BIM-based prefabricated lightweight construction control optimization method and system

Abstract

The present invention proposes a BIM-based prefabricated lightweight construction control optimization method and system, the method includes: determining the data source to collect data and preprocessing, integrating the preprocessed data and integrating it into a central database; designing a CAC-GAN model and deploying and optimizing it on the central database; building a DPT model to analyze the output of the CAC-GAN model; designing a reinforcement learning algorithm, using an adaptive learning mechanism to optimize the construction strategy through reward feedback; designing an ICAS model; building a DBMF model; building an SHMPS model; building a decision support model to provide decision support based on fusion data analysis, predicted data and real-time feedback; designing a user interface to access information and interact, and at the same time establishing an adaptive learning model based on a continuous learning mechanism to self-optimize and update according to real-time data and feedback. The present invention greatly improves the efficiency and quality of construction management, and is expected to bring significant technological innovation and benefit improvement to the construction industry.

Description

Technical Field

[0001] The present invention belongs to the field of lightweight construction, and in particular relates to a BIM-based assembled lightweight construction control optimization method and system. Background Art

[0002] In the field of modern construction engineering, with the widespread application of building information modeling (BIM) technology, the digitalization, visualization and informationization of project management have been significantly improved. BIM technology can provide accurate three-dimensional digital representation for the design, construction and operation and maintenance stages of construction projects, including multi-dimensional information such as structure, material properties, time and cost. This technology has greatly optimized the building design process, reduced the time and cost required for design modifications, and improved construction efficiency and quality. However, despite the many advantages brought by BIM technology, its application in actual construction management still faces some key challenges.

[0003] First, existing BIM-based construction management systems often fail to fully realize real-time data updates and dynamic optimization of processes. In traditional applications, once a BIM model is established, its synchronization with the actual construction progress is not timely enough, resulting in deviations between the model and the actual situation on site. This static information processing mechanism limits the ability to respond quickly to emergencies during the construction process, affecting the overall progress and quality control of the project.

[0004] Secondly, although BIM provides 3D views and detailed engineering information, the existing system's automatic optimization capabilities in resource allocation, path planning, and time management at the construction site are not strong enough. For example, material handling and equipment scheduling often rely on manual experience to arrange, lacking efficient algorithm support to achieve optimization.

[0005] In addition, existing technologies are still in their infancy in integrating emerging artificial intelligence algorithms such as deep learning or reinforcement learning for construction process simulation and predictive decision-making, which limits the extraction of value from massive construction data and the realization of refined and dynamic construction management. Summary of the invention

[0006] The purpose of the present invention is to design a BIM-based prefabricated lightweight construction control optimization method and system, using the latest artificial intelligence technology, especially generative adversarial network (GAN), Transformer model, reinforcement learning and end-to-end machine learning technology, to achieve a comprehensive innovation in construction management.

[0007] In order to achieve the above object, a first aspect of the present invention provides a BIM-based prefabricated lightweight construction control optimization method, comprising the following steps:

[0008] S1. Determine the data source to collect data, add a unified timestamp to the collected data and pre-process it, and integrate the pre-processed data into a central database; wherein the data source includes BIM model data, environmental sensor data and real-time progress reports;

[0009] S2. Design the CAC-GAN model and deploy and optimize it on the central database, including:

[0010] S21, input the data into the generator G and combine it with the environmental factor e, expressed as follows:

[0011]

[0012] Among them, z is random noise, p is construction parameter, s is scene perception data, and e is environmental factor. is the predicted construction model output;

[0013] S22, calculate the construction model generated by real-time detection The deviation between the actual construction progress x and the generator G input is dynamically adjusted according to the deviation, which is calculated as follows:

[0014]

[0015] Among them, L is the loss function, which measures the error between the actual construction progress and the prediction model;

[0016] S23. Define environment variables One-hot encoding is used to process non-continuous environmental variables, and continuous variables are directly input. As an additional input layer added to the generator G, it is expressed as follows:

[0017]

[0018] in, It integrates environmental data, where T represents temperature, H represents humidity, W represents wind speed, and P represents the probability of rainfall;

[0019] Introducing time feature T d , represents the construction day or construction time period, and then integrates time series analysis in the generator G, using the recurrent neural network layer to process the input of time features, as shown below:

[0020]

[0021] Where, t represents real-time data, T d Represents time characteristics, i.e., time coding;

[0022] S24. Use the Adam optimizer and introduce a dynamic learning rate adjustment mechanism to automatically adjust the learning rate α according to the performance of the model during training, where the learning rate α is calculated according to the rate of change of the loss function as follows:

[0023]

[0024] Among them, α t+1 represents the updated learning rate, α t represents the current learning rate, ΔL t Indicates the change in loss, L t Indicates the current loss value;

[0025] S25. Use containerization technology and microservice architecture to deploy the trained CAC-GAN model to the cloud platform to achieve large-scale data processing, and use edge computing to respond to and process real-time data;

[0026] S3. Introduce a multi-head self-attention mechanism to build a DPT model based on the Transformer model, analyze the output of the CAC-GAN model, integrate BIM data for encoding, use supervised learning to train the DPT model, and update the DPT model according to real-time data; wherein the BIM data includes dynamic data and static data;

[0027] S4. Design a reinforcement learning algorithm to adjust the construction path and resource allocation in real time, and use an adaptive learning mechanism to optimize the construction strategy through reward feedback; wherein the reinforcement learning algorithm is designed as follows:

[0028] Status t To provide a comprehensive representation combining BIM model data with real-time construction monitoring data;

[0029] Actiona t is a set of decision vectors;

[0030] Reward function r(s t ,a t ) is calculated based on completion, cost efficiency and safety standards;

[0031] S5, design the ICAS model to integrate the outputs of S1-S4 and automatically execute the BIM-based prefabricated lightweight construction control decision and control;

[0032] S6. Construct a DBMF model to collect construction site information in real time and adjust the construction strategy in time through real-time data feedback;

[0033] S7. Build a machine learning-based SHMPS model to predict potential equipment failures and construction delays by analyzing historical and real-time data, and make maintenance and adjustments in advance;

[0034] S8. Build a decision support model to provide decision support based on fusion data analysis, prediction data and real-time feedback;

[0035] S9. Design a user interface to access information and interact, and establish an adaptive learning model based on a continuous learning mechanism to self-optimize and update according to real-time data and feedback.

[0036] In the above-mentioned BIM-based prefabricated lightweight construction control optimization method, the DPT model is expressed as follows:

[0037]

[0038] Among them, Q, K, and V represent query, key, and value matrices respectively, and d k represents the dimension of the key vector;

[0039] The integrated BIM data is encoded and expressed as follows:

[0040]

[0041] Among them, E represents the position code, pos represents the time position index, i represents the dimension index, and d model Represents the dimension of the model; the loss function of the DPT model is expressed as follows:

[0042]

[0043] Where N represents the number of data points. Represents the output predicted by the model, y n represents the actual data, and n represents the current data point.

[0044] In the above-mentioned BIM-based prefabricated lightweight construction control optimization method, the reward function is calculated as follows:

[0045] r(s t ,a t )=w1·CompleteRate+w2·CostSaving-w3·SafetyViolations

[0046] Among them, w1, w2, w3 represent weight parameters, CompleteRate represents the completion rate, CostSaving represents the cost efficiency, and SafetyViolations represents the safety standard.

[0047] In the above-mentioned BIM-based prefabricated lightweight construction control optimization method, the reward function is implemented based on the Actor-Critic architecture, where the Actor is responsible for generating the best action strategy, and the Critic evaluates the state value under the current strategy, and learns through real-time feedback data from the construction site, which is expressed as follows:

[0048]

[0049] Among them, α represents the learning rate, γ represents the discount factor of future rewards, and s ′ and a ′ Representing new states and new actions respectively.

[0050] In the above-mentioned BIM-based prefabricated lightweight construction control optimization method, the DBMF model uses the collected data and BIM model data to identify deviations from the preset construction plan in real time through time series analysis and machine learning models, and updates the construction status in real time. When the status evaluation shows a significant deviation, the feedback mechanism is automatically triggered; wherein the significant deviation is preset by the user, and the real-time identification of the deviation from the preset construction plan through time series analysis and machine learning models is expressed as follows:

[0051] S t+1 =f(S t ,D(t),BIM)

[0052] Among them, S t Indicates the current construction status, S t+1 represents the updated state, D(t) represents the currently collected data, and BIM represents the parameters of the BIM model.

[0053] In the above-mentioned BIM-based prefabricated lightweight construction control optimization method, the step S7 specifically includes:

[0054] S71. Deploy various sensors at the construction site and transmit the sensor data to the central monitoring system in real time via a wireless network, as shown below:

[0055] S i (t) = sensor i (data)

[0056] Among them, S i (t) represents the reading of the i-th sensor at time t;

[0057] S72. Analyze sensor data using machine learning algorithms and then apply anomaly detection algorithms to identify potential faults, as shown below:

[0058]

[0059] Among them, f anomaly represents the anomaly detection function, S represents the sensor data set;

[0060] S73. Establish a structural health assessment model, combine real-time data with BIM model data, predict the durability and remaining life of the structure, and then use deep learning methods to analyze time series data, as shown below:

[0061]

[0062] Where L(t) represents the predicted structural health level, H i (t) represents the health indicator provided by the i-th sensor at time t, w i represents the weight factor, which is adjusted according to the importance of the sensor;

[0063] S74. Establish an early warning mechanism to automatically notify the maintenance team to intervene when a critical health indicator is detected; the early warning mechanism is as follows:

[0064]

[0065] Among them, A maint (t) is the maintenance action recommendation and ν is the critical threshold of health level.

[0066] In the above-mentioned BIM-based prefabricated lightweight construction control optimization method, the decision support model uses the model output to form decision suggestions, which are expressed as follows:

[0067] P(x)=f(x1,x2,…x i ,x n )

[0068] Among them, P(x) is the output of the prediction model, x i is the input variable;

[0069] At the same time, real-time data stream processing technology is applied to dynamically provide recommended decision actions based on the current project status and historical data, as shown below:

[0070]

[0071] Where R is the recommended decision action, is a set of optional decisions, and utility represents the decision utility function based on the current state S.

[0072] In the above-mentioned BIM-based prefabricated lightweight construction control optimization method, the design of a user interface to access information and interact with dynamic data visualization output is as follows:

[0073] V(d i)=visualize(d i )

[0074] Among them, V(d i ) represents data d i The visualization output is processed by the visualize function.

[0075] In the above-mentioned BIM-based prefabricated lightweight construction control optimization method, an adaptive learning model based on a continuous learning mechanism is established to perform self-optimization and updating according to real-time data and feedback, specifically including:

[0076] Building an adaptive learning model is expressed as follows:

[0077]

[0078] Among them, θ represents the model parameters, α represents the learning rate, L represents the loss function, and x represents the newly received data;

[0079] At the same time, a feedback mechanism is designed to allow users and system monitors to directly feedback problems and improvement suggestions during the construction process, and integrate them into the learning model, as shown below:

[0080] F adjust =f(F current ,ΔF)

[0081] Among them, F adjust represents the adjusted feedback mechanism, F current represents the current mechanism, and ΔF represents the modification suggestions based on user and system feedback;

[0082] Regularly evaluate system performance and dynamically adjust the system based on the evaluation results, as shown below:

[0083]

[0084] P eval represents the performance evaluation result, K i represents the i-th performance indicator, w i Indicates the indicator weight;

[0085] Develop a collaborative mechanism to enable the system to learn and draw on experience from similar projects to optimize the system, as shown below:

[0086]

[0087] Among them, K new represents the newly integrated knowledge, K external and K internal Represent the knowledge from external projects and internal projects respectively, Represents the fusion factor.

[0088] In a second aspect of the present invention, a BIM-based assembled lightweight construction control optimization system is provided, the system comprising:

[0089] A data collection unit is used to determine the data source to collect data, add a unified timestamp to the collected data, perform preprocessing, and integrate the preprocessed data into a central database; wherein the data source includes BIM model data, environmental sensor data, and real-time progress reports;

[0090] CAC-GAN model unit, used to design CAC-GAN models and deploy and optimize them on a central database, including:

[0091] S21, input the data into the generator G and combine it with the environmental factor e, expressed as follows:

[0092]

[0093] Among them, z is random noise, p is construction parameter, s is scene perception data, and e is environmental factor. is the predicted construction model output;

[0094] S22, calculate the construction model generated by real-time detection The deviation between the actual construction progress x and the generator G input is dynamically adjusted according to the deviation, which is calculated as follows:

[0095]

[0096] Among them, L is the loss function, which measures the error between the actual construction progress and the prediction model;

[0097] S23. Define environment variables One-hot encoding is used to process non-continuous environmental variables, and continuous variables are directly input. As an additional input layer added to the generator G, it is expressed as follows:

[0098]

[0099] in, It integrates environmental data, where T represents temperature, H represents humidity, W represents wind speed, and P represents the probability of rainfall;

[0100] Introducing time feature T d , represents the construction day or construction time period, and then integrates time series analysis in the generator G, using the recurrent neural network layer to process the input of time features, as shown below:

[0101]

[0102] Where, t represents real-time data, T d Represents time characteristics, i.e., time coding;

[0103] S24. Use the Adam optimizer and introduce a dynamic learning rate adjustment mechanism to automatically adjust the learning rate α according to the performance of the model during training, where the learning rate α is calculated according to the rate of change of the loss function as follows:

[0104]

[0105] Among them, α t+1 represents the updated learning rate, α t represents the current learning rate, ΔL t Indicates the change in loss, L t Indicates the current loss value;

[0106] S25. Use containerization technology and microservice architecture to deploy the trained CAC-GAN model to the cloud platform to achieve large-scale data processing, and use edge computing to respond to and process real-time data;

[0107] The DPT model unit introduces a multi-head self-attention mechanism to build a DPT model based on the Transformer model, analyzes the output of the CAC-GAN model, integrates BIM data for encoding, uses supervised learning to train the DPT model, and updates the DPT model based on real-time data; wherein the BIM data includes dynamic data and static data;

[0108] The reinforcement learning unit designs a reinforcement learning algorithm to adjust the construction path and resource allocation in real time, and uses an adaptive learning mechanism to optimize the construction strategy through reward feedback; wherein the reinforcement learning algorithm is designed as follows:

[0109] Status t To provide a comprehensive representation combining BIM model data with real-time construction monitoring data;

[0110] Actiona t is a set of decision vectors;

[0111] Reward function r(s t ,a t ) is calculated based on completion, cost efficiency and safety standards;

[0112] ICAS model unit, used to design the ICAS model to integrate the outputs of S1-S4 and automatically execute BIM's prefabricated lightweight construction control decisions and controls;

[0113] DBMF model unit, used to build DBMF model to collect construction site information in real time and adjust construction strategy in time through real-time data feedback;

[0114] SHMPS model unit, which is used to build a machine learning-based SHMPS model to predict potential equipment failures and construction delays by analyzing historical and real-time data, and to conduct maintenance and adjustments in advance;

[0115] Decision support model unit, used to build decision support models to provide decision support based on fusion data analysis, forecast data and real-time feedback;

[0116] The interaction and update unit is used to design a user interface to access information and interact, and at the same time establish an adaptive learning model based on a continuous learning mechanism to self-optimize and update according to real-time data and feedback.

[0117] The beneficial technical effects of the present invention are at least as follows:

[0118] (1) By introducing GAN, the present invention can update the BIM model in real time during the construction process. The generator continuously predicts the construction scene based on real-time data, while the discriminator evaluates the accuracy of the prediction according to the actual construction progress. The accuracy and practicality of the model are improved through the confrontation between the two parties. This directly solves the problem of untimely updates of the existing BIM system and ensures the dynamic synchronization of information.

[0119] (2) By deploying a Transformer-based algorithm, the present invention can effectively process and analyze complex BIM data, especially in resource allocation and construction path planning. It can deeply understand the long-distance dependencies between data through the attention mechanism and optimize the decision-making process. This provides an efficient solution to the shortcomings of existing technologies in dynamic resource allocation and optimization.

[0120] (3) This paper introduces a deep reinforcement learning framework to automatically optimize construction decisions. The system adjusts the construction strategy based on real-time feedback to minimize cost and time consumption. This approach makes the construction management process more flexible and responsive, and effectively copes with the uncertainty and complexity of the construction site.

[0121] (4) The present invention significantly improves operational efficiency and decision-making accuracy by automating the entire process from data collection to decision output. The end-to-end automated processing of this system reduces manual intervention and realizes intelligent and refined management of the construction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0122] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0123] Figure 1 This is a flow chart of the BIM-based assembled lightweight construction control optimization method of the present invention.

[0124] Figure 2 This is a framework diagram of the BIM-based assembled lightweight construction control optimization system of the present invention. DETAILED DESCRIPTION

[0125] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0126] In one or more embodiments, Figure 1 As shown, a BIM-based assembled lightweight construction control optimization method is disclosed, and the method includes the following steps S1-S9:

[0127] S1. Determine the data source to collect data, add a unified timestamp to the collected data and pre-process it, integrate the pre-processed data and integrate it into the central database; wherein the data source includes BIM model data, environmental sensor data and real-time progress report.

[0128] Specifically, determine the data sources that need to be integrated, including BIM model data, environmental sensor data (location, temperature, humidity, etc.), real-time progress reports, etc. Use standardized interfaces such as REST API to obtain data from BIM software and use OPC UA protocol to collect data from sensors. These interfaces ensure the standardization and automation of data collection.

[0129] At the same time, a unified timestamp is added to all collected data, and the Network Time Protocol (NTP) is used to ensure time consistency. According to the local time, Δt is the time difference corrected by NTP. This step is key to ensuring that data from different sources are accurately aligned in time, which facilitates subsequent data analysis and decision making.

[0130] In some cases, preprocessing is to apply linear interpolation to fill missing data points and use the Z-Score method to handle outliers, with the formula Calculate, where x is the observed value, μ and σ are the mean and standard deviation of the considered data set, respectively. The purpose of this step is to improve the quality of the data and remove noise and anomalies that may affect the analysis results.

[0131] The preprocessed and cleaned data is then integrated into a central database or data lake. Use data integration tools such as Apache NiFi or Talend to build a data flow pipeline to automatically transfer data to a specified storage system. The data model should include key fields such as source identifiers, timestamps, data types, actual values, etc. to support efficient data retrieval and analysis.

[0132] S2. Design the CAC-GAN model and deploy and optimize it on the central database, including:

[0133] S21, input the data into the generator G and combine it with the environmental factor e, expressed as follows:

[0134]

[0135] Among them, z is random noise, p is construction parameter, s is scene perception data, and e is environmental factor. is the predicted construction model output;

[0136] In the embodiment of the present invention, the generator G is fed with real-time monitored construction data s (e.g., location information, material arrival time, and employee distribution) combined with environmental factors e (e.g., weather conditions and daylight intensity);

[0137] S22, calculate the construction model generated by real-time detection The deviation between the actual construction progress x and the actual construction progress x is calculated, and the generator G input is dynamically adjusted according to the deviation. Adaptive learning rate adjustment and input parameter fine-tuning are introduced to minimize the prediction error. The calculation is as follows:

[0138]

[0139] Among them, L is the loss function, which measures the error between the actual construction progress and the prediction model;

[0140] S23. Define environment variables One-hot encoding is used to process non-continuous environmental variables (such as weather conditions), and continuous variables are directly input. As an additional input layer added to the generator G, it is expressed as follows:

[0141]

[0142] in, It integrates environmental data, where T represents temperature, H represents humidity, W represents wind speed, and P represents the probability of rainfall;

[0143] Introducing time feature T d, indicating the construction day or construction time period, and then integrating time series analysis in the generator G, using a recurrent neural network layer (such as LSTM) to process the input of time features, as shown below:

[0144]

[0145] Where, t represents real-time data, T d Represents time characteristics, i.e., time coding;

[0146] S24. Use the Adam optimizer and introduce a dynamic learning rate adjustment mechanism to automatically adjust the learning rate α according to the performance of the model during training, where the learning rate α is calculated according to the rate of change of the loss function as follows:

[0147]

[0148] Among them, α t+1 represents the updated learning rate, α t represents the current learning rate, ΔL t Indicates the change in loss, L t Indicates the current loss value;

[0149] S25. Use containerization technology and microservice architecture to deploy the trained CAC-GAN model to the cloud platform for large-scale data processing, and use edge computing to respond to and process real-time data.

[0150] In the embodiment of the present invention, the CAC-GAN model can not only accurately simulate and predict the dynamic changes in the construction process, but also adapt to the specific challenges in the construction scene through environmental and time-sensitive design. The application of this method enhances the practicality and flexibility of the model and provides strong technical support for construction control and management.

[0151] S3. Introduce a multi-head self-attention mechanism to build a DPT model based on the Transformer model, analyze the output of the CAC-GAN model, integrate BIM data for encoding, use supervised learning to train the DPT model, and update the DPT model according to real-time data; wherein the BIM data includes dynamic data and static data.

[0152] Specifically, the DPT model structure adopts a multi-head self-attention mechanism, which is specially optimized for processing the spatiotemporal characteristics of BIM data. The model not only considers the time series characteristics of the data, but also integrates the spatial correlation, such as the interaction between different construction areas, as shown below:

[0153]

[0154] Among them, Q, K, and V represent query, key, and value matrices respectively. k is the dimension of the key vector, which is the standard self-attention calculation formula used in the model to handle the correlation of various input data.

[0155] In the embodiment of the present invention, static data (such as three-dimensional information of building components) and dynamic data (such as material arrival time and construction progress) from BIM are integrated. Then, the time information is encoded using variants of sine and cosine functions to maintain the continuity and periodicity of the time series data, as shown below:

[0156]

[0157] Among them, PE is the position code, pos is the time position index, i is the dimension index, and d model is the dimension of the model.

[0158] In the embodiment of the present invention, based on supervised learning, the DPT model is trained using actual construction progress data, and the optimization goal is to minimize the difference between the predicted output and the actual data. The model outputs the optimized construction schedule, resource allocation, and predicted potential risk points, where the loss function of supervised learning is designed as:

[0159]

[0160] Where N is the number of data points, is the output predicted by the model, y n It is actual data.

[0161] Based on this, the DPT model is designed to receive new construction data and feedback in real time, update and fine-tune the model online, and adjust the model weights according to the new data to quickly adapt to changes on the construction site.

[0162] S4. Design a reinforcement learning algorithm to adjust the construction path and resource allocation in real time, and use an adaptive learning mechanism to optimize the construction strategy through reward feedback; wherein the reinforcement learning algorithm is designed as follows:

[0163] Status t To provide a comprehensive representation combining BIM model data with real-time construction monitoring data;

[0164] Actiona t is a set of decision vectors;

[0165] Reward function r(s t ,a t ) is calculated based on completion, cost efficiency and safety standards.

[0166] Specifically, the statust The design is a comprehensive representation that combines BIM model data with real-time construction monitoring data. This includes not only the construction progress and resource usage, but also specifically integrates elements in the BIM model, such as the installation status of specific construction components, the completion ratio of important structures, etc. Among them, each state variable specifically corresponds to a component in the construction plan, reflecting the direct correspondence between the real-time construction scene and the early design model.

[0167] Actiona t Designed as a set of decision vectors, including resource reallocation, construction sequence adjustment, safety measures implementation, etc. These actions are directly optimized for key nodes and construction stages defined in the BIM model. Each action is based on a specific stage or component in the BIM model, ensuring that each step can directly affect the specific aspects of the construction.

[0168] Reward function r(s t ,a t ) is designed based on completion, cost efficiency and safety standards. In particular, the reward function will include a direct assessment of the completion of key construction nodes in the BIM model, as well as consideration of the impact on budget and schedule, calculated as follows:

[0169] r(s t ,a t )=w1· CompleteRate +w2· CostSaving -w3· SafetyViolations (9)

[0170] Among them, w1, w2, and w3 are weight parameters, which are adjusted according to project priorities. These parameters reflect the specific needs and goals of project management.

[0171] In the embodiment of the present invention, the implementation is based on the Actor-Critic architecture, where the Actor is responsible for generating the best action strategy, and the Critic evaluates the state value under the current strategy. By learning from the data feedback from the construction site in real time, the strategy is continuously adjusted to adapt to the dynamic changes of the project. The calculation is as follows:

[0172]

[0173] Here, α is the learning rate, γ is the discount factor for future rewards, and s ′ and a ′ They are the new state and the new action respectively.

[0174] S5. Design the ICAS model to integrate the outputs of S1-S4 and automatically execute BIM’s prefabricated lightweight construction control decisions and controls.

[0175] In the embodiment of the present invention, all data sources (BIM data, sensor data, real-time monitoring data, etc.) are integrated into a unified data processing platform. Data bus or middleware technology (such as Apache Kafka) is used to process data streams. Then time synchronization is performed. sync , ensure real-time synchronization of data between different systems, and use stream processing technologies (such as Apache Flink) for real-time data analysis.

[0176] Furthermore, a decision engine based on rules and machine learning models is developed. The engine will automatically extract insights from the processed data and generate execution instructions. The rules include preset construction rules and best practices. The output is to integrate the aforementioned reinforcement learning model to provide optimization suggestions.

[0177] In some examples, an automatic execution system is implemented to The output of the automatic adjustment of construction plans, resource allocation and safety measures. The execution formula is as follows:

[0178] A t+1 =f(E t ,S t ), where A t+1 is the next step, E t is the current environmental state, S t Is the current system status. Establish a real-time feedback mechanism to monitor the execution results and send feedback data back to the decision engine for self-learning and adjustment.

[0179] In some possible embodiments, all components (data flow design, decision engine, execution system) are integrated into a unified console. Web technology or mobile applications are used to provide a real-time monitoring and control interface for the project management team. Key technologies for the integrated system include using modern JavaScript frameworks (such as React or Angular) to build the user interface and using SpringBoot or Node.js to implement the backend logic to ensure high availability and scalability.

[0180] S6. Build a DBMF model to collect construction site information in real time and adjust the construction strategy in time through real-time data feedback.

[0181] In the embodiment of the present invention, the construction site information is to install various types of sensors at the construction site, such as displacement sensors, environmental sensors (temperature, humidity), quality control sensors, etc., and cameras for video surveillance. Then define the collection frequency and transmission protocol of sensor data to ensure that all data is uploaded to the central database system in real time, as shown below:

[0182] D(t)={d1(t),d2(t),…,dn (t)} (11)

[0183] Where D(t) is the data set collected at time t, d i (t) is the data reading of the ith sensor at time t.

[0184] The collected data is then analyzed in real time using time series analysis and machine learning models to identify potential anomalies or deviations. Based on the collected data and the BIM model, the construction status is updated in real time to identify deviations from the preset construction plan, calculated as follows:

[0185] S t+1 =f(S t ,D(t),BIM) (12)

[0186] Among them, S t is the current construction status, S t+1 is the updated state, D(t) is the currently collected data, and BIM represents the relevant parameters of the BIM model.

[0187] In the embodiment of the present invention, an automatic feedback mechanism is designed. When the status assessment shows a significant deviation, the feedback mechanism is automatically triggered. At the same time, decision suggestions based on the latest data are provided, such as adjusting resource allocation, modifying construction methods or optimizing schedules, as shown below:

[0188] A t+1 =g(S t+1 ,C) (13)

[0189] Among them, A t+1 is the recommended action, and C is the decision criteria, including cost, time efficiency, and safety standards.

[0190] In some cases, real-time monitoring systems are integrated with existing project management software and BIM tools to provide a unified user interface. An interactive dashboard is developed to display real-time data, status assessment results, and decision recommendations, allowing project managers to adjust plans or confirm recommendations directly on the interface.

[0191] S7. Build a machine learning-based SHMPS model to predict potential equipment failures and construction delays by analyzing historical and real-time data, and make maintenance and adjustments in advance, including:

[0192] S71. Various sensors, including vibration sensors, stress sensors, and corrosion sensors, are deployed at the construction site to monitor key indicators of the building structure and construction equipment, and the sensor data is transmitted to the central monitoring system in real time via a wireless network, as shown below:

[0193] S i (t)= sensori ( data ) (14)

[0194] Among them, S i (t) represents the reading of the i-th sensor at time t;

[0195] S72. Analyze sensor data using machine learning algorithms to identify patterns ranging from small changes to significant anomalies, and then apply anomaly detection algorithms to identify potential faults, such as isolation forests or cluster-based anomaly detection, to identify potential faults at an early stage, as shown below:

[0196]

[0197] Among them, f anomaly represents the anomaly detection function, S represents the sensor data set;

[0198] S73. Establish a structural health assessment model to combine real-time data with BIM model data to predict the durability and remaining life of the structure, and then use deep learning methods to analyze time series data, such as convolutional neural network (CNN) or recurrent neural network (RNN), as shown below:

[0199]

[0200] Where L(t) represents the predicted structural health level, H i (t) represents the health indicator provided by the i-th sensor at time t, w i represents the weight factor, which is adjusted according to the importance of the sensor;

[0201] S74. Establish an early warning mechanism to automatically notify the maintenance team to intervene when a critical health indicator is detected; the early warning mechanism is as follows:

[0202]

[0203] Among them, A maint (t) is the maintenance action recommendation and ν is the critical threshold of health level.

[0204] S8. Build a decision support model to provide decision support based on fusion data analysis, predictive data and real-time feedback.

[0205] In some examples, the model integrates data from BIM models, real-time monitoring systems, and predictive maintenance systems, including project progress, resource consumption, safety indicators, and environmental impacts, to form a comprehensive decision-making database, with the data uniformly represented as:

[0206] D decision ={d1,d2,…,d n} (18)

[0207] Among them, d i Represents the decision-related data of the i-th category.

[0208] Furthermore, the data is analyzed using machine learning and statistical methods, including regression analysis, decision trees, and ensemble learning methods, to identify key decision factors and potential risk points.

[0209] Use model outputs to form decision recommendations:

[0210] P(x)=f(x1,x2,…,x n ) (19)

[0211] Where P(x) is the output of the prediction model, x i are input variables such as resource utilization, cost, time delay, etc.

[0212] Furthermore, a real-time recommendation engine is developed to dynamically provide decision suggestions based on the current project status and historical data.

[0213] Apply real-time data stream processing technologies, such as Apache Storm or Apache Flink, to ensure the timeliness and accuracy of decision recommendations.

[0214] Implement the recommended mathematical representation:

[0215]

[0216] Where R is the recommended decision action, is a set of optional decisions, and utility represents the decision utility function based on the current state S.

[0217] Under some possible design ideas, design and implement a user-friendly interface to display analysis results and recommendations with charts, dashboards, and interactive tools. Integrate feedback mechanisms to allow users to evaluate the effectiveness of decision recommendations and adjust the recommendation algorithm based on user feedback. Interface design considerations:

[0218] update (U,F)→V (21)

[0219] Where U is user input (e.g. adjusting parameters, selecting options), F is feedback data, and V is view updates.

[0220] S9. Design a user interface to access information and interact, and establish an adaptive learning model based on a continuous learning mechanism to self-optimize and update according to real-time data and feedback.

[0221] Specifically, designing a user interface to access information and interact includes the following steps:

[0222] Integrate real-time data streams into the user interface to provide dynamically updated data dashboards, including construction progress, resource usage, and safety monitoring indicators. Use advanced graphics libraries (such as D3.js or Highcharts) to design visualization components that can be updated in real time based on user actions and display the most critical data indicators. Represented as follows:

[0223] V(d i )=visualize(d i ) (twenty two)

[0224] Where V(d i ) represents data d i The visualization output is processed by the function visualize, which selects the most appropriate chart or graphical representation based on the data type.

[0225] Furthermore, an interactive interface is designed to allow users to adjust construction schedules and resource allocations through simple drag-and-drop and click operations. Decision support tools such as sliders or buttons are integrated, through which users can adjust parameters in real time and immediately see the possible impact, as shown below:

[0226] S new =adjust(S current ,Δx) (23)

[0227] Among them, S new is the adjusted state, S current is the current state, and Δx is the adjustment amount entered by the user through the interface.

[0228] At the same time, a feedback mechanism is designed so that users can submit feedback or improvement suggestions directly from the interface. The system automatically optimizes the interface layout and functions based on the collected feedback. Machine learning algorithms are applied to analyze user behavior and automatically adjust interface elements to maximize the convenience and efficiency of user operations, as shown below:

[0229] U new =f(feedback,previoususability) (24)

[0230] Among them, U new is the new configuration of the user interface, feedback is the user feedback received, and previoususability is the previous interface usability score.

[0231] Furthermore, a real-time notification system is integrated to directly display key warnings and notifications on the interface, such as equipment failures, safety hazards, or shortages of key resources. Websockets or similar technologies are used to implement real-time message push to ensure that important information is delivered to the user end without delay. The calculation is as follows:

[0232] N=g( alertconditions ) (25)

[0233] Where N represents the generated notification content, and alert_conditions is the set of conditions that trigger the notification.

[0234] In the embodiment of the present invention, establishing an adaptive learning model based on a continuous learning mechanism to perform self-optimization and updating according to real-time data and feedback specifically includes the following steps:

[0235] Design a set of machine learning models that can self-adjust and optimize according to the changing construction data and feedback, and use online learning algorithms to achieve real-time updates of the models, such as using incremental learning or transfer learning methods, as shown below:

[0236]

[0237] Among them, θ represents the model parameters, α is the learning rate, L is the loss function, and x is the newly received data.

[0238] In this embodiment of the present invention, a feedback system is created to allow users and system monitors to directly provide feedback on problems and improvement suggestions during the construction process. The feedback data is integrated into the learning model to continuously adjust and optimize the system's decision support and operation processes. The calculation is as follows:

[0239] F adjust =f(F current ,ΔF) (27)

[0240] Among them, F adjust is the adjusted feedback mechanism, F current is the current mechanism, and ΔF is the modification suggestion based on user and system feedback.

[0241] Preferably, regularly evaluate system performance, including data processing speed, decision accuracy, and user satisfaction. Apply statistical methods and performance indicators to measure the effectiveness of each part of the system, and adjust the system configuration and algorithm based on the evaluation results. The calculation is as follows:

[0242]

[0243] Among them, P eval is the performance evaluation result, K i is the i-th performance indicator, wi is the indicator weight.

[0244] As a preferred embodiment, a knowledge sharing platform is established to allow different project teams to share learning results and best practices, and promote cross-project knowledge transfer and collaboration. A collaborative mechanism is developed to enable the system to learn and draw on experience from similar projects, and speed up the launch and optimization of new projects. The calculation is as follows:

[0245]

[0246] Among them, K new is the newly integrated knowledge, K external and K internal They are knowledge from external projects and internal projects, It is the fusion factor.

[0247] In one or more embodiments, Figure 2 As shown, a BIM-based assembled lightweight construction control optimization system is disclosed, and the system includes:

[0248] The data collection unit 101 is used to determine the data source to collect data, add a unified timestamp to the collected data and perform preprocessing, and integrate the preprocessed data into a central database; wherein the data source includes BIM model data, environmental sensor data and real-time progress reports;

[0249] The CAC-GAN model unit 102 is used to design a CAC-GAN model and deploy and optimize it on a central database, including:

[0250] S21, input the data into the generator G and combine it with the environmental factor e, expressed as follows:

[0251]

[0252] Among them, z is random noise, p is construction parameter, s is scene perception data, and e is environmental factor. is the predicted construction model output;

[0253] S22, calculate the construction model generated by real-time detection The deviation between the actual construction progress x and the generator G input is dynamically adjusted according to the deviation, which is calculated as follows:

[0254]

[0255] Among them, L is the loss function, which measures the error between the actual construction progress and the prediction model;

[0256] S23. Define environment variables One-hot encoding is used to process non-continuous environmental variables, and continuous variables are directly input. As an additional input layer added to the generator G, it is expressed as follows:

[0257]

[0258] in, It integrates environmental data, where T represents temperature, H represents humidity, W represents wind speed, and P represents the probability of rainfall;

[0259] Introducing time feature T d , represents the construction day or construction time period, and then integrates time series analysis in the generator G, using the recurrent neural network layer to process the input of time features, as shown below:

[0260]

[0261] Where, t represents real-time data, T d Represents time characteristics, i.e., time coding;

[0262] S24. Use the Adam optimizer and introduce a dynamic learning rate adjustment mechanism to automatically adjust the learning rate α according to the performance of the model during training, where the learning rate α is calculated according to the rate of change of the loss function as follows:

[0263]

[0264] Among them, α t+1 represents the updated learning rate, α t represents the current learning rate, ΔL t Indicates the change in loss, L t Indicates the current loss value;

[0265] S25. Use containerization technology and microservice architecture to deploy the trained CAC-GAN model to the cloud platform to achieve large-scale data processing, and use edge computing to respond to and process real-time data;

[0266] The DPT model unit 103 introduces a multi-head self-attention mechanism to construct a DPT model based on the Transformer model, analyzes the output of the CAC-GAN model, integrates the BIM data for encoding, uses supervised learning to train the DPT model, and updates the DPT model according to real-time data; wherein the BIM data includes dynamic data and static data;

[0267] The reinforcement learning unit 104 designs a reinforcement learning algorithm to adjust the construction path and resource allocation in real time, and optimizes the construction strategy through reward feedback using an adaptive learning mechanism; wherein the reinforcement learning algorithm is designed as follows:

[0268] Status t To provide a comprehensive representation combining BIM model data with real-time construction monitoring data;

[0269] Actiona t is a set of decision vectors;

[0270] Reward function r(s t ,a t ) is calculated based on completion, cost efficiency and safety standards;

[0271] ICAS model unit 105, for designing the ICAS model to integrate the outputs of S1-S4 and automatically perform the BIM-based prefabricated lightweight construction control decision and control;

[0272] The DBMF model unit 106 is used to construct a DBMF model to collect construction site information in real time and adjust the construction strategy instantly through real-time data feedback;

[0273] A SHMPS model unit 107, used to build a SHMPS model based on machine learning, predict potential equipment failures and construction delays by analyzing historical and real-time data, and perform maintenance and adjustments in advance;

[0274] A decision support model unit 108, for building a decision support model to provide decision support based on fusion data analysis, prediction data and real-time feedback;

[0275] The interaction and updating unit 109 is used to design a user interface to access information and interact, and to establish an adaptive learning model based on a continuous learning mechanism to perform self-optimization and updating according to real-time data and feedback.

[0276] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0277] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0278] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code. It should be noted that in this article, the term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0279] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. The BIM-based prefabricated lightweight construction control optimization method is characterized by: The following steps are involved: S1. Determine the data source to collect data, add a unified timestamp to the collected data and pre-process it, and integrate the pre-processed data into a central database; wherein the data source includes BIM model data, environmental sensor data and real-time progress reports; S2. Design the CAC-GAN model and deploy and optimize it on the central database, including: S21, input the data into the generator G and combine it with the environmental factor e, expressed as follows: Among them, z is random noise, p is construction parameter, s is scene perception data, and e is environmental factor. is the predicted construction model output; S22, calculate the construction model generated by real-time detection The deviation between the actual construction progress x and the generator G input is dynamically adjusted according to the deviation, which is calculated as follows: Among them, L is the loss function, which measures the error between the actual construction progress and the prediction model; S23. Define environment variables One-hot encoding is used to process non-continuous environmental variables, and continuous variables are directly input. As an additional input layer added to the generator G, it is expressed as follows: in, It integrates environmental data, where T represents temperature, H represents humidity, W represents wind speed, and P represents the probability of rainfall; Introducing time feature T d , represents the construction day or construction time period, and then integrates time series analysis in the generator G, using the recurrent neural network layer to process the input of time features, as shown below: T d =encode(t), Where, t represents real-time data, T d Represents time characteristics, i.e., time coding; S24. Use the Adam optimizer and introduce a dynamic learning rate adjustment mechanism to automatically adjust the learning rate α according to the performance of the model during training, where the learning rate α is calculated according to the rate of change of the loss function as follows: Among them, α t+1 represents the updated learning rate, α t represents the current learning rate, ΔL t Indicates the change in loss, L t Indicates the current loss value; S25. Use containerization technology and microservice architecture to deploy the trained CAC-GAN model to the cloud platform to achieve large-scale data processing, and use edge computing to respond to and process real-time data; S3. Introduce a multi-head self-attention mechanism to build a DPT model based on the Transformer model, analyze the output of the CAC-GAN model, integrate BIM data for encoding, use supervised learning to train the DPT model, and update the DPT model according to real-time data; wherein the BIM data includes dynamic data and static data; S4. Design a reinforcement learning algorithm to adjust the construction path and resource allocation in real time, and use an adaptive learning mechanism to optimize the construction strategy through reward feedback; wherein the reinforcement learning algorithm is designed as follows: Status t To provide a comprehensive representation combining BIM model data with real-time construction monitoring data; Actiona t is a set of decision vectors; Reward function r(s t ,a t ) is calculated based on completion, cost efficiency and safety standards; S5, design the ICAS model to integrate the outputs of S1-S4 and automatically execute the BIM-based prefabricated lightweight construction control decision and control; S6. Construct a DBMF model to collect construction site information in real time and adjust the construction strategy in time through real-time data feedback; S7. Build a machine learning-based SHMPS model to predict potential equipment failures and construction delays by analyzing historical and real-time data, and make maintenance and adjustments in advance; S8. Build a decision support model to provide decision support based on fusion data analysis, prediction data and real-time feedback; S9. Design a user interface to access information and interact, and establish an adaptive learning model based on a continuous learning mechanism to self-optimize and update according to real-time data and feedback.

2. The BIM-based assembled lightweight construction control optimization method according to claim 1 is characterized in that: The DPT model is expressed as follows: Among them, Q, K, and V represent query, key, and value matrices respectively, and d k represents the dimension of the key vector; The integrated BIM data is encoded and expressed as follows: Among them, E represents the position code, pos represents the time position index, i represents the dimension index, and d model Represents the dimensions of the model; The loss function of the DPT model is expressed as follows: Where N represents the number of data points. Represents the output predicted by the model, y n represents the actual data, and n represents the current data point.

3. The BIM-based assembly lightweight construction control optimization method according to claim 1 is characterized in that: The reward function is calculated as follows: r(s t ,a t )=w1·CompleteRate+w2·CostSaving-w3·SafetyViolations Among them, w1, w2, w3 represent weight parameters, CompleteRate represents the completion rate, CostSaving represents the cost efficiency, and SafetyViolations represents the safety standard.

4. The BIM-based assembly lightweight construction control optimization method according to claim 3 is characterized in that: The reward function is implemented based on the Actor-Critic architecture, where the Actor is responsible for generating the best action strategy, and the Critic evaluates the state value under the current strategy, and learns through real-time feedback data from the construction site, as shown below: Among them, α represents the learning rate, γ represents the discount factor of future rewards, s′ and a′ represent the new state and new action respectively.

5. The BIM-based assembly lightweight construction control optimization method according to claim 1 is characterized in that: The DBMF model uses the collected data and BIM model data to identify deviations from the preset construction plan in real time through time series analysis and machine learning models, and updates the construction status in real time. When the status assessment shows a significant deviation, the feedback mechanism is automatically triggered; wherein the significant deviation is preset by the user, and the real-time identification of deviations from the preset construction plan through time series analysis and machine learning models is expressed as follows: S t+1 =f(S t ,D(t),BIM) Among them, S t Indicates the current construction status, S t+1 represents the updated state, D(t) represents the currently collected data, and BIM represents the parameters of the BIM model.

6. The BIM-based assembly lightweight construction control optimization method according to claim 1 is characterized in that: The step S7 specifically includes: S71. Deploy various sensors at the construction site and transmit sensor data to the central monitoring system in real time via a wireless network, as shown below: S i (t)=sensor i (data) Among them, S i (t) represents the reading of the i-th sensor at time t; S72. Analyze sensor data using machine learning algorithms and then apply anomaly detection algorithms to identify potential faults, as shown below: Among them, f anomaly represents the anomaly detection function, S represents the sensor data set; S73. Establish a structural health assessment model, combine real-time data with BIM model data, predict the durability and remaining life of the structure, and then use deep learning methods to analyze time series data, as shown below: Where L(t) represents the predicted structural health level, H i (t) represents the health indicator provided by the i-th sensor at time t, w i represents the weight factor, which is adjusted according to the importance of the sensor; S74. Establish an early warning mechanism to automatically notify the maintenance team to intervene when a critical health indicator is detected; the early warning mechanism is as follows: Among them, A maint (t) is the maintenance action recommendation and ν is the critical threshold of health level.

7. The BIM-based assembly lightweight construction control optimization method according to claim 1 is characterized in that: The decision support model uses the model output to form decision recommendations, which are expressed as follows: P(x)=f(x1,x2,…x i ,x n ) Among them, P(x) is the output of the prediction model, x i is the input variable; At the same time, real-time data stream processing technology is applied to dynamically provide recommended decision actions based on the current project status and historical data, as shown below: Where R is the recommended decision action, is a set of optional decisions, and utility represents the decision utility function based on the current state S.

8. The BIM-based assembly lightweight construction control optimization method according to claim 1 is characterized in that: The design of a user interface to access information and interact with dynamic data visualization output is as follows: V(d i )=visualize(d i ) Among them, V(d i ) represents data d i The visualization output is processed by the visualize function.

9. The BIM-based assembly lightweight construction control optimization method according to claim 1, characterized in that: The establishment of an adaptive learning model based on a continuous learning mechanism to self-optimize and update according to real-time data and feedback includes: Building an adaptive learning model is expressed as follows: Among them, θ represents the model parameters, α represents the learning rate, L represents the loss function, and x represents the newly received data; At the same time, a feedback mechanism is designed to allow users and system monitors to directly feedback problems and improvement suggestions during the construction process, and integrate them into the learning model, as shown below: F adjust =f(F current ,ΔF) Among them, F adjust represents the adjusted feedback mechanism, F current represents the current mechanism, and ΔF represents the modification suggestions based on user and system feedback; Regularly evaluate system performance and dynamically adjust the system based on the evaluation results, as shown below: P eval represents the performance evaluation result, K i represents the i-th performance indicator, w i Indicates the indicator weight; Develop a collaborative mechanism to enable the system to learn and draw on experience from similar projects to optimize the system, as shown below: Among them, K new represents the newly integrated knowledge, K external and K internal Represent the knowledge from external projects and internal projects respectively, Represents the fusion factor.

10. The BIM-based prefabricated lightweight construction control optimization system is characterized by: The system comprises: A data collection unit is used to determine the data source to collect data, add a unified timestamp to the collected data, pre-process it, and integrate the pre-processed data into a central database; wherein the data source includes BIM model data, environmental sensor data, and real-time progress reports; CAC-GAN model unit, used to design CAC-GAN models and deploy and optimize them on a central database, including: S21, input the data into the generator G and combine it with the environmental factor e, expressed as follows: Among them, z is random noise, p is construction parameter, s is scene perception data, and e is environmental factor. is the predicted construction model output; S22, calculate the construction model generated by real-time detection The deviation between the actual construction progress x and the generator G input is dynamically adjusted according to the deviation, which is calculated as follows: Among them, L is the loss function, which measures the error between the actual construction progress and the prediction model; S23. Define environment variables One-hot encoding is used to process non-continuous environmental variables, and continuous variables are directly input. As an additional input layer added to the generator G, it is expressed as follows: in, It integrates environmental data, where T represents temperature, H represents humidity, W represents wind speed, and P represents the probability of rainfall; Introducing time feature T d , represents the construction day or construction time period, and then integrates time series analysis in the generator G, using the recurrent neural network layer to process the input of time features, as shown below: T d =encode(t), Where, t represents real-time data, T d Represents time characteristics, i.e., time coding; S24. Use the Adam optimizer and introduce a dynamic learning rate adjustment mechanism to automatically adjust the learning rate α according to the performance of the model during training, where the learning rate α is calculated according to the rate of change of the loss function as follows: Among them, α t+1 represents the updated learning rate, α t represents the current learning rate, ΔL t Indicates the change in loss, L t Indicates the current loss value; S25. Use containerization technology and microservice architecture to deploy the trained CAC-GAN model to the cloud platform to achieve large-scale data processing, and use edge computing to respond to and process real-time data; The DPT model unit introduces a multi-head self-attention mechanism to build a DPT model based on the Transformer model, analyzes the output of the CAC-GAN model, integrates BIM data for encoding, uses supervised learning to train the DPT model, and updates the DPT model based on real-time data; wherein the BIM data includes dynamic data and static data; The reinforcement learning unit designs a reinforcement learning algorithm to adjust the construction path and resource allocation in real time, and uses an adaptive learning mechanism to optimize the construction strategy through reward feedback; wherein the reinforcement learning algorithm is designed as follows: Status t To provide a comprehensive representation combining BIM model data with real-time construction monitoring data; Actiona t is a set of decision vectors; Reward function r(s t ,a t ) is calculated based on completion, cost efficiency and safety standards; ICAS model unit, used to design the ICAS model to integrate the outputs of S1-S4 and automatically execute BIM's prefabricated lightweight construction control decisions and controls; DBMF model unit, used to build DBMF model to collect construction site information in real time and adjust construction strategy in time through real-time data feedback; SHMPS model unit, which is used to build a machine learning-based SHMPS model to predict potential equipment failures and construction delays by analyzing historical and real-time data, and to conduct maintenance and adjustments in advance; Decision support model unit, used to build a decision support model to provide decision support based on fusion data analysis, forecast data and real-time feedback; The interaction and update unit is used to design a user interface to access information and interact, and at the same time establish an adaptive learning model based on a continuous learning mechanism to self-optimize and update according to real-time data and feedback.

Images