan based on project attributes, such as setting daily data collection for the main structure stage of civil engineering, and adjusting to every three days for the decoration stage.

[0115] The construction of the high-frequency violation pattern set adopts an incremental learning method. The system continuously analyzes historical abnormal diagnosis reports to identify repeatedly occurring violation feature combinations. Each pattern records detailed context information, including the project phase where the violation frequently occurs, the associated stakeholder roles, and typical triggering factors. The pattern matching engine not only compares static feature values but also analyzes trajectory patterns of feature changes when processing new process data. For example, when it finds that the progress reports of a certain subcontracting project have presented similar delay feature curves for three consecutive times, the system will perform a similarity evaluation with the "insufficient subcontractor resources" pattern in history.

[0116] The maintenance of the association rule table adopts a confidence-based updating strategy. Each rule records the adjustment relationship between a specific violation pattern and subsequent process monitoring parameters, including three elements: precondition, operation instruction, and intensity coefficient. The rule precondition describes the identification features of the violation pattern, the operation instruction specifies the monitoring parameters that need to be adjusted, and the intensity coefficient quantifies the adjustment amplitude. The system regularly evaluates the effectiveness of the rules, automatically reduces the priority of rules that have not been triggered for a long time, and improves the confidence level of rules that have been frequently verified to be effective in recent times.

[0117] The triggering process of enhanced monitoring instructions implements a multi-level verification mechanism. When a new process feature matches a certain violation pattern, the system first checks the historical triggering records of the pattern in the current project to evaluate its persistence. Second, it analyzes the significance of the matching features to exclude false matches caused by accidental fluctuations. Finally, it judges whether an immediate response is needed in combination with the overall state of the current project phase. After these three verifications, the system generates specific enhanced monitoring instructions, avoiding the waste of monitoring resources due to excessive sensitivity.

[0118] The adjustment of monitoring sampling frequency adopts a smooth transition strategy. The system does not make sudden changes to the monitoring frequency, but determines the adjustment step and transition period according to the intensity index of the instruction. The intensity index is calculated based on the typicality of the matched pattern, historical accuracy, and current impact. For example, for a highly typical and severely affected pattern, the sampling frequency may be increased by 20% per day, completing the transition from the base frequency to the target frequency within five days. This gradual adjustment avoids sudden changes in the load of the data collection system, maintaining the stability of the monitoring process.

[0119] The monitoring strategy optimization of subsequent processes considers the influence propagation path. The system establishes influence transmission models for different types of violations by analyzing historical project data. When an abnormality is confirmed in a certain process, the model predicts the range and degree of the affected subsequent processes. These predictions are used to guide the targeted strengthening of monitoring resources, prioritizing affected processes on the critical path. The transmission model is updated regularly to improve prediction accuracy by incorporating new project data, forming a dynamic optimization cycle that becomes more refined over time.

[0120] Real-time deviation and violation pattern matching implement multi-dimensional similarity calculation. The system not only compares the current deviation value with the typical value in the pattern library, but also analyzes the time series characteristics of the deviation development process. The matching algorithm focuses on the dynamic characteristics of the deviation occurrence rhythm, duration and fluctuation form, which often reflect the nature of the problem better than static values. The matching result outputs similarity score and confidence interval for the decision engine to judge whether the monitoring strategy needs to be adjusted.

[0121] Abnormal propagation path prediction uses a modeling method based on graph neural networks. The system converts the project process network into a weighted directed graph, with nodes representing processes and edges representing dependency relationships. The edge weight reflects the influence strength. When a node is marked as abnormal, the graph neural network calculates the influence propagation probability along the edge direction to identify a high-risk set of successor nodes. This prediction method can capture the complex chain reaction in the project process, which is closer to the actual situation than simple rule inference. The prediction result is visualized as a colored path in the process network, intuitively showing the potential risk diffusion direction.

[0122] Dynamic threshold adjustment mechanism implements feedback control principle. The system regards the monitoring process as a controlled object, dynamically adjusts the monitoring sensitivity parameters by continuously measuring the difference between the actual deviation and the expected value. The adjustment algorithm considers the recent monitoring effect. If new abnormal points are found after strengthening monitoring, the sensitivity is maintained or further improved; otherwise, if no abnormalities are found for consecutive periods, the threshold is gradually relaxed. This adaptive mechanism optimizes resource use efficiency while ensuring monitoring effectiveness, avoiding unnecessary monitoring overhead.

[0123] Integrated analysis of project environmental factors enhances the rationality of adjustment decisions. The system monitors external environmental parameters, including material price index, weather warning, policy changes and other information, which may affect project execution risks. When the environment changes significantly, the system automatically assesses its impact on various processes and pre-adjusts related monitoring parameters. For example, when detecting that the monthly price of a major building material has risen beyond the historical fluctuation range, it will strengthen the cost control point monitoring related to this material in advance, without waiting for the actual deviation to occur.

[0124] The entire implementation process emphasizes the explainability of system behavior. Each monitoring frequency adjustment generates a detailed decision log, recording the trigger reason, matching pattern, adjustment basis and expected effect. Management personnel can query the monitoring history of a specific process at any time to understand the causes and effects of frequency changes. The system also provides simulation analysis functions to allow testing of the potential effects of different adjustment strategies, assisting human decision-making. This transparent design enhances the trust of management personnel in the automatic adjustment mechanism and promotes a human-machine collaborative working mode.

[0125] The long-term optimization of the monitoring strategy adopts a reinforcement learning framework. The system regards each monitoring action and its result as a state-action-reward sequence, and forms a policy optimization experience through long-term accumulation. A good strategy can timely discover major risks with minimal monitoring cost, and this balance is quantitatively expressed by the reward function. The learning process continuously tries new strategy combinations, gradually forming an optimal monitoring scheme that adapts to the specific management style of the enterprise and the characteristics of the project. The strategy library is regularly synchronized with the latest project data to ensure the timeliness and relevance of the recommendations.

[0126] Particular attention is paid to avoiding the formation of monitoring blind spots during implementation. The system regularly checks the monitoring coverage of each process to ensure that there are no long-term low monitoring links. For emerging processes or technical processes, the system automatically assigns a higher basic monitoring frequency until sufficient data is accumulated to determine its risk characteristics. This conservative strategy prevents regulatory loopholes due to lack of experience and embodies the cautious principle of risk prevention and control.

[0127] The collaborative integration with other systems expands the application value of monitoring adjustment. The interface with the progress management system obtains real-time key path change information, the interface with the cost system captures abnormal fund flow signals, and the data sharing with the quality system discovers potential associated risks. These cross-system information fusion provides more comprehensive decision-making basis, making the monitoring frequency adjustment not only based on the deviation of a single dimension, but also considering the three-dimensional state of project execution. The integrated architecture adopts a loosely coupled design, with each system running independently and exchanging necessary data through standard interfaces.

[0128] The final dynamic monitoring system has environmental adaptability. Different regional projects automatically load local regulatory requirements, special types of engineering match exclusive monitoring solutions, and emergency monitoring plans are triggered in emergency situations. The system continuously learns and optimizes from actual operation, gradually forming an intelligent supervision rhythm that matches the organization's management capabilities. This flexibility and adaptability make the process supervision of engineering cost consulting projects not only rigorous and standardized, but also efficient and practical, controlling risks while supporting the smooth implementation of the project.

[0129] It should be noted that, in this document, the terms such as first and second are used merely to distinguish one entity or action from another, and do not necessarily require or imply any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to such a process, method, article, or apparatus.

[0130] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A process execution supervision method suitable for engineering cost consulting projects, characterized by, Comprise: Obtain the whole cycle data set of the project cost consulting project through the distributed data acquisition interface, which at least contains the contract file, the design change notice, the bill of quantities and the progress payment certificate; Structural feature analysis is performed on the data set to generate a multi-dimensional project feature matrix containing time nodes, cost indicators and resource allocation; Call the pre-built project cost benchmark parameter library, calculate the real-time deviation of the multi-dimensional project feature matrix and the benchmark parameters based on the dynamic time warping algorithm, including: Load the standard duration curve, cost control curve and resource allocation curve in the benchmark parameter library; Use the dynamic time warping distance algorithm to calculate the shape difference between the actual process feature sequence and each benchmark curve; The multiple shape difference degrees are integrated into a real-time deviation by a weighted fusion algorithm; When the real-time deviation is detected to exceed the preset threshold range, the process abnormality recognition engine is automatically triggered to extract the process feature vector including the duration delay rate, the cost overrun rate and the material loss rate from the deviated process; Input the process feature vector into the pre-trained compliance verification model to output an abnormal diagnosis report containing the violation type, risk level and responsible subject.

2. The process execution monitoring method for engineering cost consulting projects according to claim 1, characterized in that, Also include: Generate a visual warning signal according to the abnormal diagnosis report, and store the warning signal in association with the original process data in the blockchain evidence module; Based on the historical abnormal diagnosis report set, the spectral clustering algorithm is used to classify and analyze the abnormal patterns to identify a high-frequency violation mode set; Use the high-frequency violation mode set to optimize the decision boundary of the compliance verification model to generate a model parameter update instruction; According to the matching result of the real-time deviation and the high-frequency violation mode set, dynamically adjust the monitoring frequency threshold of the subsequent process; Integrate the monitoring results of all processes to generate a process optimization scheme containing corrective measures, responsibility tracing path and prevention strategy.

3. The process execution monitoring method for engineering cost consulting projects according to claim 1, characterized in that, The structural feature analysis of the data set includes: Establish an engineering cost entity relationship model based on a knowledge graph to extract entity attributes from unstructured text, including contract subjects, pricing terms and change bases; Use the multi-head attention mechanism to perform time alignment processing on the extracted entity attributes to form a timestamped project feature sequence; Generate a multi-dimensional project feature matrix representing process correlation through feature cross fusion algorithm.

4. The process execution monitoring method for engineering cost consulting projects according to claim 1, characterized in that, The process abnormality recognition engine includes: Construct an abnormal feature space containing duration compression coefficient, cost fluctuation coefficient and resource conflict coefficient; Use the isolation forest algorithm to identify deviated process clusters in the abnormal feature space; Extract the feature combination of the maximum deviation dimension from the deviated process cluster to form the process feature vector.

5. The process execution monitoring method for engineering cost consulting projects according to claim 4, characterized in that, The compliance verification model includes: Map the process feature vector to the pre-defined violation mode feature space; Calculate the matching probability of the vector and each type of violation mode through a multi-layer perception network; When the highest matching probability exceeds the confidence threshold, output the corresponding violation type code and risk level label.

6. The process execution monitoring method for engineering cost consulting projects according to claim 2, wherein, The generation of visual warning signal includes: Activate the corresponding level of warning color coding according to the risk level label; Generate a three-dimensional heat map containing deviated process number, abnormal time interval and violation details; A heat map is established verifiable data link with original credentials in the blockchain storage module.

7. The process execution monitoring method for engineering cost consulting projects according to claim 2, characterized in that, The abnormal pattern classification analysis by the spectral clustering algorithm includes: Extracting the violation type code chain and time sequence characteristics in the historical abnormal diagnosis report; Building a violation pattern propagation relationship graph through a graph convolution network; Identifying high-frequency violation pattern subgraphs with stable topological structures by using a spectral clustering algorithm.

8. The process execution monitoring method for engineering cost consulting projects according to claim 7, characterized in that, The decision boundary optimization of the compliance verification model using the high-frequency violation pattern set includes: Convert the high-frequency violation pattern subgraph into a decision hyperplane in the feature space; Calculate the geometric distance between the existing decision boundary of the compliance verification model and the hyperplane; Generate neuron weight adjustment instructions when the geometric distance exceeds the tolerance threshold; The dynamic adjustment of the monitoring frequency threshold of the subsequent process includes: Establishing a correlation rule table between the high-frequency violation pattern set and the real-time deviation degree; When the new process feature matches the high-frequency violation pattern, activate the enhanced monitoring instruction in the correlation rule; According to the instruction intensity index, amplify the monitoring sampling frequency of the subsequent process by equal ratio.

9. A process execution supervision system suitable for engineering cost consulting projects, for implementing a process execution supervision method suitable for engineering cost consulting projects according to any one of claims 1 to 8, characterized in that, It includes: A distributed data acquisition module is used to acquire the whole cycle data set of the engineering cost consulting project through a distributed data acquisition interface, and the data set at least contains contract documents, design change notice, bill of quantities and progress payment vouchers; A structured feature analysis module is connected to the distributed data acquisition module and is used to analyze the structured features of the data set to generate a multi-dimensional project feature matrix containing time nodes, cost indicators and resource allocation amounts; A deviation calculation module is connected to the structured feature analysis module and has a pre-built engineering cost benchmark parameter library inside, which is used to call the benchmark parameter library and calculate the real-time deviation of the multi-dimensional project feature matrix and the benchmark parameters based on the dynamic time warping algorithm; A process anomaly identification module is connected to the deviation calculation module and is used to automatically trigger the process anomaly identification engine when it detects that the real-time deviation exceeds the preset threshold range, and extract the process feature vector including the time delay rate, cost overrun rate and material loss rate from the deviated process; A compliance verification and report generation module is connected to the process anomaly identification module and has a pre-trained compliance verification model inside, which is used to input the process feature vector into the compliance verification model and output an abnormal diagnosis report containing violation types, risk levels and responsible subjects.

Images