Engineering cost information acquisition and data mining system and method

Abstract

The invention belongs to the technical field of project cost, and particularly relates to a project cost information acquisition and data mining system and method.The project cost information acquisition and data mining system comprises a heterogeneous data acquisition module for synchronously acquiring BIM model data, Internet of Things sensor real-time data flow and a distributed historical cost database through an API gateway; and the block chain evidence storage unit is connected with the heterogeneous data acquisition module, acquires required data based on the heterogeneous data acquisition module, verifies the acquired original data through the block chain evidence storage unit, writes the original data passing the verification into a Hyperledger Fabric channel, isolates the original data not passing the verification into a sandbox, and pushes an alarm to a supervisor. It can be understood that the verification of the original data based on the block chain evidence storage unit indicates that the original data is not counterfeited if the verification is passed, and indicates that the original data is possibly counterfeited if the verification is not passed, so that the problem of uncontrollable cost caused by data tampering is fundamentally eradicated.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of engineering cost, and particularly relates to an engineering cost information collection and data mining system and method. BACKGROUND

[0002] In the existing engineering cost field, data exists in multiple dimensions and is fragmented, and the reasons are as follows: Format heterogeneity: first, BIM model data is constructed in IFC standard format, including steel reinforcement model, concrete grade, etc., and Internet of Things sensor data is real-time flow in MQTT / OPC-UA protocol, including ground scale weighing and single item progress monitoring, etc., and historical database integrates SQL / NoSQL mixed storage of cost records, and actual compatible data is less than 35%; Space-time mismatch: among them, the data of BIM model is updated at a monthly frequency, the sensor is updated at a second level, and the historical database is updated at a quarterly frequency, so the time span between different data types causes the error of cost accounting to be further enlarged; Lack of credibility: the data tampering rate of subcontractors is high, and there is a lack of on-chain notarization mechanism; Therefore, based on the existing engineering cost data storage mode, that is, the passive storage mode of the "data lake", the closed-loop processing of dynamic cleaning, credible notarization and intelligent mining cannot be realized; when dealing with sudden situations, cross-regional data cannot be shared, and when dealing with regional material price fluctuations, the cost is uncontrollable.

[0003] Therefore, the application provides an engineering cost information collection and data mining system and method. SUMMARY

[0004] In order to make up for the deficiencies of the prior art and solve at least one technical problem proposed in the background art.

[0005] The technical scheme adopted by the application to solve the technical problems is: the engineering cost information collection and data mining system comprises: The heterogeneous data collection module synchronously acquires BIM model data, Internet of Things sensor real-time data flow and distributed historical cost database through an API gateway; The blockchain notarization unit is connected to the heterogeneous data collection module, generates a SHA-256 hash fingerprint based on the collected data, verifies the authenticity of the data source based on zero-knowledge proof, and stores it in the Hyperledger Fabric channel after verification; The incremental learning engine is based on When the divergence detection feature drift occurs, The feature weight dynamic update is triggered; A knowledge graph mining unit constructs a material-process-cost correlation network and internally stores an engineering cost ontology.

[0006] Preferably, the heterogeneous data collection module comprises: A BIM parser extracts construction attribute data from an IFC file and generates a structured bill of quantities. An edge computing node is deployed at a regional construction site, performs noise reduction filtering on sensor data streams, and outputs structured parameters.

[0007] Preferably, the blockchain storage unit performs: Generates a SHA-256 hash fingerprint for each data block and attaches a timestamp and a geographic location label. Verifies the authenticity of the data source based on zero-knowledge proof technology, compares the subcontractor registration key with the historical storage records on the chain, and When the verification fails, the abnormal data stream is automatically isolated, and an alarm signal is triggered. When the verification is passed, a digital signature is attached and stored in the Hyperledger Fabric channel.

[0008] Preferably, the incremental learning engine comprises: A feature drift detector monitors changes in data distribution and uses Divergence quantifies changes in data distribution. According to the formula: Wherein, is the real-time price proportion, is the historical price benchmark proportion. A dynamic weight updater updates the feature weight matrix based on the random forest model when is detected.

[0009] Preferably, the knowledge graph mining unit comprises: An ontology construction module defines the relationship between engineering cost entities. A graph neural network inference engine predicts the cost fluctuation path and outputs the cost decision matrix.

[0010] Preferably, it further comprises a collaborative fusion based on the incremental learning engine and the knowledge graph mining unit: A local model is trained at each regional edge node to generate encrypted feature weights . A global server aggregates encrypted weights for each bid section based on secure multi-party computation. According to the formula: Wherein, is the number of bid sections, represents the first bid section; The knowledge graph receives update the association rules, and output the cross-bid optimization decision.

[0011] Preferably, the feature weight The dynamic updating method is: Based on the edge computing node, the preprocessed real-time sensor data stream is obtained as a new data vector , the real-time sensor data stream includes temperature, humidity, construction progress and geological radar scanning data; Based on the distributed storage of the historical cost database, the historical material price data set is obtained, and the historical data mean vector is calculated ; wherein, is the weight matrix of the last time step, is the current updated weight matrix, is the learning rate, used to control the weight update amplitude, is the new data vector, is the historical data mean vector.

[0012] Preferably, it also includes based on the incremental learning engine and the block chain storage unit to deal with regional risks: Based on the blockchain unit, verify the authenticity of the regional price data; Then use the incremental learning engine to adjust the feature weight based on the verified data to generate a dynamic threshold ; wherein, is the regional risk coefficient, is the standard deviation.

[0013] Preferably, it also includes based on the knowledge graph mining unit and the block chain storage unit to complete the trusted decision: Based on the association rules output by the knowledge graph, generate a hash value The cost optimization path is attached with a digital signature and stored in the Hyperledger Fabric channel; In the AR interface, the verifiable cost optimization path is displayed.

[0014] An engineering cost information collection and data mining method, comprising the following steps: S1: Collect multi-region device data stream through OPC-UA protocol, and pre-filter noise by edge node; S2: The incremental learning engine detects feature drift, and when When the global model is updated, S3: generating a material replacement scheme based on knowledge graph reasoning, and pushing the material replacement scheme to an AR terminal after being stored by a blockchain; S4: calculating a dynamic threshold , and triggering a cost redistribution instruction.

[0015] The beneficial effects of the present application are as follows: 1. The engineering cost information acquisition and data mining system and method of the present application collects required data based on a heterogeneous data acquisition module, verifies the collected raw data through a blockchain storage unit, writes the raw data that passes the verification into a Hyperledger Fabric channel, and pushes an alarm to the supervision party for the raw data that fails the verification. It can be understood that, based on the verification of the raw data by the blockchain storage unit, if the verification is passed, it indicates that the raw data is not counterfeit, and if the verification is failed, it indicates that the raw data may be counterfeit, thereby fundamentally preventing data tampering and causing uncontrollable cost.

[0016] 2. The engineering cost information acquisition and data mining system and method of the present application analyzes the divergence of the raw data that passes the verification based on an incremental learning engine, calculates the divergence of the raw data that passes the verification based on the incremental learning engine, dynamically updates the weight matrix based on the calculated divergence

[0017] , and outputs a feature weight update instruction based on the dynamically updated weight matrix, so that the knowledge graph receives the adjusted weight update instruction and outputs a dynamic decision scheme, thereby optimizing the construction process and reducing the engineering cost. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of the present application. DETAILED DESCRIPTION

[0019] In order to make the technical means, creative features, purposes and effects achieved by the present application easy to understand, the present application will be further described below in conjunction with specific embodiments.

[0020] As shown in Figure 1 , the engineering cost information acquisition and data mining system and method of the present application comprises: a heterogeneous data acquisition module that synchronously acquires BIM model data, Internet of Things sensor real-time data stream and distributed historical cost database through an API gateway; The blockchain evidence storage unit connects to the heterogeneous data acquisition module, generates SHA-256 hash fingerprints based on the acquired data, verifies the authenticity of the data source based on zero-knowledge proofs, and stores it to the Hyperledger Fabric channel after verification. Incremental learning engine, based on Divergence detection feature drift, when When this happens, the feature weights are dynamically updated; The knowledge graph mining unit constructs a material-process-cost relationship network and includes a built-in engineering cost ontology library.

[0021] In the existing field of engineering cost estimation, data suffers from multi-dimensional fragmentation, for the following reasons: Format heterogeneity: First, the BIM model data is the construction attribute in the IFC standard format, including steel bar type, concrete grade, etc., while the IoT sensor data is a real-time stream in the MQTT / OPC-UA protocol, including weighbridge weighing and single project progress monitoring, etc. The historical database integrates cost records stored in a hybrid SQL / NoSQL format, and the actual compatible data is less than 35%. Spatiotemporal mismatch: BIM model data is updated monthly, while sensor data is updated every second, and historical databases are updated quarterly. Therefore, the time span between different data types further widens the error in cost accounting. Lack of credibility: Subcontractor data tampering rate is high, and there is a lack of on-chain evidence storage mechanism; Based on the above, the existing engineering cost data storage model, namely the passive storage model of "data lake", cannot achieve a closed-loop process of dynamic cleaning, trusted evidence storage and intelligent mining. In one embodiment of the present invention, the required data is first collected based on a heterogeneous data acquisition module. Then, the collected raw data is verified through a blockchain notarization unit. Verified raw data is written into a Hyperledger Fabric channel. Raw data that fails verification is isolated to a sandbox, and an alarm is pushed to the supervisor. It can be understood that verification of the raw data based on the blockchain notarization unit, if successful, indicates that the raw data is not forged; if unsuccessful, it indicates that the raw data may have been forged. The verified raw data is then pushed to an incremental learning engine for analysis. divergence Then based on the calculations divergence The system dynamically updates the weight matrix and outputs feature weight update instructions. The knowledge graph receives the weight update instructions and outputs a dynamic decision-making scheme. The dynamic decision-making scheme is updated at the construction terminal, guiding the construction terminal to modify the construction design, thereby optimizing the construction process and reducing the project cost. Among them, divergence Used to quantify the difference between the real-time price percentage and the historical price benchmark percentage, including: when When the value is zero, it indicates a small fluctuation, which can be considered as noise. when This indicates acceptable, normal fluctuations. when If the real-time price percentage shows a significant drift compared to the historical price benchmark percentage, then the model needs to be updated at this point.

[0022] In one embodiment, the heterogeneous data acquisition module includes: The BIM parser is used to extract build attribute data from IFC files and generate a structured bill of quantities. Edge computing nodes, deployed at regional construction sites, perform noise reduction filtering on sensor data streams and output structured parameters.

[0023] In the above embodiment, the heterogeneous data acquisition module needs to collect BIM model data. It then parses the IFC file in the BIM model using a BIM parser to obtain component attribute data, such as rebar type and concrete grade. Based on this component attribute data, it generates a structured bill of quantities for the project, including the concrete usage for a specific project. For example, in a pile foundation filling project, the design concrete usage output from the BIM model is 320. If the corresponding data is output, then the data will be output accordingly. In addition to the data output, it is also necessary to combine the data input. In this embodiment, the IoT sensor data stream is obtained in real time based on the IoT sensors deployed at the regional construction site. Then, based on the edge computing node, these IoT sensor data streams are preprocessed at the regional construction site to reduce the transmission delay and provide support for the authenticity of blockchain evidence. In one embodiment, the blockchain evidence storage unit performs: Generate a SHA-256 hash fingerprint for each data block and attach a timestamp and geolocation tag; The authenticity of the data source is verified by using zero-knowledge proof technology, and the subcontractor's registration key is compared with the on-chain historical evidence records. When verification fails, the abnormal data stream is automatically isolated and an alarm signal is triggered. When verification passes, the attached digital signature is stored in the Hyperledger Fabric channel.

[0024] In one embodiment, the incremental learning engine includes: Feature drift detectors are used to monitor changes in data distribution. Changes in the distribution of dispersion-quantified data; According to the formula: wherein, is real-time price proportion, is historical price benchmark proportion; a dynamic weight updater, when detecting updates the feature weight matrix based on the random forest model.

[0025] In the process of engineering cost, there may also be problems including the price rise of raw materials, leading to periodic engineering cost report production difficulty prompt. Assuming that any raw material price rises, and the traditional model cannot update the cost in time, the engineering cost will deviate from reality. In an embodiment, assuming that the steel price of the pile pouring project suddenly fluctuates, the calculation divergence : If the steel is divided into three categories, including threaded steel, profile steel and wire rod; : Threaded steel: Profile steel: Wire rod: : Based on the above, since divergence , the random forest weight update is triggered; As described above, the threshold of divergence is set to 0.3. According to the above embodiment verification, when the price of threaded steel rises by 28% in a week, the system detects that the material cost distribution has changed in structure, and the model update is triggered. It can be understood that if the threshold of divergence is set too small, the system may be triggered to update due to regular fluctuations. If the threshold is set too large, there is no timely response to price fluctuations. Therefore, in embodiment two, based on historical data verification, the threshold of divergence is set to 0.3, which is the optimal technical compromise after engineering verification.

[0026] In an embodiment, the knowledge graph mining unit comprises: An ontology construction module for defining engineering cost entity relationships; A graph neural network inference device for predicting cost fluctuation paths and outputting cost decision matrices.

[0027] In the current field of project cost, due to the fact that the implicit cost is associated, it is easy to cause the imbalance of decision-making. Specifically, the existing project cost system only counts the direct material / labor cost, ignores the process change cost caused by supply chain interruption, and cannot quantify the overall cost of material price rise to process adjustment and final delay risk; in an embodiment, still taking the above karst landform highway as an example, assuming that the price of the main material threaded steel rises by 28% per week, the price rises from 4150 yuan / ton to 5312 yuan / ton, based on this situation, according to the traditional engineering cost calculation method, only the direct material price difference quantity is calculated to represent the new cost, but the real cost is much higher than the new cost obtained by simple calculation: First, activate the ontology construction module, define entity relationship, including: material entity-threaded steel-process entity-pile foundation binding-cost entity-mechanical bench cost-implicit cost factor-supply chain interruption coefficient, then dynamically extend the ontology library, add new steel-concrete composite material entity, and establish new conduction relationship: threaded steel shortage-process change to prefabricated component assembly-increase hoisting mechanical bench; As shown in the following table: The arrow direction in the above table refers to the conduction path between entities, such as the relationship of threaded steel→process entity, according to the specific item of process entity in entity category, i.e. pile foundation binding and prefabricated component assembly, if threaded steel is not replaced by material, the corresponding process entity can be understood as pile foundation binding, and the corresponding cost entity can be expressed as mechanical bench cost, assuming that due to the lack of threaded steel material, the cost entity also includes the loss of shutdown.

[0028] For the above, use graph neural network for reasoning, specifically: The input layer data includes main material price fluctuation + 28%, regional risk coefficient , process change feasibility 0.92 (from BIM model verification result); Then use mathematical expression total cost : Among them, is the material cost, that is, the explicit cost or understood as the direct material price difference, which comes from the real-time material price verified by the blockchain, such as the real-time price of threaded steel in this embodiment, and the difference between the benchmark price in the historical database, such as ten thousand yuan; is the implicit cost, wherein is understood as the increase of hoisting cost, is understood as the liquidated damages, and the implicit cost is the derived cost triggered by the associated process change, such as the increase of mechanical bench calculated by BIM model or the liquidated damages clause in the contract database; is a process impact factor, based on the probability distribution generated by the graph neural network reasoning, when the material is replaced, the triggered derivative cost is the lifting of the prefabricated part, then the process impact factor can be set , if emergency procurement is needed, the process impact factor can be set ; represents the number of entity relationships, represents the entity, such as the case of short supply of threaded steel, based on the entity relationship table, the entity chain of threaded steel shortage includes material entity→process entity→cost entity, then ; In addition, the number of entity relationships has a dynamic adjustment function, when the knowledge graph detects a new material entity, such as replacing threaded steel with steel-concrete composite material, the number of entity relationships is automatically expanded.

[0029] Based on the calculation, the decision matrix is output, wherein: maintain threaded steel, the explicit cost is 2.18 million, the implicit cost is 3.26 million of shutdown loss, and the total is 5.44 million; use steel-concrete prefabricated parts, the explicit cost is reduced by 1.06 million, the implicit cost is 1.42 million of hoisting delay, and the total is 3.6 million; local small steel plant emergency procurement, the explicit cost is 9.7 million, the quality sampling delay is 6.3 million, and the total is 16 million; Based on the entity relationship and cost fluctuation path, the following conclusions are drawn: through the graph neural network to construct the conduction reasoning network of material-process-cost, the implicit cost of butterfly effect in engineering cost can be revealed, so that the decision scheme with the most favorable cost reduction can be selected among many decisions, that is, the construction material replacement scheme.

[0030] In an embodiment, it also includes the fusion of the incremental learning engine and the knowledge graph mining unit: Train local models at each regional edge node to generate encrypted feature weights ; The global server aggregates the encrypted weights of each bid section based on secure multi-party computation; According to the formula: wherein, is the number of bids, represents the bid; The knowledge graph receives updates the association rules and outputs the cross-bid optimization decision.

[0031] In the process of large-scale engineering, such as cross-regional highway construction, the data of each bidding contractor is stored independently, which is difficult to share in real time, leading to global decision lag, and considering that sensitive data is uploaded to the central server, it may cause business secrets to be leaked, thereby causing the problem of declining cooperation willingness; In an embodiment of the present application, still taking the karst topography highway as an example, and the project is divided into three bidding sections, that is, there are three independent subcontractors, at this time the emergency is that the price of K2 section threaded steel rises by 22% in one day, from 4150 yuan / ton to 5060 yuan / ton, because the traditional engineering cost system cannot share the alternative scheme of each bidding section, leading to K2 section forced high-priced procurement; Based on the above, when solving the above problems, first, the edge node is locally trained, and the real-time data is uploaded to the local model for training, and the encrypted feature weight is transmitted to the global aggregation server through secure multi-party computation encryption, such as: The data collected by the K2 section sensor stream: And the local historical data: the average price of threaded steel , Upload to the local model for training, and get through full weight calculation: (The original value is 0.32) Then, based on the dynamic updating calculation of the global knowledge graph: Aggregate the encrypted weights of each bidding section: Because At this time, the alternative scheme is activated: Supply chain change: threaded steel is replaced by steel-concrete composite material; Process chain change: pile foundation binding is replaced by prefabricated component splicing; Output the cost optimization path again: K2 section uses C50 steel-concrete to replace threaded steel, and the cost is reduced; Output to each bidding section: K2 section receives the instruction: use steel-concrete to replace threaded steel; K1 section receives the instruction: share the excess threaded steel inventory of K2; K3 section receives the instruction: adjust the construction drawing according to the new process; Based on the above, through the encrypted feature weight uploaded, after secure multi-party computation encryption, it is transmitted to the global aggregation server, combined with the dynamic rule adjustment of the knowledge graph, it can solve the problem of resource mismatch and response lag in multi-bidding engineering, and reduce the response time and save the cost when responding to emergencies.

[0032] In an embodiment, the feature weight The method for dynamic updating is: The pre-processed real-time sensor data stream is obtained as a new data vector based on the edge computing node , which includes temperature, humidity, construction progress, and geological radar scanning data; The historical material price data set is obtained based on the distributed storage historical cost database, and the historical data mean vector is calculated ; wherein, is the weight matrix of the previous time step, is the current updated weight matrix, is the learning rate for controlling the weight update amplitude, is the new data vector, is the historical data mean vector.

[0033] In this embodiment, the real-time sensor data stream is sourced from the edge computing node, and the node is deployed at the construction site for real-time preprocessing of Internet of Things sensor data such as temperature, humidity, construction progress, and geological radar data, ensuring low delay of data, with transmission delay controlled within 50 msec, for example, in a specific embodiment, based on the geological radar scanning data, the cave parameters are calculated as new data input; The historical data is sourced from the distributed storage historical cost database, which contains regional material price libraries such as historical prices of deformed steel bars, sectional steel, and wire rods. The system will periodically extract historical data sets from the historical cost database to calculate the mean vector to quantify the benchmark distribution, such as the mean value of material prices; For example, based on the edge computing node deployed at the construction site, multi-dimensional sensor data streams are obtained in real time through the OPC-UA protocol, including temperature / humidity sensor data to monitor environmental conditions, construction progress monitoring data such as truck scale weighing and mechanical operation frequency, and geological radar scanning data with a sampling rate of 10 Hz and a wavelength of 0.5-100 MHz; After denoising filtering is performed at the edge node, a structured data vector is generated, such as the cave parameters output after filtering of the geological radar scanning data: Based on the distributed historical cost database, the historical data set of regional material prices such as deformed steel bars, sectional steel, and wire rods in the past 6 months is called; The historical data mean vector is calculated: wherein, is the number of months, represents the month, represents the month of purchase price; Example: historical average price of deformed steel bar Yuan / ton Since divergence , the trigger threshold, at which the random forest dynamically responds and adjusts the feature weight: If the current price of deformed steel bar is 4250 yuan / ton, then According to the formula, the increment is calculated: wherein, in order to make the weight increment fall within the interval, avoiding the weight exceeding after normalization, the scaling factor is calculated to be approximately equal to 100 based on the historical maximum fluctuation value, such as 630, divided by the weight change threshold, such as 6.3 obtained according to the engineering experience coefficient; At the same time, according to the requirement of “Construction Project Cost Dynamic Control Specification” (GB / T 51095-2015): “Price fluctuation parameters need to be dimensionless, and the base value is taken as the hundred level”, therefore, the scaling factor is taken as 100; For the learning rate , the value is taken as 0.05 in this embodiment, which can control the weight update amplitude and avoid system overshoot or oscillation according to engineering verification; then: Since the original weight is: corresponding to deformed steel bar, steel bar and wire rod respectively, then the updated new weight is: The normalized new weight is expressed as: Based on the above, if the scaling factor is not set, the calculated weight increment is 26.5, resulting in weight overflow and model collapse; if the scaling factor is set to 50, the calculated weight increment is 10.6, resulting in system severe oscillation and inaccurate decision; if the scaling factor is set to 200, the calculated weight increment is 0.1325, resulting in a sluggish system response, missing the risk window, such as a sharp rise in threaded steel prices without timely processing, leading to cost overruns; Then, the new weight output after updating is responded to engineering: that is, when the knowledge graph material substitution solution is triggered, and the decision "use C50 steel instead of threaded steel" is output. Based on the above, when the raw material procurement price fluctuates, it needs to be based on the divergence calculation to determine whether to adjust the feature weight of the raw material. As mentioned above, based on the divergence calculation of steel, the decision needs to adjust the feature weight of threaded steel using random forest, and the final output of the new weight matrix of threaded steel is Therefore, based on the knowledge graph material substitution solution, the final decision is output, that is, to use C50 steel instead of threaded steel.

[0034] In an embodiment, it also includes a regional risk response based on the incremental learning engine and the blockchain storage unit: Based on the blockchain unit, verify the authenticity of regional price data; Then use the incremental learning engine to adjust the feature weight based on the verified data to generate a dynamic threshold ; where, is the regional risk coefficient, is the standard deviation.

[0035] In large projects, such as multi-bid highway construction, the traditional cost system uses a fixed threshold. When the material price breaks through the warning line, it cannot adapt to different regional dynamic risks, such as supply chain disruption or policy mutation, resulting in false triggering of the substitution solution in low-risk areas, that is, excessive conservative decisions, leading to cost increases, causing high-risk areas to respond slowly, and thus causing cost overruns. At the same time, subcontractors may tamper with data, such as falsely reporting concrete usage, leading to cost increases, and reducing the credibility of cross-regional data; Based on the above problems, in an embodiment, first, verify the authenticity of regional price data, then calculate the risk coefficient dynamically based on the obtained data, generate a dynamic threshold based on the obtained regional risk coefficient, and respond to decisions according to the dynamic threshold; Specifically: The blockchain storage unit verifies the authenticity of regional price data in real time, specifically by receiving data from subcontractors, extracting data to generate a hash value, comparing the registered key with the on-chain record, and verifying the authenticity, including: Extract data blocks from the steel price stream reported by subcontractors, for example: Then perform zero-knowledge proof verification and compare the subcontractor's registered key with the on-chain storage record; if If the verification fails, isolate the data stream and trigger an audit alert; If the verification is passed, a digital signature is attached, and it is pushed to the incremental learning engine; Then, the risk coefficient is dynamically calculated, and the risk coefficient is generated based on the regional geological exploration report : Assume the data obtained as follows: Cave density, set weight ; Based on the "Highway Engineering Construction Risk Assessment Standard" JTG / T3650-2020, the geological risk in karst development area accounts for 35%-45% of the overall risk assessment weight, and since the cave will cause the concrete consumption deviation to exceed 15%, the value , which conforms to the standard value, and the karst topography engineering verification geological factor is the largest factor for cost overrun; Based on the geological radar scanning data, and then real-time processing by the edge node, with a sampling rate of 10Hz, structured parameters are generated: Policy compliance risk, set , and the supply chain weight Balance, The local government new regulation database is synchronized in real time through the API gateway; Then assign ; Supply chain stability, set ; Based on the Internet of Things GPS trajectory analysis of heavy trucks for real-time positioning, combined with the meteorological bureau's heavy rain warning, assuming that heavy rain causes the highway to be closed for 18 hours, and the standard transportation time is 12 hours, then Where the risk coefficient comes from the historical interruption database; Then: Based on the obtained regional risk coefficient ; After verification, the price of deformed steel bar is: According to historical data statistics, based on the cost database; Yuan / ton, standard deviation Yuan / ton; Then Yuan / ton Then according to the dynamic threshold value calculated above , if: Real-time price exceeds threshold If not, the current procurement plan is maintained, otherwise, a dynamic response is triggered, the weight is adjusted by using the incremental learning engine, and a substitute scheme is output based on the knowledge graph, such as using C50 steel instead of threaded steel, and finally the cost redistribution path is displayed on the AR terminal.

[0036] In an embodiment, the trusted decision is completed by fusing the knowledge graph mining unit and the blockchain storage unit: The association rule generated based on the knowledge graph outputs a hash value; The cost optimization path is stored in the Hyperledger Fabric channel with a digital signature; The verifiable cost optimization path is displayed on the AR interface.

[0037] As described above, assuming that the real-time price exceeds the threshold , the current procurement plan is maintained, and threaded steel is still purchased, otherwise, assuming that the real-time price of threaded steel exceeds the threshold , a dynamic response is triggered at this time, and the weight is adjusted by using the incremental learning engine, and finally the substitute scheme is output based on the knowledge graph, and the substitute scheme is displayed on the AR terminal, i.e. AR interface.

[0038] Further, the embodiment of the present application also provides an engineering cost information collection and data mining method, comprising the following steps: S1: Collecting multi-region device data streams through OPC-UA protocol, and pre-filtering noise by edge node; S2: The incremental learning engine detects feature drift, and when , the global model is updated; S3: Generating a material substitution scheme based on knowledge graph reasoning, and pushing it to the AR terminal after being stored by blockchain; S4: Calculating the dynamic threshold , and triggering the cost redistribution instruction.

[0039] The following is explained and described based on specific scenarios: Assuming that the current project is a highway construction, the geological condition is karst topography, and a 2.8m cave is found through exploration, with a depth of 15m, which needs to be reinforced by pile foundation; in the prior art, the BIM model cannot synchronize the real-time ground penetrating radar data, and the original plan is to design the concrete consumption of the pile foundation as 320 , which needs at least 48h or more through artificial processing and determination, which increases the risk of delay of construction period; In an embodiment of the present application, based on the original signal of the ground penetrating radar, the specific parameters are sampling rate 10Hz and wavelength 0.5-100MHz; based on the noise reduction filtering means, the cave feature parameters are as follows: Cave diameter calculation: based on electromagnetic wave reflection time difference Where, is the electromagnetic wave speed, ; The cave volume is estimated as: The output: structured cave parameters are: Input IFC format BIM model and cave parameters output by edge node in BIM parser: Based on the extraction of concrete compensation coefficient threshold in "Highway Bridge and Culvert Design Specification" JTG D60 ; If the cave diameter , load "stiffening rib construction library" to activate the concrete compensation coefficient ; If the cave diameter , activate the concrete compensation coefficient ; Recalculate the concrete consumption: The original design formula is expressed as: The correction formula is: Where, is the filling coefficient, which is set to ; Then based on the above, the BIM model output is: Add stiffening rib components: diameter 1.2m, spacing 0.5m; Update the pile foundation concrete amount to 253.4 , the original design is 320 .

[0040] Based on the above, the API gateway realizes the real-time issuance of construction instructions: The data flow delay from the edge node to the BIM parser is controlled within 50 ; The design change trigger condition is as follows: When: The API gateway controls the instruction issuance, and the instruction issuance path is: From the BIM parser to the API gateway to the construction machinery control system, the concrete pumping amount is automatically adjusted; Based on the above, the BIM model fails to synchronize the real-time ground penetrating radar data, resulting in an imbalance between the designed concrete amount for the pile foundation and the actual concrete amount for the pile foundation, and further leading to cost increase and material waste. Furthermore, unlike the traditional process of high time consumption from the geological report to manual design to BIM update, in the embodiment, based on the process of edge node filtering, BIM model correction, and construction instruction issuance, the total response time for engineering design changes can be greatly saved, and based on the estimation of the karst cave volume and the recalculation of the concrete amount, the material can be greatly saved, i.e., the cost is reduced. In addition, it is worth noting that, in the embodiment of the present application, based on the acquisition of the dynamic compensation coefficient , the nonlinear problem of karst cave filling is solved; Further, the pile foundation filling project is implemented in three sections, and three independent subcontractors are used for concrete pouring operations. However, in order to seek benefits, each subcontractor may report false concrete pouring amount when performing concrete pouring operations independently, resulting in a concrete usage amount that is much larger than the actual concrete usage amount. That is, in the embodiment of the present application, blockchain storage is also needed to reduce the difficulty of engineering cost data collection and mining. Based on the above, first, the sensor mounted on the transport vehicle generates a data block, such as: (Timestamp / Car Number / Weight / GPS), and then the subcontractor's registration information is verified based on zero-knowledge proof; According to the "update pile foundation concrete amount to 253.4 , originally designed as 320 ." instruction output by the BIM model, and compare it with the cumulative concrete pouring amount reported by the subcontractor. If the daily usage increases by 20%, it needs to be verified based on zero-knowledge proof. If the verification is passed, it is normally included in the cost, otherwise, if the verification is not passed, the daily concrete cumulative pouring amount data stream is isolated, and an audit alarm is simultaneously started. Based on the above, the subcontractor's false reporting of concrete pouring amount can be avoided, the engineering cost can be saved, and the concrete usage data output by the BIM model can be verified; The basic principles, main features, and advantages of the present application are shown and described above. Those skilled in the art should understand that the present application is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A system for collecting and mining engineering cost information, characterized in that: Comprise: Heterogeneous data acquisition module, through API gateway synchronization access BIM model data, real-time data stream of Internet of Things sensor and distributed historical cost database; Blockchain storage unit, connected to the heterogeneous data acquisition module, based on the collected data to generate SHA-256 hash fingerprint, based on zero-knowledge proof to verify the authenticity of data source, and store to Hyperledger Fabric channel after verification; The incremental learning engine is based on divergence detection feature drift, when triggering dynamic updating of feature weights; Knowledge graph mining unit, build material-process-cost association network, embedded engineering cost ontology library.

2. The construction cost information collection and data mining system according to claim 1, wherein: The heterogeneous data acquisition module comprises: BIM parser, for extracting construction attribute data in IFC file, and generating structured bill of quantities; Edge computing node, deployed in regional construction site, performs noise filtering on sensor data stream, and outputs structured parameters.

3. The construction cost information collection and data mining system according to claim 2, wherein: The blockchain storage unit performs: Generate SHA-256 hash fingerprint for each data block, and attach time stamp and geographic location label; Based on zero-knowledge proof technology to verify the authenticity of data source, compare subcontractor registration key with historical storage records on the chain; When the verification fails, the abnormal data stream is automatically isolated, and the alarm signal is triggered, and when the verification is passed, the digital signature is attached and stored to Hyperledger Fabric channel.

4. The construction cost information collection and data mining system according to claim 3, wherein: The incremental learning engine comprises: Feature shift detector for monitoring data distribution changes, using divergence quantifies data distribution changes; According to the formula: wherein, is a real-time price share, is a historical price benchmark share; a dynamic weight updater that, when detecting updates the feature weight matrix based on the random forest model.

5. The construction cost information collection and data mining system according to claim 4, wherein: The knowledge graph mining unit comprises: Ontology construction module, for defining engineering cost entity relationship; Graph neural network reasoner, for predicting cost fluctuation path, and outputting cost decision matrix.

6. The construction cost information collection and data mining system according to claim 5, wherein: Also includes based on incremental learning engine and knowledge graph mining unit collaborative fusion: Train local model at each regional edge node, generate encrypted feature weights ; Global server based on secure multi-party computation aggregates each bid encryption weight; According to the formula: wherein, is the number of bidders, denotes the thbidder; The knowledge graph receives The association rules are updated, and a cross-section optimization decision is output.

7. The construction cost information collection and data mining system according to claim 6, wherein: The feature weight The method for dynamically updating is: obtaining, by an edge computing node, a preprocessed real-time sensor data stream as a new data vector , the real-time sensor data stream including temperature, humidity, construction progress, and ground penetrating radar scan data; Based on the distributed storage of historical cost database to obtain historical material price data set, and calculate the historical data mean vector ; wherein, is the weight matrix of the previous time step, is the current updated weight matrix, is the learning rate used to control the weight update magnitude, is the new data vector, is the historical data mean vector.

8. The construction cost information collection and data mining system according to claim 7, wherein: Also includes based on incremental learning engine and blockchain storage unit fusion to deal with regional risk: Based on the blockchain unit to verify the authenticity of regional price data; The reusing incremental learning engine adjusts feature weights based on the validated data to generate dynamic thresholds ; wherein, is the regional risk factor, is the standard deviation.

9. The construction cost information collecting and data mining system according to claim 8, wherein: Also includes based on knowledge graph mining unit and blockchain storage unit fusion to complete trusted decision: Based on the correlation rule output by the knowledge graph to generate hash value Cost optimization path attached digital signature stored to Hyperledger Fabric channel; In the AR interface, the verifiable cost optimization path is displayed.

10. An engineering cost information collection and data mining method, characterized in that: Comprise the following steps: S1: Collect multi-region equipment data stream through OPC-UA protocol, and pre-filter noise in edge node; S2: The incremental learning engine detects feature drift, when the global model is updated; S3: Generate material substitution scheme based on knowledge graph reasoning, and push to AR terminal after blockchain storage; S4: Calculate dynamic threshold , triggering a cost reallocation instruction.

Images