A BIM-based building electromechanical full life cycle management platform and its optimization strategy

Abstract

The present invention discloses a BIM-based building mechanical and electrical full life cycle management platform and its optimization strategy, which relates to the field of building information management technology, and solves the shortcomings of existing BIM technology in processing efficiency, model accuracy, data synchronization and change management; the present invention adopts the data layer to clean and standardize data through machine learning models, and realizes distributed storage through a cloud management system; the model layer constructs and updates high-precision BIM models; the synchronization and collaboration layer uses real-time synchronization mechanisms and blockchain technology to ensure data consistency and traceability; the decision analysis layer performs data analysis and prediction through deep learning algorithms; the full life cycle management layer integrates historical data and current data, and automatically triggers management processes; the present invention greatly improves data processing efficiency and model accuracy, and significantly enhances the seamless sharing capability of information, the transparency of data synchronization and change management, thereby enhancing the overall collaboration efficiency and management level of the project.

Description

Technical Field

[0001] The present invention relates to the technical field of building information management, and more specifically to a BIM-based building electromechanical full life cycle management platform and its optimization strategy. Background Art

[0002] With the digital transformation of the construction industry, the building mechanical and electrical full life cycle management platform based on BIM (Building Information Modeling) has gradually become an important means to improve project management efficiency and quality. This platform integrates information from multiple stages such as design, construction, operation and maintenance, realizes seamless docking and sharing of data, and provides strong support for the full life cycle management of building mechanical and electrical systems. Driven by the "new infrastructure" policy, the application of BIM technology in the construction industry continues to deepen, and its advantages in improving project quality and reducing operating costs are becoming increasingly prominent, providing broad space for the development of a BIM-based building mechanical and electrical full life cycle management platform.

[0003] The BIM-based building mechanical and electrical (MEP) lifecycle management platform integrates multiple advanced technologies, including 3D modeling, data management, the Internet of Things (IoT), and artificial intelligence (AI). This platform uses BIM models to centrally manage data from the design, construction, and operation and maintenance phases of building MEP projects, enabling information sharing and collaborative work. Specifically, BIM models encompass not only the geometric information of MEP equipment but also multi-dimensional data such as its attributes, installation locations, and maintenance records. Through the cloud platform, all stakeholders can access and update this data in real time, improving project transparency and management efficiency.

[0004] However, in practical applications, BIM technology involves a massive amount of data, including design drawings, construction records, equipment parameters, and maintenance history. This makes seamless information sharing and transfer challenging. Data storage, processing, and analysis are inefficient, leading to processing delays and hindering project collaboration. Furthermore, mechanical and electrical systems typically include numerous piping, cables, and equipment, requiring high-precision data for accurate modeling. However, existing BIM software often suffers from precision loss when handling these complex elements. Data standards vary between vendors and software, leading to decreased accuracy and format incompatibilities during model import and export. Furthermore, when multiple parties are using a BIM model simultaneously, data synchronization becomes a significant challenge. Existing platforms often struggle to accurately reflect data changes from all stakeholders in real time. Design changes and equipment replacements are common occurrences in projects, but existing BIM platforms often lack effective change management mechanisms, hindering timely updates of model data. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention discloses a BIM-based building electromechanical full life cycle management platform and its optimization strategy, aiming to solve the problems in the background technology.

[0006] In order to achieve the above technical effects, the present invention adopts the following technical solutions:

[0007] A BIM-based building electromechanical full life cycle management platform, including: data layer, model layer, synchronization and collaboration layer, decision analysis layer and full life cycle management layer;

[0008] The data layer is used to collect at least design drawings, construction records, and equipment parameter data, convert non-standard format data into a standard format, and perform data cleaning, denoising, and normalization through machine learning models. The data layer is also used to distribute and store pre-processed data through a cloud management system, establish a data index and metadata management framework, and output standardized, searchable data sets to the model layer and decision analysis layer.

[0009] The model layer is used to construct and update the BIM model including building structure, electromechanical system and equipment elements based on the output data of the data layer;

[0010] The synchronization and collaboration layer is used to receive the real-time model data output by the model layer, and perform real-time synchronization and conflict detection of model parameter data through the data synchronization mechanism. When the data is changed, the blockchain is used to record the data change history. The synchronization mechanism is triggered according to the data change, and the changed data is pushed to all relevant participants in real time. The real-time updated data is output to the decision analysis layer and the full life cycle management layer. At the same time, the change information is fed back to the model layer for updating.

[0011] The decision analysis layer is used to receive the output data of the model layer and the synchronization and coordination layer, perform data analysis, prediction and anomaly detection on the electromechanical system through a comprehensive electromechanical system analysis algorithm based on deep learning, and output analysis reports, early warning information and optimization suggestions to the full life cycle management layer and relevant participants;

[0012] The full life cycle management layer is used to receive the output data of the decision analysis layer, combine it with the historical data in the data layer, and use the full life cycle fusion algorithm to integrate and manage the full life cycle information; automatically trigger the management process according to the project stage, and output the full life cycle management report, project progress information and BIM model optimization suggestions to relevant participants and decision makers.

[0013] As a further technical solution of the present invention, the working principle of the data layer for converting non-standard format data into standard format is as follows: matching specific character sequences in the data through a regular expression-based pattern matching algorithm to identify the format characteristics of the data, and analyzing the words, phrases and sentences in the data through a semantic analysis-based parsing algorithm to further identify the data format type; if the input data format matches any format in a preset standard format library, the identified non-standard data fields are matched with the corresponding fields in the standard format through a format mapping mechanism, and the non-standard data fields are converted one by one into standard format fields through key-value pair mapping and regular expression replacement; if the input data format cannot be directly matched with any item in the standard format library, feature extraction and classification of the input data are performed through clustering analysis to construct a new data format model; after the data format model is successfully constructed, the new data format model is added to the standard format library, and the mapping rules are automatically updated.

[0014] As a further technical solution of the present invention, during the data conversion process, the machine learning model first traverses each data record through an autoencoder, automatically learns and extracts effective features in the data, and suppresses noise and redundant information; during the data traversal process, if it is detected that the characteristic value exceeds the preset threshold range, the anomaly detection mechanism is triggered, and the redundant, missing or erroneous information in the abnormal characteristic value is corrected or filled by the statistical outlier detection algorithm; the statistical outlier detection algorithm fills the abnormal characteristic value by the mean, median or interpolation method; then, the machine learning model denoises the data through a generative adversarial network; if the denoised data is normally distributed, the machine learning model uses the Z-score normalization method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1; if the distribution range of the denoised data is within a preset fixed range, the minimum and maximum scaling method is used to linearly map the data to the interval [0,1] or [-1,1].

[0015] As a further technical solution of the present invention, the cloud management system includes a distributed storage module, a data index and metadata management module, a data security module and a data backup module; the distributed storage module includes a data sharding unit and a data synchronization unit; the data sharding unit is used to shard and distribute data to different storage nodes through a consistent hashing algorithm; the data synchronization unit is used to ensure that the data copies on all storage nodes remain consistent through the Raft consensus algorithm when a data update operation is detected; the data index and metadata management module includes an index construction unit and a metadata management unit; the index construction unit is used to divide the index into multiple small blocks using a block indexing strategy to improve query performance, and calculate the relevance of documents through a vector space model and TF-IDF algorithm; the metadata management unit is used to dynamically update the metadata model in the graph database when a new data type or data relationship is detected; the data security module is used to reject unauthorized user access requests and record logs through a role-based access control mechanism, and is also used to encrypt sensitive data through an encryption mechanism; the data backup module is used to regularly perform data backup operations through a scheduled task scheduler. During the backup process, the data backup module reduces backup time and storage space through an incremental backup method.

[0016] As a further technical solution of the present invention, the model layer obtains three-dimensional point cloud data through laser scanning, compares and verifies it with the design drawings, and if a difference is detected between the design drawings and the actual scanned data, the model is fine-tuned through the ICP iterative closest point algorithm; when constructing the BIM model, the model layer also assigns parametric attributes to each element in the model through parametric modeling, and if any device parameter changes, the relevant geometric shapes and attributes in the model are automatically updated through the parameter update mechanism; during the parameter update process, the parameter update mechanism automatically generates a design plan according to predefined rules through an automated rule engine, and the predefined rules include equipment layout rules and pipeline layout rules; during the parameter update process, the model layer also applies the changed data to the existing BIM model through the model update mechanism; the model update mechanism records the history of each data change through version control Git; during the use of the model, the model layer allows multiple users to edit the same BIM model at the same time through a real-time communication protocol; if multiple users edit the same part of the model at the same time, the conflict is resolved through the synchronous collaboration layer.

[0017] As a further technical solution of the present invention, the working principle of the synchronization collaboration layer is as follows: receiving the real-time model data of the model layer and synchronizing it through a synchronization protocol based on a distributed database; when a data change is detected in the model layer, the change data is processed and distributed through a distributed message queue, and the change information is encapsulated into a block through a blockchain write mechanism, and added to the blockchain after verification using a consensus algorithm. Each block contains the hash value of the previous block to form an unalterable data chain; during the change process, data conflicts are identified through a hash algorithm and data fingerprint, and data conflicts are resolved through a three-way merge algorithm; if the conflict is not resolved, the relevant participants are notified through a real-time communication protocol to manually resolve the conflict and resubmit the change; if it is detected that the data change has been confirmed to be correct, the change data is pushed to all relevant participants in real time through the real-time communication protocol; if it is detected that the data change has been confirmed to be correct and the conflict has been resolved, the real-time updated data is output to the decision analysis layer and the full life cycle management layer through the API interface, and the change information is fed back to the model layer through two-way data binding.

[0018] As a further technical solution of the present invention, the prediction method of the electromechanical system comprehensive analysis algorithm based on deep learning is: based on the input feature vector X t , including historical energy consumption data (E t-n ,...,E t-1 ), weather data (M t-n ,...M t ,) building occupancy data ( t-n ,...O t ,) and device status data (R t-n ,...R t ,), where n represents the lag order of the time series; the energy consumption at future time points is predicted by a deep hybrid neural network, which uses a multi-layer long short-term memory network to extract the input feature vector X t The time series features T in r , and calculate the time series features T through the multi-head self-attention mechanism r Correlation between different time steps, output of energy consumption forecast value at future time points The prediction formula is:

[0019]

[0020] In formula (1), W out represents the weight of the output layer; Q, K, V are query, key and value matrices respectively; d k represents the dimension of the key vector; α represents the activation function, which is used to introduce nonlinearity and enhance the expressiveness of the model; L represents the number of heads; LSTM represents the output of the LSTM layer; attn represents the output of the self-attention mechanism; WLSTM and b LSTM Represent the parameters and bias terms of the LSTM layer respectively; the anomaly detection method of the electromechanical system comprehensive analysis algorithm based on deep learning is: constructing a fault propagation model through a graph neural network, and the fault propagation model is based on the adjacency matrix A of the fault propagation graph and the device feature matrix X dev As input, the fault propagation pattern between devices is learned through multi-layer graph convolution operations, and the fault propagation probability P of each device is calculated. fault , the calculation formula is:

[0021]

[0022] In formula (2), is the weight matrix of the k-th neighbor in the l-th layer graph convolution, H (l) Represents the node feature matrix after the l-th layer of graph convolution; The attenuation coefficient that controls the spatial influence range is used to determine the degree of influence of spatial distance on weight; represents the graph convolution output feature vector of device d at layer l; X dev represents the device feature matrix; represents the parameters of the GNN model in the lth layer of graph convolution; during the prediction and anomaly detection process, the deep learning-based electromechanical system comprehensive analysis algorithm compares the calculation results with the preset threshold; if the deviation between the predicted energy consumption and the actual energy consumption exceeds the preset threshold, or the fault propagation probability of a certain equipment exceeds the preset threshold, the early warning mechanism is triggered, and the analysis report and early warning information are output to the full life cycle management and relevant participants.

[0023] As a further technical solution of the present invention, the full life cycle management layer includes an information integration module, a management process trigger module, an intelligent report generation module and an optimization suggestion and feedback module; the information integration module is used to adopt a federated learning method to fuse the output data of the decision analysis layer with the historical data in the data layer, and generate a comprehensive data representation through a graph neural network. The information integration module is also used to build a knowledge base of full life cycle information through a knowledge graph, storing data entities and data relationships; the management process trigger module is used to realize process automation through a business process management system and an event-driven architecture based on the data generated by the information integration module; the intelligent report generation module is used to generate a full life cycle management report through natural language processing and a template engine; the optimization suggestion and feedback module is used to generate optimization suggestions through reinforcement learning and operations research methods.

[0024] As a further technical solution of the present invention, the working method of the full life cycle fusion algorithm is:

[0025] S1. Calculate the weights of the time series data and spatial data of each data sample through the Softmax function;

[0026] S2, applying the calculated weights to the time series data and spatial data, and generating the fused multimodal feature vector through element-by-element multiplication;

[0027] S3. Calculate the influence factor of the fusion feature through a nonlinear activation function to determine the importance and contribution of different information sources;

[0028] S4. Combine the reduced-dimensional data with the historical data and generate an integrated full-life cycle information matrix using a weighted summation method, which includes all relevant historical and current data;

[0029] S5. Calculate the event triggering condition through linear transformation to determine whether the management process needs to be triggered;

[0030] S6. Compare the calculated event trigger condition with the preset threshold. If the trigger condition exceeds the threshold, trigger the corresponding management process; otherwise, do not trigger.

[0031] S7. Estimate the expected reward of taking an action in the current state by recursively calculating the state-action value function, select the action with the highest expected reward as the optimal action, and generate optimization suggestions.

[0032] As a further technical solution of the present invention, a BIM-based building electromechanical full life cycle management optimization strategy includes the following steps:

[0033] Step 1: Build a BIM model of the electromechanical system based on the building design drawings and equipment parameters. The BIM model of the electromechanical system integrates at least equipment operation and maintenance data and energy consumption data through an API interface.

[0034] Step 2: Based on the data integrated by the BIM model, the operating status of the electromechanical system is analyzed, predicted, and anomalies are detected through a comprehensive electromechanical system analysis algorithm based on deep learning;

[0035] Step 3: Based on the output results of step 2, combined with the project management knowledge base, the full life cycle information is integrated and managed through the full life cycle fusion algorithm to generate optimization suggestions or decision support reports, including equipment maintenance plans and energy consumption optimization plans;

[0036] Step 4: In the full lifecycle management of electromechanical systems, blockchain is used to record decision-making and change information during the design, construction, and operation and maintenance phases. Smart contracts are used to automatically execute collaborative workflows, including approval, notification, and feedback, to improve collaborative efficiency and information transparency.

[0037] Step 5: Use blockchain to record the collaborative work process and decision-making results of all parties involved, and use a real-time data stream processing engine to achieve real-time data synchronization and collaborative work during the design, construction, and operation and maintenance stages;

[0038] Step 6. Based on the actual operation of the electromechanical system and the optimization decision support report, dynamically adjust equipment parameters and maintenance plans to optimize system performance and reduce operation and maintenance costs. Continuously analyze data through the particle swarm optimization algorithm to continuously iterate and optimize the BIM model and management strategy.

[0039] Based on the above technical solutions, the positive and beneficial effects of the present invention are:

[0040] 1. The data layer of this invention utilizes machine learning models for data cleaning, denoising, and normalization, converting non-standard data into a standard format, significantly improving data availability and consistency. Furthermore, the cloud management system's distributed storage and metadata management framework enable the efficient sharing and transfer of standardized, searchable datasets across different participants. This solution not only addresses the challenges of sharing and transferring data due to massive amounts of data, but also improves the efficiency and quality of data processing, providing strong support for collaborative project management.

[0041] 2. The data layer of this invention achieves distributed data storage through a cloud management system, leveraging the powerful computing power of cloud computing to significantly increase the speed of data storage and processing. Simultaneously, the application of machine learning models makes data cleaning, denoising, and normalization more efficient, providing a reliable foundation for subsequent data analysis. Furthermore, the decision analysis layer, through real-time data analysis and prediction, can promptly identify potential problems and risks and provide optimization suggestions, helping project teams make more informed decisions. This solution not only improves data storage and processing efficiency but also makes data analysis more accurate and timely, providing strong data support for project decision-making.

[0042] 3. The model layer of the present invention is based on the output data of the data layer and can build and update high-precision BIM models. The accuracy and integrity of the model data are ensured through real-time update and synchronization mechanisms. At the same time, the model layer also has the ability to perform fine modeling of complex elements, avoiding the loss of accuracy or format incompatibility that may occur during the import and export process. This solution not only improves the accuracy and usability of the BIM model, but also provides a reliable foundation for the subsequent management and maintenance of the project. The full life cycle management layer integrates historical data and current data through a full life cycle fusion algorithm to generate a comprehensive full life cycle management report to support effective management and decision-making at all stages of the project. This real-time change management and full life cycle perspective not only improves the flexibility and adaptability of the project, but also ensures the continuous optimization and improvement of the project throughout its life cycle.

[0043] 4. The synchronization and collaboration layer of this invention achieves real-time data synchronization and conflict detection among multiple participants through a data synchronization mechanism and blockchain-based data change history recording. When data changes, the synchronization mechanism triggers an update and pushes the changed data to all relevant participants in real time. Furthermore, the application of blockchain technology ensures the traceability and immutability of data change history. This solution not only addresses the challenges of multi-party data synchronization but also improves the real-time and consistency of model data, providing a strong guarantee for collaborative project management. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which:

[0045] Figure 1 This is an architecture diagram of a BIM-based building electromechanical full life cycle management platform of the present invention;

[0046] Figure 2 A diagram showing the working principle of the data layer of the present invention for converting non-standard format data into standard format;

[0047] Figure 3 A diagram showing the working principle of the machine learning model of the present invention;

[0048] Figure 4 This is a working method diagram of the full life cycle fusion algorithm of the present invention;

[0049] Figure 5 This is a flowchart of the steps of the BIM-based building mechanical and electrical full life cycle management optimization strategy of the present invention. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0051] like Figure 1 As shown in the figure, a BIM-based building electromechanical full life cycle management platform includes: data layer, model layer, synchronization and collaboration layer, decision analysis layer and full life cycle management layer;

[0052] The data layer is used to collect at least design drawings, construction records, and equipment parameter data, convert non-standard format data into standard format, and perform data cleaning, denoising, and normalization through machine learning models; the data layer is also used to distribute the pre-processed data through the cloud management system, establish a data index and metadata management framework, and output standardized and searchable data sets to the model layer and decision analysis layer; wherein, if Figure 2As shown: the working principle of the data layer converting non-standard format data into standard format is as follows: a pattern matching algorithm based on regular expressions is used to match specific character sequences in the data to identify the format features of the data, and a parsing algorithm based on semantic analysis is used to analyze the words, phrases and sentences in the data to further identify the data format type; if the input data format matches any format in the preset standard format library, the format mapping mechanism is used to match the identified non-standard data fields with the corresponding fields in the standard format, and the non-standard data fields are converted one by one into standard format fields through key-value pair mapping and regular expression replacement; if the input data format cannot be directly matched with any item in the standard format library, the input data is subjected to feature extraction and classification through clustering analysis to construct a new data format model; after the data format model is successfully constructed, the new data format model is added to the standard format library and the mapping rules are automatically updated; and then the non-standard data fields are converted one by one into standard format fields through the format mapping mechanism, key-value pair mapping and regular expressions; in a specific embodiment, regular expressions are a powerful text processing tool that can define and match complex character patterns. Regular expressions can be used to identify specific character sequences such as dates, timestamps, numbers, etc. Regular expressions provide a rich syntax that supports a variety of complex matching rules. Semantic analysis is a branch of natural language processing (NLP) that aims to understand the meaning of text. By analyzing vocabulary, phrases, and sentences, semantic analysis can extract deeper information, such as the contextual relationships and semantic structure of the data. In the present invention, semantic analysis is used to further identify the format type of data. For example, by analyzing keywords and phrases in the data, the system can determine the data type (such as equipment parameters, construction records, etc.). Semantic analysis can also handle ambiguous and non-standard data descriptions, improving the accuracy of format recognition. Combined with preliminary matching of regular expressions, semantic analysis ensures comprehensive and accurate identification of data formats. The format mapping mechanism is a technology that matches non-standard format data fields with standard format fields. Using a preset standard format library, the system can find the correspondence between non-standard data fields and standard format fields and convert them. Key-value pair mapping is a method of mapping non-standard data fields to standard format fields, establishing a one-to-one correspondence in the form of key-value pairs. Regular expression replacement is used to find and replace specific patterns in text. Cluster analysis is an unsupervised learning method that classifies data into different categories by calculating similarities between data points. Cluster analysis can automatically discover and categorize patterns in data. New data format models are constructed based on cluster analysis results, and the system generates new data format descriptions based on the clustering results. Updating the standard format library involves adding the new data format model to the existing standard format library and automatically updating the mapping rules.

[0053] In practical applications, the data layer's operational approach demonstrates the efficiency and flexibility of its structural features. First, through regular expression pattern matching algorithms, the data layer can quickly identify and extract specific format features within the data, such as dates, numbers, and text. Next, a semantic analysis and parsing algorithm conducts in-depth analysis of the extracted data, further determining the data's format type based on vocabulary, phrases, and sentence structure. Once the data's format type is determined, a format mapping mechanism is activated, matching non-standard data fields with corresponding fields in the standard format. If a match is successful, the non-standard data fields are converted to standard format fields through key-value mapping and regular expression substitution. If a match fails, cluster analysis is initiated to extract and classify features from the input data, construct a new data format model, and automatically update the mapping rules. This operational approach not only improves data processing efficiency but also enhances the adaptability and flexibility of the data format.

[0054] Compared with the existing technology, the data layer of the present invention has the following advantages and characteristics in converting non-standard format data into standard format: First, by comprehensively applying advanced data processing technologies such as regular expression pattern matching, semantic analysis and parsing, and format mapping mechanisms, it achieves accurate identification and efficient conversion of data formats; second, through the clustering analysis algorithm and the function of automatically updating mapping rules, it achieves dynamic learning and adaptation of data formats, improving the flexibility and adaptability of data processing; finally, the working mode and logical operation process of the data layer embody high efficiency and accuracy, and can quickly convert non-standard format data into standard format data, providing reliable data support for subsequent processing layers. These advantages and characteristics make the data layer of the present invention have important application value in the BIM-based building electromechanical full life cycle management platform.

[0055] Further, such as Figure 3As shown: During the data conversion process, the machine learning model first traverses each data record through the autoencoder, automatically learns and extracts effective features from the data, and suppresses noise and redundant information. During the data traversal process, if the feature value is detected to exceed the preset threshold range, the anomaly detection mechanism is triggered, and the redundant, missing or erroneous information in the abnormal feature value is corrected or filled by the statistical outlier detection algorithm. The statistical outlier detection algorithm fills the abnormal feature value by the mean, median or interpolation method. Then, the machine learning model denoises the data through the generative adversarial network. If the denoised data is normally distributed, the machine learning model uses the Z-score normalization method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. If the distribution range of the denoised data is within a preset fixed range, the minimum and maximum scaling method is used to linearly map the data to the interval [0,1] or [-1,1]. In a specific implementation, the autoencoder is a neural network structure that compresses the input data into a low-dimensional space (encoding process) and then reconstructs the original data from the low-dimensional space (decoding process) to learn the effective feature representation of the data. In the present invention, an autoencoder is used to traverse each data record, automatically learn and extract effective features from data such as design drawings, construction records, and equipment parameters, while suppressing noise and redundant information. This process effectively reduces the data dimension and improves the efficiency of data processing, while retaining the key information of the data, providing a more accurate data foundation for subsequent data cleaning and anomaly detection. Outliers refer to values ​​in a data set that deviate from the majority of data points. They may be caused by erroneous records, equipment failures, or abnormal events. In the present invention, statistical outlier detection algorithms (such as mean, median, or interpolation methods) are used to identify and correct these abnormal feature values. When the feature value extracted by the autoencoder exceeds the preset threshold range, the anomaly detection mechanism is triggered, and the algorithm will select an appropriate statistical method to correct or fill in the gaps based on the data distribution characteristics. This step ensures the accuracy and consistency of the data and avoids the interference of abnormal data on subsequent analysis results. Generative adversarial networks (GANs) consist of two networks, a generator and a discriminator, which generate realistic data samples through competitive learning. In this paper, GANs are used for data denoising. Specifically, the generator attempts to generate data samples that are similar to the original data but with less noise, while the discriminator attempts to distinguish between real data and generated data. Through continuous iterative training, the generator is able to learn the true distribution of the data and generate high-quality denoised data. This process significantly improves the signal-to-noise ratio of the data and provides a clearer data foundation for subsequent data standardization. Z-score standardization converts data to a standard normal distribution with a mean of 0 and a standard deviation of 1. It is suitable for situations where the data distribution is unknown or has extreme values.The minimum and maximum scaling linearly maps the data to a specified interval (such as [0,1] or [-1,1]), which is suitable for situations where the data distribution range is known and the relative size relationship of the data needs to be maintained. In the present invention, these two normalization methods are used to process the denoised data to ensure the consistency and comparability of the data in subsequent analysis. Z-score normalization helps to eliminate the dimensional differences between the data, so that different features have the same weight in the model; while the minimum and maximum scaling ensures that the data is distributed within a specific range, which facilitates the training and prediction of the model.

[0056] In practice, the data layer implements distributed data storage and management through a cloud management system, establishing a data index and metadata management framework to facilitate rapid data retrieval and access. Furthermore, the data layer provides a rich library of data preprocessing tools and algorithms, enabling users to design customized data processing workflows based on their specific needs. This highly flexible and scalable approach enables machine learning models to efficiently process large-scale, multi-source, and heterogeneous building mechanical and electrical data, providing strong data support for subsequent model building and decision analysis.

[0057] Compared with the prior art, the advantages of the present invention in introducing machine learning models for data preprocessing in the BIM-based building electromechanical full life cycle management platform include: through the application of advanced technologies such as autoencoders and GANs, the present invention realizes efficient feature extraction, noise suppression and denoising of data, significantly improving the efficiency and quality of data processing. The statistical outlier detection algorithm can accurately identify and correct outliers in the data, ensure the integrity and accuracy of the data, and avoid the interference of abnormal data on subsequent analysis results. According to the distribution of the data, the appropriate standardization method (Z-score standardization or minimum and maximum scaling) is selected to ensure the consistency and comparability of the data in subsequent analysis, and enhance the generalization ability and prediction accuracy of the model. The data layer realizes distributed storage and management of data through the cloud management system, provides a rich data preprocessing tool and algorithm library, supports users to customize the data processing process design according to actual needs, and realizes the automation and intelligence of data processing. In addition, the design of the present invention fully considers the scalability and customizability of the system, supports users to expand and customize functions according to actual needs, and meets the data processing needs in different scenarios.

[0058] In a specific embodiment, the hardware working environment for data cleaning, denoising, and normalization processing by the machine learning model mainly includes: high-performance servers for running machine learning algorithms and storing large data sets; graphics processing units (GPUs); storage devices; network interfaces and power systems. A group of high-performance servers and GPUs were selected as the hardware platform, and the machine learning model (Group A) and traditional processing methods (Group B) were respectively applied to perform data cleaning, denoising, and normalization processing. The traditional processing method uses rule-based data cleaning and manual denoising, as well as manually set normalization rules. The experimental data table is shown in Table 1:

[0059] Table 1 Comparative experimental record

[0060]

[0061] Comparative experiments show that Group A significantly shortened data cleaning time, achieved superior denoising results, and achieved higher normalization accuracy. This demonstrates that machine learning models can more quickly and accurately process large amounts of data, improving data quality and providing more reliable data support for BIM-based building mechanical and electrical lifecycle management platforms. Therefore, in practical applications, using machine learning models for data preprocessing will help improve the platform's overall performance and efficiency.

[0062] Furthermore, the cloud management system includes a distributed storage module, a data index and metadata management module, a data security module and a data backup module; the distributed storage module includes a data sharding unit and a data synchronization unit; the data sharding unit is used to shard and distribute data to different storage nodes through a consistent hashing algorithm; the data synchronization unit is used to ensure that the data copies on all storage nodes remain consistent through the Raft consensus algorithm when a data update operation is detected; the data index and metadata management module includes an index construction unit and a metadata management unit; the index construction unit is used to divide the index into multiple small blocks using a block indexing strategy to improve query performance, and calculate the relevance of documents through a vector space model and TF-IDF algorithm; the metadata management unit is used to dynamically update the metadata model in the graph database when a new data type or data relationship is detected; the data security module is used to reject unauthorized user access requests and record logs through a role-based access control mechanism, and is also used to encrypt sensitive data through an encryption mechanism; the data backup module is used to regularly perform data backup operations through a scheduled task scheduler. During the backup process, the data backup module reduces backup time and storage space through an incremental backup method.

[0063] The consistent hashing algorithm is a hashing algorithm used in distributed systems. It achieves uniform data sharding by mapping data to a fixed ring space and evenly distributing nodes on this ring. When the number of nodes changes, the consistent hashing algorithm can minimize data migration, ensuring system stability and efficiency. In the present invention, the data sharding unit uses the consistent hashing algorithm to shard and distribute data across different storage nodes. This sharding method not only improves data storage efficiency but also reduces the load on individual nodes. The consistent hashing algorithm, combined with the data synchronization unit, ensures uniform data distribution across different nodes. For example, when processing a large number of design drawings, the consistent hashing algorithm can evenly distribute these drawings across multiple storage nodes, avoiding the problem of overloading a single node. The Raft consensus algorithm is a consensus protocol used in distributed systems. It ensures that data copies on all nodes remain consistent through leader election and log replication. The Raft algorithm simplifies the complexity of distributed systems through a simple state machine model and clear role definitions. In the present invention, the data synchronization unit uses the Raft consensus algorithm to ensure that data copies on all storage nodes remain consistent. This synchronization mechanism ensures data consistency and reliability. When a data update is detected, the Raft algorithm automatically synchronizes the update to all nodes, ensuring that all nodes maintain consistent data copies. The block indexing strategy divides the index into multiple small blocks to improve query performance. By dividing the index into multiple blocks, the index size can be reduced, speeding up queries. Combining the block indexing strategy with the Vector Space Model and TF-IDF algorithms further improves query relevance and accuracy. For example, when querying for specific equipment parameters, the block indexing strategy can quickly locate relevant data blocks, while the Vector Space Model and TF-IDF algorithms calculate document relevance, ensuring accurate and relevant query results. The Vector Space Model represents documents as multidimensional vectors, with each dimension representing a term. The TF-IDF (Term Frequency-Inverse Document Frequency) algorithm calculates the importance of a term in a document. A higher TF-IDF value indicates a greater importance of the term in the document. When searching for specific design drawings or construction records, the Vector Space Model and TF-IDF algorithms can quickly calculate the importance of relevant documents, providing more accurate query results. A graph database stores data in a graph structure, representing entities and their relationships using nodes and edges. A dynamic metadata model automatically updates the metadata model in a graph database when new data types or data relationships are detected. Role-based access control (RBAC) is a security management mechanism that controls access to resources by defining user roles and permissions.The RBAC mechanism manages user access rights through a role-permission mapping table. The encryption mechanism protects the security of data during transmission and storage by encrypting the data. Common encryption algorithms include AES, RSA, etc. In the present invention, the data security module encrypts sensitive data through the encryption mechanism AES to ensure the security of data during transmission and storage. The scheduled task scheduler is used to perform data backup operations regularly to ensure the reliability and integrity of the data. The incremental backup method only backs up data that has changed since the last backup, thereby reducing the backup time and storage space requirements.

[0064] The model layer is used to construct and update the BIM model including building structure, electromechanical system and equipment elements based on the output data of the data layer; the model layer obtains three-dimensional point cloud data through laser scanning, compares and verifies it with the design drawings, and if a difference is detected between the design drawings and the actual scanned data, the model is fine-tuned using the ICP iterative closest point algorithm; when constructing the BIM model, the model layer also assigns parametric attributes to each element in the model through parametric modeling, and if any equipment parameter changes, the relevant geometric shapes and attributes in the model are automatically updated through the parameter update mechanism; during the parameter update process, the parameter update mechanism automatically generates a design plan according to predefined rules through an automated rule engine, and the predefined rules include equipment layout rules and pipeline layout rules; during the parameter update process, the model layer also applies the changed data to the existing BIM model through the model update mechanism; the model update mechanism records the history of each data change through version control Git; during the use of the model, the model layer allows multiple users to edit the same BIM model at the same time through a real-time communication protocol; if multiple users edit the same part of the model at the same time, the conflict is resolved through the synchronization collaboration layer.

[0065] In a specific embodiment, laser scanning is a high-precision spatial measurement technology that generates three-dimensional point cloud data by emitting a laser beam and receiving the reflected signal. This point cloud data can accurately represent the geometry and position of real-world objects. In this application, the model layer uses laser scanning to acquire on-site three-dimensional point cloud data and compares and verifies it with the design drawings. This comparison process ensures the consistency between the design drawings and the actual situation, improving the accuracy and reliability of the model. Combining laser scanning with the ICP iterative closest point algorithm further enhances the accuracy of the model. For example, when discrepancies are detected between the design drawings and the actual scanned data, the ICP algorithm can fine-tune the model to ensure high consistency with the actual environment. The ICP (Iterative Closest Point) algorithm is an algorithm for point cloud registration that iteratively finds the optimal alignment between two sets of point clouds. By minimizing the distance error between the point clouds, the ICP algorithm gradually adjusts the position and posture of the target point cloud to align it with the reference point cloud. In this application, when discrepancies are detected between the design drawings and the actual scanned data, the ICP algorithm can fine-tune the model to ensure high consistency with the actual environment. This fine-tuning process not only improves model accuracy but also reduces the workload of manual adjustments. Parametric modeling is a method of describing the geometry and properties of a model by defining parameters. Parametric modeling allows every element in the model to be controlled by parameters, enabling flexible design and modification. Parametric modeling, combined with an automated rules engine, enables efficient parameter updates and design solution generation. For example, when device parameters change, parametric modeling can automatically update the geometry and properties of the model, while the automated rules engine automatically generates a new design solution based on predefined rules, ensuring design consistency and rationality.

[0066] Compared to existing technologies, the model layer of the present invention uses laser scanning and point cloud data processing to obtain high-precision three-dimensional point cloud data, which can then be compared and verified with the design drawings. This comparison process ensures consistency between the design drawings and the actual situation, improving the accuracy and reliability of the model. Compared to traditional manual measurement and modeling methods, laser scanning and point cloud data processing can significantly improve model accuracy and efficiency. Furthermore, using the ICP iterative closest point algorithm, the present invention can fine-tune the model when discrepancies are detected between the design drawings and the actual scanned data, ensuring high consistency between the model and the actual environment. This fine-tuning process not only improves model accuracy but also reduces the workload of manual adjustment. Compared to traditional manual adjustment methods, the ICP algorithm can complete model fine-tuning more quickly and accurately. Through parametric modeling and an automated rules engine, the present invention enables efficient parameter updates and design solution generation. Parametric modeling allows each element in the model to be controlled by parameters, enabling flexible design and modification. The automated rules engine automatically generates new design solutions based on predefined rules, ensuring design consistency and rationality. Compared with traditional manual design change methods, this parametric modeling and automated rules engine can significantly improve the efficiency and quality of design changes.

[0067] The synchronization and collaboration layer is used to receive the real-time model data output by the model layer, and perform real-time synchronization and conflict detection of model parameter data through the data synchronization mechanism; when the data is changed, the blockchain is used to record the data change history, and the synchronization mechanism is triggered according to the data change, and the changed data is pushed to all relevant participants in real time, and the real-time updated data is output to the decision analysis layer and the full life cycle management layer; at the same time, the change information is fed back to the model layer for updating; the working principle of the synchronization and collaboration layer is as follows: receiving the real-time model data of the model layer, and synchronizing it through the synchronization protocol based on the distributed database; when the data change is detected in the model layer, the changed data is processed and distributed through the distributed message queue, and the blockchain is written to the machine The change information is encapsulated into blocks, which are added to the blockchain after verification using a consensus algorithm. Each block contains the hash value of the previous block, forming an unalterable data chain. During the change process, data conflicts are identified through a hash algorithm and data fingerprints, and resolved through a three-way merge algorithm. If the conflict is not resolved, the relevant participants are notified through a real-time communication protocol to manually resolve the conflict and resubmit the change. If it is detected that the data change has been confirmed to be correct, the changed data is pushed to all relevant participants in real time through a real-time communication protocol. If it is detected that the data change has been confirmed to be correct and the conflict has been resolved, the real-time updated data is output to the decision analysis layer and the full life cycle management layer through the API interface, and the change information is fed back to the model layer through two-way data binding.

[0068] In a specific implementation, the distributed database synchronization protocol is a protocol for ensuring data consistency between multiple nodes. It defines a set of rules and mechanisms to enable data on different nodes to be synchronized. Common distributed database synchronization protocols include Raft, Paxos, etc. Applied in the present invention, when a data change is detected in the model layer, the distributed database synchronization protocol can ensure data consistency of all nodes, while the distributed message queue is responsible for processing and distributing the changed data. The distributed message queue is a middleware for processing and distributing messages. It realizes asynchronous communication by storing messages in a queue and consuming them on demand by consumers. Common distributed message queues include Kafka, RabbitMQ, etc. Applied in the present invention, when a data change is detected in the model layer, the changed data is processed and distributed through the distributed message queue. This mechanism not only improves the efficiency of data processing, but also reduces the coupling between systems.

[0069] The blockchain write mechanism is a technology used to record and verify data changes. It encapsulates change information into blocks, verifies them using a consensus algorithm, and then adds them to the blockchain. Each block contains the hash value of the previous block, forming an immutable data chain. In this application, when a data change is detected at the model layer, the blockchain write mechanism encapsulates the change information into blocks, verifies them using a consensus algorithm, and then adds them to the blockchain. This mechanism ensures the transparency and immutability of the changed data. A hash algorithm converts data of any length into a hash value of a fixed length. A data fingerprint is a unique identifier generated by hashing the data. Combining the hash algorithm with the data fingerprint can quickly identify data changes and conflicts. In this application, when a data change is detected at the model layer, the hash algorithm and the data fingerprint are used to identify data conflicts. This mechanism ensures data consistency and integrity. Combining the hash algorithm, data fingerprint, and three-way merge algorithm effectively resolves data conflicts. For example, when a data conflict is detected, the hash algorithm and data fingerprint can quickly identify the conflicting parts, while the three-way merge algorithm is used to resolve the conflict and ensure data consistency. The three-way merge algorithm is an algorithm used to resolve data conflicts. It compares three versions of data (baseline version, local version and remote version), finds differences and automatically merges them. Applied in the present invention, when a data conflict is detected, the data conflict is resolved through a three-way merge algorithm. This mechanism ensures the consistency and integrity of the data. Two-way data binding is a technology used to achieve data synchronization. It establishes a two-way connection between the data source and the view so that changes in the data source can be reflected in the view in real time, and vice versa. Applied in the present invention, when it is detected that the data change has been confirmed to be correct and the conflict has been resolved, the real-time updated data is output to the decision analysis layer and the full life cycle management layer through the API interface, and the change information is fed back to the model layer through two-way data binding. This mechanism ensures the consistency and real-time nature of the data.

[0070] Compared to existing technologies, the present invention achieves efficient data synchronization through a distributed database synchronization protocol and distributed message queues. The distributed database synchronization protocol ensures data consistency among all participants, while the distributed message queue improves data processing efficiency and system decoupling. Compared to traditional centralized synchronization mechanisms, this distributed synchronization mechanism can better meet the needs of large-scale data processing and improve system stability and scalability. Through the blockchain write mechanism, the present invention can record and verify data changes, ensuring the transparency and immutability of the changed data. The introduction of blockchain technology not only improves data security but also enhances data traceability. Compared to traditional logging mechanisms, the blockchain write mechanism provides higher data security and reliability. Through a hash algorithm, data fingerprinting, and a three-way merge algorithm, the present invention intelligently detects and resolves data conflicts. The hash algorithm and data fingerprinting can quickly identify data conflicts, while the three-way merge algorithm can automatically or assist in manual conflict resolution. Compared to traditional manual conflict resolution methods, this intelligent data conflict detection and resolution mechanism significantly improves the efficiency and accuracy of conflict resolution. Through a real-time communication protocol and two-way data binding, the present invention enables real-time data push and feedback. Real-time communication protocols ensure timely notification of data changes, while bidirectional data binding ensures data consistency and real-time performance. Compared to traditional batch processing, this real-time data push and feedback mechanism better supports multi-user collaboration and improves team efficiency.

[0071] The decision analysis layer is used to receive the output data of the model layer and the synchronous collaboration layer, perform data analysis, prediction and anomaly detection on the electromechanical system through the electromechanical system comprehensive analysis algorithm based on deep learning, and output analysis reports, early warning information and optimization suggestions to the full life cycle management layer and relevant participants; the prediction method of the electromechanical system comprehensive analysis algorithm based on deep learning is: based on the input feature vector X t , including historical energy consumption data (E t-n ,...,E t-1 ), weather data (M t-n ,...M t ,) building occupancy data ( t-n ,...O t ,) and device status data (R t-n ,...R t ,), where n represents the lag order of the time series; the energy consumption at future time points is predicted by a deep hybrid neural network, which uses a multi-layer long short-term memory network to extract the input feature vector X t The time series features T in r , and calculate the time series features T through the multi-head self-attention mechanism rCorrelation between different time steps, output of energy consumption forecast value at future time points The prediction formula is:

[0072]

[0073] In formula (1), W out represents the weight of the output layer; Q, K, V are query, key and value matrices respectively; d k represents the dimension of the key vector; α represents the activation function, which is used to introduce nonlinearity and enhance the expressiveness of the model; L represents the number of heads; LSTM represents the output of the LSTM layer; attn represents the output of the self-attention mechanism; W LSTM and b LSTM Represent the parameters and bias terms of the LSTM layer respectively; the anomaly detection method of the electromechanical system comprehensive analysis algorithm based on deep learning is: constructing a fault propagation model through a graph neural network, and the fault propagation model is based on the adjacency matrix A of the fault propagation graph and the device feature matrix X dev As input, the fault propagation pattern between devices is learned through multi-layer graph convolution operations, and the fault propagation probability P of each device is calculated. fault , the calculation formula is:

[0074]

[0075] In formula (2), is the weight matrix of the k-th neighbor in the l-th layer graph convolution, H (l) Represents the node feature matrix after the l-th layer of graph convolution; The attenuation coefficient that controls the spatial influence range is used to determine the degree of influence of spatial distance on weight; represents the graph convolution output feature vector of device d at layer l; X dev represents the device feature matrix; represents the parameters of the GNN model in the lth layer of graph convolution; during the prediction and anomaly detection process, the deep learning-based electromechanical system comprehensive analysis algorithm compares the calculation results with the preset threshold; if the deviation between the predicted energy consumption and the actual energy consumption exceeds the preset threshold, or the fault propagation probability of a certain equipment exceeds the preset threshold, the early warning mechanism is triggered, and the analysis report and early warning information are output to the full life cycle management and relevant participants.

[0076] In a specific embodiment, an LSTM is used to predict energy consumption at future time points. By combining multiple layers of LSTM, it can better capture complex time series patterns and improve prediction accuracy. Combining LSTM with a multi-head self-attention mechanism further enhances prediction accuracy and robustness. For example, when predicting energy consumption at future time points, the LSTM can capture long-term trends and cyclical changes in the time series, while the multi-head self-attention mechanism can focus on the importance of different time steps, thereby improving prediction accuracy. The multi-head self-attention mechanism is a technique used to enhance the model's attention to different parts of the input sequence. It uses multiple parallel self-attention mechanisms to extract information from different subspaces and perform a weighted combination of this information. This mechanism enables the model to better capture important features in the input sequence. In this application, the multi-head self-attention mechanism is used to enhance the performance of LSTM in time series prediction. Through the multi-head self-attention mechanism, the model can better focus on the importance of different time steps, thereby improving prediction accuracy. Graph neural networks are neural networks used to process graph-structured data. They transmit information through the connections between nodes and learn node representations through multi-layer graph convolution operations. GNNs can effectively capture local and global features within graph structures, making them suitable for modeling inter-device relationships and analyzing fault propagation. In this application, GNNs are used to construct fault propagation models. Through multi-layer graph convolution operations, GNNs can learn the fault propagation patterns between devices and calculate the fault propagation probability for each device. Combining GNNs with multi-layer graph convolution operations improves the accuracy and interpretability of fault propagation models. For example, when constructing a fault propagation model, GNNs can use multi-layer graph convolution operations to learn the fault propagation patterns between devices and calculate the fault propagation probability for each device. This mechanism not only improves fault detection accuracy but also provides a visual explanation of fault propagation between devices. A fault propagation model is used to describe and analyze fault propagation between devices. It uses a graph structure to represent the connectivity between devices and uses a graph neural network to learn the fault propagation patterns. In this application, the fault propagation model uses multi-layer graph convolution operations to learn the fault propagation patterns between devices and calculate the fault propagation probability for each device. This model can help identify fault sources and propagation paths, providing a basis for fault prevention and maintenance. Combining a fault propagation model with a graph neural network improves fault detection and prevention capabilities. For example, when detecting device failures, the fault propagation model can learn the fault propagation pattern through a graph neural network and calculate the fault propagation probability for each device. This mechanism not only improves the accuracy of fault detection but also provides a visual explanation of fault propagation between devices.

[0077] In practice, the deep learning-based electromechanical system comprehensive analysis algorithm implements data analysis, prediction, and anomaly detection of electromechanical systems through the following specific implementation methods:

[0078] The decision analysis layer receives output data from the model layer and the synchronization and coordination layer, including the real-time status of the electromechanical system, historical data, and change information. This data undergoes preprocessing, including data cleaning, noise removal, and normalization, to ensure data quality and consistency.

[0079] The decision analysis layer uses a multi-layer LSTM and multi-head self-attention mechanism to predict the energy consumption of the electromechanical system. The multi-layer LSTM captures long-term dependencies in the time series, while the multi-head self-attention mechanism focuses on the importance of different time steps. During model training, historical energy consumption data is used as input to predict energy consumption at future time points. The predicted results are compared with actual energy consumption. If the deviation exceeds a preset threshold, an early warning mechanism is triggered.

[0080] The decision analysis layer constructs a fault propagation model using a graph neural network. First, the devices in the electromechanical system are represented as nodes in a graph structure, and the connections between devices are represented as edges. Then, through multi-layer graph convolution operations, the fault propagation pattern between devices is learned. During model training, historical fault data is used as input to calculate the fault propagation probability for each device. If the fault propagation probability of a device exceeds a preset threshold, an early warning mechanism is triggered.

[0081] During the prediction and anomaly detection process, the decision analysis layer compares the calculated results with pre-set thresholds. If the deviation between the predicted and actual energy consumption exceeds the pre-set threshold, or if the probability of a device failure propagation exceeds the pre-set threshold, the system triggers an early warning mechanism. This mechanism sends analysis reports and early warning information to the full lifecycle management layer and relevant stakeholders via a real-time communication protocol. The system also records the history of early warning events for subsequent analysis and improvement.

[0082] Based on predictions and anomaly detection results, the decision analysis layer generates optimization recommendations. These recommendations include energy efficiency optimization measures, fault prevention measures, and maintenance plans. These recommendations are distributed via APIs to the full lifecycle management layer and relevant stakeholders, helping them make more informed decisions.

[0083] Compared to existing technologies, this approach, through the combination of a deep hybrid neural network and a multi-head self-attention mechanism, achieves precise predictions of future energy consumption. Compared to traditional time series forecasting methods, this method significantly improves prediction accuracy, providing a more reliable basis for energy management. Furthermore, a fault propagation model constructed using a graph neural network accurately captures the propagation patterns of faults across devices and calculates the fault propagation probability for each device. This helps to promptly identify potential fault risks and improve system stability and reliability. Furthermore, by setting appropriate thresholds, real-time monitoring and early warning of the electromechanical system status are achieved. When system anomalies occur, analysis reports and early warning information are generated promptly, providing managers with timely decision-making support. The entire algorithm is data-driven, fully leveraging both real-time monitoring and historical data from the electromechanical system to enhance the scientific and intelligent nature of management. Furthermore, the algorithm is highly scalable and flexible, adapting to electromechanical systems in buildings of varying sizes and types. Furthermore, as data accumulates, the algorithm's performance can be continuously optimized and improved.

[0084] In specific implementation, the hardware working environment of the deep learning-based electromechanical system comprehensive analysis algorithm applied to the BIM-based building electromechanical full life cycle management platform mainly includes: servers: used to run deep learning models and BIM platforms, with powerful computing power and large memory; graphics workstations: equipped with high-performance graphics cards for rendering and visualization of BIM models; data storage devices: such as SSD and NAS, used to store data of BIM models and deep learning models; sensors and Internet of Things devices: used to monitor the operating status of electromechanical systems in real time and provide data input; network communication equipment: such as switches and routers, to ensure real-time transmission and sharing of data.

[0085] A set of hardware was selected for comparison using a deep learning-based electromechanical system comprehensive analysis algorithm (Group A) and a traditional method (Group B). Traditional methods primarily analyze and diagnose electromechanical systems based on rules and experience, such as using threshold judgment and expert systems for fault warning. The experiments were conducted under identical conditions, with each group performing five runs. The mechatronic system's fault diagnosis accuracy, analysis time, and resource consumption were recorded. The experimental data is shown in Table 2.

[0086] Table 2 Algorithm comparison experiment record

[0087]

[0088] Through comparative experiments, we found that the deep learning-based electromechanical system comprehensive analysis algorithm (Group A) achieved significantly higher fault diagnosis accuracy than the traditional method (Group B), while also reducing analysis time and resource consumption. This demonstrates that deep learning algorithms offer high efficiency and accuracy in comprehensive electromechanical system analysis, contributing to the advancement of intelligent building electromechanical lifecycle management platforms.

[0089] The full lifecycle management layer receives the output data from the decision analysis layer and, in combination with historical data from the data layer, uses a full lifecycle fusion algorithm to integrate and manage full lifecycle information. It automatically triggers management processes based on project phases and outputs full lifecycle management reports, project progress information, and BIM model optimization recommendations to relevant stakeholders and decision makers. The full lifecycle management layer includes an information integration module, a management process triggering module, an intelligent report generation module, and an optimization recommendation and feedback module. The information integration module uses a federated learning approach to fuse the output data from the decision analysis layer with historical data from the data layer and generates a comprehensive data representation using a graph neural network. The information integration module also uses a knowledge graph to build a knowledge base of full lifecycle information, storing data entities and relationships. The management process triggering module uses the data generated by the information integration module to automate processes using a business process management system and an event-driven architecture. The intelligent report generation module generates full lifecycle management reports using natural language processing and a template engine. The optimization recommendation and feedback module generates optimization recommendations using reinforcement learning and operations research methods. Within the full lifecycle management layer, upon receiving the output data from the decision analysis layer, this data is fused with historical data from the data layer using a federated learning algorithm. Federated learning allows joint training of models without sharing the original data, thereby protecting data privacy. At the same time, graph neural networks are used to capture complex relationships between data and generate more comprehensive data representations. If it is detected that there is a correlation between the data but the format is inconsistent, the graph neural network is used to align and transform the features so that the data can be represented uniformly. Next, the information management unit performs information management and knowledge discovery through the knowledge graph and semantic reasoning engine. If the data fusion unit generates a unified data representation, a knowledge base of full life cycle information is constructed through the knowledge graph. The knowledge graph not only stores data entities and their relationships, but also performs logical inference through the semantic reasoning engine to discover implicit relationships and patterns. If a new data type or data relationship is detected, the knowledge graph is dynamically updated to maintain the integrity and consistency of the information. In addition, new knowledge and insights can be automatically generated through the semantic reasoning engine to provide valuable insights for decision makers.

[0090] In the automatic management process trigger module, if the data generated by the full life cycle information integration module is received, the business processes of each stage of the project are defined and managed through the BPM system. The BPM system automatically triggers the corresponding management process based on preset rules and conditions. If a specific event is detected (such as a change in project stage, completion of a key node, etc.), the corresponding process task is triggered through EDA. For example, if the design stage is detected to be completed, the construction preparation process is automatically triggered; if the construction stage is detected to be completed, the acceptance process is automatically triggered. The task scheduling unit optimizes the task execution sequence through the DAG scheduling algorithm to ensure that tasks are executed in the correct order and maximize resource utilization.

[0091] If a full lifecycle management report is required, NLP technology is used to extract key information and data points, and a template engine is used to generate a structured report. NLP technology can automatically identify and extract important information from text, such as project progress, cost analysis, and risk assessment. The template engine then populates the extracted information into the report according to the preset template format. If new reporting requirements or format changes are detected, the template is dynamically adjusted to accommodate different reporting needs. The visualization unit uses data visualization tools to generate interactive charts and dashboards, helping decision makers quickly understand project status and trends.

[0092] If optimization recommendations are needed, specific optimization strategies are generated using reinforcement learning algorithms and operations research methods. If changes in project status or new constraints are detected, the optimization strategy is recalculated to adapt to the new situation. The feedback collection unit collects user feedback through feedback collection mechanisms (such as questionnaires and user feedback systems). If user feedback is detected, the feedback content is analyzed using natural language processing technology to extract key information. If issues or suggestions are identified in the feedback, corresponding improvement measures are triggered to enhance system performance and user experience.

[0093] Further, such as Figure 4As shown: The working method of the full life cycle fusion algorithm is: S1. Calculate the weights of the time series data and spatial data of each data sample through the Softmax function; S2. Apply the calculated weights to the time series data and spatial data, and generate the fused multimodal feature vector through element-by-element multiplication; S3. Calculate the influence factor of the fused feature through the nonlinear activation function to determine the importance and contribution of different information sources; S4. Combine the reduced dimensionality data with the historical data, and generate an integrated full life cycle information matrix through the weighted summation method, which contains all relevant historical and current data; S5. Calculate the event trigger condition through linear transformation to determine whether the management process needs to be triggered; S6. Compare the calculated event trigger condition with the preset threshold. If the trigger condition exceeds the threshold, the corresponding management process is triggered; otherwise, it is not triggered; S7. Estimate the expected reward of taking a certain action in the current state by recursively calculating the state action value function, and select the action with the highest expected reward as the optimal action to generate optimization suggestions.

[0094] In a specific embodiment, the Softmax function is used to calculate the weights of the time series data and spatial data for each data sample. The Softmax function can be used to determine the importance of different types of data, thus providing a basis for subsequent data fusion. The Softmax function is combined with element-by-element multiplication to generate a fused multimodal feature vector. For example, after calculating the weights of the time series data and spatial data, the Softmax function can apply these weights to the original data and generate a fused multimodal feature vector through element-by-element multiplication. Element-by-element multiplication is used to apply the calculated weights to the time series data and spatial data to generate a fused multimodal feature vector. Element-by-element multiplication allows different types of feature data to be weighted and combined to form a more representative feature vector. After calculating the impact factors of the fused features, a nonlinear activation function can enhance the expressive power of the feature vector, while a weighted summation method combines the reduced-dimensional data with historical data to generate an integrated full-lifecycle information matrix. The weighted summation method is combined with a linear transformation to determine whether a management process needs to be triggered. For example, after generating the integrated full-lifecycle information matrix, a linear transformation can be used to calculate event trigger conditions to determine whether a management process needs to be triggered. Recursively calculating the state-action-value function (SAVF) is a method that uses recursive calculation to estimate the expected reward of taking an action in the current state. It is commonly used in reinforcement learning to recursively update the SAVF and select the optimal action. In this application, the recursive SAVF is used to estimate the expected reward of taking an action in the current state, select the action with the highest expected reward as the optimal action, and generate optimization recommendations. By recursively calculating the SAVF, optimal management strategies can be found, improving management efficiency.

[0095] In practice, the full lifecycle fusion algorithm works as follows: It receives output data from the decision analysis layer, including energy consumption forecasts, fault detection results, and optimization recommendations. Simultaneously, it obtains historical data from the data layer, such as design drawings, construction records, and equipment parameters. This data undergoes preliminary processing, including data cleaning, denoising, and normalization, to ensure data quality and consistency. Next, the Softmax function is used to convert the multidimensional vector into a probability distribution to determine the importance of different data types. For example, when processing energy consumption forecast data, the Softmax function can calculate the weights of time series data and spatial data to determine the relative importance of each type. Third, when generating the fused multimodal feature vector, element-by-element multiplication combines the time series and spatial data according to their weights. Fourth, when calculating the impact factors of the fused features, a nonlinear activation function is used to enhance the expressive power of the feature vector, allowing it to better reflect the importance and contribution of different information sources. Fifth, when generating the integrated full lifecycle information matrix, a weighted summation method is used to combine the reduced-dimensionality data and historical data according to their weights. Sixth, when calculating event trigger conditions, a linear transformation can be used to map the entire lifecycle information matrix to a simple trigger condition. Seventh, when determining whether a management process needs to be triggered, the system compares the calculated trigger condition with a preset threshold. If the trigger condition exceeds the threshold, the corresponding management process is automatically triggered. When generating optimization recommendations, the state-action-value function is recursively calculated to select the action with the highest expected reward and generate the corresponding optimization recommendation.

[0096] In the specific implementation, the hardware working environment of the full life cycle fusion algorithm includes: server, BIM model server, data acquisition equipment, visualization terminal equipment, network communication equipment; a group of identical hardware environments were selected to apply the full life cycle fusion algorithm (Group A) and the traditional method (Group B, i.e. existing commonly used algorithms, such as threshold monitoring, trend analysis, statistical control and expert system, etc.) for comparative experiments. During the experiment, both groups received the same electromechanical system operating status data and processed and analyzed them under the same experimental conditions. The experiment recorded three key indicators: data processing time, accuracy and decision-making efficiency to evaluate the performance of the two methods; among them, the traditional method is based on a series of preset rules and simple mathematical models to monitor and analyze the operating status of building electromechanical systems. These methods include:

[0097] Threshold monitoring: Set upper and lower limits for key parameters and trigger alarms when data exceeds these thresholds.

[0098] Trend analysis: Observe the performance trends of electromechanical systems over time by plotting historical data.

[0099] Statistical control: Use statistical methods (such as mean, standard deviation, etc.) to evaluate the stability of system performance.

[0100] Expert system: Provides decision support through rule-based reasoning based on expert knowledge and experience.

[0101] The experimental results are recorded in tabular form, as shown in Table 3:

[0102] Table 3. Comparative experimental record of full life cycle fusion algorithm

[0103]

[0104] Comparative experiments show that the full life cycle fusion algorithm significantly outperforms traditional methods (existing commonly used algorithms) in terms of data processing time, accuracy, and decision-making efficiency. The average data processing time for Group A was only about 120 seconds, far lower than the 300 seconds for Group B. At the same time, Group A's accuracy was as high as over 97%, which was able to more accurately reflect the true state of the electromechanical system. In addition, Group A's decision-making efficiency was also significantly higher than that of Group B, enabling it to provide effective decision support more quickly. Therefore, it can be concluded that the full life cycle fusion algorithm has significant advantages in the BIM-based building electromechanical full life cycle management platform, can significantly improve the efficiency and accuracy of data processing, and provide more powerful and intelligent support for project management and decision-making.

[0105] A BIM-based building electromechanical full life cycle management optimization strategy, such as Figure 5 As shown: It includes the following steps: Step 1, constructing a BIM model of the electromechanical system according to the building design drawings and equipment parameters, wherein the BIM model of the electromechanical system integrates at least equipment operation and maintenance data and energy consumption data information through the API interface;

[0106] Step 2: Based on the data integrated by the BIM model, the operating status of the electromechanical system is analyzed, predicted, and anomalies are detected through a comprehensive electromechanical system analysis algorithm based on deep learning;

[0107] Step 3: Based on the output results of step 2, combined with the project management knowledge base, the full life cycle information is integrated and managed through the full life cycle fusion algorithm to generate optimization suggestions or decision support reports, including equipment maintenance plans and energy consumption optimization plans;

[0108] Step 4: In the full lifecycle management of electromechanical systems, blockchain is used to record decision-making and change information during the design, construction, and operation and maintenance phases. Smart contracts are used to automatically execute collaborative workflows, including approval, notification, and feedback, to improve collaborative efficiency and information transparency.

[0109] Step 5: Use blockchain to record the collaborative work process and decision-making results of all parties involved, and use a real-time data stream processing engine to achieve real-time data synchronization and collaborative work during the design, construction, and operation and maintenance stages;

[0110] Step 6. Based on the actual operation of the electromechanical system and the optimization decision support report, dynamically adjust equipment parameters and maintenance plans to optimize system performance and reduce operation and maintenance costs. Continuously analyze data through the particle swarm optimization algorithm to continuously iterate and optimize the BIM model and management strategy.

[0111] Although specific embodiments of the present invention have been described above, those skilled in the art will appreciate that these specific embodiments are merely illustrative, and that those skilled in the art may omit, substitute, and modify the details of the methods and systems described above without departing from the principles and spirit of the present invention. For example, combining the above method steps to perform substantially the same functions in substantially the same manner to achieve substantially the same results falls within the scope of the present invention. Accordingly, the scope of the present invention is limited solely by the appended claims.

Claims

1. A BIM-based building electromechanical full life cycle management platform, characterized by: include: Data layer, model layer, synchronization and collaboration layer, decision analysis layer and full life cycle management layer; The data layer is used to collect at least design drawings, construction records, and equipment parameter data, convert non-standard format data into a standard format, and perform data cleaning, denoising, and normalization through machine learning models. The data layer is also used to distribute and store pre-processed data through a cloud management system, establish a data index and metadata management framework, and output standardized, searchable data sets to the model layer and decision analysis layer. The model layer is used to construct and update the BIM model including building structure, electromechanical system and equipment elements based on the output data of the data layer; The synchronization and collaboration layer is used to receive the real-time model data output by the model layer, and perform real-time synchronization and conflict detection of model parameter data through the data synchronization mechanism. When the data is changed, the blockchain is used to record the data change history. The synchronization mechanism is triggered according to the data change, and the changed data is pushed to all relevant participants in real time. The real-time updated data is output to the decision analysis layer and the full life cycle management layer. At the same time, the change information is fed back to the model layer for updating. The decision analysis layer is used to receive the output data of the model layer and the synchronous collaboration layer, perform data analysis, prediction and anomaly detection on the electromechanical system through the electromechanical system comprehensive analysis algorithm based on deep learning, and output analysis reports, early warning information and optimization suggestions to the full life cycle management layer and relevant participants; the prediction method of the electromechanical system comprehensive analysis algorithm based on deep learning is: based on the input feature vector , including historical energy consumption data ( ), weather data ( ), building occupancy data ( ) and device status data ( ),in Represents the lag order of the time series; predicts energy consumption at future time points through a deep hybrid neural network that uses a multi-layer long short-term memory network to extract input feature vectors Time series features in , and calculate the time series features through the multi-head self-attention mechanism Correlation between different time steps, output of energy consumption forecast value at future time points ; The prediction formula is: （1） In formula (1), represents the weight of the output layer; are query, key, and value matrices respectively; represents the dimension of the key vector; Represents the activation function, which is used to introduce nonlinearity and enhance the expressiveness of the model; Indicates the number of heads; Represents the output of the LSTM layer; Represents the output of the self-attention mechanism; and Represent the parameters and bias terms of the LSTM layer respectively; the anomaly detection method of the electromechanical system comprehensive analysis algorithm based on deep learning is: constructing a fault propagation model through a graph neural network, and the fault propagation model is based on the adjacency matrix of the fault propagation graph and device feature matrix As input, it learns the fault propagation pattern between devices through multi-layer graph convolution operations and calculates the fault propagation probability of each device , the calculation formula is: （2） In formula (2), For the The first layer of graph convolution The weight matrix of the neighbors of order , Indicates the Node feature matrix after layer graph convolution; The attenuation coefficient that controls the spatial influence range is used to determine the degree of influence of spatial distance on weight; Representation device In the The graph convolution of the layer outputs a feature vector; represents the device feature matrix; Indicates the The parameters of the GNN model in the layer graph convolution; during the prediction and anomaly detection process, the deep learning-based electromechanical system comprehensive analysis algorithm compares the calculated results with the preset threshold; if the deviation between the predicted energy consumption and the actual energy consumption exceeds the preset threshold, or the fault propagation probability of a certain device exceeds the preset threshold, the early warning mechanism is triggered and the analysis report and early warning information are output to the full life cycle management and relevant participants; The full life cycle management layer is used to receive the output data of the decision analysis layer, combine it with the historical data in the data layer, and use the full life cycle fusion algorithm to integrate and manage the full life cycle information; automatically trigger the management process according to the project stage, and output the full life cycle management report, project progress information and BIM model optimization suggestions to relevant participants and decision makers.

2. The BIM-based building electromechanical full lifecycle management platform according to claim 1 is characterized by: The data layer converts non-standard format data into standard format by matching specific character sequences in the data using a regular expression-based pattern matching algorithm to identify the format characteristics of the data, and then analyzing the words, phrases, and sentences in the data using a semantic analysis-based parsing algorithm to further identify the data format type. If the input data format matches any format in the preset standard format library, the identified non-standard data fields are matched with the corresponding fields in the standard format through the format mapping mechanism, and the non-standard data fields are converted one by one into standard format fields through key-value pair mapping and regular expression replacement; if the input data format cannot be directly matched with any item in the standard format library, the input data is feature extracted and classified through clustering analysis to build a new data format model; after the data format model is successfully built, the new data format model is added to the standard format library and the mapping rules are automatically updated.

3. The BIM-based building electromechanical full lifecycle management platform according to claim 1 is characterized by: During the data conversion process, the machine learning model first traverses each data record through the autoencoder, automatically learns and extracts effective features in the data, and suppresses noise and redundant information. During the data traversal process, if the characteristic value is detected to be beyond the preset threshold range, the anomaly detection mechanism is triggered, and the redundant, missing or erroneous information in the abnormal characteristic value is corrected or filled by the statistical outlier detection algorithm. The statistical outlier detection algorithm fills the abnormal characteristic value by the mean, median or interpolation method. Then, the machine learning model denoises the data through the generative adversarial network. If the denoised data is normally distributed, the machine learning model uses the Z-score normalization method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. If the distribution range of the denoised data is within a preset fixed range, the minimum and maximum scaling method is used to linearly map the data to the interval [0,1] or [-1,1].

4. The BIM-based building electromechanical full lifecycle management platform according to claim 1 is characterized by: The cloud management system includes a distributed storage module, a data index and metadata management module, a data security module and a data backup module; the distributed storage module includes a data sharding unit and a data synchronization unit; the data sharding unit is used to shard and distribute data to different storage nodes through a consistent hashing algorithm; the data synchronization unit is used to ensure that the data copies on all storage nodes remain consistent through the Raft consensus algorithm when a data update operation is detected; the data index and metadata management module includes an index construction unit and a metadata management unit; the index construction unit is used to divide the index into multiple small blocks using a block index strategy to improve query performance, and calculate the relevance of documents through a vector space model and a TF-IDF algorithm; the metadata management unit is used to dynamically update the metadata model in the graph database when a new data type or data relationship is detected; the data security module is used to reject unauthorized user access requests and record logs through a role-based access control mechanism, and is also used to encrypt sensitive data through an encryption mechanism; the data backup module is used to regularly perform data backup operations through a scheduled task scheduler. During the backup process, the data backup module reduces backup time and storage space through an incremental backup method.

5. The BIM-based building electromechanical full lifecycle management platform according to claim 1 is characterized by: The model layer obtains 3D point cloud data through laser scanning, compares and verifies it with the design drawings, and if a discrepancy is detected between the design drawings and the actual scanned data, fine-tunes the model using the ICP iterative closest point algorithm. When constructing a BIM model, the model layer also uses parametric modeling to assign parametric properties to each element in the model. If any equipment parameter changes, the relevant geometry and properties in the model are automatically updated through a parameter update mechanism. During the parameter update process, the parameter update mechanism automatically generates a design plan based on predefined rules using an automated rule engine. The predefined rules include equipment layout rules and piping layout rules. During the parameter update process, the model layer also applies the changed data to the existing BIM model through the model update mechanism. The model update mechanism records the history of each data change through the version control Git. During the use of the model, the model layer allows multiple users to edit the same BIM model at the same time through a real-time communication protocol; if multiple users edit the same part of the model at the same time, conflicts are resolved through the synchronization collaboration layer.

6. The BIM-based building electromechanical full lifecycle management platform according to claim 1 is characterized by: The synchronization and collaboration layer works as follows: it receives real-time model data from the model layer and synchronizes it through a distributed database-based synchronization protocol. When data changes are detected in the model layer, the changed data is processed and distributed through a distributed message queue, and the changed information is encapsulated as a block through a blockchain write mechanism. After verification using a consensus algorithm, it is added to the blockchain. Each block contains the hash value of the previous block, forming an unalterable data chain. During the change process, data conflicts are identified through hashing algorithms and data fingerprints, and resolved through a three-way merge algorithm. If the conflict is not resolved, notify the relevant parties through the real-time communication protocol to manually resolve the conflict and resubmit the changes; If the data change is detected and confirmed to be correct, the changed data will be pushed to all relevant participants in real time through the real-time communication protocol; If it is detected that the data change has been confirmed to be correct and the conflict has been resolved, the real-time updated data will be output to the decision analysis layer and the full life cycle management layer through the API interface, and the change information will be fed back to the model layer through two-way data binding.

7. The BIM-based building electromechanical full lifecycle management platform according to claim 1 is characterized by: The full life cycle management layer includes an information integration module, a management process triggering module, an intelligent report generation module and an optimization suggestion and feedback module; the information integration module is used to adopt a federated learning method to fuse the output data of the decision analysis layer with the historical data in the data layer, and generate a comprehensive data representation through a graph neural network. The information integration module is also used to build a knowledge base of full life cycle information through a knowledge graph, storing data entities and data relationships; the management process triggering module is used to realize process automation through a business process management system and an event-driven architecture based on the data generated by the information integration module; the intelligent report generation module is used to generate a full life cycle management report through natural language processing and a template engine; the optimization suggestion and feedback module is used to generate optimization suggestions through reinforcement learning and operations research methods.

8. The BIM-based building electromechanical full lifecycle management platform according to claim 1 is characterized by: The working method of the full life cycle fusion algorithm is: S1. Calculate the weights of the time series data and spatial data of each data sample through the Softmax function; S2, applying the calculated weights to the time series data and spatial data, and generating the fused multimodal feature vector through element-by-element multiplication; S3. Calculate the influence factor of the fusion feature through a nonlinear activation function to determine the importance and contribution of different information sources; S4. Combine the reduced-dimensional data with the historical data and generate an integrated full-life cycle information matrix using a weighted summation method, which includes all relevant historical and current data; S5. Calculate the event triggering condition through linear transformation to determine whether the management process needs to be triggered; S6. Compare the calculated event trigger condition with the preset threshold. If the trigger condition exceeds the threshold, trigger the corresponding management process. Otherwise, it is not triggered; S7. Estimate the expected reward of taking an action in the current state by recursively calculating the state-action value function, select the action with the highest expected reward as the optimal action, and generate optimization suggestions.

9. A BIM-based building electromechanical full life cycle management optimization strategy, characterized by: A BIM-based building electromechanical full lifecycle management platform as described in any one of claims 1 to 8 comprises the following steps: Step 1: Build a BIM model of the electromechanical system based on the building design drawings and equipment parameters. The BIM model of the electromechanical system integrates at least equipment operation and maintenance data and energy consumption data through an API interface. Step 2: Based on the data integrated by the BIM model, the operating status of the electromechanical system is analyzed, predicted, and anomalies are detected through a comprehensive electromechanical system analysis algorithm based on deep learning; Step 3: Based on the output results of step 2, combined with the project management knowledge base, the full life cycle information is integrated and managed through the full life cycle fusion algorithm to generate optimization suggestions or decision support reports, including equipment maintenance plans and energy consumption optimization plans; Step 4: In the full lifecycle management of electromechanical systems, blockchain is used to record decision-making and change information during the design, construction, and operation and maintenance phases. Smart contracts are used to automatically execute collaborative workflows, including approval, notification, and feedback, to improve collaborative efficiency and information transparency. Step 5: Use blockchain to record the collaborative work process and decision-making results of all parties involved, and use a real-time data stream processing engine to achieve real-time data synchronization and collaborative work during the design, construction, and operation and maintenance stages; Step 6. Based on the actual operation of the electromechanical system and the optimization decision support report, dynamically adjust equipment parameters and maintenance plans to optimize system performance and reduce operation and maintenance costs. Continuously analyze data through the particle swarm optimization algorithm to continuously iterate and optimize the BIM model and management strategy.

Images