AI-based intelligent control method and system for building automation system

Abstract

The present invention provides an AI-based intelligent control method and system for a building automation system. The method performs feature space conversion on the state variable monitoring data of the building automation system through an adaptive learning network model to generate a first state feature space that matches the feature pattern extracted by the building state recognition network. This method can guide the building automation system control network to make adaptive control decisions on multiple different fault disturbance events based on the first state feature space, and generate multiple adaptive control decision results associated with the building operation category. It is not only used to perform actual adaptive control on the building automation system, but also further used to optimize the building state recognition network, thereby forming a closed-loop intelligent control and optimization system. This method improves the building automation system's ability to respond to fault disturbance events and the control accuracy, enhances the system's adaptability and intelligence level, and effectively improves the building's operating efficiency and energy utilization efficiency.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an AI-based intelligent control method and system for a building automation system. Background Art

[0002] With the acceleration of urbanization, buildings, as an integral part of cities, are attracting increasing attention for their energy consumption and operational efficiency. Building automation systems, the core of building management, are responsible for monitoring and regulating various building facilities, such as air conditioning, lighting, and elevators, to ensure proper operation and reduce energy consumption and emissions. However, traditional building automation systems often rely on fixed control logic and preset parameters, making them difficult to adapt to the complex and changing conditions of actual building operations.

[0003] During building operations, various disturbances and failures occur frequently, such as equipment failures, environmental changes, and personnel turnover. These events can affect the building's status and, in turn, the effectiveness of the building's automatic control system. Traditional building automatic control systems lack the ability to adaptively respond to these disturbances and failures, often requiring manual intervention or adjustment of control parameters after an event occurs. This not only increases management complexity but can also lead to wasted energy and reduced building operational efficiency.

[0004] In recent years, with the rapid development of artificial intelligence (AI), the application of AI to building automation systems has become a new trend. By leveraging AI's powerful learning and adaptive capabilities, intelligent control of building automation systems can be achieved, improving the system's adaptability and intelligence. However, existing AI-based intelligent control methods for building automation systems often suffer from issues such as inaccurate feature extraction and inflexible control decisions, which limit their effectiveness in practical applications. Summary of the Invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides an AI-based intelligent control method for a building automation system, the method comprising:

[0006] Obtain state variable monitoring data of the building automation system;

[0007] The adaptive learning network model performs feature space conversion on the state variable monitoring data to generate a first state feature space that matches a first feature pattern, where the first feature pattern is a feature pattern matched by the state feature space extracted by the building state recognition network;

[0008] Based on the first state feature space, guiding the building automatic control system control network to perform adaptive control decisions on fault disturbance propagation for multiple different fault disturbance events, generating multiple different adaptive control decision results for building operation categories associated with the first state feature space, and the multiple different adaptive control decision results for building operation categories associated with the first state feature space are also used to optimize the building state recognition network;

[0009] Based on a plurality of different adaptive control decision results of the building operation category associated with the first state feature space, an adaptive control decision is performed on the building automatic control system.

[0010] On the other hand, an embodiment of the present invention also provides an AI-based building automatic control system intelligent control system, including a processor and a machine-readable storage medium, the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiment of the present application performs feature space conversion on the state variable monitoring data of the building automatic control system through an adaptive learning network model, and generates a first state feature space that matches the feature pattern extracted by the building state recognition network. This method can guide the building automatic control system control network to make adaptive control decisions on multiple different fault disturbance events based on the first state feature space, and generate multiple adaptive control decision results associated with the building operation category. These decision results are not only used to perform actual adaptive control on the building automatic control system, but are also further used to optimize the building state recognition network, thereby forming a closed-loop intelligent control and optimization system. This method improves the building automatic control system's response capability and control accuracy to fault disturbance events, enhances the system's adaptability and intelligence level, and effectively improves the building's operating efficiency and energy utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the execution flow of the AI-based building automatic control system intelligent control method provided by an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of the hardware architecture of the AI-based building automation system intelligent control system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1This is a flow chart of an AI-based intelligent control method for a building automation system provided by an embodiment of the present invention. The AI-based intelligent control method for a building automation system is introduced in detail below.

[0015] Step S110: Acquire state variable monitoring data of the building automation system.

[0016] In this embodiment, in a building automation system of a large commercial office building, the acquisition of state variable monitoring data is the basis for achieving effective control and management. The building automation system covers many subsystems, such as the HVAC system, lighting system, elevator system, and security system. For the HVAC system, the state variable monitoring data includes the temperature values ​​collected by the temperature sensors in each area. These temperature values ​​reflect the real-time temperature conditions of different floors, different rooms, or different office areas. For example, in the lobby on the first floor of an office building, the temperature sensor collects temperature data at regular intervals (such as 5 minutes). The temperature in the lobby may be lower before the morning rush hour, and the temperature gradually rises as people enter and equipment is turned on.

[0017] Humidity sensors in air conditioning systems also provide humidity monitoring data. Humidity levels vary significantly depending on the season and weather conditions. Humidity can be high in humid summer weather, but low in dry winter weather. Furthermore, operating parameters such as fan speed and water valve opening in the air conditioning system are monitored as state variables. Fan speed affects the air circulation rate, while water valve opening determines the flow of cold or hot water, which in turn affects the cooling or heating effect.

[0018] In terms of the lighting system, light sensors are installed on each floor, in each room, or in each public area to collect light intensity data. During the day, areas near windows may have higher light levels due to ample natural light, while interior areas away from windows may require more artificial lighting to meet the needs of work or activities. At the same time, the on / off status and dimming ratio of the lamps in the lighting system are also monitored state variables. For example, some conference rooms may need to adjust the brightness and on / off combination of lights according to different meeting scenarios (such as projection presentations or group discussions).

[0019] In elevator systems, important state variables to monitor include the elevator's operating speed, the number of floors it stops at, the number of passengers in the car, and the door's open / close status. For example, during peak hours, when elevators frequently stop at various floors and carry a large number of passengers, the operating speed may be adjusted appropriately based on the elevator's load to ensure safe and efficient operation.

[0020] Security systems include access control systems, surveillance cameras, and other equipment. Access control systems record data such as the opening and closing times of each door and the results of personnel access authorization verification. Surveillance cameras provide video image data from each monitored area. Although video images themselves are complex data, image analysis techniques can be used to extract state variables such as personnel flow direction and number of personnel. This data helps managers understand personnel activity within the building, detect abnormal behavior promptly, and take appropriate countermeasures.

[0021] By acquiring various state variable monitoring data from different subsystems, the building automatic control system can fully understand the operating status of the building and provide data support for subsequent analysis, decision-making and regulation.

[0022] In step S120 , the adaptive learning network model performs feature space conversion on the state variable monitoring data to generate a first state feature space matching a first feature pattern, where the first feature pattern is a feature pattern matched by the state feature space extracted by the building state recognition network.

[0023] Continuing with the example of the building automation system of the commercial office building, it is assumed that the adaptive learning network model has been pre-trained and has the ability to perform feature space transformation on different types of state variable monitoring data.

[0024] For the HVAC system's state variable monitoring data, the adaptive learning network model first analyzes the temperature data. Specifically, it not only focuses on the specific temperature values, but also considers factors such as the temperature trends over different time periods and the correlation between temperatures in different areas. For example, in an office area, if the temperatures of several adjacent rooms show similar upward or downward trends, this may indicate the presence of common influencing factors, such as the overall operating status of the air conditioning unit or the influence of the external climate. The model integrates this characteristic information about temperature, converting it from the original numerical space into a new feature space. This new feature space can better reflect the state characteristics of temperature in the entire building's air conditioning system. For example, it maps the temperature value to a feature space that represents temperature stability, temperature uniformity, and the relationship between temperature and comfort.

[0025] For humidity data, the adaptive learning network model combines temperature data with information such as the building's structure and occupancy density to transform the feature space. For example, in a well-sealed office area with a high occupancy density, humidity fluctuations may be closely related to the moisture exhaled by occupants and the dehumidification capacity of the air conditioning system. The model then links humidity data with these related factors to construct a feature space that encompasses the relationships between humidity and other factors. This feature space may reveal characteristics such as the impact of humidity on building comfort and equipment operation under different environmental conditions.

[0026] When processed by the adaptive learning network model, the illuminance data in the lighting system is combined with factors such as time of day (e.g., whether it is daytime or nighttime), human activity information (e.g., whether a meeting is scheduled in a certain area), and the building's orientation. For example, in a south-facing conference room on a sunny afternoon with ample natural light, the high illuminance data collected by the illuminance sensor may be associated with the conference room's planned use (whether a projection presentation is required) in the converted feature space. If a projection presentation is required, the high illuminance may be marked as a state feature that requires adjustment, while if there is no projection presentation required, the high illuminance may be considered a normal energy-saving state feature.

[0027] When transforming the elevator system's state variable monitoring data into a feature space, the adaptive learning network model considers the degree of alignment between the elevator's operating patterns and passenger flow patterns. For example, during rush hour, frequent elevator stops and high passenger loads are normal operating characteristics. The model transforms these data, such as operating speed, floors visited, and passenger loads, into a feature space that reflects elevator service efficiency, operational stability, and alignment with passenger demand. If an elevator frequently malfunctions or experiences abnormal operating speeds during peak hours, these characteristics will be highlighted in the transformed feature space for subsequent analysis and decision-making.

[0028] The adaptive learning network model combines access control data from the security system and personnel flow data obtained through video image analysis with factors such as the building's functional zoning and the security levels of different areas to perform feature space transformation. For example, in the core office area of ​​an office building, the strict verification results of the access control system and the low level of personnel flow may be considered normal security status characteristics. In public areas such as lobbies or restaurants, however, the frequent flow of personnel is a normal activity status characteristic. Through this feature space transformation, the adaptive learning network model can generate a first-state feature space that matches the first feature pattern. This first-state feature space can comprehensively and accurately reflect the characteristic relationships between the various subsystems of the building under the overall operating state, and is consistent with the feature pattern matched by the state feature space extracted by the building state recognition network, providing a suitable data foundation for subsequent fault analysis and control decisions.

[0029] Step S130: Based on the first state feature space, guide the building automatic control system control network to make adaptive control decisions on fault disturbance propagation for multiple different fault disturbance events, and generate multiple different adaptive control decision results for the building operation category associated with the first state feature space. The multiple different adaptive control decision results for the building operation category associated with the first state feature space are also used to optimize the building state recognition network.

[0030] Assume that multiple different fault disturbance events occur in the building automation system of a commercial office building.

[0031] For HVAC systems, a potential disturbance event is a sudden drop in the chiller's cooling efficiency. Based on the first-state feature space, which contains characteristic information such as temperature and humidity in each area, as well as air conditioning equipment operating parameters, the decision-making optimization unit in the building automation system's control network begins making adaptive control decisions to address the propagation of the disturbance.

[0032] In the first iteration, the decision optimization unit uses the first-state feature space as the basis for fault diagnosis, obtaining relevant information about the fault disturbance event, namely, the chiller's reduced cooling efficiency. This includes information such as the type of chiller and the air conditioning terminal equipment (such as fan coil units) that the fault may affect. It also initializes coefficients used to measure the likelihood of fault propagation. For example, based on the connection between the chiller and the air conditioning terminal equipment and previous fault experience, the probability coefficient of a chiller fault propagating to a specific fan coil unit is set to 0.3. The initial values ​​of indicators used to evaluate the effectiveness of control decisions are also initialized. For example, the initial evaluation value for keeping the indoor temperature within the comfortable range is set to 0.8. Based on the chiller's initial state information, the fault's propagation path within the current building automation system environment was analyzed. The researchers found that a decrease in the chiller's cooling efficiency would cause the temperature of the chilled water transported through the pipes to increase, thereby affecting the cooling efficiency of the connected fan coil units. The connections between the various components in the air conditioning system indicate that chilled water sequentially affects the fan coil units on each floor along the pipes. Data flows from the chiller to each fan coil unit, and the interaction mechanism is that changes in chilled water temperature directly affect the heat exchange capacity of the fan coil units. The researchers determined that the potential direction of propagation of the fault disturbance event from the chiller to the fan coil units is along the pipes, with the likelihood of this occurring depending on factors such as pipe distance and insulation. Based on these analysis results, preliminary control decision data was generated, such as increasing the chiller's compressor operating frequency or adjusting the water valve opening of some fan coil units to increase chilled water flow, in an attempt to initially suppress or adjust the propagation of the fault disturbance. The preliminary control decision data is applied to the current state information of the fault disturbance event to obtain the fault disturbance state information after the first round of iteration. At this time, the compressor operating frequency of the chiller increases, and the water valve opening of some fan coil units changes, reflecting the state change of the fault disturbance event after the implementation of the preliminary control decision.

[0033] From the second iteration to the X-1 iteration (assuming X = 5), the decision optimization unit re-evaluates the propagation of the fault disturbance event in the building automation system based on the fault disturbance state information after the previous iteration. For example, in the second iteration, it was found that although the compressor operating frequency of the chiller was increased, the cooling efficiency did not improve significantly. This may be due to scaling problems in a component of the chiller (such as the condenser). Therefore, based on the new propagation analysis results, the previous control decision data was adjusted, such as adding a cleaning operation to the condenser and further adjusting the water valve opening of the fan coil unit. After adjusting the control decision data, the new control decision data was applied to the current fault disturbance state information to obtain the fault disturbance state information after the current iteration. In the third iteration, it was found that the humidity in some areas had changed abnormally due to the adjustment of the water valve opening of the fan coil unit. Therefore, humidity adjustment measures were added to the control decision, such as turning on the dehumidification equipment in some areas. In the fourth round of iteration, the control decisions were further optimized based on the changes in fault disturbance status information. For example, the water valve opening of the fan coil unit and the operating time of the dehumidification equipment were adjusted more finely according to the actual temperature and humidity requirements of each area.

[0034] In the Xth iteration (the fifth iteration), the decision optimization unit extracts final fault diagnostic features based on the fault disturbance state information after the X-1th iteration. This analysis determines the final propagation range of the fault disturbance throughout the building automation system, affecting the air conditioning comfort levels on multiple floors. It also determines the ultimate impact on key components (such as chillers, fan coil units, and dehumidifiers). It finds that while the chiller's cooling efficiency has improved, it has not yet returned to normal. The fan coil unit's operating parameters have stabilized after multiple adjustments. The activation of the dehumidifier has had some effect on humidity control, but further optimization is required. Furthermore, no other chain reactions have been triggered, such as impacts on the elevator or lighting systems. Based on these analysis results, feature information reflecting the adaptive control decision-making outcome of the fault disturbance propagation during the Xth iteration is extracted to obtain the target fault diagnostic features.

[0035] The execution strategy unit receives the target fault diagnosis feature from the decision optimization unit. This feature contains fault disturbance propagation information optimized over multiple iterations, as well as the current configuration of the building automation system (BACS), including connectivity, functional attributes (such as cooling capacity and air volume), and operating parameter thresholds (such as temperature and humidity comfort ranges) of system components such as chillers and fan coil units. The execution strategy unit analyzes the fault propagation path, impact, and potential chain reactions in the target fault diagnosis feature, identifies the chillers and fan coil units directly affected by the fault, and the dehumidifiers indirectly affected. It prioritizes these identified affected components and generates a fault impact assessment report and a component priority list. Based on this report and component priority list, it searches for matching control strategies in a pre-built control strategy library. For example, in the case of decreased chiller cooling efficiency, it may identify control strategies such as adding auxiliary cooling equipment or optimizing chiller operating parameters. Based on the matching results, it generates a corresponding control solution. Using the BACS simulation environment or digital twin model, it simulates and tests the control solution and generates a simulation test report. Simulation tests revealed that adding auxiliary cooling equipment, while increasing cooling capacity, also increased energy consumption. Consequently, the control scheme was optimized, such as adjusting the startup conditions and operating hours of the auxiliary cooling equipment, to generate a final control decision. This final control decision was then passed to the building automation system's execution layer, which then guided the system to implement control operations based on the final control decision. This included starting the auxiliary cooling equipment according to the optimized time and conditions, and adjusting the operating parameters of the chiller and fan coil units.

[0036] Similarly, for potential fault disturbances that may occur in the lighting system, such as a faulty illuminance sensor in a certain area leading to abnormal collected data, the decision optimization unit begins making adaptive control decisions based on the relevant characteristic information of the lighting system in the first-state feature space. During the iterative process, the fault propagation path is first analyzed. A faulty illuminance sensor may affect the decisions of the connected lighting control system, thereby affecting the lighting effect in the area. After initializing the relevant parameters, preliminary control decision data is generated, such as switching to a backup illuminance sensor or making preliminary adjustments to the light brightness based on the area's historical lighting data, the current time, and personnel activities. After multiple rounds of iteration, the target fault diagnosis features are ultimately obtained. The execution strategy unit generates a control plan based on the target fault diagnosis features, such as repairing or replacing the faulty illuminance sensor, while optimizing the lighting control strategy for the area to ensure that the lighting effect meets the requirements.

[0037] The adaptive control decision results for these different fault disturbance events can also be used to optimize the building status recognition network. For example, when a chiller fails, the data accumulated during the adaptive control decision-making process on the relationship between state characteristics such as temperature, humidity, and equipment operating parameters and the fault can be fed back to the building status recognition network. Based on this data, the building status recognition network can adjust its recognition mode for fault state characteristics, improving the accuracy and timeliness of identifying similar faults.

[0038] Step S140 : performing adaptive control decision on the building automatic control system based on a plurality of different adaptive control decision results of the building operation category associated with the first state feature space.

[0039] Taking the building automation system of a commercial office building as an example, the multiple adaptive control decision results of the building operation categories associated with the first state feature space cover various aspects such as the HVAC system, lighting system, elevator system, and security system.

[0040] For the HVAC system, the building automation system makes overall adaptive control decisions based on previously generated adaptive control decisions for disturbances such as chiller cooling efficiency decline. For example, if multiple adaptive control decisions consistently reveal that the air conditioning terminal equipment (fan coil units) on certain floors are experiencing difficulty regulating or excessive energy consumption, the overall adaptive control decision may consider local upgrades to the air conditioning systems on those floors, such as replacing more efficient fan coil units or optimizing the piping layout. Furthermore, the overall HVAC system's operational strategy is adjusted based on the adaptive control decision results for different seasons and time periods. During high summer temperatures, the number of operating chillers can be increased or the cooling capacity setpoints for certain areas can be raised based on the adaptive control decision results' balance between cooling efficiency and energy consumption. During low winter temperatures, the hot water supply system's operating parameters, such as adjusting the hot water boiler's outlet water temperature and the operating frequency of the circulation pump, can be optimized based on analysis of heating demand and energy savings.

[0041] In terms of lighting systems, comprehensive adaptive control decisions are made based on the adaptive control decision results for fault disturbance events such as illuminance sensor failure and uneven brightness in the lighting area. If it is found in multiple adaptive control decision results that the lighting in certain public areas (such as corridors and stairwells) is frequently turned on and off or the brightness is excessively adjusted, the lighting control strategies for these areas may be replanned. For example, intelligent lighting algorithms can be used to more accurately control the switching and brightness adjustment of lights based on personnel activities and natural lighting conditions. For office areas, personalized lighting plans are formulated based on the working hours and lighting needs of different departments, combined with the lighting energy consumption analysis in the adaptive control decision results, such as setting different dimming schedules or adopting zoned lighting control strategies to improve the overall energy efficiency of the lighting system.

[0042] For elevator systems, adaptive control decisions are made based on the results of adaptive control decisions for fault disturbances such as abnormal elevator speeds and car overload alarms. If multiple adaptive control decision results indicate that certain elevators have low operating efficiency during peak hours, the elevator scheduling algorithm may be adjusted. For example, a strategy could be added to prioritize calls from upper or lower floors, or the elevator's stopping strategy could be adjusted to reduce unnecessary stops at different floors. Furthermore, based on the operating status and maintenance needs of the elevator equipment, combined with equipment fault analysis from the adaptive control decision results, the elevator maintenance plan can be optimized, such as performing preventive maintenance on frequently failing components and adjusting maintenance cycles.

[0043] In the security system, adaptive control decisions are made based on the results of adaptive control decisions for fault disturbances such as access control system misjudgments and blurred surveillance camera images. If multiple adaptive control decision results reveal a high misjudgment rate for the access control system at a particular entrance or exit, the access control equipment may be upgraded or the access control verification algorithm adjusted. For surveillance cameras, adaptive control decisions based on fault conditions such as blurred images can be used to clean the cameras, adjust their focus, or replace aging components. Furthermore, based on the overall operation of the security system and the security risk analysis derived from the adaptive control decision results, the security monitoring layout and strategy are optimized, such as increasing the number of surveillance cameras in key areas or adjusting the monitoring range, thereby enhancing the overall building's security capabilities. By comprehensively considering multiple adaptive control decision results associated with the first-state feature space, the building automation system can achieve more intelligent, efficient, and stable operation, improving building comfort, safety, and energy efficiency.

[0044] Based on the above steps, the embodiment of the present application performs feature space conversion on the state variable monitoring data of the building automatic control system through an adaptive learning network model, and generates a first state feature space that matches the feature pattern extracted by the building state recognition network. This method can guide the building automatic control system control network to make adaptive control decisions on multiple different fault disturbance events based on the first state feature space, and generate multiple adaptive control decision results associated with the building operation category. These decision results are not only used to perform actual adaptive control on the building automatic control system, but are also further used to optimize the building state recognition network, thereby forming a closed-loop intelligent control and optimization system. This method improves the building automatic control system's response capability and control accuracy to fault disturbance events, enhances the system's adaptability and intelligence level, and effectively improves the building's operating efficiency and energy utilization efficiency.

[0045] In a possible implementation, the building automation system control network includes a decision optimization unit and an execution strategy unit.

[0046] Step S130 includes:

[0047] Step S131: For each of the multiple different fault disturbance events, the decision optimization unit uses the first state feature space as a fault diagnosis basis to perform adaptive control decisions of fault disturbance propagation for X iteration cycles on the fault disturbance event, generating a target fault diagnosis feature obtained by the adaptive control decision of fault disturbance propagation in the Xth iteration cycle, and using the fault diagnosis feature obtained by the adaptive control decision of fault disturbance propagation in the ath iteration cycle as input to the adaptive control decision of fault disturbance propagation in the a+1th iteration cycle, where a and X are positive integers and a<X.

[0048] Step S132 : performing adaptive control decision of fault disturbance propagation on the target fault diagnosis feature through the execution strategy unit to generate an adaptive control decision result of the building operation category associated with the first state feature space.

[0049] In this example, let's continue with the example of a chiller failure in the HVAC system, a fault disturbance event. The decision optimization unit will initiate fault processing based on the first state feature space as the basis for fault diagnosis. This first state feature space contains various state feature information related to the HVAC system, such as the temperature and humidity of the entire building, and the operating parameters of each air conditioning device. Assuming X is 5, the iteration cycle begins.

[0050] In the first iteration, the decision optimization unit comprehensively analyzes the chiller failure. A chiller failure could result in abnormal chilled water supply temperatures, impacting the cooling performance of the entire building. Based on temperature data in the first state feature space, such as the current and historical temperature trends for each floor and room, as well as information about the connectivity between air conditioning system components, the decision optimization unit initializes the chiller failure propagation probability coefficient. For example, the probability coefficient of a chiller failure propagating to nearby air conditioning terminal equipment (such as fan coil units) is determined to be 0.3. The unit also initializes the initial values ​​of the control decision effectiveness evaluation indicators, such as setting the initial evaluation value for maintaining a comfortable indoor temperature range to 0.8. Based on the chiller's initial state information, such as the chiller's current cooling efficiency and chilled water flow rate, the unit analyzes the fault propagation path within the current building automation system environment. Because the chiller is connected to multiple fan coil units via pipes, a chiller failure causes the chilled water temperature to rise, potentially affecting the fan coil units on each floor along the pipes. This potential direction of fault propagation is determined, and its likelihood is related to factors such as the pipe insulation and distance. Based on these analysis results, preliminary control decision data is generated, such as increasing the chiller compressor operating frequency and adjusting the water valve opening of some fan coil units to increase the chilled water flow and suppress the propagation of the fault. This preliminary control decision data is applied to the current fault state information to obtain the fault disturbance state information after the first iteration cycle. At this time, the chiller compressor operating frequency is changed, and the water valve opening of some fan coil units is adjusted, reflecting the change in the fault state after the implementation of the preliminary control decision.

[0051] In the second iteration, the decision optimization unit uses the fault disturbance state information from the previous iteration as its basis. It was discovered that despite increasing the compressor operating frequency, the chiller's cooling efficiency did not improve significantly. Further analysis revealed that scaling in the chiller's condenser might be affecting heat exchange efficiency. The fault propagation situation was reassessed, and the control decision data was adjusted based on the new analysis results. Condenser cleaning was added, and the fan coil unit water valve opening was further refined. After these adjustments, the new control decision data was applied to the current fault disturbance state information to obtain the fault disturbance state information after the current iteration.

[0052] In the third iteration, the decision optimization unit continued to base its analysis on the fault disturbance state information from the previous round. It discovered that the humidity in some areas had changed abnormally due to adjustments to the fan coil unit water valve opening. The unit then reassessed the fault propagation and added humidity control measures to its control decisions, such as turning on dehumidification equipment in some areas. The new control decision data was then applied to the current fault disturbance state information, resulting in the fault disturbance state information for the entire iteration.

[0053] During the fourth iteration, the decision optimization unit continues to operate based on the fault disturbance state information from the previous round. At this point, it may be discovered that while the dehumidification system has a certain effect on humidity regulation, the chiller's cooling efficiency has not yet fully returned to normal, and the temperature on some floors has not yet reached a comfortable range. Therefore, further optimization of the control decisions is performed. For example, the fan coil unit water valve opening and dehumidification system operating hours are more accurately adjusted based on the actual temperature and humidity requirements of each area. The new control decision data is then applied to the current fault disturbance state information to obtain the fault disturbance state information after this iteration.

[0054] By the fifth iteration, the decision optimization unit extracts final fault diagnostic features based on the fault disturbance state information after the fourth iteration. Analyzing the current fault disturbance state information, it determines that the fault disturbance's final propagation range throughout the building automation system has affected air conditioning comfort levels on multiple floors. The chiller's cooling efficiency has improved but not fully recovered. The fan coil unit's operating parameters have stabilized after multiple adjustments. The dehumidification equipment has a certain effect on humidity control but still requires optimization. Furthermore, the system has not triggered a chain reaction in other systems, such as the elevator and lighting systems. Based on these analysis results, feature information reflecting the adaptive control decision-making results of the fault disturbance propagation in the fifth iteration is extracted to obtain the target fault diagnostic features.

[0055] The execution strategy unit receives the target fault diagnosis feature from the decision optimization unit. This target fault diagnosis feature includes fault disturbance propagation information optimized over five iterative cycles, as well as the current configuration of the building automation system (BACS). This includes the connectivity, functional attributes (such as cooling capacity and air volume), and operating parameter thresholds (such as temperature and humidity comfort ranges) of system components such as chillers and fan coil units. The execution strategy unit analyzes the fault propagation path, impact, and potential chain reactions in the target fault diagnosis feature, identifies the chillers and fan coil units directly affected by the fault, and the dehumidifiers indirectly affected. It prioritizes these affected components and generates a fault impact assessment report and a component priority list. Based on this report and component priority list, the execution strategy unit searches for matching control strategies in a pre-built control strategy library. For example, in the case of decreased chiller cooling efficiency, it may identify control strategies such as adding auxiliary cooling equipment or optimizing chiller operating parameters. Based on the matching results, the execution strategy unit generates a corresponding control solution. The control solution is simulated and tested using the BACS simulation environment or digital twin model, generating a simulation test report. During the simulation test, it was found that although adding auxiliary refrigeration equipment can increase the cooling capacity, it will also increase energy consumption. Therefore, the control plan was optimized, such as adjusting the start-up conditions and operating time of the auxiliary refrigeration equipment, generating the final control decision result, and passing the final control decision result to the execution layer of the building automatic control system to guide the building automatic control system to implement control operations according to the final control decision result, such as starting the auxiliary refrigeration equipment according to the optimized time and conditions, and adjusting the operating parameters of the chiller and fan coil unit.

[0056] Similarly, for a fault disturbance event such as a failure of an illuminance sensor in a certain area of ​​the lighting system, the decision optimization unit begins iteration based on the lighting-related state features in the first state feature space. In each iteration cycle, the fault propagation path is analyzed. For example, the failure of the illuminance sensor affects the decision-making of the lighting control system connected to it, thereby affecting the lighting effect of the area. Each iteration cycle adjusts the control decision data based on the fault disturbance state information of the previous cycle, such as first switching to a backup illuminance sensor or preliminarily adjusting the light brightness based on the historical lighting data of the area, the current time, and personnel activities. After obtaining the target fault diagnosis features through multiple iteration cycles, the execution strategy unit generates a control plan based on it, such as repairing or replacing the faulty illuminance sensor, and optimizing the lighting control strategy of the area to ensure that the lighting effect meets the requirements, etc., to generate a series of operations to generate an adaptive control decision result associated with the first state feature space.

[0057] In a possible implementation, before step S130, the method further includes:

[0058] Step A110: Obtain first template state variable monitoring data, a first template state feature space corresponding to the first template state variable monitoring data, and a set number of iteration cycles b determined for the first template state variable monitoring data. The first template state feature space is generated by performing feature space conversion on the first template state variable monitoring data using a sample building state recognition network. The sample building state recognition network is the building state recognition network or another neural network model with the same network task as the building state recognition network. b is a positive integer not greater than X.

[0059] In this embodiment, the first template state variable monitoring data originates from various subsystems of the building automation system, such as the HVAC system, lighting system, elevator system, and security system mentioned above. For example, for the HVAC system, the first template state variable monitoring data includes chiller operating parameters, such as the compressor speed, condenser and evaporator temperatures, and inlet and outlet water temperatures. It also includes status data for air conditioning terminal equipment such as fan coil units, such as fan speed and water valve opening. The first template state variable monitoring data for the lighting system includes illuminance values ​​collected by illuminance sensors in each area, the on / off status of lamps, and dimming ratios. For the elevator system, the first template state variable monitoring data includes data such as elevator speed, car position, passenger capacity, and door open / close status. The first template state variable monitoring data for the security system includes personnel entry and exit records from the access control system and personnel flow information from processed video data from surveillance cameras. Simultaneously, a first template state feature space corresponding to the first template state variable monitoring data is obtained. This feature space is generated by performing feature space transformation on the first template state variable monitoring data using the example building state recognition network. The sample building status recognition network shares the same network tasks as the actual building status recognition network, including identifying the overall building operating status and analyzing the relationships between subsystems. The number of iterations, b, is also determined. b is a positive integer not greater than X. For example, if X is 5, b can be 3.

[0060] Step A120: The feature conversion module performs feature conversion on the first template state variable monitoring data to generate a first conversion result.

[0061] The feature conversion module performs specific conversion operations for different types of data. For HVAC system data, the feature conversion module integrates and converts various operating parameters of the chiller. For example, data such as compressor speed, condenser, and evaporator temperatures are converted into features that reflect the overall performance and health of the chiller. By analyzing multiple factors, such as the relationship between compressor speed and condenser temperature, and the impact of inlet and outlet water temperatures on overall cooling efficiency, a comprehensive feature representation is generated, which forms part of the first conversion result. For the lighting system, the feature conversion module converts information such as illuminance values ​​and lamp status into features related to lighting comfort and energy efficiency. For example, by combining illuminance values ​​with lamp dimming ratios, a feature is generated that represents the lighting quality and energy efficiency of the area. Elevator system data, after being processed by the feature conversion module, generates features related to elevator service quality and operational stability. For example, based on elevator speed and cabin passenger capacity, a feature is converted to reflect the elevator's operating efficiency under different loads. The data from the security system is converted into features related to personnel safety management and regional security situation. For example, based on the personnel entry and exit records of the access control system and the personnel flow information of the surveillance camera, a feature representing the safety of personnel activities in a specific area is converted.

[0062] Step A130: The decision optimization unit performs b iterations of decision interference addition on the first conversion result to generate decision interference addition data. The decision interference addition data includes target template fault interference features obtained by the b iteration of decision interference addition.

[0063] In the first iteration, the decision optimization unit adds specific decision perturbations to each feature in the first conversion result. For example, in the HVAC system, for features related to chiller performance, a simulated fault perturbation might be added, such as one that reduces chiller cooling efficiency by, say, 10%. For the lighting system, perturbations might be added to features representing lighting quality and energy efficiency. For example, a light sensor reading deviation might be simulated, causing the lighting system to perceive the current light level as lower than the actual value, thus affecting lighting control decisions. In the elevator system, perturbations might be added to features reflecting elevator efficiency. For example, errors in the elevator car weight sensor might be simulated, causing the elevator control system to misjudge the car load, thus affecting the elevator's operating speed and docking decisions. For the security system, perturbations might be added to features representing the safety of human activity, such as simulated access control system misjudgments, causing authorized entry to be misjudged as unauthorized. As iterations progress, each iteration adjusts the decision perturbations based on the results of the previous iteration and the current system state. After b iteration cycles, decision interference addition data is generated, wherein the target template fault interference feature obtained by the decision interference addition of the b-th iteration cycle includes the state features of each subsystem after multiple decision interferences.

[0064] In step A140 , the decision optimization unit uses the first template state feature space as a fault diagnosis basis to perform adaptive control decisions on the fault disturbance propagation of b iterative cycles on the target template fault interference feature to generate a template fault diagnosis feature.

[0065] In the first iteration, the decision optimization unit analyzes the fault conditions in the target template's fault interference features based on the feature relationships in the first template state feature space. For example, in an HVAC system, due to the previously added interference of reduced chiller cooling efficiency, the decision optimization unit analyzes the potential fault propagation paths based on the relationships between the chiller and other components (such as fan coil units and cooling towers) in the first template state feature space. It may determine that the reduced chiller cooling efficiency will lead to a decrease in the fan coil unit's cooling effect, which in turn affects the indoor temperature. It then generates preliminary control decisions, such as appropriately increasing the chiller's compressor power or adjusting the fan coil unit's water valve opening. For the lighting system, based on the fault interference of light sensor reading deviation and the relationship between the lighting system and factors such as human activity and time in the first template state feature space, it may decide to adjust the dimming ratio of lamps or turn on more lamps to compensate for the perceived lack of light. In an elevator system, in response to the fault interference of misjudgment of car load, the elevator's speed limit and parking strategy are adjusted based on the relationship between elevator operation and safety protection mechanisms in the first template state feature space. In security systems, to address fault interference caused by false access control errors, the access control verification rules are adjusted or the monitoring alarm sensitivity is increased based on the relationship between the access control system, personnel authority management, and monitoring systems in the first template state feature space. As the iteration cycle progresses, each iteration adjusts based on the effectiveness of the previous cycle's control decisions and the propagation of the fault. After b iterations, a template fault diagnosis feature is generated.

[0066] Step A150 : performing adaptive control decision of fault disturbance propagation on the template fault diagnosis feature through the execution strategy unit to generate adaptive control decision data.

[0067] The strategy execution unit analyzes the fault status and correlations of each subsystem within the template's fault diagnosis features. For HVAC systems, the strategy execution unit searches a pre-defined control strategy library based on the fault status of equipment such as chillers and fan coil units and their interplay. For example, if a chiller failure results in insufficient cooling and abnormal fan coil operation, a strategy might be chosen to adjust the operating parameters of both the chiller and fan coil units simultaneously, such as further optimizing the chiller compressor frequency and the fan coil water valve opening ratio. For lighting systems, strategies such as adjusting the lighting layout or replacing the light sensor might be chosen to address lighting issues caused by a light sensor failure. For elevator systems, strategies such as recalibrating the car weight sensor or adjusting the elevator's dispatch algorithm might be used to address operational issues caused by incorrect car load assessments. For security systems, strategies such as updating the access control permission database or upgrading the access control verification algorithm might be adopted to address access control misjudgments. By executing these strategies, adaptive control decision data is generated.

[0068] Step A160 , extracting the state feature space of the adaptive control decision data through the sample building state recognition network to generate a template state feature space corresponding to the template fault diagnosis feature.

[0069] The sample building state recognition network analyzes the state information of each subsystem in the adaptive control decision data. For the HVAC system, it extracts new operating state features reflecting the entire HVAC system from the adjusted operating parameters of equipment such as chillers and fan coil units, such as whether the overall cooling efficiency has improved and whether the temperature in each area has stabilized. For the lighting system, new features related to lighting comfort and energy efficiency are extracted from information such as the adjusted lighting layout and the status of the light sensor. The elevator system extracts new features reflecting elevator operation stability and service quality from information such as the recalibrated car weight sensor and the adjusted scheduling algorithm. The security system extracts new features representing personnel safety management and regional security status from information such as the updated access control database and the upgraded verification algorithm, thereby generating a template state feature space.

[0070] Step A170: Determine a state feature space error based on the first template state feature space and the template state feature space.

[0071] Step A180: Optimizing the neuron weight information of the decision optimization unit based on the state feature space error until a first convergence requirement is met.

[0072] For example, for HVAC systems, the differences in chiller cooling efficiency characteristics and fan coil unit operating status characteristics are compared. For lighting systems, the differences in illuminance distribution characteristics and lamp energy-saving characteristics are compared. For elevator systems, the differences in elevator operating efficiency characteristics and car load judgment accuracy characteristics are compared. For security systems, the differences in personnel safety assessment characteristics and access control misjudgment rate characteristics are compared. Based on these differences, the overall state feature space error is calculated. Based on this state feature space error, the neuron weight information of the decision optimization unit is optimized until the first convergence requirement is met. During the optimization process, the neuron weights related to each subsystem in the decision optimization unit are adjusted based on the size and direction of the state feature space error. For example, if the characteristics of the chiller cooling efficiency in the HVAC system vary significantly, the neuron weights related to the chiller fault diagnosis and control decisions in the decision optimization unit will be adjusted. By continuously adjusting the weights, the decision optimization unit can perform more accurate analysis and decision-making when dealing with similar fault disturbances until the state feature space error reaches the first convergence requirement, that is, the error is within an acceptable range, thereby ensuring that the decision optimization unit can work more accurately in the subsequent adaptive control decisions of actual fault disturbance propagation.

[0073] In one possible implementation, step A170 includes:

[0074] Step A171: Determine an adjustment factor based on the set number of iteration cycles b, where the adjustment factor is inversely correlated with the set number of iteration cycles b.

[0075] Step A172: Calculate the distance between the first template state feature space and the template state feature space.

[0076] Step A173: Adjust the distance according to the adjustment factor to generate a state feature space error.

[0077] In this embodiment, in the context of a building automation system for a commercial office building, setting the number of iteration cycles, b, plays a crucial role in this process. Since the adjustment factor is inversely correlated with the number of iteration cycles, b, and assuming a value of 3 for b (this is based on the previous example and is not a logical assumption), the smaller the value of b, the larger the adjustment factor. This is because with fewer iteration cycles, each iteration has a relatively greater impact on the result, so a larger adjustment factor is needed to accurately measure the error.

[0078] When calculating the distance between the first template state feature space and the template state feature space, for the HVAC system, the chiller cooling efficiency features and fan coil unit operating state features in the first template state feature space are compared and calculated with the corresponding features in the template state feature space. For example, the chiller cooling efficiency may be represented as a specific numerical range or vector form in the first template state feature space, but in the template state feature space, due to a series of operations (such as fault disturbances, adaptive control decisions, etc.), it changes to another numerical range or vector form. The distance between the two is calculated using a specific mathematical calculation method (such as the Euclidean distance calculation method if the feature space is in vector form). Similarly, for the lighting system, the illumination distribution features and lamp energy-saving features are also calculated in this way. The elevator operating efficiency features and car load judgment accuracy features of the elevator system, as well as the personnel safety assessment features and access control false positive rate features of the security system, are also calculated according to their respective feature representation forms.

[0079] For example, for the HVAC system, if the calculated distance for a characteristic like chiller efficiency is a certain value, the adjustment factor is multiplied by the distance value based on its inverse relationship with b (assuming the adjustment factor is 0.5, based on the previous example). This value is then combined with the similarly adjusted values ​​of other subsystems to form the state feature space error. Lighting, elevator, and security systems also contribute to the overall state feature space error process based on their respective adjustment results.

[0080] In a possible implementation manner, the adaptive control decision result further includes disturbance control estimated in the adaptive control decision process of fault disturbance propagation in b iterative cycles.

[0081] The method further comprises:

[0082] The disturbance control error is determined based on the fault disturbance feature added by adding the decision disturbance for b iterative cycles to the first conversion result and the disturbance control feature estimated in the adaptive control decision process of propagating the fault disturbance for b iterative cycles to the fault disturbance feature of the target template.

[0083] Step A180 may include:

[0084] Step A181: Fusion calculation is performed on the state feature space error and the disturbance control error to generate a first training cost.

[0085] Step A182: Optimize the neuron weight information of the decision optimization unit based on the first training cost until a first convergence requirement is met.

[0086] In this embodiment, during the b-iteration cycle of adding decision disturbances to the first conversion result, for the HVAC system, the added fault disturbance feature may be a simulation of a reduction in the chiller's cooling efficiency or an abnormal change in the fan coil unit water valve opening. During the b-iteration cycle of the adaptive control decision-making process for fault disturbance propagation based on the target template fault disturbance feature, the estimated disturbance control feature may be, for example, an estimate of the chiller compressor power adjustment or an estimate of the fan coil unit water valve opening adjustment. Based on these, the disturbance control error is determined. The difference between the fault disturbance feature added during the decision disturbance addition and the actual estimated disturbance control feature during the adaptive control decision-making process for fault disturbance propagation is calculated. For example, the difference between the simulated value of the chiller's cooling efficiency reduction and the estimated cooling efficiency recovery value in the actual control decision, or the difference between the simulated value of the abnormal change in the fan coil unit water valve opening and the actual estimated adjustment value, etc., is combined to determine the disturbance control error.

[0087] Then, for the HVAC system, the chiller efficiency-related portion of the state feature space error and the chiller-related portion of the disturbance control error are calculated using a specific fusion method (such as weighted summation). Similarly, the two fan coil unit-related errors are fused and then combined with the fusion results from other subsystems to form a first training cost. Based on this first training cost, the neuron weights of the decision optimization unit are optimized until the first convergence requirement is met. During this process, the first training cost guides the direction and magnitude of adjustments to the neuron weights of the decision optimization unit. If the first training cost is large, indicating that the decision result of the current decision optimization unit deviates significantly from the expected result, the neuron weights are adjusted significantly, for example, in the neuron weights related to chiller fault diagnosis and control decisions. With continuous adjustments, the first training cost gradually decreases. When the first convergence requirement is met (for example, the first training cost is less than a set threshold), the decision-making ability of the decision optimization unit has reached a relatively ideal state within the current error standard.

[0088] In a possible implementation, step A180 may also include: locking the neuron weight information of the feature conversion module and the execution strategy unit, and optimizing the neuron weight information of the decision optimization unit based on the state feature space error until the first convergence requirement is met.

[0089] When optimizing the neuron weights of the decision optimization unit, the neuron weights of the feature conversion module and the execution strategy unit are locked. This means that during the optimization process, the neuron weights of the feature conversion module and the execution strategy unit remain unchanged, and the focus is on adjusting the neuron weights of the decision optimization unit. This ensures that changes in the weights of other modules will not interfere with the direction and effect of the decision optimization unit weight adjustment during the optimization process, making the optimization of the decision optimization unit more independent and accurate until it meets the first convergence requirement.

[0090] In a possible implementation, before step A120, the method includes:

[0091] Step A111: Obtain second template state variable monitoring data.

[0092] Step A112: performing feature conversion on the second template state variable monitoring data through the feature conversion module to generate a second conversion result.

[0093] Step A113: Performing an adaptive control decision of fault disturbance propagation on the second conversion result through the execution strategy unit to generate state variable monitoring data of an adaptive control decision template of fault disturbance propagation.

[0094] Step A114: Determine a second training cost based on the second template state variable monitoring data and the adaptive control decision template state variable monitoring data for fault disturbance propagation.

[0095] Step A115: Optimize the neuron weight information of the feature conversion module and the execution strategy unit based on the second training cost until a second convergence requirement is met.

[0096] Before the feature conversion module performs feature conversion on the first template state variable monitoring data, some pre-processing operations are required. The second template state variable monitoring data is obtained. This second template state variable monitoring data also comes from the various subsystems of the building automation system. For the HVAC system, this includes chiller operating parameters, fan coil status data, etc.; lighting system illuminance sensor values ​​and lamp status data; elevator system operating speed, car position, passenger capacity, and other data; security system access control system personnel entry and exit records, surveillance camera personnel flow information, etc.

[0097] Next, taking the HVAC system as an example, the operating parameters of the chiller might be converted into features more closely related to overall performance and faults, while the status data of the fan coil unit could be converted into features that reflect its impact on the indoor environment. In the lighting system, the illuminance sensor values ​​and lamp status data could be converted into features related to lighting comfort and energy efficiency. Elevator system operating data could be converted into features related to service quality and operational stability, and the personnel entry and exit and flow data of the security system could be converted into features related to regional safety management.

[0098] Furthermore, for the HVAC system, the execution strategy unit, based on the chiller and fan coil unit characteristics in the second conversion result, makes a control decision based on a pre-set control strategy library (containing control strategies for different fault scenarios) assuming a certain chiller fault (such as condenser scaling affecting cooling efficiency). For example, it adjusts the chiller compressor operating frequency or the fan coil unit water valve opening, thereby generating state variable monitoring data for the adaptive control decision template for fault disturbance propagation. In the lighting system, if the light sensor fails or the lamp status is abnormal, the execution strategy unit adjusts the lamp dimming ratio or turns some lamps on and off to generate corresponding data. In the elevator system, if the elevator speed is abnormal or the car load is incorrectly determined, it adjusts the elevator's parking strategy or speed limit to generate relevant data. In the security system, if the access control system misjudges or the surveillance camera data is abnormal, it adjusts the access control verification rules or increases the monitoring alarm sensitivity to generate relevant data.

[0099] For the HVAC system, the original operating parameters of the chiller in the second template state variable monitoring data are compared with the adjusted operating parameters of the chiller in the state variable monitoring data of the adaptive control decision template for fault disturbance propagation, and the difference between the two is calculated. Similarly, the difference between the original state data and the adjusted data of the fan coil unit is also calculated. The lighting system compares the difference between the original and adjusted values ​​of the illuminance sensor, the difference between the original and adjusted states of the lamps, etc. The elevator system compares the difference between the original data of the operating speed, car position, passenger capacity, etc. and the adjusted data. The security system compares the difference between the original data of the access control system personnel entry and exit records and the adjusted data, and the difference between the original information of personnel flow from the surveillance camera and the adjusted information, etc. These differences are combined according to a specific calculation method to determine the second training cost.

[0100] If the second training cost is large, it means that the current performance of the feature conversion module and the execution strategy unit deviates greatly from the expected results, and then their neuron weights will be adjusted. For example, in the feature conversion module, if it is found that the features after the conversion of the operating parameters of the HVAC system chiller cannot reflect the actual fault situation well, the neuron weights related to the chiller will be adjusted. In the execution strategy unit, if it is found that the control decision for the chiller fault cannot effectively improve the system state, the neuron weights related to the chiller fault control decision will be adjusted. With continuous adjustment, the second training cost gradually decreases. When the second convergence requirement is met (for example, the second training cost is less than a set threshold), it means that the feature conversion module and the execution strategy unit have reached a relatively ideal state under the current error standard.

[0101] In a possible implementation, step S131 includes:

[0102] Step S1311: The decision optimization unit obtains the first state feature space as a fault diagnosis basis, and simultaneously obtains information related to the fault disturbance event to be processed, wherein the information related to the fault disturbance event includes the type of the fault disturbance event and the affected building automation system component information.

[0103] Step S1312: Initializing relevant parameters of the fault disturbance event through the decision optimization unit, wherein the relevant parameters include a coefficient for measuring the possibility of fault propagation and an initial value of an indicator for evaluating the effect of the control decision.

[0104] Step S1313: Initialize information indicating an initial state of the fault disturbance event according to the fault disturbance event related information to obtain initial state information.

[0105] Step S1314: In the first round of iteration, the decision optimization unit analyzes the fault disturbance event based on the initialized relevant parameters and the initial state information of the fault to generate an analysis result. Specifically, the propagation path of the fault disturbance event in the current building automation system environment, the connection relationship between the various components in the building automation system, the data flow direction and the interaction mechanism are first analyzed to determine the potential direction and possibility of the fault disturbance event propagating from the initially affected component to other components.

[0106] Step S1315: Generate preliminary control decision data based on the analysis results of the fault disturbance propagation. The preliminary control decision data is generated based on the preliminary analysis of the global operating state of the building reflected by the first state feature space and the current fault disturbance, and is used to preliminarily suppress or adjust the propagation of the fault disturbance.

[0107] Step S1316: After generating the preliminary control decision data, apply the preliminary control decision data to the current state information of the fault disturbance event to obtain the fault disturbance state information after the first round of iteration. The fault disturbance state information reflects the state change of the fault disturbance event after the implementation of the preliminary control decision.

[0108] Step S1317, starting from the second round of iteration to the X-1 round of iteration, the decision optimization unit re-evaluates the propagation of the fault disturbance event in the building automatic control system based on the fault disturbance status information after the previous round of iteration, and adjusts the previous control decision data according to the new propagation analysis results. After adjusting the control decision data, the new control decision data is applied to the current fault disturbance status information to obtain the fault disturbance status information after the current iteration. The fault disturbance status information after the current iteration reflects the latest status information of the fault disturbance event after multiple rounds of iteration and control decision adjustment.

[0109] In step S1318, in the Xth iteration, the decision optimization unit extracts final fault diagnostic features based on the fault disturbance state information after the X-1th iteration. Specifically, the current fault disturbance state information is analyzed to determine the final propagation range of the fault disturbance throughout the building automation system, the ultimate impact on each key component, and whether other chain reactions have been triggered. Based on the analysis results, feature information reflecting the adaptive control decision results of the fault disturbance propagation in the Xth iteration cycle is extracted to obtain the target fault diagnostic features.

[0110] In a possible implementation, step S132 includes:

[0111] Step S1321: The execution strategy unit receives the target fault diagnosis feature from the decision optimization unit. The target fault diagnosis feature includes the fault disturbance propagation information optimized through multiple iterative cycles, and the current configuration information of the building automatic control system. The current configuration information includes the connection relationship, functional attributes, and operating parameter thresholds of the system components.

[0112] Step S1322: Analyze the fault propagation path, impact level, and potential chain reactions in the target fault diagnosis characteristics, identify the affected components that are directly and indirectly affected by the fault, prioritize the identified affected components, and generate a fault impact range assessment report and a component priority ranking list.

[0113] Step S1323: Based on the fault impact scope assessment report and the component priority ranking list, a matching control strategy is searched in a pre-built control strategy library, and a corresponding control plan is generated based on the matching result. The control strategy library contains a set of predefined control strategies for various fault scenarios.

[0114] Step S1324: Use the simulation environment or digital twin model of the building automation system to perform simulation testing on the control scheme and generate a simulation test report.

[0115] Step S1325: Based on the simulation test report, the control scheme is optimized to generate a final control decision result, wherein the optimization process involves adjusting control parameters, modifying control steps, and adding or deleting local control measures.

[0116] Step S1326: delivering the final control decision result to the execution layer of the building automatic control system, instructing the building automatic control system to implement the control operation according to the final control decision result.

[0117] Take, for example, the fault disturbance event of scaling on the chiller evaporator in a HVAC system. First, the decision optimization unit acquires the first state feature space as a basis for fault diagnosis, while also obtaining information related to the current fault disturbance event to be processed. The fault disturbance event of scaling on the chiller evaporator is clearly classified as a scaling issue on the evaporator. The affected building automation system components primarily include the chiller itself and its connected components, such as fan coil units and cooling towers connected via pipes.

[0118] Next, the decision optimization unit initializes the relevant parameters of the fault disturbance event. In terms of the coefficient used to measure the possibility of fault propagation, since evaporator scaling will affect the cooling efficiency of the chiller, which may in turn affect the temperature regulation function of the entire HVAC system, based on the connection relationship between the chiller and other components and previous experience data, the coefficient of the chiller fault propagation to the fan coil is set to 0.4, and the coefficient of propagation to the cooling tower is set to 0.3, etc. In terms of the initial values ​​of the indicators used to evaluate the effectiveness of the control decision, the initial evaluation value for controlling the indoor temperature within a comfortable range (for example, 22-26 degrees Celsius) is set to 0.8, and the initial evaluation value for controlling the energy consumption of the entire HVAC system within a reasonable range is set to 0.7, etc.

[0119] Based on the information related to the fault disturbance event, information representing the initial state of the fault disturbance event is initialized to obtain initial state information. In the initial state of scaling in the chiller evaporator, the evaporator's heat exchange efficiency decreases, the chiller's cooling capacity begins to decline, and the chilled water outlet temperature gradually increases. This information constitutes the initial state information of the fault.

[0120] In the first iteration, the decision optimization unit analyzes the fault disturbance event based on initialized parameters and initial fault state information. When analyzing the propagation path of the fault disturbance event in the current building automation system environment, the connections between the various components in the building automation system are taken into account. The chiller delivers chilled water to each fan coil unit through pipes, while the cooling tower is responsible for cooling the circulating water. Evaporator scaling reduces cooling efficiency and increases the chilled water temperature. This change propagates along the pipes to the fan coil units, affecting their cooling effect. Data flows from the chiller to the fan coil units. The interaction mechanism provides a cooling source for the chiller, and the fan coil units use chilled water for heat exchange to regulate the indoor temperature. Based on these circumstances, the potential direction for the fault disturbance event to propagate from the initially affected component (the chiller) to other components (the fan coil units and the cooling tower) is along the pipes, with the probability as previously determined.

[0121] Based on the analysis of the fault disturbance propagation, preliminary control decision data is generated. Based on the global building operating state reflected by the first state feature space and a preliminary analysis of the current fault disturbance, since the chiller's cooling capacity is decreasing and affecting indoor temperature, preliminary control decision data may include appropriately increasing the chiller compressor operating frequency to attempt to increase cooling capacity, while also adjusting the water valve opening of some fan coil units near the core area and slightly increasing the chilled water flow to enhance the cooling effect, thereby providing preliminary suppression or adjustment of the fault disturbance propagation.

[0122] After generating the preliminary control decision data, apply it to the current state information of the fault disturbance event to obtain the fault disturbance state information after the first round of iteration. At this point, the chiller compressor operating frequency increases, the water valve opening of some fan coil units increases, the chiller's cooling capacity may increase to a certain extent, the rising trend of the chilled water outlet temperature slows, and the cooling effect of the fan coil units is partially improved. These reflect the state changes of the fault disturbance event after the implementation of the preliminary control decision.

[0123] From the second iteration to the X-1 iteration (assuming X = 5), the decision optimization unit re-evaluates the propagation of the fault disturbance event in the building automation system based on the fault disturbance state information after the previous iteration. In the second iteration, it was found that although the operating frequency of the chiller compressor was increased, the cooling efficiency improvement was limited due to the unresolved evaporator scaling problem. In addition, energy consumption increased due to the increased compressor operating frequency. Based on the new propagation analysis results, the previous control decision data was adjusted. Considering the fundamental problem of evaporator scaling, the control decision was added to the schedule for evaporator cleaning. At the same time, the adjustment strategy for the fan coil water valve opening was adjusted to make more refined adjustments based on the actual temperature requirements of different areas, such as appropriately reducing the water valve opening in areas close to windows. After adjusting the control decision data, the new control decision data was applied to the current fault disturbance state information to obtain the fault disturbance state information after the current iteration. During the third iteration, it was discovered that the humidity in some areas had changed abnormally due to adjustments to the fan coil unit water valve openings. Consequently, the fault propagation was reassessed and humidity control measures were added to the control decisions. These measures included activating dehumidification equipment in some areas and adjusting their operating hours. Furthermore, the evaporator cleaning plan was further optimized, considering the use of chemical cleaning methods and determining the specific cleaning time and operating procedures. The new control decision data was then applied to the current fault disturbance state information, resulting in the fault disturbance state information for this iteration. During the fourth iteration, based on the changes in the fault disturbance state information, it was discovered that despite the implementation of dehumidification measures, humidity control was still suboptimal, and the chiller's cooling efficiency had not fully returned to normal levels. The control decisions were further optimized, adjusting the dehumidification equipment's operating parameters, such as the humidity setpoint and air volume, and adjusting the chiller's compressor operating frequency based on cooling capacity demand and energy consumption. The effectiveness of the evaporator cleaning was estimated, and the cleaning operation details were adjusted. The new control decision data was then applied to the current fault disturbance state information, resulting in the fault disturbance state information for this iteration.

[0124] In the Xth iteration (the fifth iteration), the decision optimization unit performed final fault diagnostic feature extraction based on the fault disturbance state information after the X-1th iteration. Analysis of the current fault disturbance state information determined that the fault disturbance had propagated throughout the building automation system, impacting air conditioning comfort on multiple floors, particularly those near the chillers. Regarding the final impact on each key component, despite some control measures, the chiller's evaporator scaling had not yet fully recovered, and cooling efficiency remained below normal. Due to operations such as water valve opening and humidity control, the operating parameters of some fan coil units were in a non-ideal but relatively stable state. The operation of the dehumidifier contributed to humidity control to some extent, but further optimization was required. Furthermore, no other chain reactions were triggered, such as impacts on the elevator or lighting systems. Then, based on the analysis results, characteristic information reflecting the adaptive control decision-making results of the fault disturbance propagation in the Xth iteration cycle is extracted, such as the specific value of the cooling efficiency in the final state of scaling of the chiller evaporator, the final adjustment value of the water valve opening of each fan coil unit, and the final operating parameters of the dehumidification equipment, to obtain the target fault diagnosis features.

[0125] The process of generating an adaptive control decision result of a building operation category associated with a first state feature space for adaptive control decision making by performing fault disturbance propagation on a target fault diagnosis feature through an execution strategy unit.

[0126] The execution strategy unit receives the target fault diagnosis features from the decision optimization unit. These target fault diagnosis features include fault disturbance propagation information optimized over multiple iterations, as well as the current configuration of the building automation system. For example, the current configuration information shows the connection between system components: a chiller connected to each fan coil unit and a cooling tower via pipes. The chiller's functional attribute is to provide a certain cooling capacity, and its operating parameter thresholds include the normal range of the chiller's cooling efficiency and the range of chilled water inlet and outlet temperatures. The fan coil unit's functional attribute is to regulate the indoor temperature, and the normal range of its water valve opening is also included.

[0127] Analyze the fault propagation path, impact, and potential chain reactions within the target fault diagnostic features, identify the chillers and fan coil units directly affected by the fault, and the dehumidifiers indirectly affected, and prioritize the identified affected components. Since the chiller is the source of the fault and fundamentally impacts the operation of the entire HVAC system, the chiller is given the highest priority. Fan coil units are directly affected by the chiller's cooling performance and are ranked second. Although the dehumidifier regulates humidity caused by adjusting the fan coil water valve opening, it has an indirect impact but also has a certain impact on the indoor environment and is ranked third. Generate a fault impact scope assessment report and a component priority list.

[0128] Based on the fault impact scope assessment report and the component priority list, a matching control strategy is searched for in the pre-built control strategy library. The control strategy library contains a set of predefined control strategies for various fault scenarios. For the case of scaling of the chiller evaporator, control strategies such as thorough cleaning of the evaporator, replacement of some damaged components, and optimization of the chiller operating parameters may be found. Based on the matching results, a corresponding control plan is generated. For example, it is determined that the evaporator should be cleaned first, and the operating parameters of the chiller are adjusted during the cleaning process, such as reducing the compressor operating frequency to reduce energy consumption and protect the equipment. At the same time, the fan coil water valve opening and the operating parameters of the dehumidification equipment are adjusted according to the actual needs of different areas.

[0129] Using the building automation system's simulated environment or digital twin model, the control scheme was simulated and tested, generating a simulation test report. During the simulation, the chiller evaporator was simulated to simulate actual scaling. The control scheme was then implemented to observe changes in operating parameters of various components, the effectiveness of controlling indoor temperature and humidity, and energy consumption. Evaporator cleaning improved cooling efficiency, but if the compressor operating frequency was completely reduced during the cleaning process, the cleaning process would take too long, impacting the normal operation of the entire system.

[0130] Based on the simulation test report, the control plan is optimized to generate the final control decision. The optimization process involves adjusting control parameters, such as adjusting the compressor operating frequency during the evaporator cleaning process to ensure cleaning effectiveness without excessively extending the cleaning time; modifying control procedures, such as adding a step to jointly debug the chiller and fan coil unit after the cleaning is completed; and adding local control measures, such as conducting a comprehensive inspection of the entire HVAC system before cleaning the evaporator to ensure there are no potential problems with other components.

[0131] The final control decision results are transmitted to the building automation system's execution layer, which instructs the building automation system to implement control operations based on the final control decision results. For example, the execution layer initiates evaporator cleaning according to the optimized time and conditions, adjusts the compressor operating frequency, performs joint commissioning of the chiller and fan coil units according to the new commissioning procedures, and controls the fan coil unit water valve opening and dehumidification equipment operation according to the adjusted parameters. This effectively controls HVAC system faults and ensures the normal operation of the building.

[0132] For a fault disturbance event, a light intensity sensor failure in a certain area of ​​the lighting system, the decision optimization unit obtains the first state feature space as a basis for fault diagnosis, and simultaneously obtains relevant information about the fault disturbance event. The fault type is a light intensity sensor failure, and the affected components are the lighting control system associated with the sensor and the lamps in the area.

[0133] Initialize relevant parameters. Regarding the coefficient measuring the likelihood of fault propagation, a coefficient of 0.8 was set to measure the likelihood of fault propagation, as a light sensor failure could lead to a misjudgment by the lighting control system, potentially affecting all lamps in the area. Regarding the initial values ​​of indicators used to evaluate the effectiveness of control decisions, the initial evaluation value for the area's lighting brightness meeting normal operating requirements (illuminance standards vary depending on the area's function) was set at 0.8, and the initial evaluation value for the entire lighting system's energy consumption being within a reasonable range was set at 0.7.

[0134] Initialize the initial state information of the fault. When the light sensor fails, the collected light data is inaccurate, which may cause the lighting control system to believe that the light level in the area is lower than the actual value.

[0135] In the first iteration, the fault propagation path was analyzed. A light sensor failure, through its connection to the lighting control system, affected the control signals of the connected lamps. Data flowed from the light sensor to the lighting control system and then to the lamps. The interaction mechanism enabled the light sensor to provide light information to the lighting control system, which then controlled the on / off and brightness of the lamps based on this information. It was determined that the fault was likely to propagate from the light sensor to the lamps. Based on these analysis results, preliminary control decision data was generated, such as switching to a backup light sensor (if available) or making preliminary adjustments to the lamp brightness based on the area's historical light data, the current time, and occupant activity, such as appropriately increasing the brightness during office hours.

[0136] The preliminary control decision data is applied to the current state information of the fault to obtain the fault disturbance state information after the first round of iteration. At this time, if the backup light intensity sensor is switched to, the lighting control system receives the new accurate light intensity information, and the brightness of the lamp is adjusted according to the new control decision.

[0137] From the second round of iteration to the X-1 round of iteration, the fault propagation situation is re-evaluated based on the fault disturbance status information after the previous round of iteration. In the second round of iteration, it was found that although the backup light sensor was switched or the brightness of the lamps was adjusted, the lighting effect in some areas was still not ideal due to changes in personnel activities in the area. According to the new propagation analysis results, the control decision data is adjusted, such as dynamically adjusting the brightness of the lamps according to the specific location of the personnel, and using the partition control method to divide the area into different areas such as office areas and corridors, and adjust the brightness separately. The new control decision data is applied to the current fault disturbance status information to obtain the fault disturbance status information after the current iteration. In subsequent iterations, the control decision will continue to be optimized based on factors such as personnel activities and time changes. For example, during the meeting time, the lights in the conference room area are adjusted to a brightness mode suitable for the meeting.

[0138] In the Xth iteration, final fault diagnostic feature extraction is performed to determine that the fault's propagation throughout the lighting system is limited to this area. The impact on key components (lamps and lighting control systems) is as follows: the lamp brightness meets demand after multiple adjustments, but the lighting control system undergoes a complex adjustment process due to a faulty light sensor. No other chain reactions are triggered, such as impacts on the HVAC or elevator systems. Feature information reflecting the adaptive control decision-making results of the fault disturbance propagation in the Xth iteration is extracted, such as the final lamp brightness adjustment value and the final operating mode of the lighting control system, to obtain the target fault diagnostic features.

[0139] After receiving the target fault diagnostic signature, the execution strategy unit analyzes the fault propagation path, impact, and potential chain reactions. It identifies the light sensors and lamps directly affected by the fault, prioritizing light sensors and lamps. It then generates a fault impact assessment report and a component priority list.

[0140] According to the report and list, the matching control strategy is searched in the control strategy library. It may be possible to find strategies such as repairing or replacing faulty light intensity sensors, optimizing the control algorithm of the lighting control system, etc., and generate corresponding control plans.

[0141] Using simulation environments or digital twin models, control solutions were tested, such as by simulating human activity at different times, to observe lighting effects and energy consumption. It was discovered that replacing the illuminance sensor immediately could cause fluctuations in the lighting system within a short period of time.

[0142] Based on the simulation test report, the control plan is optimized. For example, the light sensor replacement time is adjusted to a period of low human activity, and the control algorithm of the lighting control system is optimized to better adapt to changes in human activity during sensor failures. The final control decision result is generated and passed to the execution layer. The execution layer replaces the light sensor at the appropriate time based on the result and controls the brightness of the lamps according to the optimized algorithm.

[0143] For the fault disturbance event of the elevator car overload alarm device misjudging in the elevator system, the decision optimization unit obtains the first state feature space and related information of the fault disturbance event. The fault type is the misjudgment of the car overload alarm device, and the affected components are the elevator control system and the operation of the elevator car.

[0144] Initialize relevant parameters. Regarding the coefficient measuring the probability of fault propagation, set the coefficient to 0.9, as misjudgments by the overload alarm device can lead to incorrect operation of the elevator control system. Regarding the initial values ​​of indicators for evaluating the effectiveness of control decisions, set the initial evaluation value for safe and reliable elevator operation (no false overload alarms or abnormal operation) to 0.8, and the initial evaluation value for elevator operating efficiency within the normal range to 0.7.

[0145] Initialization fault initial status information, when the car overload alarm device misjudges, it may issue an alarm signal when it is not overloaded, causing the elevator to stop running or limit the passenger capacity.

[0146] In the first iteration, the fault propagation path was analyzed. The car overload alarm device is connected to the elevator control system. Misjudged signals could affect the elevator control system's assessment of the car load, and thus affect the elevator's operational decisions. Data flows from the overload alarm device to the elevator control system, and the interaction mechanism is that the overload alarm device provides the elevator control system with car load information. It was determined that the fault was highly likely to propagate from the overload alarm device to the elevator control system. Based on the analysis results, preliminary control decision data was generated. This included preliminary calibration of the overload alarm device and checking the proper connection of the sensor. The elevator control system's temporary handling strategy for overload alarms was also adjusted, such as performing a simple load review upon an alarm, rather than immediately stopping operation.

[0147] Apply the preliminary control decision data to the current state information of the fault to obtain the fault disturbance state information after the first round of iteration. At this time, the overload alarm device may have improved its misjudgment situation after calibration, and the temporary processing strategy of the elevator control system begins to take effect.

[0148] From the second iteration to the X-1 iteration, the fault propagation situation was re-evaluated. In the second iteration, it was found that despite preliminary calibration and temporary handling strategy adjustments, the overload alarm device still occasionally misjudged due to factors such as vibration during elevator operation. Based on the new propagation analysis results, the control decision data was adjusted, such as further in-depth inspection of the overload alarm device's sensors and replacement of potentially damaged components. At the same time, the elevator control system's processing logic for overload alarms was optimized, and a comprehensive judgment on whether the car is truly overloaded was made based on information such as the car's operating speed and acceleration. The new control decision data was applied to the current fault disturbance state information to obtain the fault disturbance state information after the current iteration. In subsequent iterations, the control decision was continuously optimized based on the actual operation of the elevator, such as adjusting the elevator's tolerance for overload alarms based on the passenger demand of different floors.

[0149] During the Xth iteration, final fault diagnostic feature extraction was performed, confirming that the fault's propagation throughout the elevator system was primarily concentrated in the elevator car and control system. The impact on key components (the overload alarm and elevator control system) was as follows: After multiple adjustments, the overload alarm significantly reduced misjudgments, and the optimized operating logic of the elevator control system was better able to handle overload alarms. No other chain reactions were triggered, such as impacts on the lighting or HVAC systems. Feature information reflecting the adaptive control decision-making results of the fault disturbance propagation during the Xth iteration was extracted, including the final calibration parameters of the overload alarm and the final operating logic of the elevator control system, to obtain the target fault diagnostic features.

[0150] After receiving the target fault diagnostic signature, the execution strategy unit analyzes the fault propagation path, impact, and potential chain reactions. It identifies the overload alarm and elevator control system directly affected by the fault, prioritizing the overload alarm and the elevator control system. It then generates a fault impact assessment report and a prioritized list of components.

[0151] According to the report and list, the matching control strategy is searched in the control strategy library. It may be possible to find strategies such as comprehensive inspection of the overload alarm device and upgrade the algorithm of the elevator control system, and generate corresponding control plans.

[0152] By using simulation environments or digital twin models to simulate and test control schemes, such as simulating the operation of elevators under different passenger capacities and operating scenarios, it was found that directly upgrading the elevator control system algorithm may cause compatibility issues with existing equipment.

[0153] Based on simulation test reports, control plans are optimized. For example, compatibility testing of the elevator control system is first performed, followed by a phased algorithm upgrade. Advanced testing equipment is used to ensure quality during a comprehensive overhaul of the overload alarm system. Final control decision results are generated and passed to the execution layer, which then conducts overhauls of the overload alarm system and upgrades the elevator control system using the optimized algorithm.

[0154] For the fault disturbance event of the access control system card reader failure in the security system, the decision optimization unit obtains the first state feature space and related information of the fault disturbance event. The fault type is the access control system card reader failure, and the affected components are the access control system and the associated monitoring system (used to record the entry and exit of people).

[0155] Initialize the relevant parameters. Regarding the coefficient measuring the probability of fault propagation, set the coefficient of fault propagation to 0.9, as card reader failure may cause the access control system to be unable to properly identify personnel. Regarding the initial values ​​of the indicators for initializing the effectiveness of control decisions, set the initial evaluation value for the access control system to accurately identify personnel (with an error rate below a certain value) to 0.8, and the initial evaluation value for maintaining the overall security of the security system at a normal level to 0.7.

[0156] Initialization fault initial status information. When the access control system card reader fails, it cannot correctly read the personnel access control card information, resulting in personnel being unable to enter and exit normally, or illegal personnel may try to enter by exploiting the fault loophole.

[0157] In the first iteration, the fault propagation path was analyzed. The card reader is connected to the access control system. A fault could prevent the access control system from obtaining correct personnel identity information, thus affecting the normal opening and closing of the access control. Data flows from the card reader to the access control system, and the interaction mechanism is that the card reader transmits the access card information it reads to the access control system for authentication. It was determined that the fault was highly likely to propagate from the card reader to the access control system. Based on the analysis results, preliminary control decision data was generated, such as restarting the card reader and checking the connection between the card reader and the access control system for looseness. At the same time, the access control system's temporary response strategy for the card reader failure was adjusted. For example, during the fault period, manual identity verification (such as checking other valid documents such as work permits) was used as a supplementary means to control personnel access.

[0158] Apply the preliminary control decision data to the current status information of the fault to obtain the fault disturbance status information after the first round of iteration. At this time, the card reader may partially restore its function after restarting. If the loose connection line is found and repaired after inspection, the temporary response strategy of the access control system will take effect, and the entry and exit of personnel will be managed to a certain extent.

[0159] From the second iteration to the X-1 iteration, the fault propagation situation was re-evaluated. In the second iteration, it was found that despite restarting and checking the lines, the card reader still frequently malfunctioned, possibly due to damage to the internal components of the card reader. Based on the new propagation analysis results, the control decision data was adjusted, such as arranging for professionals to conduct internal inspections of the card reader and preparing to replace potentially damaged components. At the same time, the personnel identity verification process of the access control system during maintenance was optimized and the level of identity verification was increased. For example, in addition to checking the ID card, personnel information must be verified with the relevant departments. The new control decision data was applied to the current fault disturbance status information to obtain the fault disturbance status information after the current iteration. In subsequent iterations, the control decision was optimized based on the card reader's detection results and the actual situation of personnel entering and exiting. For example, the strictness of manual identity verification was adjusted according to the security level of different areas.

[0160] In the Xth iteration, final fault diagnostic feature extraction was performed, confirming that the fault's propagation throughout the security system was primarily concentrated in the access control system's card reader and access control system. The impact on key components (card reader and access control system) was as follows: the card reader gradually regained stability after multiple inspections and repairs, and the access control system's personnel authentication process was optimized to better handle card reader failures. No other chain reactions were triggered, such as impacts on the elevator or lighting systems. Feature information reflecting the adaptive control decision-making results of the fault disturbance propagation during the Xth iteration was extracted, including the final repair status of the card reader and the final personnel authentication process of the access control system, to obtain the target fault diagnostic features.

[0161] After receiving the target fault diagnostic signature, the strategy execution unit analyzes the fault propagation path, impact, and potential chain reactions. It identifies the card readers and access control systems directly affected by the fault, prioritizing the card readers and access control systems. It then generates a fault impact assessment report and a prioritized list of components.

[0162] According to the report and list, the matching control strategy is searched in the control strategy library. It may be possible to find strategies such as completely replacing the card reader and upgrading the authentication algorithm of the access control system, and generate corresponding control plans.

[0163] Use simulation environments or digital twin models to simulate and test the control scheme, such as simulating the operation of the access control system under different personnel flow and security threat scenarios. It is found that if the card reader is directly replaced, it may cause incompatibility with some functions of the existing access control system, such as problems with the linkage with the monitoring system.

[0164] Based on the simulation test report, the control plan is optimized. For example, the new card reader is first tested for compatibility to ensure that all functions of the access control system and monitoring system can work properly before replacement. At the same time, the authentication algorithm of the access control system is gradually upgraded to ensure that the upgrade process does not affect the normal entry and exit of personnel. The final control decision result is generated and passed to the execution layer. The execution layer replaces the card reader (if necessary) according to the result and upgrades the access control system according to the optimized algorithm, thus ensuring the normal operation of the security system and maintaining the safety of the building.

[0165] In a possible implementation, before step S120, the method further includes:

[0166] Step S101 , obtaining third template state variable monitoring data and a third template state feature space corresponding to the third template state variable monitoring data, wherein the third template state feature space is generated by the sample building state recognition network performing feature space conversion on the third template state variable monitoring data.

[0167] In this embodiment, the third template state variable monitoring data originates from various subsystems of the building automation system, including the HVAC system, lighting system, elevator system, and security system. Taking the HVAC system as an example, the third template state variable monitoring data includes various operating parameters of the chiller, such as compressor speed, condenser and evaporator temperatures, chilled water inlet and outlet temperatures, and flow rate. It also includes status data for air conditioning terminal equipment such as fan coil units, such as fan speed and water valve opening. The third template state variable monitoring data in the lighting system includes illuminance values ​​collected by the illumination sensors in each area, the on / off status of lamps, and dimming ratios. The third template state variable monitoring data in the elevator system includes elevator speed, car position, passenger capacity, and door open / close status. The security system includes personnel entry and exit records from the access control system and processed personnel flow information collected by surveillance cameras. The third template state feature space is generated by the sample building state recognition network through feature space transformation of the third template state variable monitoring data. This feature space can reflect the characteristic relationship of these state variable data under specific modes. For example, in the HVAC system, it may reflect the correlation between the operating parameters of the chiller and the indoor ambient temperature and humidity, as well as the mutual influence relationship between the various components of the air-conditioning system.

[0168] Step S102: The feature pattern editing model performs feature pattern editing on the third template state feature space to a conversion result pattern to generate estimated feature pattern information. The conversion result pattern is used to represent the feature pattern matched by the conversion result generated by the feature conversion module.

[0169] Taking the HVAC system as an example, the characteristic pattern editing model will edit the characteristic relationship between the chiller and the fan coil unit in the third template state feature space. If the cooling efficiency of the chiller and the cooling effect of the fan coil unit in the third template state feature space is a more complex nonlinear relationship, the characteristic pattern editing model may edit it into a linear relationship or a specific functional relationship form that is easier to understand and handle according to the requirements of the conversion result pattern, thereby generating estimated characteristic pattern information. For the lighting system, if the relationship between the illuminance value and the dimming ratio of the lamp in the third template state feature space is based on a certain statistical law, the characteristic pattern editing model will edit it into a relationship form that meets the expected conversion result of the feature conversion module according to the conversion result pattern, for example, converting it into a pattern relationship directly related to lighting comfort.

[0170] Step S103 : performing adaptive learning according to the estimated feature pattern information through the adaptive learning network model to generate a template adaptive learning feature that matches the first feature pattern.

[0171] The adaptive learning network model analyzes and estimates the various feature pattern relationships in the feature pattern information. For HVAC systems, based on the edited relationship pattern between chillers and fan coil units, the adaptive learning network model learns a new feature pattern. For example, under different indoor and outdoor temperature differences and different occupant densities, how do chillers and fan coil units work together to achieve the best cooling or heating effect and energy efficiency? This is part of the generated template adaptive learning feature. For lighting systems, based on the edited relationship pattern between illuminance and dimming ratio, the adaptive learning network model learns how to adjust the dimming ratio of lamps in different time periods (such as daytime and nighttime) and different area functions (such as office areas and rest areas) to achieve the best lighting effect and energy saving goals. This also becomes part of the template adaptive learning feature.

[0172] Step S104 : determining a third training cost according to the template adaptive learning feature and the third template state feature space.

[0173] For the HVAC system, the features of the chiller and fan coil unit's coordinated operation in the template's adaptive learning features are compared with the original features in the third template's state feature space. If the optimal operating parameters of the chiller and fan coil unit under certain operating conditions in the template's adaptive learning features differ from the original parameters in the third template's state feature space, this difference is quantified. Similarly, the difference between the illumination intensity and dimming ratio in the lighting system's adaptive learning features and the third template's state feature space is calculated. These differences are combined to determine the third training cost. Similar difference calculations are performed for the elevator system and security system based on their respective features, and these differences are also involved in determining the third training cost.

[0174] Step S105 : optimizing the neuron weight information of the adaptive learning network model and the characteristic pattern editing model based on the third training cost until a third convergence requirement is met.

[0175] Finally, if the third training cost is large, it means that the current state of the adaptive learning network model and the feature pattern editing model deviates significantly from the expected state. For the adaptive learning network model, in the HVAC system, if the feature learning results related to the chiller and fan coil unit deviate significantly, the neuron weights related to the chiller and fan coil unit will be adjusted so that the required feature patterns can be generated more accurately in subsequent learning. For the feature pattern editing model, if editing the relationship between the chiller and the fan coil unit results in a large training cost, the neuron weights related to the editing of this relationship will be adjusted. With continuous adjustment, the third training cost gradually decreases. When the third convergence requirement is met (for example, the third training cost is less than a set threshold), it means that the adaptive learning network model and the feature pattern editing model have reached a relatively ideal state under the current error standard, and can provide an accurate basis for the subsequent feature space conversion of the state variable monitoring data.

[0176] Figure 2 FIG. 1 shows the hardware structure of the AI-based building automation system intelligent control system 100 for implementing the above-mentioned AI-based building automation system intelligent control method provided by an embodiment of the present invention, as shown in FIG. Figure 2 As shown, the AI-based building automation system intelligent control system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130 and a communication unit 140.

[0177] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions used by the AI-based building automation system intelligent control system 100 to execute or perform the exemplary methods described in the present invention.

[0178] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the AI-based building automatic control system intelligent control method of the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0179] The specific implementation process of the processor 110 can be found in the various method embodiments executed by the above-mentioned AI-based building automatic control system intelligent control system 100. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0180] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned AI-based intelligent control method for a building automatic control system is implemented.

[0181] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. An AI-based intelligent control method for building automation systems, characterized in that: The method comprises: Obtain state variable monitoring data of the building automation system; The adaptive learning network model performs feature space conversion on the state variable monitoring data to generate a first state feature space that matches a first feature pattern, where the first feature pattern is a feature pattern matched by the state feature space extracted by the building state recognition network; Based on the first state feature space, guiding the building automatic control system control network to perform adaptive control decisions on fault disturbance propagation for multiple different fault disturbance events, generating multiple different adaptive control decision results for building operation categories associated with the first state feature space, and the multiple different adaptive control decision results for building operation categories associated with the first state feature space are also used to optimize the building state recognition network; Performing adaptive control decisions on the building automatic control system based on a plurality of different adaptive control decision results of the building operation category associated with the first state feature space; The building automation system control network includes a decision optimization unit and an execution strategy unit; The method of guiding the building automatic control system control network to perform adaptive control decisions on fault disturbance propagation for a plurality of different fault disturbance events based on the first state feature space, and generating a plurality of different adaptive control decision results for the building operation category associated with the first state feature space, includes: For each of the multiple different fault disturbance events, the decision optimization unit uses the first state feature space as a fault diagnosis basis to perform adaptive control decisions of fault disturbance propagation for X iterations on the fault disturbance event, generating a target fault diagnosis feature obtained by the adaptive control decision of fault disturbance propagation in the Xth iteration, and using the fault diagnosis feature obtained by the adaptive control decision of fault disturbance propagation in the ath iteration as an input to the adaptive control decision of fault disturbance propagation in the a+1th iteration, where a and X are positive integers and a<X; The execution strategy unit performs adaptive control decision of fault disturbance propagation on the target fault diagnosis feature to generate an adaptive control decision result of the building operation category associated with the first state feature space.

2. The AI-based intelligent control method for building automation systems according to claim 1 is characterized in that: Before the step of guiding the building automation system control network to perform adaptive control decisions on fault disturbance propagation for a plurality of different fault disturbance events based on the first state feature space, and generating a plurality of different adaptive control decision results for building operation categories associated with the first state feature space, the method further includes: Obtaining first template state variable monitoring data, a first template state feature space corresponding to the first template state variable monitoring data, and a set number of iteration cycles b determined for the first template state variable monitoring data, wherein the first template state feature space is generated by performing feature space conversion on the first template state variable monitoring data by a sample building state recognition network, and the sample building state recognition network is the building state recognition network or another neural network model with the same network task as the building state recognition network; b is a positive integer not greater than X; The feature conversion module performs feature conversion on the first template state variable monitoring data to generate a first conversion result; Performing b iterations of decision interference addition on the first conversion result by the decision optimization unit to generate decision interference addition data; the decision interference addition data includes a target template fault interference feature obtained by the decision interference addition of the b iteration cycle; The decision optimization unit uses the first template state feature space as a fault diagnosis basis to perform adaptive control decisions on the fault disturbance propagation of b iterative cycles on the target template fault interference feature to generate a template fault diagnosis feature; Performing adaptive control decision of fault disturbance propagation on the template fault diagnosis feature by the execution strategy unit to generate adaptive control decision data; Performing state feature space extraction on the adaptive control decision data through the sample building state recognition network to generate a template state feature space corresponding to the template fault diagnosis feature; determining a state feature space error based on the first template state feature space and the template state feature space; Based on the state feature space error, the neuron weight information of the decision optimization unit is optimized until a first convergence requirement is met.

3. The AI-based intelligent control method for building automation systems according to claim 2 is characterized in that: The determining of a state feature space error based on the first template state feature space and the template state feature space includes: Determine an adjustment factor based on the set number of iteration cycles b, wherein the adjustment factor is inversely correlated with the set number of iteration cycles b; Calculating the distance between the first template state feature space and the template state feature space; The distance is adjusted according to the adjustment factor to generate a state feature space error.

4. The AI-based intelligent control method for building automation systems according to claim 2 is characterized in that: The adaptive control decision result also includes the disturbance control estimated in the adaptive control decision process of the fault disturbance propagation in b iterative cycles; The method further comprises: Determining a disturbance control error based on a fault disturbance feature added by performing decision disturbance addition for b iterative cycles on the first conversion result and a disturbance control feature estimated during an adaptive control decision process of fault disturbance propagation for b iterative cycles on the target template fault disturbance feature; Optimizing the neuron weight information of the decision optimization unit based on the state feature space error until a first convergence requirement is met includes: Fusing the state feature space error with the disturbance control error to generate a first training cost; Based on the first training cost, the neuron weight information of the decision optimization unit is optimized until a first convergence requirement is met.

5. The AI-based intelligent control method for building automation system according to claim 2 is characterized in that: Optimizing the neuron weight information of the decision optimization unit based on the state feature space error until a first convergence requirement is met includes: The neuron weight information of the feature conversion module and the execution strategy unit is locked, and the neuron weight information of the decision optimization unit is optimized based on the state feature space error until a first convergence requirement is met.

6. The AI-based intelligent control method for building automation systems according to claim 2 is characterized in that: Before the feature conversion module performs feature conversion on the first template state variable monitoring data to generate a first conversion result, the method includes: Obtaining second template state variable monitoring data; Performing feature conversion on the second template state variable monitoring data by the feature conversion module to generate a second conversion result; Performing an adaptive control decision of fault disturbance propagation on the second conversion result by the execution strategy unit to generate state variable monitoring data of an adaptive control decision template of fault disturbance propagation; determining a second training cost based on the second template state variable monitoring data and the adaptive control decision template state variable monitoring data of the fault disturbance propagation; The neuron weight information of the feature conversion module and the execution strategy unit is optimized based on the second training cost until a second convergence requirement is met.

7. The AI-based intelligent control method for building automation systems according to claim 1 is characterized in that: The step of performing, by the decision optimization unit, an adaptive control decision of fault disturbance propagation for X iteration cycles on each of the multiple different fault disturbance events using the first state feature space as a fault diagnosis basis, and generating a target fault diagnosis feature obtained by the adaptive control decision of fault disturbance propagation for the Xth iteration cycle, includes: The decision optimization unit obtains the first state feature space as a fault diagnosis basis, and simultaneously obtains information related to the fault disturbance event to be processed, wherein the information related to the fault disturbance event includes the type of the fault disturbance event and information about the affected building automation system components; Initializing relevant parameters of the fault disturbance event through the decision optimization unit, wherein the relevant parameters include a coefficient for measuring the possibility of fault propagation and an initial value of an indicator for evaluating the effect of the control decision; Initializing information indicating an initial state of the fault disturbance event according to the relevant information of the fault disturbance event to obtain initial state information; In the first iteration, the decision optimization unit analyzes the fault disturbance event based on the initialized relevant parameters and the initial state information of the fault to generate analysis results. Specifically, the propagation path of the fault disturbance event in the current building automation system environment, the connection relationship between the various components in the building automation system, the data flow direction and the interaction mechanism are analyzed to determine the potential direction and probability of the fault disturbance event propagating from the initially affected component to other components. Generating preliminary control decision data based on the analysis results of the fault disturbance propagation. The preliminary control decision data is generated based on the preliminary analysis of the global building operation state reflected by the first state feature space and the current fault disturbance, and is used to preliminarily suppress or adjust the propagation of the fault disturbance; After generating the preliminary control decision data, applying the preliminary control decision data to the current state information of the fault disturbance event to obtain fault disturbance state information after the first round of iteration, wherein the fault disturbance state information reflects the state change of the fault disturbance event after the implementation of the preliminary control decision; From the second round of iteration to the X-1th round of iteration, the decision optimization unit re-evaluates the propagation of the fault disturbance event in the building automatic control system based on the fault disturbance state information after the previous round of iteration, and adjusts the previous control decision data according to the new propagation analysis results. After adjusting the control decision data, the new control decision data is applied to the current fault disturbance state information to obtain the fault disturbance state information after the current iteration, and the fault disturbance state information after the current iteration reflects the latest state information of the fault disturbance event after multiple rounds of iteration and control decision adjustment; In the Xth iteration, the decision optimization unit performs final fault diagnosis feature extraction based on the fault disturbance state information after the X-1th iteration. Specifically, the current fault disturbance state information is analyzed to determine the final propagation range of the fault disturbance in the entire building automation system, the ultimate impact on each key component, and whether other chain reactions have been triggered. Then, based on the analysis results, feature information reflecting the adaptive control decision result of the fault disturbance propagation in the Xth iteration cycle is extracted to obtain the target fault diagnosis feature. The step of performing adaptive control decision of fault disturbance propagation on the target fault diagnosis feature by the execution strategy unit to generate an adaptive control decision result of the building operation category associated with the first state feature space includes: The execution strategy unit receives target fault diagnosis features from the decision optimization unit, the target fault diagnosis features including fault disturbance propagation information optimized through multiple iterative cycles and current configuration information of the building automation system, the current configuration information including connection relationships, functional attributes, and operating parameter thresholds of system components; Analyze the fault propagation path, impact level, and potential chain reactions in the target fault diagnosis characteristics, identify the directly and indirectly affected components, prioritize the identified affected components, and generate a fault impact scope assessment report and a component priority list; Based on the fault impact assessment report and component priority list, a matching control strategy is searched in a pre-built control strategy library, which contains a set of predefined control strategies for various fault scenarios, and a corresponding control solution is generated based on the matching results. Conduct simulation tests on the control scheme using a simulation environment or digital twin model of the building automation system, and generate a simulation test report; Based on the simulation test report, the control scheme is optimized to generate a final control decision result, wherein the optimization process involves adjusting control parameters, modifying control steps, and adding or deleting local control measures; The final control decision result is delivered to the execution layer of the building automatic control system, and the building automatic control system is instructed to implement the control operation according to the final control decision result.

8. The AI-based intelligent control method for building automation system according to any one of claims 2 to 6, characterized in that: Before the adaptive learning network model performs feature space conversion on the state variable monitoring data to generate a first state feature space matching the first feature pattern, the method further includes: Acquire third template state variable monitoring data and a third template state feature space corresponding to the third template state variable monitoring data, wherein the third template state feature space is generated by performing feature space conversion on the third template state variable monitoring data by the sample building state recognition network; The feature pattern editing model performs feature pattern editing on the third template state feature space to a conversion result pattern to generate estimated feature pattern information; the conversion result pattern is used to represent the feature pattern matched by the conversion result generated by the feature conversion module; Performing adaptive learning based on the estimated feature pattern information through the adaptive learning network model to generate a template adaptive learning feature that matches the first feature pattern; determining a third training cost based on the template adaptive learning feature and the third template state feature space; Based on the third training cost, the neuron weight information of the adaptive learning network model and the characteristic pattern editing model is optimized until a third convergence requirement is met.

9. An AI-based intelligent control system for building automation systems, characterized in that: The AI-based building automatic control system intelligent control system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the AI-based building automatic control system intelligent control method described in any one of claims 1 to 8 above.

Images