AI-Driven Online Intelligent Internet of Things Monitoring and Evaluation Method, System, Device, and Medium for Building Complex Services

Abstract

The present invention relates to an AI-driven online intelligent Internet of Things monitoring and evaluation method, system, device, and medium for building complexes. The method includes: collecting and preprocessing environmental parameters, device status, and user data through a sensor network and a user interface, and using a dynamic Gaussian mixture model for clustering analysis to identify and divide multiple clusters or groups; analyzing and processing the clusters or groups through a deep learning network to obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports, comprehensively evaluating the status of the building complex through fuzzy composition operations and generating control strategies; constructing an online sharing platform, and using transfer learning to transfer the control strategies to other building complexes to achieve the sharing of models, strategies, and experiences. The present invention solves many problems existing in traditional Internet of Things systems, realizes real-time data analysis, intelligent decision-making, and system integration, and significantly improves the management efficiency and user experience of building complexes.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent Internet of Things for building complexes, and particularly to an AI-driven online intelligent Internet of Things monitoring and evaluation method, system, device, and medium for building complex services. Background Art

[0002] With the continuous progress of Internet of Things (IoT) technology, more and more devices are connected to the Internet, enabling real-time data collection and exchange. These devices are widely used in various fields such as home automation, industrial monitoring, and environmental monitoring. However, traditional IoT systems mainly focus on data collection and transmission functions and do not conduct in-depth analysis and intelligent control of the collected data, which to a certain extent limits the potential and application value of IoT systems.

[0003] Existing IoT solutions usually perform operations based on preset rules, which are relatively rigid and difficult to flexibly respond to dynamic environmental changes or meet users' personalized needs. Although some systems have tried to adopt basic data analysis techniques, these analysis processes are often carried out offline, unable to achieve real-time response to environmental changes and not reaching the level of true intelligent control.

[0004] The development of artificial intelligence (AI) technology has brought new opportunities to IoT systems. By integrating advanced AI algorithms, IoT systems can not only perform real-time analysis of data but also make intelligent decisions and controls based on the analysis results. However, the current relevant solutions on the market still have significant deficiencies in how to effectively integrate IoT monitoring, AI algorithm models, AI control, and comprehensive evaluation. These systems usually fail to organically integrate these components into a coordinated whole system and instead process each part in isolation, resulting in problems such as low system efficiency and long response time.

[0005] In addition, there are several key issues with existing technologies in real-time data monitoring and processing. First, many systems cannot achieve real-time data processing and response, especially when processing large amounts of data or when network conditions are not ideal, there is often a significant delay in data processing, which limits the application effect of the system in emergency or critical situations. Second, the decision-making process of current systems mostly relies on predefined rules, lacks adaptive capabilities, and cannot automatically adjust according to environmental changes or user behavior patterns, so it performs poorly when dealing with unforeseen or non-standard situations. Furthermore, the poor integration between IoT devices, AI algorithms, and control systems leads to inefficient transmission of data flows and control commands between system components. This segmented architecture has a negative impact on the overall performance and reliability of the system. Finally, existing technologies are usually under-optimized in terms of energy consumption and operating costs, resulting in inefficient system operation, which in turn increases unnecessary energy waste and maintenance costs. Summary of the invention

[0006] 1. Technical issues to be resolved

[0007] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides an AI-driven online intelligent Internet of Things monitoring and evaluation method, system, equipment and medium for building complex services, which solves the technical problems that traditional Internet of Things systems lack real-time data analysis, intelligent decision-making and system integration, resulting in low efficiency, slow response, and inability to meet personalized needs.

[0008] (II) Technical solution

[0009] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0010] In a first aspect, an embodiment of the present invention provides an AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services, including:

[0011] Collect and pre-process environmental parameters, equipment status, and user data through a dynamically configurable sensor network and user interface pre-deployed in the building complex;

[0012] Performing cluster analysis on at least one of the preprocessed environmental parameters, device status, and user data using a dynamic Gaussian mixture model to identify and divide multiple clusters or groups each representing a specific comfort zone;

[0013] Through deep learning network analysis and processing of clusters or groups, we can obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports;

[0014] Based on the environmental data analysis results, model prediction outputs, and user behavior analysis reports, obtain a comprehensive status assessment of the building complex services through fuzzy synthesis operations, and generate at least one control strategy for the building complex based on the comprehensive status assessment;

[0015] Build an online sharing platform among building complexes, and use transfer learning to transfer the control strategies of the building complex generated on the current building complex to other building complexes, enabling different building complexes to exchange and share the learned models, control strategies, and optimization experiences.

[0016] Optionally, through a dynamically configurable sensor network and user interface pre-deployed in the building complex, collect and preprocess environmental parameters, device status, and user data, including:

[0017] Divide the space of the building complex into grids or regions, introduce a grid importance index, assign initial sensor deployment nodes to each grid or region, and preset a basic activation strategy for each node;

[0018] Construct and hybridize an optimization algorithm framework based on the simulated annealing algorithm and genetic algorithm, perform iterative search based on the hybrid optimization algorithm framework, and combine the introduced innovative operators including node position adjustment and activation strategy recombination to generate a new sensor deployment plan and a new activation strategy;

[0019] In each iteration, evaluate the deployment positions and activation strategies of sensor nodes according to the fitness function;

[0020] Dynamically update the sensor deployment plan and activation strategy using the probability acceptance criterion of simulated annealing, and synchronously update the grid importance index and fitness function parameters;

[0021] After deploying the dynamic sensor network based on the new sensor deployment nodes, dynamically activate or deactivate the sensors in the corresponding regions based on the activation strategy or real-time requirements input by the user to collect environmental parameters, device status, and user behavior data;

[0022] By analyzing the user behavior data, obtain the hot spots of user activities, and determine the deployment location of the user interface in combination with environmental characteristics;

[0023] After deploying the user interface, receive and store user feedback data through the user interface;

[0024] Preprocess the collected environmental parameters, device status, user behavior data, and user feedback data;

[0025] Among them,

[0026] The grid importance index is:

[0027] ;

[0028] In the formula, E G is the rate of change of environmental parameters, S is the space complexity including the number of spatial elements and the number of connected elements, P is the density of people flow, E n is the rate of change of environmental parameters, α , β , γ is the basic weight coefficient, is an exponential coefficient, which is used to adjust the nonlinear effect of the corresponding factor. A is the activity level within the grid, including the number of business transactions and social interactions. A max is the maximum activity level, η is the weight coefficient of the activity level, T It's the time factor. T max is the maximum value of the time factor, θ is the weight coefficient of the time factor, D is the distance between the grid and a specific location, λ is the distance attenuation coefficient, is the distance decay function;

[0029] The fitness function is:

[0030] ;

[0031] In the formula, F is the total value of the fitness function, is the weight coefficient of each indicator and satisfies , n is the number of indicators of target coverage, m is the number of indicators of target accuracy, C ti It is i The actual value of the target coverage, C mi It is i The maximum or expected value of target coverage, A dj It is j The actual data value of the target accuracy, A mj It is j The maximum or expected value of the target accuracy, E c is the actual value of energy efficiency, S t is the actual value of communication stability,R s is the actual value of the response time.

[0032] Optionally, a dynamic Gaussian mixture model is used to perform clustering analysis on at least one of the preprocessed environmental parameters, device states, and user data to identify and divide into multiple clusters or groups each representing a specific comfort zone, including:

[0033] Select at least one of the environmental parameters, device states, and user data as the object to be analyzed;

[0034] Determine the number of Gaussian components corresponding to the number of clusters or groups formed in the Gaussian mixture model K , and initialize the parameter configuration of the Gaussian mixture model;

[0035] Divide the object to be analyzed into multiple consecutive time windows, and move the time window according to a set step size to cover the entire time series of the object to be analyzed. Each new time window contains a part of new data and a part of data overlapping with the previous time window;

[0036] Within each new time window, use the initialized Gaussian mixture model to fit the object to be analyzed, and alternately perform the expectation step and the maximization step in an expectation-maximization manner until a preset number of iterations is reached or the change in model parameters is less than a threshold. Among them, in the expectation step, calculate the responsibility degree of each data point belonging to each Gaussian component according to the current model parameters, and in the maximization step, update the parameter configuration of the Gaussian mixture model according to the responsibility degree;

[0037] Through the fitting process, divide the data points of the object to be analyzed into K clusters or groups, and determine the clusters or groups corresponding to the data points within each time window that represent specific comfort zones;

[0038] where the responsibility degree is:

[0039] ;

[0040] In the formula, z pq is the responsibility degree of the p th data point for the q th Gaussian component, w q is the weight of the q th Gaussian component, σ q is the standard deviation of the q th Gaussian component, x p is the p th data point, μq is the mean of the q th Gaussian component, is the p th data point and the q th Gaussian component mean squared Euclidean distance between, K is the total number of Gaussian components, w k is the k th Gaussian component weight, σ k is the k th Gaussian component standard deviation, μ k is the k th Gaussian component mean, χ is the sharpening parameter, an adjustable parameter used to control the sharpness of the responsibility calculation.

[0041] Optionally, analyze and process the clusters or groups through a deep learning network to obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports including:

[0042] Construct a deep learning network architecture containing a multi-layer perceptron and a recurrent neural network, and initialize the parameters of the network architecture;

[0043] Through forward propagation calculation, use the activation function to learn the non-linear relationship from the input clusters or groups, and adjust the network parameters through the backpropagation algorithm and optimizer to minimize the prediction error to complete the training of the multi-layer perceptron of the deep learning network architecture;

[0044] Train the recurrent neural network of the deep learning network architecture according to the input clusters or groups and the way of time series backpropagation;

[0045] Apply the trained deep learning network architecture to the analysis task of clusters or groups, and through the collaborative analysis and processing of the multi-layer perceptron and the recurrent neural network, output the analysis results of environmental parameters, device status, and user data;

[0046] Based on the analysis results of environmental parameters and device status, combined with expert experience, output an environmental data analysis report;

[0047] Based on the analysis results of user data, generate a user behavior analysis report;

[0048] Use the trained deep learning network architecture to predict future environmental parameters, device status, or user behavior to obtain model prediction outputs.

[0049] Optionally, based on the environmental data analysis results, model prediction outputs, and user behavior analysis reports, a comprehensive status assessment of the building complex services is obtained through fuzzy synthesis operations, and at least one control strategy for the building complex is generated based on the comprehensive status assessment, including:

[0050] Align and perform multi-dimensional fusion on the environmental data analysis results, model prediction outputs, and user behavior analysis reports in the time and space dimensions, and eliminate information redundancy and conflicts based on a predefined rule set;

[0051] Analyze the data after eliminating information redundancy and conflicts, and select evaluation indicators that can reflect the environmental status, prediction trends, and user needs of the building complex to construct a comprehensive status indicator system;

[0052] Use the fuzzy evaluation matrix and weight vector determined by integrating the evaluations of all experts to perform fuzzy synthesis operations on the data in the comprehensive status indicator system to generate a comprehensive status assessment of the building complex services;

[0053] Based on the comprehensive status assessment of the building complex services, analyze the control objectives and control requirements of the building complex;

[0054] Construct a reinforcement learning model, and use the comprehensive status of the building complex services as the state space of the reinforcement learning environment. Integrate one or more agents in the reinforcement learning environment, and each agent can perceive the environmental status and make decisions according to the learned strategy;

[0055] According to the control objectives and control requirements of the building complex, combined with predefined performance indicators, simulate the operation process of the building complex services in the constructed environmental model, so that the agents can perceive the environmental status and select control actions according to the current status, observe the changes in the environmental status and the received reward signals after executing the actions, and continuously adjust the control strategy according to the received reward signals with the goal of maximizing the cumulative reward;

[0056] By allowing the agents to continuously interact and try and error in the simulation environment, gradually converge the strategies of the agents to obtain at least one control strategy;

[0057] Among them, the rule set includes data verification rules, redundancy detection rules, and conflict detection and resolution rules:

[0058] The data verification rules include: checking whether the data is within the threshold range calculated by at least one statistical method; using statistical verification, logical verification, expert review, and a predefined format template to cross-verify the data format;

[0059] The redundancy detection rules include: if two data sources provide the same type of data and the time stamps differ within a predefined time window, it is regarded as redundant data; for redundant data, select the data source with the latest time stamp as the valid data;

[0060] The conflict detection and resolution rules include: constructing an association graph model based on the data provided by multiple data sources, traversing each node and edge of the association graph model, triggering conflict detection when differences containing inconsistent data values, contradictory association relationships, or abnormal data patterns are found; when a conflict occurs, recording the detected conflict nodes and edges, and performing self-processing such as data correction, merging, or deletion, feeding back the conflict resolution result to the association graph model, updating the status of nodes and edges, and if the same problem still exists in the next round, requesting manual intervention.

[0061] Optionally, after obtaining a comprehensive status assessment based on the environmental data analysis results, model prediction outputs, and user behavior analysis reports, and generating a control strategy for the building complex, it further includes:

[0062] Creating a digital twin building complex model that can real-time simulate the status of the building complex under various conditions;

[0063] Mapping the generated control strategy to the control instructions of the building complex;

[0064] Applying the mapped control instructions to the digital twin building complex model, observing and tracking the environmental parameter change information, equipment status change information, and user feedback information within the building complex;

[0065] Feeding the environmental parameter change information, equipment status change information, and user feedback information within the building complex into the constructed reinforcement learning model to adaptively optimize and adjust the control strategy to obtain the optimal control strategy.

[0066] Optionally, constructing an online sharing platform between building complexes, and using transfer learning to transfer the control strategy of the building complex generated on the current building complex to other building complexes, enabling different building complexes to exchange and share the learned models, control strategies, and optimization experiences, including:

[0067] Constructing an online sharing platform between building complexes based on a cloud computing platform;

[0068] Responding to the transfer instruction input by the user, determining the source building complex and the target building complex;

[0069] Analyzing the control action sets of the source building complex and the target building complex, identifying common actions and specific actions, directly mapping the common actions, and tailoring or expanding the specific actions according to the control requirements of the target building complex to ensure that the transferred strategy can adapt to the action space of the target building complex;

[0070] Using the policy function learned on the source building complex as the initial point to construct the basis of the transfer strategy;

[0071] On the target building complex, based on the migration formula and the set optimization objectives, adjust or continue to learn the policy function. During the migration process, continuously evaluate the performance of the migrated policy on the target building complex;

[0072] According to the evaluation results received in each round, adjust the parameters in the migration formula, and repeat multiple times until the preset migration performance requirements or the maximum number of iterations are reached;

[0073] Among them, the migration formula is:

[0074] ;

[0075] In the formula, J m is the total return of the migration process, T b is the total number of time steps of the migration process, τ is the discount factor, which is used to balance the importance of immediate return and future return, is in the target building complex T BG in the state s t when taking the action a t the immediate return obtained, λ is the state difference weight, which is used to adjust the influence of the state space difference on the migration process, is the state of the source building complex and the state of the target building complex the difference metric between.

[0076] In the second aspect, an AI-driven online intelligent Internet of Things monitoring and evaluation system for building complex services provided by an embodiment of the present invention includes:

[0077] A collection and processing module, configured to collect and preprocess environmental parameters, device status, and user data through a dynamically configurable sensor network and a user interface pre-deployed in the building complex;

[0078] A clustering analysis module, configured to perform clustering analysis on at least one of the preprocessed environmental parameters, device status, and user data by using a dynamic Gaussian mixture model to identify and divide multiple clusters or groups each representing a specific comfort area;

[0079] A deep learning module, configured to analyze and process the clusters or groups through a deep learning network to obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports;

[0080] A status evaluation and strategy output module, configured to obtain a comprehensive status evaluation of the building complex service through fuzzy synthesis operations based on the environmental data analysis results, model prediction outputs, and user behavior analysis reports, and generate at least one control strategy for the building complex based on the comprehensive status evaluation;

[0081] A strategy sharing module, configured to build an online sharing platform among building complexes, and use transfer learning to transfer the control strategies of the building complexes generated on the current building complex to other building complexes, so that different building complexes can exchange and share the learned models, control strategies, and optimization experiences.

[0082] Thirdly, an AI-driven online intelligent IoT monitoring and evaluation device for building complex services provided by an embodiment of the present invention includes:

[0083] A cloud computing platform, configured to connect multiple building complexes into a network and use transfer learning to achieve data sharing and collaboration among building complexes;

[0084] A central control platform, configured to perform the AI-driven online intelligent IoT monitoring and evaluation method for building complex services as described above; and,

[0085] A database, communicatively connected to the central control platform, configured to store the monitoring data of the building complex, the constructed models, control strategies, and optimization experiences.

[0086] Fourthly, an embodiment of the present invention provides a computer-readable medium, on which computer-executable instructions are stored, and when the executable instructions are executed by a processor, the AI-driven online intelligent IoT monitoring and evaluation method for building complex services as described above is implemented.

[0087] (III) Advantageous Effects

[0088] The beneficial effects of the present invention are:

[0089] Through the dynamically configurable sensor network and user interface pre-deployed in the building complex, the present invention can comprehensively and accurately collect and preprocess environmental parameters, device status, and user data. This measure provides a rich and high-quality information basis for subsequent data analysis and intelligent decision-making, thereby ensuring the accuracy and timeliness of system responses. This not only provides a rich data source for subsequent data analysis but also greatly improves the reliability of the data.

[0090] Secondly, the dynamic Gaussian mixture model is used to perform clustering analysis on the preprocessed data. The present invention can identify and divide multiple clusters or groups representing specific comfort zones. This analysis method not only helps to more deeply understand the complexity and variability of the building complex environment, but also provides strong data support for subsequent deep learning and intelligent control strategy generation. Further, through in-depth analysis and processing of the clusters or groups by the deep learning network, the present invention obtains accurate environmental data analysis results, model prediction outputs, and user behavior analysis reports. These analysis results not only reveal the internal laws and trends of the building complex environment, but also provide a scientific basis for comprehensive status assessment and control strategy generation. In addition, based on the above analysis results, the present invention obtains a comprehensive status assessment of the building complex service through fuzzy synthesis operation, and generates targeted building complex control strategies accordingly. These control strategies not only consider the real-time state of the environment, but also fully integrate the actual needs and behavior patterns of users, thus ensuring that the building complex can maintain an efficient, comfortable and energy-saving operating state under various circumstances.

[0091] Furthermore, the present invention also constructs an online sharing platform among building complexes, and uses transfer learning technology to transfer the control strategies generated on the current building complex to other building complexes. This cross-building complex knowledge sharing and transfer learning not only accelerates the intelligent process of new building complexes, but also promotes the exchange and cooperation among different building complexes in terms of models, control strategies, and optimization experiences, thus promoting the continuous innovation and development in the field of building complex intelligent control.

[0092] Thus, by integrating methods such as Internet of Things monitoring, artificial intelligence algorithm models, intelligent control, comprehensive evaluation, and online sharing, the present invention significantly improves the response speed and accuracy of the system. Using real-time data processing and online learning mechanisms, the present invention can quickly and accurately adapt to environmental changes and user needs. In addition, the intelligent control strategy of the present invention effectively optimizes energy use, reduces operating costs, and at the same time enhances the system's adaptability to complex usage environments and long-term stability through continuous learning and adaptive updates. This highly adaptive and user-friendly design significantly improves user satisfaction and provides a solid foundation for the long-term operation and future upgrade of the system, demonstrating its important value and potential in technological innovation and practical applications. Generally speaking, this solution solves many problems existing in traditional Internet of Things systems, realizes real-time data analysis, intelligent decision-making, and system integration, and significantly improves the management efficiency and user experience of building complexes. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 It is a schematic flowchart of the method provided by the embodiment of the present invention;

[0094] Figure 2 It is a specific schematic flowchart of step S1 of the method provided by the embodiment of the present invention;

[0095] Figure 3 It is a schematic diagram of the specific process of step S2 of the method provided by the embodiment of the present invention;

[0096] Figure 4 It is a schematic diagram of the specific process of step S3 of the method provided by the embodiment of the present invention;

[0097] Figure 5 It is a schematic diagram of the specific process of step S4 of the method provided by the embodiment of the present invention;

[0098] Figure 6 It is a schematic diagram of the specific process of step S5 of the method provided by the embodiment of the present invention;

[0099] Figure 7 It is a schematic diagram of the composition of the system provided by the embodiment of the present invention. Detailed implementation manners

[0100] In order to better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the accompanying drawings through specific implementation manners.

[0101] As Figure 1 shown, an AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services proposed by an embodiment of the present invention includes: collecting and preprocessing environmental parameters, device status, and user data through a dynamically configurable sensor network and user interface pre-deployed in the building complex; using a dynamic Gaussian mixture model to perform clustering analysis on at least one of the preprocessed environmental parameters, device status, and user data to identify and divide multiple clusters or groups each representing a specific comfort area; analyzing and processing the clusters or groups through a deep learning network to obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports; based on the environmental data analysis results, model prediction outputs, and user behavior analysis reports, obtaining a comprehensive status evaluation of the building complex services through fuzzy synthesis operations, and generating at least one control strategy for the building complex based on the comprehensive status evaluation; building an online sharing platform among building complexes, and using transfer learning to transfer the control strategy of the building complex generated on the current building complex to other building complexes, so that different building complexes can exchange and share learned models, control strategies, and optimization experiences.

[0102] Through the dynamically configurable sensor network and user interface pre-deployed in the building complex, the present invention can comprehensively and accurately collect and preprocess environmental parameters, device status, and user data. This measure provides a rich and high-quality information basis for subsequent data analysis and intelligent decision-making, thus ensuring the accuracy and timeliness of system responses. This not only provides a rich data source for subsequent data analysis but also greatly improves the reliability of the data.

[0103] Secondly, the dynamic Gaussian mixture model is used to perform clustering analysis on the preprocessed data. The present invention can identify and divide multiple clusters or groups representing specific comfort zones. This analysis method not only helps to understand the complexity and variability of the building complex environment more deeply, but also provides strong data support for subsequent deep learning and intelligent control strategy generation. Further, through in-depth analysis and processing of the clusters or groups by the deep learning network, the present invention obtains accurate environmental data analysis results, model prediction outputs, and user behavior analysis reports. These analysis results not only reveal the internal laws and trends of the building complex environment, but also provide a scientific basis for comprehensive status assessment and control strategy generation. In addition, based on the above analysis results, the present invention obtains a comprehensive status assessment of the building complex service through fuzzy synthesis operation, and generates targeted building complex control strategies accordingly. These control strategies not only consider the real-time state of the environment, but also fully integrate the actual needs and behavior patterns of users, thus ensuring that the building complex can maintain an efficient, comfortable and energy-saving operation state under various circumstances.

[0104] Furthermore, the present invention also constructs an online sharing platform between building complexes, and uses transfer learning technology to transfer the control strategies generated on the current building complex to other building complexes. This cross-building complex knowledge sharing and transfer learning not only accelerates the intelligent process of new building complexes, but also promotes the exchange and cooperation between different building complexes in terms of models, control strategies, and optimization experiences, thus promoting the continuous innovation and development of the entire building complex intelligent control field.

[0105] Thus, the present invention significantly improves the response speed and accuracy of the system by integrating methods such as Internet of Things monitoring, artificial intelligence algorithm models, intelligent control, comprehensive evaluation, and online sharing. Using real-time data processing and online learning mechanisms, the present invention can quickly and accurately adapt to environmental changes and user needs. In addition, the intelligent control strategy of the present invention effectively optimizes energy use, reduces operating costs, and at the same time enhances the adaptability and long-term stability of the system to complex usage environments through continuous learning and adaptive updates. This highly adaptive and user-friendly design significantly improves user satisfaction and provides a solid foundation for the long-term operation and future upgrade of the system, demonstrating its important value and potential in technological innovation and practical applications. Generally speaking, this solution solves many problems existing in traditional Internet of Things systems, realizes real-time data analysis, intelligent decision-making, and system integration, and significantly improves the management efficiency and user experience of building complexes.

[0106] To better understand the above technical solution, the exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more clear and thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0107] Specifically, the embodiments of the present invention provide an AI-driven online intelligent Internet of Things monitoring and evaluation method for building complexes, including:

[0108] S1. Collect and preprocess environmental parameters, device status, and user data through a dynamically configurable sensor network and user interface pre-deployed in the building complex.

[0109] The sensor network covers all areas that need to be monitored, ensuring the comprehensiveness and accuracy of the data. Environmental parameters include, but are not limited to, temperature, humidity, light intensity, etc. These parameters are crucial for evaluating the comfort and energy efficiency of the building complex. At the same time, the sensors also continuously monitor the status of various devices, such as the on / off status of air conditioners, lights, and curtains, providing data support for subsequent intelligent control.

[0110] Furthermore, as Figure 2 shown, step S1 includes:

[0111] S11. Divide the space of the building complex into grids or regions by using the structure, usage frequency, and environmental parameter change gradient of the building obtained. Among them, the environmental parameter change gradient refers to the rate of change and direction of environmental parameters (such as temperature, humidity, light intensity, air quality, etc.) as the spatial position or time changes within a certain spatial range. This division needs to ensure that each grid or region can reflect the internal spatial characteristics, usage conditions, and environmental change characteristics.

[0112] S12. Introduce a grid importance index, assign an initial sensor deployment node to each grid or region, and preset a basic activation strategy for each node.

[0113] Introduce a grid importance index, which comprehensively considers the structural complexity, usage frequency, and dynamic variability of environmental parameters within the grid.

[0114] According to the grid importance index, assign an initial sensor deployment node to each grid or region. The position of the node should be selected at a location that can maximize the monitoring effect.

[0115] Meanwhile, preset a basic activation policy for each initially deployed sensor node, including but not limited to working mode, data acquisition frequency, and communication method, to ensure the realization of preliminary environmental monitoring and data acquisition functions.

[0116] Specifically, the grid importance index is as follows:

[0117] ;

[0118] In the formula, E G is the environmental parameter change rate, S is the spatial complexity including the number of spatial elements and the number of connection elements. For example, a simple single-layer space may be given a relatively low S value, such as 1 or 2, while a multi-layer, multi-channel, and complex-structured building may be given a relatively high S value, such as 7 or 8; P is the pedestrian flow density, E n is the environmental parameter change rate, α , β , γ is the basic weight coefficient, is the exponential coefficient used to adjust the non-linear influence of various factors, A is the activity level within the grid including the number of commercial transactions and the number of social interactions, Amax is the maximum value of the activity level, η is the weight coefficient of the activity level, T is the time factor, which can be quantified by a timestamp or a time period, such as peak hours of the day or off-peak / peak seasons, Tmax is the maximum value of the time factor, θ is the weight coefficient of the time factor, D is the distance between the grid and a specific location, λ is the distance attenuation coefficient. If the distance has a great influence on the grid importance, λ can be set relatively large, so that the importance of grids slightly farther away will decrease rapidly, is a distance attenuation function used to simulate the effect of the grid importance gradually decreasing with the increase of distance.

[0119] S13. Construct a hybrid optimization algorithm framework based on the simulated annealing algorithm and the genetic algorithm. The present invention combines the advantages of the simulated annealing algorithm and the genetic algorithm to construct a hybrid optimization algorithm framework. This framework aims to improve the search efficiency and the quality of the solution through the combination of global search and local optimization. In this framework, the initial parameters of the algorithm are clearly set, including but not limited to the initial temperature, annealing rate, population size, crossover probability, mutation probability, and selection strategy.

[0120] S14. Conduct iterative search based on the hybrid optimization algorithm framework, and combine the introduced innovative operators that include node position adjustment and activation strategy recombination to generate a new sensor deployment plan and a new activation strategy; conduct iterative search based on the hybrid optimization algorithm framework. Then, in each iteration, evaluate the deployment positions and activation strategies of sensor nodes according to the fitness function.

[0121] During the search process, combine innovative operators, such as the fine-tuning algorithm for node positions and the recombination mechanism for activation strategies, to generate a new sensor deployment plan and activation strategy. The design of these innovative operators aims to increase the diversity and depth of the search, so as to find a better solution.

[0122] In each iteration, evaluate the current deployment positions and activation strategies of sensor nodes according to a comprehensive fitness function. Among them, the fitness function is:

[0123] ;

[0124] In the formula, F is the total value of the fitness function, is the weight coefficient of each index, and satisfies , C ti is the actual value of the i th target coverage rate, C mi is the i th maximum or expected value of the target coverage rate, A dj is the j th actual data value of the target accuracy, A mj is the j th maximum or expected value of the target accuracy, E c is the actual value of the energy consumption efficiency. Taking the reciprocal here is to make the lower the energy consumption, the higher the fitness. S t is the actual value of the communication stability. The higher this value, the more stable the communication. R s is the actual value of the system response time. The lower this value, the faster the system response.

[0125] S15. Use the probability acceptance criterion of simulated annealing, adjust the acceptance probability according to the temperature parameter, balance global and local search, and decide whether to accept the new deployment plan and information activation plan to obtain the optimal sensor deployment nodes and activation strategies.

[0126] S16. If a new deployment plan and information activation plan are received, the grid importance index and fitness function are dynamically updated according to the operation data under the new deployment plan and information activation plan.

[0127] S17. After deploying the dynamic sensor network based on the optimal sensor deployment nodes, the sensors in the corresponding area are dynamically activated or deactivated based on the activation policy or real-time requirements input by the user to collect environmental parameters, device status, and user behavior data.

[0128] S18. By analyzing the user behavior data, the hot spot areas of user activities are obtained, and the deployment location of the user interface is determined in combination with the environmental characteristics.

[0129] Specifically, collect and analyze user behavior data, including but not limited to the movement trajectory, stay time, interaction frequency, etc. of the user in a specific space to determine the hot spot areas and preferences of user activities. At the same time, combine environmental characteristics, such as spatial layout, lighting conditions, noise level, etc., to comprehensively evaluate and determine the best deployment location of the user interface. This location should be convenient for users to observe and operate, while ensuring the visibility and usability of the user interface in the environment, thereby improving the user experience and interaction efficiency.

[0130] S19. After deploying the user interface, receive and store the user feedback data through the user interface, clean the collected environmental parameters, device status, and user feedback data to remove duplicate, invalid, or incorrect data entries; perform noise filtering on the cleaned data; perform outlier detection on the data after noise filtering to identify and process those data points that significantly deviate from the normal range, and, perform normalization or standardization on the data after outlier detection, and synchronize and align the data from different sensors according to the collected timestamps.

[0131] During the data collection process, duplicate, invalid, or incorrect data entries may be mixed in. To ensure data quality, the present invention cleans all the collected data. The cleaning process includes removing duplicate data, identifying and deleting invalid data (such as incorrect formats, values outside the reasonable range, etc.), and correcting or marking incorrect data.

[0132] Although the data has been cleaned, there may still be certain noise in the data. The noise may be caused by the errors of the sensors themselves, environmental interference, or problems in the data transmission process. To further improve the accuracy of the data, the present invention performs noise filtering on the cleaned data.

[0133] An outlier refers to a value that significantly deviates from the normal range in data, and it may be caused by abnormal situations, faults, or incorrect operations. If these outliers are directly used for analysis without being processed, they may lead to deviations or misleading in the results. Therefore, the present invention performs outlier detection on the data after noise filtering.

[0134] For the convenience of subsequent data analysis and model training, the present invention normalizes or standardizes the data after outlier detection. Normalization is to scale the data to a specified range (such as between 0 and 1), while standardization is to convert the data into a distribution with a mean of 0 and a standard deviation of 1. These processes help to eliminate the dimensional differences and numerical range differences between different features and improve the comparability of the data. At the same time, according to the collected timestamp information, the data from different sensors are time-synchronized and aligned to ensure their consistency in the time dimension. This step is crucial for subsequent time-series data analysis.

[0135] S2. Use a dynamic Gaussian mixture model to perform clustering analysis on at least one of the preprocessed environmental parameters, device states, and user data to identify and divide multiple clusters or groups that each represent a specific comfort area.

[0136] Further, as Figure 3 shown, step S2 includes:

[0137] S21. Select at least one of the environmental parameters, device states, and user data as the object to be analyzed.

[0138] S22. Determine the number of Gaussian components corresponding to the number of clusters or groups formed in the Gaussian mixture model K , K the value corresponds to the number of clusters or groups expected to be formed, and initialize the parameter configuration of the Gaussian mixture model, such as the mixing coefficient, mean vector, and covariance matrix.

[0139] Determine the number K of Gaussian components in the Gaussian mixture model, that is, the number of clusters or groups to be formed. Initialize the mixing coefficient, mean vector, and covariance matrix of each Gaussian component. These parameters will be used to describe the data distribution characteristics of each potential comfort area.

[0140] S23. Divide the object to be analyzed into multiple consecutive time windows, and move the time window according to the set step size to cover the entire time series of the object to be analyzed. Each new time window contains a part of new data and a part of data overlapping with the previous time window. The specific range of overlap = window size - step size.

[0141] S24. In each new time window, use the initialized Gaussian mixture model to fit the object to be analyzed. By means of the expectation-maximization method, alternately perform the expectation step and the maximization step until the preset number of iterations is reached or the change in model parameters is less than the threshold. Among them, in the expectation step, calculate the responsibility degree of each data point belonging to each Gaussian component according to the current model parameters. In the maximization step, update the parameter configuration of the Gaussian mixture model according to the responsibility degree.

[0142] Among them, the responsibility degree is:

[0143] ;

[0144] In the formula, z pq is the responsibility degree of the p th data point for the q th Gaussian component, w q is the weight of the q th Gaussian component, σ q is the standard deviation of the q th Gaussian component, x p is the p th data point, μ q is the mean of the q th Gaussian component, is the square of the Euclidean distance between the p th data point and the mean of the q th Gaussian component, K is the total number of Gaussian components, w k is the weight of the k th Gaussian component, σ k is the standard deviation of the k th Gaussian component, μ k is the k th Gaussian component mean, χ is the sharpening parameter, which is an adjustable parameter used to control the sharpening degree of the responsibility degree calculation. When the sharpening parameter = 1, the responsibility degree formula degenerates into the standard Gaussian responsibility degree calculation; when the sharpening parameter > 1, the responsibility degree calculation will be sharper, making the data point more inclined to belong to the Gaussian component closer to it; when 0 < the sharpening parameter < 1, the responsibility degree calculation will be smoother.

[0145] It should be clear that the present invention introduces an adjustable sharpening parameter, which can flexibly adjust the calculation method of the responsibility degree as needed. By adjusting the value of the sharpening parameter, a better balance can be achieved between the fitting ability and generalization ability of the model. For example, when dealing with data having an obvious clustering structure, the value of the sharpening parameter can be increased to emphasize the membership relationship between the data points and their nearest Gaussian components; while when dealing with data with a relatively uniform distribution or more noise, we can reduce the value of the sharpening parameter to obtain a smoother responsibility degree distribution.

[0146] S25. Through the fitting process, divide the data points of the object to be analyzed into K clusters or groups, and determine the clusters or groups corresponding to the data points within each time window that represent specific comfort zones.

[0147] Through the fitting process, divide the data points of the object to be analyzed into K clusters or groups. Determine the classification attribution of the clusters or groups corresponding to the data points within each time window, that is, each data point is assigned to the cluster corresponding to the Gaussian component with the highest probability density. Thus, each cluster or group represents a specific comfort zone or other related characteristic zone.

[0148] Next, the present invention uses a Gaussian mixture model to construct the comfort range division. The Gaussian mixture model can be a weighted combination of multiple Gaussian distributions for fitting data, and its mathematical formula is as follows:

[0149] ;

[0150] In the formula, is the probability density of the data point x , K is the total number of Gaussian components, that is, the number of different Gaussian distributions in the model, is the mixing coefficient of the k th Gaussian distribution, satisfying and . These coefficients are the weights of each Gaussian distribution. is a multivariate normal distribution with a mean of and a covariance matrix of .

[0151] Next, the probability density function of each Gaussian component can be:

[0152] ;

[0153] In the formula, x is d a dimensional data vector, is the mean vector of the k th component, is the covariance matrix of the k th component, which is the diffusion and shape of the component data. is the determinant of the covariance matrix.

[0154] Thus, the Gaussian mixture model is a model composed of multiple Gaussian distributions. Each Gaussian distribution has its own mean and standard deviation (or covariance matrix), and the contribution of each Gaussian distribution to the overall model is determined by the weight coefficient. This model can fit any type of distribution, especially suitable for the case where the data in the same set contains multiple different distributions.

[0155] S3. Analyze and process the clusters or groups through a deep learning network to obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports.

[0156] Furthermore, as Figure 4 shown, step S3 includes:

[0157] S31. Construct a deep learning network architecture that includes a multi-layer perceptron and a recurrent neural network, and initialize the parameters of the network architecture.

[0158] S32. Through forward propagation calculation, use the activation function to learn the non-linear relationship from the input clusters or groups, and adjust the network parameters through the backpropagation algorithm and optimizer to minimize the prediction error, so as to complete the training of the multi-layer perceptron of the deep learning network architecture.

[0159] S33. Train the recurrent neural network of the deep learning network architecture according to the input clusters or groups and the way of time series backpropagation.

[0160] S34. Apply the trained deep learning network architecture to the analysis task of clusters or groups, and through the collaborative analysis and processing of the multi-layer perceptron and the recurrent neural network, output the analysis results of environmental parameters, device status, and user data.

[0161] S35. Based on the analysis results of environmental parameters and device status, combined with expert experience, output an environmental data analysis report.

[0162] S36. Based on the analysis results of user data, generate a user behavior analysis report.

[0163] S37. Use the trained deep learning network architecture to predict future environmental parameters, device status, or user behavior to obtain model prediction outputs.

[0164] In another specific embodiment, a deep learning network is used to analyze and process data. The network design includes a multi-layer perceptron and a recurrent neural network, which are used to capture the complex patterns and dependencies of time series data.

[0165] Design a multi-layer perceptron network MLP, which includes multiple hidden layers, and each layer consists of multiple neurons. Initialize the weight matrix of the network W and the bias vector b . Through forward propagation calculation, use the activation function to learn the non-linear relationship from the input features, and adjust the network parameters through the backpropagation algorithm and optimizer to minimize the prediction error. Its forward propagation calculation is as follows:

[0166] ;

[0167] In the formula, W and b are the weight matrix and the bias vector respectively, x is the input vector, h is the output after passing through the activation function, is the activation function, such as ReLU or Sigmoid.

[0168] Next, design a recurrent neural network, especially a long short-term memory network (LSTM), to capture the complex patterns and dependencies in time series data. Initialize the weight matrix of the LSTM (such as W xh and W hh ) and the bias vector b h . Use the hidden state update formula of the LSTM, combined with the activation function (usually tanh), to process the input vector t at each time step x t , and update the hidden state h t . Train the LSTM network through the Backpropagation Through Time (BPTT) algorithm, and optimize the network parameters to capture the long-term dependencies in time series data. Its hidden state update formula is:

[0169] .

[0170] Then, apply the trained deep learning model to the analysis task of clustering or grouping, and output the analysis results.

[0171] S4. Based on the environmental data analysis results, model prediction outputs, and user behavior analysis reports, obtain a comprehensive status assessment of the building complex service through fuzzy composition operation, and generate at least one control strategy for the building complex based on the comprehensive status assessment.

[0172] Furthermore, as shown in Figure 5 , step S4 includes:

[0173] S41. Align the environmental data analysis results, model prediction outputs, and user behavior analysis reports in terms of time and space dimensions and perform multi-dimensional fusion, and eliminate information redundancy and conflicts based on a predefined rule set.

[0174] Specifically, the rule set includes data verification rules, redundancy detection rules, and conflict detection and resolution rules:

[0175] The data verification rules include: checking whether the data is within the threshold range calculated by at least one statistical method; cross-verifying the data format using statistical verification, logical verification, expert review, and a preset format template.

[0176] The redundancy detection rules include: if two data sources provide the same type of data and the time stamps differ within a preset time window, the data is considered redundant; for redundant data, select the data source with the latest time stamp as the valid data.

[0177] The conflict detection and resolution rules include: constructing an association graph model based on the data provided by multiple data sources, traversing each node and edge of the association graph model, and triggering conflict detection when differences containing inconsistent data values, contradictory association relationships, or abnormal data patterns are found; when a conflict occurs, record the detected conflict nodes and edges, and perform self-processing of data correction, merging, or deletion, and feedback the conflict resolution result to the association graph model to update the status of the nodes and edges. If the same problem still exists in the next round, request manual intervention.

[0178] S42. Analyze the data after eliminating information redundancy and conflicts, and select evaluation indicators that can reflect the environmental status, prediction trends, and user needs of the building complex to construct a comprehensive status indicator system.

[0179] S43. Use the fuzzy evaluation matrix and weight vector determined by integrating the evaluations of all experts to perform fuzzy synthesis operations on the data in the comprehensive status indicator system to generate a comprehensive status assessment of the building complex service.

[0180] In a specific embodiment, first, it is necessary to clarify all evaluation factors involved in the building complex status assessment, and these factors should cover all aspects such as environment, equipment, operation, and user satisfaction. Each evaluation factor can be further divided into more specific sub-factors, thus forming a hierarchical evaluation factor set.

[0181] For each evaluation factor, invite experts or use relevant data to conduct a fuzzy evaluation. Fuzzy evaluation can be carried out using membership functions, that is, determining the degree to which each factor belongs to different evaluation levels (such as excellent, good, medium, poor). By integrating the evaluations of all experts or the results of data analysis, a fuzzy evaluation matrix can be constructed, where each row represents an evaluation factor, each column represents an evaluation level, and the elements in the matrix are the degrees to which the factor belongs to the corresponding level.

[0182] The importance of different evaluation factors in the evaluation of the building complex status is different. Therefore, it is necessary to assign a weight to each factor. The determination of weights can adopt subjective weighting methods (such as the Delphi method, the analytic hierarchy process) or objective weighting methods (such as the entropy weight method, the principal component analysis method). By determining the weight vector, the influence of each factor in the comprehensive evaluation can be quantified.

[0183] Perform fuzzy composition operations using the fuzzy evaluation matrix and the weight vector. The purpose of this step is to synthesize the fuzzy evaluation results of each factor to obtain an overall evaluation result. Different operators can be used for fuzzy composition operations, such as the weighted average operator, the max-min operator, etc. Select a suitable operator for the operation according to the specific situation.

[0184] After the fuzzy composition operation, a comprehensive evaluation value or evaluation vector can be obtained. This value or vector reflects the comprehensive status of the building complex in all aspects.

[0185] S44. Based on the comprehensive status evaluation of the building complex services, analyze the control objectives and control requirements of the building complex.

[0186] The control objectives of the building complex include: such as energy efficiency improvement, comfort improvement, equipment life extension, etc., and specific control requirements are formulated according to these objectives. Based on these objectives, the system formulates specific control requirements, such as adjusting the operation strategy of the air conditioning system to reduce energy consumption, or optimizing the lighting system to improve user comfort.

[0187] S45. Construct a reinforcement learning model and use the comprehensive status of the building complex services as the state space of the reinforcement learning environment.

[0188] S46. Integrate one or more agents in the reinforcement learning environment, and each agent can perceive the environmental state and make decisions according to the learned strategy.

[0189] S47. Based on the control objectives and control requirements of the building complex, combined with predefined performance indicators, simulate the operation process of the building complex services in the constructed environmental model, enabling the agents to perceive the environmental state and select control actions according to the current state, observe the changes in the environmental state and the received reward signals after executing the actions, and continuously adjust the control strategy according to the received reward signals with the goal of maximizing the cumulative reward.

[0190] Among them, predefined performance metrics, such as the percentage reduction in energy consumption, comfort index, etc., are used to quantitatively evaluate the effectiveness of the control strategy. In the embodiments of the present invention, optimization mechanisms of reinforcement learning algorithms, such as gradient descent, policy gradient, etc., are used to iteratively update and optimize the control strategy of the agent.

[0191] S48. By allowing the agent to continuously interact and try out in the simulation environment, the strategy of the agent is gradually converged to obtain at least one control strategy.

[0192] After sufficient times of interaction and trial and error, the agent will gradually converge to one or more optimal control strategies. These optimal control strategies can maximize or minimize the performance metrics while meeting the control objectives and requirements of the building complex.

[0193] Additionally, after step S4, it further includes: creating a digital twin building complex model that can real-time simulate the state of the building complex under various conditions; mapping the generated control strategy to the control instructions of the building complex, and then applying the mapped control instructions to the digital twin building complex model to observe and track the change information of environmental parameters, equipment status, and user feedback information within the building complex; feeding the change information of environmental parameters, equipment status, and user feedback information within the building complex into the constructed reinforcement learning model to adaptively optimize and adjust the control strategy to obtain the optimal control strategy.

[0194] In yet another specific embodiment, based on the current environmental state s t , the control strategy obtains the control instruction through the policy function π ( s t ):

[0195] ;

[0196] where π(⋅) is the policy function, which may be based on reinforcement learning or neural network. a t is the control action at time step t , such as adjusting the temperature, turning on or off the lights, etc.

[0197] Create a digital twin building complex model that can simulate the state of the building complex in real time under various conditions, and configure it with user agents that can simulate user input commands and feedback information. Then, apply the mapped control commands to the digital twin building complex model and enter the observation and tracking phase. The goal of this phase is to collect data on environmental changes (such as temperature, humidity, light, etc.) and feedback information from user agents (such as whether the user manually adjusted the device settings or expressed satisfaction or dissatisfaction with the environmental state in some way). This information is crucial for evaluating the effectiveness of control strategies and is also the basis for subsequent strategy optimization.

[0198] Finally, feed the collected information on environmental parameter changes, device state changes, and user feedback into the reinforcement learning model. This information is used as input data for model training to help the model learn how to better generate control strategies based on the current environmental state and user preferences. Through continuous iteration and optimization, the model can gradually adapt to different environments and user needs, thereby achieving more intelligent and efficient control while updating the comfort range. This can be the update rule for the strategy:

[0199] ;

[0200] where, is the learning rate, which controls the step size of the control strategy update, is the loss function, which is used to quantify the deviation between the generated strategy and the performance metric, such as energy efficiency optimization or user satisfaction maximization.

[0201] S5. Build an online sharing platform among building complexes, and use transfer learning to transfer the control strategies of the building complex generated on the current building complex to other building complexes, enabling different building complexes to exchange and share the learned models, control strategies, and optimization experiences.

[0202] Furthermore, as Figure 6 shown, step S5 includes:

[0203] S51. Build an online sharing platform among building complexes based on the cloud computing platform.

[0204] S52. Respond to the transfer instruction input by the user and determine the source building complex and the target building complex.

[0205] S53. Analyze the control action sets of the source building complex and the target building complex, identify the common actions and specific actions. For the common actions, directly perform mapping. For the specific actions, perform trimming or expansion according to the control requirements of the target building complex to ensure that the transferred strategy can adapt to the action space of the target building complex.

[0206] S54. Use the policy function learned from the source building complex as the initial point to build the basis for the transfer policy.

[0207] S55. On the target building complex, based on the transfer formula and the set optimization objective, adjust or continue to learn the policy function. Through the reinforcement learning algorithm, iteratively update the transfer policy on the target building complex to maximize the performance metric. At the same time, during the transfer process, continuously evaluate the performance of the transferred policy on the target building complex.

[0208] S56. According to the evaluation results received in each round, adjust the parameters in the transfer formula, and repeat multiple times until the preset transfer performance requirement or the maximum number of iterations is reached.

[0209] According to the evaluation results, adjust the parameters in the transfer formula, such as the state difference weight and the discount factor, to optimize the transfer effect. Repeat the above steps until the preset transfer performance requirement or the maximum number of iterations is reached. Among them, the transfer formula is:

[0210] ;

[0211] In the formula, J m is the total return of the transfer process, T b is the total number of time steps of the transfer process, τ is the discount factor, which is used to balance the importance of immediate rewards and future rewards, is in the target building complex T BG in state s t when taking action a t the immediate reward obtained, λ is the state difference weight, which is used to adjust the influence of the state space difference on the transfer process, is the state of the source building complex and the state of the target building complex the difference measure between them.

[0212] It should be emphasized that the present invention also provides a user-friendly interface that allows users to view the environmental status, control device operations, and provide feedback. The interface design is simple and intuitive, supporting access by multiple devices. And there is a continuous learning mechanism that can continuously optimize its performance according to historical data and continuous user interactions to adapt to the changing needs of users and the dynamic changes of environmental conditions.

[0213] In a specific embodiment, IoT sensors are used to obtain various environmental parameters in the room in real time, such as temperature, humidity, light intensity, etc. The thermal sensation of the current occupants in the room is evaluated through a dynamic Gaussian mixture model. This model models the joint distribution of temperature and humidity, divides different thermal sensation intervals, and determines whether the current state is within the comfort interval. After receiving the feedback, the control system uses machine learning algorithms (such as reinforcement learning) to calculate the optimal control strategy to adjust the temperature and humidity parameters, so that the indoor environment returns to the comfort range. This process includes controlling the operation of devices such as air conditioners, humidifiers, and dehumidifiers to ensure comfort through real-time response.

[0214] Additionally, an AI-driven online intelligent IoT monitoring and evaluation system for building complexes according to an embodiment of the present invention is as Figure 7 shown, including:

[0215] A collection and processing module for collecting and preprocessing environmental parameters, device status, and user data through a dynamically configurable sensor network and user interface pre-deployed in the building complex.

[0216] A clustering analysis module for performing clustering analysis on at least one of the preprocessed environmental parameters, device status, and user data using a dynamic Gaussian mixture model to identify and divide multiple clusters or groups each representing a specific comfort area.

[0217] A deep learning module for analyzing and processing the clusters or groups through a deep learning network to obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports.

[0218] A status evaluation and policy output module for obtaining a comprehensive status evaluation of the building complex service through fuzzy synthesis operations based on the environmental data analysis results, model prediction outputs, and user behavior analysis reports, and generating at least one control policy for the building complex based on the comprehensive status evaluation.

[0219] A policy sharing module for building an online sharing platform between building complexes, and using transfer learning to transfer the control policies of the building complex generated on the current building complex to other building complexes, so that different building complexes can exchange and share the learned models, control policies, and optimization experiences.

[0220] Next, an embodiment of the present invention provides an AI-driven online intelligent Internet of Things monitoring and evaluation device for building complexes, including: a cloud computing platform for connecting multiple building complexes into a network and realizing data sharing and collaboration between building complexes through transfer learning; a central control platform for the AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services as described above; and a database communicatively connected to the central control platform for storing monitoring data of building complexes, constructed models, control strategies, and optimization experiences.

[0221] By integrating an advanced cloud computing platform, a central control platform, and a database, this device provides a comprehensive and efficient intelligent management solution for building complexes.

[0222] First of all, as one of the core components of this device, the cloud computing platform has powerful data processing and computing capabilities. It can tightly connect multiple building complexes to build a huge network system. In this network, through the application of transfer learning technology, seamless data sharing and efficient collaboration can be achieved between building complexes. This not only breaks the limitations of data islands in traditional building management but also greatly improves the utilization value of data resources, laying a solid foundation for the overall intelligent management of building complexes.

[0223] Secondly, the central control platform is responsible for executing the above-mentioned AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services. Relying on powerful algorithm support, it can collect, analyze, and process data information from each building complex in real time, thereby realizing precise monitoring and intelligent evaluation of the status of building complexes. Through the central control platform, users can easily master the operation status of building complexes, timely discover potential problems, and formulate effective countermeasures.

[0224] The database, as an important storage unit of this device, maintains a close communication connection with the central control platform. It is responsible for storing various monitoring data of building complexes, constructed models, control strategies, and optimization experiences. These valuable data resources not only provide strong support for the continuous learning and optimization of the device but also can provide rich references for users in subsequent building complex management.

[0225] Furthermore, an embodiment of the present invention provides a computer-readable medium storing computer-executable instructions that, when executed by a processor, implement the above-mentioned AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services.

[0226] An embodiment of the present invention further provides a computer-readable medium, on which specific computer-executable instructions are stored. When these instructions are executed by a processor, the AI-driven online intelligent Internet of Things monitoring and evaluation method designed specifically for building complexes can be accurately implemented. This innovative medium not only provides a convenient carrier for the implementation of related methods, but also further broadens the application scope of AI in the field of intelligent Internet of Things monitoring and evaluation, thereby greatly improving the intelligent level of building complex management.

[0227] In summary, an embodiment of the present invention provides an AI-driven online intelligent Internet of Things monitoring and evaluation method, system, device, and medium for building complex services. The core advantage of the present invention lies in its advanced learning mechanism, which is mainly reflected in the following aspects:

[0228] Firstly, it is the multi-modal data fusion ability, which enables the system to comprehensively process data from different sources, thereby obtaining a more comprehensive view of the operation situation.

[0229] Secondly, it is the ability of continuous learning and online learning. The system design of the present invention introduces an innovative continuous learning framework, enabling the system to continuously absorb new data and user feedback, and update and optimize the decision-making algorithm in real time. Compared with traditional static AI models, the model of the present invention has stronger environmental adaptability and the ability to cope with changes in behavior patterns, and can continuously improve the decision-making quality and operation efficiency over time. At the same time, the present invention also emphasizes the importance of online learning, allowing the system to perform model training and optimization without interrupting the service. This means that the system can learn new interactions, events, and data in real time during actual operation, without offline processing or manual intervention.

[0230] Furthermore, a sharing mechanism is provided to transfer the control strategy of the building complex generated on the current building complex to other building complexes by using transfer learning, so that different building complexes can exchange and share the learned models, control strategies, and optimization experiences.

[0231] Through the above several mechanisms, the present invention can provide a highly adaptive and intelligent solution, which is significantly superior to traditional systems that rely on preset rules and periodic manual updates. This adaptive learning ability is a major innovation point of the present invention, bringing significant benefits to users and operators, including higher system reliability, lower maintenance costs, and better user experience.

[0232] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0233] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can also be implemented.

[0234] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications after learning the basic creative concepts. Therefore, the technical solution should be construed to include the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0235] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the technical solution of the present invention and its equivalent technologies, the present invention should also include these modifications and variations.

Claims

1. An AI-driven online intelligent IoT monitoring and evaluation method for building complex services, characterized in that: include: Collect and pre-process environmental parameters, equipment status, and user data through a dynamically configurable sensor network and user interface pre-deployed in the building complex; A dynamic Gaussian mixture model is used to perform cluster analysis on at least one of the preprocessed environmental parameters, device status, and user data to identify and divide a plurality of clusters or groups each representing a specific comfort zone, including: selecting at least one of the environmental parameters, device status, and user data as an object to be analyzed; determining the number of Gaussian components in the Gaussian mixture model corresponding to the number of clusters or groups formed K , and initialize the parameter configuration of the Gaussian mixture model; divide the object to be analyzed into multiple continuous time windows, move the time window according to the set step size to cover the entire time series of the object to be analyzed, and each new time window contains a part of new data and a part of data overlapping with the previous time window; in each new time window, use the initialized Gaussian mixture model to fit the object to be analyzed, and use the expectation maximization method to alternately perform the expectation step and the maximization step until the preset number of iterations is reached or the model parameter change is less than the threshold, wherein in the expectation step, the responsibility of each data point belonging to each Gaussian component is calculated according to the current model parameters, and in the maximization step, the parameter configuration of the Gaussian mixture model is updated according to the responsibility; through the fitting process, the data points of the object to be analyzed are divided into K clusters or groups, and determine the clusters or groups representing specific comfort zones corresponding to the data points in each time window; wherein the degree of responsibility is: ； In the formula, z pq For the p Data points for q The responsibility of the Gaussian component, w q It is q The weights of the Gaussian components, σ q It is q The standard deviation of the Gaussian components, x p It is p data points, μ q It is q The mean of the Gaussian components, For the p Data points and q The square of the Euclidean distance between the means of the Gaussian components, K is the total number of Gaussian components, w k It is k The weights of the Gaussian components, σ k It is k The standard deviation of the Gaussian components, μ k It is k The mean of the Gaussian components, χ is the sharpening parameter, which is an adjustable parameter used to control the sharpening degree of responsibility calculation; Through deep learning network analysis and processing of clusters or groups, we can obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports; Based on the environmental data analysis results, model prediction output and user behavior analysis report, a comprehensive status evaluation of the building complex services is obtained through fuzzy synthesis operation, and at least one control strategy for the building complex is generated based on the comprehensive status evaluation; Build an online sharing platform between building complexes, and use transfer learning to migrate the control strategies of the building complexes generated on the current building complex to other building complexes, so that different building complexes can exchange and share the learned models, control strategies and optimization experiences.

2. The AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services according to claim 1, characterized in that: Through a dynamically configurable sensor network and user interface pre-deployed in the building complex, environmental parameters, device status and user data are collected and pre-processed, including: The space of the building complex is divided into grids or areas, the grid importance index is introduced, the initial sensor deployment node is allocated to each grid or area, and the basic activation strategy is preset for each node; A hybrid optimization algorithm framework is constructed based on simulated annealing algorithm and genetic algorithm. Iterative search is performed based on the hybrid optimization algorithm framework. In combination with the introduced innovative operators including node position adjustment and activation strategy reorganization, new sensor deployment schemes and new activation strategies are generated. In each iteration, the deployment location and activation strategy of the sensor nodes are evaluated according to the fitness function; The sensor deployment scheme and activation strategy are dynamically updated using the simulated annealing probability acceptance criterion, and the grid importance index and fitness function parameters are updated synchronously. After the dynamic sensor network is deployed based on the new sensor deployment node, the sensors in the corresponding area are dynamically activated or dormant based on the activation strategy or the real-time needs input by the user to collect environmental parameters, device status and user behavior data; By analyzing user behavior data, we can obtain hot spots of user activity and determine the deployment location of the user interface based on environmental characteristics. After the user interface is deployed, receiving and storing user feedback data through the user interface; Pre-process the collected environmental parameters, equipment status, user behavior data, and user feedback data; in, The grid importance index is: ； In the formula, E G is the rate of change of environmental parameters, S is the space complexity including the number of spatial elements and the number of connected elements, P is the density of people flow, E n is the rate of change of environmental parameters, α , β , γ is the basic weight coefficient, δ , ϵ , ζ is an exponential coefficient, which is used to adjust the nonlinear effect of the corresponding factor. A is the activity level within the grid, including the number of business transactions and social interactions. A max is the maximum activity level, η is the weight coefficient of the activity level, T It's the time factor. T max is the maximum value of the time factor, θ is the weight coefficient of the time factor, D is the distance between the grid and a specific location, λ is the distance attenuation coefficient, is the distance decay function; The fitness function is: ； In the formula, F is the total value of the fitness function, is the weight coefficient of each indicator and satisfies , n is the number of indicators of target coverage, m is the number of indicators of target accuracy, C ti It is i The actual value of the target coverage, C mi It is i The maximum or expected value of target coverage, A dj It is j The actual data value of the target accuracy, A mj It is j The maximum or expected value of the target accuracy, E c is the actual value of energy efficiency, S t is the actual value of communication stability, R s is the actual value of the response time.

3. The AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services according to claim 1, characterized in that: Through deep learning network analysis and processing of clusters or groups, we can obtain environmental data analysis results, model prediction outputs, and user behavior analysis reports including: Build a deep learning network architecture consisting of a multi-layer perceptron and a recursive neural network, and initialize the parameters of the network architecture; Through forward propagation calculation, the activation function is used to learn nonlinear relationships from the input clusters or groups, and the network parameters are adjusted through the back-propagation algorithm and optimizer to minimize the prediction error to complete the training of the multi-layer perceptron of the deep learning network architecture; Training a recurrent neural network of a deep learning network architecture based on clustering or grouping of inputs and sequential backpropagation; Apply the trained deep learning network architecture to cluster or group analysis tasks, and output the analysis results of environmental parameters, device status, and user data through the collaborative analysis and processing of multi-layer perceptron and recurrent neural network; Based on the analysis results of environmental parameters and equipment status, combined with expert experience, an environmental data analysis report is output; Generate user behavior analysis reports based on the analysis results of user data; Use the trained deep learning network architecture to predict future environmental parameters, device status, or user behavior and obtain model prediction output.

4. The AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services according to claim 1, characterized in that: Based on the environmental data analysis results, model prediction output and user behavior analysis report, a comprehensive status evaluation of the building complex service is obtained through fuzzy synthesis operation, and at least one control strategy for the building complex is generated based on the comprehensive status evaluation, including: Align and integrate environmental data analysis results, model prediction outputs, and user behavior analysis reports in time and space dimensions, and eliminate information redundancy and conflict based on a predefined set of rules; Analyze the data after eliminating information redundancy and conflict, and select evaluation indicators that can reflect the environmental status of the building complex, predict trends and user needs to build a comprehensive status indicator system; The fuzzy evaluation matrix and weight vector determined by integrating the evaluations of all experts are used to perform fuzzy synthesis operations on the data in the comprehensive status indicator system to generate a comprehensive status assessment of the building complex services; Analyze the control objectives and control requirements of the building complex based on the comprehensive status assessment of the building complex services; Construct a reinforcement learning model and use the overall state of the building complex services as the state space of the reinforcement learning environment. Integrate one or more intelligent agents in the reinforcement learning environment. Each intelligent agent can perceive the state of the environment and make decisions based on the learned strategy. According to the control objectives and control requirements of the building complex, combined with predefined performance indicators, the operation process of the building complex service is simulated in the constructed environment model, so that the intelligent agent can perceive the environmental state and select control actions according to the current state. After executing the action, the change of the environmental state and the reward signal obtained are observed, and the control strategy is continuously adjusted according to the received reward signal, with the goal of maximizing the cumulative reward. By allowing the agent to continuously interact and trial and error in the simulated environment, the agent's strategy is gradually converged to obtain at least one control strategy; The rule set includes data validation rules, redundancy detection rules, and conflict detection and resolution rules: Data validation rules include: checking whether the data is within a threshold range derived from at least one statistical calculation; cross-validating the data format using statistical validation, logical validation, expert review, and preset format templates; Redundancy detection rules include: if two data sources provide the same type of data and the timestamp difference is within a preset time window, it is considered redundant data; for redundant data, the data source with the latest timestamp is selected as the valid data; The conflict detection and resolution rules include: building an association graph model based on data provided by multiple data sources, traversing each node and edge of the association graph model, and triggering conflict detection when differences are found that contain inconsistent data values, contradictory association relationships, or abnormal data patterns; when a conflict occurs, recording the detected conflicting nodes and edges, and performing self-processing of data correction, merging, or deletion, feeding back the results of the conflict resolution to the association graph model, updating the status of the nodes and edges, and if the same problem still exists in the next round, requesting manual intervention.

5. The AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services according to claim 1, characterized in that: After obtaining a comprehensive status assessment and generating a control strategy for the building complex based on environmental data analysis results, model prediction output, and user behavior analysis reports, it also includes: Create a digital twin building complex model that can simulate the state of the building complex under various conditions in real time; Mapping the generated control strategies to the control instructions of the building complex; Apply the mapped control instructions to the digital twin building complex model to observe and track the environmental parameter change information, equipment status change information and user feedback information within the building complex; The environmental parameter change information, equipment status change information and user feedback information within the building complex are fed into the constructed reinforcement learning model to adaptively optimize the control strategy and obtain the optimal control strategy.

6. The AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services according to claim 1, characterized in that: Build an online sharing platform between building clusters, and use transfer learning to migrate the control strategy of the building cluster generated on the current building cluster to other building clusters, so that different building clusters can exchange and share the learned models, control strategies and optimization experience, including: Build an online sharing platform among building complexes based on cloud computing platform; In response to a migration instruction input by a user, determining a source building complex and a target building complex; Analyze the control action sets of the source and target building complexes, identify common actions and specific actions, directly map common actions, and cut or expand specific actions according to the control requirements of the target building complex to ensure that the migrated strategy can adapt to the action space of the target building complex; Use the policy function learned on the source building cluster as the starting point to build the basis of the migration strategy; On the target building complex, adjust the policy function or continue to learn based on the migration formula and the set optimization goal. During the migration process, continuously evaluate the performance of the migrated policy on the target building complex. According to the evaluation results received in each round, adjust the parameters in the migration formula and repeat multiple times until the preset migration performance requirements or the maximum number of iterations are met; The migration formula is: ； In the formula, J m is the total return of the migration process, T b is the total number of time steps of the migration process, τ is the discount factor used to balance the importance of immediate returns and future returns, In the target building complex T BG Medium status s t Take action a t Immediate returns, λ is the state difference weight, which is used to adjust the impact of state space differences on the migration process. Is the source complex status The target building status The difference measure between .

7. An AI-driven online intelligent IoT monitoring and evaluation system for building complex services, characterized in that: include: A collection and processing module for collecting and preprocessing environmental parameters, device status, and user data through a dynamically configurable sensor network and user interface pre-deployed in the building complex; A cluster analysis module, for performing cluster analysis on at least one of the preprocessed environmental parameters, device status, and user data using a dynamic Gaussian mixture model to identify and divide multiple clusters or groups each representing a specific comfort zone, including: selecting at least one of the environmental parameters, device status, and user data as an object to be analyzed; determining the number of Gaussian components in the Gaussian mixture model corresponding to the number of clusters or groups formed K , and initialize the parameter configuration of the Gaussian mixture model; divide the object to be analyzed into multiple continuous time windows, move the time window according to the set step size to cover the entire time series of the object to be analyzed, and each new time window contains a part of new data and a part of data overlapping with the previous time window; in each new time window, use the initialized Gaussian mixture model to fit the object to be analyzed, and use the expectation maximization method to alternately perform the expectation step and the maximization step until the preset number of iterations is reached or the model parameter change is less than the threshold, wherein in the expectation step, the responsibility of each data point belonging to each Gaussian component is calculated according to the current model parameters, and in the maximization step, the parameter configuration of the Gaussian mixture model is updated according to the responsibility; through the fitting process, the data points of the object to be analyzed are divided into K clusters or groups, and determine the clusters or groups representing specific comfort zones corresponding to the data points in each time window; wherein the degree of responsibility is: ； In the formula, z pq For the p Data points for q The responsibility of the Gaussian component, w q It is q The weights of the Gaussian components, σ q It is q The standard deviation of the Gaussian components, x p It is p data points, μ q It is q The mean of the Gaussian components, For the p Data points and q The square of the Euclidean distance between the means of the Gaussian components, K is the total number of Gaussian components, w k It is k The weights of the Gaussian components, σ k It is k The standard deviation of the Gaussian components, μ k It is k The mean of the Gaussian components, χ is the sharpening parameter, which is an adjustable parameter used to control the sharpening degree of responsibility calculation; Deep learning module, used to process clusters or groups through deep learning network analysis, obtain environmental data analysis results, model prediction output and user behavior analysis report; A state assessment and strategy output module is used to obtain a comprehensive state assessment of building complex services through fuzzy synthesis operations based on environmental data analysis results, model prediction outputs, and user behavior analysis reports, and to generate at least one control strategy for the building complex based on the comprehensive state assessment; The strategy sharing module is used to build an online sharing platform between building complexes. It uses transfer learning to migrate the control strategy of the building complex generated on the current building complex to other building complexes, so that different building complexes can exchange and share the learned models, control strategies and optimization experiences.

8. An AI-driven online intelligent IoT monitoring and evaluation device for building complex services, characterized in that: include: A cloud computing platform is used to connect multiple building clusters into a network and use transfer learning to achieve data sharing and collaboration between building clusters; A central control platform for the AI-driven online intelligent IoT monitoring and evaluation method for building complex services as described in any one of claims 1 to 6; as well as, The database is connected to the central control platform and is used to store the monitoring data of the building complex, the constructed models, control strategies and optimization experience.

9. A computer-readable medium having computer-executable instructions stored thereon, characterized in that: When the executable instructions are executed by the processor, the AI-driven online intelligent Internet of Things monitoring and evaluation method for building complex services is implemented as described in any one of claims 1-6.

Images