Management method, device, equipment and storage medium of smart space system

Abstract

The present application relates to the field of digital twin technology, and discloses a management method, device, equipment and storage medium for a smart space system. The method includes: multi-dimensional fusion processing of the physical data and business data of the smart space to construct a digital twin model of the smart space; scenario analysis to obtain integrated scene data of the smart space; bidirectional long-short term memory network training to obtain a predictive maintenance model of the smart space; reinforcement learning optimization to obtain a dynamic energy management solution for the smart space; integration of the digital twin model of the smart space and the dynamic energy management solution for the smart space to obtain a full-cycle value evaluation system for the smart space; self-learning processing of the full-cycle value evaluation system for the smart space to generate a target smart space management strategy. The present application constructs a complete digital twin model of the smart space to describe the smart space from different levels and granularities, thereby improving the problem response speed and processing efficiency.

Description

Technical Field

[0001] The present application relates to the field of digital twin technology, and in particular to a management method, device, equipment and storage medium for a smart space system. Background Art

[0002] With the acceleration of urbanization and the rapid development of digital technology, smart space systems have become an important part of modern urban management and building operations. This system aims to integrate physical space with digital technology, improve space utilization efficiency, optimize energy management, and enhance user experience. However, traditional smart space management methods often have problems such as data silos, system fragmentation, and delayed response, making it difficult to achieve comprehensive optimization of space resources and maximize long-term value.

[0003] Current smart space systems usually lack deep integration of physical space and business data, making it difficult to fully grasp the space's operating status and potential problems. At the same time, existing management methods are mostly based on static rules, which are difficult to adapt to complex and changing space usage scenarios and energy needs. In addition, most systems lack the ability to evaluate and predict the value of the space throughout its life cycle, and cannot provide strong support for long-term decision-making. Summary of the invention

[0004] The present application provides a management method, device, equipment and storage medium for a smart space system, which is used to build a complete smart space digital twin model to describe the smart space from different levels and granularities, thereby improving problem response speed and processing efficiency.

[0005] In a first aspect, the present application provides a management method for a smart space system, the management method for the smart space system comprising:

[0006] Perform multi-dimensional fusion processing on the physical data and business data of the smart space to build a digital twin model of the smart space;

[0007] Perform scenario analysis on the smart space digital twin model to obtain smart space scenario integrated data;

[0008] Performing bidirectional long short-term memory network training on the smart space scene integrated data to obtain a smart space predictive maintenance model;

[0009] Performing reinforcement learning optimization on the smart space predictive maintenance model to obtain a smart space dynamic energy management solution;

[0010] Integrate the smart space digital twin model and the smart space dynamic energy management solution to obtain a full-cycle value evaluation system for the smart space;

[0011] The smart space full-cycle value assessment system is self-learned to generate a target smart space management strategy.

[0012] In a second aspect, the present application provides a management device for a smart space system, the management device for the smart space system comprising:

[0013] The fusion module is used to perform multi-dimensional fusion processing on the physical data and business data of the smart space and build a digital twin model of the smart space;

[0014] An analysis module, used to perform scenario analysis on the smart space digital twin model to obtain smart space scenario integrated data;

[0015] A training module, used for performing bidirectional long short-term memory network training on the smart space scene integrated data to obtain a smart space predictive maintenance model;

[0016] An optimization module, used to perform reinforcement learning optimization on the smart space predictive maintenance model to obtain a smart space dynamic energy management solution;

[0017] An integration module, used to integrate the smart space digital twin model and the smart space dynamic energy management solution to obtain a full-cycle value evaluation system for the smart space;

[0018] A generation module is used to perform self-learning processing on the full-cycle value evaluation system of the smart space and generate a target smart space management strategy.

[0019] The third aspect of the present application provides a management device for a smart space system, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the management device of the smart space system executes the above-mentioned management method of the smart space system.

[0020] The fourth aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned smart space system management method.

[0021] In the technical solution provided by this application, a complete digital twin model of smart space is constructed through multi-dimensional fusion processing of physical data and business data, and accurate mapping of physical space and digital space is achieved. The scenario analysis based on the digital twin model can capture and identify the dynamic changes of the smart space in real time. The bidirectional long short-term memory network is used for training to obtain a maintenance model with predictive capabilities, which can identify potential problems in advance and reduce the risk of equipment failure and maintenance costs. By optimizing the predictive maintenance model through reinforcement learning, the generation of dynamic energy management solutions is realized, energy efficiency is improved, and operating costs are reduced. The digital twin model and the dynamic energy management solution are integrated to build a full-cycle value evaluation system. Through self-learning processing, the management strategy can be automatically adjusted and optimized according to historical data and real-time conditions, which improves the flexibility and adaptability of management. The multi-scale modeling method can describe the smart space from different levels and granularities. Through time series analysis and pattern matching, abnormal situations and key events can be quickly identified, which improves the problem response speed and processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic diagram of an embodiment of a management method of a smart space system in an embodiment of the present application;

[0024] Figure 2 This is a schematic diagram of an embodiment of a management device for a smart space system in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The embodiments of the present application provide a management method, device, equipment and storage medium for a smart space system. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the management method of the smart space system in the embodiment of the present application includes:

[0027] Step S101: Perform multi-dimensional fusion processing on the physical data and business data of the smart space to build a digital twin model of the smart space;

[0028] It is understandable that the execution subject of the present application may be a management device of the smart space system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.

[0029] Specifically, the physical data of the smart space is processed by three-dimensional modeling. By using three-dimensional modeling tools or scanning technology, the physical structure and layout information of the smart space is converted into a three-dimensional geometric model. The model describes the various physical entities and their geometric relationships in the smart space. At the same time, the business data of the smart space is semantically annotated. Based on the business logic and application requirements, semantic labels and annotations are added to the business data to form the semantic information layer of the smart space. The information layer contains the meaning, purpose and association of the business data. According to the three-dimensional geometric model and the semantic information layer, the topological relationship analysis of the smart space is carried out. Based on the joint data of the three-dimensional geometric model and the semantic information layer, the spatial relationship and business association between the various entities in the smart space are analyzed to generate the topological relationship diagram of the smart space. The topological relationship diagram reflects the spatial connection and positional relationship of each physical entity, and reveals the logical association and interaction between business data. The three-dimensional geometric model, the semantic information layer and the topological relationship diagram are mapped to the graph database to construct the knowledge graph of the smart space. The knowledge graph is a graphical representation of all physical and business data and their relationships in the smart space. The complex multidimensional data relationships are stored and managed in the form of nodes and edges through the graph database technology. At the same time, the real-time sensor data of the smart space is analyzed in time series. The time series analysis technology is used to process and analyze the time change rules in the sensor data, and the dynamic attribute model of the smart space is constructed. The dynamic attribute model reflects the change characteristics and behavior patterns of the smart space in the time dimension. According to the constructed knowledge graph and dynamic attribute model, the smart space is modeled in multi-scale. By comprehensively considering the different scales and hierarchical characteristics of the smart space, a multi-scale digital model of the smart space containing multi-level and multi-scale information is constructed. This model can more comprehensively and meticulously reflect the complex structure and dynamic behavior of the smart space. The multi-scale digital model is checked for data consistency, and data correction rules for the smart space are formulated. The data correction is used to ensure that the data in the multi-scale digital model is accurate, consistent and correct. According to the correction rules, the multi-scale digital model is corrected to obtain a corrected and accurate digital model of the smart space. The corrected digital model is visualized and rendered. Through graphics processing and rendering technology, the corrected digital model is converted into an intuitive and easy-to-understand three-dimensional visualization model. The three-dimensional visualization model and the knowledge graph are fused to obtain a digital twin model of the smart space.

[0030] Step S102: Perform scenario analysis on the smart space digital twin model to obtain smart space scenario integrated data;

[0031] Specifically, CAD technology is used to accurately model the smart space, and two-dimensional and three-dimensional basic models including floor plans, elevations, and sections are created. Based on these basic models, a BIM model containing rich building information is constructed, covering key elements such as building components, equipment systems, and material information. In order to improve the visual effect and spatial expression of the model, the BIM model is converted into a high-precision 3D model, and geometric optimization and detail improvement are performed. By checking the consistency of the BIM model and the 3D model, the accurate correspondence between the digital model and the actual building is ensured. In the physical space, an IoT sensor network is deployed to collect real-time non-graphic data such as temperature, humidity, and energy consumption. These data are cleaned, filtered, and standardized to ensure data quality and consistency. A data fusion platform is constructed to integrate BIM models, 3D models, and processed IoT data into a unified multi-source heterogeneous data management system. Semantic annotation and association analysis are performed on the data in the system to establish a semantic network for the smart space and improve the comprehensibility and usability of the data. Using real-time data and semantic networks, the 3D model is dynamically updated to achieve synchronization between the virtual model and the physical building, forming a digital twin prototype. By integrating machine learning algorithms, the model is endowed with intelligent analysis and prediction capabilities, and ultimately a fully functional digital twin model of the smart space is built.

[0032] Step S103: Perform bidirectional long short-term memory network training on the smart space scene integrated data to obtain a smart space predictive maintenance model;

[0033] Specifically, the integrated data of the smart space scene is segmented into time windows to obtain a time series data sample set, and the time series data sample set is subjected to data standardization to obtain a standardized time series data set. A bidirectional long short-term memory network structure is constructed based on the standardized time series data set to obtain an initial bidirectional LSTM model. In the process of constructing the initial bidirectional LSTM model, the number of layers of the bidirectional LSTM network, the number of neurons in each layer, and other network hyperparameters are designed to ensure that the network structure can effectively capture the complex patterns in the time series data. The initial bidirectional LSTM model is forward propagated and the output of the neurons is calculated layer by layer to obtain the output result of the model, and the loss function value is calculated based on the model output result and the actual label to obtain the model error. The loss function value usually uses indicators such as mean square error or cross entropy to measure the difference between the model output and the actual label. According to the calculated model error, back propagation is performed, the gradient of the network parameters is calculated by the back propagation algorithm, and the model parameters are optimized and updated according to the network parameter gradient to obtain an updated bidirectional LSTM model. In the parameter optimization process, a gradient descent or its variant algorithm (such as the Adam optimization algorithm) is used to gradually adjust the network parameters to minimize the loss function value and improve the prediction accuracy of the model. Based on the updated bidirectional LSTM model, cross-validation is performed to evaluate the generalization ability of the model by dividing the data set into multiple subsets, repeatedly training and validating the model, and obtaining the model performance evaluation indicators. These evaluation indicators can include precision, recall, F1 score, etc. to comprehensively measure the performance of the model. According to the model performance evaluation indicators, the network structure is adjusted to optimize the number of layers, number of neurons in each layer, activation function and other parameters of the bidirectional LSTM model to ensure the good performance of the model on complex time series data, and obtain the optimized bidirectional LSTM model. The optimized bidirectional LSTM model is trained for predictive maintenance recognition. By introducing more training data and optimization algorithms, the model can accurately identify and predict potential maintenance needs in the smart space, and obtain the predictive maintenance model of the smart space. The model can monitor various types of data in the smart space in real time, detect possible problems in advance, and provide maintenance suggestions, thereby effectively improving the operating efficiency and reliability of the smart space.

[0034] Step S104: Perform reinforcement learning optimization on the smart space predictive maintenance model to obtain a smart space dynamic energy management solution;

[0035] Specifically, the state space modeling of the smart space predictive maintenance model is carried out to obtain the smart space energy state vector, and the action space is defined according to the smart space energy state vector to obtain the energy management action set. By analyzing various energy consumption data and equipment operation status in the smart space, the state vector that can reflect the overall energy usage is extracted. These state vectors include the current energy consumption of the equipment, historical energy consumption trends, environmental factors, etc. According to the state vector, the corresponding action space is defined. The action space includes various possible energy management measures, such as turning the equipment on or off, adjusting the energy load, etc., to form an energy management action set. The reward function is designed for the energy management action set to obtain the energy management reward model, and the Q value network is constructed according to the energy management reward model to obtain the initial Q learning model. The reward function design quantifies the energy management goal into a specific reward value by defining a reasonable reward function, such as saving energy costs, improving equipment efficiency, reducing carbon emissions, etc. The reward model can guide the intelligent agent to select the optimal energy management action in the reinforcement learning process. According to the designed reward model, the Q value network is constructed, and the initial Q learning model is used to estimate the long-term reward value that can be obtained by performing a certain action in a specific state. The initial Q learning model is sampled by experience replay to obtain a training sample batch, and the target Q value is calculated based on the training sample batch to obtain the Q value update target. Experience replay sampling stores the historical decisions and feedback of the intelligent agent and randomly extracts samples for training to avoid the problem of training instability caused by data correlation. Through the training sample batch, the target Q value is calculated, and the actual reward value and the expected future reward value are combined to obtain the update target. According to the Q value update target, the Q value network is subjected to time difference learning, and the network parameters are gradually adjusted to optimize the Q value network so that it can accurately estimate the long-term reward value of each state-action pair. The optimized Q value network can more accurately reflect the effect of different energy management decisions. According to the optimized Q value network, exploration is carried out. By balancing exploration and utilization, the intelligent agent continuously optimizes the decision-making strategy while trying different energy management actions, and finally obtains the optimal energy management decision sequence. The optimal decision sequence refers to a series of energy management actions that can obtain the maximum long-term reward under the current state. According to the optimal energy management decision sequence and the smart space predictive maintenance model, the energy allocation of the smart space is optimized to obtain the dynamic energy management solution of the smart space. Energy allocation optimization combines the output of the optimal decision sequence and predictive maintenance model to dynamically adjust the energy usage strategy of the smart space and realize intelligent energy management.

[0036] Step S105: Integrate the smart space digital twin model and the smart space dynamic energy management solution to obtain a full-cycle value evaluation system for the smart space;

[0037] Specifically, asset attributes of the digital twin model of smart space are extracted to obtain the smart space asset feature vector, and a smart space value evaluation index system is constructed based on the smart space asset feature vector and the smart space dynamic energy management solution. By analyzing various asset data in the smart space, feature vectors that can reflect asset value and performance are extracted. These feature vectors include the operating status, maintenance history, and service life of the equipment. Combined with the smart space dynamic energy management solution, a comprehensive value evaluation index system is constructed to ensure that the evaluation indicators can cover all aspects of the smart space, including energy efficiency, equipment utilization, maintenance cost, etc. A hierarchical analysis is performed on the smart space value evaluation index system to obtain an indicator weight matrix, and the historical operation data of the smart space is weighted according to the indicator weight matrix to obtain a smart space value benchmark data set. The hierarchical analysis method is used to determine the importance of each evaluation indicator. By constructing a comparison matrix and calculating the feature vector, the weight of each indicator is obtained. These weights are applied to the historical operation data of the smart space, and the data is weighted to obtain a benchmark data set that comprehensively reflects the value of the smart space. The smart space value benchmark data set is decomposed to obtain a smart space value trend model, and a smart space future value prediction sequence is generated based on the smart space value trend model. The value trend model reveals the changing trend and cyclical characteristics of the value of smart space through time series analysis of the benchmark data set. Based on this model, a future value prediction sequence is generated to predict the expected value of smart space in different time periods. The future value prediction sequence of smart space is analyzed to obtain the value distribution probability map of smart space, and the smart space risk assessment matrix is ​​calculated based on the value distribution probability map of smart space. The value distribution probability map shows the possible distribution of future value through probabilistic statistical methods and reveals the probability of occurrence of different value levels. Based on this probability map, the risk assessment matrix is ​​calculated to evaluate the risk level of smart space in different scenarios. A comprehensive evaluation of the smart space risk assessment matrix is ​​performed to obtain the full-cycle value score of the smart space, and a value mapping relationship is constructed based on the full-cycle value score of the smart space and the digital twin model of the smart space to obtain the full-cycle value assessment system of the smart space.

[0038] Step S106: Perform self-learning processing on the smart space full-cycle value assessment system to generate a target smart space management strategy.

[0039] Specifically, data aggregation is performed on the smart space full-cycle value evaluation system to obtain a smart space comprehensive score table. The correlation coefficients of the indicators in the comprehensive score table are calculated to reveal the mutual relationship and influence between the indicators, and generate an indicator correlation matrix. Hierarchical cluster analysis is performed on the indicator correlation matrix to obtain the key factor set of smart space management, and the quantitative indicators of smart space management effect are calculated based on the key factor set of smart space management. Hierarchical cluster analysis can effectively identify the key factors in smart space management. Through cluster analysis of the correlation matrix, the set of factors that have the greatest impact on the management effect is extracted. Based on these key factors, the quantitative indicators of smart space management effect are calculated, and the quantitative indicators can specifically reflect the effect and effectiveness of management measures. A smart space management trend chart is generated based on the quantitative indicators of smart space management effect. The trend chart can intuitively show the changes in management effect over time, and the abnormal points and inflection points are identified based on the smart space management trend chart to obtain a list of key events in the smart space. The management trend chart reveals the long-term trend and change law of smart space management. By identifying the abnormal points and inflection points, emergencies and important turning points in the management process can be discovered, and a list of key events can be generated. A causal analysis is performed on the list of key events in smart space to obtain the experience rule base for smart space management, and a preliminary management strategy model is constructed based on the experience rule base for smart space management. The purpose of causal analysis is to explore the causes and impacts of key events. Through the analysis of key events, valuable management experience and rules are extracted to form a management experience rule base. Based on the experience rule base, a preliminary management strategy model is constructed to provide preliminary strategic guidance for the management of smart space. The preliminary management strategy model is verified by historical data. By applying the model to historical data, its performance in different historical scenarios is verified to obtain a strategy reliability evaluation report. The reliability evaluation report reflects the effectiveness and reliability of the management strategy in practical applications. The advantages and disadvantages of the strategy model can be identified through the evaluation report. Iterative optimization is performed based on the strategy reliability evaluation report and the full-cycle value evaluation system of smart space. According to the feedback in the reliability evaluation report, the management strategy model is gradually adjusted and improved to ensure that it can show good management effects in different scenarios and obtain the target smart space management strategy.

[0040] In the embodiment of the present application, a complete digital twin model of smart space is constructed through multi-dimensional fusion processing of physical data and business data, and accurate mapping of physical space and digital space is achieved. The scenario analysis based on the digital twin model can capture and identify the dynamic changes of the smart space in real time. The bidirectional long short-term memory network is used for training to obtain a maintenance model with predictive capabilities, which can identify potential problems in advance and reduce the risk of equipment failure and maintenance costs. By optimizing the predictive maintenance model through reinforcement learning, the generation of dynamic energy management solutions is realized, energy efficiency is improved, and operating costs are reduced. The digital twin model and the dynamic energy management solution are integrated to build a full-cycle value assessment system. Through self-learning processing, the management strategy can be automatically adjusted and optimized according to historical data and real-time conditions, which improves the flexibility and adaptability of management. The multi-scale modeling method can describe the smart space from different levels and granularities. Through time series analysis and pattern matching, abnormal situations and key events can be quickly identified, which improves the problem response speed and processing efficiency.

[0041] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0042] (1) Perform CAD modeling on the smart space to obtain a two-dimensional basic model and a three-dimensional basic model;

[0043] (2) constructing a BIM model based on the two-dimensional basic model and the three-dimensional basic model;

[0044] (3) performing three-dimensional conversion and optimization on the BIM model to obtain a 3D model, and performing consistency check on the BIM model and the 3D model to obtain model correction data;

[0045] (4) deploying an IoT sensor network in the smart space to obtain a real-time non-graphic data stream, and performing data cleaning and standardization on the non-graphic data stream to obtain a normalized data set;

[0046] (5) constructing a data fusion platform according to the BIM model, the 3D model and the standardized data set to obtain a multi-source heterogeneous data management system;

[0047] (6) Performing semantic annotation and association analysis on the data in the multi-source heterogeneous data management system to obtain a smart space semantic network;

[0048] (7) Dynamically update the 3D model according to the smart space semantic network and real-time data to obtain a smart space digital twin model.

[0049] Specifically, computer-aided design (CAD) technology is used to accurately model the smart space. This step not only includes the creation of traditional two-dimensional floor plans, elevations, and sections, but also involves preliminary three-dimensional space modeling. The accuracy of the CAD model directly affects the quality of subsequent steps, so special attention should be paid to the accuracy and completeness of the details at this stage. Based on the CAD basic model, a building information model (BIM) is constructed. The BIM model not only contains geometric information, but also integrates rich non-geometric information such as building components, equipment systems, and material properties. This step realizes the leap from pure geometric description to information expression, providing a comprehensive data foundation for subsequent intelligent analysis and management. In order to further enhance the visual effect and spatial expression of the model, the BIM model is converted into a high-precision three-dimensional model. This process involves complex geometric optimization algorithms and detail improvement techniques such as texture mapping and light and shadow effect optimization to ensure that the 3D model is both accurate and intuitive. Model consistency checking is a key step to ensure the accuracy of the digital twin model. By comparing the BIM model and the 3D model, possible inconsistencies such as geometric shape deviation and missing attribute information can be identified and resolved. This step may require multiple iterations until the preset consistency standard is reached. Deploying IoT sensor networks in physical space is a bridge between the virtual and real worlds. This involves multiple aspects such as the selection of sensor types, the formulation of layout strategies, and the determination of data collection frequency. The real-time data collected includes but is not limited to multi-dimensional information such as temperature, humidity, energy consumption, and pedestrian flow. Data cleaning and standardization are key steps to ensure data quality. This process includes multiple subtasks such as outlier detection, missing value processing, and data format unification. Through complex algorithms and rules, the raw data is converted into structured and standardized data sets. The construction of a data fusion platform is the core of achieving unified management of multi-source heterogeneous data. This platform needs to design a flexible data model to support the seamless integration of BIM models, 3D models, and IoT data. The semantic annotation and association analysis steps are designed to improve the comprehensibility and usability of data. This involves complex natural language processing and knowledge graph technologies, and by establishing a semantic network of smart space, logical associations and intelligent reasoning between data are achieved. The virtual-real digital twin prototype is achieved by dynamically updating the 3D model. This requires the design of an efficient data synchronization mechanism to ensure that changes in the physical space can be reflected in the digital model in real time. This step may involve complex state estimation algorithms and data fusion technologies. By integrating machine learning algorithms, the digital twin model is endowed with intelligent analysis and prediction capabilities. This may include multiple advanced functional modules such as anomaly detection, performance optimization, and predictive maintenance. The selection and training process of the algorithm needs to consider the specific needs and data characteristics of the smart space.

[0050] Optionally, in one embodiment, the process of executing step S101 may further include the following steps:

[0051] (1) Perform three-dimensional modeling on the physical data of the smart space to obtain the three-dimensional geometric model of the smart space, and perform semantic annotation on the business data of the smart space to obtain the semantic information layer of the smart space;

[0052] (2) Based on the three-dimensional geometric model and semantic information layer, the topological relationship of the smart space is analyzed to obtain the topological relationship diagram of the smart space;

[0053] (3) Map the three-dimensional geometric model, semantic information layer, and topological relationship diagram to the graph database to obtain the knowledge graph of the smart space, and perform time series analysis on the real-time sensor data of the smart space to obtain the dynamic attribute model of the smart space;

[0054] (4) Based on the knowledge graph and dynamic attribute model, multi-scale modeling of the smart space is carried out to obtain a multi-scale digital model of the smart space;

[0055] (5) Performing data consistency check on the multi-scale digital model to obtain the data correction rules of the smart space, and performing data correction on the multi-scale digital model according to the data correction rules to obtain the corrected digital model of the smart space;

[0056] (6) Visualize and render the corrected digital model to obtain a three-dimensional visualization model of the smart space, and fuse the three-dimensional visualization model with the knowledge graph to obtain a digital twin model of the smart space.

[0057] Specifically, physical data in the smart space is collected, including information such as the structure of the building, the layout of the equipment, and the size of the space. By using technologies such as laser scanning, photogrammetry, or BIM (Building Information Modeling), these physical data are converted into a three-dimensional geometric model. The model describes each physical object in the smart space and its position and relationship in the space. For example, by using laser scanning technology, accurate point cloud data inside the building is obtained, and through point cloud data processing, a three-dimensional geometric model of the smart space is generated, which is represented as ,in Represents the coordinates of each point in three-dimensional space, It is the point set of the entire geometric model. The business data of the smart space is semantically annotated. The business data includes information such as the purpose of the equipment, working status, maintenance records, and user behavior. By associating business data with physical objects, a semantic information layer is constructed. For example, in an office building, a room (an area in the geometric model) is labeled as a "conference room", and its related business data includes information such as "frequency of use", "reservation status", and "equipment status". After semantic processing, these data form a semantic information layer. ,in represents a physical object, Represents the semantic tags and business data related to it. Based on the 3D geometric model and the semantic information layer, the topological relationship analysis of the smart space is performed to obtain the topological relationship diagram of the smart space. The topological relationship diagram describes the spatial and functional relationship between physical objects. For example, the connection relationship between two rooms, the relationship between a device and the room it is in, etc. By analyzing the points and surfaces in the 3D geometric model and the business data in the semantic information layer, a topological relationship diagram is constructed. ,in Representing physical objects and The relationship between them is represented by the graph database, and R is the set of all possible relationships. The three-dimensional geometric model, the semantic information layer, and the topological relationship graph are mapped to the graph database to obtain the knowledge graph of the smart space. The knowledge graph uses the graph database technology to store and manage the points in the three-dimensional geometric model, the labels in the semantic information layer, and the relationships in the topological relationship graph. The knowledge graph is represented as ,in is a set of nodes, representing physical objects and semantic labels, is an edge set, representing the relationship between objects. is an attribute set, which represents the attribute information of nodes and edges. At the same time, the real-time sensor data of the smart space is analyzed in time series to obtain the dynamic attribute model of the smart space. The sensor data includes temperature, humidity, light energy consumption, etc. Through time series analysis, the changing trends and laws of these data are understood, which are expressed as ,in It's the time point. is the sensor at time According to the knowledge graph and dynamic attribute model, the smart space is modeled at multiple scales to obtain a multi-scale digital model of the smart space. Multi-scale modeling integrates different scale information of the smart space (such as rooms, floors, and buildings) to construct a hierarchical digital model, which is expressed as ,in Indicates The layer model, is the number of layers of the model. The multi-scale digital model is checked for data consistency to obtain the data correction rules of the smart space. According to the data correction rules, the multi-scale digital model is corrected to obtain the corrected digital model of the smart space. The data consistency check ensures the consistency and accuracy of data from different sources and scales by comparing and verifying them. The data correction rule is expressed as ,in is the correction function, is the correction condition. According to these rules, the multi-scale digital model is corrected to obtain the corrected model. The corrected digital model is visualized and rendered to obtain a three-dimensional visualization model of the smart space, and the three-dimensional visualization model and the knowledge graph are fused to obtain a digital twin model of the smart space. Visual rendering converts the corrected digital model into an intuitive three-dimensional image through graphics processing technology. By fusing the three-dimensional visualization model and the knowledge graph, a comprehensive digital twin model is formed, which contains the physical and business information of the smart space and can also reflect its dynamic changes in real time.

[0058] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0059] (1) The digital twin model of the smart space is divided into functional areas to obtain a functional area map. Based on the functional area map, the crowd density of the smart space is analyzed to obtain a crowd heat map.

[0060] (2) Perform time series decomposition on the crowd flow heat map to obtain a crowd flow trend model, and conduct correlation analysis on the use of smart space equipment based on the crowd flow trend model to obtain a device use correlation matrix;

[0061] (3) Extract features from the device usage association matrix to obtain the device usage feature vector, and perform cluster analysis on the smart space based on the device usage feature vector to obtain a set of smart space scene types;

[0062] (4) Perform semantic annotation on the smart space scene type set to obtain a smart space semantic scene library, and perform pattern matching on the current state of the smart space based on the smart space semantic scene library to obtain real-time scene recognition results;

[0063] (5) Perform time series analysis on the real-time scene recognition results to obtain the smart space scene conversion probability matrix. Based on the scene conversion probability matrix and the smart space semantic scene library, perform multimodal data fusion on the smart space to obtain the smart space scene integrated data.

[0064] Specifically, the digital twin model of the smart space is divided into functional areas, and each area in the smart space is divided according to different uses and functions. For example, in a large shopping mall, it can be divided into different functional areas such as shopping area, dining area, entertainment area, parking area, etc. Through the digital twin model, the boundaries and scope of each functional area are determined to obtain a functional area map. The functional area map is represented as ,in Indicates functional areas, is the total number of functional areas. According to the functional area map, the crowd density analysis of the smart space is carried out. By monitoring and counting the flow of people in each functional area, a crowd heat map is generated. The crowd heat map can intuitively show the density and flow of people in each area. The crowd flow data can be obtained in real time through the cameras, sensors and other equipment installed in the smart space, expressed as a function of time and location. ,in Indicates time, and Indicates the location coordinates. By statistically analyzing the crowd flow data, we can get the crowd flow heat map. , where the value of each point represents the density of people flow at that location within a specific time. The people flow heat map is decomposed into time series to obtain the people flow trend model. Time series decomposition is to decompose the people flow data into trend, seasonal and random components to better understand and predict the changes in people flow. The people flow trend model is expressed as ,in is the trend component, It is a seasonal ingredient. is a random component, , and is the corresponding weight coefficient. According to the pedestrian flow trend model, the usage of smart space equipment is analyzed and the device usage association matrix is ​​obtained. By counting the correlation between device usage data and pedestrian flow data, the relationship between device usage and pedestrian flow changes is found. The device usage data can be expressed as ,in Represents the device identification. The device usage association matrix is ​​represented as ,in Indicates The device in The usage of each functional area under the change of passenger flow, is the total number of devices. The device usage association matrix is ​​subjected to feature extraction to obtain the device usage feature vector. By performing dimension reduction and data mining on the association matrix, feature vectors that can reflect the device usage rules are extracted. The device usage feature vector is expressed as ,in Indicates Device Features, is the total number of features. Based on the device usage feature vector, cluster analysis is performed on the smart space to obtain a set of smart space scene types. Cluster analysis is to group devices and areas with similar features to form different scene types. The scene type set is represented as ,in Indicates scene types, is the total number of scene types. The set of smart space scene types is semantically annotated to obtain a smart space semantic scene library. Semantic annotation forms a semantic library by adding semantic information to each scene type. For example, the scene types of a shopping area can be annotated as "peak hours", "normal hours", etc. The semantic scene library is represented as ,in Indicates The semantic label of the scene type. According to the smart space semantic scene library, the current state of the smart space is pattern matched to obtain the real-time scene recognition result. The current scene type is identified by comparing the current device usage feature vector with the features in the semantic scene library. The real-time scene recognition result is expressed as ,in is the scene type that best matches the current feature. The real-time scene recognition results are analyzed in time series to obtain the probability matrix of scene transition in the smart space. Time series analysis calculates the transition probability between different scenes by counting the historical data of scene transition. The scene transition probability matrix is ​​expressed as ,in Indicates the scene type Switch to scene type According to the scene transition probability matrix and the smart space semantic scene library, multimodal data fusion is performed on the smart space to obtain the smart space scene integration data. Multimodal data fusion forms a comprehensive scene data set by combining data from different sources and types. The scene integration data is represented as ,in is the real-time scene recognition result, is the scene transition probability matrix, It is a semantic scene library. is the total number of time points.

[0065] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0066] (1) Perform time window segmentation on the integrated data of smart space scenes to obtain a time series data sample set, and perform data standardization on the time series data sample set to obtain a standardized time series data set;

[0067] (2) Construct a bidirectional long short-term memory network structure based on the standardized time series data set to obtain the initial bidirectional LSTM model;

[0068] (3) Perform forward propagation calculation on the initial bidirectional LSTM model to obtain the model output result, and calculate the loss function value based on the model output result and the actual label to obtain the model error;

[0069] (4) Back-propagate the model error to obtain the network parameter gradient, and optimize and update the model parameters according to the network parameter gradient to obtain the updated bidirectional LSTM model;

[0070] (5) Cross-validate the updated bidirectional LSTM model to obtain the model performance evaluation index, and adjust the network structure according to the model performance evaluation index to obtain the optimized bidirectional LSTM model;

[0071] (6) Perform predictive maintenance recognition training on the optimized bidirectional LSTM model to obtain the smart space predictive maintenance model.

[0072] Specifically, the scene integration data of the smart space is collected, which includes scene types, crowd density, equipment usage and other information at different time points. Through time window segmentation, these data are divided into multiple time windows, each of which contains all the data within a fixed time period. Time window segmentation can be expressed as ,in Represents a time window The data sample set in is the total number of time windows. The time series data sample set is normalized. By subtracting the mean and dividing by the standard deviation, the data is converted into a standard normal distribution with a mean of 0 and a standard deviation of 1 to obtain a standardized time series data set. A bidirectional long short-term memory (LSTM) network structure is constructed based on the standardized time series data set to obtain an initial bidirectional LSTM model. Bidirectional LSTM is a special recurrent neural network that can use both forward and backward time information and is suitable for processing time series data. The bidirectional LSTM network structure consists of two parts: the forward LSTM and the backward LSTM. The forward LSTM processes the forward information of the time series, and the backward LSTM processes the reverse information of the time series. The initial bidirectional LSTM model can be expressed as ,in It's time point The hidden state of the initial bidirectional LSTM model is forward propagated and the output of the model is obtained by calculating the output of neurons layer by layer. The forward propagation calculation formula is:

[0073] ;

[0074] in is the model at time point The output result is, is the weight matrix from hidden state to output, is the bias vector of the output layer. According to the model output results and the actual labels, the loss function value is calculated. The loss function usually uses mean square error or cross entropy, which is expressed as:

[0075] L ;

[0076] Where L is the loss function value, is the total number of samples, is the predicted value of the model, is the actual label. The model error is back-propagated, and the gradient of the loss function to the model parameters is calculated by the chain rule. The back-propagation calculation formula is:

[0077] ;

[0078] in is the loss function over the weight matrix Based on these gradients, the model parameters are optimized and updated, and the gradient descent method or its variants (such as the Adam optimization algorithm) are used to gradually adjust the model parameters to minimize the loss function. The updated bidirectional LSTM model is expressed as:

[0079] ;

[0080] in is the learning rate. The updated bidirectional LSTM model is cross-validated. By dividing the data set into a training set and a validation set, the model is repeatedly trained and validated to evaluate the generalization ability of the model. Cross-validation can effectively avoid the overfitting problem of the model and obtain the performance evaluation indicators of the model, including accuracy, precision, recall, F1 score, etc. According to the model performance evaluation indicators, the network structure and hyperparameters are adjusted, such as the number of LSTM layers, the number of hidden units, the learning rate, etc., to obtain the optimized bidirectional LSTM model. The optimized bidirectional LSTM model is trained for predictive maintenance recognition. By introducing more training data and optimization algorithms, the model can accurately identify and predict potential maintenance needs in the smart space. Predictive maintenance recognition training gradually improves the accuracy and reliability of the model through long-term model training and parameter optimization. The predictive maintenance model of the smart space is obtained, which can monitor various data of the smart space in real time, detect possible problems in advance, and provide maintenance suggestions.

[0081] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0082] (1) The state space modeling of the smart space predictive maintenance model is carried out to obtain the smart space energy state vector, and the action space is defined according to the smart space energy state vector to obtain the energy management action set;

[0083] (2) Design a reward function for the energy management action set to obtain an energy management reward model, and construct a Q-value network based on the energy management reward model to obtain an initial Q-learning model;

[0084] (3) Perform experience replay sampling on the initial Q learning model to obtain a training sample batch, and calculate the target Q value based on the training sample batch to obtain the Q value update target;

[0085] (4) Performing temporal difference learning on the Q-value update target to obtain an optimized Q-value network, and exploring based on the optimized Q-value network to obtain the optimal energy management decision sequence;

[0086] (5) Based on the optimal energy management decision sequence and the predictive maintenance model of the smart space, the energy allocation of the smart space is optimized to obtain a dynamic energy management solution for the smart space.

[0087] Specifically, first, various energy consumption and supply data of smart space are modeled. State space modeling extracts feature vectors that can fully reflect the energy state of smart space by analyzing the operating status, energy consumption level, environmental parameters, etc. of various devices and systems in smart space. The energy state vector of smart space can be expressed as ,in Indicates time point The total energy consumption, Indicates power consumption, Indicates the ambient temperature, Represents the state of the cooling system. Each element represents an energy state feature of the smart space at a specific time point. According to the energy state vector of the smart space, the action space is defined to obtain the energy management action set. The action space refers to the set of energy management measures that the system can take in each state. Energy management actions can include adjusting the operating state of the equipment, adjusting the energy consumption strategy, changing the environmental control settings, etc. The action space is represented by A , where each action Represents a specific energy management operation, such as lowering the air conditioning temperature, reducing the lighting intensity, etc. The reward function is designed for the energy management action set to obtain the energy management reward model. The reward function refers to the reward value obtained by the system after taking a certain action, which is used to measure the effect of the action. The design of the reward function needs to take into account multiple factors such as energy efficiency, user comfort, and equipment life. The reward function can be expressed as ,in Yes Status Take action Rewards received, is the change in energy consumption, is the change in user comfort, is the change in equipment life, , , is the corresponding weight coefficient. According to the energy management reward model, a Q value network is constructed to obtain the initial Q learning model. The Q value network is a neural network structure used to estimate the long-term reward value of taking an action in a specific state. The Q learning model optimizes the decision-making strategy by continuously updating the Q value. The initial Q learning model can be expressed as:

[0088] ;

[0089] in Yes Status Next action The Q value, are the feature vectors of states and actions, is the weight matrix, is the bias vector. The initial Q-learning model is sampled by experience replay to obtain a training sample batch. Experience replay sampling randomly extracts samples for training by storing and reusing the historical decisions and feedback of the intelligent agent to avoid the problem of training instability caused by data correlation. Through experience replay, a training sample batch is obtained. ,in is the current state, is the action taken, It is the reward obtained. is the next state. Calculate the target Q value based on the training sample batch and get the Q value to update the target. The target Q value calculation formula is:

[0090] ;

[0091] in is the target Q value, is the discount factor, is the maximum Q value of the next state. By calculating the target Q value, the model parameter update is guided. Temporal difference learning is performed on the Q value update target to obtain the optimized Q value network. Temporal difference learning updates the parameters of the Q value network by comparing the difference between the predicted Q value and the target Q value. The update formula is:

[0092] ;

[0093] in is the learning rate, is the gradient of the loss function with respect to the weight matrix. Based on the optimized Q-value network, the optimal energy management decision sequence is obtained. The agent gradually optimizes the decision strategy by exploring and selecting actions that maximize the Q-value in the state space. The optimal decision sequence is expressed as ,in It's time point According to the optimal energy management decision sequence and the smart space predictive maintenance model, the energy allocation of the smart space is optimized to obtain the dynamic energy management plan of the smart space. Energy allocation optimization refers to dynamically adjusting the operating status of each system and equipment in the smart space under the guidance of the optimal decision sequence to achieve the best energy utilization effect. The final energy management plan can be expressed as , where each element represents the state The best action to take .

[0094] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0095] (1) Extract asset attributes from the smart space digital twin model to obtain the smart space asset feature vector, and build a smart space value assessment index system based on the smart space asset feature vector and the smart space dynamic energy management solution;

[0096] (2) Conduct hierarchical analysis on the smart space value assessment indicator system to obtain the indicator weight matrix, and perform weighted processing on the historical operation data of the smart space according to the indicator weight matrix to obtain the smart space value benchmark data set;

[0097] (3) Decompose the smart space value benchmark data set to obtain the smart space value trend model, and generate the smart space future value prediction sequence based on the smart space value trend model;

[0098] (4) Analyze the future value prediction sequence of smart space to obtain the smart space value distribution probability map, and calculate the smart space risk assessment matrix based on the smart space value distribution probability map;

[0099] (5) Conduct a comprehensive evaluation of the smart space risk assessment matrix to obtain the full-cycle value score of the smart space, and construct a value mapping relationship based on the full-cycle value score of the smart space and the smart space digital twin model to obtain the full-cycle value assessment system of the smart space.

[0100] Specifically, various types of assets in the smart space are analyzed and extracted. Asset attribute extraction extracts important parameters that can reflect the characteristics of these assets by analyzing various types of equipment, systems, and infrastructure in the smart space. The asset feature vector of the smart space can be expressed as ,in Indicates The characteristics of each asset, such as the service life of the equipment, maintenance cost, energy efficiency, etc. These characteristics can fully reflect the status and performance of each asset. According to the smart space asset feature vector and the smart space dynamic energy management plan, a smart space value evaluation index system is constructed. The value evaluation index system defines a set of indicators for evaluating the overall value of the smart space by comprehensively considering various characteristics of assets and energy management plans. The value evaluation indicators of smart space can include energy efficiency, maintenance cost, user comfort, equipment reliability, etc. These indicators can be expressed as ,in Indicates Value evaluation indicators. For example, the energy efficiency index can measure the energy consumption of each device in the smart space, the maintenance cost index can measure the maintenance cost of the equipment, and the user comfort index can measure the environmental quality of the space. Perform a hierarchical analysis on the smart space value evaluation index system to obtain an indicator weight matrix. The hierarchical analysis method is a multi-index decision-making method that constructs a hierarchical model, analyzes the relative importance of each indicator layer by layer, and determines the weight of each indicator. The result of the hierarchical analysis method is a weight matrix. ,in Indicates The weight of the value assessment indicator is calculated. According to the indicator weight matrix, the historical operation data of the smart space is weighted to obtain the smart space value benchmark data set. The weighted processing formula is:

[0101] ;

[0102] in is the value benchmark dataset, It is The historical operation data corresponding to the value evaluation indicators are decomposed into the smart space value benchmark data set to obtain the smart space value trend model. The value trend model reveals the changing trends and laws of the smart space value through time series analysis of historical operation data. The value trend model can be expressed as:

[0103] ;

[0104] in It's time point The value trend of is the coefficient of the model. According to the value trend model, the future value prediction sequence of the smart space is generated to predict the value at each time point in the future. The future value prediction sequence of the smart space is analyzed to obtain the value distribution probability map of the smart space. The value distribution probability map shows the probability of occurrence of different value levels by performing probability statistics on the future value prediction sequence. The value distribution probability map can be expressed as ,in Indicates the value level The probability of occurrence of . According to the value distribution probability diagram, the smart space risk assessment matrix is ​​calculated. The risk assessment matrix quantifies the risks faced by the smart space by analyzing the risks at different value levels. The risk assessment matrix can be expressed as ,in Indicates the value level The smart space risk assessment matrix is ​​comprehensively evaluated to obtain the full-cycle value score of the smart space. The full-cycle value score evaluates the overall value of the smart space throughout its life cycle by comprehensively considering the risks and benefits at different time points and value levels. The full-cycle value score can be expressed as:

[0105] ;

[0106] in is the full-cycle value score, Is the value level The probability of occurrence, Is the value level According to the full-cycle value score of smart space and the digital twin model of smart space, a value mapping relationship is constructed to obtain the full-cycle value evaluation system of smart space. The value mapping relationship forms a complete value evaluation system by associating the full-cycle value score with various assets and management strategies in the digital twin model.

[0107] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0108] (1) Aggregate data on the smart space full-cycle value evaluation system to obtain a comprehensive score sheet for smart space, and calculate the correlation coefficients of various indicators based on the comprehensive score sheet for smart space to obtain an indicator correlation matrix;

[0109] (2) Perform hierarchical cluster analysis on the indicator correlation matrix to obtain the key factor set of smart space management, and calculate the quantitative indicators of smart space management effect based on the key factor set of smart space management;

[0110] (3) Generate a smart space management trend chart based on the quantitative indicators of smart space management effects, identify abnormal points and inflection points based on the smart space management trend chart, and obtain a list of key events for the smart space;

[0111] (4) Conduct causal analysis on the list of key events of smart space to obtain the smart space management experience rule base, and build a preliminary management strategy model based on the smart space management experience rule base;

[0112] (5) Verify the preliminary management strategy model with historical data to obtain a strategy reliability assessment report, and iterate and optimize it based on the strategy reliability assessment report and the smart space full-cycle value assessment system to obtain the target smart space management strategy.

[0113] Specifically, various relevant data are extracted from the smart space full-cycle value evaluation system, including indicators of energy consumption, equipment maintenance, user satisfaction, etc., and data from different sources and different time periods are integrated to form a comprehensive comprehensive score sheet. The comprehensive score sheet is expressed as ,in Indicates The comprehensive score at each time point, is the total number of time points. Each comprehensive score is obtained by weighted calculation of each indicator. According to the comprehensive score table, the correlation coefficient of each indicator is calculated to obtain the indicator correlation matrix. The correlation coefficient is used to measure the degree of correlation between two variables. For example, the Pearson correlation coefficient is expressed as:

[0114] ;

[0115] in, is a variable and The correlation coefficient between yes and The covariance of and They are and By calculating the correlation coefficients between the various indicators, we can get the indicator correlation matrix ,in Indicators and indicators The correlation coefficient between them is calculated. Hierarchical clustering analysis is performed on the indicator correlation matrix to obtain the key factor set of smart space management. Hierarchical clustering analysis clusters similar samples together by calculating the similarity between samples. Hierarchical clustering can be divided into agglomerative hierarchical clustering and divisive hierarchical clustering. Commonly used agglomerative hierarchical clustering methods include single connection method, full connection method and average connection method. The clustering result can be expressed as a tree diagram, in which each node represents an indicator or a group of indicators, and finally the key factor set of smart space management is obtained. ,in Indicates Key management factors is the total number of key factors. Based on the key factor set of smart space management, the quantitative index of smart space management effect is calculated. The quantitative index is used to measure the effect of management measures. The quantitative index is expressed as ,in Indicates A quantitative indicator, is the total number of quantitative indicators. For example, a key factor set may include the average energy consumption of equipment, maintenance frequency, and number of user complaints. A smart space management trend chart is generated based on the quantitative indicators of smart space management effect. The trend chart shows the changing trend of management effect through time series analysis of historical data. The trend chart can help identify abnormal points and inflection points in the data and find key events in the management process. The trend chart is represented as ,in Indicates quantitative indicators at a point in time By analyzing the trend chart, we can identify abnormal points and inflection points and obtain a list of key events in the smart space. ,in Indicates A key event, is the total number of key events. Perform causal analysis on the key event list of smart space to obtain the smart space management experience rule base. Causal analysis extracts useful management experience and rules by analyzing the causes and results of events. For example, by comparing and analyzing the data before and after a key event, the cause of the event can be found. The experience rule base is represented as ,in Indicates Management rules of thumb, is the total number of rules. Based on the smart space management experience rule base, a preliminary management strategy model is constructed. The management strategy model forms a systematic management plan through the comprehensive application of experience rules to guide the operation and maintenance of smart space. The preliminary management strategy model is expressed as ,in Indicates management strategies, is the total number of management strategies. The preliminary management strategy model is validated with historical data to obtain a strategy reliability evaluation report. Historical data validation evaluates the effectiveness and reliability of the management strategy model by applying it to historical data. The strategy reliability evaluation report is expressed as ,in Indicates The evaluation results of each management strategy, including effect evaluation, advantages and disadvantages analysis, etc. Iterative optimization is performed based on the strategy reliability evaluation report and the smart space full-cycle value evaluation system to obtain the target smart space management strategy. Iterative optimization gradually improves the effect and reliability of the management strategy by continuously adjusting and improving it. The target smart space management strategy is expressed as ,in Indicates The final management strategy, is the total number of final strategies.

[0116] The management method of the smart space system in the embodiment of the present application is described above. The management device of the smart space system in the embodiment of the present application is described below. Figure 2 In one embodiment of the present application, a management device of the smart space system includes:

[0117] The fusion module 201 is used to perform multi-dimensional fusion processing on the physical data and business data of the smart space to build a digital twin model of the smart space;

[0118] The analysis module 202 is used to perform scenario analysis on the smart space digital twin model to obtain smart space scenario integrated data;

[0119] The training module 203 is used to perform bidirectional long short-term memory network training on the smart space scene integrated data to obtain a smart space predictive maintenance model;

[0120] The optimization module 204 is used to perform reinforcement learning optimization on the smart space predictive maintenance model to obtain a smart space dynamic energy management solution;

[0121] The integration module 205 is used to integrate the smart space digital twin model and the smart space dynamic energy management solution to obtain a smart space full-cycle value evaluation system;

[0122] The generation module 206 is used to perform self-learning processing on the smart space full-cycle value evaluation system and generate a target smart space management strategy.

[0123] Through the collaboration of the above components and the multi-dimensional fusion processing of physical data and business data, a complete digital twin model of smart space is constructed, and the precise mapping of physical space and digital space is realized. The scenario analysis based on the digital twin model can capture and identify the dynamic changes of the smart space in real time. The bidirectional long short-term memory network is used for training to obtain a maintenance model with predictive capabilities, which can identify potential problems in advance and reduce the risk of equipment failure and maintenance costs. The predictive maintenance model is optimized through reinforcement learning, and the generation of dynamic energy management solutions is realized, which improves energy efficiency and reduces operating costs. The digital twin model and dynamic energy management solution are integrated to build a full-cycle value evaluation system. Through self-learning processing, the management strategy can be automatically adjusted and optimized according to historical data and real-time conditions, which improves the flexibility and adaptability of management. The multi-scale modeling method can describe the smart space from different levels and granularities. Through time series analysis and pattern matching, abnormal situations and key events can be quickly identified, which improves the problem response speed and processing efficiency.

[0124] The present application also provides a management device for a smart space system, the management device for the smart space system comprising a memory and a processor, the memory storing computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the management method for the smart space system in the above-mentioned embodiments.

[0125] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the management method of the smart space system.

[0126] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0127] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0128] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A management method for a smart space system, characterized in that: The method comprises: Perform multi-dimensional fusion processing on the physical data and business data of the smart space to build a digital twin model of the smart space; A scenario analysis is performed on the digital twin model of the smart space to obtain integrated scenario data of the smart space; specifically comprising: functionally zoning the digital twin model of the smart space to obtain a functional area map, and performing crowd density analysis on the smart space according to the functional area map to obtain a crowd heat map; performing time series decomposition on the crowd heat map to obtain a crowd trend model, and performing association analysis on the use of smart space equipment according to the crowd trend model to obtain an equipment use association matrix; performing feature extraction on the equipment use association matrix to obtain a device use feature vector, and performing cluster analysis on the smart space according to the device use feature vector to obtain a set of smart space scene types; performing semantic annotation on the set of smart space scene types to obtain a smart space semantic scene library, and performing pattern matching on the current state of the smart space according to the smart space semantic scene library to obtain a real-time scene recognition result; performing time series analysis on the real-time scene recognition result to obtain a smart space scene conversion probability matrix, and performing multimodal data fusion on the smart space according to the scene conversion probability matrix and the smart space semantic scene library to obtain integrated scenario data of the smart space; The smart space scene integrated data is trained with a bidirectional long short-term memory network to obtain a smart space predictive maintenance model; specifically comprising: performing time window segmentation on the smart space scene integrated data to obtain a time series data sample set, and performing data standardization processing on the time series data sample set to obtain a standardized time series data set; constructing a bidirectional long short-term memory network structure according to the standardized time series data set to obtain an initial bidirectional LSTM model; performing forward propagation calculation on the initial bidirectional LSTM model to obtain a model output result, and calculating a loss function value according to the model output result and the actual label to obtain a model error; performing back propagation on the model error to obtain a network parameter gradient, and optimizing and updating the model parameters according to the network parameter gradient to obtain an updated bidirectional LSTM model; cross-validating the updated bidirectional LSTM model to obtain a model performance evaluation index, and adjusting the network structure according to the model performance evaluation index to obtain an optimized bidirectional LSTM model; performing predictive maintenance recognition training on the optimized bidirectional LSTM model to obtain a smart space predictive maintenance model; Performing reinforcement learning optimization on the smart space predictive maintenance model to obtain a smart space dynamic energy management solution; Integrate the smart space digital twin model and the smart space dynamic energy management solution to obtain a full-cycle value evaluation system for the smart space; The smart space full-cycle value assessment system is self-learned to generate a target smart space management strategy.

2. The management method of the smart space system according to claim 1, characterized in that: The multi-dimensional fusion processing of the physical data and business data of the smart space to build a digital twin model of the smart space includes: Carry out CAD modeling of the smart space to obtain a two-dimensional basic model and a three-dimensional basic model; Constructing a BIM model according to the two-dimensional basic model and the three-dimensional basic model; Performing three-dimensional conversion and optimization on the BIM model to obtain a 3D model, and performing consistency check on the BIM model and the 3D model to obtain model correction data; Deploy an IoT sensor network in the smart space to obtain a real-time non-graphic data stream, and perform data cleaning and standardization on the non-graphic data stream to obtain a normalized data set; Building a data fusion platform according to the BIM model, the 3D model and the normalized data set to obtain a multi-source heterogeneous data management system; Performing semantic annotation and association analysis on the data in the multi-source heterogeneous data management system to obtain a smart space semantic network; The 3D model is dynamically updated according to the smart space semantic network and real-time data to obtain a smart space digital twin model.

3. The management method of the smart space system according to claim 1, characterized in that: The smart space predictive maintenance model is subjected to reinforcement learning optimization to obtain a smart space dynamic energy management solution, including: Performing state space modeling on the smart space predictive maintenance model to obtain a smart space energy state vector, and defining an action space according to the smart space energy state vector to obtain an energy management action set; Designing a reward function for the energy management action set to obtain an energy management reward model, and constructing a Q value network according to the energy management reward model to obtain an initial Q learning model; Performing experience replay sampling on the initial Q learning model to obtain a training sample batch, and calculating a target Q value based on the training sample batch to obtain a Q value update target; Performing temporal difference learning on the Q-value update target to obtain an optimized Q-value network, and performing exploration based on the optimized Q-value network to obtain an optimal energy management decision sequence; According to the optimal energy management decision sequence and the smart space predictive maintenance model, the energy distribution of the smart space is optimized to obtain a dynamic energy management solution for the smart space.

4. The management method of the smart space system according to claim 1, characterized in that: The smart space digital twin model and the smart space dynamic energy management solution are integrated to obtain a smart space full-cycle value evaluation system, including: Extract asset attributes from the smart space digital twin model to obtain a smart space asset feature vector, and construct a smart space value assessment index system based on the smart space asset feature vector and the smart space dynamic energy management solution; Performing hierarchical analysis on the smart space value assessment indicator system to obtain an indicator weight matrix, and performing weighted processing on the smart space historical operation data according to the indicator weight matrix to obtain a smart space value benchmark data set; Decomposing the smart space value benchmark data set to obtain a smart space value trend model, and generating a smart space future value prediction sequence according to the smart space value trend model; Analyze the future value prediction sequence of the smart space to obtain a smart space value distribution probability map, and calculate the smart space risk assessment matrix based on the smart space value distribution probability map; A comprehensive evaluation is performed on the smart space risk assessment matrix to obtain a full-cycle value score of the smart space, and a value mapping relationship is constructed based on the full-cycle value score of the smart space and the smart space digital twin model to obtain a full-cycle value assessment system for the smart space.

5. The management method of the smart space system according to claim 1, characterized in that: The self-learning process of the smart space full-cycle value assessment system to generate a target smart space management strategy includes: Aggregate data on the smart space full-cycle value evaluation system to obtain a smart space comprehensive scoring table, and calculate the correlation coefficient of each indicator based on the smart space comprehensive scoring table to obtain an indicator correlation matrix; Performing hierarchical cluster analysis on the indicator correlation matrix to obtain a set of key factors for smart space management, and calculating quantitative indicators of smart space management effects based on the set of key factors for smart space management; Generate a smart space management trend chart according to the smart space management effect quantitative indicators, and identify abnormal points and inflection points according to the smart space management trend chart to obtain a list of key events of the smart space; Performing causal analysis on the list of key events of the smart space to obtain a smart space management experience rule base, and constructing a preliminary management strategy model based on the smart space management experience rule base; The preliminary management strategy model is verified by historical data to obtain a strategy reliability assessment report, and iterative optimization is performed based on the strategy reliability assessment report and the smart space full-cycle value assessment system to obtain the target smart space management strategy.

6. A management device for a smart space system, characterized in that: The device is used to execute the management method of the smart space system according to any one of claims 1 to 5, comprising: The fusion module is used to perform multi-dimensional fusion processing on the physical data and business data of the smart space and build a digital twin model of the smart space; The analysis module is used to perform scenario analysis on the digital twin model of the smart space to obtain integrated scene data of the smart space. Specifically, the module includes: functionally partitioning the digital twin model of the smart space to obtain a functional area map, and performing crowd density analysis on the smart space according to the functional area map to obtain a crowd heat map; performing time series decomposition on the crowd heat map to obtain a crowd trend model, and performing association analysis on the use of smart space equipment according to the crowd trend model to obtain a device use association matrix; performing feature extraction on the equipment use association matrix to obtain a device use feature vector, and performing cluster analysis on the smart space according to the device use feature vector to obtain a set of smart space scene types; performing semantic annotation on the set of smart space scene types to obtain a smart space semantic scene library, and performing pattern matching on the current state of the smart space according to the smart space semantic scene library to obtain a real-time scene recognition result; performing time series analysis on the real-time scene recognition result to obtain a smart space scene conversion probability matrix, and performing multimodal data fusion on the smart space according to the scene conversion probability matrix and the smart space semantic scene library to obtain integrated scene data of the smart space; A training module is used to perform bidirectional long short-term memory network training on the smart space scene integrated data to obtain a smart space predictive maintenance model; specifically comprising: performing time window segmentation on the smart space scene integrated data to obtain a time series data sample set, and performing data standardization processing on the time series data sample set to obtain a standardized time series data set; constructing a bidirectional long short-term memory network structure according to the standardized time series data set to obtain an initial bidirectional LSTM model; performing forward propagation calculation on the initial bidirectional LSTM model to obtain a model output result, and calculating a loss function value according to the model output result and the actual label to obtain a model error; performing back propagation on the model error to obtain a network parameter gradient, and optimizing and updating the model parameters according to the network parameter gradient to obtain an updated bidirectional LSTM model; cross-validating the updated bidirectional LSTM model to obtain a model performance evaluation index, and adjusting the network structure according to the model performance evaluation index to obtain an optimized bidirectional LSTM model; performing predictive maintenance recognition training on the optimized bidirectional LSTM model to obtain a smart space predictive maintenance model; An optimization module, used to perform reinforcement learning optimization on the smart space predictive maintenance model to obtain a dynamic energy management solution for the smart space; An integration module, used to integrate the smart space digital twin model and the smart space dynamic energy management solution to obtain a full-cycle value evaluation system for the smart space; A generation module is used to perform self-learning processing on the full-cycle value evaluation system of the smart space and generate a target smart space management strategy.

7. A management device for a smart space system, characterized in that: The management device of the smart space system includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instruction in the memory so that the management device of the smart space system executes the management method of the smart space system as described in any one of claims 1-5.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the management method of the smart space system as described in any one of claims 1 to 5 is implemented.

Images