BIM-based electromechanical installation resource management and control system

Abstract

The invention discloses a BIM-based electromechanical installation resource management and control system, and the system comprises a data sensing layer which integrates a laser radar, an RFID, a UWB and an environment sensor, collects multi-modal data, and forms a total-factor digital twin base data stream after the multi-modal data is preprocessed by a mobile edge computing node; the model fusion layer is used for binding the construction progress with BIM component attributes, dynamically deducing a component-level progress path through an AI algorithm, identifying key path conflicts, integrating ERP resource data to generate a thermodynamic diagram and a cost deviation curve, and automatically detecting pipeline collision, progress lag and cost hyper-branched risks through a rule engine machine learning dual-mode engine; triggering multi-stage early warning and dynamically adjusting a threshold value; the intelligent decision-making layer is used for predicting the progress deviation probability of the future seven days by adopting LSTM, generating a non-tampering deviation correction scheme in combination with a block chain, and realizing hierarchical response through a yellow-orange-red three-level alarm system; and the user interaction layer is used for realizing multi-terminal collaboration and executing verification data feedback through AR glasses, a mobile terminal and a large-screen monitoring center.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical and electrical installation of hydropower engineering, and particularly relates to a mechanical and electrical installation resource management and control system based on BIM. BACKGROUND

[0002] In hydropower engineering, resource management and control needs to coordinate multi-dimensional factors such as accurate positioning of large equipment, long-distance material flow control, special environmental parameter adaptation, etc., to ensure that the progress, cost and safety targets of key paths such as dam pouring, unit installation and underground chamber construction are achieved synchronously.

[0003] However, the traditional mechanical and electrical installation resource management and control system generally has weak early warning capability in the hydropower engineering scene; for example, a large hydropower station project once failed to detect the space interference between the stator and the rotor in the installation stage of the hydroelectric generating unit in real time, resulting in a 20-day delay in the work period and an increase in costs of more than 3 million yuan; another example is that a pumped storage power station project failed to timely warn due to lagging progress prediction, causing key equipment such as the speed regulation system to arrive late and unable to be installed on time, resulting in a delay in the entire link work period and causing serious economic and property losses, and therefore a mechanical and electrical installation resource management and control system based on BIM is proposed. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art and provide a mechanical and electrical installation resource management and control system based on BIM.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: A mechanical and electrical installation resource management and control system based on BIM, comprising: A data perception layer: by integrating laser radar, RFID, UWB and environmental sensor arrays, real-time multi-modal data of spatial deviation of equipment installation, material flow path, construction environment parameters are collected, after local preprocessing by a mobile edge computing node, full-element digital twin base data flow is transmitted to a model fusion layer, providing high-fidelity basic data support for BIM model point cloud fusion and physical-digital mapping, forming a bottom layer data source for "physical-digital" two-way mapping; A model fusion layer: based on real-time data transmitted by the data perception layer, construction progress plan and BIM component attributes are bound, AI algorithm is used to dynamically deduce component-level progress path and identify key path conflicts; ERP resource data and cost budget are synchronously integrated, resource consumption heat map and cost deviation curve are generated; through rule engine + machine learning dual-mode conflict early warning engine, risks such as pipeline collision, progress lag, cost overrun are automatically detected, multi-level early warning is triggered and early warning threshold is dynamically adjusted, the deduced results and early warning signals are transmitted to an intelligent decision layer, realizing "model-progress-resource" three-synchronous intelligent linkage; Intelligent decision layer: receives the dynamic reasoning results and early warning signals transmitted by the model fusion layer, uses LSTM neural network to predict the progress deviation probability in the next 7 days, combines with the blockchain technology to generate an unalterable deviation correction scheme (such as construction sequence adjustment, resource allocation); through the "yellow-orange-red" three-level alarm system, the red alarm directly triggers the shutdown instruction and is pushed to the user interaction layer in real time through the 5G network, at the same time, the execution verification data fed back by the user interaction layer is received for effect evaluation and algorithm optimization, forming a "prediction-decision-execution-verification" closed-loop management chain, ensuring that the decision is traceable and the risk is controllable; User interaction layer: relying on the decision instructions and alarm signals issued by the intelligent decision layer, multi-terminal collaboration is realized through the digital twin visualization platform: AR glasses superimpose BIM models and real-time field images to guide workers to install accurately; the mobile terminal integrates voice broadcast and picture annotation functions, real-time pushes early warning information and receives on-site operation feedback; the large screen monitoring center integrates progress reasoning dynamics, resource heat map, risk warning map and other multi-dimensional visual dashboards, supports remote decision-making and execution verification, and reversely feeds back the on-site operation results and verification data to the intelligent decision layer, finally realizes the whole life cycle intelligent management and control from design to operation, and forms a complete closed-loop link of "data perception-model fusion-intelligent decision-user interaction".

[0006] The above technical solution further comprises: Further, the real-time collection of multi-modal data of spatial deviation of equipment installation, material flow path, and construction environment parameters through the integration of laser radar, RFID, UWB, and environmental sensor array comprises the following steps: Obtain three-dimensional point cloud data of the equipment installation site through high-precision scanning, and calculate the spatial deviation of the equipment and the preset position of the BIM model in real time, capture abnormal states such as equipment tilt and displacement; The RFID tag binds materials / equipment, and the UWB base station tracks the flow path of materials from the warehouse to the installation point in real time through the time difference of signal arrival, records the movement trajectory, residence time, and final installation position of each batch of materials, and forms a full-process traceability data chain; Deploy temperature and humidity sensors to monitor the installation environment (such as concrete curing temperature and equipment operating environment), vibration sensors to detect the stability of the equipment installation foundation, and light sensors to evaluate the influence of light conditions on installation accuracy, and all environmental parameters are collected synchronously.

[0007] Further, after local preprocessing by the mobile edge computing node, the full-factor digital twin base data stream is transmitted to the model fusion layer, comprising the following steps: The mobile edge computing node receives real-time multi-modal raw data streams from lidar (spatial deviation data), RFID / UWB (material flow path), and environmental sensor array (temperature / humidity / vibration / illumination parameters). The node first performs data integrity checks, such as checking whether the lidar point cloud is missing key frames, whether the RFID path data contains valid timestamps, and whether the environmental parameters are within a reasonable range (e.g., temperature -40°C~80°C). Invalid or abnormal data is removed to ensure the reliability of the input source.

[0008] Noise filtering is performed on the raw data (e.g., outlier removal in lidar point cloud), missing value interpolation (e.g., linear interpolation to recover the trajectory when UWB positioning data is lost), and abnormal value correction (e.g., smoothing of sudden high values in environmental sensors using historical mean value). Based on the unified timestamp, the multi-source data is aligned to ensure the synchronization of "device spatial position-material flow node-environment parameter". For example, the pipe three-dimensional coordinates scanned by lidar are matched with the material arrival position recorded by RFID at the same time, forming a spatio-temporal consistency data chain of "space-process-environment".

[0009] Through point cloud registration algorithm (e.g., ICP iterative closest point), the real-time point cloud of lidar is fused with the initial point cloud of BIM model to generate a high-precision digital twin base. At the same time, the material flow path (e.g., "material A arrives at installation point B at 10:00") and environmental parameters (e.g., "current temperature 25°C, humidity 60%") are embedded as dynamic attributes into the corresponding BIM components, forming a three-dimensional fusion of "spatial position-process trajectory-environment state" full-element data stream. A quality label is attached to each data packet to mark the reliability and update frequency of the data source, supporting high-fidelity applications in the subsequent model fusion layer. The full-element digital twin base data stream is packaged into structured data packets according to a unified protocol, including spatial deviation matrix, material flow coordinate sequence, and environmental parameter time sequence, ensuring standardized data format and direct parsing by the model fusion layer. A lightweight encryption algorithm is used to encrypt the data packets, which are transmitted in real time to the model fusion layer through 5G network or gigabit Ethernet. The edge computing node's caching mechanism is used to handle network fluctuations during transmission, ensuring that data is not lost or delayed, and providing high-fidelity, traceable basic data support for the model fusion layer, directly driving subsequent BIM point cloud fusion and physical-digital mapping analysis.

[0010] Further, the model fusion layer associates and maps the construction schedule plan (such as WBS decomposed processes, time nodes) with the attributes of specific components in the BIM model (such as component ID, spatial coordinates, installation sequence); for example, the installation process of a certain section of pipeline is marked as "3rd day-5th day" in the schedule plan, which is bound to the "installation time attribute" of the corresponding pipeline component in BIM, forming a "schedule-component" one-to-one correspondence, ensuring that subsequent deduction can be accurately positioned to specific components; Based on the bound data, AI algorithms are used to dynamically simulate the component-level schedule path; the algorithm calculates the deviation between the planned installation time of each component and the actual collected data, and identifies conflicts on the critical path; for example, if the planned installation times of two adjacent components overlap and the spatial positions conflict (such as pipeline intersection), the system will automatically mark it as a "critical path conflict" and calculate the probability of time delay caused by the conflict; Synchronize resource data (manpower hours, material inventory, equipment utilization) in the ERP system with cost budget data, aggregate resource consumption by component, area, and time dimension, generate "resource consumption heat map" (show material consumption in a certain area through color depth) and "cost deviation curve", these visual charts can intuitively reflect the risk areas of resource over-consumption or cost overrun.

[0011] Further, the rule engine + machine learning dual-mode conflict warning engine automatically detects pipeline collision, schedule lag, and cost overrun risk, triggers multi-level warning and dynamically adjusts the warning threshold, including the following steps: Rule engine hard conflict detection Pre-set rule library: based on industry standards and project customization requirements, establish a rule library, for example: pipeline spacing ≥ 10 cm, daily progress deviation ≤ 5%, cost overrun threshold ≤ 3% of the budget.

[0012] Real-time rule matching: compare the real-time data transmitted by the data perception layer (such as pipeline spatial coordinates scanned by laser radar, daily resource consumption recorded by ERP) with the rule library. If a rule violation is detected, immediately trigger a "pipeline collision" warning; if the daily progress completion rate is less than 95%, trigger a "schedule lag" warning.

[0013] Machine learning engine forward-looking risk prediction Historical data training: use historical project data to train the machine learning model, the model learns the implicit rules such as "continuous 3-day schedule lag will result in a probability of total project duration overrun > 30%" and "material loss rate increases by 15% during the rainy season".

[0014] Real-time risk prediction: input current data into the trained model to predict future risk probability. For example, if the model predicts that the progress lag probability in a certain area in the next 7 days is >80%, generate a "high risk of progress lag" warning; if the predicted cost overrun probability is >60%, generate a "cost overrun warning".

[0015] Multi-level warning trigger and hierarchical response Warning grading standards: combine rule engine and machine learning results, and grade according to risk severity: Yellow warning: rule engine trigger or machine learning prediction risk probability 20%-50%, prompt potential risk, require on-site verification; Orange warning: rule engine and machine learning trigger at the same time or predicted risk probability 50%-80%, need human intervention to develop rectification scheme; Red warning: rule engine triggers hard conflict or machine learning predicts risk probability >80%, directly associate shutdown instruction and push to responsible person.

[0016] Warning signal packaging: package warning type, risk location, probability value, and recommended measures into a structured data packet, and synchronize to the user interaction layer (AR glasses / mobile terminal / large screen) through the digital twin platform.

[0017] Dynamic threshold self-adaptive adjustment mechanism Threshold feedback optimization: the system dynamically adjusts the threshold value according to historical warning effect (such as the consistency of actual risk occurrence frequency and predicted probability). For example, if a certain type of "progress lag" warning is frequently false, the system analyzes the reasons through the machine learning model and automatically adjusts the warning threshold in this scenario; if the "cost overrun" warning accuracy rate is continuously >90%, the threshold value is maintained or slightly increased to avoid false negatives.

[0018] Threshold update period: the threshold adjustment is based on "day", combined with the changes in construction environment (such as increased material loss due to increased humidity during the rainy season) and real-time feedback data (such as deviation values verified by on-site operations), to ensure that the threshold value always matches the actual risk level and avoids the rigidity of fixed threshold values.

[0019] Further, the receiving model fusion layer transmits the dynamic deduction result and the warning signal, and uses the LSTM neural network to predict the progress deviation probability in the next 7 days, including the following steps: Receiving dynamic deduction result and warning signal The data such as component-level progress path, critical path conflict identification, resource consumption heat map, and cost deviation curve transmitted by the model fusion layer are received by the intelligent decision layer in the form of structured data stream. For example, the "planned installation time-actual installation time deviation value" Δt = actual time - planned time of a component is the core input of progress deviation.

[0020] LSTM neural network progress deviation probability prediction: ) wherein, is the future 7th day progress deviation probability vector; Softmax is an activation function that converts linear output into a probability distribution; is the weight matrix of the hidden layer to the output layer; is the bias term; is the hidden state at time t, which matches the dimension of the output layer, and is used to pass time series dependency information; wherein, is the output gate, which controls the proportion of cell state information output: wherein, σ is the Sigmoid activation function; is the output gate weight matrix; is the output gate bias vector; is the cell state, the "memory" core of LSTM, which is dynamically updated through the input gate and the forget gate wherein is the candidate state; and are the input gate and the forget gate, which control the writing of new information and the retention of old information: wherein, is the input gate weight matrix; is the input gate bias vector;[ , ] is the spliced vector, is the previous time step hidden state; is the current input; is the forget gate weight matrix; is the forget gate bias vector; LSTM is trained by historical progress deviation data to capture time dependency and predict the progress deviation probability of each component in the future 7 days, providing a quantitative basis for risk classification.

[0021] Further, the combination of blockchain technology generates an unalterable correction scheme, adopts a yellow-orange-red three-level alarm system, and receives execution verification data from the user interaction layer for effect evaluation and algorithm optimization, including the following steps:​ Blockchain rectification scheme generation mechanism Scheme structured packaging: the intelligent decision-making layer generates rectification schemes based on dynamic inference results (such as progress lag prediction, resource conflict analysis). The scheme content includes specific adjustment measures, responsible person, execution time window and expected effect. Each scheme is a block of the blockchain, and the block body contains scheme details, generation timestamp, and responsible person digital signature. The block header contains the hash value of the previous block, forming a chain structure.

[0022] Tamper-proofing: SHA-256 algorithm is used to calculate the block hash value. Any modification of the scheme content will cause the hash value to change, which will be detected by other nodes in the network and rejected, ensuring that the scheme is tamper-proof from generation to execution, meeting the audit compliance and responsibility tracing requirements.

[0023] Three-level alarm system hierarchical response logic Yellow warning (potential risk): When the machine learning predicts the progress lag probability of 20-50% or the rule engine detects a slight rule violation (such as material consumption close to the budget threshold), the system triggers a yellow warning, and pushes a prompt message through the mobile terminal, requiring the on-site responsible person to check the reason and submit feedback.

[0024] Orange warning (moderate risk): If the predicted risk probability is 50-80% or the rule conflict intensifies (such as multiple regional progress deviations stacking up), the system triggers an orange warning, requiring manual intervention to develop detailed rectification schemes, and displaying risk heat maps and resource allocation suggestions through the large screen monitoring center.

[0025] Red warning (high risk emergency): When a hard conflict (such as pipeline collision, cost overrun exceeding budget by 10%) or predicted risk probability > 80% is detected, the system directly triggers a red warning, automatically generates a stop work order and pushes it to the responsible person's mobile terminal and AR glasses through the 5G network, forcing the site to suspend construction until the problem is resolved.

[0026] Execution verification data feedback and algorithm optimization closed loop On-site operation feedback collection: After receiving the stop work order, the user interaction layer receives the rectification operation feedback from the on-site personnel, and the operation process is real-time collected. The execution verification data is transmitted back to the intelligent decision-making layer through an encrypted channel, serving as an input source for algorithm optimization.

[0027] Effect evaluation and model optimization: The intelligent decision-making layer compares the predicted results with the actual execution data to evaluate the effectiveness of the correction scheme (e.g., whether the progress deviation is reduced, and whether the cost is controlled within the budget). Based on the evaluation results, the system dynamically adjusts the LSTM model parameters (weight matrix, bias term) and the early warning threshold (according to the historical false alarm rate, the trigger threshold of a certain type of risk is lowered), forming a "prediction-execution-verification-optimization" self-learning closed loop, continuously improving the early warning accuracy and decision-making scientificity.

[0028] Further, the user interaction layer adopts multi-terminal collaborative data transmission. The decision-making instructions and alarm signals issued by the intelligent decision-making layer are standardized packaged through the digital twin visualization platform, forming a unified data format. The platform automatically adapts the data content according to the terminal type: AR glasses focus on spatial superposition instructions, mobile terminals focus on early warning information and feedback channels, and large screens focus on global visualization data. The data is synchronized in real time to each terminal through 5G network or enterprise private network, ensuring the timeliness of the "decision-making-execution-verification" link. AR glasses capture real-time images through built-in cameras. The digital twin platform aligns the BIM model three-dimensional components with the live images, generating a "virtual-reality" superimposed view. Workers wearing AR glasses can directly see the accurate position of "virtual components" in the real scene (e.g., "the pipeline should be installed here, with a deviation of ≤2mm"), while receiving voice instructions (e.g., "rotate 5 degrees to the left"). If installation deviation is detected, the system automatically triggers secondary superposition correction until the accuracy meets the standard. The mobile terminal integrates a voice broadcast module, which converts alarm signals into natural language (e.g., "red alert: A area pipeline installation progress lags behind, triggering a stop work order") and superimposes picture annotations in real time. Workers can input operation feedback (e.g., "stop work has been done, checking the reason") through the mobile terminal touch screen. The feedback data is encrypted and returned to the intelligent decision-making layer for verification. In addition, the mobile terminal supports historical warning query and correction scheme viewing, forming a complete closed loop of "push-feedback-record".

[0029] The large screen monitoring center integrates multi-dimensional visualization dashboards, including: Progress deduction dynamic dashboard: shows the component-level progress path deduction results in the form of a time axis, with color gradient indicating "planned-actual-predicted" progress deviation (e.g., green for normal, red for severe lag); Resource consumption heat map: visually displays the resource (manpower / materials / equipment) consumption intensity in each area through a heat map, assisting in resource allocation decisions; Risk early warning map: superimposes risk indicators (e.g., "pipeline collision point" "progress lag area") on the BIM model, supporting click-to-view details and displaying correction schemes in conjunction.

[0030] Remote decision-makers can interact in real time through large screens (such as dragging to adjust the construction sequence), and decision instructions are synchronized to the intelligent decision layer to generate new deviation correction schemes and push them to the field terminal for execution, forming a closed-loop management link of "remote decision-making - field execution - real-time verification".

[0031] The present application has the following advantages: In the present application, integrated laser radar, RFID, UWB and environmental sensor array are used to collect multi-modal data such as equipment space deviation, material flow path, construction environment parameters, etc. in real time, and form a full-factor digital twin base data stream through local preprocessing by mobile edge computing nodes. Construction progress is bound to BIM component attributes, and AI algorithm is used to dynamically deduce component-level progress path and identify key path conflicts. ERP resources and cost data are integrated to generate resource consumption heat map and cost deviation curve. Through rule engine + machine learning dual-mode conflict warning engine, pipeline collision, progress lag, cost overrun and other risks are automatically detected and multi-level warning is triggered. LSTM neural network is used to predict the progress deviation probability in the next 7 days, and blockchain technology is used to generate an unalterable deviation correction scheme. Through a three-level alarm system of yellow-orange-red, risk classification response is realized. Red alarm directly triggers a stop work order, and receives execution verification data for effect evaluation and algorithm optimization, forming a "prediction - execution - verification - optimization" closed loop, effectively solving the problem of weak early warning capability in the existing mechanical and electrical installation resource management system. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 A system block diagram of a BIM-based mechanical and electrical installation resource management system is provided. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0034] Please refer to Figure 1 As shown in the drawings, the present application is a BIM-based mechanical and electrical installation resource management system, which comprises: A BIM-based mechanical and electrical installation resource management system comprises: Data perception layer: By integrating laser radar, RFID, UWB and environmental sensor array, real-time collection of equipment installation space deviation, material flow path, construction environment parameter multi-modal data, after local preprocessing by mobile edge computing node, full-factor digital twin base data stream is transmitted to model fusion layer, providing high-fidelity basic data support for BIM model point cloud fusion and physical-digital mapping, forming the bottom layer data source of "physical-digital" two-way mapping; Model fusion layer: Based on the real-time data transmitted by the data perception layer, the construction progress plan is bound with the BIM component attributes, and the AI algorithm is used to dynamically deduce the component-level progress path and identify the critical path conflict; synchronously integrate ERP resource data and cost budget, generate resource consumption heat map and cost deviation curve; through the rule engine + machine learning dual-mode conflict warning engine, automatically detect pipeline collision, progress lag, cost overrun and other risks, trigger multi-level warning and dynamically adjust the warning threshold, transmit the deduction results and warning signals to the intelligent decision layer, realize the three-synchronous intelligent linkage of "model-progress-resource"; Intelligent decision layer: receiving the dynamic deduction results and warning signals transmitted by the model fusion layer, using LSTM neural network to predict the progress deviation probability in the next 7 days, combining with the block chain technology to generate an unalterable correction scheme (such as construction sequence adjustment, resource allocation); through the "yellow-orange-red" three-level alarm system, the red alarm directly triggers the stop work instruction and is pushed to the user interaction layer in real time through the 5G network, at the same time, receiving the execution verification data feedback from the user interaction layer for effect evaluation and algorithm optimization, forming a closed-loop management and control chain of "prediction-decision-execution-verification", ensuring that the decision is traceable and the risk is controllable; User interaction layer: relying on the decision instructions and alarm signals issued by the intelligent decision layer, realizing multi-terminal collaboration through the digital twin visualization platform: AR glasses superimpose BIM model and real-time scene, guiding workers to install accurately; mobile terminal integrates voice broadcast and picture annotation functions, real-time push warning information and receive on-site operation feedback; large screen monitoring center integrates progress deduction dynamic, resource heat map, risk warning map and other multi-dimensional visual dashboards, supporting remote decision-making and execution verification, feeding back the on-site operation results and verification data to the intelligent decision layer, finally realizing the whole life cycle intelligent management and control from design to operation, and forming a complete closed-loop link of "data perception-model fusion-intelligent decision-user interaction".

[0035] In one embodiment, the real-time collection of equipment installation space deviation, material flow path, construction environment parameter multi-modal data by integrating laser radar, RFID, UWB and environmental sensor array includes the following steps: Obtain three-dimensional point cloud data of equipment installation site by high-precision scanning, real-time calculate the spatial deviation of equipment and BIM model preset position, capture equipment tilt, displacement and other abnormal states; RFID tag binding materials / equipment, UWB base station through signal arrival time difference real-time tracking of materials from warehouse to installation point, recording the movement trajectory, residence time and final installation position of each batch of materials, forming a full-process traceability data chain; Deploy temperature and humidity sensors to monitor the installation environment (such as concrete curing temperature, equipment operating environment), vibration sensors to detect the stability of the equipment installation foundation, and light sensors to evaluate the impact of light conditions on the installation precision. All environmental parameters are collected synchronously.

[0036] In one embodiment, after the mobile edge computing node local preprocessing, the full-element digital twin base data flow is transmitted to the model fusion layer, including the following steps: The mobile edge computing node receives real-time multi-modal raw data streams from laser radar (spatial deviation data), RFID / UWB (material flow path), and environmental sensor array (temperature and humidity / vibration / light parameters). The node first performs data integrity check, such as checking whether the laser radar point cloud is missing key frames, whether the RFID path data contains valid timestamps, and whether the environmental parameters are within a reasonable range (such as temperature -40℃~80℃), and eliminating invalid or abnormal data to ensure the reliability of the input source;

[0037] Noise filtering (such as outlier removal in laser radar point cloud), missing value interpolation (such as linear interpolation to recover the trajectory when UWB positioning data is lost), and abnormal value correction (such as smoothing the history mean value when the environmental sensor has a sudden high value) are performed on the raw data. Align multi-source data based on a unified timestamp to ensure synchronization of "device spatial position-material flow node-environmental parameter". For example, match the pipe three-dimensional coordinates scanned by laser radar with the material arrival position recorded by RFID at the same time to form a spatio-temporal consistency data chain of "space-process-environment";

[0038] Fuse the real-time point cloud of laser radar with the initial point cloud of BIM model through point cloud registration algorithm (such as ICP iterative closest point) to generate high-precision digital twin base; at the same time, embed the material flow path (such as "material A arrives at installation point B at 10:00") and environmental parameters (such as "current temperature 25℃, humidity 60%") as dynamic attributes into the corresponding BIM component to form a three-dimensional fusion of full-element data flow of "spatial position-process trajectory-environmental state"; Attach a quality label to each data packet to mark the reliability and update frequency of the data source, supporting high-fidelity applications in the subsequent model fusion layer; The full-factor digital twin base data stream is packaged as a structured data packet according to a unified protocol, including core fields such as a spatial deviation matrix, a material flow coordinate sequence, and an environmental parameter time sequence, to ensure that the data format is standardized and can be directly parsed by the model fusion layer; The data packet is encrypted using a lightweight encryption algorithm and transmitted in real time to the model fusion layer through a 5G network or a gigabit Ethernet network. The cache mechanism of the edge computing node is used to cope with network fluctuations during transmission, ensuring that data is not lost or delayed, and ultimately providing high-fidelity, traceable basic data support for the model fusion layer, directly driving subsequent BIM point cloud fusion and physical-digital mapping analysis.

[0039] In one embodiment, the model fusion layer correlates and maps the construction schedule plan (such as WBS decomposed processes and time nodes) with the attributes of specific components in the BIM model (such as component ID, spatial coordinates, and installation sequence); for example, the installation process of a certain section of pipe is marked as "3rd day-5th day" in the schedule plan, which is bound to the "installation time attribute" of the corresponding pipe component in the BIM, forming a one-to-one correspondence between "schedule" and "component", ensuring that subsequent reasoning can accurately locate to specific components; Based on the bound data, AI algorithms are used to dynamically simulate component-level schedule paths; the algorithm calculates the deviation between the planned installation time of each component and the actual collected data, and identifies conflicts on the critical path; for example, if the planned installation times of two adjacent components overlap and the spatial positions conflict (such as pipeline intersection), the system will automatically mark it as a "critical path conflict" and calculate the probability of schedule delay caused by the conflict; Synchronize resource data (man-hour, material inventory, equipment utilization) and cost budget data in the ERP system, aggregate resource consumption by component, area, and time dimension, generate "resource consumption heat map" (display material consumption in a certain area through color depth) and "cost deviation curve", these visual charts can intuitively reflect the risk areas of resource over-consumption or cost overrun.

[0040] In one embodiment, the rule engine + machine learning dual-mode conflict warning engine automatically detects pipeline collision, schedule lag, and cost overrun risk, triggers multi-level warning and dynamically adjusts the warning threshold, including the following steps: Rule engine hard conflict detection Pre-set rule library: based on industry standards and project customization requirements, establish a rule library, for example: pipeline spacing ≥ 10 cm, single-day schedule deviation ≤ 5%, cost overrun threshold ≤ 3% of budget.

[0041] Real-time rule matching: Compare real-time data transmitted by the data perception layer (such as pipeline spatial coordinates scanned by laser radar, daily resource consumption recorded by ERP) with the rule library. If a rule violation is detected, trigger a "pipeline collision" warning immediately; if the daily progress completion rate is less than 95%, trigger a "progress lag" warning.

[0042] Machine learning engine forward-looking risk prediction Historical data training: Train machine learning models using historical project data, and the models learn implicit rules such as "a 3-day progress lag will result in a 30% probability of total project duration extension" and "material loss rate increases by 15% during the rainy season."

[0043] Real-time risk prediction: Input current data into the trained model to predict future risk probability. For example, if the model predicts a 80% probability of progress lag in a certain area in the next 7 days, generate a "high risk of progress lag" warning; if the predicted cost overrun probability is >60%, generate a "cost overrun warning".

[0044] Multi-level warning triggering and hierarchical response Warning classification criteria: Based on the results of rule engine and machine learning, classify by risk severity: Yellow warning: Rule engine triggered or machine learning predicted risk probability 20%-50%, indicating potential risk, requiring on-site verification; Orange warning: Rule engine and machine learning triggered simultaneously or predicted risk probability 50%-80%, requiring human intervention to develop a correction plan; Red warning: Rule engine triggered hard conflict or machine learning predicted risk probability >80%, directly linked to stop work order and pushed to responsible person.

[0045] Warning signal packaging: Package warning type, risk location, probability value, and recommended measures into a structured data package and synchronize to the user interaction layer (AR glasses / mobile devices / large screens) through the digital twin platform.

[0046] Dynamic threshold self-adaptive adjustment mechanism Threshold feedback optimization: The system dynamically adjusts the threshold based on historical warning effectiveness (such as the consistency of actual risk occurrence frequency and predicted probability). For example, if a certain type of "progress lag" warning is frequently false, the system analyzes the cause through the machine learning model and automatically adjusts the warning threshold for that scenario; if the "cost overrun" warning accuracy rate is consistently >90%, maintain or slightly increase the threshold to avoid false negatives.

[0047] Threshold update cycle: Threshold adjustment is in "day" unit, combined with the construction environment changes of the day (such as the increase of material loss caused by the humidity in the rainy season) and real-time feedback data (such as the deviation value verified by on-site operation), to ensure that the threshold always fits the actual risk level, avoiding the rigid defects of fixed threshold.

[0048] In one embodiment, the dynamic inference result and early warning signal transmitted by the model fusion layer are received, and an LSTM neural network is used to predict the progress deviation probability in the next 7 days, including the following steps: Dynamic inference result and early warning signal receiving The component-level progress path, critical path conflict identification, resource consumption heat map and cost deviation curve data transmitted by the model fusion layer are received by the intelligent decision layer in the form of structured data flow. For example, the "planned installation time-actual installation time deviation value" Δt = actual time-planned time of a component is used as the core input of progress deviation.

[0049] LSTM neural network progress deviation probability prediction: ) Wherein, is the progress deviation probability vector in the next 7 days; Softmax is an activation function that converts linear output to probability distribution; is the weight matrix from the hidden layer to the output layer; is the bias term; is the hidden state of time t, which matches the dimension of the output layer and is used to pass time series dependency information; Wherein, is the output gate, which controls the information output proportion of the cell state : Wherein, σ is the Sigmoid activation function; is the output gate weight matrix; is the output gate bias vector; is the cell state, the "memory" core of LSTM, which is dynamically updated through the input gate and the forget gate : Wherein is the candidate state; and are the input gate and the forget gate, which control the writing of new information and the retention of old information: wherein, is an input gate weight matrix; is an input gate bias vector; , is a concatenation vector, is a previous time step hidden state; is a current input; is a forget gate weight matrix; is a forget gate bias vector; The LSTM is trained by historical progress deviation data, captures time dependence, predicts the probability of progress deviation of each component in the next 7 days, and provides a quantitative basis for risk classification.

[0050] In one embodiment, the blockchain technology is used to generate an unalterable correction scheme, a yellow-orange-red three-level alarm system is adopted, and execution verification data received from the user interaction layer is used for effect evaluation and algorithm optimization, including the following steps: Blockchain correction scheme generation mechanism Scheme structured packaging: the intelligent decision layer generates a correction scheme based on dynamic deduction results (such as progress lag prediction and resource conflict analysis), and the scheme content includes specific adjustment measures, responsible person, execution time window, and expected effect. Each scheme is a block of the blockchain, the block body contains scheme details, generation timestamp, and digital signature of the responsible person, and the block header contains the hash value of the previous block, forming a chain structure.

[0051] Unalterable guarantee: SHA-256 algorithm is used to calculate the block hash value, any modification of the scheme content will cause the hash value to change, which is detected by other nodes in the network and rejected, ensuring that the scheme is unalterable from generation to execution, meeting the audit compliance and responsibility tracing requirements.

[0052] Three-level alarm system grading response logic Yellow warning (potential risk): when the machine learning predicts the progress lag probability of 20-50% or the rule engine detects a slight rule violation (such as material consumption close to the budget threshold), the system triggers a yellow warning, pushes a prompt message through the mobile terminal, and requires the on-site responsible person to check the reason and submit feedback.

[0053] Orange warning (moderate risk): if the predicted risk probability is 50-80% or the rule conflict is intensified (such as multiple regional progress deviation superposition), the system triggers an orange warning, which requires manual intervention to develop a detailed correction scheme, and displays a risk heat map and resource allocation suggestion through the large screen monitoring center.

[0054] Red alert (high risk emergency): When hard conflict (such as pipeline collision, cost overrun by more than 10% of the budget) is detected or the predicted risk probability is > 80%, the system directly triggers a red alert, automatically generates a stop work order and pushes it to the mobile terminal and AR glasses of the responsible person in real time through the 5G network, forcing the site to suspend construction until the problem is resolved.

[0055] Execution verification data feedback and algorithm optimization closed loop On-site operation feedback collection: After receiving the stop work order, the user interaction layer receives the stop work order, the on-site personnel execute the correction operation, and the execution verification data is collected in real time during the operation. These data are returned to the intelligent decision-making layer through an encrypted channel as the input source for algorithm optimization.

[0056] Effect evaluation and model optimization: The intelligent decision-making layer compares the predicted results with the actual execution data to evaluate the effectiveness of the correction scheme (such as whether the progress deviation is reduced, and whether the cost is controlled within the budget). Based on the evaluation results, the system dynamically adjusts the LSTM model parameters (weight matrix, bias term) and the warning threshold (according to the historical false alarm rate, the trigger threshold of a certain type of risk is lowered), forming a "prediction-execution-verification-optimization" self-learning closed loop, continuously improving the warning accuracy and decision-making scientificity.

[0057] In one embodiment, the user interaction layer adopts multi-terminal collaborative data transmission. The decision-making instructions and alarm signals issued by the intelligent decision-making layer are standardized packaged through the digital twin visualization platform, forming a unified data format; the platform automatically adapts the data content according to the terminal type: AR glasses focus on spatial superposition instructions, mobile terminals focus on warning information and feedback channels, and large screens focus on global visualization data; data is synchronized in real time to each terminal through 5G network or enterprise private network, ensuring the timeliness of the "decision-making-execution-verification" link; AR glasses capture real-time images through built-in cameras, and the digital twin platform aligns BIM model three-dimensional components with on-site images to generate "virtual-reality" superimposed views; workers wearing AR glasses can directly see the accurate position of "virtual components" in the real scene (such as "the pipeline should be installed here, deviation ≤2mm"), while receiving voice instructions (such as "rotate 5 degrees to the left"); if installation deviation is detected, the system automatically triggers secondary superposition correction until the accuracy meets the standard; The mobile terminal integrates a voice broadcast module, which converts alarm signals into natural language in real time (such as "red alert: A area pipeline installation progress lags behind, triggering stop work order") and superimposes picture annotations. Workers can input operation feedback (such as "stop work, checking the reason") through the mobile terminal touch screen, and the feedback data is returned to the intelligent decision-making layer for verification after encryption. In addition, the mobile terminal supports historical warning query and correction scheme viewing, forming a complete closed loop of "push-feedback-record".

[0058] The large-screen monitoring center integrates multi-dimensional visual panels, including: Progress deduction dynamic panel: shows the component-level progress path deduction results in the form of a time axis, and identifies the "plan-actual-forecast" progress deviation by color gradient (such as green for normal and red for serious lag); Resource consumption heat map: visually displays the resource (manpower / materials / equipment) consumption intensity of each area through a heat map to assist in resource allocation decisions; Risk early warning map: superimposes risk identifiers (such as "pipeline collision points" and "progress lag areas") on the BIM model to support details viewing and display of rectification schemes.

[0059] Remote decision makers can interact in real time through the large screen (such as dragging to adjust the construction sequence), and the decision instructions are synchronized to the intelligent decision layer to generate new rectification schemes and push them to the site terminal for execution, forming a closed-loop management link of "remote decision-making-site execution-real-time verification".

[0060] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A BIM-based mechanical and electrical installation resource management and control system, characterized in that, Comprise: Data perception layer: by integrating laser radar, RFID, UWB and environmental sensor array, real-time acquisition of equipment installation space deviation, material flow path, construction environment parameter multi-modal data, after local preprocessing by mobile edge computing node, full element digital twin basement data flow is transmitted to model fusion layer, providing high-fidelity basic data support for BIM model point cloud fusion and physical-digital mapping; Model fusion layer: based on real-time data transmitted by data perception layer, construction progress plan and BIM component attribute are bound, and AI algorithm is used to dynamically deduce component level progress path and identify key path conflict; Synchronize ERP resource data and cost budget to generate resource consumption heat map and cost deviation curve; through rule engine + machine learning dual mode conflict warning engine, automatically detect pipeline collision, progress lag, cost overrun risk, trigger multi-level warning and dynamically adjust warning threshold, transmit the deduction result and warning signal to intelligent decision layer; Intelligent decision layer: receiving dynamic deduction result and warning signal transmitted by model fusion layer, using LSTM neural network to predict progress deviation probability in the next 7 days, combining with blockchain technology to generate tamper-proof correction scheme; Using yellow-orange-red three-level alarm system, red alarm directly triggers shutdown instruction and is pushed to user interaction layer in real time through 5G network, at the same time, receiving execution verification data feedback from user interaction layer for effect evaluation and algorithm optimization; User interaction layer: relying on decision instruction and alarm signal issued by intelligent decision layer, multi-terminal collaboration is carried out through digital twin visualization platform; AR glasses superimpose BIM model and real-time scene to guide workers to install accurately; mobile terminal integrates voice broadcast and picture annotation function, real-time push warning information and receive on-site operation feedback; large screen monitoring center integrates multi-dimensional visualization board, supports remote decision-making and execution verification, and feeds back on-site operation results and verification data to intelligent decision layer.

2. The BIM-based mechanical and electrical installation resource management and control system according to claim 1, characterized in that, The real-time acquisition of multi-modal data by integrating laser radar, RFID, UWB and environmental sensor array comprises the following steps: Obtain three-dimensional point cloud data of equipment installation site by high-precision scanning, real-time calculate spatial deviation of equipment and BIM model preset position, capture equipment tilt, displacement abnormal state; RFID tag binds materials / equipment, UWB base station tracks material flow path from warehouse to installation point through signal arrival time difference, records moving track, residence time and final installation position of each batch of materials, forms full-process traceability data chain; deploy temperature and humidity sensor to monitor installation environment, vibration sensor to detect equipment installation foundation stability, light sensor to evaluate the influence of light condition on installation accuracy, all environmental parameters are collected synchronously.

3. The BIM-based mechanical and electrical installation resource management and control system according to claim 1, characterized in that, After local preprocessing by mobile edge computing node, full element digital twin basement data flow is transmitted to model fusion layer, comprising the following steps: The mobile edge computing node receives multi-modal raw data streams from lidar, RFID / UWB, and environmental sensor arrays in real time; the node first performs data integrity checking to check whether the lidar point cloud is missing key frames, whether the RFID path data contains valid timestamps, and whether the environmental parameters are within a reasonable range, and eliminates invalid or abnormal data; Noise filtering, missing value interpolation, and abnormal value correction are performed on the raw data; multi-source data is aligned based on a unified timestamp to synchronize the "device spatial position-material flow node-environmental parameter" three; the lidar real-time point cloud is fused with the BIM model initial point cloud through a point cloud registration algorithm to generate a digital twin base; at the same time, the material flow path and environmental parameters are embedded as dynamic attributes into the corresponding BIM components to form a three-dimensional fusion of spatial position-process trajectory-environmental state full-element data stream; A quality label is attached to each data packet to mark the data source reliability and update frequency, supporting high-fidelity applications in the subsequent model fusion layer; the full-element digital twin base data stream is packaged into a structured data packet according to a unified protocol, including spatial deviation matrix, material flow coordinate sequence, and environmental parameter time sequence core fields, to standardize the data format and make it directly interpretable by the model fusion layer; A lightweight encryption algorithm is used to encrypt the data packet, which is transmitted in real time to the model fusion layer through 5G network or gigabit Ethernet; the cache mechanism of the edge computing node is used to cope with network fluctuations during transmission, providing high-fidelity and traceable basic data support for the model fusion layer, directly driving subsequent BIM point cloud fusion and physical-digital mapping analysis.

4. The BIM-based mechanical and electrical installation resource management and control system according to claim 1, characterized in that, The model fusion layer associates the construction progress plan with the attributes of specific components in the BIM model; based on the bound data, AI algorithms are used to dynamically simulate the component-level progress path; the algorithm calculates the deviation between the planned installation time of each component and the actual collected data, and identifies conflicts on the critical path; resource data and cost budget data from the ERP system are integrated synchronously to aggregate resource consumption by component, region, and time dimension, generating "resource consumption heat maps" and "cost deviation curves"; these visual charts can intuitively reflect the risk areas of excessive resource consumption or cost overrun.

5. The BIM-based mechanical and electrical installation resource management and control system according to claim 1, characterized in that, The rule engine + machine learning dual-mode conflict warning engine automatically detects pipeline collision, progress lag, and cost overrun risks, triggers multi-level warnings, and dynamically adjusts the warning threshold, including the following steps: Based on industry standards and project customization requirements, a rule library is established to compare the real-time data transmitted by the data perception layer with the rule library; if a rule violation is detected, a "pipeline collision" warning is triggered immediately; if the daily progress completion rate is less than 95%, a "progress lag" warning is triggered; A machine learning model is trained using historical project data; the model learns implicit rules such as "three consecutive days of progress lag will result in a probability of total project duration overrun greater than 30%" and "material loss rate increases by 15% during rainy season construction"; the current data is input into the trained model to predict future risk probability; Multi-level early warning trigger hierarchical response, combined with rule engine and machine learning results, is classified as yellow warning, orange warning and red warning according to risk severity, and the early warning type, risk location, probability value and suggested measures information are packaged as structured data packets and synchronized to the user interaction layer through the digital twin platform; The system dynamically adjusts the threshold value according to the historical early warning effect. If a certain type of "progress lag" early warning is frequently misreported, the system analyzes the reasons through a machine learning model and automatically adjusts the early warning threshold value in this scenario. If the "cost overrun" early warning accuracy rate is consistently greater than 90%, the threshold value is maintained or slightly increased to avoid false negatives. The threshold value is adjusted in units of "days" and is combined with the daily construction environment changes and real-time feedback data.

6. The BIM-based mechanical and electrical installation resource management and control system according to claim 1, characterized in that, The receiving model fusion layer transmits the dynamic deduction results and early warning signals, and uses an LSTM neural network to predict the progress deviation probability for the next 7 days, including the following steps: The model fusion layer transmits the component-level progress path, critical path conflict identification, resource consumption heat map, and cost deviation curve data, which are received by the intelligent decision-making layer in the form of structured data streams. The LSTM neural network predicts the progress deviation probability: ) where, is the future 7th day progress bias probability vector; Softmax is the activation function that converts linear output to probability distribution; is the hidden layer to output layer weight matrix; is the bias term; is the hidden state at time t, with dimensions matching the output layer, used to pass time series dependency information; wherein, is an output gate, controlling the cell state information output ratio: wherein σ is a Sigmoid activation function; is an output gate weight matrix; is an output gate bias vector; For cell state, the "memory" core of the LSTM, is dynamically updated by the input gate and the forget gate gate wherein is a candidate state; and are input and forget gates, controlling new information writing and old information retention: wherein, is an input gate weight matrix; is an input gate bias vector; , is a concatenation vector, is a previous time step hidden state; is a current input; is a forget gate weight matrix; is a forget gate bias vector; The LSTM is trained on historical progress deviation data to capture time dependencies and predict the progress deviation probability for each component in the next 7 days, providing a quantitative basis for risk classification.

7. The BIM-based mechanical and electrical installation resource management and control system according to claim 1, characterized in that, The combination of blockchain technology generates tamper-proof correction schemes, adopts a yellow-orange-red three-level alarm system, and simultaneously receives execution verification data from the user interaction layer for effect evaluation and algorithm optimization, including the following steps: The intelligent decision-making layer generates correction schemes based on dynamic deduction results. The scheme content includes specific adjustment measures, responsible persons, execution time window, and expected effects. Each scheme is a block in the blockchain. The block body contains scheme details, generation timestamp, and digital signature of the responsible person. The block header contains the hash value of the previous block, forming a chain structure. SHA-256 algorithm is used to calculate the block hash value. Any modification of the scheme content will cause the hash value to change, which will be detected by other nodes in the network and rejected, making the entire life cycle of the scheme from generation to execution tamper-proof. Three-level alarm system hierarchical response: Yellow warning: When the machine learning predicts a progress lag probability of 20-50% or the rule engine detects a slight rule violation, the system triggers a yellow warning and pushes a prompt message through the mobile terminal, requiring the site responsible person to check the reason and submit feedback. Orange warning: If the predicted risk probability is 50-80% or the rule conflict intensifies, the system triggers an orange warning, requiring manual intervention to develop detailed correction schemes, and displaying risk heat maps and resource allocation suggestions through the large screen monitoring center. Red warning: When a hard conflict is detected or the predicted risk probability is greater than 80%, the system directly triggers a red warning, automatically generates a stop work order and pushes it to the mobile terminal and AR glasses of the responsible person through the 5G network, forcing the site to suspend construction until the problem is resolved. After receiving the stop work order, the user interaction layer receives the stop work order, and the site personnel execute the correction operation, and real-time execution verification data is collected during the operation. These data are returned to the intelligent decision-making layer through an encrypted channel as input sources for algorithm optimization. The intelligent decision-making layer compares the prediction results with the actual execution data to evaluate the effectiveness of the correction scheme. Based on the evaluation results, the system dynamically adjusts the LSTM model parameters and the early warning threshold, forming a self-learning closed loop of prediction-execution-verification-optimization, continuously improving the early warning accuracy and decision-making scientificity.

8. The BIM-based mechanical and electrical installation resource management and control system according to claim 1, characterized in that, The user interaction layer uses multi-terminal collaborative data transmission. The decision-making instructions and alarm signals issued by the intelligent decision-making layer are standardized packaged through the digital twin visualization platform, forming a unified data format. The platform automatically adapts the data content according to the terminal type. AR glasses focus on spatial superposition instructions, mobile terminals focus on early warning information and feedback channels, and large screens focus on global visualization data. Data is synchronized in real time to each terminal through 5G network or enterprise private network. AR glasses capture real-time images through the built-in camera. The digital twin platform aligns the BIM model three-dimensional components with the live images, generating a virtual-reality superimposed view. After wearing AR glasses, workers can directly see the accurate position of "virtual components" in the real scene, while receiving voice instructions. If installation deviation is detected, the system automatically triggers secondary superposition correction until the accuracy meets the standard. The mobile terminal integrates a voice broadcast module, which converts alarm signals into natural language in real time and superimposes picture annotations. Workers input operation feedback through the mobile terminal touch screen. The feedback data is encrypted and returned to the intelligent decision-making layer for verification. The mobile terminal supports historical early warning query and correction scheme viewing. The large screen monitoring center integrates multi-dimensional visualization dashboards, including: Progress inference dynamic dashboard: shows component-level progress path inference results in timeline form, with color gradient to identify planned-actual-predicted progress deviation. Resource consumption heat map: visually displays resource consumption intensity in each area through heat map, assisting resource allocation decision-making. Risk early warning map: superimposes risk identification on the BIM model, supports click-to-view details and displays correction schemes. Remote decision-makers interact in real time through the large screen. Decision-making instructions are synchronized to the intelligent decision-making layer to generate new correction schemes and push them to the on-site terminal for execution.

Images