An air conditioner adaptive control method and device based on user operation

Abstract

The application relates to the technical field of air conditioner control, and discloses an air conditioner adaptive control method and device based on user operation. The method comprises the following steps: acquiring an environment state log, a system control log and a user operation log in an air conditioner running process; taking an automatic control event as an anchor point to construct a feedback identification time window, and identifying relevant user operation events in the window; monitoring the environment state until it is stable after the user operation event occurs, generating a target comfort label based on the stable environment state; updating a personalized comfort target model based on the target comfort label, and controlling the air conditioner to run based on the updated model. The application automatically converts user operation behavior into a model training label, solves the problem that air conditioner personalized control lacks continuous learning data sources in the related art, and improves the air conditioner adaptive control capability.

Description

Technical Field

[0001] This application relates to the field of air conditioning control technology, specifically to an air conditioning adaptive control method and device based on user operation. Background Technology

[0002] In related technologies, personalized air conditioning control schemes typically record users' manual adjustments as isolated parameter change events, failing to recognize them as feedback signals to the system's automatic control results. This leads to a significant amount of behavioral data containing user comfort preferences not being effectively utilized. While related technologies attempt to learn from user adjustment behavior, they directly use the parameter values ​​after user adjustments as training labels, ignoring the thermal inertia of the room environment under air conditioning system control—environmental parameters require a transition process to reach a stable state after user adjustments, and the parameters at the moment of adjustment cannot accurately reflect the user's true comfort preferences.

[0003] In other words, the relevant technology cannot automatically convert user actions into model training labels, resulting in a lack of continuous learning data sources for personalized air conditioning control and insufficient adaptive capabilities. Summary of the Invention

[0004] This application provides a user-operated adaptive control method and device for air conditioning to solve the problems of insufficient data sources for continuous learning and inadequate adaptive capabilities in personalized air conditioning control.

[0005] In a first aspect, this application provides an adaptive control method for air conditioning based on user operation, the method comprising: Acquire environmental status logs, system control logs, and user operation logs during the operation of the air conditioner; Using the automatic control events in the system control log as anchor points, a feedback identification time window is constructed for each automatic control event; Within the feedback recognition time window, identify user operation events related to automatic control events; After a user action event occurs, the environmental state is monitored, and once the environmental state enters a stable state, a target comfort label is generated based on the stable environmental state. Update the personalized comfort target model based on the target comfort label; The air conditioning operation is controlled based on the updated personalized comfort target model.

[0006] In one optional implementation, using automatic control events in the system control log as anchor points, a feedback identification time window is constructed for each automatic control event, including: Filter control events automatically executed by the air conditioning system from the system control log and determine the actual effective time of the control event as the time anchor point; Starting from a time anchor point, a feedback recognition time window with a duration equal to the preset feedback window length is constructed. The preset feedback window length is configured based on at least one of the control action type of the automatic control event and the thermal inertia of the room where the air conditioner is located.

[0007] In one alternative implementation, it further includes: When multiple automatic control events exist in the same time period, the user operation event is associated with the target automatic control event among the multiple automatic control events. The target automatic control event is the control event whose occurrence time is closest to the occurrence time of the user operation event and is in an active state when the user operation event occurs.

[0008] In one alternative implementation, within the feedback identification time window, user operation events related to automatic control events are identified, including: If a user action event occurs within the feedback recognition time window, the user action event is determined to be related to the automatic control event based on at least one of the following conditions: The operation parameters of user operation events and the control parameters of automatic control events belong to the same control dimension or are directly related dimensions. When a user operation event occurs, the corresponding control for the automatic control event is still in effect; The target device identifier for a user operation event is the same as the device identifier for an automatic control event.

[0009] In one optional implementation, after a user action event occurs, the environmental state is monitored, and once the environmental state reaches a stable state, a target comfort label is generated based on the stable environmental state, including: A stable observation window is launched after the user action event occurs; Within a stable observation window, monitor the rate of change of environmental parameters related to user action events; When the change in environmental parameters is less than the corresponding threshold in multiple consecutive sampling periods, the environmental state is determined to have entered a stable state. Extract the statistics of the environment state under the steady state, and use them as the steady environment state; The target comfort label is calculated based on the stable environmental conditions.

[0010] In one alternative implementation, the target comfort label is calculated based on a stable environmental state, including: The stable comfort value is calculated based on the stable environmental state, and this stable comfort value is used as the target comfort label; or... The target comfort range is determined based on the stable environmental state, and the target comfort range is used as the target comfort label.

[0011] In one alternative implementation, the personalized comfort target model is updated based on the target comfort label, including: Construct training samples, which include sample features, target comfort labels, and sample weights. The sample features include at least environmental parameters and time period information. The training samples are added to the historical training sample set, and the personalized comfort target model is trained and updated based on the historical training sample set.

[0012] In one alternative implementation, the sample weights include at least one of the following: The time distance weighting is such that the closer the time of occurrence of the user operation event is to the time of occurrence of the automatic control event, the higher the time distance weighting. Stability weight: The faster the environment enters a stable state after a user action event, the higher the stability weight.

[0013] Secondly, this application provides an air conditioning adaptive control device based on user operation, the device comprising: The log acquisition module is used to acquire environmental status logs, system control logs, and user operation logs during the operation of the air conditioner. The window building module is used to build a feedback recognition time window for each automatic control event, using the automatic control events in the system control log as anchor points. The feedback recognition module is used to identify user operation events related to automatic control events within the feedback recognition time window; The tag generation module is used to monitor the environmental state after a user operation event occurs, and generate target comfort tags based on the stable environmental state after the environmental state enters a stable state. The model update module is used to update the personalized comfort target model based on the target comfort label; The control module is used to control the operation of the air conditioner based on the updated personalized comfort target model.

[0014] Thirdly, this application provides an electronic device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the user-operated adaptive air conditioning control method of the first aspect or any corresponding embodiment described above.

[0015] Fourthly, this application provides a computer-readable storage medium storing computer instructions for causing a computer to execute the user-operated adaptive air conditioning control method of the first aspect or any corresponding embodiment described above.

[0016] Fifthly, this application provides a computer program product, including computer instructions for causing a computer to execute the user-operated adaptive air conditioning control method of the first aspect or any corresponding embodiment described above.

[0017] According to the user-operated adaptive control method for air conditioning provided in this application, the following beneficial technical effects can be achieved compared with the prior art: By acquiring environmental status logs, system control logs, and user operation logs during air conditioning operation, a complete data foundation is provided for subsequent identification of the correlation between user operations and automatic control events. Using automatic control events in the system control logs as anchor points, a feedback identification time window is constructed for each automatic control event, limiting user operations to a reasonable range that is temporally relevant to the automatic control event, thus eliminating interference from irrelevant operations. Within the feedback identification time window, user operation events related to the automatic control event are identified, transforming operations originally recorded as isolated parameter change events into feedback signals to the automatic control results, solving the problem in related technologies where supervisory signals cannot be extracted from user operations. After a user operation event occurs, the system monitors the environmental state. Once the environment stabilizes, a target comfort label is generated based on this stable state. This avoids the impact of environmental transition caused by the thermal inertia of the air conditioner on the label accuracy, ensuring that the generated label truly reflects the user's desired comfort level in the current environment. Based on the target comfort label, the personalized comfort target model is updated, achieving automated conversion from user operation to model training labels. This allows the model to continuously obtain training data from the user's daily usage behavior. Based on the updated personalized comfort target model, the air conditioner operation is controlled, enabling the air conditioner control to continuously approach the user's true comfort preferences, ultimately improving the adaptive capability of the personalized air conditioner control. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating an application scenario according to an embodiment of this application; Figure 2 This is a flowchart of an air conditioning adaptive control method based on user operation according to an embodiment of this application; Figure 3 This is a flowchart illustrating the extraction of a stable environmental state after artificial intervention and the generation of a target comfort label, according to an embodiment of this application. Figure 4 This is a flowchart illustrating a feedback window and manual intervention identification according to an embodiment of this application; Figure 5 This is a structural block diagram of an air conditioning adaptive control device based on user operation according to an embodiment of this application; Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] It is understood that before using the technical solutions disclosed in the various embodiments of this application, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this application in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0022] As one optional application scenario in the embodiments of this application, such as Figure 1 As shown, the user-operated adaptive air conditioning control system may include at least one terminal device and at least one server. Figure 1 The example shows that the system includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0023] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0024] In related technologies, personalized air conditioning control schemes typically record users' manual adjustment operations as isolated parameter change events, failing to identify them as feedback signals to the system's automatic control results. This leads to a significant amount of behavioral data containing user comfort preferences not being effectively utilized. While related technologies attempt to learn from user adjustment behavior, they directly use the parameter values ​​after user adjustments as training labels, ignoring the thermal inertia of the room environment under air conditioning system control. Consequently, the parameters at the moment of adjustment cannot accurately reflect the user's true comfort preferences.

[0025] To address this, this application provides an air conditioning adaptive control method based on user operation. By constructing a feedback recognition time window with automatic control events as anchor points, user operation events related to automatic control events are identified. After the operation event occurs, the environmental state is monitored until it stabilizes, and a target comfort label is generated. Based on this label, the personalized comfort target model is updated and the air conditioning operation is controlled. In this way, user operation behavior is automatically converted into model training labels, providing a continuous learning data source for personalized control and improving the adaptive control capability of the air conditioning.

[0026] According to an embodiment of this application, an embodiment of an adaptive control method for air conditioning based on user operation is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0027] This embodiment provides an air conditioning adaptive control method based on user operation, which can be used in air conditioning controllers. Figure 2 This is a flowchart of an air conditioning adaptive control method based on user operation according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain the environmental status log, system control log, and user operation log during the operation of the air conditioner.

[0028] Specifically, the environmental status log refers to time-series data recording environmental parameters and their changes within the space where the air conditioner is located, including at least timestamps, space identifiers, temperature, humidity, carbon dioxide concentration, and occupancy status. The system control log refers to log data recording control events automatically issued and executed by the air conditioning system, including at least the automatic control event identifier, corresponding space identifier, corresponding target air conditioner identifier, actual effective time of the control, control command issued by the system, execution status of the control, and the model version used to generate the control. The user operation log refers to log data recording manual adjustments to air conditioning operating parameters by users through applications or control terminals, including at least the corresponding space identifier, corresponding target air conditioner identifier, operation time, operation source, operation type, parameter values ​​before the operation, and parameter values ​​after the operation.

[0029] This step, by collecting the three types of logs mentioned above, provides a complete data foundation for subsequent implicit feedback identification and label generation. The environmental status log provides the environmental context of the air conditioner's operation, the system control log records the control events initiated by the system and their timing, and the user operation log records the user's response behavior to these control events. Together, these three constitute a complete observation chain from system actions to user feedback to environmental changes.

[0030] In one possible implementation, the environmental status log is collected by environmental sensors deployed in the space where the air conditioner is located at a preset sampling period and uploaded to the cloud learning platform via an edge device. The environmental sensors include temperature sensors, humidity sensors, carbon dioxide sensors, and human infrared sensors, with a preset sampling period of once every 30 seconds. The system control log is automatically generated and recorded by the air conditioner control system each time a control command is automatically issued. In fully automatic mode, the actual effective time is the moment when the system directly issues and executes the control command. In suggested mode, the actual effective time is the moment when the control is actually executed after user confirmation. The user operation log is automatically recorded by the application or control terminal when it detects that the user has manually modified the air conditioner operating parameters. The operation sources include the system application, the system control terminal, and user confirmation, rejection, or adjustment records in suggested mode.

[0031] Step S202: Using the automatic control events in the system control log as anchor points, construct a feedback identification time window for each automatic control event.

[0032] Specifically, automatic control events refer to control actions actively issued and actually executed by the air conditioning system, including but not limited to temperature adjustment, humidity adjustment, fresh air activation, fan speed adjustment, and mode switching. These events are obtained by filtering from the system control log. The feedback identification time window is a continuous time interval constructed from the actual effective time of the automatic control event to a preset feedback window length. It is used to capture manual operations performed by the user after the automatic control occurs that may be related to this automatic control. By constructing this window, user operations can be limited to a reasonable range related to the time of the automatic control event, eliminating interference from irrelevant operations performed by the user long after the automatic control occurred.

[0033] This step constructs a feedback recognition time window using automatic control events as anchor points, limiting user operations that may be related to the automatic control in time to a reasonable observation range. This eliminates the interference of independent user operations long after the automatic control occurred on implicit feedback recognition, thus improving the accuracy of subsequent recognition steps.

[0034] In one possible implementation, control events automatically executed and actually effective by the air conditioning system are selected from the system control log. The actual effective time of the control event is recorded as a time anchor point. Starting from this anchor point, a feedback recognition time window with a preset feedback window length is constructed. For temperature adjustment actions, since the room temperature response is relatively fast, the preset feedback window length can be configured to 15 minutes; for fresh air or air quality intervention actions, since the air environment parameter response is relatively slow, the preset feedback window length can be configured to 30 minutes. If no user operation event occurs within the feedback recognition time window, this automatic control will not generate training samples, or will only be retained as a no-feedback sample.

[0035] For example, if the air conditioning system automatically performs a temperature adjustment control at 14:30, this moment is recorded as the time anchor point t0. A feedback recognition time window of 15 minutes is constructed starting from t0, i.e., [14:30, 14:45]. User actions occurring within this window will be used as candidate feedback for subsequent judgment.

[0036] Step S203: Within the feedback recognition time window, identify user operation events related to automatic control events.

[0037] Specifically, user operation events refer to records of user actions that manually adjust air conditioner operating parameters through applications, control terminals, or suggested modes, including but not limited to temperature adjustment, fan speed adjustment, mode switching, and power on / off. In this step, user operation events specifically refer to candidate operation events that fall within the feedback recognition time window.

[0038] This step involves determining the relevance of user operation events within the feedback identification time window to filter out the operation events that truly constitute a user response to the automatic control result. The relevance determination is based on a comprehensive assessment of multiple dimensions, including parameter dimension correlation, effective status correlation, and device identifier consistency. This avoids misclassifying independent operations that are close in time but completely unrelated in control dimensions as valid feedback, thus ensuring the quality of subsequently generated training samples.

[0039] In one possible implementation, if a user action event occurs within the feedback recognition time window, the user action event is determined to be related to an automatic control event based on at least one of the following conditions: Condition 1: The operation parameters of the user operation event and the control parameters of the automatic control event belong to the same control dimension or a directly related dimension. A control dimension refers to an independently adjustable functional category in an air conditioning system, including temperature, humidity, fan speed, and mode dimensions. A directly related dimension refers to a physical relationship between two control dimensions; for example, the fresh air dimension and the temperature dimension are directly related because the introduction of fresh air changes the indoor temperature distribution.

[0040] Condition 2: When a user operation event occurs, the corresponding automatic control event is still within its effective period. The effective period refers to the time period from the actual effective time of the automatic control event until the system issues a new control command to override that control.

[0041] Condition 3: The target device identifier corresponding to the user operation event is consistent with the device identifier corresponding to the automatic control event.

[0042] If an operation is only close in time but completely unrelated in control dimensions, it will not be included in the training samples.

[0043] For example, if the system detects an increase in indoor carbon dioxide concentration at 14:30 and automatically activates fresh air, and the user adjusts the air conditioner's set temperature from 25.5℃ to 26℃ via the system application at 14:45, the system determines that the user's action is related to the automatic control event: First, the operating parameter temperature and the automatic control parameter fresh air are directly related dimensions, because the entry of fresh air affects the perceived indoor temperature; second, the automatic fresh air control was still active when the user's action occurred; third, the user's action and the automatic control are both directed at the same air conditioning unit. This action is determined to be valid feedback. If the system automatically controls the air conditioner at 14:30, and the user initiates the activation control again via the system application at 16:30 due to re-entering the room, this action is not considered feedback to the 14:30 control event.

[0044] Step S204: After a user operation event occurs, monitor the environmental state, and after the environmental state enters a stable state, generate a target comfort label based on the stable environmental state.

[0045] Specifically, environmental state refers to the set of values ​​for various environmental parameters within the space where the air conditioner is located, including at least temperature, humidity, and carbon dioxide concentration. Steady state refers to the convergent state of environmental parameters when the variation amplitude of relevant environmental parameters is less than the corresponding threshold over multiple consecutive sampling periods. Target comfort label refers to a quantitative indicator used to characterize the comfort state that a user expects to achieve in the current environmental context, including stable comfort values ​​or target comfort ranges.

[0046] This step is one of the core aspects that distinguishes this embodiment from related technologies. Related technologies typically record the control parameter values ​​adjusted by the user. However, due to the thermal inertia of the air conditioning system, the environment does not immediately reach a steady state after user adjustments. Directly using environmental parameter values ​​or user-set values ​​as labels at this time results in significant deviations. This step does not immediately sample after user operation but continues to monitor the environmental state, waiting for the environmental parameters to gradually converge to a stable state before extracting the environmental state and generating the target comfort label. This eliminates the interference of air conditioning thermal inertia and environmental fluctuations on the label accuracy, ensuring that the generated label truly reflects the user's desired target comfort state.

[0047] In one possible implementation, a stable observation window is initiated, starting from the moment the user operation event occurs. The duration of this window is a preset stable observation window, configured based on room size, air conditioning response speed, and operation type; for example, it can be configured to be 15 to 30 minutes. Within the stable observation window, key environmental parameters related to the user operation are continuously monitored at preset sampling intervals (e.g., every 30 seconds), and the variation amplitude of each environmental parameter over multiple consecutive sampling intervals is calculated. When the variation amplitude of the relevant environmental parameter is less than the corresponding threshold for a preset number of consecutive sampling intervals, the environmental state is considered to have entered a stable state. Variation amplitude thresholds are preset for different types of environmental parameters; for example, the temperature variation amplitude threshold can be set to 0.2℃, the humidity variation amplitude threshold can be set to 2%, and the carbon dioxide concentration variation amplitude threshold can be set to 50ppm. The environmental state under stable conditions, or its statistical measure (e.g., average value) under stable conditions, is extracted as the stable environmental state. Based on the stable environmental state, the corresponding stable comfort value is calculated and used as the target comfort label.

[0048] For example, if a user adjusts the air conditioner setting from 25.5℃ to 26℃ at 14:45, a 20-minute stable observation window is initiated starting at 14:45. Temperature data is collected every 30 seconds within the stable observation window. When the temperature change is less than 0.2℃ for five consecutive sampling periods (i.e., five consecutive samples, totaling 2.5 minutes), the temperature environment is considered to have entered a stable state. The average temperature at each sampling time within the stable zone, along with the average values ​​of other environmental parameters, are extracted and combined to form the stable environmental state. Based on this stable environmental state, a stable comfort value is calculated as the target comfort label.

[0049] In another possible implementation, a target comfort range can be determined based on a stable environmental state rather than a single comfort value, and this target comfort range can be used as a target comfort label. The target comfort range can be an interval centered on the stable comfort value, extending upwards and downwards within a preset fluctuation range. For example, it could be an interval centered on the stable comfort value, defined by comfort ranges corresponding to ±0.5℃. Using an interval-based label provides a more flexible optimization objective for model training, better accommodating subtle fluctuations in user comfort preferences under similar environmental conditions.

[0050] If a reliable stable region cannot be identified within the stable observation window (e.g., environmental parameters consistently fail to meet the stability criteria), then this user operation will not be included in the current training sample.

[0051] Step S205: Update the personalized comfort target model based on the target comfort label.

[0052] Specifically, a personalized comfort goal model refers to a machine learning model that can predict a user's desired target comfort state based on input environmental scenario information. Model types include, but are not limited to, regression models, decision trees, neural networks, etc., as long as they can predict the target comfort state based on the input scenario.

[0053] This step establishes a complete closed loop from user behavior to continuous model evolution by constructing training samples based on target comfort labels automatically generated from user actions and updating the personalized comfort target model. Compared with training methods in related technologies that rely on explicit user feedback, this step does not require users to actively provide any ratings or preference settings. The model can automatically learn and evolve from the user's daily usage behavior, continuously approaching the user's true comfort preferences.

[0054] In one possible implementation, training samples are constructed, including sample features, target comfort labels, and sample weights. The sample features include at least environmental parameters (such as temperature, humidity, and carbon dioxide concentration) and time period information (dividing the day into multiple time periods, such as morning, afternoon, and night), and optionally also occupancy status and previous system actions. The constructed training samples are added to a historical training sample set, and the personalized comfort target model is trained and updated based on this historical training sample set. After training, the updated model is published to the edge-end air conditioning control device for subsequent automatic control.

[0055] In one possible implementation, sample weights include at least one of temporal distance weights and stability weights. The temporal distance weight is negatively correlated with the time interval between the occurrence of the user action event and the occurrence of the automatic control event; the shorter the interval, the higher the weight. This means that samples more directly related to the automatic control event have a higher proportion in the training process. The stability weight is positively correlated with the speed at which the environment reaches a stable state after the user action event; the faster the environment reaches a stable state, the higher the weight. This means that samples with faster environment convergence are more representative of the environment. By introducing these weights into the training samples, the model can distinguish the contributions of samples of different quality during training. Samples with strong correlation and good environmental stability have a greater impact on model parameter updates, while the impact of samples with weak correlation or poor environmental stability is appropriately reduced, thereby improving the model's training effect and prediction accuracy.

[0056] For example, after a user action, a target comfort label is generated, along with the time period (afternoon), occupancy status (occupied), corresponding environmental parameters, and previous system actions. The system constructs this information into a training sample. If the action occurs within 3 minutes of an automatic control event, it is assigned a higher temporal distance weight; if the environment stabilizes within 8 minutes of the action, it is assigned a higher stability weight. This sample is added to the historical training sample set for the next model training. After the model is updated, under similar time periods and environmental conditions, the system can automatically output a control target that better matches the user's comfort preferences.

[0057] Step S206: Control the operation of the air conditioner based on the updated personalized comfort target model.

[0058] Specifically, controlling the operation of the air conditioner refers to the process of generating corresponding control commands based on the target comfort state output by the updated personalized comfort target model and sending them to the air conditioner for execution. Control commands include, but are not limited to, temperature setpoints, fan speed settings, and operating modes.

[0059] This step is the final execution stage of the learning process described above. The purpose of updating the personalized comfort target model is to make subsequent control decisions more closely aligned with the user's actual comfort preferences. By inputting the current environmental state into the updated model, the model outputs the target comfort state that the user expects to achieve in the current scenario. Based on this, the system generates corresponding control commands and sends them to the air conditioner for execution, thereby achieving adaptive control of the air conditioner.

[0060] In one possible implementation, after the personalized comfort target model is updated, the current environmental state (including temperature, humidity, carbon dioxide concentration, etc.) and scene information (including time period and occupancy status, etc.) are acquired and input into the updated personalized comfort target model. The model predicts and outputs the target comfort state (e.g., target comfort value or target comfort range) that the user prefers to achieve in the current scene based on the input information. The system generates corresponding control commands based on the target comfort state output by the model and the differences between the current environmental parameters and the target state. These control commands are then sent to the air conditioner for execution, causing the air conditioner to operate according to the control commands.

[0061] For example, in a family living room scenario, the system automatically turns on the fresh air system at 14:30, and the user manually adjusts the set temperature at 14:45. Based on this, the system generates training samples and updates the model. The next day, around 14:30, the system detects an increase in indoor CO2 concentration and prepares to turn on the fresh air system again. At this time, the system inputs the current environmental status (temperature 26℃, humidity 55%, CO2 concentration 1200ppm) and scenario information (afternoon, occupied status) into the updated model. Based on the user preferences learned the previous day, the model outputs the target comfort state (e.g., the corresponding stable comfort value) that the user prefers to achieve in this scenario. The system generates the corresponding temperature setpoint based on this target comfort state and sends it to the air conditioner as the control parameter for this automatic control.

[0062] Thus, a complete "perception → learning → execution" closed loop is formed, from collecting user operation behavior, identifying implicit feedback, generating target comfort labels, updating the model, to finally issuing control commands. As users continue to generate operation behavior in daily use, the above process is repeated continuously, and the personalized comfort target model continues to evolve, making air conditioning control increasingly closer to the user's true comfort preferences.

[0063] The user-operated adaptive control method for air conditioning provided in this embodiment acquires environmental status logs, system control logs, and user operation logs during air conditioning operation, providing a complete data foundation for subsequent identification of the correlation between user operations and automatic control events. Using automatic control events in the system control logs as anchor points, a feedback identification time window is constructed for each automatic control event, limiting user operations to a reasonable range that is temporally related to the automatic control event, thus eliminating interference from irrelevant operations. Within the feedback identification time window, user operation events related to the automatic control event are identified, transforming operations originally recorded as isolated parameter change events into feedback signals to the automatic control results. This solves the problem in related technologies where it is impossible to extract data from user operations... The system addresses the issue of extracting supervisory signals. After a user action event occurs, the system monitors the environmental state and generates target comfort labels based on the stable environmental state. This avoids the impact of environmental transition processes caused by the thermal inertia of the air conditioner on the accuracy of the labels, ensuring that the generated labels truly reflect the comfort state the user expects to achieve in the current environmental context. Based on the target comfort labels, the system updates the personalized comfort target model, achieving automated conversion from user actions to model training labels, enabling the model to continuously obtain training data from the user's daily usage behavior. Based on the updated personalized comfort target model, the system controls the air conditioner's operation, allowing the air conditioner control to continuously approach the user's true comfort preferences, ultimately improving the adaptive capability of the personalized air conditioner control.

[0064] In an optional implementation, step S202 above includes the following steps: Step a1: Filter control events automatically executed by the air conditioning system from the system control log and determine the actual effective time of the control event as the time anchor point.

[0065] Step a2: Starting from the time anchor point, construct a feedback recognition time window with a duration equal to the preset feedback window length. The preset feedback window length is configured based on at least one of the control action type of the automatic control event and the thermal inertia of the room where the air conditioner is located.

[0066] Specifically, the time anchor point refers to the actual effective time of the control event automatically executed by the air conditioning system, selected from the system control log; that is, the time when the control command truly takes effect on the air conditioning equipment. The preset feedback window length refers to the duration of the pre-set feedback recognition time window; for example, temperature adjustment actions can be set to 15 minutes, and fresh air or air quality intervention actions can be set to 30 minutes. The control action type refers to the types of control commands automatically issued by the air conditioning system, including but not limited to temperature adjustment, humidity adjustment, fresh air activation, fan speed adjustment, and mode switching. Room thermal inertia refers to the room's resistance to changes in environmental parameters; specifically, it is the time required for the room's environmental parameters to reach a stable state after user operation or system control. The larger the room volume and the better the insulation performance of the building envelope, the greater the thermal inertia, and the longer the time required for environmental parameters to stabilize.

[0067] This implementation method standardizes the observation interval for feedback of each automatic control event by filtering executed control events from the system control log and using their effective time as the time anchor point, and constructing a unified feedback identification time window with a preset window length. By configuring the window length according to the control action type and the room's thermal inertia, reasonable feedback observation durations can be set for different types of control events and rooms with different characteristics. This ensures that effective user operation feedback is not missed due to an excessively short window, nor is irrelevant operation interference introduced due to an excessively long window, thereby improving the accuracy of subsequent correlation determination.

[0068] In one possible implementation, automatic control events with an execution status of "executed" are selected from the system control log. The actual effective time of this event is recorded as the time anchor point t0. A feedback recognition time window [t0, t0+Tw] with a preset feedback window length Tw is constructed starting from this anchor point. The preset feedback window length Tw is configured differently according to the type of control action: for temperature adjustment actions, since the room temperature response is relatively fast, Tw can be configured to 15 minutes; for fresh air activation or air quality intervention actions, since the air environment parameter response is slower, Tw can be configured to 30 minutes. If the thermal inertia of the room where the air conditioner is located is large (such as a large space or a room with good insulation), Tw can be appropriately extended; if the room thermal inertia is small (such as a small space or a room with poor sealing), Tw can be appropriately shortened. If no user operation event occurs within the feedback window, no training sample is generated for this automatic control, or it is only retained as a no-feedback sample.

[0069] For example, the room where the air conditioner is located is a large space with good insulation and high thermal inertia. The system performs a temperature adjustment control at 14:30. Based on the configuration of temperature adjustment actions, the basic window length is 15 minutes. Considering the high thermal inertia of this room, the window length is extended to 20 minutes, and the feedback recognition time window is constructed as [14:30, 14:50]. User operation events occurring within this window will be used as candidate feedback for subsequent judgment.

[0070] In an optional implementation, the method further includes: when multiple automatic control events exist in the same time period, associating the user operation event with a target automatic control event among the multiple automatic control events, wherein the target automatic control event is a control event whose occurrence time is closest to the occurrence time of the user operation event and is in an active state when the user operation event occurs.

[0071] Specifically, a target automatic control event refers to the automatic control event among multiple automatic control events existing in the same time period that occurs most recently with the occurrence time of the user operation event and is still in effect when the user operation event occurs.

[0072] This implementation determines the target automatic control event corresponding to the user's operation event based on two conditions: "most recent" and "effective status," in scenarios with overlapping multiple events. When multiple events coexist, a single user operation may correspond to multiple possible sources of control events. Without attribution determination, it is impossible to accurately establish the correlation between the user operation and the automatic control event. This implementation effectively solves the attribution problem in multi-event scenarios by using the above two conditions, ensuring the accuracy of subsequent relevance determination and tag generation.

[0073] In one possible implementation, when a user operation event occurs within the feedback recognition time window, the occurrence time of the user operation event is first determined. Then, among multiple automatic control events existing in the same time period, the one with the closest occurrence time to the user operation event is selected as a candidate automatic control event. Next, it is checked whether the candidate automatic control event is still active at the time the user operation event occurred. If it is active, the candidate automatic control event is identified as the target automatic control event, and the user operation event is associated with the target automatic control event. If it is not active, the next most recent automatic control event is searched, and the above judgment is repeated.

[0074] For example, if the system automatically adjusts the humidity at 14:20 and automatically turns on the fresh air system at 14:30, and the user manually adjusts the temperature via the app at 14:45, the user's action is first determined to occur at 14:45. Among the automatic control events existing within the same time period, the closest event to 14:45 is the fresh air system activation event at 14:30, and the fresh air control is still active at that time. Therefore, the user's action is associated with the fresh air control event at 14:30, rather than the humidity control event at 14:20. This attribution rule can accurately establish the association between user actions and corresponding automatic control events in complex scenarios with multiple overlapping events.

[0075] In an optional implementation, step S203 includes: If a user action event occurs within the feedback recognition time window, the user action event is determined to be related to the automatic control event based on at least one of the following conditions: The operation parameters of user operation events and the control parameters of automatic control events belong to the same control dimension or are directly related dimensions. When a user operation event occurs, the corresponding control for the automatic control event is still in effect; The target device identifier for a user operation event is the same as the device identifier for an automatic control event.

[0076] Specifically, the operation parameters of user operation events refer to the control parameter values ​​set by the user after manual adjustment, such as "26℃" when the user adjusts the temperature from 25.5℃ to 26℃, or "high fan speed" when the user adjusts the fan speed from automatic to high. The control parameters of automatic control events refer to the parameter values ​​carried in the control commands automatically issued by the system, such as "26℃" in the system's "set temperature 26℃" command, or "fresh air on" in the "turn on fresh air" command. The same control dimension refers to functional parameters whose operation parameters and control parameters belong to the same category, such as temperature and temperature belonging to the same dimension, or fan speed and fan speed belonging to the same dimension. Directly related dimensions refer to dimensions where the operation parameters and control parameters, although not belonging to the same category, have a physical linkage relationship. For example, turning on fresh air will change the indoor temperature distribution, therefore the fresh air dimension and the temperature dimension are directly related dimensions; humidity adjustment and temperature adjustment also have a mutual influence relationship. The target device identifier refers to the name or number used to uniquely identify the target device, such as the device's MAC address or serial number, used to determine whether the user operation and automatic control are targeting the same air conditioning device.

[0077] This implementation method comprehensively determines the correlation between user operation events and automatic control events by considering three dimensions: parameter dimension correlation, effective status correlation, and device identifier consistency. It cross-validates from multiple perspectives whether the operation event constitutes a valid response to the automatic control result. Compared to a single determination method that relies solely on temporal proximity, multi-dimensional determination can effectively filter out independent operations that are temporally close but substantially unrelated, improving the purity of implicit feedback samples and reducing the negative impact of misjudged samples on model training.

[0078] In one possible implementation, when a user operation event occurs within the feedback recognition time window, the following conditions are checked sequentially: First, determine whether the operation parameters of the user operation event and the control parameters of the automatic control event belong to the same control dimension (e.g., temperature to temperature, wind speed to wind speed) or a directly related dimension (e.g., fresh air to temperature, humidity to temperature); second, determine whether the automatic control event is still in effect when the user operation event occurs; finally, determine whether the target device identifier corresponding to the user operation event is consistent with the device identifier corresponding to the automatic control event. If at least one of the above conditions is met, the user operation event is determined to be related to the automatic control event.

[0079] For example, if the system automatically turns on the fresh air at 14:30, and the user adjusts the air conditioner's set temperature from 25.5℃ to 26℃ via the App at 14:45, then the following conditions are met: First, the operation parameter "26℃" and the control parameter "fresh air turned on" are directly related (fresh air affects the perceived temperature); second, the fresh air control is still in effect when the user operates; and third, the device identifiers are consistent.

[0080] If the system automatically adjusts the humidity at 14:30, and the user adjusts the air conditioner's fan speed from automatic to high via the App at 14:40, then the operation parameter "fan speed" and the control parameter "humidity" do not belong to the same control dimension or a directly related dimension (humidity and fan speed have no direct physical linkage). Even if the time is close, the operation will not be judged as related.

[0081] In an optional implementation, step S204 includes: Step b1: After the user operation event occurs, start the stable observation window; Step b2: Within the stable observation window, monitor the rate of change of environmental parameters related to user operation events; Step b3: When the change amplitude of environmental parameters in multiple consecutive sampling periods is less than the corresponding threshold, the environmental state is determined to have entered a stable state. Step b4: Extract the statistics of the environment state under the steady state as the steady environment state; Step b5: Calculate the target comfort label based on the stable environmental conditions.

[0082] Specifically, a stable observation window refers to a continuous observation interval constructed for a preset duration, starting from the moment the user's operation event occurs. This interval is used to wait for the air conditioning system and room environment to reach thermodynamic equilibrium again after user intervention. The duration can be configured based on room size, air conditioning response speed, and the type of operation, for example, 15 to 30 minutes. A statistic refers to a representative value obtained by statistically processing environmental parameter values ​​from multiple sampling moments within the stable region, including but not limited to the average or median. For example, averaging the temperature values ​​from 10 consecutive sampling moments within the stable region yields the average temperature for that period. A stable environmental state refers to the set of environmental parameters corresponding to the convergence of environmental parameters and the fulfillment of stability criteria. This can be the environmental parameter value at a specific moment within the stable region, or a statistic obtained by statistically processing environmental parameters from multiple moments within the stable region.

[0083] This implementation method incorporates the environmental convergence process after user interaction into the label generation logic through a complete process: activating a stable observation window, continuously monitoring the rate of change of environmental parameters, determining the stable zone based on a threshold of change amplitude, extracting the stable environmental state, and finally calculating the target comfort label. Using a threshold value within multiple consecutive sampling periods as the criterion for determining the stable zone effectively distinguishes between normal fluctuations in environmental parameters during the convergence process and the truly thermodynamically balanced stable state, avoiding label deviations caused by premature sampling. By generating labels after waiting for environmental convergence, the interference of air conditioning thermal inertia and environmental fluctuations on label accuracy is fundamentally eliminated, ensuring that the generated labels accurately reflect the user's desired target comfort state.

[0084] In one possible implementation, a stable observation window is initiated, with the occurrence of the user action event as the starting point and a duration equal to the preset stable observation window length. Within the stable observation window, key environmental parameters related to the current user action are continuously monitored at preset sampling periods, and the variation amplitude of each environmental parameter over multiple consecutive sampling periods is calculated. When the variation amplitude of the relevant environmental parameter is less than the corresponding threshold over a preset number of consecutive sampling periods, the environmental state is determined to have entered a stable state. The environmental state under the stable state, or its statistical measure (e.g., mean or median) under the stable state, is extracted as the stable environmental state. Based on the stable environmental state, a stable comfort value is calculated using a comfort calculation model (e.g., a Predicted Mean Vote (PMV) model), and this stable comfort value is used as the target comfort label. If a credible stable region cannot be identified within the stable observation window, the current user action is not included in the current training sample.

[0085] For example, at 14:45, a user adjusts the air conditioner setting from 25.5℃ to 26℃, initiating a 20-minute stable observation window. Within this window, temperature data is collected every 30 seconds. When the temperature change is less than 0.2℃ for five consecutive sampling periods (i.e., five consecutive samples, totaling 2.5 minutes), the temperature environment is considered to have entered a stable state. The average temperature at each sampling time within this stable period, along with the average values ​​of other environmental parameters (humidity, CO2 concentration, etc.), are extracted and combined to form the stable environmental state. This stable environmental state is input into the PMV comfort model to calculate the corresponding stable comfort value as the target comfort label. If, after the 20-minute stable observation window ends, the temperature change consistently fails to meet the condition of being less than 0.2℃ for five consecutive sampling periods, a reliable stable zone is not identified, and this user action is not included in the training samples.

[0086] Figure 3 This is a flowchart illustrating the extraction of a stable environmental state after artificial intervention and the generation of a target comfort label, according to an embodiment of this application. Figure 3 As shown, the process includes the following stages: Detection of human intervention phase: After identifying user operation events related to automatic control events within the feedback recognition time window, record the human intervention behavior.

[0087] Recording control parameters after intervention: Record the time of occurrence of user operation events and the modified control parameter values ​​as the starting point for subsequent steady-state monitoring.

[0088] Observation of the stabilization window phase: After the user operation event occurs, the stabilization observation window is started to continuously monitor the changing trends of environmental parameters related to this intervention and determine whether the environment has converged to a stable state.

[0089] Extracting the stable environmental state stage: When the change amplitude of environmental parameters is less than the corresponding threshold in multiple consecutive sampling periods, the environmental state is determined to have entered the stable region. The environmental state under the stable state or the statistics of the environmental state under the stable state are extracted as the stable environmental state.

[0090] Stable comfort value calculation stage: Based on the extracted stable environmental state, the stable comfort value is calculated using a comfort calculation model (such as the PMV model).

[0091] Target comfort label generation stage: The calculated stable comfort value is output as the target comfort label for subsequent model training. If a reliable stable region cannot be identified, no label is generated and the region is not included in the training samples.

[0092] For a detailed description of each of the above stages, please refer to the relevant content in step S204 and its optional embodiments in this specification, which will not be repeated here.

[0093] In one optional implementation, step b5 above includes the following steps: Step c1: Calculate the stable comfort value based on the stable environmental state, and use the stable comfort value as the target comfort label; or, Step c2: Determine the target comfort range based on the stable environmental state, and use the target comfort range as the target comfort label.

[0094] Specifically, the stable comfort value refers to an index value used to quantify user comfort, which is calculated based on a combination of environmental parameters such as temperature, humidity, and carbon dioxide concentration under stable environmental conditions. Examples include specific values ​​of comfort indicators such as PMV or Standard Effective Temperature (SET).

[0095] The target comfort range refers to the comfort value range formed by expanding a preset range upward and downward from the stable comfort value as the center. For example, the range formed by expanding upward and downward from the stable comfort value as the center and the comfort range corresponding to ±0.5℃ is used to accommodate the slight fluctuations in the user's comfort preference under similar environmental conditions.

[0096] This implementation provides two forms of target comfort labels: stable comfort values ​​in point form and target comfort intervals in interval form. Point-value labels are straightforward to calculate and have a clear training objective, making them suitable for control scenarios with high accuracy requirements. Interval-value labels provide a more flexible optimization objective for model training, better accommodating subtle fluctuations in user comfort preferences under similar environmental conditions, thus improving the model's robustness and generalization ability.

[0097] In one possible implementation, if a point-value format is used, the extracted stable environmental state is input into a comfort calculation model (such as a PMV model). The model outputs a specific comfort value as the stable comfort value, which is then directly used as the target comfort label. If an interval format is used, the stable comfort value is first calculated as described above, and then the target comfort interval is formed by expanding upwards and downwards from this value with a preset fluctuation range. This preset fluctuation range can be dynamically determined based on the dispersion of the user's historical behavior data, or it can be set to a fixed value.

[0098] For example, if the stable environmental conditions extracted after a user operation are a temperature of 26°C, humidity of 55%, and CO2 concentration of 800 ppm, and these parameters are input into the PMV model to calculate a PMV value of +0.3, then +0.3 can be directly used as the target comfort label. Alternatively, with the PMV value of +0.3 as the center and a fluctuation range of ±0.2, the target comfort interval can be determined as [+0.1, +0.5], and this interval can be used as the target comfort label. When using the interval form, if the model's predicted value falls within this interval during subsequent model training, it can be considered as meeting the user's comfort preference, thus providing a more lenient optimization objective.

[0099] In an optional implementation, step S205 above includes the following steps: Step d1: Construct training samples. The training samples include sample features, target comfort labels, and sample weights. The sample features include at least environmental parameters and time period information.

[0100] Step d2: Add the training samples to the historical training sample set, and train and update the personalized comfort target model based on the historical training sample set.

[0101] Specifically, a training sample refers to a single data unit used for training the personalized comfort target model. It contains model input features and corresponding label values, i.e., a combination of sample features, target comfort labels, and sample weights. Sample features refer to the environmental scene description information input into the personalized comfort target model, including at least environmental parameters (such as temperature, humidity, and carbon dioxide concentration) and time information (such as morning, afternoon, and night). Optionally, it may also include occupancy status and previous system actions. Sample weights are the importance coefficients of training samples during model training, used to adjust the influence of samples of different quality on model parameter updates. The historical training sample set refers to the collection of all training samples accumulated from historical usage, used to train and update the personalized comfort target model, and continuously expanded with daily user use.

[0102] This implementation constructs structured training samples containing features, labels, and weights, enabling the model training process to learn the scenario-dependent nature of user comfort preferences using multi-dimensional environmental information. The time-based information and environmental parameters in the sample features together provide the model with contextual information, allowing it to learn the differentiated patterns of user comfort preferences in different scenarios. By continuously adding newly generated samples to the historical training sample set and updating the model, continuous evolution and adaptation of the model are achieved.

[0103] In one possible implementation, after each target comfort label is generated, the environmental parameters (temperature, humidity, carbon dioxide concentration), time period information, occupancy status (if recorded), and previous system actions (if recorded) corresponding to that label are combined into sample features. These sample features are then combined with the target comfort label and sample weights determined based on correlation analysis to construct a complete training sample. This training sample is added to the historical training sample set, and the updated historical training sample set is used to retrain or incrementally train the personalized comfort target model, generating updated model parameters.

[0104] For example, after a user action, a target comfort label is generated, along with the recorded time period (afternoon), occupancy status (occupied), corresponding environmental parameters (temperature 26℃, humidity 55%, CO2 concentration 800ppm), and previous system action ("fresh air turned on"). The system constructs this information into a training sample: Sample Features = {Time Period: Afternoon, Occupancy Status: Occupied, Temperature: 26℃, Humidity: 55%, CO2: 800ppm, Previous Action: Fresh Air Turned On}, Target Comfort Label = Stable Comfort Value (e.g., PMV = +0.3), Sample Weight = Time Distance Weight 0.9 + Stability Weight 0.8. This sample is added to the historical training sample set for the next model training. After the model is updated, under similar time periods and environmental conditions, the system can automatically output control targets that better match the user's comfort preferences.

[0105] In one optional implementation, the sample weights in step d1 above include at least one of the following: The time distance weighting is such that the closer the time of occurrence of the user operation event is to the time of occurrence of the automatic control event, the higher the time distance weighting. Stability weight: The faster the environment enters a stable state after a user action event, the higher the stability weight.

[0106] Specifically, the time distance weight is a weighting coefficient determined based on the length of the time interval between the occurrence of a user operation event and the occurrence of an automatic control event. This weight is negatively correlated with the time interval—the shorter the interval, the higher the weight; the longer the interval, the lower the weight. Its physical meaning is: the more timely the user operation, the more likely it is to be a direct response to the automatic control, and the higher the sample quality. The stability weight is a weighting coefficient determined based on the speed at which the environment reaches a stable state after a user operation event. This weight is positively correlated with the speed of reaching a stable state—the faster the stability is reached, the higher the weight; the slower the stability is reached, the lower the weight. Its physical meaning is: the faster the environment converges, the more representative the environmental state after the operation, and the higher the sample quality.

[0107] This implementation introduces temporal distance weights and stability weights into the training samples, enabling the model to distinguish the contributions of samples of different quality during training. Samples with strong correlation (short time intervals) and good environmental stability (rapid convergence) have a greater impact on model parameter updates, while the impact of samples with weak correlation or poor environmental stability is appropriately reduced, thereby improving the model's training effect and prediction accuracy.

[0108] In one possible implementation, when constructing training samples, the system calculates a time distance weight based on the time interval between the occurrence of a user operation event and the occurrence of an automatic control event. Specifically, a time interval threshold can be set: a weight of 1.0 for intervals within 0-5 minutes, 0.8 for intervals within 5-10 minutes, 0.6 for intervals within 10-15 minutes, and 0.4 for intervals exceeding 15 minutes. The system also calculates a stability weight based on the time required for the environment to reach a stable state after a user operation event; the shorter the time to reach a stable state, the higher the weight. The time distance weight and the stability weight are then combined according to a preset rule (such as weighted summation or multiplication) to form a comprehensive weight for the sample, which serves as the final weight for that training sample.

[0109] For example, if a user action occurs 3 minutes after an automatic control event, the time distance weight is 1.0; if the environment stabilizes 6 minutes after the action, the stability weight is 0.9; the final sample weight is the product of these two weights, 0.9. Another user action occurs 12 minutes after an automatic control event, with a time distance weight of 0.6; if the environment stabilizes 18 minutes after the action, the stability weight is 0.5; the final sample weight is 0.3. During model training, the former sample has a much greater influence on model parameter updates than the latter, thus ensuring the quality of the training samples.

[0110] In one alternative implementation, Figure 4 This is a flowchart illustrating a feedback window and manual intervention identification according to an embodiment of this application. Figure 4 As shown, the process includes the following steps: First, locate the automatic control event in the system control log and record the actual effective time of the event as the time anchor point t0.

[0111] Then, starting from the time anchor point t0, a feedback recognition time window [t0, t0+Tw] is constructed, where Tw is the preset feedback window length.

[0112] Within the feedback recognition time window, it is determined whether a user operation event (i.e., manual intervention) has occurred. If no user operation event occurs within the window, no implicit feedback sample is generated for this automatic control, and the process ends.

[0113] If a user operation event occurs within the window, it is further determined whether the user operation event is related to an automatic control event. The correlation determination is based on at least one of the following conditions: the operation parameter and the control parameter belong to the same control dimension or a directly related dimension; the automatic control is still in effect when the user operation occurs; the target device identifier corresponding to the user operation is consistent with the device identifier corresponding to the automatic control.

[0114] If the result is deemed relevant, the label generation process begins; if the result is deemed irrelevant, no implicit feedback sample is generated, and the process ends.

[0115] For a detailed description of each of the above steps, please refer to steps S202 to S203 and their optional embodiments in this specification, which will not be repeated here.

[0116] This embodiment also provides an air conditioning adaptive control device based on user operation, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0117] This embodiment provides an air conditioning adaptive control device based on user operation, such as... Figure 5 As shown, it includes: The log acquisition module 501 is used to acquire environmental status logs, system control logs, and user operation logs during the operation of the air conditioner. The window construction module 502 is used to construct a feedback identification time window for each automatic control event, using the automatic control events in the system control log as anchor points. The feedback recognition module 503 is used to recognize user operation events related to automatic control events within the feedback recognition time window; The tag generation module 504 is used to monitor the environmental state after a user operation event occurs, and generate a target comfort tag based on the stable environmental state after the environmental state enters a stable state. The model update module 505 is used to update the personalized comfort target model based on the target comfort label; Control module 506 is used to control the operation of the air conditioner based on the updated personalized comfort target model.

[0118] In one alternative implementation, the window building module 502 includes: The event filtering unit is used to filter control events automatically executed by the air conditioning system from the system control log and determine the actual effective time of the control event as the time anchor point; A window construction unit is used to construct a feedback recognition time window with a duration equal to the preset feedback window length, starting from a time anchor point. The preset feedback window length is configured based on at least one of the control action type of the automatic control event and the thermal inertia of the room where the air conditioner is located.

[0119] In an alternative implementation, the window building module 502 further includes: The event association unit is used to associate a user operation event with a target automatic control event among multiple automatic control events when multiple automatic control events exist at the same time. The target automatic control event is the control event whose occurrence time is closest to the occurrence time of the user operation event and is in an active state when the user operation event occurs.

[0120] In one optional implementation, the feedback identification module 503 includes: The correlation determination unit is used to determine whether a user operation event is related to an automatic control event based on at least one of the following conditions when a user operation event occurs within the feedback recognition time window: The operation parameters of user operation events and the control parameters of automatic control events belong to the same control dimension or are directly related dimensions. When a user operation event occurs, the corresponding control for the automatic control event is still in effect; The target device identifier for a user operation event is the same as the device identifier for an automatic control event.

[0121] In one alternative implementation, the label generation module 504 includes: A stable window startup unit is used to start a stable observation window after a user operation event occurs. The environmental monitoring unit is used to monitor the rate of change of environmental parameters related to user operation events within a stable observation window; The stability determination unit is used to determine that the environmental state has entered a stable state when the change amplitude of environmental parameters is less than the corresponding threshold in multiple consecutive sampling periods. The state extraction unit is used to extract the statistics of the environment state under the steady state, and use them as the steady environment state. The tag calculation unit is used to calculate the target comfort tag based on a stable environmental state.

[0122] In one alternative implementation, the tag calculation unit is specifically used for: The stable comfort value is calculated based on the stable environmental state, and this stable comfort value is used as the target comfort label; or... The target comfort range is determined based on the stable environmental state, and the target comfort range is used as the target comfort label.

[0123] In one alternative implementation, the model update module 505 includes: The sample construction unit is used to construct training samples. The training samples include sample features, target comfort labels, and sample weights. The sample features include at least environmental parameters and time period information. The training update unit is used to add training samples to the historical training sample set and to train and update the personalized comfort target model based on the historical training sample set.

[0124] In one alternative implementation, the sample weights include at least one of the following: The time distance weighting is such that the closer the time of occurrence of the user operation event is to the time of occurrence of the automatic control event, the higher the time distance weighting. Stability weight: The faster the environment enters a stable state after a user action event, the higher the stability weight.

[0125] The user-operated adaptive air conditioning control device provided in this application can execute the user-operated adaptive air conditioning control method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0126] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0127] The following is a detailed reference. Figure 6 This diagram illustrates a suitable structural schematic for implementing the electronic device described in the embodiments of this application. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 601, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 602 or a program loaded from memory 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0128] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0129] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a memory 608, or installed from a ROM 602. When the computer program is executed by the processor 601, it performs the functions defined in the user-operated adaptive air conditioning control method of embodiments of this application.

[0130] Figure 6 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0131] This application also provides a computer-readable storage medium. The methods described above according to this application can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the user-operated adaptive air conditioning control method shown in the above embodiments is implemented.

[0132] A portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0133] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. An adaptive control method for air conditioning based on user operation, characterized in that, include: Acquire environmental status logs, system control logs, and user operation logs during the operation of the air conditioner; Using the automatic control events in the system control log as anchor points, a feedback identification time window is constructed for each automatic control event; Within the feedback recognition time window, user operation events related to the automatic control event are identified; After the user operation event occurs, the environmental state is monitored, and after the environmental state enters a stable state, a target comfort label is generated based on the stable environmental state. Based on the target comfort labels, update the personalized comfort target model; The operation of the air conditioner is controlled based on the updated personalized comfort target model.

2. The method according to claim 1, characterized in that, The step of constructing a feedback identification time window for each automatic control event, using the automatic control events in the system control log as anchor points, includes: Filter control events automatically executed by the air conditioning system from the system control log, and determine the actual effective time of the control event as the time anchor point; Starting from the time anchor point, a feedback recognition time window with a duration equal to the preset feedback window length is constructed, wherein the preset feedback window length is configured based on at least one of the control action type of the automatic control event and the thermal inertia of the room where the air conditioner is located.

3. The method according to claim 1, characterized in that, Also includes: When multiple automatic control events exist at the same time, the user operation event is associated with a target automatic control event among the multiple automatic control events. The target automatic control event is the control event whose occurrence time is closest to the occurrence time of the user operation event and is in an active state when the user operation event occurs.

4. The method according to claim 1, characterized in that, The step of identifying user operation events related to the automatic control event within the feedback identification time window includes: If a user operation event occurs within the feedback recognition time window, the user operation event is determined to be related to the automatic control event based on at least one of the following conditions: The operation parameters of the user operation event and the control parameters of the automatic control event belong to the same control dimension or are directly related dimensions. When the user operation event occurs, the control corresponding to the automatic control event is still in effect; The target device identifier corresponding to the user operation event is the same as the device identifier corresponding to the automatic control event.

5. The method according to claim 1, characterized in that, The step of monitoring the environmental state after the user operation event occurs, and generating a target comfort label based on the stable environmental state after the environmental state enters a stable state, includes: A stable observation window is started after the occurrence of the user operation event; Within the stable observation window, the rate of change of environmental parameters related to the user operation event is monitored; When the change in the environmental parameter is less than the corresponding threshold in multiple consecutive sampling periods, the environmental state is determined to have entered a stable state. Extract the statistics of the environmental state under the steady state, and use them as the steady environmental state; The target comfort label is calculated based on the stable environmental state.

6. The method according to claim 5, characterized in that, The calculation of the target comfort label based on the stable environmental state includes: A stable comfort value is calculated based on the stable environmental state, and this stable comfort value is used as the target comfort label; or... The target comfort range is determined based on the stable environmental state, and the target comfort range is used as the target comfort label.

7. The method according to claim 1, characterized in that, The process of updating the personalized comfort target model based on the target comfort label includes: Construct training samples, which include sample features, target comfort labels, and sample weights, wherein the sample features include at least environmental parameters and time period information; The training samples are added to the historical training sample set, and the personalized comfort target model is trained and updated based on the historical training sample set.

8. The method according to claim 7, characterized in that, The sample weights include at least one of the following: The time distance weight is higher as the time of occurrence of the user operation event is closer to the time of occurrence of the automatic control event. The stability weight is defined as follows: the faster the environment enters a stable state after the user operation event, the higher the stability weight.

9. An air conditioning adaptive control device based on user operation, characterized in that, include: The log acquisition module is used to acquire environmental status logs, system control logs, and user operation logs during the operation of the air conditioner. The window construction module is used to construct a feedback identification time window for each automatic control event, using the automatic control events in the system control log as anchor points. The feedback identification module is used to identify user operation events related to the automatic control event within the feedback identification time window; The tag generation module is used to monitor the environmental state after the user operation event occurs, and generate a target comfort tag based on the stable environmental state after the environmental state enters a stable state. The model update module is used to update the personalized comfort target model based on the target comfort label; A control module is used to control the operation of the air conditioner based on the updated personalized comfort target model.

10. An electronic device, characterized in that, include: The system includes a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the user-operated adaptive air conditioning control method according to any one of claims 1 to 8.

Images