Green building engineering cost collaborative management system oriented to information processing and storage support service

Abstract

The present application belongs to the technical field of construction engineering cost collaborative management, and relates to a green building engineering cost collaborative management system facing information processing and storage support services. The present application comprises: a physical perception layer which collects environmental parameters and material data and completes cleaning and filling; a semantic fusion processing layer which analyzes building information model files to construct component topology graphs, extracts contract text semantic features by using a deep learning network, generates cross-modal fusion node features through a gated attention fusion mechanism, calculates and outputs component-level cost deviation risk indexes by using a graph neural network, and tracks and conducts paths; a distributed storage support layer which implements three-level storage migration according to a multi-factor heat decay function and processes write conflicts through a multi-version concurrency control mechanism; and a cost collaborative application layer which executes multi-objective optimization with green constraints and triggers deviation early warning. The present application improves the accuracy, response speed and system consistency of green building engineering cost collaborative management.

Description

Technical Field

[0001] This invention belongs to the field of collaborative management technology for construction project costs, and specifically relates to a green building project cost collaborative management system oriented towards information processing and storage support services. Background Technology

[0002] Green building project cost management requires processing three distinct types of data simultaneously: carbon emission flow data generated by the physical sensing layer, static structural data carried by the building information model, and unstructured constraints constituted by contract texts and policy clauses. These three types of data lack a unified semantic anchoring framework within the existing technological system. Carbon emission constraints cannot be precisely mapped to each building component, and contract risk clauses cannot directly participate in component-level cost calculations, thus cost management remains at the level of macro-indicator benchmarking.

[0003] Existing technologies have attempted to introduce graph neural networks into construction project cost prediction. Patent document CN121258187A discloses an intelligent cost assessment method based on graph neural networks, which trains a hierarchical graph neural network model by constructing engineering structural data, historical cost data, and external context data. However, in this scheme, the input features of the graph nodes are limited to the structured data dimension; the semantic information of contract terms and policy documents is not included in the feature space, and a mathematical mapping relationship is not established between legal constraints and physical engineering entities. Patent document CN120317908A discloses an artificial intelligence-based cost management method that calculates the weights of each modality of data and fuses feature vectors through a self-attention mechanism. The fusion objects are structured data such as environmental data, market data, and project-specific data; similarly, the semantic dimension of the contract text is not included in the fusion system. The patent document with publication number CN120524568A uses a hierarchical attention mechanism and a temporal graph neural network to extract semantic parameters from unstructured text and fuse them with geometric data. However, its application scenario is three-dimensional modeling and construction management, and it has not extended to the technical path of coupling text semantics with component-level cost calculation in the field of cost management.

[0004] At the data storage level, patent document CN120872240A discloses a multi-level caching optimization method based on adaptive adjustment of data popularity. The popularity scoring model is constructed based on access frequency, time decay, and user priority. This scheme does not distinguish the differentiated impact weights of different business operation types on data popularity. In the engineering cost collaboration scenario, collaborative operations involving amount modification and read-only audit query operations have fundamentally different requirements for cache residency, and the general popularity scoring model cannot identify this difference. High-frequency, low-value query operations may squeeze high-value collaborative data out of the cache space, resulting in a structural disconnect between storage scheduling logic and cost business value.

[0005] Regarding the transmission of cost fluctuations, existing solutions typically rely on statistical methods or empirical coefficients to estimate indirect impacts. When material prices fluctuate, managers cannot trace the scope and extent of affected components along physical dependencies, making it difficult for cost management to switch from post-event accounting to a proactive early warning model. Summary of the Invention

[0006] The purpose of this invention is to provide a collaborative management system for green building project cost that supports information processing and storage services. It solves the technical problems in green building project cost management, such as the lack of a unified semantic anchoring framework among physical perception data, building information model data and unstructured contract text, the disconnect between storage scheduling logic and cost business value, and the lack of mathematical modeling based on physical topology for cost transmission caused by material price fluctuations.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A green building engineering cost collaborative management system for information processing and storage support services, the system consists of a physical perception layer, a semantic fusion processing layer, a distributed storage support layer, and a cost collaborative application layer, and the system performs the following steps:

[0009] Step 1: The physical sensing layer collects environmental parameter sequences and material flow data of the construction site. The abnormal noise in the environmental parameter sequence is identified and corrected by a sliding window anomaly detection algorithm deployed on the edge side. The non-continuous missing breakpoints in the corrected sequence are interpolated and filled by a bidirectional temporal neural network to generate a continuous environmental sensing data stream.

[0010] Step two: The semantic fusion processing layer parses the building information model file, extracts the geometric attributes, material attributes, and spatial and logical relationships between components, and reconstructs the building information model file into a topology graph with components as nodes and connections between components as edges. A pre-trained language model is used to extract semantics from unstructured contract text and policy clauses, generating text semantic feature vectors. A gating attention mechanism is used to adaptively weight and fuse the text semantic feature vectors with the structural feature vectors of the components, generating cross-modal fusion node features. A graph neural network is used to perform neighborhood aggregation calculations on the cross-modal fusion node features, outputting the component-level cost deviation risk index for each component in a dynamic environment. Based on the component-level cost deviation risk index, the transmission path of cost fluctuations is tracked in the topology graph.

[0011] Step 3: The distributed storage support layer establishes a three-level storage model including a cache layer, a warm storage layer, and a cold archive layer. It constructs a multi-factor heat decay function based on access frequency and operation type, calculates the heat value of each data shard in real time, and performs adaptive migration of data shards between the three-level storage models according to the heat value. It also adopts a multi-version concurrency control mechanism to handle write conflicts of high-concurrency nodes.

[0012] Step four: The cost collaboration application layer receives the component-level cost deviation risk index, executes a multi-objective optimization model with green constraints, and solves for the optimal balance between economic cost and environmental cost. When the cost deviation is detected to exceed the warning threshold, the settlement process of subsequent related nodes is blocked and change approval is triggered.

[0013] Furthermore, in step one, the sliding window anomaly detection algorithm dynamically defines the window length based on the sensor sampling frequency for the environmental parameter observation sequence. For the current data point, it calculates the mean and standard deviation of the current data point within the sliding window. When the absolute value of the deviation between the current data point and the mean exceeds the product of the dynamic threshold coefficient and the standard deviation, the current data point is determined to be abnormal noise. The dynamic threshold coefficient is determined based on the quantiles of the historical noise distribution.

[0014] Specifically, the sliding window anomaly detection algorithm targets environmental parameter observation sequences. The window length is dynamically defined based on the sensor sampling frequency. For the current data point Calculate the current data point mean within the sliding window with standard deviation When satisfied Time determination This is abnormal noise, among which The dynamic threshold coefficient is determined based on the quantiles of the historical noise distribution.

[0015] Furthermore, in step one, the bidirectional temporal neural network consists of a forward hidden state extraction layer and a backward hidden state extraction layer. The forward hidden state extraction layer extracts the contextual features of the temporal sequence before the missing breakpoint, and the backward hidden state extraction layer extracts the contextual features of the temporal sequence after the missing breakpoint. The forward hidden state and the backward hidden state are concatenated and then mapped to the interpolated filling value of the missing breakpoint through a fully connected layer.

[0016] Furthermore, in step two, the process of generating cross-modal fusion node features by the gated attention fusion mechanism includes: performing element-wise weighted summation of the structural feature vector of the component and the text semantic feature vector according to the gate coefficients calculated by the self-attention mechanism, wherein the gate coefficients are used to adjust the contribution ratio of the structural modality and the text modality in the current cost calculation scenario.

[0017] Specifically, the gated attention fusion mechanism calculates cross-modal fusion node features. The expression is:

[0018]

[0019] in The structural feature vector of the component. The text semantic feature vector. These are the gating coefficients calculated using the self-attention mechanism. This indicates element-wise multiplication.

[0020] Furthermore, in step two, the process of the graph neural network calculating the component-level cost deviation risk index for each component node includes: weighted aggregation of cross-modal fusion node features of each node in the neighborhood of the target node; normalization of the aggregation weights based on the graph structure degree of the target node and its neighboring nodes; and outputting the component-level cost deviation risk index after processing with a nonlinear activation function. The graph neural network calculates the component-level cost deviation risk index for each component node through multi-layer graph convolution operations. :

[0021]

[0022] in It is a non-linear activation function. For nodes The set of neighboring nodes, For node-based With nodes The normalization factor determined by the degree. For trainable weight matrix, For bias terms, Neighboring nodes Cross-modal fusion node features.

[0023] Furthermore, in step two, tracking the transmission path of cost fluctuations includes: when the change in the component-level cost deviation risk index of the source node is identified, the transmission impact is recursively calculated layer by layer along the edges in the topology graph to the associated nodes; the transmission impact is the sum of the product of the change in the component-level cost deviation risk index of the upstream node, the edge dependency weight, and the transmission attenuation factor determined based on the material type.

[0024] Specifically: when the source node is identified Change in component-level cost deviation risk index At that time, along the edges in the topology graph to the associated nodes Calculate the transmission effect by recursion layer by layer :

[0025]

[0026] in For nodes The neighborhood set, For the edge Dependency weights For nodes The conduction attenuation factor is determined based on the material type. upstream node The change in component-level cost deviation risk index. For associated nodes The amount of transmission influence.

[0027] Furthermore, the conduction attenuation factor is dynamically adjusted based on the market supply and demand tension index of the materials used in the component. The adjusted conduction attenuation factor is jointly determined by the initial conduction attenuation factor, the market supply and demand tension index, and the adjustment coefficient, and the adjusted conduction attenuation factor is positively correlated with the market supply and demand tension index.

[0028] The adjustment expression is as follows:

[0029]

[0030] in This is the adjusted conduction attenuation factor. For components The market supply and demand tension index of the materials used This is the adjustment coefficient.

[0031] Furthermore, in step three, the process of updating the data shard popularity value using the multi-factor popularity decay function includes: accumulating the remaining portion of the initial popularity value after time decay with the weighted value of the operation type for each access within the statistical period; wherein, the degree of time decay is determined by the time decay coefficient and the time difference from the most recent access, and the weight of each access is jointly determined by the time weight of the access and the business operation type coefficient. The multi-factor popularity decay function updates the data shard popularity value. The formula for calculation is:

[0032]

[0033] in This is the initial heat value. The time decay coefficient, The time difference since the most recent access. This represents the total number of visits within the statistical period. For the first Time weighting for each visit. For the first The coefficient of the business operation type for each visit.

[0034] Furthermore, the business operation type coefficient is preset based on the data modification permissions and audit frequency of the cost business. A first weight coefficient is assigned to collaborative operations involving amount modification, and a second weight coefficient is assigned to read-only audit query operations. The first weight coefficient is greater than the second weight coefficient.

[0035] Furthermore, when the distributed storage support layer detects that the data hit rate of the cache layer is lower than a preset hit rate threshold, it sends a calculation degradation instruction to the semantic fusion processing layer. The semantic fusion processing layer dynamically reduces the neighborhood aggregation depth of the graph neural network from the current aggregation layer to a two-layer aggregation based on the calculation degradation instruction.

[0036] Furthermore, in step three, the multi-version concurrency control mechanism includes: each write operation to the cost node generates a new version data snapshot with a strictly incrementing logical timestamp, and the read operation reads the corresponding historical version snapshot based on the logical timestamp of the time of issuance.

[0037] Furthermore, in step four, the objective function of the multi-objective optimization model with green constraints is the sum of economic cost terms and environmental cost terms; the economic cost term is the sum of the products of each resource unit price, consumption quantity, and market fluctuation adjustment coefficient; the environmental cost term is the sum of the products of each emission equivalent and environmental compensation coefficient, and is subject to adjustment by a green weight factor.

[0038] The objective function of the multi-objective optimization model with green constraints is:

[0039]

[0040] in For the first The unit price of the resource, For the first The consumption of this resource For the first Market fluctuation adjustment coefficient for the resource. For the first Emission equivalents of Class II emissions, For the first Environmental compensation coefficients corresponding to class of emissions, The green weighting factor.

[0041] Furthermore, in step two, the gating attention fusion mechanism introduces a semantic and structural consistency verification layer during the fusion process. The semantic and structural consistency verification layer extracts key entities from the contract text and policy clauses to construct a text-side pseudo-structure graph, calculates the graph matching degree between the text-side pseudo-structure graph and the topology graph, and triggers manual review or automatic correction when the graph matching degree is lower than a preset matching degree threshold.

[0042] Furthermore, in step one, the physical sensing layer also parses the construction schedule file to extract the total float of each component and obtains the carbon quota trading price volatility in real time through an external data interface; in step two, in the step of tracking the transmission path of cost fluctuations in the topology graph based on the component-level cost deviation risk index, the calculation of the transmission impact introduces a carbon-schedule coupling dynamic adjustment mechanism, and the calculation expression for the carbon-schedule coupling transmission attenuation factor is:

[0043]

[0044] in, The carbon-progression coupling conduction attenuation factor. The initial conduction attenuation factor for the node is determined based on the material type. The market supply and demand tension index for materials used in components. This is the market adjustment coefficient. This represents the year-on-year change rate of carbon allowance trading prices within the current accounting period. To preset the carbon price sensitivity coefficient, The total float of the components, To preset the progress sensitivity coefficient, This is the preset time constant.

[0045] Furthermore, the multi-version concurrency control mechanism of the distributed storage support layer is configured with a spatiotemporal adaptive concurrency lock unit. When it is detected that the maximum value of the shortest path distance between any two nodes in the topology graph corresponding to multiple write operations arriving within a preset time window is less than a preset topology interference threshold, the spatiotemporal adaptive concurrency lock unit dynamically upgrades the node-level exclusive lock of each node corresponding to the multiple write operations to a regional topology lock covering the shortest path connected subgraph. During the existence of the regional topology lock, all nodes in the connected subgraph share the same transaction-level logical timestamp and generate a unified new version data snapshot. After the new version data snapshot is persisted, the regional topology lock is downgraded and released.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] This invention constructs a four-layer collaborative architecture consisting of a physical perception layer, a semantic fusion processing layer, a distributed storage support layer, and a cost collaborative application layer. It integrates physical environment perception data, static structural data of building information model, and unstructured contract text into a unified semantic framework for correlation calculation, thereby solving the technical problem of lack of unified anchoring basis for multi-source heterogeneous data in green building project cost management.

[0048] A gated attention fusion mechanism is employed to adaptively weight and fuse the structural feature vectors of the Building Information Model (BIM) with the semantic feature vectors of the text. This allows contract risk clauses and policy constraints to directly impact each building component, thereby improving the accuracy and semantic consistency of cost accounting. Furthermore, a graph neural network is used to perform neighborhood aggregation calculations on the cross-modal fusion node features and output a component-level cost deviation risk index. This enables the computable transmission and tracking of cost fluctuations along the building's physical topology, elevating cost management from post-event statistics to a pre-event topology early warning technology level.

[0049] In terms of storage scheduling, a multi-factor heat decay function based on access frequency and business operation type is introduced to achieve structured alignment between storage resource allocation and cost business value, improving storage response efficiency in high-concurrency collaborative scenarios. The carbon-schedule coupling dynamic adjustment mechanism introduced in the additional technical solution couples the volatility of carbon quota trading prices and the total construction float of components into the calculation of the transmission decay factor. This allows the intensity of cost volatility transmission to dynamically scale according to changes in the carbon market and the urgency of construction schedules, enhancing the sensitivity and targeting of risk warnings.

[0050] Meanwhile, the spatiotemporal adaptive concurrent locking unit configured in the distributed storage support layer dynamically upgrades node locks to regional topology locks by detecting the shortest path distance between nodes in the topology graph. This eliminates write-after-write conflicts in extreme concurrency scenarios of graph neural network computing, ensuring strong consistency of cost data under multi-version concurrency control. Combining the above technical means, this invention effectively improves the accuracy, response speed, and system robustness of collaborative cost management for green building projects. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0052] Figure 1 This is the overall flowchart of the present invention.

[0053] Figure 2This is a flowchart of the carbon-progress coupling conduction path tracing subprocess of the present invention.

[0054] Figure 3 This is a flowchart of the spatiotemporal adaptive concurrent locking mechanism subprocess of the present invention.

[0055] Figure 4 The diagram shown is of the system operation interface used in this invention. Detailed Implementation

[0056] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0057] The following is in conjunction with the appendix Figures 1-4 The embodiments of the present invention will be described in detail below.

[0058] This invention discloses a green building engineering cost collaborative management system for information processing and storage support services. The system consists of a physical perception layer, a semantic fusion processing layer, a distributed storage support layer, and a cost collaborative application layer.

[0059] The physical sensing layer is the front-end sensing network that acquires parameters of the physical world. Carbon emission concentration sensors and temperature and humidity sensors are deployed at the perimeter of the construction site, key construction work surfaces, and material storage areas to monitor carbon emission dynamics and changes in environmental parameters in real time. All green building materials entering the construction site are equipped with material identification modules. Fixed reading devices installed at entrances, exits, and tower crane cantilever points uniquely identify building materials and track their location throughout the entire process, providing raw physical flow data for material cost accounting.

[0060] The raw data captured by the physical sensing layer is cleaned and preprocessed by the data aggregation module through an edge computing processing unit deployed at the edge of the field. The edge computing processing unit runs an outlier detection algorithm based on a sliding time window to remove abnormal noise generated by the sensor due to electromagnetic interference or extreme weather.

[0061] The algorithm is based on the environmental parameter observation sequence. Define a length of Sliding window, window length It is dynamically adjusted based on the sensor's sampling frequency. When the current data point is acquired... satisfy When this occurs, the algorithm determines that the data point is abnormal noise and removes it. In the formula, The mean of all data points within the window. The standard deviation of all data points within the window. The dynamic threshold coefficient is determined based on the quantiles of the historical noise distribution.

[0062] For data identified as noise, the edge computing processing unit employs a bidirectional temporal neural network to impute missing values. The bidirectional temporal neural network consists of a forward hidden state extraction layer and a backward hidden state extraction layer. The forward hidden state extraction layer extracts the contextual features of the time sequence before the missing breakpoint, while the backward hidden state extraction layer extracts the contextual features of the time sequence after the missing breakpoint. The forward and backward hidden states are concatenated and then mapped to the interpolated imputed value for the missing breakpoint through a fully connected layer. Before deployment, the bidirectional temporal neural network has been pre-trained using historical field-measured time-series data, and its parameters are fixed and burned into the dedicated acceleration module of the edge computing processing unit.

[0063] The edge computing processing unit also integrates a frequency domain analysis module based on discrete Fourier transform, which filters out electromagnetic interference noise through a low-pass filter, so that the signal-to-noise ratio of environmental parameters transmitted to the semantic fusion processing layer is stably maintained above 40dB.

[0064] After data cleaning, it enters the semantic fusion processing layer. The semantic fusion processing layer uses a building information model (BIM) parsing engine to perform deep analysis on the BIM file conforming to the industrial basic standards, extracting the geometric coordinates, material properties, sub-item project codes, and spatial and logical relationships between building components. The semantic fusion processing layer then reconstructs the BIM file into a topological graph with components as nodes and connections between components as edges. Each node represents an independent building component, and each edge represents the component's load-bearing, connection, or subordinate relationship in physical space.

[0065] The semantic fusion processing layer uses a semantic mapping model based on graph neural networks to organically fuse the dynamic environmental data of the physical perception layer with the static structural data of the building information model.

[0066] The initial feature vector of each node in this model It consists of the following feature dimensions:

[0067] Component type code, spatial location coordinates, floor code, material carbon emission factor, material unit price, design quantity, current cumulative progress, estimated completion time, historical cost deviation rate, environmental monitoring correlation index, construction process type code, construction machinery energy consumption level, waste classification and recycling rate, dust monitoring value of construction area, and noise monitoring value of construction area.

[0068] For unstructured policy documents and contract clauses, the semantic fusion processing layer introduces a text semantic extraction unit based on a pre-trained language model. This unit, fine-tuned on a professional corpus in the cost engineering field, can automatically identify risk-sharing clauses, cost adjustment coefficients, and quota constraints in contracts, transforming them into dense text semantic feature vectors. The semantic fusion processing layer then adaptively weights and fuses these text semantic feature vectors with the structural feature vectors of building information model components using a gated attention fusion mechanism, generating cross-modal fusion node features. The mathematical expression of the gated attention fusion mechanism is as follows:

[0069]

[0070] In the formula, This represents the fused cross-modal fusion node feature vector. For the structural feature vectors of building information model components, This is a text semantic feature vector. These are the gating coefficients calculated using the self-attention mechanism. This represents element-wise multiplication. The gating coefficient is used. The system automatically learns the relative importance weights of textual and structural modalities in the current cost calculation scenario through a self-attention mechanism. When there are risk clauses in the contract that are directly related to the current component, the weight of the textual modality is automatically increased; when the cost of the component is dominated by purely structural factors, the weight of the structural modality is dominant.

[0071] In one implementation, the semantic fusion processing layer introduces a semantic and structural consistency verification layer during the execution of the gated attention fusion mechanism. This semantic and structural consistency verification layer extracts key entities from the contract text and policy clauses, including component names, material types, and construction techniques. Based on the extracted key entities, it constructs a text-side pseudo-structure graph and calculates the graph matching degree between the text-side pseudo-structure graph and the building information model topology graph. When the graph matching degree is lower than a preset matching degree threshold, a manual review or automatic correction process is triggered.

[0072] The fused cross-modal node features are input into a semantic mapping model based on a graph neural network. This model employs a multi-layer graph convolutional network architecture: the first layer aggregates first-order neighborhood features to extract the stress-environment coupling features of local component groups; the second layer aggregates second-order neighborhood features to extract cost correlation features within floors or functional zones; the third layer outputs the final component-level cost deviation risk index for each node. The layer-by-layer propagation rule of the graph convolution is as follows:

[0073]

[0074] In the formula, For the first Layer nodes eigenvectors, For the first Layer Neighborhood Nodes eigenvectors, For nodes The set of neighboring nodes, and They are nodes and nodes The degree, For the first Layer trainable weight matrix, For the first Layer bias term, It is a non-linear activation function.

[0075] The graph neural network ultimately outputs the component-level cost deviation risk index for each component. The calculation formula is as follows:

[0076]

[0077] In the formula, For nodes The component-level cost deviation risk index. It is a non-linear activation function. For nodes The set of neighboring nodes, For node-based With nodes The normalization factor determined by the degree. For trainable weight matrix, For bias terms, For neighboring nodes The cross-modal fusion node characteristics. This index characterizes the degree of cost deviation risk of building components under the dual effects of real-time carbon emission constraints and market price fluctuations.

[0078] The component-level cost deviation risk index, after calculation, is further used to track the transmission path of cost fluctuations. When the market price of a certain material fluctuates, causing a source node... The change in the component-level cost deviation risk index At that time, the disturbance travels along the edges in the topology graph to the associated nodes. Layer-by-layer recursive diffusion, transmission of influence The calculation formula is:

[0079]

[0080] In the formula, For nodes The neighborhood set, For the edge Dependency weights For nodes The conduction attenuation factor is determined based on the material type. upstream node The change in component-level cost deviation risk index. For associated nodes The amount of conduction influence. Conduction attenuation factor. The value is determined based on the material type of the component: the main structural components have low material substitutability and the conduction attenuation factor approaches 1; the decorative and finishing components have high material substitutability and the conduction attenuation factor approaches 0.3.

[0081] In one implementation, the conduction attenuation factor Further adjustments are made dynamically based on the market supply and demand tension index of the materials used in the components. The adjustment expression is as follows:

[0082]

[0083] In the formula, This is the adjusted conduction attenuation factor. The initial conduction attenuation factor is determined based on the material type. For components The market supply and demand tension index for the materials used is obtained from an external supply chain data interface and comprehensively reflects changes in material inventory turnover and supplier delivery cycles. This is the adjustment coefficient.

[0084] The system generates a cost fluctuation transmission topology diagram in the form of a directed weighted graph, which shows the degree of impact of each component on the collaborative dashboard of the cost collaboration application layer with differentiated visual markers.

[0085] The semantic fusion processing layer also includes a topological conflict detection algorithm. This algorithm uses axis-aligned bounding box technology to calculate and compare the 3D envelope spaces of components bound to different professional cost items in real time within 3D space. When volumetric interference overlap is detected between the envelope boxes of components from two independent professional fields, the system determines that a topological conflict exists and automatically freezes the cost settlement process for all sub-items within the interference area and its associated impact space until the relevant professional responsible persons complete the online countersigning correction.

[0086] The processed deep semantic data is supported by a distributed storage layer. This layer establishes a three-tiered storage model based on data lifecycle and access density. The first tier is a high-performance caching layer, used to store real-time cost data streams during the current construction phase and intermediate calculations for high-frequency concurrent modifications. The second tier is a warm data storage layer, employing a large-capacity storage cluster and erasure coding technology to encode every four original data blocks into two redundant check blocks, distributed across different physical racks. The third tier is a cold data archiving layer, utilizing cloud object storage technology to store carbon footprint tracking data and audit backups after the project's lifecycle ends.

[0087] In the three-tier storage model, data migration is driven by a multi-factor heat decay function. For each individual data shard in the storage system, the system calculates its heat value in real time. The formula for updating the popularity value is:

[0088]

[0089] In the formula, The initial heat value for data sharding. The time decay coefficient, The time difference since the last access to this data shard. This is the total number of cumulative operations within the preset swipe access frequency statistics window. For the first time recorded by the system Time weighting for each visit. For the first The coefficient of the business operation type for each visit.

[0090] Business operation type coefficient Based on the preset data modification permissions and audit frequency for cost estimation operations: Cost estimation collaborative modification operations involving monetary changes are assigned a first weight coefficient, read-only audit query operations are assigned a second weight coefficient, and historical report generation operations are assigned a third weight coefficient. The first weight coefficient is greater than the second weight coefficient, and the second weight coefficient is greater than the third weight coefficient. Even if audit query requests poll a data shard at extremely high frequency, its cumulative popularity remains lower than that of a data shard that has undergone a single collaborative modification operation because the upper limit of the business operation type coefficient corresponding to the audit query operation is relatively low. Therefore, storage resources are consistently allocated to core cost estimation collaborative operations.

[0091] When the heat value of a certain data shard When the data fragment's popularity value falls below a preset popularity threshold, the storage controller automatically migrates the data fragment from the high-performance cache layer to the warm data storage layer. When the popularity value of the data fragment further decreases and the time since the last modification exceeds a preset duration, the storage controller archives the data fragment from the warm data storage layer to the cold data archive layer.

[0092] In one implementation, a backpressure transmission link is established between the distributed storage support layer and the semantic fusion processing layer. When the distributed storage support layer detects that the data hit rate of the high-performance cache layer is lower than a preset hit rate threshold, it sends a computational degradation instruction to the semantic fusion processing layer. Based on the computational degradation instruction, the semantic fusion processing layer dynamically reduces the neighborhood aggregation depth of the graph neural network from the current aggregation layer to a two-layer aggregation.

[0093] The distributed storage support layer introduces a distributed consistency protocol and a multi-version concurrency control mechanism. Each successful write operation to a cost node generates a new version data snapshot with a strictly incrementing logical timestamp at the underlying level. Read operations retrieve the corresponding historical version snapshot based on the logical timestamp at the time the request was issued; read and write operations do not block each other.

[0094] For large-scale building information model (BIM) point cloud data, the distributed storage support layer adopts an octree-based spatial index structure. This structure recursively divides the three-dimensional space into eight subspaces until the amount of point cloud data contained in each subspace is less than the preset leaf node capacity. Each octree node records the point cloud density distribution and associated cost attributes within its coverage area. Cost attributes include the average component-level cost deviation risk index, cumulative cost amount, and main material type distribution for all building components within the spatial area. The retrieval algorithm has a logarithmic time complexity. The distributed storage support layer integrates content-addressable storage-based deduplication technology, extracting fingerprints from uploaded cost documents using cryptographic hash functions, ensuring that data blocks with identical content retain a single copy in physical storage.

[0095] The cost coordination application layer maintains a persistent connection with the lower layer via a full-duplex communication protocol. Its core coordination engine runs a multi-objective optimization model with green constraints, the objective function of which is:

[0096]

[0097] In the formula, For the first The unit price of the resource, For the first The consumption of this resource For the first Market fluctuation adjustment coefficient for the resource. For the first Emission equivalents of Class II emissions, For the first Environmental compensation coefficients corresponding to class of emissions, Green weighting factor. The objective function is dynamically configured by management personnel based on the project's green certification target level. The first term of the objective function is the economic cost term, and the second term is the environmental cost term.

[0098] The cost coordination application layer also introduces an automated bridging mechanism between green building certification and cost settlement. The system converts various indicators in the green building certification standards into a cost adjustment coefficient matrix and binds it to the settlement logic. When the actual carbon emissions reported by the physical sensing layer are lower than the preset benchmark value, the system automatically triggers the corresponding cost adjustment coefficient for cost settlement deduction.

[0099] The cost coordination application layer introduces traffic shaping and circuit breaking mechanisms. When the instantaneous input / output load of the distributed storage support layer exceeds the preset fault tolerance threshold, the system automatically performs rate limiting and interception on non-core data access requests.

[0100] To facilitate a better understanding of the present invention by those skilled in the art, the present invention will be further described below with reference to specific embodiments.

[0101] Example 1: This example describes the specific application of the method of the present invention in a large-scale green public building project.

[0102] The physical sensing layer deploys carbon emission concentration sensors and temperature and humidity sensors at the perimeter of the construction site, key construction work surfaces, and material storage areas. The carbon emission concentration sensors have a range of 0 to 1000 ppm, with an accuracy error controlled within ±5% across the entire range; the temperature and humidity sensors have a temperature sensitivity of 0.1℃. All green building materials entering the controlled areas of the site are equipped with material identification modules.

[0103] The data aggregation module cleanses the raw data stream using edge computing processing units deployed at the edge of the field. (Sliding window length) The value is determined based on the sensor's sampling frequency. Dynamic adaptive adjustment, the adjustment formula is as follows ,in This represents the window time span coefficient. Threshold coefficient. The parameters are determined based on the quantiles of historical noise distribution. The bidirectional temporal neural network has been pre-trained using historical field-measured time-series data before deployment, and its parameters are fixed and burned into the dedicated acceleration module of the edge computing processing unit.

[0104] The semantic fusion processing layer employs a graph neural network-based semantic mapping model with a three-layer graph convolutional network architecture. The pre-trained language model is fine-tuned on a corpus containing millions of professional terms in the field of construction cost. The gating coefficients of the gated attention fusion mechanism are... The initial value was set to 0.5, and adaptive optimization was performed through backpropagation during training. The training dataset for the semantic mapping model based on graph neural networks used anonymized data from multiple similar green building projects over the past five years, containing a large number of component nodes and their corresponding dynamic cost records.

[0105] The high-performance caching layer of the distributed storage support layer meets the real-time requirements of high-frequency concurrent collaboration. The erasure coding strategy of the warm data storage layer adopts a four-data-block-to-two-parity-block pattern, allowing simultaneous corruption of any two physical storage units without data loss, achieving a storage utilization rate of 66.7%. The time decay coefficient in the multi-factor heat decay function... The value is 0.05 per day, which is the coefficient for business operation type. The value standard is as follows: 1.0 for collaborative modification operations, 0.3 for audit query operations, and 0.1 for report generation operations.

[0106] Green weight factor in the multi-objective optimization model of cost coordination application layer The initial value is set to 1.0, and can be dynamically adjusted according to the project's green certification target level. An improved non-dominated sorting genetic algorithm is used for optimization.

[0107] Example 2: Based on Example 1, Example 2 further defines the semantic mapping model and propagation tracking mechanism based on graph neural networks in the semantic fusion processing layer, and sets a comparative example to verify the beneficial effects of the technical solution of the present invention.

[0108] In this embodiment, the semantic mapping model based on graph neural networks is implemented using a graph convolutional neural network. The node feature vectors of the building information model components... It includes fifteen feature dimensions: component type code, spatial location coordinates, floor code, material carbon emission factor, material unit price, design quantity, current cumulative progress, estimated completion time, historical cost deviation rate, environmental monitoring correlation index, construction process type code, construction machinery energy consumption level, waste classification and recycling rate, dust monitoring value of construction area, and noise monitoring value of construction area.

[0109] In the function of tracking the transmission path of cost fluctuations, the transmission attenuation factor The values ​​are refined as follows: 0.8 to 0.95 for main structural components, 0.5 to 0.7 for electromechanical installation components, and 0.3 to 0.5 for decorative components. When the market price fluctuation of a certain material exceeds the preset 5% threshold, the semantic mapping model based on graph neural networks performs transmission calculations layer by layer along the edges of the topology graph. The system generates a cost fluctuation transmission topology graph in the form of a directed weighted graph, showing the degree of impact of each component on the collaborative dashboard of the cost collaboration application layer using differentiated visual markers.

[0110] The scaling ratio is set as follows.

[0111] Comparative Example 1: A conventional distributed cost estimation system. It uses a traditional distributed database to store cost estimation data, lacks cross-modal semantic fusion capabilities, relies on manual input and mobile terminal feedback for data collection, and employs a standard consistency protocol and a common least recently used caching strategy in the storage layer, without hierarchical storage optimization. The system architecture consists of an application server plus a single-layer database.

[0112] Comparative Example 2: A centralized Building Information Modeling (BIM) cost management system. It uses a centralized server to store all BIM and cost data, lacks real-time physical sensing integration, and collaborative work is achieved through file uploads and downloads. There is no multi-version concurrency control mechanism, and all cost calculation tasks are processed on a single server.

[0113] For Example 2, compared with Comparative Examples 1 and 2, continuous operation tests were conducted under the same network bandwidth and concurrent user count. Test metrics included system response time, storage throughput, cost calculation accuracy, and green indicator calculation coverage. Experimental data comparison is shown in Table 1.

[0114] Table 1. Comparison of performance data between Example 2 and the comparative example;

[0115]

[0116] Note: In Table 1, the green indicator coverage rate of Comparative Example 1 and Comparative Example 2 refers to the proportion of macro indicators that can be covered by manual entry of static reports or ex-post estimation, rather than the component-level dynamic coverage rate based on real-time sensing data flow achieved by this invention.

[0117] As shown in Table 1, Example 2 achieves a response time of less than 200ms in high-concurrency collaborative average response latency, which is superior to Comparative Example 1's 642ms and Comparative Example 2's 2150ms. Regarding peak throughput of the storage system, Example 2 achieves a throughput of 18.5GB / s, approximately 4.4 times higher than Comparative Example 1's 4.2GB / s and approximately 12.3 times higher than Comparative Example 2's 1.5GB / s. In terms of dynamic cost prediction error rate, Example 2 achieves 2.1%, superior to Comparative Example 1's 8.5% and Comparative Example 2's 12.4%. Finally, Example 2 achieves 98.5% coverage of green indicator calculations.

[0118] Example 3: Based on Example 2, Example 3 further specifies the multi-factor heat decay mechanism and backpressure transmission mechanism of the distributed storage support layer.

[0119] In this embodiment, the data sharding heat value of the distributed storage support layer In the calculation, the time decay coefficient The value is 0.05 per day. Business operation type coefficient. The value standards are refined as follows: 1.0 for cost collaborative modification operations, 0.5 for building information model spatial retrieval operations, 0.3 for audit batch query operations, 0.1 for historical report generation operations, and 0.15 for carbon footprint traceability query operations. When the data sharding heat value... When the data fragment's popularity value falls below the popularity threshold of 0.15, the system automatically migrates the data fragment from the high-performance cache layer to the warm data storage layer; when the popularity value of the data fragment... If the value falls below 0.05 and the last modification occurred more than 180 days ago, the system will archive the data fragment from the warm data storage layer to the cold data archiving layer.

[0120] This embodiment further specifies the backpressure transmission mechanism between the distributed storage support layer and the semantic fusion processing layer. When the distributed storage support layer detects that the data hit rate of the high-performance cache layer is lower than the preset hit rate threshold of 60%, it sends a computation degradation instruction to the semantic fusion processing layer. Based on the computation degradation instruction, the semantic fusion processing layer dynamically reduces the graph convolution aggregation depth in the graph neural network-based semantic mapping model from the default three layers to two layers. The first layer of graph convolution still performs local neighborhood feature aggregation, while the second layer directly outputs an approximate value of the component-level cost deviation risk index. When the data hit rate of the cache layer recovers to above the preset recovery threshold of 80% and remains stable for a preset duration, the semantic fusion processing layer automatically restores the graph convolution aggregation depth to three layers.

[0121] This embodiment also specifies the dynamic adjustment mechanism of the conduction attenuation factor. When the external supply chain data interface obtains the material market supply and demand tension index... When the preset threshold is exceeded, the adjustment coefficient is adjusted. The adjusted conduction attenuation factor is set to a value of 0.2. The value is increased by the corresponding percentage compared to the initial value.

[0122] Example 4: Based on Example 3, Example 4 systematically reconstructs the physical coupling model of cost fluctuation transmission and the concurrent control logic of the underlying distributed storage. This example specifically discloses the complete engineering operation steps for implementing carbon-schedule risk coupling calculation and preventing topology write conflicts.

[0123] During the environmental and business data perception phase, the edge computing processing unit of the physical perception layer integrates an engineering plan parsing component. This component periodically reads construction schedule files in XML or MPP format, uses a built-in critical path method engine to perform forward and backward time extrapolation, calculates the earliest and latest start times of each construction task in the project network diagram, and obtains the total float for each task by subtracting the two. The system uses a globally unique identifier (GUID) to map and inject the calculated total time difference parameter into the physical component node features corresponding to the building information model. The system establishes an external secure network communication link and sends HTTP requests to the national or regional carbon emission trading center via an application programming interface (API) at a fixed polling period of 5 minutes to retrieve the latest average carbon allowance transaction price. The semantic fusion processing layer receives this price stream, uses a sliding time window to compare the average price difference between the current period and the previous benchmark period, and calculates and outputs in real time the month-on-month change rate of the carbon allowance trading price, i.e., the carbon allowance trading price volatility. .

[0124] Once the data is ready, the semantic fusion processing layer fully utilizes the carbon-schedule coupling dynamic adjustment mechanism when performing cost fluctuation transmission tracking. The system calculates the risk impact transmitted to related nodes according to the following core algorithm:

[0125]

[0126]

[0127] The calculation and derivation are performed using the core load-bearing component of the project, the "C50 cast-in-place prestressed concrete main beam". The system queries the material property library to extract the initial conduction attenuation factor of this high-strength cement-based component. With carbon price sensitivity coefficient The supply chain interface provides feedback on the current market supply and demand tension index for this grade of concrete. The external interface calculates the current volatility of carbon allowance trading prices. (That is, an increase of 12%).

[0128] The market regulation coefficient is set globally by the system. Schedule sensitivity coefficient time constant sky.

[0129] If the engineering plan analysis component indicates that the main beam is on the critical path of the current construction phase, its total float... sky.

[0130] The system substitutes the variables into the natural exponential function to obtain... The central processing unit performs floating-point operations to derive the correction attenuation factor. .

[0131] at this time The float is significantly greater than 1. The system determines that due to the extremely tight construction schedule coupled with the rising carbon prices of high-carbon emission materials, the cost risk at this node has been drastically amplified, with the impact cascading through to adjacent floor slabs and secondary beam nodes at a multiplier of 1.39. If this component belongs to a non-critical supporting area with ample construction time, its total float... sky.

[0132] System calculates index term The result of the calculation The sharp drop in the risk scaling factor indicates that the buffer formed by the time elasticity effectively absorbed the dual shockwaves from carbon prices and the market. This coupled calculation process ensures that the computing power of the cost early warning system is always tilted towards key nodes with real crisis attributes.

[0133] With the cascading diffusion of the aforementioned risk multipliers, graph neural networks generate a massive number of neighborhood update requests in a short period. To prevent state machine crashes caused by concurrent writes to nodes with strong topological associations, the distributed storage support layer enables a spatiotemporal adaptive concurrent locking unit. The working mechanism of this unit is based on a graph-structured memory adjacency list.

[0134] When a batch of concurrent write transactions enters the memory buffer queue, the spatiotemporal adaptive concurrent lock unit extracts the set of node IDs that the batch of transactions intends to modify (e.g., the node set contains nodes A and B). The system immediately runs Dijkstra's algorithm in memory to calculate the shortest path hop count between node A and node B in the graph neural network structure. .

[0135] System pre-configured topological interference threshold When detected (That is, there is a strong structural coupling between the two, satisfying) When this occurs, the system interrupts the regular multi-version concurrent write process.

[0136] The spatiotemporal adaptive concurrent lock unit sends an interception command to the distributed lock manager, stripping the transaction queue of its independent operation rights over nodes A and B (i.e., revoking the node-level exclusive lock). The algorithm initiates a breadth-first search (BFS) with nodes A and B as the two source nodes, expanding outward by two hops to precisely delineate the minimum connected subgraph containing the source nodes and their shortest path intermediate nodes. The distributed lock manager then generates an exclusive regional topology lock for this specific connected subgraph in the global registry.

[0137] During the exclusive period of this regional topology lock, the multi-version concurrency control mechanism treats all affected component nodes within the connected subgraph as a black-box state machine. The system's underlying layer allocates a globally unique transaction-level logical timestamp (e.g., transaction identifier: TxID-20240901-08442A) to uniformly record the change states of nodes A and B and their coupled calculation results. After the changed data is serialized and packaged, it is written to the warm data storage layer at once to generate a new version data snapshot of the connected subgraph. After the write confirmation beacon from the storage hardware returns to the control bus, the spatiotemporal adaptive concurrency lock unit actively destroys the regional topology lock, seamlessly downgrading the access granularity of graph data back to the discrete node level. This spatiotemporally aware lock escalation mechanism fundamentally eliminates phantom reads and write-after-write (WAW) conflicts in active concurrency scenarios, ensuring the physical boundaries of cost network data consistency.

[0138] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0139] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A green building engineering cost collaborative management system for information processing and storage support services, the system comprising a physical perception layer, a semantic fusion processing layer, a distributed storage support layer, and a cost collaborative application layer, characterized in that, The system performs the following steps: Step 1: The physical sensing layer collects environmental parameter sequences and material flow data of the construction site. The abnormal noise in the environmental parameter sequence is identified and corrected by a sliding window anomaly detection algorithm deployed on the edge side. The non-continuous missing breakpoints in the corrected sequence are interpolated and filled by a bidirectional temporal neural network to generate a continuous environmental sensing data stream. Step two: The semantic fusion processing layer parses the building information model file, extracts the geometric attributes, material attributes, and spatial and logical relationships between components, and reconstructs the building information model file into a topology graph with components as nodes and connections between components as edges. A pre-trained language model is used to extract semantics from unstructured contract text and policy clauses, generating text semantic feature vectors. A gating attention mechanism is used to adaptively weight and fuse the text semantic feature vectors with the structural feature vectors of the components, generating cross-modal fusion node features. A graph neural network is used to perform neighborhood aggregation calculations on the cross-modal fusion node features, outputting the component-level cost deviation risk index for each component in a dynamic environment. Based on the component-level cost deviation risk index, the transmission path of cost fluctuations is tracked in the topology graph. Step 3: The distributed storage support layer establishes a three-level storage model including a cache layer, a warm storage layer, and a cold archive layer. It constructs a multi-factor heat decay function based on access frequency and operation type, calculates the heat value of each data shard in real time, and performs adaptive migration of data shards between the three-level storage models according to the heat value. It also adopts a multi-version concurrency control mechanism to handle write conflicts of high-concurrency nodes. Step four: The cost collaboration application layer receives the component-level cost deviation risk index, executes a multi-objective optimization model with green constraints, and solves for the optimal balance between economic cost and environmental cost. When the cost deviation is detected to exceed the warning threshold, the settlement process of subsequent related nodes is blocked and change approval is triggered.

2. The green building engineering cost collaborative management system for information processing and storage support services according to claim 1, characterized in that, In step one, the sliding window anomaly detection algorithm dynamically defines the window length based on the sensor sampling frequency for the environmental parameter observation sequence. For the current data point, it calculates the mean and standard deviation of the current data point within the sliding window. When the absolute value of the deviation between the current data point and the mean exceeds the product of the dynamic threshold coefficient and the standard deviation, the current data point is determined to be abnormal noise. The dynamic threshold coefficient is determined based on the quantiles of the historical noise distribution.

3. The green building engineering cost collaborative management system for information processing and storage support services according to claim 1, characterized in that, In step one, the bidirectional temporal neural network consists of a forward hidden state extraction layer and a backward hidden state extraction layer. The forward hidden state extraction layer extracts the context features of the temporal sequence before the missing breakpoint, and the backward hidden state extraction layer extracts the context features of the temporal sequence after the missing breakpoint. The forward hidden state and the backward hidden state are concatenated and then mapped to the interpolated filling value of the missing breakpoint through a fully connected layer.

4. The green building engineering cost collaborative management system for information processing and storage support services according to claim 1, characterized in that, In step two, the process of generating cross-modal fusion node features by the gated attention fusion mechanism includes: summing the structural feature vector of the component and the text semantic feature vector element by element according to the gate coefficients calculated by the self-attention mechanism, wherein the gate coefficients are used to adjust the contribution ratio of the structural modality and the text modality in the current cost calculation scenario.

5. The green building engineering cost collaborative management system for information processing and storage support services according to claim 1, characterized in that, In step two, the process of the graph neural network calculating the component-level cost deviation risk index of each component node includes: weighted aggregation of the cross-modal fusion node features of each node in the neighborhood of the target node, the aggregation weight being normalized based on the graph structure degree of the target node and the neighborhood nodes, and the aggregation result being processed by a nonlinear activation function to output the component-level cost deviation risk index.

6. The green building engineering cost collaborative management system for information processing and storage support services according to claim 1, characterized in that, In step two, tracking the transmission path of cost fluctuations includes: when the change in the component-level cost deviation risk index of the source node is identified, the transmission impact is recursively calculated layer by layer along the edges of the topology graph to the associated nodes; the transmission impact is the sum of the product of the change in the component-level cost deviation risk index of the upstream node, the edge dependency weight, and the transmission attenuation factor determined based on the material type.

7. A collaborative management system for green building project cost estimation based on information processing and storage support services as described in claim 6, characterized in that, The conduction attenuation factor is dynamically adjusted based on the market supply and demand tension index of the materials used in the component. The adjusted conduction attenuation factor is determined by the initial conduction attenuation factor, the market supply and demand tension index, and the adjustment coefficient. The adjusted conduction attenuation factor is positively correlated with the market supply and demand tension index.

8. A collaborative management system for green building project cost estimation based on information processing and storage support services as described in claim 1, characterized in that, In step three, the process of updating the data shard popularity value using the multi-factor popularity decay function includes: accumulating the remaining part of the initial popularity value after time decay with the weighted value of the operation type of each access within the statistical period; wherein, the degree of time decay is determined by the time decay coefficient and the time difference from the most recent access, and the weight of each access is jointly determined by the time weight of the access and the business operation type coefficient.

9. A collaborative management system for green building project cost estimation based on information processing and storage support services as described in claim 8, characterized in that, The business operation type coefficient is preset based on the data modification permissions and audit frequency of the cost business. A first weight coefficient is assigned to collaborative operations involving amount modification, and a second weight coefficient is assigned to read-only audit query operations. The first weight coefficient is greater than the second weight coefficient.

10. A collaborative management system for green building project cost estimation based on information processing and storage support services as described in claim 1, characterized in that, In step one, the physical sensing layer also parses the construction schedule file to extract the total float of each component and obtains the carbon quota trading price volatility in real time through an external data interface; In step two, in the step of tracking the transmission path of cost fluctuations in the topology graph based on the component-level cost deviation risk index, the calculation of the transmission impact introduces a carbon-schedule coupling dynamic adjustment mechanism, and the calculation expression of the carbon-schedule coupling transmission attenuation factor is as follows: in, The carbon-progression coupling conduction attenuation factor. The initial conduction attenuation factor for the node is determined based on the material type. The market supply and demand tension index for materials used in component manufacturing. This is the market adjustment coefficient. This represents the year-on-year change rate of carbon allowance trading prices within the current accounting period. To preset the carbon price sensitivity coefficient, The total float of the components, To preset the progress sensitivity coefficient, This is the preset time constant.

Images