Multi-dimensional intelligent management method and system for whole-process cost

Abstract

The invention discloses a multi-dimensional intelligent management method and system for whole-process cost, and the method comprises the steps: generating a time-space associated structured cost data cube according to heterogeneous cost data of the stages of project planning, design, construction and completion; outputting a dynamic cost prediction curve and deviation sensitive nodes based on the structured cost data cube; according to the dynamic cost prediction curve, performing multi-party task allocation optimization by using a block chain enabled BIM / CIM collaboration platform, and generating a collaboration instruction set of smart contract coding; outputting a risk probability matrix and an advanced early warning signal based on the collaborative instruction set and the real-time engineering data flow; and according to the risk probability matrix, adopting a multi-objective optimization algorithm to generate an anti-interference decision scheme set, and outputting an optimal cost control strategy after digital twinborn simulation verification. By using the embodiment of the invention, the cost data of each stage of the project can be efficiently integrated, dynamic cost prediction is realized, and collaborative decision and effective risk management are optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of cost management, and particularly relates to a multi-dimensional intelligent management method and system for the whole-process cost. Background Art

[0002] In the field of construction project management, cost control and cost management have always been the key factors affecting the success of a project. With the increasing complexity and scale of engineering projects, the whole-process cost management is facing unprecedented challenges. The cost data involved in each stage of an engineering project - from planning, design to construction and completion - are often heterogeneous and scattered. How to efficiently integrate and manage these data to achieve scientific cost prediction and control has become an urgent problem to be solved in the development of the industry.

[0003] Traditional cost management methods mainly rely on manual data collection and analysis, which are prone to problems such as information lag and inaccurate cost prediction. At the same time, the lack of cross-stage collaborative management makes the phenomenon of data islands in different stages increasingly serious, and the information interaction between different participants (such as owners, designers, constructors, etc.) is not smooth, resulting in untimely and inaccurate decision-making. In addition, due to the influence of factors such as market price fluctuations and construction progress changes, traditional cost management often cannot achieve dynamic adjustment, making it easy for projects to exceed the budget during the implementation process, causing huge economic losses. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-dimensional intelligent management method and system for the whole-process cost to solve the deficiencies in the prior art, which can efficiently integrate the cost data of each stage of the project, realize dynamic cost prediction, optimize collaborative decision-making, and effective risk management.

[0005] An embodiment of the present application provides a multi-dimensional intelligent management method for the whole-process cost, and the method includes:[[]] According to the heterogeneous cost data in each stage of project planning, design, construction, and completion, perform cross-stage feature alignment through a data fusion engine driven by federated learning to generate a spatio-temporal associated structured cost data cube, wherein the data fusion engine uses a differential privacy protection mechanism to eliminate data islands; Based on the structured cost data cube, combine the market price fluctuation trend and the construction progress change factor to construct a spatio-temporal graph neural network prediction model, and output a dynamic cost prediction curve and deviation-sensitive nodes, wherein the spatio-temporal graph neural network prediction model captures the associated influence of potential cost overrun risk factors through an attention mechanism; According to the dynamic cost prediction curve, a multi-party task allocation optimization is performed using a blockchain-enabled BIM / CIM collaborative platform to generate a collaborative instruction set encoded with a smart contract. Among them, the BIM / CIM collaborative platform realizes privacy-protected data synchronization among the owner, the designer, and the constructor through zero-knowledge proof; Based on the collaborative instruction set and the real-time project data stream, a potential risk path is parsed through a dynamic knowledge graph engine, and a risk probability matrix and an early warning signal are output. Among them, the dynamic knowledge graph engine integrates an industry knowledge base and a reinforcement learning strategy to achieve risk evolution simulation; According to the risk probability matrix, a multi-objective optimization algorithm is used to generate a set of anti-interference decision-making solutions, and an optimal cost control strategy is output after verification through digital twin simulation. Among them, the optimal cost control strategy synchronously drives the three-dimensional risk heat map and the decision path deduction animation in the visualization interface.

[0006] Optionally, according to the heterogeneous cost data in the engineering planning, design, construction, and completion stages, a cross-stage feature alignment is performed through a data fusion engine driven by federated learning to generate a spatio-temporal associated structured cost data cube. Among them, the data fusion engine uses a differential privacy protection mechanism to eliminate data islands, including: According to the BIM model parameters in the engineering planning stage, the bill of quantities in the design stage, the progress log in the construction stage, and the settlement documents in the completion stage, a field semantic mapping is performed through a heterogeneous data adapter to generate a four-dimensional tensor containing time stamps, spatial coordinates, and cost attributes, and a standardized feature vector set is output; Based on the feature vector set, a cross-stage alignment algorithm under the federated learning framework is used to optimize the local model parameters of each participant through projected gradient descent, and a distributed feature space alignment is completed under the constraint of differential privacy to generate a cross-stage joint feature matrix; The joint feature matrix is input into a spatio-temporal association engine, and the coupling relationship between project progress - cost consumption - geographical distribution is captured through a three-dimensional convolution kernel to construct a six-dimensional cost data cube containing time series association, spatial topology association, and cost conduction association, and a structured data cube with a privacy protection label is output.

[0007] Optionally, based on the structured cost data cube, combined with the market price fluctuation trend and the construction progress change factor, a spatio-temporal graph neural network prediction model is constructed to output a dynamic cost prediction curve and deviation-sensitive nodes. Among them, the spatio-temporal graph neural network prediction model captures the associated influence of potential cost overrun risk factors through an attention mechanism, including: According to the time series cost data in the structured data cube, cost fluctuation features are extracted through a multi-scale sliding window, and a time-dimensional dynamic embedding vector is constructed in combination with the external market price index; Based on the construction progress change factor, model the spatial dependence relationship between construction nodes through the graph attention mechanism to generate a spatial topology weight matrix; Input the time dynamic embedding vector and the spatial topology weight matrix into the spatio-temporal graph convolutional layer, fuse the spatio-temporal features through a bidirectional gated recurrent unit, and output the correlation influence intensity distribution map of potential overspending risk factors; According to the risk influence intensity distribution map, adopt a dynamic threshold segmentation algorithm to identify deviation-sensitive nodes and generate a dynamic cost prediction curve with confidence annotations.

[0008] Optionally, according to the dynamic cost prediction curve, use the blockchain-enabled BIM / CIM collaborative platform to optimize multi-party task allocation and generate a collaborative instruction set encoded by a smart contract. Among them, the BIM / CIM collaborative platform realizes privacy-protected data synchronization among the owner, the designer, and the constructor through zero-knowledge proof, including: According to the dynamic cost prediction curve, homomorphically encrypt privacy data such as the owner's budget constraint, the designer's drawing changes, and the constructor's resource scheduling through the zero-knowledge proof protocol to generate a verifiable ciphertext data set; Based on the ciphertext data set, decompose the collaborative tasks using multi-party secure computing, solve the optimal task allocation scheme under the balance of multi-party interests through the Hungarian algorithm, and output a task allocation matrix with weight constraints; Encode the task allocation matrix into a smart contract, define the contract trigger conditions through an event-driven state machine, and generate a smart contract template containing an execution logic chain; Deploy lightweight blockchain nodes on the BIM / CIM collaborative platform, achieve privacy-protected data synchronization across participating parties through the sharding consensus mechanism, and generate a collaborative instruction hash chain with timestamps; According to the real-time engineering data stream, verify the authenticity of off-chain data through an oracle, dynamically update the execution status of the smart contract, and output a traceable collaborative instruction set.

[0009] Optionally, based on the collaborative instruction set and the real-time engineering data stream, parse potential risk paths through a dynamic knowledge graph engine and output a risk probability matrix and an early warning signal. Among them, the dynamic knowledge graph engine fuses the industry knowledge base and the reinforcement learning strategy to realize risk evolution simulation, including: According to the collaborative instruction set, extract risk semantic fragments from engineering contract terms and construction specification documents through a domain adaptation pre-training model to generate risk knowledge triples with probability annotations; Based on the risk knowledge triples, combined with abnormal events in the real-time engineering data stream, construct a risk propagation path topology graph through a temporal graph convolutional network; Input the risk propagation path into the reinforcement learning environment, simulate the risk evolution process through the double deep Q-network, and generate a risk probability transition matrix containing the intensity of the cascade effect; Based on the risk probability transition matrix, use the backpropagation importance sampling algorithm to identify the key risk paths and output early warning signals with time window constraints; Compress the risk model parameters through knowledge distillation technology, deploy them to the edge computing node to achieve real-time risk monitoring, and generate a lightweight risk evolution simulator.

[0010] Optionally, according to the risk probability matrix, use a multi-objective optimization algorithm to generate a set of anti-interference decision-making solutions, and output the optimal cost control strategy after verification through digital twin simulation. Among them, the optimal cost control strategy synchronously drives the three-dimensional risk heat map and the decision path deduction animation in the visualization interface, including: According to the risk probability matrix, synchronously optimize the three objectives of the cost control rate, the engineering quality compliance rate, and the risk aversion coefficient through the NSGA-III algorithm, and generate an initial Pareto front containing 300-500 groups of solutions; Input the Pareto solution set into the digital twin engine, simulate the construction process under different decision-making solutions through the physical engine, and generate a multi-dimensional simulation data set containing cost deviation, quality defects, and risk outbreak points; Based on the simulation data set, use the fuzzy comprehensive evaluation method to calculate the fitness scores of each solution, and screen the optimal cost control strategy through the gradient boosting decision tree; Fuse the optimal strategy with the BIM model, render the three-dimensional risk heat map through ray tracing technology, and generate a decision path deduction animation, and output a visualization decision support interface.

[0011] Optionally, based on the construction progress change factor, model the spatial dependence relationship between construction nodes through the graph attention mechanism, and generate a spatial topology weight matrix, including: According to the node delay data in the construction progress change factor, extract the spatio-temporal correlation features between construction nodes through the spatio-temporal graph convolutional layer, and generate an initial dependence graph with timestamps; Based on the initial dependence graph, combined with the geographical coordinate data in the BIM model, use the multi-modal attention mechanism to calculate the spatial correlation weights between nodes, where the three-channel features of the physical distance between nodes, the construction process dependence degree, and the resource flow direction are fused, and a multi-modal attention score matrix is output; Input the multi-modal attention score matrix into the dynamic residual connection network, and fuse the implicit dependence relationship in the historical construction log through the gated recurrent unit to generate a spatio-temporal enhanced node dependence intensity distribution map; Based on the above-mentioned dependence strength distribution map, a graph pruning algorithm with differential privacy protection is adopted to eliminate weak dependence edges with confidence lower than the threshold, and a spatial topology weight matrix with a hierarchical structure is generated through spectral clustering, and a spatial topology of construction nodes with security protection labels is output.

[0012] Optionally, based on the ciphertext dataset, a multi-party secure computing decomposition collaborative task is adopted, and the optimal task allocation scheme under the balance of multi-party interests is solved through the Hungarian algorithm, and a task allocation matrix with weight constraints is output, including: According to the ciphertext dataset of homomorphic encryption, the collaborative task is decomposed through the secure multi-party computing protocol to generate a task requirement vector with privacy protection, where the secret sharing technology is used to split the task parameters into multiple shadow shares; Based on the task requirement vector, a multi-party interest game model is constructed, and the initial task allocation scheme is calculated through the Nash equilibrium solution algorithm, where the model constraint conditions include the upper limit of the owner's budget, the resource capacity of the construction party, and the change tolerance of the design party; The initial allocation scheme is input into the Hungarian algorithm, and a dynamic weight adjustment strategy is introduced to optimize the conflict task matching, where the dynamic weight is adaptively updated according to the real-time supply chain risk index, and an intermediate task allocation matrix with elastic constraints is output; Verify the fairness of the intermediate task allocation matrix through zero-knowledge proof, and use the zk-SNARK protocol to generate a verifiable allocation proof, and finally output a non-tamperable task allocation matrix with weight constraints that meets the balance of multi-party interests.

[0013] Specifically, based on the simulation dataset, the fuzzy comprehensive evaluation method is used to calculate the fitness scores of each scheme, and the optimal cost control strategy is screened through the gradient boosting decision tree, including: According to the multi-dimensional engineering parameters in the simulation dataset, through a dynamic weight generator based on Monte Carlo sampling, adaptive weight allocation is performed on three objectives of the cost deviation rate, the quality defect index, and the risk outbreak probability, and a dynamic evaluation weight vector with a confidence interval is generated; Based on the dynamic evaluation weight vector, the fuzzy comprehensive evaluation algorithm is used to calculate the fitness scores of the schemes. The fuzzy comprehensive evaluation algorithm introduces a time decay factor to correct the contribution degree of historical simulation data, and maps the high-dimensional non-linear pattern through the radial basis function kernel, and outputs a fitness score matrix with timeliness labels; According to the fitness score matrix, a multi-modal feature fusion channel is constructed, and the hyperplane curvature feature, the decision variable correlation feature, and the simulation process stability feature of the Pareto solution set are extracted, and a unified dimension enhanced feature vector set is generated through the heterogeneous feature embedding network; Input the enhanced feature vector set into the gradient boosting decision tree model, adopt the multi-objective splitting criterion to optimize the growth direction of the tree structure, strengthen the search ability of the high-dimensional non-convex solution space through the cost-sensitive learning strategy, and output the global ranking result of the optimal cost control strategy; According to the global ranking result, combined with the physical constraint verification of the digital twin engine, iteratively optimize the strategy parameters through the backpropagation correction mechanism, and generate the execution sequence of the optimal cost control strategy that meets the construction feasibility boundary.

[0014] Another embodiment of the present application provides a multi-dimensional intelligent management system for the whole-process cost, and the system includes: An alignment module, configured to perform cross-stage feature alignment on heterogeneous cost data in each stage of project planning, design, construction, and completion through a data fusion engine driven by federated learning, and generate a spatio-temporally correlated structured cost data cube, wherein the data fusion engine uses a differential privacy protection mechanism to eliminate data islands; A construction module, configured to construct a spatio-temporal graph neural network prediction model based on the structured cost data cube, combined with the market price fluctuation trend and the construction progress change factor, and output a dynamic cost prediction curve and deviation-sensitive nodes, wherein the spatio-temporal graph neural network prediction model captures the associated impact of potential cost overrun risk factors through an attention mechanism; An optimization module, configured to perform multi-party task allocation optimization on the dynamic cost prediction curve by using a BIM / CIM collaborative platform empowered by blockchain, and generate a collaborative instruction set encoded with smart contracts, wherein the BIM / CIM collaborative platform realizes privacy-protected data synchronization among the owner, the designer, and the constructor through zero-knowledge proof; An analysis module, configured to analyze potential risk paths based on the collaborative instruction set and the real-time project data stream through a dynamic knowledge graph engine, and output a risk probability matrix and an early warning signal, wherein the dynamic knowledge graph engine fuses the industry knowledge base and the reinforcement learning strategy to realize risk evolution simulation; An output module, configured to generate an anti-interference decision solution set by using a multi-objective optimization algorithm according to the risk probability matrix, and output the optimal cost control strategy after verification by digital twin simulation, wherein the optimal cost control strategy synchronously drives the three-dimensional risk heat map and the decision path deduction animation in the visualization interface.

[0015] Compared with the prior art, a multi-dimensional intelligent management method for the whole-process cost provided by the present invention generates a spatio-temporal associated structured cost data cube according to the heterogeneous cost data in the engineering planning, design, construction, and completion stages; based on the structured cost data cube, outputs a dynamic cost prediction curve and deviation-sensitive nodes; according to the dynamic cost prediction curve, uses a blockchain-enabled BIM / CIM collaborative platform to optimize multi-party task allocation, and generates a collaborative instruction set encoded with a smart contract; based on the collaborative instruction set and real-time engineering data stream, outputs a risk probability matrix and an early warning signal; according to the risk probability matrix, uses a multi-objective optimization algorithm to generate an anti-interference decision-making solution set, and outputs an optimal cost control strategy after verification by digital twin simulation, so as to be able to efficiently integrate the cost data of each stage of the project, realize dynamic cost prediction, optimize collaborative decision-making, and effective risk management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a hardware structure block diagram of a computer terminal for a multi-dimensional intelligent management method for the whole-process cost provided by an embodiment of the present invention; Figure 2 It is a flowchart of a multi-dimensional intelligent management method for the whole-process cost provided by an embodiment of the present invention; Figure 3 It is a structural diagram of a multi-dimensional intelligent management system for the whole-process cost provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] An embodiment of the present invention first provides a multi-dimensional intelligent management method for the whole-process cost, which can be applied to electronic devices, such as computer terminals, specifically, ordinary computers, etc.

[0019] The following takes running on a computer terminal as an example to describe it in detail. Figure 1 It is a hardware structure block diagram of a computer terminal for a multi-dimensional intelligent management method for the whole-process cost provided by an embodiment of the present invention. As Figure 1 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory may include a non-volatile storage medium and an internal memory.

[0020] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any multi-dimensional intelligent management method for the whole-process cost.

[0021] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0022] The internal memory provides an environment for the operation of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any multi-dimensional intelligent management method for the whole-process cost.

[0023] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 1 the structure shown in

[0024] It should be understood that the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0025] See Figure 2 , embodiments of the present invention provide a multi-dimensional intelligent management method for the whole-process cost, which may include the following steps: S201, according to the heterogeneous cost data in the engineering planning, design, construction, and completion stages, perform cross-stage feature alignment through a data fusion engine driven by federated learning to generate a spatio-temporal associated structured cost data cube, wherein the data fusion engine uses a differential privacy protection mechanism to eliminate data islands; Specifically, according to the BIM model parameters in the engineering planning stage, the bill of quantities in the design stage, the progress log in the construction stage, and the settlement documents in the completion stage, perform field semantic mapping through a heterogeneous data adapter to generate a four-dimensional tensor including time stamps, spatial coordinates, and cost attributes, and output a standardized feature vector set; The heterogeneous data adapter adopts a modular design and configures dedicated parsers for data features in different stages: BIM Model Parameter Analysis: Extract component attributes (such as concrete strength grade C30, steel bar specification HRB400) from IFC (Industry Foundation Classes) files, associate geometric data with material properties through an IFC-to-RDF converter, and generate semantic triples with spatial coordinates (x, y, z). For example, a beam component is parsed as: (Component ID - 2038, Type: Frame Beam, Coordinates (12.5, 8.2, 3.0), Concrete volume 2.8 m³).

[0026] Bill of Quantities Processing: Use NLP techniques (such as the BERT-CRF model) to identify key fields (project code, unit, quantity) in the bill text and match them with BIM component IDs. For example: (Project Code 010502003, Rectangular Column, m³, 15.6) → Map to BIM Component ID - 2041.

[0027] Progress Log Integration: Parse the semi-structured text of the construction log (such as "2023-05-20, 80% of the 3rd floor beam-slab pouring completed"), and extract the timestamp (20230520), progress value (80%), and associated component ID through regular expressions.

[0028] Settlement Document Standardization: Use OCR (Optical Character Recognition, accuracy ≥ 98%) to identify the amount and visa number in the scanned document and align them with the progress log timestamp.

[0029] Four-Dimensional Tensor Construction: Dimension 1 (Time): Unix timestamp (such as 1684540800); Dimension 2 (Space): Three-dimensional coordinates (x, y, z), with an accuracy of 0.01 meters; Dimension 3 (Cost): Sub-item costs (such as material costs, labor costs, etc.), in yuan; Dimension 4 (Attribute): Classification code (such as 010101 - Earthwork).

[0030] The final output standardized feature vector set contains approximately 500,000 records, each record being a 128-dimensional vector (8 dimensions for time + 24 dimensions for space + 32 dimensions for cost + 64 dimensions for attributes), and is stored in Apache Parquet format for efficient querying.

[0031] Based on the feature vector set, adopt the cross-stage alignment algorithm under the federated learning framework, optimize the local model parameters of each participant through projected gradient descent, and complete the distributed feature space alignment under differential privacy constraints to generate a cross-stage joint feature matrix; The federated learning system includes three types of participant nodes: Owner Node: Holds BIM data in the planning stage; Design Institute Node: Store the list in the design phase; Construction Unit Node: Maintain construction / completion data.

[0032] Cross-phase Alignment Process: Local Feature Extraction: Each node uses the ResNet-50 model (removing the last fully connected layer) to extract local features and outputs a 512-dimensional feature vector. For example, the Design Institute Node maps the list entry "010101001 Earth excavation 100m³" to the vector [0.12, -0.05, ..., 0.33].

[0033] Differential Privacy Protection: Add Laplace noise (ε = 0.5, sensitivity Δf = 1.0) before feature upload to ensure that individual data points are indistinguishable. The formula for the amount of noise is: noise = Laplace(0, Δf / ε), and the actual range of added noise values is [-0.2, 0.2].

[0034] Projected Gradient Descent: Initialize the global projection matrix W ∈ R^(512×256); In each iteration, each node calculates the local gradient ∇W; After gradient aggregation, update W ← W - η·∇W (learning rate η = 0.01).

[0035] Feature Space Alignment: After 100 iterations, the cosine similarity of the features of each node in the 256-dimensional shared space reaches above 0.85 (initial value 0.3).

[0036] Joint Feature Matrix Generation: Concatenate the aligned features by timestamp to form a 2000×256 matrix (2000 key time points, each time point with 256-dimensional features), and reduce the dimension to 64 through principal component analysis (PCA) to improve computational efficiency.

[0037] Input the joint feature matrix into the spatio-temporal correlation engine, capture the coupling relationship of project progress - cost consumption - geographical distribution through a three-dimensional convolution kernel, construct a six-dimensional cost data cube including time series correlation, spatial topology correlation, and cost conduction correlation, and output a structured data cube with privacy protection labels.

[0038] The spatio-temporal correlation engine adopts a three-layer processing architecture: Time Series Correlation Layer: Use a 1D convolution kernel (width 7, stride 1) to scan the time axis and extract patterns such as weekly and monthly cycles. For example, it is detected that the cost peak appears on the 25th of each month (related to the progress payment).

[0039] Output the time feature map (dimension: 50×64, 50 time segments).

[0040] Spatial topological association layer: A 3D convolution kernel (3×3×3) is used to process BIM coordinate data and identify spatial dependencies. For example, it is found that the overrun of concrete in the underground 2-story column will affect the amount of steel bars in the above-ground 1-story beam.

[0041] Output spatial feature map (dimension: 30×30×30×32, corresponding to 30m×30m×30m spatial grid).

[0042] Cost transmission association layer: The cost flow between sub-projects is modeled through the Graph Attention Network (GAT), and the edge weight reflects the transfer ratio of materials / labor. For example, the weight of steel bar processing (010516) to concrete pouring (010502) is 0.7.

[0043] Output graph embedding matrix (dimension: 100×16, 100 sub-projects).

[0044] Six-dimensional data cube construction: Dimension 1-2: time (year-month, such as 2023-05); Dimensions 3-5: Space (x, y, z, grid accuracy 1 meter); Dimension 6: Cost transmission path (such as 010501→010502→010503).

[0045] Each cell stores the result of multi-dimensional feature concatenation (total dimension = 50+32+16=98 dimensions) and generates a privacy protection tag through HMAC-SHA256. For example: (202305, (12,8,3), 010501→010502): [0.12,...,0.45], tag=0x3a7d...

[0046] The cube supports OLAP (Online Analytical Processing) queries, such as "Analysis of structural engineering cost transmission path in the area with coordinates (12,8,3) in May 2023", with a response time of <200ms.

[0047] This step integrates heterogeneous data (such as BIM parameters, bill of quantities, progress logs, and settlement documents) in the entire engineering life cycle (planning, design, construction, completion) without sharing the original data through federated learning technology. It uses the projected gradient descent algorithm to align cross-stage features and models the coupling relationship of time, space, and cost through a three-dimensional convolutional kernel, finally generating a six-dimensional structured data cube. The differential privacy mechanism injects noise during the feature alignment process to ensure the data privacy of each participating party (such as the owner and the construction party), breaks the data silo problem in traditional cost management, and realizes the deep integration of cross-stage data; enhances multi-party collaboration trust through privacy protection technology, and provides a high-dimensional and high-precision data basis for subsequent prediction and decision-making.

[0048] S202, based on the structured cost data cube, combines the market price fluctuation trend and the construction progress change factor to construct a spatio-temporal graph neural network prediction model, and outputs a dynamic cost prediction curve and deviation-sensitive nodes. Among them, the spatio-temporal graph neural network prediction model captures the associated impact of potential cost overrun risk factors through an attention mechanism; Specifically, according to the time series cost data in the structured data cube, cost fluctuation features can be extracted through a multi-scale sliding window, and a dynamic embedding vector in the time dimension can be constructed by combining with the external market price index; The structured cost data cube contains cost time series data for each stage of the project (such as daily material costs and labor costs). The multi-scale sliding window is designed in three levels: Short-term window (7 days): Capture sudden cost fluctuations (such as a 5% single-day increase in steel prices), and use first-order differences to extract change rate features; Medium-term window (30 days): Identify periodic fluctuations (such as cash flow tension caused by end-of-month settlements), and extract the main frequency components through the fast Fourier transform (FFT); Long-term window (90 days): Analyze trend changes (such as cost increases caused by inflation), and use linear regression to fit the slope.

[0049] External market price index integration: Data source: Access real-time data streams such as the futures price of rebar (code RB) and the cement price index (CEMPI); Feature fusion: Align the time axis of the external index and the project cost data through the dynamic time warping (DTW) algorithm to eliminate the impact of market data delay. For example, when the steel market price delays for 3 days to affect project procurement, DTW automatically shifts the market data forward by 3 days for matching.

[0050] Dynamic embedding vector generation: Vector dimension: 128 dimensions, including short-term fluctuation intensity (16 dimensions), medium-term cycle phase (32 dimensions), long-term trend coefficient (16 dimensions), and external market correlation (64 dimensions); Example: The dynamic vector of a certain week may be encoded as [0.35, -0.12,..., 0.08], indicating that the short-term steel cost has increased by 35%, but the long-term trend is still lower than expected.

[0051] Based on the construction progress change factor, the spatial dependence relationship between construction nodes is modeled through the graph attention mechanism to generate a spatial topology weight matrix; Specifically, based on the construction progress change factor, the spatial dependence relationship between construction nodes is modeled through the graph attention mechanism to generate a spatial topology weight matrix. The spatio-temporal correlation features between construction nodes can be extracted through the spatio-temporal graph convolutional layer according to the node delay data in the construction progress change factor, and an initial dependence graph with timestamps is generated; The construction progress change factor includes the delay days of each node (such as "foundation pouring", "steel structure installation") (e.g., node A is delayed by 3 days) and the process dependence relationship (e.g., node B must start after node A is completed). The spatio-temporal graph convolutional layer consists of a spatial graph convolution (Spatial GCN) and a temporal gated convolution (Temporal GRU). The specific process is as follows: Spatial graph construction: Using construction nodes as graph vertices, the physical adjacency relationship (such as node spacing <50 meters) and process dependence as edges to construct an initial adjacency matrix. For example, if there is a process dependence between node A and node B, the edge weight is set to 1.0; if there is only physical proximity but no process relationship, the weight is set to 0.3.

[0052] Temporal feature extraction: Perform time window sliding (window size 7 days) on the delay data of each node (such as [3 days, 0 days, 5 days]), and extract temporal features through GRU (64 hidden units) to output a time embedding vector with a dimension of 128.

[0053] Spatio-temporal convolution fusion: Use two layers of GCN (output dimension of each layer is 64) to process the spatial relationship, take the time embedding output by GRU as the node feature input, and generate spatio-temporal fusion features through the ReLU activation function. For example, the feature vector of node A fuses its own 3-day delay and the progress influence of its adjacent node B.

[0054] Dependence graph generation: Calculate the cosine similarity between node features, establish dependence edges for node pairs with similarity >0.7, and finally generate an initial dependence graph with timestamps. Each edge in the graph is marked with the dependence strength (0~1) and the latest update time (such as 2023-10-05 14:00).

[0055] Based on the initial dependency graph, combined with the geographical coordinate data in the BIM model, a multi-modal attention mechanism is used to calculate the spatial correlation weights between nodes. Among them, the three-channel features of the physical distance between nodes, the construction process dependency degree, and the resource flow direction are fused to output a multi-modal attention score matrix. The multi-modal attention mechanism processes different modal features through three independent attention heads respectively, followed by a feature fusion layer: Physical distance attention head: Input: Euclidean distance between nodes (e.g., the distance from node A to B is 35 meters), normalized to [0, 1]; Calculation: It is mapped to Query and Key through a fully connected layer (input 1D, output 16D), and the attention score is obtained after dot product and scaling. For example, the score of node pairs with a distance < 20 meters is increased to 0.9.

[0056] Process dependency degree attention head: Input: Process dependency strength (dependency strength value from step 1); Calculation: Multi-head attention (4 heads) is adopted. Each head maps the dependency strength to a 16D vector to calculate cross-attention. For example, the attention score of a certain head for strongly dependent (0.8) node pairs reaches 0.95.

[0057] Resource flow direction attention head: Input: Resource flow direction (e.g., "concrete pump truck from node A to B" is encoded as a direction vector); Calculation: A direction encoding matrix (8D) is used to represent the flow direction. The direction features are extracted through a bidirectional GRU (32 hidden units) to calculate the attention score.

[0058] Feature fusion: The attention scores of the three heads are concatenated (3×16 = 48D), fused through a fully connected layer (output 16D), and then normalized by Softmax to generate a multi-modal attention score matrix. For example, the final attention score between node A and B is 0.88, reflecting their strong spatial correlation.

[0059] The multi-modal attention score matrix is input into a dynamic residual connection network, and the implicit dependency relationship in the historical construction log is fused through a gated recurrent unit to generate a spatio-temporal enhanced node dependency strength distribution map; The dynamic residual connection network consists of residual blocks and GRU modules: Residual block processing: Input: Multi-modal attention score matrix (N×N, N is the number of nodes); Operation: Two convolutional layers (kernel size 3×3, number of channels 64→128), and the residual skip connection retains the original features; Output: An enhanced attention feature map with 128 channels.

[0060] Historical Log Fusion: Data: Construction logs for the past 30 days (e.g., "Node C was delayed by 2 days due to material shortage"), encoded as a time series vector (dimension 64); GRU Processing: Use a bidirectional GRU (64 hidden units) to extract temporal dependency features and output a latent relationship vector with a dimension of 128.

[0061] Feature Concatenation and Gating: Concatenate the residual feature map and the GRU output to obtain a 256-dimensional joint feature; control the information flow through a gating mechanism (Sigmoid function).

[0062] Dependency Strength Distribution Map Generation: Map the gating output through a fully connected layer (256→1 dimension), convert it to a dependency strength value (0~1) through the Sigmoid function, and generate a spatio-temporal enhanced distribution map. For example, the dependency strength between Node A and B increases from 0.8 to 0.92, reflecting the influence of their frequent linkage in the historical logs.

[0063] Based on the described dependency strength distribution map, use a differentially private graph pruning algorithm to remove weak dependency edges with a confidence lower than the threshold, and generate a hierarchical spatial topology weight matrix through spectral clustering, outputting the spatial topology of construction nodes with security protection labels.

[0064] Differentially Private Pruning: Noise Injection: Add Laplace noise (noise scale ϵ = 0.1) to the dependency strength, and the strength value is adjusted to: strength′ = strength + Laplace(0, 1 / ϵ).

[0065] Threshold Filtering: Remove edges with a strength < 0.4 (e.g., the strength of Node C-D is 0.35 and is removed), and retain high-confidence dependency relationships.

[0066] Spectral Clustering Hierarchy: Construct the Laplacian matrix: L = D - W, where D is the degree matrix and W is the pruned weight matrix; Eigenvalue Decomposition: Calculate the first 3 smallest eigenvectors of L to form a low-dimensional embedding (3 dimensions); K-means Clustering: Divide the nodes into 5 clusters (e.g., "Foundation Construction Cluster" "Decoration Cluster"), and the weighted average of the dependency strength within the cluster is used as the hierarchical weight.

[0067] Topology Matrix Generation: Weight within the Hierarchy: The weight between nodes in the same cluster is increased by 20% (e.g., from 0.8→0.96); Weight between Hierarchies: The weight between nodes across clusters is decreased by 30% (e.g., from 0.6→0.42); Safety label addition: Label each edge with the privacy budget consumption (e.g., ϵ = 0.05) and the clustering membership label.

[0068] Example output: Nodes A, B, and C belong to the "structural construction cluster" with internal weights ranging from 0.92 to 0.95; Nodes D and E belong to the "equipment installation cluster" with a cross-cluster weight of 0.4 with the structural cluster; All edges are labeled with the differential privacy parameter ϵ = 0.1, meeting the ISO 27001 data security standard.

[0069] Input the time dynamic embedding vector and the spatial topology weight matrix into the spatio-temporal graph convolutional layer, fuse the spatio-temporal features through a bidirectional gated recurrent unit, and output the association influence intensity distribution map of the potential overspending risk factors; Customized design of spatio-temporal graph convolution (ST-GCN): Spatial convolution: Based on the topological weight matrix, perform neighborhood aggregation, and each node aggregates the features of its neighbors within 3 hops; Use Chebyshev polynomials to approximate the graph convolution kernel (order K = 3) to reduce the computational complexity.

[0070] Temporal convolution: Adopt dilated causal convolution along the time axis (dilation = 2) to capture long-term temporal dependencies; The convolution kernel size = 5, the stride = 1, and the output time dimension is compressed to 1 / 3 of the original length.

[0071] Improvement of the bidirectional gated recurrent unit (Bi-GRU): Forward GRU: Learn the impact of historical cost data on the current situation (e.g., the price increase of steel last month led to the current budget tension); Backward GRU: Predict the reverse constraint of future risks on the current decision (e.g., it is expected that there will be a labor shortage in three months and reserves need to be made in advance); Gating mechanism: Control information forgetting and updating through the sigmoid function (output range 0 - 1). For example, when an abnormal fluctuation is detected, the forgetting gate value drops to 0.2, forcing the retention of the historical stable pattern.

[0072] Generation of the association influence intensity distribution map: Heat map generation: Map the risk influence intensity (scalar value) of each node to the corresponding position in the BIM model; Dynamic threshold segmentation: Red area (intensity > 0.8): Immediate intervention is required (e.g., the risk value of the "main structure node" reaches 0.91 due to the price increase of steel bars); Yellow area (0.5 - 0.8): Monitoring is required (e.g., the risk value of "curtain wall installation" is 0.63); Green area (<0.5): Safe area.

[0073] According to the risk impact intensity distribution map, a dynamic threshold segmentation algorithm is used to identify deviation-sensitive nodes and generate a dynamic cost prediction curve with confidence level annotations.

[0074] Dynamic threshold segmentation algorithm: Initial threshold setting: Based on the statistics of historical engineering data, the 90th percentile (such as 0.75) of the risk intensity distribution is taken as the benchmark threshold; Adaptive adjustment: When more than 5 nodes break through the threshold for 3 consecutive days, the threshold is automatically lowered by 10% (such as reduced to 0.675); When no node breaks through the threshold for 7 consecutive days, the threshold is raised by 5%.

[0075] Sensitive node identification: Feature extraction: Calculate for each node exceeding the threshold: Risk conduction speed (change rate of risk intensity of adjacent nodes); Resource sensitivity (cost increase caused by unit resource shortage of this node); Decision tree classification: Use the C4.5 algorithm to classify nodes into: Key sensitive nodes (to be processed within 48 hours); General sensitive nodes (to be monitored weekly).

[0076] Dynamic cost prediction curve generation: Benchmark prediction: Generate the cost trajectory without intervention based on the ARIMA model (p = 2, d = 1, q = 1); Risk correction: Use the risk intensity of sensitive nodes as the correction factor and adjust the predicted value through ridge regression (λ = 0.5); Confidence level calculation: Use Bootstrap sampling to generate 100 groups of prediction sequences; Calculate the 95% confidence interval, for example, "Q3 predicted cost = 2.85 million ± 0.12 million (confidence level 92%)".

[0077] Visualization output: Curve annotation: Mark the names of sensitive nodes and corresponding suggestions at key time points (such as milestone nodes); Interactive function: Click on the sensitive node to drill down to view the details of associated risk factors.

[0078] This step combines the structured data cube with external market indices and construction change factors to construct a spatio-temporal graph neural network model. The model extracts the cost fluctuation characteristics in the time dimension through a multi-scale sliding window and analyzes the spatial dependence relationship between construction nodes using the graph attention mechanism. Finally, it outputs a dynamic cost prediction curve and key risk nodes (such as the sensitive period of material price increase and the process delay conduction path), accurately predicts the dynamic changes of project costs, and identifies "vulnerable nodes" that may lead to cost overruns; through spatio-temporal correlation analysis, it reveals hidden risks (such as the cascading impact of supply chain delays on multiple processes) to support proactive intervention.

[0079] S203, according to the dynamic cost prediction curve, use the blockchain-enabled BIM / CIM collaborative platform to optimize multi-party task allocation and generate a collaborative instruction set encoded by smart contracts, where the BIM / CIM collaborative platform realizes privacy-protected data synchronization among the owner, the designer, and the constructor through zero-knowledge proof; Specifically, according to the dynamic cost prediction curve, homomorphic encryption can be performed on privacy data such as the owner's budget constraint, the designer's drawing changes, and the constructor's resource scheduling through the zero-knowledge proof protocol to generate a verifiable ciphertext data set; The zero-knowledge proof protocol (Zero-Knowledge Proof, ZKP) is used in this stage to achieve the secure verification of multi-party privacy data. Taking zk-SNARK (succinct non-interactive zero-knowledge proof) as an example, sensitive data such as the owner's budget constraint (such as a total budget of 50 million yuan), the designer's drawing change records (such as 3 structure modifications), and the constructor's resource scheduling plan (such as a daily concrete supply of 200 tons) are first encrypted through the Paillier homomorphic encryption algorithm to generate ciphertext forms. For example, the plaintext of the owner's budget of 50 million yuan becomes the ciphertext E(5000)=0x3a7d...f2c1 after encryption, ensuring that the data cannot be reverse-cracked during transmission and calculation.

[0080] Implementation details of homomorphic encryption: Key generation: Use a 2048-bit RSA key pair, where the public key is used for encryption and the private key is kept by the data owner; Encryption operation: Perform scalar encryption on numerical data (such as budget amounts) and block encryption on text data (such as design change descriptions); Verifiability: Generate proof parameters (CRS, common reference string) through the Groth16 protocol of zero-knowledge proof to ensure that the encryption process complies with the protocol specifications.

[0081] Generation of verifiable ciphertext data set: Data Sharding: The encrypted data is split into multiple shadow shards according to the participating parties (owner, designer, constructor). For example, the owner's data shards are [S1, S2, S3], which are stored on different blockchain nodes respectively; Proof Generation: Each participating party uses zk-SNARK to generate a proof. For example, the owner proves that the budget ciphertext E(5000) indeed corresponds to the plaintext of 50 million yuan and does not exceed the preset range (such as a 100 million upper limit); Verification and Aggregation: The verification node verifies the validity of all shards through elliptic curve pairing (EC Pairing), and finally generates a globally verifiable ciphertext set VSet={E(Budget), E(Design), E(Resource)}.

[0082] Based on the ciphertext data set, use multi-party secure computing to decompose collaborative tasks, solve the optimal task allocation scheme under the balance of multi-party interests through the Hungarian algorithm, and output a task allocation matrix with weight constraints; Specifically, based on the ciphertext data set, use multi-party secure computing to decompose collaborative tasks, solve the optimal task allocation scheme under the balance of multi-party interests through the Hungarian algorithm, and output a task allocation matrix with weight constraints. It is possible to decompose collaborative tasks according to the homomorphic encrypted ciphertext data set through a secure multi-party computing protocol to generate a task requirement vector with privacy protection, where the secret sharing technology is used to split the task parameters into multiple shadow shares; In the engineering collaborative task allocation, the demand data (such as budget, resource capacity, change tolerance) of the owner, designer, and constructor are converted into a ciphertext data set through homomorphic encryption (such as the Paillier algorithm). For example, the owner's budget constraint of "50 million yuan" is encrypted as ciphertext C1, the design party's drawing change times limit of "3 times" is encrypted as C2, and the constructor's equipment scheduling capacity of "20 units / day" is encrypted as C3.

[0083] The secure multi-party computing (MPC) protocol uses Shamir secret sharing technology to decompose task parameters: Parameter Splitting: Each ciphertext parameter is split into 5 shadow shares (threshold k = 3). For example, the budget ciphertext C1 is split into shares S 11 、S 12 、…、S 15 , and at least 3 shares are required to recover; Distributed Storage: The shares are scattered and stored on 5 independent nodes (such as blockchain shard nodes) to ensure that a single point of failure cannot leak data; Task Requirement Vector Generation: Each participating party collaboratively calculates through the MPC protocol to generate a task requirement vector without exposing the plaintext. For example, when calculating "total budget ≥ equipment procurement cost + labor cost", the ciphertext is directly operated on through the additive homomorphic property, and the output result is an encrypted boolean value (1 indicates satisfaction, 0 indicates dissatisfaction).

[0084] The finally generated task requirement vector contains the following privacy protection fields: Requirement 1: Encrypted budget for equipment procurement (threshold 3 / 5); Requirement 2: Limit on the number of design changes (threshold 2 / 5); Requirement 3: Elasticity coefficient of construction resources (range 0.5 - 1.2).

[0085] Based on the task requirement vector, a multi - party interest game model is constructed, and the initial task allocation plan is calculated through the Nash equilibrium solution algorithm. Among them, the model constraint conditions include the upper limit of the owner's budget, the resource capacity of the construction party, and the change tolerance of the design party; The multi - party interest game model models the owner, the design party, and the construction party as three game participants, and the objective functions are respectively: Owner: Minimize the total cost (objective function f1 = Σ(equipment cost + labor cost)); Design party: Minimize the number of design changes (f2 = Σ(number of drawing modifications)); Construction party: Maximize resource utilization rate (f3 = equipment utilization rate × man - hour efficiency).

[0086] Constraint conditions: Owner's budget ≤ 50 million yuan; Design changes ≤ 3 times; Construction equipment scheduling ≤ 20 units / day.

[0087] The Nash equilibrium solution adopts the Alternating Direction Method of Multipliers (ADMM): Variable splitting: Decompose the global variables (such as the equipment allocation plan) into local copies, and each participant needs to consider the consistency of other parties' copies when optimizing its own objective; Iterative update: In each round of iteration, each party alternately updates the local variables and coordinates the conflicts through the Lagrange multiplier. For example, the owner proposes an equipment procurement plan in the t - th round, and the construction party adjusts the equipment scheduling plan in the (t + 1) - th round to match the procurement plan; Convergence determination: Terminate when the change rate of the objective function < 0.1% or the maximum number of iterations (such as 100 times) is reached.

[0088] An example of the initial task allocation plan is as follows: Equipment procurement: 10 excavators (budget 12 million yuan); Design changes: 2 times (involving structural reinforcement); Resource scheduling: Equipment utilization rate is 85%, and man-hour efficiency is 0.9.

[0089] Input the initial allocation plan into the Hungarian algorithm, introduce a dynamic weight adjustment strategy to optimize the conflict task matching. Among them, the dynamic weight is adaptively updated according to the real-time supply chain risk index, and an intermediate matrix of task allocation with elastic constraints is output; The Hungarian algorithm is used to solve the bipartite graph matching problem of tasks and resources. When constructing the cost matrix, introduce dynamic weight adjustment: Definition of the cost matrix: Rows represent tasks (such as earth excavation, structural pouring), columns represent resources (equipment, personnel), and the matrix element Cᵢⱼ represents the cost of task i using resource j; Dynamic weight: Supply chain risk index (0 - 100, calculated from logistics delay rate, raw material price fluctuations, etc.): When the index > 70, the equipment scheduling cost weight increases by 30%; Elastic constraint: The resource capacity of the construction party allows a ±10% fluctuation, which is reflected in the expansion of the resource column in the matrix (e.g., "excavator" is split into "excavator - normal" and "excavator - elastic").

[0090] Optimization process of the Hungarian algorithm: Initial matching: Find the minimum line set that covers all rows and columns with zero elements; Adjust the weight: Update the cost matrix according to the real-time risk index. For example, when the risk index rises to 75, adjust the cost of "earth excavation - excavator" from 100 to 130; Iterative optimization: Search for the optimal matching through augmenting path search, with a time complexity of O(n³), supporting up to 1000 task nodes.

[0091] Exemplarily, the information contained in the output intermediate matrix of task allocation is as shown in Table 1 below: Table 1 The optimal matching is: Earth excavation → Excavator - elastic (cost 150), Structural pouring → Crane (200), Pipeline laying → Excavator - normal (180).

[0092] Verify the fairness of the intermediate matrix of task allocation through zero-knowledge proof, use the zk - SNARK protocol to generate a verifiable allocation proof, and finally output a weighted-constrained task allocation matrix that is tamper-proof and satisfies the multi-party interest balance.

[0093] Zero-knowledge proof (zk - SNARK) is used to verify the fairness of the task allocation scheme and ensure no malicious manipulation: Arithmetic Circuit Construction: Encoding the allocation logic into a circuit. For example, verifying "Total Cost ≤ Budget" can be represented as a gate circuit: Σ(Allocation Option × Cost) ≤ 5000; Trusted Setup: Generating a proof key (pk) and a verification key (vk), and avoiding the single-point trust problem through a multi-party computation ceremony (such as the Tau ceremony); Proof Generation: Inputting the intermediate matrix of task allocation, generating a proof π (about 1KB), which proves that the allocation satisfies all constraints and does not disclose sensitive data; Verification: Any verifier can confirm the validity of the scheme within milliseconds using vk and π without knowing the specific allocation details.

[0094] The final output weighted constraint task allocation matrix includes: Resource Allocation Details: Encrypted storage on the blockchain (such as Hyperledger Fabric); Dynamic Weight Log: Recording the temporal relationship between the supply chain risk index and weight adjustment; zk-SNARK Proof: Ensuring the immutability and audibility of the scheme.

[0095] For example, after the zk-SNARK proof of a certain allocation scheme is automatically verified through a smart contract, the following operations are triggered: The owner releases the first installment to the smart contract escrow account; The designer uploads a new version of the BIM model to the collaboration platform; The constructor schedules equipment according to the matrix, and the real-time data is uploaded to the blockchain for evidence storage.

[0096] Encoding the task allocation matrix into a smart contract, defining the contract trigger conditions through an event-driven state machine, and generating a smart contract template containing an execution logic chain; The smart contract is encoded in the Solidity language and realizes automated execution based on an event-driven architecture. Taking the "Foundation Pouring" task as an example, the contract logic includes: State Definition: State 0: Task to be allocated; State 1: The constructor confirms the resources; State 2: The owner pays the advance payment; State 3: Task in execution; State 4: Acceptance completed.

[0097] Event Trigger: Event_ResourceConfirm: The constructor uploads a proof of resource preparation (such as a concrete test report); Event_Payment: The owner pays 30% of the advance payment through an on-chain transfer; Event_ProgressUpdate: The construction party submits the daily progress (such as 100 m³ of pouring volume).

[0098] Execution logic chain: When Event_ResourceConfirm is triggered and the signature verification passes, the status changes from 0 to 1; When Event_Payment detects that the amount is ≥ 30% of the budget, the status changes from 1 to 2; When Event_ProgressUpdate has cumulatively completed ≥ 80%, the status changes from 3 to 4.

[0099] Intelligent contract template generation: Automatically generate contract instances according to different task parameters (such as budget, construction period) and deploy them to the blockchain network.

[0100] Deploy lightweight blockchain nodes on the BIM / CIM collaborative platform, and achieve privacy-protected data synchronization across participating parties through the sharding consensus mechanism to generate a timestamped collaborative instruction hash chain; The lightweight blockchain nodes are optimized based on the Hyperledger Fabric architecture. Each participating party (owner, designer, construction party) deploys a Peer node to jointly form a consortium chain. The sharding consensus adopts the PBFT (Practical Byzantine Fault Tolerance) algorithm, divides the network into multiple shards (such as shard 1 processes design data, shard 2 processes construction data), and each shard contains 3 nodes (tolerating 1 faulty node).

[0101] Data synchronization process: Transaction proposal: The participating party submits a data update request (such as the designer uploads a new version of the BIM model) to generate a transaction proposal TxProposal; Consensus within the shard: The shard nodes reach an agreement through three rounds of PBFT voting (Pre-prepare, Prepare, Commit). For example, nodes A / B / C in shard 1 all agree that the model version V2.3 is valid; Cross-shard synchronization: Ensure the integrity of cross-shard transactions through the Atomic Commit protocol. For example, when the design model is updated and needs to be synchronized to the construction shard, if any shard rejects, the whole will be rolled back; Hash chain generation: Each transaction is packaged into a block, and the block header contains the previous hash, timestamp (UNIX millisecond level), and Merkle root. For example, block #7821 records "Design change V2.3", hash 0x5b9e...d4a3, timestamp 1630000000000.

[0102] Privacy protection mechanism: Channel Isolation: Sensitive data (such as budgets) is transmitted through private channels and can only be decrypted by relevant participants. Data Masking: Non-key fields (such as log IDs) are hashed to avoid exposing business details.

[0103] Based on the real-time engineering data stream, the authenticity of off-chain data is verified through an oracle, the execution status of the smart contract is dynamically updated, and a traceable collaborative instruction set is output.

[0104] The oracle, as a bridge for off-chain data, collects real-time data from Internet of Things devices (such as concrete humidity sensors) and ERP systems (such as resource inventory databases), and encrypts and transmits it to the blockchain through TLS (Transport Layer Security Protocol). Taking "concrete pouring progress" as an example: Data Collection: The sensor uploads humidity data (such as humidity 72%) every 5 minutes, and the ERP system updates the inventory every hour (such as 500 tons of remaining concrete). Oracle Verification: The Chainlink oracle node is used to ensure data authenticity through cross-verification of multiple data sources (such as comparing the average value of 3 sensors). Smart Contract Trigger: When the humidity ≥ 70% and the inventory ≥ 200 tons, Event_ProgressUpdate is triggered to update the task status to "in progress". Instruction Set Generation: All status change records generate an instruction sequence in chronological order, for example: {Time: 1630000000, Cmd: "StartFoundation", Params: {Concrete: 200t}}, {Time: 1630003600, Cmd: "PauseDueToRain", Params: {Delay: 2h}}, {Time: 1630010800, Cmd: "ResumeWork"} 。

[0105] Traceability Assurance: Full-chain Audit: Each instruction is associated with a blockchain transaction hash and can be traced back to the original data. Digital Signature: Participants sign key instructions (such as the signature of the construction party leader 0x3a7d...) to ensure traceability of responsibilities.

[0106] ​Based on the cost prediction results, homomorphic encryption is used for multi-party private data (such as owner budgets, design change requirements, construction resource plans) using zero-knowledge proofs. The task allocation scheme is optimized through the Hungarian algorithm and encoded into an automatically executable smart contract. The blockchain sharding consensus mechanism ensures that the instruction set is synchronized and immutable in the BIM / CIM platform, enabling efficient multi-party collaboration while protecting business secrets; the smart contract automatically triggers task instructions (such as material procurement, process adjustment), reducing human communication delays and errors.

[0107] S204, based on the collaborative instruction set and real-time engineering data stream, parse potential risk paths through a dynamic knowledge graph engine, and output a risk probability matrix and early warning signals. Among them, the dynamic knowledge graph engine integrates an industry knowledge base and a reinforcement learning strategy to achieve risk evolution simulation; Specifically, according to the collaborative instruction set, risk semantic fragments in engineering contract terms and construction specification documents can be extracted through a domain adaptive pre-training model to generate risk knowledge triples with probability annotations; The domain adaptive pre-training model uses the RoBERTa model (Robustly Optimized BERT model) based on the Transformer architecture, and is secondarily pre-trained through professional corpora in the engineering field to make it adapt to engineering risk semantic understanding. The model input is the text paragraphs of contract terms and construction specifications (such as "The concrete strength grade shall not be lower than C30"), and the output is risk entities and their associated relationships.

[0108] Data preprocessing and training process: Entity annotation: The BIO (Begin-Inside-Outside) annotation system is adopted to define risk entity types (such as "material risk", "construction period risk", "safety risk") and relationships (such as "causes", "associates"). For example, in the sentence "The unqualified steel bar specifications may cause structural instability", "The unqualified steel bar specifications" is annotated as a material risk (B-MAT, I-MAT), "structural instability" is annotated as a safety risk (B-SAFE, I-SAFE), and the relationship is "causes".

[0109] Domain adaptive training: Based on a general Chinese corpus (such as WikiZh), load engineering field texts (100,000 pieces), and perform pre-training using a dynamic masking strategy (masking probability 15%), with a learning rate set to 2e-5, a batch size of 32, and training for 3 epochs.

[0110] Probability annotation: When the model outputs entity relationships, confidence probabilities (0~1) are generated through the Softmax layer. For example, the confidence of "The unqualified steel bar specifications → causes → structural instability" is 0.92.

[0111] Risk knowledge triple generation: Format: (head entity, relationship, tail entity, probability, time window). For example: (Insufficient concrete strength, causes, risk of crack propagation, 0.85, [t0 + 7d, t0 + 14d]).

[0112] Derivation of time window: Based on the statistics of historical engineering data, for example, concrete strength problems usually cause cracks 7 - 14 days after construction.

[0113] Based on the risk knowledge triple, combined with abnormal events in the real - time engineering data stream, construct a topological graph of the risk propagation path through a temporal graph convolutional network; The Temporal Graph Convolutional Network (T - GCN) is composed of alternating Time Convolutional Layers (TCN) and Graph Convolutional Layers (GCN) for modeling the propagation of risks in the time and space dimensions.

[0114] Data processing and model construction: Abnormal event detection: Extract features from the real - time engineering data stream (such as excessive concrete humidity monitored by sensors, progress delay alarm) through a sliding window (window size 1 hour, step size 10 minutes), and use the Isolation Forest algorithm to detect abnormalities (abnormal score > 0.6 is determined as an abnormality).

[0115] Graph initialization: Construct an initial graph with risk knowledge triples as nodes and edges. Node attributes include risk type, probability, and time window; edge weights are confidence levels.

[0116] Temporal graph convolution: Time Convolutional Layer: Use dilated causal convolution (dilation coefficient 2, convolution kernel size 3) to capture the delayed effect of risks over time (such as material defects causing progress risks after 3 days).

[0117] Graph Convolutional Layer: Aggregate the features of adjacent nodes. For example, the node of "Insufficient concrete strength" aggregates the features of "Incorrect cement grade" and "Improper curing".

[0118] Dynamic edge weight update: Adjust the edge weights according to real - time abnormal events. For example, when "Daily progress delay ≥ 8%" is detected, the edge weight of "Project duration risk → Cost overrun" is increased from 0.7 to 0.9.

[0119] Output of the risk propagation path topological graph: Nodes: 200 - 500 risk points (depending on the project scale); Edges: Cascade paths (such as A → B → C) are highlighted in red, and independent risks are shown in gray; Topological attributes: Each path is labeled with the propagation time delay (such as "Material procurement delay → Shutdown risk: 2 days").

[0120] Input the risk propagation path into the reinforcement learning environment, simulate the risk evolution process through the double deep Q-network, and generate a risk probability transition matrix containing the intensity of the cascade effect; The reinforcement learning environment is customized based on the OpenAI Gym framework, and the agent learns risk intervention strategies through the double deep Q-network (DDQN).

[0121] Environment settings: State space: A 300-dimensional vector, including the probability of each risk node, project progress, and resource margin; Action space: 20 intervention measures (such as "increase the concrete inspection frequency", "urgent purchase of steel bars"); Reward function: R = 0.5×(1 - risk probability) + 0.3×progress compliance rate + 0.2×cost savings rate.

[0122] DDQN training process: Network structure: Both the main network and the target network are 3-layer fully connected (512-256-128), ReLU activation, learning rate 0.001, and the experience replay cache capacity is 10,000; Training process: Sample 100 experiences (state-action-reward-new state) in each iteration, update the parameters of the main network, and synchronize the target network every 100 steps; Risk evolution simulation: The agent tries different actions in the virtual environment. For example, selecting "enable alternative suppliers" can reduce the risk probability of "material shortage" from 0.8 to 0.3, but the cost increases by 5%.

[0123] Risk probability transition matrix generation: Matrix dimension: N×N (N is the number of risk nodes), and the element Pij represents the probability that risk i triggers risk j; Cascade intensity calculation: Statistically calculate the path activation frequency through Monte Carlo simulation (1000 times). For example, the cascade intensity of the path "design change → project duration delay → cost overrun" is 0.75.

[0124] Based on the risk probability transition matrix, use the backpropagation importance sampling algorithm to identify the key risk paths and output early warning signals with time window constraints; The backpropagation importance sampling (BP-IS) algorithm locates the key paths through gradient backpropagation and combines importance sampling to reduce the computational complexity.

[0125] Algorithm steps: Gradient calculation: Take the total risk probability as the objective function and calculate the gradient of each path. For example, the gradient of the path A→B is ∂P_total / ∂P_AB = 0.6, where P_total is the total risk probability and P_AB is the risk probability of the path A→B, indicating that it has a greater impact on the overall risk; Importance Sampling: Allocate sampling weights according to the absolute value of the gradient, and preferentially explore high-impact paths (for example, the sampling probability of the top 10% of paths in terms of gradient is increased to 50%); Critical Path Screening: Define the criticality index CI = gradient × frequency / cost, and screen paths where CI > 0.5.

[0126] Early Warning Signal Generation: Signal Format: JSON structure, including risk paths, expected outbreak time, and recommended measures. For example: { "risk_path": ["Insufficient concrete strength", "Structural cracks", "Acceptance failure"], "time_window": ["2023-10-05", "2023-10-12"], "action": "Immediately conduct third-party strength testing" }

[0127] Dynamic Push: Real-time send to the mobile terminals of management personnel through 5G messages (5G MSG), with a time delay < 200ms.

[0128] Compress the risk model parameters through knowledge distillation technology, deploy them to edge computing nodes to achieve real-time risk monitoring, and generate a lightweight risk evolution simulator.

[0129] Knowledge distillation adopts a teacher-student model architecture, compressing the complex T-GCN + DDQN model (teacher) into a lightweight simulator (student).

[0130] Distillation Process: Output of the Teacher Model: Prediction results for historical risk cases (such as risk probability, propagation path); Construction of the Student Model: Adopt the MobileNetV3 small architecture (with 1M parameters), and compress the input dimension to 64; Loss Function: Combine cross-entropy (student output and true label) and KL divergence (similarity between student and teacher output distributions), with a weight ratio of 6:4; Training Configuration: Learning rate 0.01, batch size 64, training for 50 rounds.

[0131] Edge Deployment and Optimization: Hardware Adaptation: Deploy to the Huawei Atlas 500 edge intelligent small station (computing power 16 TOPS), with memory occupancy < 500MB; Real-time Inference: Input real-time data (such as "Today's progress is delayed by 10%"), and output a risk warning within 1 second; Dynamic update: Synchronize new knowledge of the cloud teacher model weekly and incrementally update the parameters of the student model.

[0132] Lightweight simulator performance: Accuracy retention: Compared with the teacher model, the risk prediction accuracy of the student model decreases by ≤ 3% (F1-score drops from 0.91 to 0.88); Latency optimization: The single inference time is reduced from 2.1 seconds to 0.3 seconds, meeting the real-time requirements of the engineering site.

[0133] The dynamic knowledge graph engine extracts risk knowledge triples from engineering contracts and construction specifications (such as "material shortage → project duration delay → liquidated damages"), constructs a topological graph of the risk propagation path in combination with real-time data (such as logistics anomaly events), and simulates the risk evolution process through reinforcement learning to output a probabilistic risk matrix (such as the probability of cascading risk outbreaks) and early warning signals (such as a 3-week advance warning of concrete supply interruption), combining industry experience with real-time data to quantify the risk conduction path; realizing "predictive management" of risks through simulation and deduction, and supporting the formulation of targeted prevention and control strategies.

[0134] S205, according to the risk probability matrix, use a multi-objective optimization algorithm to generate a set of anti-interference decision-making solutions, and output the optimal cost control strategy after verification by digital twin simulation, where the optimal cost control strategy synchronously drives the 3D risk heat map and decision path deduction animation in the visualization interface.

[0135] Specifically, according to the risk probability matrix, the NSGA-III algorithm can be used to synchronously optimize three objectives: cost control rate, project quality compliance rate, and risk aversion coefficient, generating an initial Pareto front containing 300 - 500 groups of solutions; NSGA-III (the third-generation non-dominated sorting genetic algorithm) is used to find the Pareto optimal solution set in multi-objective optimization problems. In this step, the optimization objectives are the cost control rate (objective 1), project quality compliance rate (objective 2), and risk aversion coefficient (objective 3).

[0136] Parameter setting and population initialization: Population size: Set to 500 to cover a wider solution space; Variable dimension: Includes 10 decision variables such as material procurement cost (0 - 100% of the budget), construction period (10 - 365 days), and risk response resource allocation ratio (0 - 100%); Reference point generation: Use the Das-Dennis method to generate uniformly distributed reference points, with a quantity of 100, to ensure that the solution set is evenly distributed in the objective space.

[0137] Genetic operation: Crossover: Simulated Binary Crossover (SBX) is adopted with a crossover probability of 0.9 and a distribution index η = 20. For example, parental solutions A (cost control rate 70%, quality compliance rate 85%) and B (cost control rate 65%, quality compliance rate 90%) generate offspring solution C (68%, 88%) through crossover; Mutation: Polynomial mutation with a mutation probability of 0.1. For example, the risk aversion coefficient of a certain solution mutates from 60% to 55%.

[0138] Non-dominated Sorting and Selection: Non-dominated Layers: The population solutions are divided into multiple non-dominated layers. For example, the first layer contains 50 solutions and the second layer contains 80; Reference Point Association: The solutions are mapped to reference points, and the solution belonging to the reference point that is the closest and not occupied is selected to ensure diversity.

[0139] Example: In a certain optimization, 480 groups of solutions are generated, and the Pareto front contains: Solution A: Cost control rate 82%, quality compliance rate 92%, risk aversion coefficient 75%; Solution B: Cost control rate 78%, quality compliance rate 95%, risk aversion coefficient 80%.

[0140] These solutions ensure convergence through the elitist retention strategy of NSGA-III (retaining the top 10% of the solutions in each generation).

[0141] Input the Pareto solution set into the digital twin engine, and simulate the construction process under different decision-making scenarios through the physical engine to generate a multi-dimensional simulation data set containing cost deviations, quality defects, and risk outbreak points; The digital twin engine constructs a virtual construction environment based on BIM (Building Information Modeling) and real-time sensor data, and the physical engine uses NVIDIA PhysX to achieve high-precision dynamics simulation.

[0142] Simulation Scenario Construction: Model Import: Convert the BIM model to USD (Universal Scene Description) format, containing more than 2000 components such as structural members and mechanical and electrical pipelines; Parameter Binding: Map the decision variables in the Pareto solution (such as material cost) to the model attributes (such as concrete strength grade C30 → C35).

[0143] Physical Simulation Process: Time Step: Set to 1 day to simulate a 365-day construction period; Event Trigger: When the risk value of a certain node in the risk probability matrix > 0.7, trigger an emergency resource scheduling event; Data collection: Record the daily cost deviation (actual cost - budget), quality defects (such as the wall flatness error > 3mm), and risk outbreak points (such as the number of supply chain disruptions).

[0144] Example: The simulation of solution A shows that on the 120th day, the cost deviation is +12% due to the increase in steel prices, on the 200th day, insufficient concrete strength is found in a certain area (quality defect marked), and on the 280th day, the logistics is interrupted due to a typhoon (risk outbreak).

[0145] Dataset generation: Dimension definition: Include time series cost curve (365 points), quality defect distribution heat map (partitioned by floor), and spatio-temporal coordinates of risk events (latitude and longitude + time); Data storage: Adopt Parquet columnar storage, with a compression ratio of 5:1, supporting fast querying.

[0146] Based on the simulation dataset, use the fuzzy comprehensive evaluation method to calculate the fitness scores of each solution, and screen the optimal cost control strategy through the gradient boosting decision tree; The fuzzy comprehensive evaluation method quantifies the satisfaction of each objective through the membership function and calculates the comprehensive score by combining the weights.

[0147] Membership function design: Cost control rate: Trapezoidal function, ideal interval [80%, 100%], membership degree 0 below 70%; Quality compliance rate: Gaussian function, mean 95%, standard deviation 2%; Risk aversion coefficient: S-shaped function, threshold 60% (membership degree drops sharply below 60%).

[0148] Weight allocation: Owner's preference: Cost control (weight 0.5), quality (0.3), risk (0.2); Constructor's preference: Quality (0.5), risk (0.3), cost (0.2).

[0149] Determine the final weights through AHP (Analytic Hierarchy Process) as cost 0.4, quality 0.4, and risk 0.2.

[0150] Gradient boosting decision tree (GBDT) screening: Feature engineering: Input features include 15-dimensional indicators such as the mean cost deviation, maximum quality defect level, and risk event frequency; Model parameters: Number of trees 100, learning rate 0.1, maximum depth 6; Training and prediction: Use 80% of the data for training and 20% for validation, and output the fitness scores of each solution (0 - 100 points).

[0151] Example: The fitness score of solution A is 92 (cost 88, quality 90, risk 85), and the score of solution B is 89 (cost 85, quality 95, risk 80). Finally, solution A is selected as the optimal strategy.

[0152] Additionally, based on the simulation dataset, the fuzzy comprehensive evaluation method is used to calculate the fitness scores of each solution, and the gradient boosting decision tree is used to screen the optimal cost control strategy, which may include: According to the multi-dimensional engineering parameters in the simulation dataset, through a dynamic weight generator based on Monte Carlo sampling, adaptive weight allocation is performed for the three objectives of cost deviation rate, quality defect index, and risk outbreak probability to generate a dynamic evaluation weight vector with a confidence interval. Based on the engineering parameter dataset (including 12-dimensional parameters such as cost deviation rate, quality defect index, and risk outbreak probability) generated by digital twin simulation, a dynamic weight generator is constructed. First, the Monte Carlo method is used to randomly sample the priorities of the three objectives: set a group of weight combinations for each sampling (such as cost deviation rate weight w1 = 0.4, quality defect index w2 = 0.3, risk outbreak probability w3 = 0.3), and perform 1000 iterations of sampling. In each iteration, according to the success criteria in the historical engineering case library (such as cost deviation rate ≤ 5%, quality defect index ≤ 0.2, risk outbreak probability ≤ 10%), calculate the fitness score of the current weight combination, and screen out the effective weight interval that meets the constraint conditions. For example, through statistical distribution, it is found that when w1 ∈ [0.35, 0.45], w2 ∈ [0.25, 0.35], and w3 ∈ [0.2, 0.3], the confidence interval of the fitness score (95% confidence level) is [82.3, 89.5]. Finally, a dynamic evaluation weight vector (w1 = 0.42 ± 0.03, w2 = 0.28 ± 0.02, w3 = 0.30 ± 0.02) is generated, and confidence labels for each weight are marked (such as the confidence level of the cost deviation rate weight is 0.92).

[0153] Based on the dynamic evaluation weight vector, the fuzzy comprehensive evaluation algorithm is used to calculate the fitness score of the solution. The fuzzy comprehensive evaluation algorithm introduces a time decay factor to correct the contribution degree of historical simulation data, and outputs a fitness score matrix with a timeliness label through a radial basis function kernel mapping of high-dimensional non-linear patterns. ‌1. Time Decay Factor and Fuzzy Comprehensive Evaluation‌ When calculating the fitness score, considering the time series characteristics of the simulation dataset, a time decay factor (decay factor λ = 0.85) is designed to dynamically adjust the contribution of historical data. For example, the weight of the simulation data at the current time t is 1.0, while the weight of the data at time t - 1 is λ = 0.85, and at time t - 2 is λ² = 0.72, and so on, ensuring that the latest data has a greater impact on the evaluation result. At the same time, a fuzzy membership function (such as the membership function of the cost deviation rate is defined as a "low deviation" trapezoidal function, with a threshold interval [0, 5%]) is used to fuzzify the three objectives. Combining the dynamic weight vector with the fuzzy membership degree, the fitness score of each scheme is calculated. For example, for a certain scheme, the cost deviation rate is 3.2% (membership degree 0.86), the quality defect index is 0.15 (membership degree 0.92), and the risk outbreak probability is 8% (membership degree 0.78), and its comprehensive score is 0.42×0.86 + 0.28×0.92 + 0.30×0.78 = 0.847.

[0154] 2. Radial Basis Function Kernel Mapping and Timeliness Label To process high-dimensional non-linear data (such as the complex relationship between construction process parameters and cost deviation), a Radial Basis Function (RBF) kernel (parameter gamma = 0.1) is used to perform feature mapping on 12-dimensional engineering parameters. Through the kernel function, the original data is projected into a high-dimensional space to extract non-linear relationship features (such as the U-shaped correlation between concrete strength and quality defects). In the finally generated fitness score matrix, each scheme is labeled with a timeliness label (such as "the proportion of data in the past week is 70%") for subsequent decision-making screening.

[0155] Based on the fitness score matrix, a multi-modal feature fusion channel is constructed to extract the hyperplane curvature feature, decision variable correlation feature, and simulation process stability feature of the Pareto solution set, and an enhanced feature vector set with a unified dimension is generated through a heterogeneous feature embedding network; 1. Multi-modal Feature Extraction and Fusion Hyperplane Curvature Feature: For the Pareto solution set (300 - 500 groups of solutions) generated by the NSGA-III algorithm, principal component analysis (PCA) is used to reduce the dimension to 3-dimensional space, and the curvature radius of the solution set front (such as the average curvature radius R = 12.5) is calculated. The greater the curvature, the higher the diversity of the solution set and the more complex the decision space.

[0156] Decision Variable Correlation Feature: The Pearson correlation coefficient is used to analyze the correlation between cost control strategy parameters (such as material procurement cycle, number of construction teams) and the three objectives. For example, it is found that the correlation coefficient r between the procurement cycle and the cost deviation rate is 0.62 (strong positive correlation), and r with the quality defect rate is -0.33 (weak negative correlation).

[0157] Stability characteristics of the simulation process: Statistically analyze the variance of the results of 10 repeated experiments of the same strategy in digital twin simulation (e.g., the variance of cost deviation σ² = 0.8). The smaller the variance, the stronger the robustness of the strategy.

[0158] ‌2. Heterogeneous feature embedding network‌ Construct a heterogeneous network that includes a convolutional layer (to extract spatial features), an LSTM layer (to extract temporal features), and a fully connected layer (to fuse features). Input the curvature feature (dimension 3), the correlation feature (dimension 6), and the stability feature (dimension 1) into the network, and generate an enhanced feature vector of 128 dimensions with a unified dimension through a weight sharing mechanism (e.g., the size of the convolutional kernel is 3×3, and the stride is 1). For example, the feature vector of a certain strategy is expressed as [curvature = 0.12, procurement cycle correlation = 0.62, stability = 0.92,...].

[0159] Input the enhanced feature vector set into the gradient boosting decision tree model, adopt a multi-objective splitting criterion to optimize the growth direction of the tree structure, and strengthen the search ability of the high-dimensional non-convex solution space through a cost-sensitive learning strategy, and output the global ranking result of the optimal cost control strategy; 1. Gradient boosting decision tree (GBDT) modeling‌ Set the GBDT model parameters: the maximum depth of the tree is 5 (to prevent overfitting), the learning rate lr = 0.05 (to balance the training speed and accuracy), and the number of iterations n_estimators = 200. Adopt a multi-objective splitting criterion, and simultaneously optimize the weighted information gain of three objectives (cost, quality, and risk) during each node split. For example, when a certain node splits according to "procurement cycle ≤ 7 days", the information gain of the left subtree is the cost objective gain 0.15 + the quality objective gain 0.08 + the risk objective gain 0.06 = 0.29, and the right subtree is 0.21. Select the left subtree as the splitting direction.

[0160] 2.‌ Cost-sensitive learning and global ranking‌ Introduce a cost-sensitive learning strategy, impose a higher penalty weight (penalty = 3.0) on misclassified samples with high cost deviation (e.g., deviation rate > 10%), and set the penalty weight for low-risk samples (risk outbreak probability < 5%) to 1.0. After model training, output the global ranking result of the strategy (e.g., strategy A scores 92.3, strategy B scores 88.7), and label the advantages and disadvantages of each strategy (e.g., "cost control priority" for strategy A, "risk aversion priority" for strategy B).

[0161] According to the global ranking result, combined with the physical constraint verification of the digital twin engine, iteratively optimize the strategy parameters through a backpropagation correction mechanism to generate an optimal cost control strategy execution sequence that meets the construction feasibility boundary.

[0162] 1. Physical Constraint Verification and Backpropagation Correction Input the top 10 strategies into the digital twin engine, simulate the construction process based on a physics engine (such as NVIDIA PhysX), and verify whether the strategies comply with actual constraints (such as tower crane bearing capacity ≤ 50 tons, single-day concrete pouring volume ≤ 300 m³). For example, a certain strategy is marked as infeasible due to tower crane overload (simulation shows that 55 tons need to be lifted). Through the backpropagation mechanism, feedback the constraint violation signal to the GBDT model, adjust the feature weights (such as increasing the weight of "tower crane utilization rate" by 20%), and regenerate the strategy ranking.

[0163] 2. Execution Sequence Generation and Visualization Finally, 3 groups of feasible strategies (such as Strategy A, Strategy D, Strategy G) are selected, and the execution sequence is generated according to the priority: First stage (1 - 30 days): Execute Strategy A (cost deviation rate controlled at 3.5%); Second stage (31 - 60 days): Switch to Strategy D (to cope with sudden risk events); Third stage (61 - 90 days): Enable Strategy G (to ensure the quality of completion acceptance).

[0164] In the BIM collaboration platform, render the 3D deduction animation through Unreal Engine to display the cost curve changes of strategies in each stage, the diffusion process of the risk heat map, and the trigger logic of key decision nodes.

[0165] Integrate the optimal strategy with the BIM model, render the 3D risk heat map through ray tracing technology, and generate the decision path deduction animation to output the visual decision support interface.

[0166] The visualization system is developed based on the Unity engine, integrating the BIM model and real-time data stream to achieve dynamic rendering and interaction.

[0167] 3D Risk Heat Map Rendering: Data mapping: Map the risk values (0 - 1) in the risk probability matrix to a color gradient (green → yellow → red); Ray tracing: Use the NVIDIA RTX 6000 GPU to calculate ray reflection in real time. For example, high-risk areas (>0.8) are displayed in dark red and superimposed with a flashing effect; Interaction function: Click on the heat map area to pop up a detailed risk analysis (such as "the risk value of the support structure in the second basement floor is 0.78").

[0168] Decision Path Deduction Animation: Timeline control: The animation duration is 5 minutes, and the 365-day construction process is compressed and displayed; Key event annotation: For example, when the cost is overspent on the 120th day, the screen focuses on the procurement module and pops up optimization suggestions ("enable backup suppliers"); Multi-perspective switching: supports global bird's-eye view, first-person inspection, component sectioning and other perspectives.

[0169] Example: In the animation, users can observe the following process: Day 0-100: Green heat is the main source, and cost control is stable; Day 120: A certain area turns yellow, indicating the impact of steel price increases; Day 200: Part of the surface turns red, indicating concrete defects, and the system automatically triggers quality rectification instructions; Day 365: The overall situation returned to green, and the signage project was successfully accepted.

[0170] Output interface: Main panel: 3D heat map on the left and fitness score ranking on the right; Console: supports parameter adjustment (such as re-running optimization after modifying weights); Report export: Generate decision analysis reports in PDF / Excel format.

[0171] The NSGA-III algorithm is used to balance multiple objectives such as cost control rate, quality compliance rate, and risk aversion coefficient, and hundreds of candidate solutions are generated. The digital twin engine is used to simulate the performance of each solution in the virtual construction environment (such as cost deviation and quality defects), and finally the optimal strategy is selected. The visualization interface locates high-risk areas through a three-dimensional heat map, and uses animation to deduce the effect of decision implementation, providing a global optimal solution under complex constraints, avoiding the local optimal trap caused by single-objective optimization; through simulation verification and visualization, the credibility of decisions is enhanced, helping managers to intuitively understand the impact of strategies.

[0172] It can be seen that according to the heterogeneous cost data of each stage of project planning, design, construction, and completion, a structured cost data cube with temporal and spatial correlation is generated; based on the structured cost data cube, a dynamic cost prediction curve and deviation-sensitive nodes are output; according to the dynamic cost prediction curve, the blockchain-enabled BIM / CIM collaborative platform is used to optimize multi-party task allocation and generate a collaborative instruction set encoded in a smart contract; based on the collaborative instruction set and real-time engineering data stream, a risk probability matrix and an advance warning signal are output; according to the risk probability matrix, a multi-objective optimization algorithm is used to generate an interference-resistant decision-making solution set, and the optimal cost control strategy is output after verification through digital twin simulation, so that the cost data of each stage of the project can be efficiently integrated to achieve dynamic cost prediction, optimized collaborative decision-making and effective risk management.

[0173] Another embodiment of the present invention provides a multi-dimensional intelligent management system for the whole process cost.Figure 3 , the system may include: An alignment module 301, configured to perform cross-stage feature alignment on heterogeneous cost data in each stage of project planning, design, construction, and completion through a data fusion engine driven by federated learning to generate a spatio-temporally correlated structured cost data cube, wherein the data fusion engine uses a differential privacy protection mechanism to eliminate data islands; A construction module 302, configured to construct a spatio-temporal graph neural network prediction model based on the structured cost data cube, combined with market price fluctuation trends and construction progress change factors, and output a dynamic cost prediction curve and deviation-sensitive nodes, wherein the spatio-temporal graph neural network prediction model captures the associated impacts of potential cost overrun risk factors through an attention mechanism; An optimization module 303, configured to perform multi-party task allocation optimization on a blockchain-enabled BIM / CIM collaboration platform according to the dynamic cost prediction curve to generate a collaborative instruction set encoded with smart contracts, wherein the BIM / CIM collaboration platform realizes privacy-protected data synchronization among the owner, the designer, and the constructor through zero-knowledge proof; An analysis module 304, configured to analyze potential risk paths based on the collaborative instruction set and real-time project data streams through a dynamic knowledge graph engine, and output a risk probability matrix and an early warning signal, wherein the dynamic knowledge graph engine fuses an industry knowledge base and a reinforcement learning strategy to achieve risk evolution simulation; An output module 305, configured to generate an anti-interference decision plan set using a multi-objective optimization algorithm according to the risk probability matrix, and output an optimal cost control strategy after verification through digital twin simulation, wherein the optimal cost control strategy synchronously drives a three-dimensional risk heat map and a decision path deduction animation in a visualization interface.

[0174] It can be seen that, based on heterogeneous cost data in each stage of project planning, design, construction, and completion, a spatio-temporally correlated structured cost data cube is generated; based on the structured cost data cube, a dynamic cost prediction curve and deviation-sensitive nodes are output; according to the dynamic cost prediction curve, multi-party task allocation optimization is performed on a blockchain-enabled BIM / CIM collaboration platform to generate a collaborative instruction set encoded with smart contracts; based on the collaborative instruction set and real-time project data streams, a risk probability matrix and an early warning signal are output; according to the risk probability matrix, an anti-interference decision plan set is generated using a multi-objective optimization algorithm, and an optimal cost control strategy is output after verification through digital twin simulation, thereby enabling efficient integration of cost data in each stage of the project, realizing dynamic cost prediction, optimized collaborative decision-making, and effective risk management.

[0175] The structure, features and effects of the present invention have been described in detail based on the embodiments shown in the drawings. The above is only the preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or equivalent embodiments modified into equivalent changes, which still do not exceed the spirit covered by the specification and the drawings, shall fall within the protection scope of the present invention.

Claims

1. A multi-dimensional intelligent management method for the whole-process cost, characterized in that The method includes: According to the heterogeneous cost data in the engineering planning, design, construction, and completion stages, through the data fusion engine driven by federated learning, cross-stage feature alignment is performed to generate a spatio-temporally correlated structured cost data cube. Among them, the data fusion engine uses a differential privacy protection mechanism to eliminate data islands; Based on the structured cost data cube, combined with the market price fluctuation trend and the construction progress change factor, a spatio-temporal graph neural network prediction model is constructed to output a dynamic cost prediction curve and deviation-sensitive nodes. Among them, the spatio-temporal graph neural network prediction model captures the associated impact of potential cost overrun risk factors through an attention mechanism; According to the dynamic cost prediction curve, a multi-party task allocation optimization is carried out using the blockchain-enabled BIM / CIM collaborative platform to generate a collaborative instruction set encoded with smart contracts. Among them, the BIM / CIM collaborative platform realizes privacy-protected data synchronization among the owner, the design party, and the construction party through zero-knowledge proof; Based on the collaborative instruction set and the real-time engineering data stream, the potential risk path is analyzed through a dynamic knowledge graph engine to output a risk probability matrix and an early warning signal. Among them, the dynamic knowledge graph engine fuses the industry knowledge base and the reinforcement learning strategy to realize risk evolution simulation; According to the risk probability matrix, a multi-objective optimization algorithm is used to generate an anti-interference decision plan set, and the optimal cost control strategy is output after being verified by digital twin simulation. Among them, the optimal cost control strategy synchronously drives the three-dimensional risk heat map and the decision path deduction animation in the visualization interface.

2. The method according to claim 1, wherein The step of, according to the heterogeneous cost data in the engineering planning, design, construction, and completion stages, through the data fusion engine driven by federated learning, performing cross-stage feature alignment to generate a spatio-temporally correlated structured cost data cube, where the data fusion engine uses a differential privacy protection mechanism to eliminate data islands, includes: According to the BIM model parameters in the engineering planning stage, the bill of quantities in the design stage, the progress log in the construction stage, and the settlement documents in the completion stage, field semantic mapping is performed through a heterogeneous data adapter to generate a four-dimensional tensor containing timestamps, spatial coordinates, and cost attributes, and a standardized feature vector set is output; Based on the feature vector set, using a cross-stage alignment algorithm under the federated learning framework, the local model parameters of each participant are optimized through projected gradient descent, and distributed feature space alignment is completed under differential privacy constraints to generate a cross-stage joint feature matrix; The joint feature matrix is input into the spatio-temporal correlation engine, and the coupling relationship between project progress-cost consumption-geographical distribution is captured through a three-dimensional convolution kernel to construct a six-dimensional cost data cube containing time series correlation, spatial topology correlation, and cost conduction correlation, and a structured data cube with a privacy protection label is output.

3. The method according to claim 2, characterized in that, The step of, based on the structured cost data cube, combined with the market price fluctuation trend and the construction progress change factor, constructing a spatio-temporal graph neural network prediction model to output a dynamic cost prediction curve and deviation-sensitive nodes, where the spatio-temporal graph neural network prediction model captures the associated impact of potential cost overrun risk factors through an attention mechanism, includes: Extract cost fluctuation features through multi-scale sliding windows based on the time series cost data in the structured data cube, and construct a dynamic embedding vector in the time dimension by combining with the external market price index; Based on the construction progress change factor, model the spatial dependence relationship between construction nodes through the graph attention mechanism to generate a spatial topology weight matrix; Input the time dynamic embedding vector and the spatial topology weight matrix into the spatio-temporal graph convolutional layer, fuse spatio-temporal features through the bidirectional gated recurrent unit, and output the correlation influence intensity distribution map of potential overspending risk factors; According to the risk influence intensity distribution map, use the dynamic threshold segmentation algorithm to identify deviation-sensitive nodes and generate a dynamic cost prediction curve with confidence annotations; 4. The method according to claim 3, wherein According to the dynamic cost prediction curve, use the blockchain-enabled BIM / CIM collaborative platform to optimize multi-party task allocation and generate a collaborative instruction set encoded by smart contracts. Among them, the BIM / CIM collaborative platform realizes privacy-protected data synchronization among the owner, the designer, and the constructor through zero-knowledge proof, including: According to the dynamic cost prediction curve, homomorphically encrypt privacy data such as the owner's budget constraint, the designer's drawing changes, and the constructor's resource scheduling through the zero-knowledge proof protocol to generate a verifiable ciphertext data set; Based on the ciphertext data set, decompose collaborative tasks using multi-party secure computing, and solve the optimal task allocation scheme under multi-party interest balance through the Hungarian algorithm to output a task allocation matrix with weight constraints; Encode the task allocation matrix into a smart contract, define the contract trigger condition through an event-driven state machine, and generate a smart contract template containing an execution logic chain; Deploy lightweight blockchain nodes on the BIM / CIM collaborative platform, and achieve privacy-protected data synchronization across participating parties through the sharding consensus mechanism to generate a collaborative instruction hash chain with timestamps; According to the real-time engineering data stream, verify the authenticity of off-chain data through an oracle, dynamically update the execution status of the smart contract, and output a traceable collaborative instruction set.

5. The method according to claim 4, wherein Based on the collaborative instruction set and the real-time engineering data stream, parse potential risk paths through a dynamic knowledge graph engine, and output a risk probability matrix and an early warning signal. Among them, the dynamic knowledge graph engine fuses the industry knowledge base and the reinforcement learning strategy to realize risk evolution simulation, including: According to the collaborative instruction set, extract risk semantic fragments from engineering contract terms and construction specification documents through a domain adaptive pre-training model to generate risk knowledge triples with probability annotations; Based on the risk knowledge triples, combined with abnormal events in the real-time engineering data stream, construct a risk propagation path topology graph through a temporal graph convolutional network; Input the risk propagation path into the reinforcement learning environment, simulate the risk evolution process through a double deep Q network, and generate a risk probability transition matrix containing cascade effect intensity; Based on the risk probability transition matrix, use the backpropagation importance sampling algorithm to identify key risk paths and output an early warning signal with time window constraints; Compress the risk model parameters through knowledge distillation technology, deploy them to edge computing nodes to realize real-time risk monitoring, and generate a lightweight risk evolution simulator.

6. The method according to claim 5, wherein According to the risk probability matrix, a multi-objective optimization algorithm is used to generate a set of anti-interference decision-making solutions, and the optimal cost control strategy is output after being verified by digital twin simulation. Among them, the optimal cost control strategy synchronously drives the three-dimensional risk heat map and the decision path deduction animation in the visualization interface, including: According to the risk probability matrix, the NSGA-III algorithm is used to synchronously optimize the three objectives of the cost control rate, the engineering quality compliance rate, and the risk aversion coefficient, and an initial Pareto front containing 300-500 groups of solutions is generated; The Pareto solution set is input into the digital twin engine, and the physical engine is used to simulate the construction process under different decision-making solutions, and a multi-dimensional simulation data set containing cost deviation, quality defects, and risk outbreak points is generated; Based on the simulation data set, the fuzzy comprehensive evaluation method is used to calculate the fitness scores of each solution, and the gradient boosting decision tree is used to screen the optimal cost control strategy; The optimal strategy is fused with the BIM model, the three-dimensional risk heat map is rendered through ray tracing technology, and the decision path deduction animation is generated, and the visual decision support interface is output.

7. The method according to claim 3, characterized in that, Based on the construction progress change factor, the spatial dependence relationship between construction nodes is modeled through the graph attention mechanism to generate a spatial topology weight matrix, including: According to the node delay data in the construction progress change factor, the spatio-temporal correlation features between construction nodes are extracted through the spatio-temporal graph convolutional layer, and an initial dependence graph with time stamps is generated; Based on the initial dependence graph, combined with the geographical coordinate data in the BIM model, the multi-modal attention mechanism is used to calculate the spatial correlation weights between nodes, where the three-channel features of the physical distance between nodes, the construction process dependence degree, and the resource flow direction are fused, and the multi-modal attention score matrix is output; The multi-modal attention score matrix is input into the dynamic residual connection network, and the implicit dependence relationship in the historical construction log is fused through the gated recurrent unit to generate a spatio-temporal enhanced node dependence intensity distribution map; Based on the dependence intensity distribution map, a graph pruning algorithm with differential privacy protection is used to remove the weak dependence edges with a confidence level lower than the threshold, and a spatial topology weight matrix with a hierarchical structure is generated through spectral clustering, and the spatial topology of construction nodes with security protection labels is output.

8. The method according to claim 5, characterized in that, Based on the ciphertext data set, the multi-party secure calculation is used to decompose the collaborative task, and the optimal task allocation scheme under the multi-party interest balance is solved through the Hungarian algorithm, and the task allocation matrix with weight constraints is output, including: According to the ciphertext data set encrypted homomorphically, the collaborative task is decomposed through the secure multi-party calculation protocol to generate a task demand vector with privacy protection, where the secret sharing technology is used to split the task parameters into multiple shadow shares; Based on the task demand vector, a multi-party interest game model is constructed, and the initial task allocation scheme is calculated through the Nash equilibrium solution algorithm, where the model constraint conditions include the upper limit of the owner's budget, the resource capacity of the construction party, and the change tolerance of the design party; The initial allocation scheme is input into the Hungarian algorithm, and a dynamic weight adjustment strategy is introduced to optimize the conflict task matching, where the dynamic weight is adaptively updated according to the real-time supply chain risk index, and an intermediate task allocation matrix with elastic constraints is output; Verify the fairness of the intermediate matrix for task assignment through zero-knowledge proof, generate a verifiable assignment proof using the zk-SNARK protocol, and finally output a weighted constraint task assignment matrix that is tamper-proof and meets the multi-party interest balance.

9. The method according to claim 6, wherein Based on the simulation dataset, use the fuzzy comprehensive evaluation method to calculate the fitness scores of each scheme, and screen the optimal cost control strategy through the gradient boosting decision tree, including: According to the multi-dimensional engineering parameters in the simulation dataset, through a dynamic weight generator based on Monte Carlo sampling, perform adaptive weight allocation for the three objectives of cost deviation rate, quality defect index, and risk outbreak probability, and generate a dynamic evaluation weight vector with a confidence interval; Based on the dynamic evaluation weight vector, use the fuzzy comprehensive evaluation algorithm to calculate the fitness scores of the schemes. The fuzzy comprehensive evaluation algorithm introduces a time decay factor to correct the contribution degree of historical simulation data, and maps the high-dimensional non-linear pattern through a radial basis function kernel, and outputs a fitness score matrix with a timeliness label; According to the fitness score matrix, construct a multi-modal feature fusion channel, extract the hyperplane curvature feature, decision variable correlation feature, and simulation process stability feature of the Pareto solution set, and generate an enhanced feature vector set with a unified dimension through a heterogeneous feature embedding network; Input the enhanced feature vector set into the gradient boosting decision tree model, use the multi-objective splitting criterion to optimize the growth direction of the tree structure, and strengthen the search ability of the high-dimensional non-convex solution space through the cost-sensitive learning strategy, and output the global ranking result of the optimal cost control strategy; According to the global ranking result, combined with the physical constraint verification of the digital twin engine, iteratively optimize the strategy parameters through the backpropagation correction mechanism, and generate an optimal cost control strategy execution sequence that meets the construction feasibility boundary.

10. A multi-dimensional intelligent management system for the whole-process cost, characterized in that, The system includes: An alignment module, which is used to generate a spatio-temporal correlated structured cost data cube through cross-stage feature alignment by a data fusion engine driven by federated learning according to the heterogeneous cost data in each stage of project planning, design, construction, and completion. Among them, the data fusion engine uses a differential privacy protection mechanism to eliminate data islands; A construction module, which is used to construct a spatio-temporal graph neural network prediction model based on the structured cost data cube, combined with the market price fluctuation trend and the construction progress change factor, and output a dynamic cost prediction curve and deviation-sensitive nodes. Among them, the spatio-temporal graph neural network prediction model captures the associated impact of potential cost overrun risk factors through an attention mechanism; An optimization module, which is used to perform multi-party task assignment optimization using a blockchain-enabled BIM / CIM collaboration platform according to the dynamic cost prediction curve, and generate a collaborative instruction set encoded by a smart contract. Among them, the BIM / CIM collaboration platform realizes privacy-protected data synchronization among the owner, designer, and constructor through zero-knowledge proof; An analysis module, which is used to analyze potential risk paths based on the collaborative instruction set and real-time project data streams through a dynamic knowledge graph engine, and output a risk probability matrix and an early warning signal. Among them, the dynamic knowledge graph engine fuses the industry knowledge base and the reinforcement learning strategy to realize risk evolution simulation; An output module, configured to generate a set of anti-interference decision-making schemes by using a multi-objective optimization algorithm according to the risk probability matrix, and output an optimal cost control strategy after verification through digital twin simulation, wherein the optimal cost control strategy synchronously drives a three-dimensional risk heat map and a decision path deduction animation in a visualization interface.

Images