Intelligent supervision management and control method and system

Abstract

The invention discloses an intelligent supervision management and control method and system, and the method comprises the steps: generating a violation operation recognition result and an equipment abnormality early warning signal according to construction scene multi-mode data collected by an intelligent camera and real-time monitoring information of an Internet of Things sensor; generating a non-tampering block chain evidence storage data chain based on the violation operation identification result and the equipment abnormity early warning signal; according to the engineering change information in the block chain evidence storage data chain, updating a BIM collaboration platform through an incremental three-dimensional model synchronization technology, and generating a multi-party real-time interaction enhanced three-dimensional model; based on an enhanced three-dimensional model and Internet of Things environment monitoring data, a digital twin mirror image space is constructed, a construction process is simulated through a multi-source data fusion deduction engine, and a risk pre-judgment map and a resource configuration optimization strategy are generated. According to the embodiment of the invention, visual supervision, credible tracing and prospective decision making of the construction process can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of project management and control, and particularly relates to an intelligent supervision and control method and system. Background Art

[0002] In the field of construction project supervision, traditional supervision methods mainly rely on manual inspections and paper document records, which have problems such as low supervision efficiency, difficult data traceability, and slow collaborative response. With the popularization of BIM technology and Internet of Things devices, although existing digital supervision systems can achieve partial data collection and visual display, there are still obvious deficiencies in multi-source heterogeneous data fusion, real-time risk warning, and trustworthy evidence storage: on the one hand, the visual data, equipment status information, and project documents at the construction site are fragmented and difficult to be associated and analyzed through a unified model; on the other hand, the storage of key supervision data is easily tampered with, and there is a lack of a trustworthy evidence storage mechanism in links such as the acceptance of concealed works. In addition, the dynamic simulation of the construction process in existing systems is mostly based on static BIM models and cannot predict potential risks by combining real-time environmental data, resulting in a lag in resource allocation optimization. Especially in large and complex projects, problems such as design conflicts and schedule delays caused by frequent engineering changes are difficult to discover and solve in a timely manner. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent supervision and control method and system to solve the deficiencies in the prior art and be able to achieve visual supervision, trustworthy traceability, and forward-looking decision-making during the construction process.

[0004] An embodiment of the present application provides an intelligent supervision and control method, and the method includes: According to the multi-modal data of the construction scene collected by intelligent cameras and the real-time monitoring information of Internet of Things sensors, fuse visual features and equipment status features through a multi-modal graph neural network to generate illegal operation identification results and equipment anomaly warning signals; Based on the illegal operation identification results and equipment anomaly warning signals, combine the construction material entry records and the acceptance video data of concealed works, and use a dynamic fragmentation encryption algorithm to block and upload key information to the blockchain to generate an immutable blockchain evidence storage data chain; According to the project change information in the blockchain evidence storage data chain, update the BIM collaboration platform through incremental three-dimensional model synchronization technology to generate an enhanced three-dimensional model for multi-party real-time interaction, supporting cross-terminal annotation conflict detection and schedule simulation visualization; Based on the enhanced three-dimensional model and the Internet of Things environmental monitoring data, construct a digital twin mirror space, simulate the construction process through a multi-source data fusion deduction engine, and generate a risk prediction map and a resource allocation optimization strategy.

[0005] Optionally, based on the multi-modal data of the construction scenario collected by the intelligent camera and the real-time monitoring information of the Internet of Things sensors, the visual features and device status features are fused through a multi-modal graph neural network to generate an illegal operation recognition result and a device anomaly warning signal, including: According to the video timestamp of the intelligent camera and the acquisition frequency of the Internet of Things sensors, an adaptive interpolation algorithm is used to align the visual data stream and the device status signal to generate a multi-modal data stream with synchronized time series; Model construction workers, construction machinery and building materials as graph nodes, visual features and device parameters as node attributes, and construction movement line relationships as edge weights to construct a three-dimensional spatio-temporal heterogeneous graph; Based on the three-dimensional spatio-temporal heterogeneous graph, a dual-channel graph attention network is constructed. The first channel extracts the illegal behavior patterns in the visual features, and the second channel extracts the device status anomaly patterns. The features are fused through a cross-modal distillation loss function to generate a joint feature vector; According to the construction stage, the classification threshold is dynamically adjusted, and a fuzzy support vector machine is used to classify the joint feature vector to output the illegal operation recognition result and the device anomaly warning signal in real time.

[0006] Optionally, based on the illegal operation recognition result and the device anomaly warning signal, combined with the construction material entry record and the concealed works acceptance image data, a dynamic fragmentation encryption algorithm is used to block and chain the key information to generate an immutable blockchain evidence data chain, including: According to the key level of the illegal recognition result, the video fragments, sensor data and acceptance images are segmented into dynamic fragments of 8KB - 32KB; Generate a unique weight coefficient for each data fragment, and dynamically select the encryption algorithm based on the chaotic mapping algorithm. The greater the weight, the higher the encryption intensity; Adopt the Merkle Patricia tree structure to calculate the double-layer hash value by associating the encrypted fragments with the entry record to generate a globally unique fingerprint across data sources; When the concealed works acceptance data is detected, the consortium chain intelligent contract is automatically activated, and the fragment chaining process is started after verifying the data integrity according to the preset rules; Store the encrypted fragments through IPFS, record the hash value and access rights on the Ethereum side chain, and generate a blockchain evidence data chain with timestamp proof.

[0007] Optionally, based on the project change information in the blockchain evidence data chain, the BIM collaboration platform is updated through incremental three-dimensional model synchronization technology to generate an enhanced three-dimensional model for multi-party real-time interaction, supporting cross-terminal annotation conflict detection and progress simulation visualization, including: Analyze the engineering change orders in blockchain-based evidence storage, use natural language processing to extract key parameters including the changed parts and material specifications, and generate a structured change feature vector; Perform spatial octree segmentation on the BIM model according to the change feature vector, only update the 3D mesh data of the affected area, and generate a lightweight incremental model package; Deploy a rule inference engine on the collaborative platform, compare the incremental model with the design specification database, detect elevation conflicts and pipeline collisions, and generate a conflict heat map; Integrate the kinematic model of construction machinery, simulate the construction process sequence after the change, and generate the optimal equipment scheduling path through the particle swarm optimization algorithm; Adopt WebGL streaming technology to push the incremental model and conflict detection results to each terminal in real time, and support cross-terminal annotation of conflict detection and visualization of progress simulation.

[0008] Optionally, based on the enhanced 3D model and IoT environmental monitoring data, construct a digital twin mirror space, simulate the construction process through a multi-source data fusion deduction engine, and generate a risk prediction map and a resource allocation optimization strategy, including: Fuse the enhanced 3D model with IoT environmental data, and generate a digital twin basic model with millimeter-level accuracy through point cloud semantic segmentation; Access the real-time stream containing concrete curing monitoring data and crane stress data, and construct a material strength attenuation model and an equipment fatigue prediction model in the digital twin basic model; Adopt the Monte Carlo method to simulate equipment failure emergencies, calculate the structural safety factor decline curve through finite element analysis, and generate a risk probability distribution map; Construct a resource constraint satisfaction model, dynamically adjust the material distribution plan and mechanical scheduling plan in combination with the risk map, and output a resource allocation optimization strategy with redundancy.

[0009] Another embodiment of the present application provides a smart supervision and control system, and the system includes: A fusion module for fusing visual features and equipment status features through a multi-modal graph neural network according to the multi-modal data of the construction scene collected by an intelligent camera and the real-time monitoring information of IoT sensors, and generating an illegal operation recognition result and an equipment anomaly warning signal; An encryption module for, based on the illegal operation recognition result and the equipment anomaly warning signal, combining the construction material entry record and the hidden project acceptance image data, and using a dynamic fragmentation encryption algorithm to block-chain the key information in blocks to generate an immutable blockchain evidence storage data chain; An update module, configured to update the BIM collaboration platform through incremental 3D model synchronization technology according to the engineering change information in the blockchain evidence data chain, generate an enhanced 3D model for multi-party real-time interaction, and support cross-terminal annotation conflict detection and progress simulation visualization; A generation module, configured to construct a digital twin mirror space based on the enhanced 3D model and the Internet of Things environment monitoring data, simulate the construction process through a multi-source data fusion deduction engine, and generate a risk prediction map and a resource allocation optimization strategy.

[0010] Another embodiment of the present application provides a storage medium, in which a computer program is stored, and the computer program is configured to execute the method described in any one of the above when running.

[0011] Another embodiment of the present application provides an electronic device, including a memory and a processor, a computer program is stored in the memory, and the processor is configured to run the computer program to execute the method described in any one of the above.

[0012] Compared with the prior art, a smart supervision and control method provided by the present invention generates an illegal operation identification result and a device anomaly warning signal according to the multi-modal data of the construction scene collected by the intelligent camera and the real-time monitoring information of the Internet of Things sensors; based on the illegal operation identification result and the device anomaly warning signal, generate an immutable blockchain evidence data chain; according to the engineering change information in the blockchain evidence data chain, update the BIM collaboration platform through incremental 3D model synchronization technology to generate an enhanced 3D model for multi-party real-time interaction; based on the enhanced 3D model and the Internet of Things environment monitoring data, construct a digital twin mirror space, simulate the construction process through a multi-source data fusion deduction engine, and generate a risk prediction map and a resource allocation optimization strategy, so as to realize the visual supervision, trustworthy traceability and forward-looking decision-making of the construction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a hardware structure block diagram of a computer terminal of a smart supervision and control method provided by an embodiment of the present invention; Figure 2 It is a flowchart of a smart supervision and control method provided by an embodiment of the present invention; Figure 3 It is a structural diagram of a smart supervision and control system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation to the present invention.

[0015] An embodiment of the present invention first provides an intelligent supervision and control method, which can be applied to an electronic device, such as a computer terminal, specifically, an ordinary computer, etc.

[0016] The following takes the operation on a computer terminal as an example for a detailed description. Figure 1 It is a hardware structure block diagram of a computer terminal for an intelligent supervision and control method provided by an embodiment of the present invention. As Figure 1 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory can include a non-volatile storage medium and an internal memory.

[0017] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any intelligent supervision and control method.

[0018] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0019] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any intelligent supervision and control method.

[0020] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 1 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0021] It should be understood that the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0022] See Figure 2 , an embodiment of the present invention provides an intelligent supervision and control method, which can include the following steps: S201. Based on the multi-modal data of the construction scene collected by the intelligent camera and the real-time monitoring information of the Internet of Things sensors, fuse the visual features and device status features through a multi-modal graph neural network to generate the recognition results of illegal operations and the device anomaly warning signals. Specifically, according to the video timestamp of the intelligent camera and the acquisition frequency of the Internet of Things sensors, an adaptive interpolation algorithm can be used to align the visual data stream and the device status signal to generate a multi-modal data stream with synchronized time series. In the construction scene, the intelligent camera usually collects video streams at a rate of 25 frames per second (fps), while the sampling frequencies of Internet of Things sensors (such as vibration sensors and temperature sensors) may be 10 Hz or higher. Due to the microsecond-level deviation of the clock sources of different devices, directly fusing the data will cause misalignment of the time axes. To solve this problem, an adaptive interpolation algorithm is used to achieve time series synchronization: Clock alignment: Calibrate the system clocks of all devices based on NTP (Network Time Protocol) to control the time error within ±1 millisecond. For example, if the timestamp of a camera is 2023-10-01 08:00:00.123 and the timestamp of a vibration sensor is 2023-10-01 08:00:00.125, with a deviation of 2 milliseconds between them, they need to be aligned to the same benchmark.

[0023] Frequency matching: For low-frequency sensor data (such as 10 Hz), the cubic spline interpolation method is used to increase the data to 25 Hz synchronized with the video stream. For example, if the readings of a temperature sensor at time points t = 0 ms, t = 100 ms, and t = 200 ms are 25 °C, 26 °C, and 27 °C respectively, the temperature value at the intermediate time point (such as t = 40 ms) generated after interpolation is 25.4 °C.

[0024] Anomaly rejection: Statistically calculate the mean and variance of the sensor data through a sliding window (window size 1 second). If a data point deviates from the mean by more than 3 times the variance (3σ principle), it is marked as an outlier and replaced with a forward fill value. For example, if a vibration sensor suddenly changes to 1000 g (normal range 0 - 50 g) at t = 150 ms, the system automatically replaces it with 50 g at t = 100 ms.

[0025] The finally generated multi-modal data stream with synchronized time series is based on the time axis, and each data unit is 40 milliseconds (corresponding to 25 Hz), including fields such as video frames, sensor values, and device status labels (such as "tower crane in operation").

[0026] Model construction workers, mechanical equipment, and building materials as graph nodes, with visual features and equipment parameters as node attributes and construction movement line relationships as edge weights to construct a three-dimensional spatio-temporal heterogeneous graph; The three-dimensional spatio-temporal heterogeneous graph is a structured representation that integrates spatial location, time series, and entity relationships. Its construction process is as follows: Node definition and attribute extraction: Construction worker nodes: Extract the person bounding boxes from video frames through the YOLOv5 object detection model, and use ResNet-50 to extract visual features. The attributes include position coordinates (x, y, z), safety helmet wearing status (0 / 1), and action categories (walking, climbing, etc.).

[0027] Mechanical equipment nodes: Internet of Things sensor data (such as tower crane inclination angle, motor temperature) is used as equipment parameters, and combined with visual features (such as robotic arm posture) to generate node attributes. For example, the attributes of the tower crane node are {inclination angle: 2°, temperature: 65°C, load: 5 tons}.

[0028] Building material nodes: RFID tags record material types (such as steel bars, concrete), arrival time, and specification parameters (such as steel bar diameter 20mm), and visual features are extracted through a material surface texture classification model (such as ViT-Base).

[0029] Edge weight calculation: Construction movement line relationship: Calculate the movement trajectories of workers and equipment through a multi-object tracking algorithm (such as DeepSORT). If the distance between the two is less than 2 meters within 10 seconds, an edge is established, and the weight is the interaction frequency (such as the number of encounters per hour). For example, a welder and a welding equipment interact 30 times per hour, and the edge weight is 0.8 (normalized value).

[0030] Spatial relationship: Calculate the Euclidean distance between nodes based on three-dimensional coordinates. If the distance is less than 5 meters and there is a logical association (such as a concrete pump truck and a pouring area), an edge is established, and the weight is 1 / (1 + distance).

[0031] Spatio-temporal dimension expansion: Attach a time stamp sequence to each node to record its historical state changes (such as the temperature change curve of the tower crane in the past 5 minutes); The graph structure is dynamically updated according to time slices (every 5 minutes) to form a spatio-temporal evolution sequence.

[0032] Example: The heterogeneous graph of a construction scene contains the following nodes and edges: Node A (construction worker): Location (10, 5, 0), safety helmet = 1, action = welding, visual feature vector [0.25, -1.32, 0.87, 0.03, -0.45, 1.28, -0.67, 0.91, 0.12, -0.33, 0.75, -0.21]; Node B (welding machine): Temperature = 120°C, current = 200A, visual feature vector [-0.12, 0.89, 0.45, -0.67, 1.05, 0.33, -0.78, 0.21, 2.15, -0.09]; Edge AB: Weight = 0.9 (due to continuous interaction between personnel and equipment); Node C (steel bar material): Specification = 20mm, RFID = XYZ123, visual feature vector [0.55, 0.02, -0.88, 1.24, -0.15, 0.37, -1.42, 0.68].

[0033] Based on a three-dimensional spatio-temporal heterogeneous graph, a dual-channel graph attention network is constructed. The first channel extracts the violation behavior patterns in visual features, and the second channel extracts the abnormal device state patterns. The features are fused through a cross-modal distillation loss function to generate a joint feature vector; The dual-channel graph attention network (DC-GAT) consists of two parallel channels, which process visual features and device parameters respectively: Visual feature channel: Input: Node visual features (such as ResNet features); Network structure: 3-layer graph attention layer (GAT), with 128 attention heads in each layer, aggregating the visual information of neighboring nodes. For example, a welder node focuses on the welding equipment node it interacts with through the attention mechanism and calculates the weighted features.

[0034] Violation behavior detection: Extract abnormal action patterns in consecutive time slices through a spatio-temporal convolution module. For example, when detecting "not wearing a safety helmet", the model compares the current node attributes with the safety specification library. If the safety helmet status = 0 and it lasts for more than 10 seconds, it is marked as a violation.

[0035] Device state channel: Input: Numerically normalized device parameters (such as temperature, vibration value); Network structure: 2-layer graph convolutional network (GCN), with 64 dimensions in each layer, combined with an LSTM module to capture temporal anomalies. For example, the temperature of a tower crane rises from 60°C to 85°C within 5 minutes, and the LSTM outputs an anomaly probability of 0.75.

[0036] Abnormal pattern recognition: Set the equipment vibration threshold based on the ISO 10816 standard. If the vibration value of a node exceeds the threshold (such as 7.1 mm / s), a warning is triggered.

[0037] Cross-modal distillation loss function: Knowledge distillation: The violation detection results in the visual channel (such as "not wearing a safety helmet") are used as soft labels to guide the equipment channel to learn associated features (such as abnormal equipment status near the person). Loss calculation: Use the Kullback-Leibler Divergence to measure the distribution difference between the outputs of the two channels, with a weight ratio of 6:4 (visual dominant). Feature fusion: Concatenate the 128-dimensional output vectors of the two channels into a 256-dimensional joint feature vector, and reduce the dimension to 64 through a fully connected layer.

[0038] Training process: Dataset: 100,000 groups of labeled construction site data (5,000 cases of violation behaviors and 3,000 cases of equipment anomalies). Optimizer: AdamW, learning rate 0.001, weight decay 0.01. Number of training epochs: 100 epochs, batch size 32. Evaluation metrics: F1-score for violation detection is 92.3%, and AUC for equipment anomaly warning is 0.89.

[0039] Dynamically adjust the classification threshold according to the construction stage, use the fuzzy support vector machine to classify the joint feature vector, and output the violation operation recognition result and equipment anomaly warning signal in real time.

[0040] The fuzzy support vector machine (Fuzzy SVM) enhances the classification robustness for uncertain samples by introducing the membership function: Dynamic threshold adjustment: Construction stage division: Divided into foundation construction (stage 1), main structure (stage 2), and decoration (stage 3) according to the progress. Threshold rule: Stage 1 focuses on equipment anomalies (threshold 0.6), stage 3 focuses on personnel violations (threshold 0.7), and stage 2 is balanced (threshold 0.65). For example, the vibration warning threshold of the tower crane in stage 1 is 7.1 mm / s, and it is adjusted to 6.8 mm / s in stage 2.

[0041] Fuzzy membership calculation: Calculate the membership degree of each sample belonging to "violation" or "normal". For example, a welder not wearing a safety helmet but only staying briefly in a low-risk area has a membership degree of 0.4 (close to normal); if in a high-altitude operation area, the membership degree rises to 0.9.

[0042] The membership function uses a Gaussian kernel: , where c is the class center and γ = 0.1 controls the fuzzy range.

[0043] Classification decision: If the "violation" membership degree of a sample exceeds the dynamic threshold, a warning is triggered. For example, in stage 2, the temperature membership degree of a certain device is 0.68 (threshold 0.65), which is determined to be abnormal; The kernel function of the support vector machine selects the radial basis function (RBF), with parameter C = 1.0 (penalty coefficient) and gamma = 0.01.

[0044] Real-time output example: Violation operation identification: Time 08:00:05, location (15, 20, 5), personnel ID 003, not wearing a seat belt, confidence level 0.91; Device anomaly warning: Time 08:00:10, device ID tower crane - 002, vibration value 8.2 mm / s, exceeding the threshold by 23%, it is recommended to stop the machine for inspection.

[0045] S202, based on the violation operation identification result and the device anomaly warning signal, combined with the construction material entry record and the concealed works acceptance video data, uses a dynamic fragmentation encryption algorithm to divide the key information into blocks and upload them to the chain, generating an immutable blockchain evidence data chain; Specifically, according to the key level of the violation identification result, the video segments, sensor data, and acceptance images can be segmented into dynamic fragments of 8KB - 32KB; The key level of the violation identification result is dynamically determined by a predefined rule library. For example, a worker not wearing a safety helmet belongs to a first-level violation (high risk), while irregular temporary material stacking belongs to a second-level violation (medium risk). According to different levels, the data block division strategy is dynamically adjusted: First-level violation: The associated video segments are cut in units of 8KB to ensure that a single block of data contains a complete violation behavior (such as a 5-second video segment, resolution 1080P, frame rate 25fps, about 8KB after H.265 encoding); Second-level violation: The cutting granularity is relaxed to 16KB, allowing a single block to cover a longer time window (such as a 10-second video); Sensor data (such as tower crane inclination, concrete temperature) is segmented according to time windows, and each 32KB contains 10 minutes of sampled data (sampling frequency 1Hz, single parameter occupies 4 bytes).

[0046] The cutting process adopts a dynamic boundary detection algorithm, scans the data stream through a sliding window (window size: 512 bytes, step size: 256 bytes), and forces segmentation at the position of I-frames (key frames) to avoid video decoding distortion. For example, if the I-frame interval of a certain video is 2 seconds, the cutting is preferentially performed at the boundary of every 2 seconds to ensure that each fragment can be decoded independently.

[0047] Generate a unique weight coefficient for each data fragment, and dynamically select an encryption algorithm based on the chaotic mapping algorithm. The greater the weight, the higher the encryption intensity. The weight coefficient is determined by the criticality and timeliness of the fragment content, and the calculation rule is: weight = violation level coefficient × time decay factor. Violation level coefficient: The first-level violation is 1.0, and the second-level is 0.7. Time decay factor: e^(-0.1t), where t is the number of hours from the current time (for example, it decays to 0.9 after 1 hour).

[0048] The chaotic mapping algorithm uses the Logistic equation (parameter μ = 3.99, initial value ) to generate a pseudo-random sequence, which is mapped to the encryption algorithm selection space: Chaotic value ∈ [0, 0.33): Select AES-128 (fast encryption speed, suitable for low-weight fragments); Chaotic value ∈ [0.33, 0.66): Select SM4 (national cryptographic algorithm, medium strength); Chaotic value ∈ [0.66, 1]: Select AES-256 (highest strength, used for fragments with weight > 0.8).

[0049] For example, if the weight of a certain fragment is 0.85 (first-level violation, just occurred) and the generated value of the chaotic sequence is 0.72, then AES-256 encryption is triggered. The key is managed by the Hardware Security Module (HSM). The encryption time increases by 30%, but the security is improved to the FIPS140-2 Level 3 standard.

[0050] Adopt the Merkle Patricia tree structure, associate the encrypted fragments with the arrival records to calculate the double-layer hash value, and generate a globally unique fingerprint across data sources. The Merkle Patricia tree (MPT) combines the efficient verification of the Merkle tree and the space optimization characteristics of the radix tree. Its construction process is as follows: Leaf nodes: The metadata of each encrypted fragment (such as fragment ID, storage path, timestamp) is used as the key (Key), and its SHA3-256 hash value is used as the value (Value). Intermediate nodes: Build a hierarchical relationship in the chronological order of arrival records. For example, "Steel bar arrival on September 15, 2023" is the parent node, with multiple fragment child nodes below. Root hash calculation: Merge hashes layer by layer from bottom to top, and finally generate a 64-byte global root hash.

[0051] The double-layer hash mechanism further strengthens data integrity: The first layer: The original data hash of a single fragment (such as SHA3-256); The second layer: The ciphertext hash of the encrypted fragment (such as BLAKE2b).

[0052] For example, the hash of a certain video fragment before encryption is 0x3a7d…, and after encryption it becomes 0x5b9e…, and the two together constitute the double-layer fingerprint of the fragment.

[0053] When detecting the acceptance data of concealed works, the consortium chain smart contract is automatically activated, and after verifying the data integrity according to the preset rules, the fragment on-chain process is started; The acceptance data of concealed works (such as the binding of foundation steel bars) needs to meet the multi-signature verification conditions: Participants: Supervisor, Construction Party, Design Institute Representative; Verification rules: At least 2 / 3 signatures pass, and the timestamp is within the acceptance window period (such as 48 hours before concrete pouring).

[0054] The consortium chain smart contract (based on Hyperledger Fabric) executes the following logic: Data integrity verification: Compare whether the submitted fragment hash is consistent with the MPT root hash; Multi-party signature verification: Use the Ed25519 algorithm to verify the validity of the signature, and the signature threshold is dynamically adjustable; Trigger condition: When the deviation of the key node (such as the steel bar spacing) in the acceptance image from the BIM model is <5mm, the on-chain process is automatically triggered.

[0055] For example, an acceptance image of a certain foundation shows that the steel bar spacing is 98mm (design value 100mm), with a deviation of 2mm. After the smart contract verification passes, the relevant fragments (including images and sensor data) are marked as "ready for on-chain".

[0056] ‌Store the encrypted fragments through IPFS, record the hash value and access rights on the Ethereum side chain, and generate a blockchain evidence data chain with timestamp proof.

[0057] IPFS storage uses the content addressing protocol, and each fragment generates a unique CID (Content Identifier), such as "QmXoypizjW3WknFiJnKLwHCnL72vedxjQkDDP1mXWo6uco". Fragments are stored by heat level: High-frequency access data: Reserved on edge nodes (such as local construction site servers), with a response latency <50ms; Low-frequency data: archived to cloud storage and retrieved on demand through the IPFS gateway.

[0058] The Ethereum sidechain (using Polygon) records the following metadata: CID: points to the IPFS storage location; Access Control List (ACL): defines read and write permissions (e.g. the supervisor can read and write, the construction party can only read); Timestamp: Adopts IEEE 1588 precision time protocol with error <1μs.

[0059] After the data is uploaded to the chain, privacy protection is achieved through zero-knowledge proofs (such as zk-SNARKs). For example, when verifying the eligibility of a batch of cement, only the hash value is disclosed without revealing the supplier information, ensuring commercial confidentiality.

[0060] S203, based on the engineering change information in the blockchain evidence data chain, the BIM collaboration platform is updated through the incremental 3D model synchronization technology to generate an enhanced 3D model for real-time interaction among multiple parties, supporting cross-terminal annotation conflict detection and progress simulation visualization; Specifically, the engineering change instructions stored in the blockchain can be parsed, and key parameters including the changed parts and material specifications can be extracted using natural language processing to generate a structured change feature vector; The engineering change instructions in the blockchain evidence data chain exist in the form of unstructured text (such as "change the steel bar specifications in the 3rd floor B area from HRB400Φ20 to HRB500Φ25"). In order to extract key parameters, a natural language processing (NLP) model based on BERT (Bidirectional Encoder Representations from Transformers) is used for semantic analysis. The model is fine-tuned on the Chinese engineering corpus (containing 100,000 change instructions), focusing on identifying entities such as changed parts, material types, and specification parameters.

[0061] Processing flow: Text preprocessing: remove punctuation and stop words, and convert full-width characters to half-width characters, for example, "3 层" is converted to "3 层"; Entity Recognition: Use the Conditional Random Field (CRF) layer to annotate entity boundaries. For example, "3rd floor area B" is annotated as <location>,"HRB400 Φ20” is labeled as <Material_Old>; Relationship extraction: Establish associations between entities through the multi-head attention mechanism. For example, "changed to" connects <Material_Old> and <Material_New>; Structured output: Generate a feature vector in JSON format, including fields: Changed part: Three-dimensional coordinates (x = 120.5, y = 45.3, z = 3.0); Old material: {type: steel bar, grade: HRB400, diameter: 20mm}; New material: {type: steel bar, grade: HRB500, diameter: 25mm}; Effective time: 2023-08-15T14:30:00.

[0062] Perform spatial octree segmentation on the BIM model according to the change feature vector, only update the 3D mesh data of the affected area, and generate a lightweight incremental model package; The BIM model usually contains millions of 3D mesh patches, and full-scale update is inefficient. Use the spatial octree segmentation technology to divide the model into cube blocks with a side length of 1 meter, and only update the blocks that intersect with the changed part.

[0063] Operation process: Spatial index construction: The original BIM model is pre-segmented into an octree structure, and each node stores the geometric and attribute data of the corresponding area; Changed area mapping: Locate the octree nodes to be updated according to the three-dimensional coordinates in the feature vector. For example, the coordinates (150.2, 78.4, 5) fall within the node OCT-3A7B; Local mesh update: Delete old data: Remove all meshes related to the old material within the OCT-3A7B node (such as C30 concrete patches); Insert new data: Generate the geometry of C35 concrete according to the new material parameters and associate the attributes (compressive strength, supplier information); Lightweight compression: Use the Draco algorithm to compress the incremental mesh data, and the compression rate can reach 80%. The structure of the incremental package is as follows: Header: Change version number (such as v2.1.5), list of affected block IDs; Data body: Compressed mesh data, material texture hash value; Checksum: CRC32 checksum to ensure transmission integrity.

[0064] Performance comparison: For full model updates, 500MB of data needs to be transmitted, while for incremental packages, only 15MB is required. The transmission time is reduced from 5 minutes to 20 seconds.

[0065] Deploy a rule inference engine on the collaborative platform, compare the incremental model with the design specification database, detect elevation conflicts and pipeline collisions, and generate a conflict heat map. The rule inference engine is developed based on the Drools framework, integrating building industry specifications (such as the Steel Structure Design Standard GB50017-2017) and project-specific constraints (such as pipeline spacing ≥ 0.5 meters).

[0066] Conflict detection process: Specification rule encoding: Convert text specifications into executable rules.

[0067] Model data import: Convert the geometric data of the incremental model into objects recognizable by the inference engine (such as Pipeline, Beam). Conflict detection execution: The engine traverses all object combinations and triggers matching rules. Detection types include: Elevation conflict: The elevation of the floor and the beam overlaps. Pipeline collision: The water pipe crosses the cable tray. Material conflict: The load-bearing structure is not covered by fireproof materials.

[0068] Heat map generation: Spatial meshing: Divide the model space into grids of 0.1m³. Conflict density calculation: Count the number of conflicts in each grid and map it to a color gradient (green → yellow → red). Visualization overlay: Render the heat map in the BIM model, supporting clicking to view conflict details.

[0069] Example: 3 pipeline spacing conflicts are detected in a certain area. The corresponding grid in the heat map is shown in red. Clicking on it can view the specific pipeline ID and the proposed adjustment plan.

[0070] Integrate the kinematic model of construction machinery, simulate the construction process sequence after changes, and generate the optimal equipment scheduling path through the particle swarm optimization algorithm. The kinematic models of construction machinery (such as tower cranes, concrete pumps) include: Tower crane: slewing speed (0.8rad / s), lifting speed (1.2m / s), working radius (50m); Concrete pump: boom deployment angle (0 - 270°), concrete delivery rate (30m³ / h).

[0071] Process simulation and optimization process: Task decomposition: Decompose the construction tasks involved in the change into atomic operations (such as "lifting steel bars with a tower crane" and "pouring concrete with a pump truck"); Timing constraint modeling: Define task dependencies (such as pouring should start after steel bar binding is completed) and resource constraints (such as the same tower crane cannot serve two areas simultaneously); Application of Particle Swarm Optimization (PSO) algorithm: Particle encoding: Each particle represents a scheduling plan, encoded as an equipment ID - time matrix. For example: Particle 1: [Tower crane A, 09:00 - 09:30, Area X]; [Pump truck B, 09:30 - 10:00, Area X].

[0072] Fitness function: Evaluate the total project duration (weight 0.6), equipment idle rate (weight 0.3), and energy consumption (weight 0.1) of the plan; Iterative optimization: The population size is 100, the inertia weight ω linearly decreases from 0.9 to 0.4, and the learning factors c1 = c2 = 2.0. It converges after 50 generations of iteration; Path visualization: Output Gantt charts and 3D spatial path animations, marking the equipment movement trajectories and key time nodes.

[0073] Adopt WebGL streaming technology to push the incremental model and conflict detection results to each terminal in real time, supporting cross - terminal annotation of conflict detection and progress simulation visualization.

[0074] WebGL streaming is implemented based on the Three.js framework. The key technical points include: Data chunking: Divide the incremental model into multiple LOD (Level of Detail) levels according to octree nodes. Load high - precision models for blocks close to the viewing point and low - precision models for those far away; Progressive transmission: Adopt the HTTP / 2 Server Push technology to preferentially transmit data in the visible area. The transmission protocol is as follows: Metadata frame: Contains block ID, location, and LOD level; Data frame: Compressed geometric data and textures; Conflict data synchronization: The conflict heat map is transmitted independently in JSON format and aligned with the model data through timestamps; Cross - terminal annotation: Annotation protocol: Based on the Operational Transformation (OT) algorithm to achieve multi - user collaborative editing. For example, when user A marks a conflict somewhere on an iPad, the operation (location, annotation content) is synchronized to the PC side in real time; Conflict Resolution: When multiple users annotate the same location, the majority voting mechanism is triggered to retain the high-frequency annotations. Progress Simulation Interaction: Users can view the device positions and construction status at different stages by sliding the timeline, and zooming and rotating the perspective are supported.

[0075] S204. Based on the enhanced three-dimensional model and the Internet of Things environment monitoring data, construct a digital twin mirror space, simulate the construction process through a multi-source data fusion deduction engine, and generate a risk prediction map and a resource allocation optimization strategy.

[0076] Specifically, the enhanced three-dimensional model can be fused with the Internet of Things environment data, and a digital twin basic model with millimeter-level accuracy can be generated through point cloud semantic segmentation. The construction of the digital twin basic model depends on the deep fusion of high-precision point cloud data and real-time Internet of Things monitoring data. First, the construction site is scanned station by station using a three-dimensional laser scanner (such as FARO Focus S 350) to generate the original point cloud data (with a density of 5000 points per square meter). At the same time, Internet of Things sensors (such as temperature and humidity sensors, vibration sensors) collect environmental data (temperature, humidity, mechanical vibration frequency) at a frequency of 10 times per second. After the point cloud data is denoised (outlier points are removed using the RANSAC algorithm), it is aligned in space and time with the Internet of Things data - taking the construction coordinate system as the reference, the time axis is aligned through the dynamic time warping (DTW) algorithm to ensure that the timestamp error of the data is less than ±50ms.

[0077] The point cloud semantic segmentation uses an improved PointNet++ model, and the specific optimizations include: Hierarchical Sampling: The point cloud is divided into octrees (with a depth of 6 layers), and different sampling rates are set for each layer (the sampling rate of the bottom layer is 0.1, and the top layer is 0.8) to retain the key structural features. Multi-modal Feature Fusion: Internet of Things data (such as the concrete curing temperature of 25°C, the stress of the crane boom of 120MPa) is used as an additional feature channel and jointly input into the network together with the point cloud coordinates (x, y, z) and color (RGB). Loss Function Optimization: The Focal Loss is used to alleviate the class imbalance problem (such as the volume ratio differences of steel bars, formwork, and concrete), and α = 0.25 and γ = 2 are set.

[0078] The final generated digital twin model has an accuracy of ±2mm and can clearly distinguish the differences in steel bar diameters (such as 16mm and 20mm ribbed steel bars). For example, in a beam-column joint model, the steel bar arrangement error is controlled within 3mm, and at the same time, a red warning area with abnormal concrete curing temperature is marked.

[0079] Access the real-time streams containing concrete curing monitoring data and crane stress data, and construct a material strength attenuation model and an equipment fatigue prediction model in the digital twin basic model; The material strength attenuation model is constructed based on the principle of concrete hydration heat dynamics: Data input: The concrete curing monitoring data includes temperature (collected every 5 minutes), humidity (accuracy ±3%RH), and age (in days); Strength prediction: Use a time series convolutional network (TCN) to model the strength growth curve. The network input is the temperature-humidity sequence within 72 hours (sampling interval 1 hour), and the output is the predicted values of the compressive strength at 3 days, 7 days, and 28 days (unit: MPa). For example, input the temperature sequence [20°C, 22°C, 25°C...] and the humidity sequence [85%, 80%, 78%...], and predict the 28-day strength to be 35.2 MPa (error ±1.5 MPa); Abnormal warning: When the predicted strength is lower than the design value (e.g., for C30 concrete, it should be ≥30 MPa), trigger a yellow warning and recommend a curing plan (such as covering with plastic film for moisture retention).

[0080] The equipment fatigue prediction model targets the key components of cranes (such as tower cranes): Stress-life curve modeling: Based on the Miner linear cumulative damage theory, input the real-time stress data (sampling frequency 100Hz) to calculate the remaining life of wire ropes and hooks. For example, when the stress amplitude of a certain tower crane hook exceeds 200 MPa, the life loss coefficient increases by 0.05%; Real-time monitoring and feedback: In the digital twin model, high-fatigue areas (such as stress concentration sites) are marked with flashing red lights, and warnings are pushed to the operator terminal in real-time through the edge computing node (NVIDIA Jetson AGX Xavier); Dynamic threshold adjustment: Combine the equipment's historical maintenance records (such as the average first major overhaul cycle of a certain type of tower crane is 8000 hours), and use the exponential smoothing method to dynamically correct the alarm threshold (such as triggering an emergency shutdown command when the remaining life is less than 200 hours).

[0081] Adopt the Monte Carlo method to simulate equipment failure emergencies, calculate the structural safety factor decline curve through finite element analysis, and generate a risk probability distribution map; The Monte Carlo simulation is used to quantify the impact of emergencies: Fault scenario library construction: Define 10 common types of faults (such as tower crane wire rope breakage, concrete pump truck pipe blockage), and set the occurrence probability for each type of fault (such as the probability of wire rope breakage is 0.01% / hour); Random Sampling: Generate a sequence of fault events based on a Poisson process. The simulation period is 30 days (720 hours), and the total number of sampling times is 100,000. For example, in a certain simulation, a pipeline blockage event of a pump truck occurred at the 153rd hour, resulting in a 6-hour shutdown; Impact Assessment: Calculate the impact of each fault on the project duration delay (unit: days) and cost increase (unit: ten thousand yuan), and generate a probability distribution histogram. For example, the simulation results show that the probability of a tower crane fault causing a project duration delay of ≥ 3 days is 12.7%.

[0082] Finite Element Analysis (FEA) Focuses on Structural Safety: Model Parameterization: Import the digital twin model into ANSYS Workbench, and define material properties (such as the elastic modulus of concrete is 30 GPa) and boundary conditions (such as fixed constraints on the foundation); Load Condition Simulation: Set extreme conditions (such as wind speed of 25 m / s, seismic acceleration of 0.3 g), and calculate the stress distribution of key nodes. For example, in a certain high-rise building core tube, the maximum displacement under an 8-level wind load reaches 35 mm, exceeding the code limit of 30 mm; Dynamic Update of Safety Factor: According to real-time monitoring data (such as the development of concrete strength, the corrosion rate of steel bars), use the Back Propagation Neural Network (BPNN) to predict the safety factor decay curve. For example, when it is detected that the thickness of the steel bar protection layer of a certain floor slab is insufficient, the safety factor drops from 1.8 to 1.5, triggering an orange warning.

[0083] The risk probability distribution map is presented in the form of a heat map: Horizontal Axis: Risk Types (Quality, Safety, Progress); Vertical Axis: Risk Levels (Level 1 - 5); Color Depth: Risk Occurrence Probability (0% - 100%). For example, in a certain deep foundation pit project, the risk probability of "instability of the retaining structure" is 8.3%, marked as a dark red area.

[0084] Construct a resource constraint satisfaction model, dynamically adjust the material distribution plan and mechanical scheduling plan in combination with the risk map, and output an optimized resource allocation strategy with redundancy.

[0085] The Resource Constraint Satisfaction Model (RCSP) includes the following core modules: Constraint Condition Library: Hard Constraints: The maximum lifting capacity of the tower crane is 10 tons, and the maximum number of daily transports of the concrete mixer truck is 4 times; Soft Constraints: The utilization rate of the steel bar processing production capacity ≤ 85% (reserving a 15% buffer); Optimization Objectives: Minimize the total cost (weight 0.6); Minimize the risk of project duration delay (weight 0.4); The dynamic adjustment algorithm adopts an improved genetic algorithm (GA): Coding scheme: Encode the delivery plan as a chromosome, and the gene segment represents the material batch (e.g., the first gene: steel bar Φ20mm, quantity 50 tons, arrival time D+3); Fitness function: Among them, the risk-weighted construction period = Σ (construction period of each process × risk probability); Genetic operations: Crossover: Two-point crossover (crossover probability 0.7); Mutation: Gaussian mutation (mutation probability 0.1, standard deviation σ = 2 days); Redundancy design is achieved through "safety stock + standby machinery": Material redundancy: Set a 5% - 10% safety stock for main materials such as steel bars and cement (e.g., planned usage 100 tons, actual purchase 105 tons); Mechanical redundancy: Configure a standby tower crane (rental rate 5000 yuan per shift) for key path processes (such as main structure construction), and immediately activate it when the main tower crane fails; Human resource redundancy: Cross-train workers to ensure that there is at least 1 substitute for key positions (such as welders and scaffolders).

[0086] Output case: The optimization strategies for a high-rise residential project include: Material delivery: Adjust the concrete pouring time from the original plan of 8:00 - 12:00 daily to 6:00 - 10:00, avoiding the morning traffic peak, and the transportation efficiency is increased by 20%; Mechanical scheduling: Configure 2 sets of standby steel wire ropes for tower crane A, and automatically trigger the replacement process when the wear rate is detected to exceed the standard; Risk hedging: Purchase construction period delay insurance to cover the losses caused by suspension of work due to extreme weather (insured amount 2 million yuan, premium 150,000 yuan).

[0087] It can be seen that according to the multi-modal data of the construction scene collected by the intelligent camera and the real-time monitoring information of the Internet of Things sensors, the identification results of illegal operations and the early warning signals of equipment abnormalities are generated; based on the identification results of illegal operations and the early warning signals of equipment abnormalities, an immutable blockchain evidence data chain is generated; according to the project change information in the blockchain evidence data chain, the BIM collaborative platform is updated through the incremental 3D model synchronization technology to generate an enhanced 3D model for multi-party real-time interaction; based on the enhanced 3D model and the Internet of Things environmental monitoring data, a digital twin mirror space is constructed, and the construction process is simulated through the multi-source data fusion deduction engine to generate a risk prediction map and a resource allocation optimization strategy, so as to realize the visual supervision, credible traceability and forward-looking decision-making of the construction process.

[0088] Another embodiment of the present invention provides an intelligent supervision and control system. Refer to Figure 3 , the system may include: A fusion module 301, configured to fuse visual features and device status features through a multi-modal graph neural network according to the multi-modal data of the construction scene collected by an intelligent camera and the real-time monitoring information of an Internet of Things sensor, and generate an illegal operation identification result and a device anomaly warning signal; An encryption module 302, configured to, based on the illegal operation identification result and the device anomaly warning signal, combine the construction material entry record and the concealed project acceptance video data, and use a dynamic fragmentation encryption algorithm to block-chain the key information in chunks to generate an immutable blockchain evidence data chain; An update module 303, configured to update the BIM collaboration platform through an incremental three-dimensional model synchronization technology according to the project change information in the blockchain evidence data chain, generate an enhanced three-dimensional model for multi-party real-time interaction, and support cross-terminal annotation conflict detection and progress simulation visualization; A generation module 304, configured to construct a digital twin mirror space based on the enhanced three-dimensional model and the Internet of Things environment monitoring data, simulate the construction process through a multi-source data fusion deduction engine, and generate a risk prediction map and a resource allocation optimization strategy.

[0089] It can be seen that, according to the multi-modal data of the construction scene collected by an intelligent camera and the real-time monitoring information of an Internet of Things sensor, an illegal operation identification result and a device anomaly warning signal are generated; based on the illegal operation identification result and the device anomaly warning signal, an immutable blockchain evidence data chain is generated; according to the project change information in the blockchain evidence data chain, the BIM collaboration platform is updated through an incremental three-dimensional model synchronization technology to generate an enhanced three-dimensional model for multi-party real-time interaction; based on the enhanced three-dimensional model and the Internet of Things environment monitoring data, a digital twin mirror space is constructed, and the construction process is simulated through a multi-source data fusion deduction engine to generate a risk prediction map and a resource allocation optimization strategy, so as to realize visual supervision, credible traceability, and forward-looking decision-making of the construction process.

[0090] The embodiment of the present invention also provides a storage medium, in which a computer program is stored, and the computer program is configured to execute the steps in any one of the above method embodiments when running.

[0091] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for executing the following steps: S201, according to the multi-modal data of the construction scene collected by an intelligent camera and the real-time monitoring information of an Internet of Things sensor, fuse visual features and device status features through a multi-modal graph neural network, and generate an illegal operation identification result and a device anomaly warning signal; S202. Based on the illegal operation identification result and the device anomaly warning signal, combined with the construction material entry record and the concealed work acceptance video data, use the dynamic fragmented encryption algorithm to block-chain the key information in chunks, and generate an immutable blockchain evidence data chain; S203. According to the project change information in the blockchain evidence data chain, update the BIM collaboration platform through the incremental 3D model synchronization technology to generate an enhanced 3D model for multi-party real-time interaction, supporting cross-terminal annotation conflict detection and progress simulation visualization; S204. Based on the enhanced 3D model and the Internet of Things environment monitoring data, construct a digital twin mirror space, simulate the construction process through a multi-source data fusion deduction engine, and generate a risk prediction map and a resource allocation optimization strategy.

[0092] It can be seen that according to the multi-modal data of the construction scene collected by the intelligent camera and the real-time monitoring information of the Internet of Things sensors, an illegal operation identification result and a device anomaly warning signal are generated; based on the illegal operation identification result and the device anomaly warning signal, an immutable blockchain evidence data chain is generated; according to the project change information in the blockchain evidence data chain, the BIM collaboration platform is updated through the incremental 3D model synchronization technology to generate an enhanced 3D model for multi-party real-time interaction; based on the enhanced 3D model and the Internet of Things environment monitoring data, a digital twin mirror space is constructed, and the construction process is simulated through a multi-source data fusion deduction engine to generate a risk prediction map and a resource allocation optimization strategy, so as to realize the visual supervision, credible traceability and forward-looking decision-making of the construction process.

[0093] The embodiment of the present invention also provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0094] Specifically, the above electronic device may further include a transmission device and an input / output device. Among them, the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0095] Specifically, in this embodiment, the above processor may be configured to execute the following steps through a computer program: S201. According to the multi-modal data of the construction scene collected by the intelligent camera and the real-time monitoring information of the Internet of Things sensors, fuse the visual features and device status features through a multi-modal graph neural network to generate an illegal operation identification result and a device anomaly warning signal; S202. Based on the illegal operation identification result and the device anomaly warning signal, combined with the construction material entry record and the concealed work acceptance video data, use the dynamic fragmented encryption algorithm to block-chain the key information in chunks, and generate an immutable blockchain evidence data chain; S203. Update the BIM collaboration platform through incremental 3D model synchronization technology according to the engineering change information in the blockchain evidence data chain, generate an enhanced 3D model for multi-party real-time interaction, and support cross-terminal annotation conflict detection and progress simulation visualization; S204. Based on the enhanced 3D model and IoT environmental monitoring data, construct a digital twin mirror space, simulate the construction process through a multi-source data fusion deduction engine, and generate a risk prediction map and a resource allocation optimization strategy.

[0096] It can be seen that according to the multi-modal data of the construction scene collected by the intelligent camera and the real-time monitoring information of the IoT sensors, an illegal operation recognition result and an equipment anomaly warning signal are generated; based on the illegal operation recognition result and the equipment anomaly warning signal, an immutable blockchain evidence data chain is generated; according to the engineering change information in the blockchain evidence data chain, the BIM collaboration platform is updated through incremental 3D model synchronization technology to generate an enhanced 3D model for multi-party real-time interaction; based on the enhanced 3D model and IoT environmental monitoring data, a digital twin mirror space is constructed, and the construction process is simulated through a multi-source data fusion deduction engine to generate a risk prediction map and a resource allocation optimization strategy, thereby enabling visual supervision, trustworthy traceability, and forward-looking decision-making of the construction process.

[0097] The structure, features, and function effects of the present invention have been described in detail based on the embodiments shown in the drawings. The above are only the preferred embodiments of the present invention, but the present invention is not limited to the scope defined by the drawings. Any changes made according to the concept of the present invention, or equivalent embodiments modified into equivalent changes, still within the spirit covered by the description and the drawings, shall fall within the protection scope of the present invention.< / location>

Claims

1. A smart supervision and control method, characterized in that, The method includes: Based on the multi-modal data of the construction scene collected by the intelligent camera and the real-time monitoring information of the Internet of Things sensors, the visual features and equipment status features are fused through a multi-modal graph neural network to generate the identification results of illegal operations and equipment anomaly warning signals; Based on the identification results of illegal operations and equipment anomaly warning signals, combined with the construction material entry records and the video data of concealed project acceptance, the key information is divided into blocks and chained using a dynamic fragmentation encryption algorithm to generate an immutable blockchain evidence data chain; According to the engineering change information in the blockchain evidence data chain, the BIM collaboration platform is updated through incremental 3D model synchronization technology to generate an enhanced 3D model for multi-party real-time interaction, supporting cross-terminal annotation conflict detection and progress simulation visualization; Based on the enhanced 3D model and the Internet of Things environmental monitoring data, a digital twin mirror space is constructed, and the construction process is simulated through a multi-source data fusion deduction engine to generate a risk prediction map and a resource allocation optimization strategy.

2. The method according to claim 1, characterized in that, The step of based on the multi-modal data of the construction scene collected by the intelligent camera and the real-time monitoring information of the Internet of Things sensors, fusing the visual features and equipment status features through a multi-modal graph neural network to generate the identification results of illegal operations and equipment anomaly warning signals includes: According to the video timestamp of the intelligent camera and the acquisition frequency of the Internet of Things sensors, an adaptive interpolation algorithm is used to align the visual data stream and the equipment status signal to generate a multi-modal data stream with time series synchronization; Model construction workers, construction machinery, and building materials as graph nodes, with visual features and equipment parameters as node attributes and construction movement line relationships as edge weights to construct a three-dimensional spatio-temporal heterogeneous graph; Based on the three-dimensional spatio-temporal heterogeneous graph, a dual-channel graph attention network is constructed. The first channel extracts the illegal behavior patterns in the visual features, and the second channel extracts the equipment status anomaly patterns. The features are fused through a cross-modal distillation loss function to generate a joint feature vector; According to the construction stage, the classification threshold is dynamically adjusted, and a fuzzy support vector machine is used to classify the joint feature vector to output the identification results of illegal operations and equipment anomaly warning signals in real time.

3. The method according to claim 2, wherein The step of based on the identification results of illegal operations and equipment anomaly warning signals, combined with the construction material entry records and the video data of concealed project acceptance, using a dynamic fragmentation encryption algorithm to divide the key information into blocks and chain them to generate an immutable blockchain evidence data chain includes: According to the key level of the illegal identification results, the video clips, sensor data, and acceptance images are segmented into dynamic fragments of 8KB - 32KB; Generate a unique weight coefficient for each data fragment, and dynamically select an encryption algorithm based on the chaotic mapping algorithm. The greater the weight, the higher the encryption intensity; Adopt a Merkle Patricia tree structure to calculate the double-layer hash value by associating the encrypted fragments with the entry records to generate a globally unique fingerprint across data sources; When the concealed project acceptance data is detected, the consortium chain smart contract is automatically activated, and after verifying the data integrity according to the preset rules, the fragment chaining process is started; Store the encrypted fragments through IPFS, record the hash value and access rights on the Ethereum side chain, and generate a blockchain evidence data chain with timestamp proof.

4. The method according to claim 3, characterized in that, Updating the BIM collaborative platform through incremental 3D model synchronization technology according to the engineering change information in the blockchain evidence storage data chain, generating an enhanced 3D model for multi-party real-time interaction, and supporting cross-terminal annotation conflict detection and progress simulation visualization, including: Analyzing the engineering change instructions in the blockchain evidence storage, extracting key parameters including the changed parts and material specifications using natural language processing, and generating a structured change feature vector; Performing spatial octree segmentation on the BIM model according to the change feature vector, only updating the 3D mesh data of the affected area, and generating a lightweight incremental model package; Deploying a rule inference engine on the collaborative platform, comparing the incremental model with the design specification database, detecting elevation conflicts and pipeline collision problems, and generating a conflict heat map; Integrating a construction machinery kinematics model, simulating the construction process time sequence after the change, and generating the optimal equipment scheduling path through the particle swarm optimization algorithm; Adopting WebGL streaming technology to push the incremental model and conflict detection results to each terminal in real time, supporting cross-terminal annotation conflict detection and progress simulation visualization.

5. The method according to claim 4, wherein Building a digital twin mirror space based on the enhanced 3D model and Internet of Things environment monitoring data, simulating the construction process through a multi-source data fusion deduction engine, and generating a risk prediction map and a resource allocation optimization strategy, including: Fusing the enhanced 3D model with Internet of Things environment data, and generating a digital twin basic model with millimeter-level accuracy through point cloud semantic segmentation; Accessing real-time streams including concrete curing monitoring data and crane stress data, and building a material strength attenuation model and an equipment fatigue prediction model in the digital twin basic model; Simulating equipment failure emergencies using the Monte Carlo method, calculating the structural safety factor decline curve through finite element analysis, and generating a risk probability distribution map; Building a resource constraint satisfaction model, dynamically adjusting the material distribution plan and mechanical scheduling plan in combination with the risk map, and outputting a resource allocation optimization strategy with redundancy.

6. An intelligent supervision and control system, characterized in that, The system includes: A fusion module for generating an illegal operation recognition result and an equipment anomaly warning signal by fusing visual features and equipment status features through a multi-modal graph neural network according to the multi-modal data of the construction scene collected by an intelligent camera and the real-time monitoring information of Internet of Things sensors; An encryption module for, based on the illegal operation recognition result and the equipment anomaly warning signal, combining the construction material entry record and the hidden project acceptance video data, and using a dynamic fragmentation encryption algorithm to chain the key information in blocks, generating an immutable blockchain evidence storage data chain; An update module for updating the BIM collaborative platform through incremental 3D model synchronization technology according to the engineering change information in the blockchain evidence storage data chain, generating an enhanced 3D model for multi-party real-time interaction, and supporting cross-terminal annotation conflict detection and progress simulation visualization; A generation module for building a digital twin mirror space based on the enhanced 3D model and Internet of Things environment monitoring data, simulating the construction process through a multi-source data fusion deduction engine, and generating a risk prediction map and a resource allocation optimization strategy.

7. The system according to claim 6, wherein The fusion module is specifically used for: According to the video timestamps of the intelligent camera and the acquisition frequency of the Internet of Things sensors, an adaptive interpolation algorithm is used to align the visual data stream and the device status signals, generating a multi-modal data stream with synchronized time series. The construction workers, construction machinery and equipment, and building materials are modeled as graph nodes, with visual features and device parameters as node attributes and the construction movement line relationships as edge weights to construct a three-dimensional spatio-temporal heterogeneous graph. Based on the three-dimensional spatio-temporal heterogeneous graph, a dual-channel graph attention network is constructed. The first channel extracts the violation behavior patterns in the visual features, and the second channel extracts the device status anomaly patterns. The features are fused through a cross-modal distillation loss function to generate a joint feature vector. According to the construction stage, the classification threshold is dynamically adjusted, and a fuzzy support vector machine is used to classify the joint feature vector, and the violation operation recognition results and device anomaly warning signals are output in real time.

8. The system according to claim 7, wherein The encryption module is specifically used for: According to the critical level of the violation recognition result, the video clips, sensor data, and acceptance images are segmented into dynamic fragments of 8KB - 32KB. A unique weight coefficient is generated for each data fragment, and the encryption algorithm is dynamically selected based on the chaotic mapping algorithm. The greater the weight, the higher the encryption intensity. The Merkle Patricia tree structure is adopted to calculate the double-layer hash value by associating the encrypted fragments with the entry records, generating a globally unique fingerprint across data sources. When the concealed project acceptance data is detected, the consortium chain intelligent contract is automatically activated, and after verifying the data integrity according to the preset rules, the fragment uploading process is started. The encrypted fragments are stored through IPFS, and the hash value and access permissions are recorded on the Ethereum side chain, generating a blockchain evidence data chain with timestamp proof.

9. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is set to execute the method described in any one of claims 1 - 5 when running.

10. An electronic device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is set to run the computer program to execute the method described in any one of claims 1 - 5.

Images