Intelligent operation control method and device for air purifier

Abstract

The embodiment of the invention discloses an intelligent operation control method and device for an air purifier. According to the scheme, the intelligent operation control method comprises the steps that S1, indoor environment data and indoor behavior data of a user are collected; s2, the indoor environment data and the behavior data are preprocessed, and a joint feature matrix fusing the indoor environment data and the behavior data is constructed; s3, inputting the joint feature matrix into a pre-trained behavior environment joint prediction model for processing to obtain an air quality prediction value and a user state prediction value in a future specified time period; s4, according to the air quality predicted value and the user state predicted value, a preliminary control strategy is generated, the preliminary control strategy is optimized, a final intelligent control strategy is generated and executed, and decision output of the final intelligent control strategy at least comprises a target air speed gear, a target air exchange amount and a target state of a mute mode; wherein the reinforcement learning model performs strategy optimization based on a comprehensive reward function for balancing air quality, equipment energy consumption and operation noise.

Description

Technical Field

[0001] This application relates to the field of smart home technology, and in particular to a smart operation control method and device for an air purifier. Background Technology

[0002] With increasing public concern for a healthy living environment, air purifiers have become essential devices for improving indoor air quality in homes and offices. Current air purifier technology generally employs automated control schemes based on fixed thresholds or preset modes. These schemes automatically adjust the fan speed or switch operating modes by monitoring real-time concentrations of pollutants such as PM2.5, VOCs, and CO2, as well as temperature and humidity parameters. While these schemes can react to immediate detected pollution, their control logic is inherently low-dimensional and passive. The operational decisions rely solely on instantaneous readings from isolated environmental sensors, completely ignoring the dynamic factor of the behavioral patterns of people active indoors.

[0003] This "single-dimensional environmental parameter → passive response" control method has inherent flaws, mainly in two aspects. First, the lack of user participation in operational decisions makes it difficult to balance energy efficiency and comfort. For example, the equipment may continue to operate at high power after the user leaves the room, resulting in unnecessary energy waste. Or, while the user is sleeping at night, a continuously high fan speed will generate noise interference, affecting the quality of rest. The equipment cannot recognize the short-term changes in air quality indicated by the user's habitual activities such as opening windows or cooking, and therefore cannot adjust in advance to alleviate purification lag. Second, the system lacks the ability to jointly model and predict user status and environmental dynamics. Existing solutions cannot couple and analyze the user's presence time, activity intensity, and daily routine with the generation, diffusion, and dissipation trends of environmental pollutants, thus failing to achieve predictive control for the future and making it difficult to adjust personalized operating strategies according to individual preferences. Summary of the Invention

[0004] This specification provides an intelligent operation control method and device for an air purifier to solve at least one of the technical problems mentioned above.

[0005] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows: According to a first aspect of the embodiments of this specification, an intelligent operation control method for an air purifier is provided, comprising: S1. Collect indoor environmental data and user behavior data in the room. The behavior data includes a state parameter U(t) indicating whether the user is in the room. Wherein, U(t) = 1 indicates that the user is in the room at time step t; U(t) = 0 indicates that the user is not in the room at time step t. S2. Preprocess the indoor environment data and the behavioral data, and construct a joint feature matrix that integrates the two. ; S3, the joint feature matrix The pre-trained behavioral environment joint prediction model is processed to obtain the predicted air quality and user status values ​​for a specified future time period. S4. Based on the predicted air quality value and the predicted user status value, a preliminary control strategy is generated, and the preliminary control strategy is optimized using a reinforcement learning model to generate and execute a final intelligent control strategy. The decision output of the final intelligent control strategy includes at least the target wind speed level, the target air exchange rate, and the target status of the silent mode. The reinforcement learning model optimizes the strategy based on a comprehensive reward function that balances air quality, equipment energy consumption, and operating noise.

[0006] In some optional implementations, step S2 specifically includes: Construct an environmental data matrix based on the indoor environmental data. ,in Among them, symbols Indicates time step time Concentration, symbol Indicates time step The concentration of volatile organic compounds at that time, symbol Indicates time step Carbon dioxide concentration at time, symbol Indicates time step Humidity at that time; Construct a behavior data matrix based on the behavior data. ; wherein, the behavioral data matrix Includes user entry and exit status User activity intensity and night / day status ; The environmental data matrix With the behavioral data matrix The joint feature matrix is ​​obtained by concatenating the features. ; In step S3, the joint prediction model for behavior and environment is a Long Short-Term Memory (LSTM) network model, which is determined by the formula... and To make a prediction, where the symbol The joint feature matrix At time step eigenvectors, symbols Indicates the current time step Hidden state, symbol Indicates the previous time step The hidden state, For including future time The output vector contains air quality predictions and user status predictions.

[0007] In some optional implementations, step S4 involves generating the wind speed settings in the initial control strategy. The method is as follows: Among them, symbols Indicates prediction Concentration value, and For the preset threshold, the symbol Indicates the maximum operating wind speed of the equipment, symbol The lower limit of high-level wind speed, symbol This represents the lower limit of the medium-speed wind, indicated by the symbol. This is the minimum operating wind speed for the equipment; If it is currently nighttime, the calculated wind speed setting will be downgraded. The adjustment method is as follows: ,in The gear is calculated according to the formula. This is the downshift offset. This is the lowest setting.

[0008] In some alternative implementations, in step S4, the ventilation volume in the preliminary control strategy is generated. The method is as follows: in, This is the weighting coefficient; if the user is not in the room, i.e. Then, energy-saving adjustments are made to the ventilation volume: ,in .

[0009] In some optional implementations, step S4 involves generating the silent mode state in the initial control strategy. The method is as follows: in, Nighttime mode, For the user's presence status, This is the wind speed setting. Set to the lowest setting; and use a smooth switching algorithm to update the silent mode status: , where the symbol For the target state, This is the smoothing coefficient.

[0010] In some optional implementations, step S4 further includes filter life prediction management and closed-loop feedback steps, including: Based on the current lifespan of the filter Based on the predicted pollution load, calculate the predicted future filter life. The calculation formula is: in, This is the loss coefficient; when When needed, a filter replacement reminder will be generated; Furthermore, the final intelligent control strategy is adjusted in real time based on the error between the predicted air quality value and the actual measured value after execution.

[0011] In some optional implementations, the behavior-environment joint prediction model in step S3 and the reinforcement learning model in step S are trained and optimized by a cloud server. The method further includes step S5: the cloud server collects anonymized feature data from multiple air purifiers and performs centralized training and parameter updates on the behavior-environment joint prediction model and reinforcement learning model. And step S6: The cloud server sends the updated model parameters to each local air purifier; steps S3 and S4 are executed by the lightweight model on the local air purifier, and when the cloud server is unavailable, intelligent control operation is maintained based on the most recently sent model parameters.

[0012] In some optional implementations, the method further includes step S7: displaying information to the user through a user interface, the information including at least one of real-time air quality, air quality prediction, equipment operating status, filter life information, and intelligent control strategy; and receiving manual control commands input by the user through the user interface.

[0013] According to a second aspect of the embodiments of this specification, an intelligent operation control device for an air purifier is provided, comprising: The data acquisition module is used to collect indoor environmental data and user behavior data within the indoor space. The preprocessing and feature extraction module is used to preprocess the indoor environmental data and the behavioral data, and to construct a joint feature matrix that fuses the two. ; The data prediction module is used to process the joint feature matrix. The pre-trained behavioral environment joint prediction model is processed to obtain the predicted air quality and user status values ​​for a specified future time period. The intelligent control module is used to generate a preliminary control strategy based on the predicted air quality value and the predicted user status value, and to optimize the preliminary control strategy using a reinforcement learning model to generate and execute a final intelligent control strategy. The decision output of the final intelligent control strategy includes at least the target wind speed level, the target air exchange rate, and the target status of the silent mode. The reinforcement learning model optimizes the strategy based on a comprehensive reward function that balances air quality, equipment energy consumption, and operating noise.

[0014] One embodiment of this specification can achieve at least the following beneficial effects: This application's technical solution, by simultaneously collecting user behavior data and environmental parameter data, constructing a joint feature matrix of behavior and environment, and inputting it into a pre-trained joint prediction model of behavior and environment, can achieve joint prediction of future air quality and user presence. This transforms the traditional "single parameter - passive response" control mode into a "multi-source prediction - active regulation" mode. Then, based on the prediction results, a smart control strategy optimized using reinforcement learning is generated, enabling coordinated dynamic adjustment of wind speed, air exchange rate, and quiet mode. This effectively resolves the contradictions in traditional methods, such as purification lag and the difficulty in balancing energy efficiency and comfort, improving both energy efficiency and user experience while ensuring purification effectiveness. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of an intelligent operation control method for an air purifier provided in the embodiments of this specification; Figure 2 This is a complete architecture diagram of an intelligent operation control device for an air purifier provided in the embodiments of this specification; Figure 3 This is a structural diagram of the behavior-environment joint model used in the intelligent operation control method of the air purifier provided in the embodiments of this specification; Figure 4 This is a flowchart illustrating the cloud-local collaboration in the intelligent operation control method for an air purifier provided in the embodiments of this specification; Figure 5 This is a schematic diagram of the signaling interaction of each module in the intelligent operation control system of the air purifier in the technical solution of this application; Figure 6 For corresponding Figure 1 A schematic diagram of the structure of an intelligent operation control device for an air purifier. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of one or more embodiments of this specification.

[0018] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another.

[0019] Figure 1 This is a flowchart of an intelligent operation control method for an air purifier provided in the embodiments of this specification; Figure 2 This is a complete architecture diagram of an intelligent operation control device for an air purifier provided in the embodiments of this specification; Figure 3 This is a structural diagram of the behavior-environment joint model used in the intelligent operation control method of the air purifier provided in the embodiments of this specification; Figure 4 This is a flowchart illustrating the cloud-local collaboration in the intelligent operation control method for an air purifier provided in the embodiments of this specification.

[0020] This application provides an intelligent operation control method for an air purifier, such as... Figure 1 As shown, the method may include the following steps: Step S1: Collect indoor environmental data and user behavior data in the room. The behavior data includes a state parameter U(t) indicating whether the user is in the room. Wherein, U(t) = 1 indicates that the user is in the room at time step t; U(t) = 0 indicates that the user is not in the room at time step t.

[0021] This step involves real-time monitoring of multiple indoor air quality indicators, including PM2.5 (fine particulate matter) concentration, VOC (volatile organic compound) concentration, CO2 (carbon dioxide) concentration, and humidity. This data can be acquired through integrated air purifiers or dedicated sensor modules deployed indoors, forming a basic quantitative description of the environmental conditions. Simultaneously, behavioral data collection aims to capture users' activity patterns and presence indoors; sources may include, but are not limited to, […]. Figure 2The infrared motion sensor, door magnetic switch, presence sensing signal linked to the smart home system, or permission information indirectly obtained through user devices are shown. The collected behavioral data can at least cover the binary state of the user entering and leaving a specific room (such as being in or leaving the room), quantitative indicators characterizing the intensity of user activity, and night / day status markers based on time or lighting conditions.

[0022] S2: Preprocess the indoor environmental data and the behavioral data, and construct a joint feature matrix that integrates the two. .

[0023] The purpose of this step is to preprocess the collected indoor environmental data and user behavior data and construct a fused feature matrix. Preprocessing begins with the environmental data sequence to improve data regularity and comparability. This includes filtering raw sensor readings to suppress noise, interpolating missing values ​​due to transmission or sampling, and normalizing environmental parameters (such as PM2.5, VOC, CO2 concentration, and humidity) to scale them to a uniform numerical range. Simultaneously, user behavior data can be serialized and state-encoded. For example, discrete entry and exit events can be converted into continuous presence time-series signals, activity intensity can be quantified into values ​​within a specified interval, and each moment can be identified and labeled as nighttime or daytime based on time information. After preprocessing, the time-aligned environmental and behavioral data can be constructed into structured matrices. Each column of the environmental data matrix represents a time step, and each row corresponds to an environmental parameter. The behavioral data matrix organizes various behavioral states in a similar structure. Finally, by concatenating these two matrices along the feature dimension, a joint feature matrix that integrates multidimensional environmental information and user state information can be obtained. .

[0024] Step S3: Combine the joint feature matrix The pre-trained behavioral environment joint prediction model is processed to obtain air quality predictions and user status predictions for a specified future time period.

[0025] In this step, the joint feature matrix will be constructed. The data is input into a pre-trained joint prediction model for the behavioral environment. This model can be trained on a large amount of historical time-series data to learn the complex dynamic relationships and evolution patterns between environmental parameters and user behavior sequences. During processing, the trained model then... As input sequences, the system analyzes and fuses coupled information about the environment and behavior step-by-step through its internal structure (e.g., recurrent neural network units), capturing their cross-temporal dependencies. The final output is a prediction for a specified future time period (e.g., the next few tens of minutes to several hours). This result is presented as a vector and may include predicted air quality values ​​(e.g., predicted PM2.5, VOC, CO2 concentrations, and humidity) and predicted user status values ​​(e.g., user presence probability and activity intensity). Figure 3 As shown, Figure 3 This is a structural diagram of the behavioral-environment joint model used in the intelligent operation control method for the air purifier provided in the embodiments of this specification. Figure 3 We can determine what data to collect and how to process it if the model uses an LSTM model.

[0026] Step S4: Based on the predicted air quality value and the predicted user status value, generate a preliminary control strategy, and optimize the preliminary control strategy using a reinforcement learning model to generate and execute a final intelligent control strategy. The decision output of the final intelligent control strategy includes at least the target wind speed level, the target air exchange rate, and the target status of the silent mode. The reinforcement learning model optimizes the strategy based on a comprehensive reward function that balances air quality, equipment energy consumption, and operating noise.

[0027] In this step, a preliminary control strategy is first generated based on the air quality prediction and user status prediction obtained in step S3. This preliminary strategy is then input into a pre-trained reinforcement learning model for optimization. This model, through its built-in agent-environment interaction mechanism, evaluates the cumulative effect of different control actions over long-term operation and outputs an optimized final intelligent control strategy. The decision-making instructions of this strategy can at least explicitly specify the target fan speed, target air exchange rate, and target state of silent mode for future periods, in order to achieve multi-objective coordinated and forward-looking intelligent regulation of the air purifier's operating parameters.

[0028] This application's technical solution, by simultaneously collecting user behavior data and environmental parameter data, constructing a joint feature matrix of behavior and environment, and inputting it into a pre-trained joint prediction model of behavior and environment, can achieve joint prediction of future air quality and user presence. This transforms the traditional "single parameter - passive response" control mode into a "multi-source prediction - active regulation" mode. Then, based on the prediction results, a smart control strategy optimized using reinforcement learning is generated, enabling coordinated dynamic adjustment of wind speed, air exchange rate, and quiet mode. This effectively resolves the contradictions in traditional methods, such as purification lag and the difficulty in balancing energy efficiency and comfort, improving both energy efficiency and user experience while ensuring purification effectiveness.

[0029] Based on the technical solutions described above, this specification also provides some specific implementation schemes, which are described below.

[0030] In some optional implementations, step S2 may specifically include: Construct an environmental data matrix based on the indoor environmental data. ,in Among them, symbols Indicates time step time Concentration, symbol Indicates time step The concentration of volatile organic compounds at that time, symbol Indicates time step Carbon dioxide concentration at time, symbol Indicates time step Humidity at that time; Construct a behavior data matrix based on the behavior data. ; wherein, the behavioral data matrix Includes user entry and exit status User activity intensity and night / day status ; The environmental data matrix With the behavioral data matrix The joint feature matrix is ​​obtained by concatenating the features. ; In step S3, the joint prediction model for behavior and environment is a Long Short-Term Memory (LSTM) network model, which is determined by the formula... and To make a prediction, where the symbol The joint feature matrix At time step eigenvectors, symbols Indicates the current time step Hidden state, symbol Indicates the previous time step The hidden state, For including future time The output vector contains air quality predictions and user status predictions.

[0031] In this embodiment, in step S2, the collected multi-source heterogeneous data is first organized into a unified time-series matrix format to provide structured input for the subsequent joint prediction model. Specifically, for indoor environmental data, it is constructed into an environmental data matrix. Each time step corresponding feature vector It consists of four relatively important environmental indicators, namely This represents the concentration of fine particulate matter in the air at that moment. This represents the concentration of volatile organic compounds. Represents the concentration of carbon dioxide, and Represents air humidity. Matrix All observation vectors from time 1 to time T are stacked sequentially, thus providing a complete characterization of the continuous trajectory of environmental parameters evolving over time. For user behavior data, this embodiment simultaneously constructs a behavior data matrix. Each row of the matrix corresponds to a time step and contains three dimensions of behavioral state encoding, namely the user's entry and exit from the room. It is a binary variable (1 indicates the user is in the room, 0 indicates the user is away), used to characterize the user's spatial presence; user activity intensity. It is a continuous or discrete value normalized to the interval [0,1] (0 represents stillness, 1 represents high-intensity activity), used to quantify the degree of pollutant release or diffusion that user activities may cause; night / day status Similarly, a binary variable (e.g., 1 for nighttime, 0 for daytime) is used to identify time patterns with different noise tolerances and purification requirements. Finally, the environmental data matrix is ​​concatenated using a matrix stitching operation. With behavioral data matrix Merge along the feature dimensions to generate a joint feature matrix. This operation can align and merge the originally independent environmental sequences and behavioral sequences on the same timeline, forming a multidimensional time series that simultaneously includes both environmental and user states.

[0032] The behavior-environment joint prediction model used in step S3 can be a Long Short-Term Memory (LSTM) network. The operating mechanism of this model can be described by two formulas. Specifically, the first formula... The update process of the model's internal state is defined, at each time step. The model receives data from the joint feature matrix. The feature vector at the current time step As input, and combined with the hidden state passed down from the previous time step. The computation is performed through a complex gating mechanism within the LSTM unit (including input gates, forget gates, and output gates). This process selectively remembers long-term historical information, forgets irrelevant information, and incorporates new observational data, ultimately outputting the hidden state representing the integrated memory and information at the current moment. The second formula This defines the process for generating the predicted output, which is the hidden state calculated at the current time step. Input to an output function In the middle, the mapping generates a future specified time period. Prediction vector The output vector contains both the predicted air quality values ​​for future times (such as the predicted concentrations of PM2.5, VOCs, CO2, and humidity) and the predicted user status values ​​(such as the predicted probability of whether the user is present).

[0033] In some optional implementations, step S4 involves generating the wind speed settings in the initial control strategy. The methods can be: Among them, symbols Indicates prediction Concentration value, and For the preset threshold, the symbol Indicates the maximum operating wind speed of the equipment, symbol The lower limit of high-level wind speed, symbol This represents the lower limit of the medium-speed wind, indicated by the symbol. This is the minimum operating wind speed for the equipment; If it is currently nighttime, the calculated wind speed setting will be downgraded. The adjustment method is as follows: ,in The gear is calculated according to the formula. This is the downshift offset. This is the lowest setting.

[0034] In the technical solution of this embodiment, the wind speed setting... The generation of PM2.5 is mainly based on the predicted values ​​of PM2.5 concentrations for future periods. And the user's real-time status in the room This comprehensive judgment is achieved through the piecewise function mentioned above, specifically when the predicted PM2.5 concentration is higher than or equal to a preset high threshold. And the user is currently in the room (i.e. When setting the fan speed to the high setting, that is, between the preset lower limit of the high setting, the fan speed will be adjusted accordingly. With the equipment's maximum wind speed This is to ensure strong purification capabilities when pollution is severe and users are present. When the predicted PM2.5 concentration is within the middle threshold... When the wind speed is between the high threshold and the user is present, the wind speed setting will be in the medium range, i.e., between the lower limit of the medium range. With the lower limit of the high level The goal is to balance purification effectiveness with operational noise and energy consumption. When the predicted PM2.5 concentration is below the medium threshold, or the user is not currently indoors (i.e.,...), the system will be optimized to achieve the desired purification effect while minimizing noise and energy consumption. When the wind speed setting is low, that is, between the minimum wind speed of the equipment, the wind speed setting should be set to a low setting. With the lower limit of the high level This design allows the control strategy to proactively adjust based on future pollution trends and the actual situation of the user, avoiding the lag and blindness of traditional methods that rely solely on the current concentration threshold for response. This provides sufficient purification intensity when needed by the user and automatically reduces operating power to achieve energy savings when the user leaves or pollution levels are low.

[0035] Furthermore, to accommodate the varying noise sensitivity of users at different times, especially the need for quiet operation during sleep at night, this embodiment also introduces a dynamic adjustment mechanism based on time status. That is, if it is determined that the current state is nighttime, the noise level will be adjusted to the initial setting calculated by the piecewise function. The gear is downshifted based on the existing gear. The specific adjustment method is as follows: .in, The preset downshift offset represents the number of gears that need to be downshifted at night. This is the lowest permissible operating speed. Through this mechanism, even when higher pollution levels are predicted at night and a higher speed would normally be triggered, the user's sleep comfort will be prioritized, and the fan speed will be proactively reduced by one or more speeds. This maintains basic purification capabilities while reducing the noise disturbance to the user.

[0036] In some alternative implementations, in step S4, the ventilation volume in the preliminary control strategy is generated. The methods can be: in, This is the weighting coefficient; if the user is not in the room, i.e. Then, energy-saving adjustments are made to the ventilation volume: ,in .

[0037] In the technical solution of this embodiment , and These represent the predicted concentrations of PM2.5, volatile organic compounds, and carbon dioxide for a specified future time period, obtained through a joint behavioral-environment prediction model; while , and These are the corresponding preset weighting coefficients, which can be used to quantify the impact of different types of pollutants on the required ventilation volume. Their specific values ​​can be preset through experiments or model optimization based on the volume of different rooms, the health hazard weight of pollutants, or the purification efficiency of the equipment. By comprehensively calculating the predicted values ​​of multiple pollutants in a weighted sum form, this strategy can overcome the limitations of single-parameter control, and can more comprehensively and evenly address complex air pollution, ensuring that the calculated ventilation volume command matches the expected comprehensive pollution load in the future. Furthermore, in order to optimize energy utilization while ensuring air quality, this strategy introduces an energy-saving adjustment logic, that is, when the system detects that the user is currently not in the room (i.e., When this happens, the initially calculated ventilation volume will be affected. To reduce the ratio, adjust the formula as follows: ,in This is an energy-saving adjustment coefficient between 0 and 1. This adjustment is based on the premise that when the room is unoccupied, the immediate requirements for air quality can be appropriately relaxed, eliminating the need to maintain the same air exchange rate as when someone is present. This allows for energy savings by reducing fan power consumption. The technical solution in this embodiment can map multi-pollutant predictions to corresponding airflow commands and dynamically optimize based on real-time feedback from the user's presence.

[0038] In some optional implementations, step S4 involves generating the silent mode state in the initial control strategy. The method is as follows: in, Nighttime mode, For the user's presence status, This is the wind speed setting. Set to the lowest setting; and use a smooth switching algorithm to update the silent mode status: , where the symbol For the target state, This is the smoothing coefficient.

[0039] In the technical solution of this embodiment, the silent mode state The generation of this mechanism aims to intelligently manage the noise output of devices based on environmental time periods, user status, and device operating intensity. Its logic includes two stages: target state determination and smooth switching. First, the target state is determined based on a multi-condition combination rule, namely, a state is determined if and only if the current state is simultaneously nighttime (…). The user is in the room. And the wind speed settings generated based on the prediction results. Higher than the device's preset lowest setting The system will only enter the target state of silent mode when these three conditions are met. Set to "On"; otherwise, set all target states to "Off".

[0040] This rule in the technical solution of this embodiment can meet the user's need for a low-noise environment when resting at night and the equipment needs to operate at a speed higher than the lowest setting. It ensures that the silent mode is actively activated to reduce operating noise in this scenario. During the day, when the user leaves the premises, or when the equipment only needs to operate at the lowest setting, the silent mode is turned off to prioritize the equipment's full purification capabilities. Secondly, to avoid sudden changes in fan speed or uncomfortable noise jumps that might occur when the silent mode abruptly switches between "on" and "off" states, this technical solution also introduces a smooth switching algorithm: In this algorithm, This represents the current state value of the silent mode. The target state value determined by the above rules. It is a smoothing coefficient between 0 and 1. This formula allows the output of the state at the next time step to be determined. It is the weighted average of the current state and the target state. This can be achieved by appropriately setting the smoothing coefficient. The value of the variable (e.g., a small positive number) can control the rate of state switching, thereby achieving a smooth and gradual transition from the current state to the target state. This smoothing mechanism in the technical solution of this embodiment can effectively suppress frequent and rapid mode switching caused by instantaneous fluctuations in environmental parameters or accidental changes in user state. It not only improves the smoothness of changes in equipment operating noise and the comfort of the user experience, but also reduces frequent impacts on the fan control system, which is beneficial to the stable operation of the equipment.

[0041] In some optional implementations, step S4 further includes filter life prediction management and closed-loop feedback steps, which may include: Based on the current lifespan of the filter Based on the predicted pollution load, calculate the predicted future filter life. The calculation formula is: in, This is the loss coefficient; when When needed, a filter replacement reminder will be generated; Furthermore, the final intelligent control strategy is adjusted in real time based on the error between the predicted air quality value and the actual measured value after execution.

[0042] In this embodiment, the filter life prediction management is based on the estimation of the future cumulative load of pollutants. In this model, filter wear is considered a continuous process, and its consumption rate is proportional to the concentration of major particulate matter and gaseous pollutants in the air flowing through the filter. The formula... and Represents the current moment To the future Within this range, the predicted curves of PM2.5 concentration and volatile organic compound concentration over time, derived from the behavioral-environment joint prediction model. (Coefficient) and These are pre-calibrated loss coefficients, which quantify the loss to filter lifespan caused by unit concentrations of PM2.5 and VOCs. By integrating the predicted concentration curve over a time interval, the model can cumulatively calculate the filter's lifespan in future periods. The total pollutant load that the interior is expected to bear, and its remaining lifespan from the currently known total lifespan. Subtracting from the mean yields a forward-looking lifespan prediction. When this prediction falls below a preset safety threshold... When the filter wears out, the system will automatically generate a filter replacement reminder. Compared to the traditional method of estimating based solely on runtime or cumulative airflow, this method more accurately reflects the actual wear and tear of the filter in real-world polluted environments, preventing premature or delayed filter replacement due to fluctuations in pollution load.

[0043] Meanwhile, this embodiment also dynamically calibrates and optimizes the intelligent control strategy by continuously comparing the error between the predicted air quality value and the actual measured value after the strategy is executed. In this mechanism, when the final intelligent control strategy (such as the set wind speed and ventilation rate) generated based on the prediction is executed, the corresponding actual air quality data (such as PM2.5 concentration) is collected. Subsequently, the actual measured value is compared with the predicted value of the corresponding period in real time to calculate the prediction error (such as absolute error or root mean square error). This error signal is used in the subsequent generation and optimization process of the control strategy. If the error remains at a low level, it indicates that the prediction model is accurate and the control strategy is effective; if the error is significant and continues to deviate, it indicates that there is a deviation between the actual environmental dynamics and the prediction, or that the execution effect of the control strategy has not met expectations. At this time, the control strategy will be corrected in real time according to the preset adjustment algorithm (such as proportional-integral adjustment or rule-based parameter fine-tuning method). For example, the concentration threshold on which the wind speed level decision is based can be adaptively fine-tuned according to the magnitude and direction of the error, or the weight coefficient in the ventilation rate calculation formula can be adjusted. The scheme in this embodiment not only relies on forward prediction, but also has the ability to correct backward based on actual feedback, thereby effectively overcoming the control performance degradation caused by initial model errors, sudden environmental disturbances, or slow changes in equipment performance.

[0044] In some optional implementations, the behavior-environment joint prediction model in step S3 and the reinforcement learning model in step S4 are trained and optimized by a cloud server. The method further includes step S5: the cloud server collects anonymized feature data from multiple air purifiers and performs centralized training and parameter updates on the behavior-environment joint prediction model and reinforcement learning model. And step S6: The cloud server sends the updated model parameters to each local air purifier; steps S3 and S4 are executed by the lightweight model on the local air purifier, and when the cloud server is unavailable, intelligent control operation is maintained based on the most recently sent model parameters.

[0045] In this embodiment, the training and parameter optimization processes of the computationally complex "joint prediction model of behavior and environment" and the "reinforcement learning model," which require a large amount of data, are deployed on a cloud server. The specific logic can be found in the appendix. Figure 4 As shown. To support this training process, in this embodiment, the cloud server collects anonymized feature data (i.e., joint feature matrices and other data with specific user identity information desensitized) from a massive number of air purifier terminal devices connected to the network. Using this vast dataset aggregated from diverse real-world scenarios, the model undergoes centralized and periodic deep training and parameter updates. This swarm intelligence-based learning approach improves the accuracy of model predictions and the universality of the strategy. Subsequently, in step S6, the cloud server distributes the latest model parameters (such as neural network weights, reinforcement learning strategy network parameters, etc.) after training convergence and optimization to the local control units of each air purifier via the network. Under normal operation, steps S3 (joint prediction based on input features) and S4 (generating and optimizing control strategies) are executed by a lightweight model deployed locally on the air purifier, which has been trimmed and optimized, ensuring real-time control response and low latency. Meanwhile, when the device loses connection to the cloud server due to network failure or other reasons, the local system can automatically switch to offline mode and drive the local lightweight model based on the model parameters that were most recently successfully received and stored, thereby maintaining the basic functions of intelligent control.

[0046] In some optional implementations, the method may further include step S7: displaying information to the user through a user interface, the information including at least one of real-time air quality, air quality prediction, equipment operating status, filter life information, and intelligent control strategy; and receiving manual control commands input by the user through the user interface.

[0047] like Figure 5As shown, this is a schematic diagram of the signaling interaction of each module of the intelligent operation control system for an air purifier in the technical solution of this application. The following explanation is based on this diagram and describes the complete implementation process of the intelligent operation control method in this embodiment and the interaction relationship of each module.

[0048] After system startup, the first step is "Phase One: Data Acquisition and Model Upload." The environmental data acquisition module first collects environmental parameters such as PM2.5, VOC, CO2, temperature, and humidity. Simultaneously, the user behavior sensing source collects behavioral data such as user entry and exit from rooms, activity intensity, and day / night status. After time-series processing, the two types of data are concatenated into a joint feature matrix X_seq, which is then uploaded to the cloud learning module. The cloud learning module cleans the data uploaded from multiple devices, trains and updates the behavior-environment joint model, and then distributes the optimized model parameters and policy weights to the local inference and real-time control module to complete the initial configuration of the model and policy.

[0049] After initialization, the system enters a continuous running loop marked "loop" to execute the "Phase Two: Local Inference and Policy Generation" process. The local inference and real-time control modules load the latest joint feature matrix. The system processes the input behavior and environment joint model, outputting future air quality predictions and user status predictions. Based on these predictions, the intelligent control strategy module calculates control parameters such as target fan speed, air exchange rate, silent mode status, and filter operating status, generates control commands, and sends them to the air purifier actuator. Upon receiving the commands, the air purifier actuator performs operations such as adjusting fan speed and air exchange rate, while simultaneously collecting real-time environmental data after execution, forming a closed-loop logic of "prediction-control-feedback".

[0050] The "Phase Three: User Interaction and Continuous Optimization" process is executed in parallel within the main control loop. The user interaction module synchronously receives information such as device operating status and filter lifespan, displays real-time air quality, predicted air quality values, device operating status, and filter lifespan information to the user, and receives manually input control commands. The local inference and real-time control module reports data packets containing environmental parameters, control strategies, and user feedback from each run to the cloud learning module. This module continuously optimizes model parameters and strategy weights by analyzing historical data and sends the updated parameters back to the local system. Simultaneously, when the predicted filter lifespan falls below a preset threshold, the system automatically sends a filter replacement reminder to the user, achieving intelligent operation and continuous iteration across the entire process, ensuring a dynamic balance between purification effect, energy consumption, and user experience.

[0051] It should be understood that in the methods described in one or more embodiments of this specification, the order of some steps may be adjusted according to actual needs, or some steps may be omitted.

[0052] Based on the foregoing technical solutions, the present invention also provides an intelligent operation control device for an air purifier, such as... Figure 6 As shown, the device, from a macroscopic perspective, may include the following modules: Data acquisition module 602 is used to collect indoor environmental data and user behavior data in the indoor environment; The preprocessing and feature extraction module 604 is used to preprocess the indoor environmental data and the behavioral data, and construct a joint feature matrix that fuses the two. ; Data prediction module 606 is used to process the joint feature matrix The pre-trained behavioral environment joint prediction model is processed to obtain the predicted air quality and user status values ​​for a specified future time period. The intelligent control module 608 is used to generate a preliminary control strategy based on the predicted air quality value and the predicted user status value, and to optimize the preliminary control strategy using a reinforcement learning model to generate and execute a final intelligent control strategy. The decision output of the final intelligent control strategy includes at least the target wind speed level, the target air exchange rate, and the target status of the silent mode. The reinforcement learning model optimizes the strategy based on a comprehensive reward function that balances air quality, equipment energy consumption, and operating noise.

[0053] Those skilled in the art will understand that the modules in the apparatus of the foregoing embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules, that is, the module division can be flexibly performed to implement the method embodiments described above.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent operation control of an air purifier, characterized in that, The method includes: S1. Collect indoor environmental data and user behavior data in the room. The behavior data includes a state parameter U(t) indicating whether the user is in the room. Wherein, U(t) = 1 indicates that the user is in the room at time step t; U(t) = 0 indicates that the user is not in the room at time step t. S2. Preprocess the indoor environment data and the behavioral data, and construct a joint feature matrix that integrates the two. ; S3, the joint feature matrix The pre-trained behavioral environment joint prediction model is processed to obtain the predicted air quality and user status values ​​for a specified future time period. S4. Based on the predicted air quality value and the predicted user status value, a preliminary control strategy is generated, and the preliminary control strategy is optimized using a reinforcement learning model to generate and execute a final intelligent control strategy. The decision output of the final intelligent control strategy includes at least the target wind speed level, the target air exchange rate, and the target status of the silent mode. The reinforcement learning model optimizes the strategy based on a comprehensive reward function that balances air quality, equipment energy consumption, and operating noise.

2. The intelligent operation control method for an air purifier according to claim 1, characterized in that, Step S2 specifically includes: Construct an environmental data matrix based on the indoor environmental data. ,in Among them, symbols Indicates time step time Concentration, symbol Indicates time step The concentration of volatile organic compounds at that time, symbol Indicates time step Carbon dioxide concentration at time, symbol Indicates time step Humidity at that time; Construct a behavior data matrix based on the behavior data. ; wherein, the behavioral data matrix Includes user entry and exit status User activity intensity and night / day status ; The environmental data matrix With the behavioral data matrix The joint feature matrix is ​​obtained by concatenating the features. ; In step S3, the joint prediction model for behavior and environment is a Long Short-Term Memory (LSTM) network model, which is determined by the formula... and To make a prediction, where the symbol The joint feature matrix At time step eigenvectors, symbols Indicates the current time step Hidden state, symbol Indicates the previous time step The hidden state, For including future time The output vector contains air quality predictions and user status predictions.

3. The intelligent operation control method for an air purifier according to claim 1, characterized in that, In step S4, the wind speed settings in the preliminary control strategy are generated. The method is as follows: Among them, symbols Indicates prediction Concentration value, and For the preset threshold, the symbol Indicates the maximum operating wind speed of the equipment, symbol The lower limit of high-level wind speed, symbol This represents the lower limit of the medium-speed wind, indicated by the symbol. This is the minimum operating wind speed for the equipment; If it is currently nighttime, the calculated wind speed setting will be downgraded. The adjustment method is as follows: ,in The gear is calculated according to the formula. This is the downshift offset. This is the lowest setting.

4. The intelligent operation control method for an air purifier according to claim 1, characterized in that, In step S4, the ventilation volume in the preliminary control strategy is generated. The method is as follows: in, This is the weighting coefficient; if the user is not in the room, i.e. Then, energy-saving adjustments are made to the ventilation volume: ,in .

5. The intelligent operation control method for an air purifier according to claim 1, characterized in that, In step S4, the silent mode state in the preliminary control strategy is generated. The method is as follows: in, Nighttime mode, For the user's presence status, This is the wind speed setting. Set to the lowest setting; and use a smooth switching algorithm to update the silent mode status: , where the symbol For the target state, This is the smoothing coefficient.

6. The intelligent operation control method for an air purifier according to claim 1, characterized in that, Step S4 further includes filter life prediction management and closed-loop feedback steps, including: Based on the current lifespan of the filter Based on the predicted pollution load, calculate the predicted future filter life. The calculation formula is: in, This is the loss coefficient; when When needed, a filter replacement reminder will be generated; Furthermore, the final intelligent control strategy is adjusted in real time based on the error between the predicted air quality value and the actual measured value after execution.

7. The intelligent operation control method for an air purifier according to claim 1, characterized in that, The behavior-environment joint prediction model in step S3 and the reinforcement learning model in step S4 are trained and optimized by a cloud server. The method further includes step S5: the cloud server collects anonymized feature data from multiple air purifiers and performs centralized training and parameter updates on the behavior-environment joint prediction model and reinforcement learning model. And step S6: The cloud server sends the updated model parameters to each local air purifier; steps S3 and S4 are executed by the lightweight model on the local air purifier, and when the cloud server is unavailable, intelligent control operation is maintained based on the most recently sent model parameters.

8. The intelligent operation control method for an air purifier according to claim 1, characterized in that, The method further includes step S7: displaying information to the user through a user interface, the information including at least one of real-time air quality, air quality prediction, equipment operating status, filter life information, and intelligent control strategy; and receiving manual control commands input by the user through the user interface.

9. An intelligent operation control device for an air purifier, characterized in that, The device includes: The data acquisition module is used to collect indoor environmental data and user behavior data within the indoor space. The preprocessing and feature extraction module is used to preprocess the indoor environmental data and the behavioral data, and to construct a joint feature matrix that fuses the two. ; The data prediction module is used to process the joint feature matrix. The pre-trained behavioral environment joint prediction model is processed to obtain the predicted air quality and user status values ​​for a specified future time period. The intelligent control module is used to generate a preliminary control strategy based on the predicted air quality value and the predicted user status value, and to optimize the preliminary control strategy using a reinforcement learning model to generate and execute a final intelligent control strategy. The decision output of the final intelligent control strategy includes at least the target wind speed level, the target air exchange rate, and the target status of the silent mode. The reinforcement learning model optimizes the strategy based on a comprehensive reward function that balances air quality, equipment energy consumption, and operating noise.

Images