Engineering project full-cycle cost intelligent accounting system based on BIM technology

Abstract

The invention discloses an engineering project full-cycle cost intelligent accounting system based on a BIM technology, and relates to the technical field of constructional engineering informatization, and the system comprises a BIM visualization platform which is in communication connection with the following modules: a dynamic optimization scheduling module which is used for fusing a target engineering project BIM model and a real-time digital twinborn body, and introducing a bat algorithm to carry out dynamic optimization on construction resource nodes influenced by risks. According to the method, a dynamic resource topology network is constructed by fusing a BIM model and real-time digital twinborn data, accurate tracking and state perception of construction resources are realized, on the basis, a bat algorithm and a graph neural network are introduced to perform dynamic optimization and conflict coordination on resource configuration in a multi-risk concurrent scene, and the resource allocation efficiency is improved. And a rescheduling scheme and an optimal cost increment which conform to actual engineering logic are generated, the limitation of traditional static risk conduction simulation is broken through, and the adaptability of resource scheduling and the refinement level of cost control are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of information technology in building engineering, specifically to an intelligent cost accounting system for the entire lifecycle of engineering projects based on BIM technology. Background Technology

[0002] With the rapid advancement of information technology and the deepening of digital transformation in the construction industry, BIM technology provides an integrated visualization platform for building projects, enabling the effective integration of data from different stages such as design, construction, operation, and maintenance. The intelligent cost accounting throughout the entire project lifecycle not only improves the accuracy of cost budgeting but also reduces cost risks through real-time data analysis and feedback during project implementation. The application of BIM technology enables cost accounting to be automated and intelligent, reducing errors caused by manual intervention and traditional manual calculations.

[0003] For example, the engineering cost data visualization monitoring method and system in Chinese Patent Publication No. CN120598373A provides engineering managers with a decision-making view that combines overall trend control with the ability to trace the source of emergencies. This fundamentally promotes the shift of engineering cost management from post-event accounting to pre-event control, and significantly improves the cost risk resistance of complex engineering projects.

[0004] In the process of cost accounting for engineering projects, although existing systems have a certain foundation in risk transmission simulation and cost prediction, when faced with multiple concurrent risk events, they mostly use static graph traversal methods to deduce the impact path. This lacks the ability to dynamically schedule and coordinate conflicts among construction resources, resulting in overly rigid cost increments that fail to reflect the cost control potential achievable through resource reallocation in actual projects. Secondly, although resource optimization simulations are performed for single risk events, the output cost adjustment value remains an isolated numerical result, failing to be deeply integrated with historical full-cycle cost data and experience extraction. This makes it difficult for the cost accounting process to form a continuously accumulating and self-evolving knowledge system, and it is difficult to proactively identify abnormal cost patterns and recurring risks across projects. This limits the intelligent evolution of the accounting system and the long-term improvement of its decision support capabilities. Therefore, this paper proposes an intelligent cost accounting system for the entire lifecycle of engineering projects based on BIM technology to solve the above-mentioned problems. Summary of the Invention

[0005] To solve the above technical problems, the present invention is implemented through the following technical solution: an intelligent cost accounting system for the entire life cycle of engineering projects based on BIM technology, including a BIM visualization platform, wherein the BIM visualization platform is communicatively connected to the following modules: The dynamic optimization scheduling module is used to integrate the BIM model of the target project with the real-time digital twin. It introduces the bat algorithm to dynamically optimize the construction resource nodes affected by risks. Based on the BIM component association relationship and twin status data, it optimizes and adjusts the configuration path and scheduling logic of manpower, machinery and materials in real time to obtain the optimal cost increment after rescheduling, thereby improving the resource response speed and the scenario adaptability of cost control. The knowledge graph construction module is used to semantically associate and temporally embed optimal cost increments with historical project data, build a cross-project cost knowledge graph, and form a structured and queryable cost knowledge base, providing a data foundation for subsequent intelligent analysis. The temporal graph convolutional mining module extracts spatiotemporal features from the cost knowledge graph across projects based on temporal graph convolutional networks, identifies cost fluctuation patterns and risk transmission paths, discovers abnormal cost patterns across cycles and projects, and performs trend prediction. The intelligent evolution analysis module is used to dynamically integrate the optimized cost data of the current project with the historical cost knowledge graph. Through an incremental learning mechanism, it updates the graph structure and node relationships to form a continuously evolving and self-improving cost knowledge system. The early warning decision support module uses graph neural networks and rule engines to detect abnormal cost nodes and recurring risk patterns, triggering early warning signals in real time. It combines the BIM model to locate risk sources, provides targeted resource optimization suggestions, and binds the early warning information and optimization suggestions to the BIM model for presentation on the BIM visualization platform. This allows users to intuitively understand the cost impact and resource scheduling effects, improving decision-making efficiency and risk response capabilities.

[0006] Preferably, the dynamic optimization scheduling module includes a BIM-twin fusion resource modeling unit and a conflict coordination unit; The BIM-twin fusion resource modeling unit, based on the component attributes of the target project's BIM model and the twin status data of the real-time digital twin, constructs an integrated resource-task-space association model, dynamically binding construction resources to BIM component nodes to form a calculable and traceable resource topology network, thereby identifying construction resource nodes affected by risks. The conflict coordination unit, relying on the resource-task-space integrated association model, uses the bat algorithm to perform multi-objective dynamic optimization of construction resource nodes affected by risks. Combined with the real-time progress and resource status feedback from the real-time digital twin, it generates and iterates resource allocation schemes. It also uses graph neural networks to model conflicts between resources, realizes automatic conflict identification and collaborative resolution, and outputs the optimal cost increment after rescheduling.

[0007] Preferably, the BIM-twin fusion resource modeling unit is configured to perform the following steps: The system analyzes the component attributes in the BIM model of the target project and extracts the twin status data of the real-time digital twin, including the component construction status, resource occupancy and progress completion rate, to achieve real-time fusion of BIM static data and twin dynamic data, ensuring data consistency and timeliness. Based on the component attributes and twin state data, an integrated resource-task-space association model is constructed, which dynamically binds construction resources with corresponding BIM component nodes to form a traceable and computable resource topology network, providing a structured foundation for risk transmission simulation. Based on the resource topology network, construction resource nodes that are directly affected or indirectly dependent on risk events are identified, forming a set of construction resource nodes. A structured input containing resource type, quantity, location, and constraints is generated, providing an accurate set of nodes and their relationships for subsequent dynamic scheduling. The set of resource nodes affected by risks is accurately identified, providing structured input conditions for subsequent scheduling optimization.

[0008] Preferably, the conflict coordination unit is configured to perform the following steps: Based on the resource topology network and the set of construction resource nodes affected by risks, a multi-objective optimization model is constructed. The optimization objective is to minimize the cost increment and resource idle time. The optimization objective function is obtained, and time, space and resource capacity constraints are set to achieve the global optimal allocation of construction resources under multi-risk scenarios and ensure that the scheduling scheme meets the actual engineering constraints. The Bat Algorithm is used to dynamically optimize resource allocation schemes. Combined with progress and status data fed back by real-time digital twins, multiple scheduling schemes are iteratively generated and evaluated. Graph neural networks are used to model the conflict relationships between resource nodes, which improves the convergence speed and accuracy of scheduling optimization and effectively avoids secondary delays caused by resource conflicts. Based on the conflict identification results and the optimization objective function, the rescheduled resource allocation scheme and its corresponding optimal cost increment are output. The scheme includes the adjusted task order, resource allocation path and conflict resolution strategy, forming an executable and quantifiable scheduling instruction set, which significantly reduces the net cost increase caused by risk events.

[0009] Preferably, the knowledge graph construction module includes a semantic association embedding unit and a temporal structure modeling unit; The semantic association embedding unit is used to extract the component attributes and cost node semantic features of the BIM model of the target project, perform semantic matching and vectorization embedding of the optimal cost increment after rescheduling and historical project nodes, establish a traceable association mapping across projects and stages, analyze project progress time sequence information, enhance the interpretability and reusability of cost data, improve the reusability of cross-project cost experience, and support rapid matching of similar historical events. The time-series structure modeling unit is used to combine project progress time-series information to construct a time-series graph structure of cost nodes, form a cost knowledge graph across projects, capture the time-series dependencies and stage characteristics in the cost evolution process, provide time-series-aware data support for full-cycle pattern mining, enhance the ability to identify long-term dependencies on cost fluctuations, and improve the accuracy of trend prediction.

[0010] Preferably, the semantic association embedding unit is configured to perform the following steps: Extract the semantic attributes and cost node feature descriptions of each component in the BIM model of the target project, and obtain similar node data from historical projects to build a cross-project semantic feature library. Effectively integrate multi-source heterogeneous data to form a standardized and reusable cost semantic library to support accurate matching and knowledge reuse. A pre-trained language model is used to vectorize the semantic features of cross-projects. The optimal cost increment after rescheduling is semantically matched and similarity is calculated with historical project nodes to establish a traceable semantic association mapping, realize intelligent association and experience tracing of cross-project cost events, and improve the predictive ability and decision-making basis of cost control. Based on the semantic association mapping results, the current project cost node vector is embedded into a unified semantic space to perform interpretable alignment and knowledge reuse of cost data across projects and stages, build a unified semantic understanding framework, support multi-dimensional cost knowledge retrieval and intelligent reuse, and promote the continuous evolution of the accounting system.

[0011] Preferably, the temporal structure modeling unit is configured to perform the following steps: Obtain the time sequence information of each cost node throughout the project lifecycle, including the occurrence time, duration, and progress stage labels, and construct a cost node sequence with time as the axis to ensure that cost events have traceability and structured representation capabilities in the time dimension, laying the time sequence foundation for subsequent graph construction; Based on the cost node sequence and semantic embedding results, a graph structure with temporal edges and semantic edges is established to form a cost knowledge graph that spans projects and stages. Nodes represent cost events, and edges represent temporal dependencies or semantic associations, enabling the association and fusion of cost data across multiple projects and stages, forming a knowledge network with both semantic and temporal dimensions. A temporal attention mechanism is introduced into the cost knowledge graph to capture the stage characteristics and long-term dependencies in the cost evolution process. This provides temporal-aware data support for subsequent spatiotemporal feature extraction and pattern mining, enhances the model's ability to model long-term temporal dependencies, and improves the accuracy of risk transmission path identification and trend prediction.

[0012] Preferably, the time-series graph convolutional mining module is configured to perform the following steps: Load a cross-project cost knowledge graph, extract the feature vectors of each node in the graph and their temporal connection relationships, construct a spatiotemporal graph data structure, realize the spatiotemporal structured unified expression of multi-source heterogeneous cost data, and provide high-quality input for deep learning; Using temporal graph convolutional networks, multi-layer feature aggregation and propagation are performed on spatiotemporal graph data structures to capture spatiotemporal dependencies between nodes, extract implicit spatiotemporal features of cost fluctuation patterns and risk transmission paths, deeply explore the spatiotemporal correlation patterns in cost evolution, and identify implicit risk transmission links. Based on the extracted implicit spatiotemporal features, abnormal cost patterns across cycles and projects are identified, and future cost trends are predicted. The system outputs a risk transmission path diagram and a set of abnormal nodes, enabling proactive discovery and trend prediction of abnormal costs, and supporting risk visualization and decision-making intervention.

[0013] Preferably, the intelligent evolution analysis module is configured to perform the following steps: By combining the risk transmission path graph and abnormal node set output by the time series graph convolutional mining module, the optimized cost data of the current project and its associated semantic embedding vectors are dynamically aligned and fused with the historical cost knowledge graph. The node attributes and time series edges in the graph are updated to ensure that the historical knowledge graph data is continuously updated, so as to achieve seamless accumulation and instant reuse of cross-project experience. An incremental learning mechanism is used to adaptively adjust the graph structure, including inserting new nodes, updating edge weights, and reorganizing the structure, to ensure that the graph continues to evolve over time, thereby achieving adaptive evolution of the knowledge graph and continuously improving the accuracy and timeliness of risk pattern recognition and prediction. Based on the updated cost knowledge graph, the strength of associations and temporal dependencies between nodes are recalculated to form a self-improving and continuously accumulating cost knowledge system. This provides real-time knowledge support for subsequent early warning and decision-making, and constructs a dynamic knowledge base with millisecond-level response, providing accurate and real-time panoramic data support for intelligent early warning and decision-making.

[0014] Preferably, the early warning decision support module is configured to perform the following steps: Based on graph neural networks and a pre-defined rule engine, anomaly detection is performed on nodes in the cost knowledge graph to identify recurring risk patterns and cost mutation nodes, triggering early warning signals in real time. This transforms risk perception from manual inspection to intelligent real-time monitoring, reducing early warning response time to the second level. By combining the spatial location of abnormal nodes and related components in the BIM model, targeted resource optimization suggestions are generated, including resource reallocation, schedule adjustment, process optimization and risk mitigation strategies. The abstract cost risk is accurately mapped to three-dimensional spatial components, making the optimization suggestions intuitive and operable. By dynamically binding early warning information and resource optimization suggestions to the BIM model, and highlighting and interactively presenting them on the BIM visualization platform, users can intuitively understand the cost impact and scheduling effect, build a visual closed loop of data-model-decision, and significantly improve managers' situational understanding and decision-making efficiency.

[0015] This invention provides an intelligent cost accounting system for the entire lifecycle of engineering projects based on BIM technology. It has the following beneficial effects: (I) This intelligent cost accounting system for the entire life cycle of engineering projects based on BIM technology constructs a dynamic resource topology network by integrating BIM models and real-time digital twin data, thereby achieving accurate tracking and status perception of construction resources. On this basis, the system introduces the Bat algorithm and graph neural network to dynamically optimize resource allocation and coordinate conflicts in scenarios with multiple concurrent risks, generating rescheduling schemes and optimal cost increments that fit the actual engineering logic. This breaks through the limitations of traditional static risk transmission simulation and significantly improves the adaptability of resource scheduling and the level of precision in cost control.

[0016] (II) This intelligent cost accounting system for the entire lifecycle of engineering projects based on BIM technology constructs a cost knowledge graph across projects and stages. By using semantic association embedding and temporal structure modeling, it deeply integrates optimized cost increments with massive historical project data. This not only forms a structured and queryable knowledge base, but also uses temporal graph convolutional networks to deeply mine the spatiotemporal patterns of cost evolution and risk transmission paths, achieving a leap from isolated event analysis to system pattern recognition, and providing a solid knowledge foundation for cost prediction and risk tracing.

[0017] (III) This intelligent cost accounting system for the entire life cycle of engineering projects based on BIM technology detects abnormal patterns through a dual-channel graph neural network and rule engine, locates risk sources based on a real-time knowledge base and BIM model, and automatically generates resource optimization suggestions. It seamlessly connects anomaly detection, spatial positioning, and strategy generation, and realizes rapid automatic response from risk perception to response strategy suggestions, which greatly improves the timeliness and pertinence of management decisions. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the workflow of an intelligent cost accounting system for the entire life cycle of engineering projects based on BIM technology, as described in this invention. Figure 2 This is a data flow diagram of an intelligent cost accounting system for the entire lifecycle of engineering projects based on BIM technology, as described in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: an intelligent cost accounting system for the entire lifecycle of engineering projects based on BIM technology, including a BIM visualization platform, which has the following communication modules: The dynamic optimization scheduling module is used to integrate the BIM model of the target project with the real-time digital twin. It introduces the bat algorithm to dynamically optimize the construction resource nodes affected by risks. Based on the BIM component association relationship and twin status data, it optimizes and adjusts the configuration path and scheduling logic of manpower, machinery and materials in real time to obtain the optimal cost increment after rescheduling, and improves the resource response speed and cost control scenario adaptability. The dynamic optimization scheduling module includes a BIM-twin fusion resource modeling unit and a conflict coordination unit. The BIM-twin fusion resource modeling unit, based on the component attributes of the target project's BIM model and the twin status data of the real-time digital twin, constructs an integrated resource-task-space association model. It dynamically binds construction resources to BIM component nodes, forming a calculable and traceable resource topology network. This identifies construction resource nodes affected by risks, providing precise input conditions and constraints for subsequent optimization. It achieves real-time fusion of static BIM data and dynamic construction status, constructing a high-precision resource topology network. This provides a reliable data foundation for risk transmission simulation, analyzes the component attributes in the target project's BIM model, and extracts the twin status data of the real-time digital twin, including component construction status, resource occupancy, and progress. To achieve high efficiency, real-time fusion of BIM static data and BIM twin dynamic data is realized, ensuring data consistency and timeliness. Based on component attributes and BIM twin status data, an integrated resource-task-space association model is constructed, dynamically binding construction resources with corresponding BIM component nodes to form a traceable and computable resource topology network. This provides a structured foundation for risk transmission simulation. Based on the resource topology network, construction resource nodes directly affected or indirectly dependent on risk events are identified, forming a set of construction resource nodes. A structured input containing resource type, quantity, location, and constraints is generated, providing accurate node sets and associations for subsequent dynamic scheduling. The set of resource nodes affected by risks is accurately identified, providing structured input conditions for subsequent scheduling optimization. The specific work content is as follows: During the BIM model analysis phase of the target project, by reading the IFC format BIM model file of the target project, component attribute data containing geometric attributes, material parameters, and bill of quantities is extracted. Simultaneously, twin status data is obtained from the real-time digital twin platform through a real-time data interface, including component construction status (component installation completion status), resource occupancy, and progress completion rate (calculated based on the percentage of planned and actual completed work). A unified coding system is used to associate and map BIM component IDs with twin status data items, forming a structured record table containing component number, status code, timestamp, and resource identifier fields. Based on the analyzed component attributes and twin status data, a resource-task-space integrated association model is constructed using a graph structure. Nodes include BIM components (including spatial coordinates, work...). The system comprises a resource topology network consisting of components, materials, construction tasks (including duration and process type), and resource entities (including resource code, type, quantity, and unit labor cost). Edge relationships include component-task association (execution relationship), task-resource dependency (occupancy relationship), and spatial adjacency (construction area constraints). Resource instances are dynamically bound to corresponding component nodes using preset mapping rules, establishing a time-space resource topology network for resource flow paths. Based on this network, a dependency transitive algorithm is used to identify the set of resource nodes affected by risk events. First, directly affected nodes are located. Then, a breadth-first traversal is performed along task dependency edges and resource association edges to identify indirectly dependent nodes. Finally, a structured resource node set is output, with each record containing resource type, quantity requirements, spatial location (construction zone number), and constraints, ensuring that the resource allocation scheme conforms to actual engineering constraints and logical relationships. The conflict coordination unit, relying on an integrated resource-task-space correlation model, employs the Bat Algorithm to dynamically optimize construction resource nodes affected by risks across multiple objectives. Combined with real-time progress and resource status feedback from a digital twin, it generates and iterates resource allocation schemes. Furthermore, it models inter-resource conflicts using graph neural networks, enabling automatic conflict identification and collaborative resolution. The unit outputs the optimal cost increment after rescheduling. This combination of intelligent scheduling and conflict prediction significantly improves resource allocation efficiency and reduces the actual impact of unforeseen risks on project costs. Based on the resource topology network and the set of construction resource nodes affected by risks, a multi-objective optimization model is constructed, minimizing cost increments and resource idle time as optimization objectives. This yields the optimization objective function, and time, space, and resource tolerance are set. To address quantity constraints, the system achieves globally optimal allocation of construction resources under multiple risk scenarios, ensuring that scheduling schemes conform to actual engineering constraints. The Bat Algorithm is used to dynamically optimize resource allocation schemes. Combined with real-time progress and status data from digital twins, multiple scheduling schemes are iteratively generated and evaluated. Furthermore, graph neural networks are used to model conflict relationships between resource nodes, improving the convergence speed and accuracy of scheduling optimization and effectively avoiding secondary delays caused by resource conflicts. Based on conflict identification results and the optimization objective function, the system outputs a rescheduled resource allocation scheme and its corresponding optimal cost increment. This scheme includes adjusted task order, resource allocation paths, and conflict resolution strategies, forming an executable and quantifiable set of scheduling instructions, significantly reducing the net cost increase caused by risk events. The specific work involves: obtaining the set of risk-affected resource nodes generated from the resource topology network, and constructing an accurate multi-objective optimization model to guide resource rescheduling. This model's core optimization objectives are minimizing the total cost increment and minimizing the total idle time of critical resources. The calculation of the total cost increment covers direct losses caused by risks (rush work fees, idle labor costs, material premiums, etc.), indirect derived costs (amortization of management expenses), and additional costs incurred by resource rescheduling (such as secondary equipment transfer fees). The total idle time of critical resources is calculated for high-value, low-substitutability bottleneck resources (such as large tower cranes and special work teams). The model constraints are based on the work... The actual process settings include: time constraints such as the earliest / latest start time of tasks and the immediate / infinite relationships determined by the process logic; spatial constraints based on the spatial coordinates of the BIM model and the safe operating radius, limiting the number of conflicting devices that can coexist within the same construction zone at the same time; and resource capacity constraints based on enterprise quotas and actual on-site supply capacity, setting a maximum daily available quantity limit for each type of resource. All parameters are directly extracted from the project schedule, BIM model attribute library, and enterprise resource database or derived through standardized conversion to ensure that the optimization problem closely matches the real construction scenario. An improved bat algorithm is used to solve the multi-objective optimization model. During algorithm initialization, each bat represents a... A complete resource allocation and task sequencing scheme is proposed, with its position vector encoding the planned start time offset of each affected task and the resource combination allocated to it. During the algorithm iteration process, on the one hand, it receives feedback from the real-time digital twin on the current progress completion rate and real-time resource occupancy status to dynamically update the feasible region boundary of the solution space; on the other hand, a pre-trained graph neural network submodule is introduced. This graph neural network submodule takes the resource topology network as input to identify and quantify the potential resource conflict relationships in different bat individual schemes. The conflict intensity score output by the graph neural network submodule is used as a penalty term and incorporated into the fitness function of the bat algorithm. The sonar pulse frequency and response are then compared. The iterative update mechanism of the simulation searches the solution space, continuously generating and evaluating new solutions, gradually approaching the Pareto front that can balance cost and efficiency while satisfying constraints. After multiple rounds of iterative convergence, the Bat Algorithm outputs one or more non-dominated optimal solutions, from which the solution with the highest comprehensive fitness is selected as the final rescheduling instruction. This solution is a structured set of executable instructions, specifically including: 1) an adjusted task timing network, which clarifies the new start / end time of each affected task and its logical relationship; 2) a detailed resource allocation path table, which lists the configuration objects, quantities, construction zones, and handover nodes of each resource on a daily basis during the rescheduling cycle.3) An explicit list of conflict resolution strategies: For specific spatial conflicts identified by the graph neural network submodule, specific measures are given, and the optimal cost increment bound to the solution is output. The optimal cost increment has deducted the potential losses recovered through optimized scheduling. The entire output will be packaged in a machine-readable standardized format and seamlessly connected to the project management system and BIM visualization platform. The formula for calculating the optimal cost increment is as follows:

[0021] In the formula: This represents the optimal cost increment, which is the net cost change value that is ultimately output and bound to the rescheduling scheme; This represents the original cost increment, which is the estimated total cost increase directly and indirectly caused by the risk event without resource rescheduling optimization. This refers to the cost of recovering losses, that is, the potential cost losses avoided or reduced through resource rescheduling optimization; Indicates the first Direct loss costs include cost items directly triggered by risk events, such as rush work fees, idle labor costs, material premiums, and equipment lease default penalties. This represents the total number of direct loss costs. Indicates the first Indirect costs include on-site management fees, headquarters management fees, and financial expenses that are amortized proportionally due to the extended construction period; This represents the total number of indirectly derived costs. To indicate the first Additional costs arising from resource reallocation include secondary equipment transfer fees, temporary rental price differences, personnel dispatch fees, and costs for purchasing or renting small machinery, which are new costs incurred due to adjustments in the resource allocation plan. The total number of items that incur additional costs for resource rescheduling; This indicates the total time period covered by the rescheduling scheme, corresponding to the scheduling period in the resource allocation path table; This represents the total number of tasks affected by the risk, consistent with the number of tasks adjusted in the task timing network; Indicates the first The penalty costs or opportunity losses incurred for each day of delay in a task include contractually agreed-upon penalties for project delays and the discounted value of owner claims risks. Let be a function relating a 0 / 1 decision variable, representing the result of the rescheduling scheme. The task in the Whether the project schedule has been shortened compared to the original plan; if so, the value is positive, indicating the equivalent amount of the number of days of delay recovered. This refers to a set of critical resources, namely bottleneck resources that are high in value and have low substitutability, such as large tower cranes and special work teams. Indicates the first The unit time idle cost of key resources, including depreciation, rent, basic labor costs and other costs that must be borne even when idle; This indicates that under the rescheduling scheme, the first... Class of key resources in the first The reduction in idle time compared to the original scheduling scheme; Indicates the first Class of key resources in the first The maximum reduction in idle time per day is constrained by process logic and task dependencies. It represents the effective coefficient for recovering losses, used to convert the theoretical recoverable value into actual realizable economic value, taking into account factors such as execution losses and risk reserves; The knowledge graph construction module is used to semantically associate and temporally embed the optimal cost increment with historical project data, build a cross-project cost knowledge graph, form a structured and queryable cost knowledge base, and provide a data foundation for subsequent intelligent analysis. The knowledge graph construction module includes a semantic association embedding unit and a temporal structure modeling unit. The semantic association embedding unit is used to extract the semantic features of component attributes and cost nodes from the BIM model of the target project. It performs semantic matching and vectorized embedding of the optimal cost increment after rescheduling with historical project nodes, establishing a traceable association mapping across projects and phases. This analyzes project progress timeline information, enhances the interpretability and reusability of cost data, improves the reusability of cross-project cost experience, supports rapid matching of similar historical events, extracts the semantic attributes and feature descriptions of cost nodes from each component in the BIM model of the target project, obtains similar node data from historical projects, constructs a cross-project semantic feature library, effectively integrates multi-source heterogeneous data, and forms a standardized and reusable [database / system]. By reusing the cost semantic library to support accurate matching and knowledge reuse, a pre-trained language model is used to vectorize the semantic features of cross-projects. The optimal cost increment after rescheduling is semantically matched and similarity is calculated with historical project nodes to establish a traceable semantic association mapping. This enables intelligent association and experience tracing of cost events across projects, improving the predictive ability and decision-making basis of cost control. Based on the semantic association mapping results, the current project cost node vector is embedded into a unified semantic space to perform interpretable alignment and knowledge reuse of cost data across projects and stages. A unified semantic understanding framework is constructed to support multi-dimensional cost knowledge retrieval and intelligent reuse, driving the continuous evolution of the accounting system. The specific work content is as follows: In the semantic feature library construction phase, the semantic attributes of the BIM model components of the target project are extracted using the Industrial Foundation Class (IFC) data pattern parser. The semantic attributes follow the "Classification and Coding Standard for Building Information Models" (GB / T51269). The key fields of the extracted semantic attributes include the component classification code, functional description, and material specification grade. At the same time, standardized feature descriptions of cost nodes are extracted from the project cost breakdown structure (CBS), including cost type (labor cost / material cost / machinery cost), bill of quantities item code (according to the "Construction Engineering Quantity List Pricing Specification" GB50500), unit of measurement, and comprehensive unit price composition description. Historical project data is extracted from the enterprise project database using ETL tools. The selection rules are: projects with a similarity of project type >75%, the same structural form, and a completion time within 5 years. Node data items including component classification code, cost feature vector, and final settlement unit price are extracted. Additionally, a method based on term frequency-inverse document frequency (TF-IDF) weighting and the Word2Vec word vector model is used to generate a 300-dimensional vector of the descriptive text of the component semantic attributes. The system underwent processing to construct a cross-project semantic feature library containing over 100,000 standardized records. A BERT-Base (768-dimensional) pre-trained language model, finely tuned on a construction engineering corpus (containing 3 million technical specifications, construction plans, and cost indicators), was used as the encoder for semantic vectorization and matching. This BERT pre-trained language model encodes the text descriptions in the cross-project semantic feature library into 768-dimensional dense vectors. For the optimal cost increment output after rescheduling of the current project, the associated set of component nodes and their cost adjustment descriptions are output. The same encoder is used to generate the semantic vector of the current node. An improved cosine similarity algorithm (introducing a Euclidean distance weight coefficient γ=0.3) is used to calculate the similarity between the current vector and the vectors of historical project nodes. A matching threshold of 0.85 is set. When the similarity exceeds the matching threshold, a bidirectional traceable semantic association mapping is automatically established. The recorded content includes: mapping pair ID, source node ID, target historical node ID, similarity value, and mapping timestamp. At the same time, a semantic association confidence report is generated, which includes a detailed comparative analysis of the top-5 matching results. The similarity calculation expression is as follows:

[0022] In the formula: Similarity; The Euclidean distance weighting coefficient is set to... ; The semantic vector of the current node is a 768-dimensional dense vector generated by the fine-tuned BERT model encoding a specific component node and its cost adjustment description after the current project is rescheduled. The historical node semantic vector is a 768-dimensional dense vector generated by the same BERT model from the standardized descriptive text encoding of a historical project node in the cross-project semantic feature library. Standard cosine similarity measures vector similarity. and The degree of proximity in direction; The L2 norm (magnitude) of a vector is used to represent the vector. For distance-based similarity transformation, it means converting the Euclidean distance between two vectors into a similarity metric; For vectors and The Euclidean distance (L2 distance) between them; The scaling parameter controls the rate at which the influence of distance on similarity decays. This parameter is determined through optimization on the validation set before model deployment to ensure that the distance similarity term and the cosine similarity term match on a numerical scale. All validated semantic association mapping pairs (confidence ≥ 0.85) are input into the unified semantic space projection module for semantic space alignment and knowledge reuse. This module uses the t-SNE dimensionality reduction algorithm (initial perplexity set to 30, learning rate 200) to project 768-dimensional vectors onto a 50-dimensional main semantic space, ensuring that cost data vectors across projects and stages retain their original semantic structure after dimensionality reduction. Over 85% of the cost data is automatically identified by performing density-based hierarchical clustering analysis (HDBSCAN, with a minimum cluster sample size of 5) within a unified semantic space. Each cluster is labeled with a topic tag. By calculating the Mahalanobis distance between the current node vector and the center vectors of each semantic cluster, the interpretability alignment of the cost data is achieved. Finally, a structured knowledge reuse package is generated, which includes: a semantic alignment matrix (50×50 dimensions), a cross-project cost pattern rule set (encapsulated in JSON-LD format), and a knowledge index file that can be queried later. This index supports multi-dimensional related queries based on component code, cost type, and time stage. The temporal structure modeling unit is used to combine project schedule temporal information to construct a time-series graph structure of cost nodes, forming a cross-project cost knowledge graph. This captures the temporal dependencies and stage characteristics in the cost evolution process, providing time-aware data support for full-cycle pattern mining. It enhances the ability to identify long-term dependencies in cost fluctuations, improves trend prediction accuracy, and obtains the temporal information of each cost node throughout the project lifecycle, including occurrence time, duration, and schedule stage labels. It constructs a time-based sequence of cost nodes, ensuring that cost events have traceability and structured representation capabilities in the time dimension, laying a time-series foundation for subsequent graph construction. By combining cost node sequences with semantic embedding results, a graph structure with temporal and semantic edges is established, forming a cost knowledge graph that spans projects and stages. Nodes represent cost events, and edges represent temporal dependencies or semantic associations, enabling the association and fusion of cost data across multiple projects and stages. This forms a knowledge network with both semantic and temporal dimensions. A temporal attention mechanism is introduced into the cost knowledge graph to capture the stage characteristics and long-term dependencies in the cost evolution process, providing temporally aware data support for subsequent spatiotemporal feature extraction and pattern mining. This enhances the model's ability to model long-term temporal dependencies and improves the accuracy of risk transmission path identification and trend prediction. The specific work involves: when acquiring the full-cycle cost node time-series information, extracting multi-source heterogeneous time-series data from the project management system, schedule documents, and construction logs; extracting precise time information for each cost node at a daily granularity; and setting specific parameters as follows: the time-series attributes of each cost node include the occurrence timestamp, duration, and progress stage label (based on the eight standard construction stages defined in GB / T50326 "Construction Project Management Standard": foundation and substructure, main structure, building decoration and finishing, building roof, building water supply and drainage and heating, building electrical, intelligent building, and ventilation and air conditioning). This information is then processed through an ETL process. Multi-source heterogeneous time-series data is cleaned, aligned, and standardized to ensure continuous and conflict-free timelines, constructing a cost node time series covering the entire project lifecycle. Based on the generated cost node sequences and the 768-dimensional semantic vectors output by semantic embedding units, a cross-project, cross-stage cost knowledge graph is constructed. The cost knowledge graph is stored and represented using an attribute graph model, where node types are defined as cost events, and node attributes include at least: event ID, timestamp, duration, stage label, semantic vector (768 dimensions), and cost amount. Edge relationships are divided into two categories: time-series edges are used to connect projects with sequential dependencies. The cost knowledge graph is structured around cost events, with attributes including dependency type and time lag. Semantic edges connect cost events from different projects that are associated through semantic matching (similarity ≥ 0.85), with attributes including similarity value and mapping source ID. The cost knowledge graph is stored in a graph database (Neo4j) and supports the nGQL query language, forming a traceable and reasonable cost knowledge network. A temporal attention mechanism is introduced into the graph structure of the cost knowledge graph to enhance its ability to capture the dynamic patterns of cost evolution. This mechanism is implemented based on the Transformer encoder architecture, specifically configured as follows: the input layer receives cost events. The temporal attributes of nodes are concatenated with semantic vectors; the number of attention heads is set to 8, the hidden layer dimension is 512, and the feedforward network dimension is 2048; in the self-attention calculation, in addition to the regular semantic attention, a relative position encoding based on time interval is introduced, and the maximum relative distance parameter in its calculation formula is set to 60 (corresponding to a two-month construction period). Through multi-layer attention stacking (the number of layers is set to 4), the model can automatically learn the dependence strength of cost events on the time axis, identify long-term cost transmission patterns across stages, and provide node and edge feature representations with strong temporal awareness capabilities for the downstream temporal graph convolutional mining module; The temporal graph convolutional mining module extracts spatiotemporal features from the cost knowledge graph across projects based on temporal graph convolutional networks, identifies cost fluctuation patterns and risk transmission paths, discovers abnormal cost patterns across cycles and projects, and performs trend prediction, providing data-driven quantitative predictions for future cost trends to assist decision-making and resource pre-allocation. The intelligent evolution analysis module is used to dynamically integrate the optimized cost data of the current project with the historical cost knowledge graph. Through an incremental learning mechanism, it updates the graph structure and node relationships to form a continuously evolving and self-improving cost knowledge system. The early warning decision support module uses graph neural networks and rule engines to detect abnormal cost nodes and recurring risk patterns, triggering early warning signals in real time. It combines the BIM model to locate risk sources, provides targeted resource optimization suggestions, and binds the early warning information and optimization suggestions to the BIM model for presentation on the BIM visualization platform. This allows users to intuitively understand the cost impact and resource scheduling effects, improving decision-making efficiency and risk response capabilities.

[0023] Example 2, as Figure 1 , Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: the temporal graph convolutional mining module is configured to perform the following steps: loading a cross-project cost knowledge graph, extracting the feature vectors of each node in the graph and their temporal connection relationships, constructing a spatiotemporal graph data structure, realizing a unified spatiotemporal structured expression of multi-source heterogeneous cost data, providing high-quality input for deep learning, using a temporal graph convolutional network to perform multi-layer feature aggregation and propagation on the spatiotemporal graph data structure, capturing the spatiotemporal dependencies between nodes, extracting the implicit spatiotemporal features of cost fluctuation patterns and risk transmission paths, deeply mining the spatiotemporal correlation patterns in cost evolution, identifying implicit risk transmission links, based on the extracted implicit spatiotemporal features, identifying abnormal cost patterns across cycles and projects, predicting future cost trends, outputting a risk transmission path graph and an abnormal node set, realizing the proactive discovery and trend prediction of abnormal costs, and supporting risk visualization and decision intervention; The specific work involves: In practice, using a graph database query interface, loading a cross-project cost knowledge graph generated by the temporal structure modeling unit, and precisely extracting two types of key data from the graph to construct a spatiotemporal graph data structure that can be directly used for neural network calculations: First, extracting the feature vector of each cost event node. This vector is 1536-dimensional, composed of a 768-dimensional semantic embedding vector and a 768-dimensional temporal attention enhancement feature vector, representing the semantic attributes and dynamic temporal state of the node; Second, extracting the temporal connection relationships between nodes. Here, the temporal edges are defined according to the project management logic. A connection is established only when there is a logical dependency between two cost event nodes confirmed by the schedule plan. The attributes of each time-series edge are recorded, including dependency type (preceding / following) and standardized time delay value (in working days). All extracted data is serialized into an adjacency matrix and a feature matrix. The adjacency matrix is ​​a sparse matrix to improve storage and computation efficiency, with a dimension of N×N (N is the total number of loaded nodes). The feature matrix has a dimension of N×1536, forming a standardized spatiotemporal graph tensor input. The constructed spatiotemporal graph data structure is then input into a Temporal Graph Convolutional Network (T-GCN). For deep feature learning, the T-GCN model structure is designed to include four consecutive temporal graph convolutional layers. Each layer contains two core operations: temporal feature capture and spatial feature aggregation. In the temporal dimension, a one-dimensional gated recurrent unit (GRU) is used to model the historical feature sequence of each node (the length of which is set to 30 time steps by the sliding window, corresponding to approximately 1.5 months of construction time) to capture the trend and cycle of cost changes. In the spatial dimension, graph convolution operations are used to perform feature aggregation on each node's neighboring nodes (i.e., its immediate predecessors or successors) based on the temporal dependencies defined by the adjacency matrix. Features of successor events are weighted and aggregated, with the aggregation weights calculated from a pre-defined symmetric normalized Laplacian matrix. Through multi-layer stacking (4 layers), node features are abstracted and propagated layer by layer in the spatiotemporal dimension. Finally, a high-dimensional implicit spatiotemporal feature matrix with a dimension of N×512 is output at the 4th layer. Using the 512-dimensional implicit spatiotemporal feature matrix output by T-GCN, the analysis task is performed. First, abnormal cost pattern recognition is performed: density-based spatial clustering with noise algorithm (DBSCAN) is used to perform unsupervised clustering analysis on the feature matrix, with the parameter set to neighborhood radius ε=0.5. With a minimum sample size of ms=5, nodes identified as noise points or belonging to small outlier clusters are marked as anomalous cost nodes. Simultaneously, by tracing their connections in the adjacency matrix, a risk transmission path subgraph with this node as the source or key transmission node is automatically generated. Next, future cost trend prediction is performed: for each cost event node to be predicted, its feature vector is input into a fully connected regression network (structure 512→256→128→1). This network has been trained using historical data (using a 70% training, 15% validation, and 15% test split), and can output the predicted cost change for that node in a specific future time window. Finally, two structured results are integrated and output: one is a visualization map of the risk transmission path marked with high-risk nodes; the other is a report of the anomalous node set containing a list of anomalous node IDs, transmission paths, and key prediction indicators. The intelligent evolution analysis module is configured to perform the following steps: combining the risk transmission path graph and abnormal node set output by the time-series graph convolutional mining module, dynamically aligning and fusing the optimized cost data of the current project and its associated semantic embedding vectors with the historical cost knowledge graph, updating the node attributes and temporal edges in the graph, ensuring continuous updates of the historical knowledge graph data, achieving seamless accumulation and immediate reuse of cross-project experience, adopting an incremental learning mechanism to adaptively adjust the graph structure, including inserting new nodes, updating edge weights, and reorganizing the structure, ensuring that the graph continues to evolve over time, achieving adaptive evolution of the knowledge graph, continuously improving the accuracy and timeliness of risk pattern recognition and prediction, and recalculating the correlation strength and temporal dependence between nodes based on the updated cost knowledge graph, forming a self-improving and continuously accumulating cost knowledge system, providing real-time knowledge support for subsequent early warning and decision-making, and constructing a dynamic knowledge base with millisecond-level response, providing accurate and real-time panoramic data support for intelligent early warning and decision-making; The specific work involves: immediately initiating the data alignment process after completing the risk transmission analysis of the current project. Specifically, this involves matching the set of abnormal nodes involved in the risk transmission path graph output by the time-series graph convolutional mining module, along with their associated optimal cost increment data and corresponding 768-dimensional semantic embedding vectors, with the historical cost knowledge graph stored in the Neo4j graph database in real time. The matching process is based on a unified node encoding system and timestamp alignment rules, with time alignment accuracy required to reach the daily level and the semantic vector similarity threshold maintained at 0.85. For existing nodes that are successfully matched, their node attributes are automatically updated, including: supplementing... The system automatically updates the actual cost amount for the current period, refreshes the semantic vector weighted average, and updates the timestamp to the latest data collection time. For newly added temporal dependencies in the transmission path, it creates new temporal edges in the graph database, accurately recording their time lag attributes (in working days). The entire process is executed automatically through a predefined Cypher query template (containing 15 standard update statements), ensuring the atomicity and consistency of data updates and achieving seamless integration of current project data with historical knowledge systems. After data fusion, an incremental learning mechanism is activated to dynamically optimize the overall structure of the cost knowledge graph. The core of this incremental learning mechanism is a lightweight graph. The spectral adaptive adjustment algorithm, with a learning rate of 0.01 and a batch size of 32, first identifies newly added isolated cost event nodes. Based on their semantic vectors and spatiotemporal features, it automatically links them to the most relevant existing clusters in the graph using the K-nearest neighbor algorithm (K=5, with cosine similarity as the distance metric), and establishes preliminary semantic edges (confidence > 0.75). For existing edges, the weights of their temporal edges are dynamically adjusted according to the actual transmission effect of the newly integrated projects. The adjustment formula is: new weight = original weight × 0.9 + current observation intensity × 0.1, where the observation intensity is calculated by normalizing the proportion of actual cost changes. Simultaneously, every 2... A structural reorganization check is performed every 4 hours. The LouvainMethod algorithm is used to repartition node communities to discover emerging cost pattern clusters, ensuring that the graph structure reflects the latest project practices and risk association patterns, and enabling continuous knowledge accumulation and evolution. After the structural update is completed, the knowledge system reconstruction calculation is initiated. Based on the latest graph topology and node attributes, the association strength between any two nodes in the entire graph is recalculated. This association strength is a comprehensive index composed of three weighted components: semantic similarity (weight 0.5), temporal dependency strength (weight 0.3), and cost transmission correlation coefficient (weight 0).2) The calculation cycle is 1 hour. Simultaneously, a time series analysis model is used to recalibrate the time lag relationships on the critical path to optimize the accuracy of time series predictions. Finally, a version-marked, queryable real-time cost knowledge snapshot is generated. This snapshot contains complete nodes, edges, weights, and metadata. Through the graph database's indexing mechanism (a composite index based on cost type, stage, and time has been established), it provides millisecond-level multi-dimensional query responses. This provides the downstream early warning decision support module with a real-time knowledge base covering historical patterns, current status, and evolutionary trends, directly supporting the triggering of risk warning rules and the intelligent generation of optimization strategies. The early warning decision support module is configured to perform the following steps: Based on graph neural networks and a preset rule engine, it performs anomaly detection on nodes in the cost knowledge graph, identifies recurring risk patterns and cost mutation nodes, triggers early warning signals in real time, and realizes the transformation of risk perception from manual inspection to intelligent real-time monitoring, shortening the early warning response time to the second level. Combining the BIM model to locate the spatial location and related components of abnormal nodes, it generates targeted resource optimization suggestions, including resource reallocation, schedule adjustment, process optimization and risk mitigation strategies, accurately mapping abstract cost risks to three-dimensional spatial components, making the optimization suggestions intuitive and operable. It dynamically binds early warning information, resource optimization suggestions and BIM model, and highlights and interactively presents them on the BIM visualization platform, supporting users to intuitively understand the cost impact and scheduling effect, and constructing a data-model-decision visualization closed loop, which greatly improves the manager's situational understanding and decision-making efficiency. The specific work content is as follows: In actual operation, the early warning decision support module calls a pre-trained graph neural network model (the model structure is a three-layer GAT, with 4 attention heads per layer, 256 hidden layer dimensions, LeakyReLU activation function, and a negative slope of 0.2) to perform node-level feature analysis and community discovery on the cost knowledge subgraph extracted in real time from the Neo4j graph database (extraction rules: time range traced back 90 natural days, node types include cost events and resource entities). The model outputs the anomaly score (score ≥ 0.7 is marked as an anomaly) and the risk pattern cluster ID to which each node belongs. At the same time, the built-in Drools rule engine works in parallel. Its knowledge base has 32 pre-defined discrimination rules defined by domain experts. When the output of either the graph neural network or the rule engine meets the early warning conditions (anomaly score exceeds the threshold or rule is triggered), a structured early warning signal is immediately generated. Its data packet follows the ISO29481-1 standard and includes early warning ID, trigger time, anomaly node ID set, risk pattern encoding, and confidence level. After the early warning is triggered, it immediately acts according to the anomalies in the early warning signal. The node ID set, through a unified encoding mapping interface (the mapping table is maintained in a central database, with an update cycle of ≤5 minutes), obtains the precise spatial coordinates, geometric attributes, and current construction status of the corresponding components from the BIM model server (supporting the IFC4 standard). Then, a strategy generator (whose core is a multi-objective optimization decision tree model) is launched, aiming to minimize cost increments and resource conflicts. Input parameters include: the spatial constraint matrix of the affected components, the current resource topology network status, and a list of resources available within 72 hours and their unit prices extracted from the enterprise quota library (based on the "Construction Engineering Cost Data Standard" GB / T51161-2016). After the model runs (the solution time is limited to within 30 seconds), it outputs a set of non-dominated, ranked optimization suggestion schemes, automatically selecting the top-ranked scheme and generating optimization suggestions containing specific instructions. The generated warning information and optimization suggestions are dynamically associated with the corresponding components and spatial locations in the BIM visualization platform through a lightweight data binding protocol (based on WebSocket, with a heartbeat interval of 15 seconds). The BIM visualization platform's rendering engine (based on Three.net) is used for this purpose.Upon receiving the binding command (js152 version), the system, according to predefined visual coding specifications, renders the BIM components corresponding to high-risk abnormal nodes as flashing red highlights (flash frequency 2Hz), renders associated impact path components as orange semi-transparent, and renders resource entities involved in movement in the optimization suggestions as blue wireframe models with path arrows. Simultaneously, the interactive panel is updated to display warning details and optimization steps in a structured list format. Users can click on highlighted components to view their associated cost deviation data, risk transmission path analysis diagrams, and optimization scheme simulation animations. All visualization statuses and interactive operation logs are recorded, and an audit report conforming to the "Building Information Modeling Delivery Standard" GB / T51301-2018 can be exported through the platform.

[0024] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0025] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A BIM technology-based engineering project full-cycle cost intelligent accounting system comprising a BIM visualization platform, characterized in that, The BIM visualization platform is communicatively connected with the following modules: A dynamic optimization scheduling module is configured to fuse a target engineering project BIM model and a real-time digital twin, introduce a bat algorithm to dynamically optimize a construction resource node affected by a risk, rely on a BIM component correlation and a twin state data to real-time optimize and adjust a configuration path and a scheduling logic of manpower, machinery and materials, and obtain an optimal cost increment after rescheduling; A knowledge graph construction module is configured to perform semantic association and time sequence embedding on the optimal cost increment and historical project data, construct a cross-project cost knowledge graph, and form a cost knowledge base; A time sequence graph convolution mining module is configured to extract a space-time feature in the cross-project cost knowledge graph based on a time sequence graph convolution network, identify a cost fluctuation mode and a risk transmission path, discover an abnormal cost mode across cycles and projects, and perform trend prediction; An intelligent evolution analysis module is configured to dynamically fuse cost data of a current project after optimization and a historical cost knowledge graph, update a graph structure and a node relationship through an incremental learning mechanism; An early warning decision support module is configured to detect an abnormal cost node and a repetitive risk mode by using a graph neural network and a rule engine, real-time trigger an early warning signal, locate a risk source based on a BIM model, provide a targeted resource optimization suggestion, and bind the early warning information, the optimization suggestion and the BIM model to present them on the BIM visualization platform.

2. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 1, characterized in that: The dynamic optimization scheduling module includes a BIM-twin fusion resource modeling unit and a conflict coordination unit. The BIM-twin fusion resource modeling unit is configured to construct a resource-task-space integrated correlation model based on component attributes of a target engineering project BIM model and twin state data of a real-time digital twin, dynamically bind construction resources to BIM component nodes to form a resource topology network, and determine construction resource nodes affected by a risk. The conflict coordination unit is configured to perform multi-objective dynamic optimization on the construction resource nodes affected by the risk by using a bat algorithm, generate and iterate a resource configuration scheme in combination with real-time progress and resource state feedback of the real-time digital twin, model conflicts between resources by using a graph neural network, realize automatic identification and collaborative resolution of the conflicts, and output an optimal cost increment after rescheduling.

3. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 2, characterized in that: The BIM-twin fusion resource modeling unit is configured to perform the following steps: Parse component attributes in a target engineering project BIM model, and extract twin state data of a real-time digital twin, including component construction state, resource occupation and progress completion rate; Based on the component attributes and the twin state data, construct a resource-task-space integrated correlation model, dynamically bind construction resources to corresponding BIM component nodes, and form a traceable resource topology network; According to the resource topology network, identify construction resource nodes directly affected by a risk event or indirectly dependent on the risk event, form a construction resource node set, and generate a structured input containing resource type, quantity, location and constraint conditions.

4. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 1, characterized in that: The conflict coordination unit is configured to perform the following steps: Based on the resource topology network and the set of construction resource nodes affected by risks, a multi-objective optimization model is constructed to minimize the cost increment and resource idle time as the optimization objective, obtain an optimization objective function, and set time, space, and resource capacity constraints; A bat algorithm is used to dynamically optimize the resource allocation scheme, and combined with the progress and state data fed back by the real-time digital twin, multiple scheduling schemes are iteratively generated and evaluated, and a graph neural network is used to model the conflict relationship between resource nodes; Based on the conflict identification result and the optimization objective function, the resource allocation scheme after rescheduling and the corresponding optimal cost increment are output, which includes the adjusted task sequence, resource allocation path, and conflict resolution strategy.

5. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 2, characterized in that: The knowledge graph construction module includes a semantic association embedding unit and a time sequence structure modeling unit; The semantic association embedding unit is configured to perform the following steps: Extract the semantic properties of each component in the target engineering project BIM model and the feature description of the cost node, and obtain similar node data in the historical project to build a cross-project semantic feature library; 6. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 5, characterized in that: A pre-trained language model is used to vectorize the cross-project semantic features, and the optimal cost increment after rescheduling is semantically matched and similarity calculated with the historical project nodes to establish a traceable semantic association mapping; Based on the semantic association mapping result, the current project cost node vector is embedded into a unified semantic space for cross-project, cross-stage cost data alignment and knowledge reuse. The time sequence structure modeling unit is configured to perform the following steps: Obtain the time sequence information of each cost node in the project life cycle, including the occurrence time, duration, and progress stage label, and construct a cost node sequence with time as the axis; 7. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 5, characterized in that: Based on the cost node sequence and semantic embedding result, a graph structure with time sequence edges and semantic edges is established to form a cross-project, cross-stage cost knowledge graph, where nodes represent cost events and edges represent time sequence dependencies or semantic associations; Introduce a time sequence attention mechanism in the cost knowledge graph to capture the stage characteristics and long-term dependencies in the cost evolution process. The time sequence graph convolution mining module is configured to perform the following steps: Load the cross-project cost knowledge graph, extract the feature vectors of each node and their time sequence connection relationships, and construct a spatio-temporal graph data structure; 8. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 5, characterized in that: Use a time sequence graph convolution network to perform multi-layer feature aggregation and propagation on the spatio-temporal graph data structure to capture the spatio-temporal dependencies between nodes and extract the implicit spatio-temporal features of cost fluctuation patterns and risk transmission paths; ​ ​ Based on the extracted implicit spatio-temporal features, abnormal cost patterns across cycles and projects are identified, and future cost trends are predicted, outputting a risk transmission path diagram and an abnormal node set.

9. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 8, characterized in that: The intelligent evolution analysis module is configured to perform the following steps: In combination with the risk transmission path diagram and the abnormal node set output by the time series diagram convolution mining module, the cost data of the current project after optimization and its associated semantic embedding vector are dynamically aligned and fused with the historical cost knowledge graph, updating the node attributes and time sequence edges in the graph; An incremental learning mechanism is adopted to adaptively adjust the graph structure, including insertion of new nodes, updating of edge weights, and restructuring of the structure; Based on the updated cost knowledge graph, the correlation strength and time sequence dependence between nodes are recalculated, forming a self-improving and continuously accumulating cost knowledge system.

10. The BIM technology-based engineering project whole-cycle cost intelligent accounting system according to claim 9, characterized in that: The early warning decision support module is configured to perform the following steps: Based on the graph neural network and the preset rule engine, the nodes in the cost knowledge graph are subjected to abnormal detection, identifying repetitive risk patterns and cost mutation nodes, and triggering real-time warning signals; In combination with the BIM model, the spatial position and associated components of the abnormal nodes are located, generating targeted resource optimization suggestions, including resource reallocation, schedule adjustment, process optimization, and risk mitigation strategies; The early warning information, resource optimization suggestions, and BIM model are dynamically bound, and highlighted display and interactive presentation are performed on the BIM visualization platform.

Images