Personalized control method and device for air conditioner in passenger compartment, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a passenger compartment air conditioner personalized control method and device, electronic equipment and a storage medium, and relates to the technical field of new energy automobile air conditioner systems. The method comprises the steps that passenger compartment environment parameters, passenger biological characteristic data and air conditioner adjusting behavior data are collected, and a universal control strategy is generated; based on the passenger compartment environment parameters, the passenger biological characteristic data and the air conditioner adjusting behavior data, user personalized data are obtained; and obtaining an action vector increment based on the user personalized data, and carrying out personalized adjustment on the universal control strategy based on the action vector increment to obtain a personalized control strategy. According to the method, the universal control strategy is generated based on the generality of the personalized data, personalized adjustment is carried out based on the personalized data, the personalized control strategy is generated, the personalized requirements of passengers are met, and the problem that the personalized requirements of different passengers are difficult to meet due to the fact that a fixed temperature set value and a unified control strategy are adopted in an existing method is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new energy vehicle air conditioning systems, in particular to a passenger cabin air conditioning individualized control method and device, electronic equipment and storage medium. BACKGROUND

[0002] At present, the new energy vehicle air conditioning system mainly adopts a control method based on real vehicle calibration, and realizes passenger cabin temperature regulation by querying a preset calibration table and combining a PID control algorithm. The existing technology usually relies on fixed temperature set values and unified control strategies, and it is difficult to meet the individualized needs of different passengers for temperature, air speed and air outlet mode. SUMMARY

[0003] The purpose of the embodiments of the present application is to provide a passenger cabin air conditioning individualized control method and device, electronic equipment and storage medium, generate a general control strategy based on the commonness of individualized data, and make individualized adjustments based on individualized data to generate an individualized control strategy, meet the individualized needs of passengers, and solve the problem that the existing method uses fixed temperature set values and unified control strategies, which is difficult to meet the individualized needs of different passengers.

[0004] In a first aspect, the present application provides a passenger cabin air conditioning individualized control method, which comprises: collecting passenger cabin environmental parameters, passenger biological feature data and air conditioning adjustment behavior data; generating a general control strategy based on the passenger cabin environmental parameters, passenger biological feature data and air conditioning adjustment behavior data; obtaining user individualized data based on the passenger cabin environmental parameters, passenger biological feature data and air conditioning adjustment behavior data; obtaining an action vector increment based on the user individualized data, and making individualized adjustments to the general control strategy based on the action vector increment to obtain an individualized control strategy; and controlling the passenger cabin air conditioning to operate based on the individualized control strategy.

[0005] In the technical solution of the embodiments of the present application, the commonness of individualized data is learned, the data is aggregated by de-individualization, a general control strategy is generated, and individualized adjustments are made based on individualized data to generate an individualized control strategy, meet the individualized needs of passengers, and solve the problem that the existing method uses fixed temperature set values and unified control strategies, which is difficult to meet the individualized needs of different passengers.

[0006] In some embodiments, the generating the general control strategy based on the passenger cabin environment parameters, passenger biological feature data and air conditioner adjustment behavior data comprises: generating a state vector based on the passenger cabin environment parameters, passenger biological feature data and current air conditioner state parameters; generating an action vector based on the air conditioner adjustment behavior data; generating a reward function based on the action vector and power of the air conditioner; and performing reinforcement learning using a reinforcement learning algorithm with the reward function as an optimization target, and generating the general control strategy based on the state vector. The multi-source perception data such as the passenger cabin environment parameters, passenger biological feature data and air conditioner adjustment behavior data are used as a basis to provide a comprehensive data basis and basis, so that the obtained general control strategy is more scientific and reasonable.

[0007] In some embodiments, the performing reinforcement learning using a reinforcement learning algorithm with the reward function as an optimization target, and generating the general control strategy based on the state vector comprises: obtaining control parameters of similar users based on cloud collaboration or clustering algorithm; and performing evolutionary self-learning based on the control parameters and the state vector to obtain the general control strategy. The reinforcement learning algorithm is used to find users with similar comfort through the collaborative filtering or clustering algorithm of the cloud, and then gradually capture the personalized data for evolutionary self-learning, so as to provide a basic control strategy for all users by learning the general decision rule.

[0008] In some embodiments, the obtaining an action vector increment based on the user personalized data, and performing personalized adjustment on the general control strategy based on the action vector increment to obtain a personalized control strategy comprises: quantifying user behavior quantity based on the personalized data; determining a personalized adapter based on the user behavior quantity; obtaining an action vector increment using the personalized data and the personalized adapter; and performing personalized adjustment on the general control strategy based on the action vector increment to obtain a personalized control strategy. The personalized adapter is used as a tool to fine-tune the general control strategy according to the personalized data of the user to obtain a personalized control strategy to meet the personalized needs of the user.

[0009] In some embodiments, the determining a personalized adapter based on the user behavior quantity comprises: if the user behavior quantity is less than a first threshold value, the model parameter quantity that needs to be adjusted is 0; if the user behavior quantity is between the first threshold value and a second threshold value, the model parameter quantity is adjusted in a fine-tuning manner; if the user behavior quantity is between the second threshold value and a third threshold value, the model parameter quantity is adjusted based on a low-rank matrix combined with small-scale playback data; and if the user behavior quantity is greater than the third threshold value, all model parameter quantities are adjusted. The model parameter adjustment quantity is determined by the user behavior quantity, so that the update frequency of the model parameters can be adjusted according to the actual needs.

[0010] In some embodiments, obtaining action vector increments using the personalized data and the personalized adapter includes: quantizing the action vector increments according to the data type and preference weights of the personalized data. Mapping the personalized data to specific action vector increments makes the personalized adjustments more aligned with user preferences.

[0011] In some embodiments, quantifying the action vector increment based on the data type and preference weight of the personalized data includes: calculating a one-dimensional action vector increment based on the data type and corresponding preference weight of the personalized data, and obtaining the action vector increment based on the one-dimensional action vector increment; if there is dynamic adjustment, updating the mapping coefficient, and obtaining the one-dimensional dynamic action vector increment based on the updated mapping coefficient, wherein the initial value of the mapping coefficient is the corresponding preference weight. Determining the one-dimensional action vector increment based on the data type and preference weight achieves the quantization of the action vector increment based on user personalized data.

[0012] Secondly, this application provides a personalized control device for passenger cabin air conditioning. The device includes: a data acquisition module for collecting passenger cabin environmental parameters, passenger biometric data, and air conditioning adjustment behavior data; a general control strategy generation module for generating a general control strategy based on the passenger cabin environmental parameters, passenger biometric data, and air conditioning adjustment behavior data; a personalized data acquisition module for acquiring user-specific data based on the passenger cabin environmental parameters, passenger biometric data, and air conditioning adjustment behavior data; a personalized adjustment module for adjusting the general control strategy based on the user-specific data to obtain a personalized control strategy; and an execution module for controlling the operation of the passenger cabin air conditioning based on the personalized control strategy. This device dynamically adjusts the air conditioning control parameters according to the individual characteristics of each passenger, achieving personalized comfort adjustment tailored to each individual, significantly improving the passenger comfort experience.

[0013] Thirdly, this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the above-described personalized control method for passenger cabin air conditioning.

[0014] Fourthly, this application provides a readable storage medium storing computer program instructions, which are read and executed by a processor to perform the aforementioned personalized control method for passenger cabin air conditioning. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating a personalized control method for passenger cabin air conditioning provided in this application embodiment; Figure 2 A flowchart illustrating personalized control based on a shared master model and a user-level lightweight adapter, provided for embodiments of this application; Figure 3 A flowchart for generating a general control strategy provided in the embodiments of this application; Figure 4 A flowchart illustrating the specific generation process of the general control strategy provided in the embodiments of this application; Figure 5 A flowchart illustrating the generation of personalized control strategies provided in this application embodiment; Figure 6 A flowchart of partition control provided for embodiments of this application; Figure 7 A flowchart illustrating the implementation of personalized control in embodiments of this application; Figure 8 A schematic diagram of the temperature change curve after adaptive adjustment using an AI reinforcement learning algorithm, provided as an embodiment of this application; Figure 9 This is a structural block diagram of a personalized control device for passenger cabin air conditioning provided in an embodiment of this application.

[0017] icon: 110 - Data acquisition module; 120 - General control strategy generation module; 130 - Personalized data acquisition module; 140 - Personalized adjustment module; 150 - Execution module. Detailed Implementation

[0018] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0019] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0020] With the intelligent development of new energy vehicles, traditional air conditioning system control methods can no longer meet the comfort needs of passengers. Existing control methods mainly adopt vehicle-calibrated control methods, which use preset calibration tables and PID control algorithms to regulate the temperature of the passenger compartment. This method usually relies on fixed temperature setpoints and uniform control strategies, making it difficult to meet the personalized needs of different passengers for temperature, fan speed, and airflow mode.

[0021] To address the aforementioned technical issues, this application provides a personalized control method for passenger cabin air conditioning. This method is based on multi-dimensional perception data fusion, utilizes AI reinforcement learning to generate general decision-making rules, generates a general control strategy, and uses personalized data to adjust the general control strategy to obtain a personalized control strategy that meets the personalized needs of passengers. This solves the problem that existing methods, which use fixed temperature setpoints and uniform control strategies, are unable to meet the personalized needs of different passengers.

[0022] Please refer to Figure 1 , Figure 1 A flowchart of a personalized control method for passenger cabin air conditioning provided in this application embodiment is included, the method comprising the following steps: S110: Collects passenger cabin environmental parameters, passenger biometric data, and air conditioning regulation behavior data; S120: Generates a general control strategy based on crew cabin environmental parameters, crew biometric data, and air conditioning regulation behavior data; S130: Personalized user data is obtained based on passenger cabin environmental parameters, passenger biometric data, and air conditioning adjustment behavior data; S140: Obtain the action vector increment based on user personalized data, and adjust the general control strategy in a personalized manner based on the action vector increment to obtain a personalized control strategy. S150: Controls the operation of the passenger cabin air conditioning based on personalized control strategies.

[0023] Please refer to Figure 2 , Figure 2This is a flowchart illustrating personalized control based on a shared master model and a user-level lightweight adapter. Environmental data, user behavior data, and biometric data are acquired through multi-source sensing devices and input into the shared master model. This model learns general rules to generate general strategies, which are then generated by a personalized adapter and executed by the air conditioning execution module. The personalized adapter can select the fine-tuning method for model parameters based on user behavior data. For example, various data within the passenger cabin can be acquired through multi-source sensing devices, including but not limited to passenger cabin environmental parameters, occupant biometric data, and air conditioning adjustment behavior data. Passenger cabin environmental parameters can be acquired through environmental sensors such as temperature and humidity sensors and light sensors. Occupant biometric data can be acquired through devices such as cameras and infrared sensors, capturing occupant BMI, age, clothing thickness, body temperature, and movement data. Air conditioning adjustment behavior data records user actions on the air conditioning system, such as temperature settings, fan speed adjustments, and airflow mode switching, serving as feedback signals of user preferences.

[0024] Personalized user data, such as users' manual adjustment records, body temperature / clothing, etc., reflects users' personalized needs for air conditioning.

[0025] By acquiring multi-dimensional biometric data of occupants through multi-source sensing devices, a more comprehensive basis for modeling human thermal comfort is provided, making air conditioning regulation more scientific and rational, and compensating for the lack of multi-dimensional biometric sensing capabilities in traditional solutions. Reinforcement learning algorithms are used to learn general decisions and patterns to generate general control strategies, and personalized data is used to make personalized adjustments to general control measurements to obtain personalized control strategies that meet users' individual control needs.

[0026] Please refer to Figure 3 , Figure 3 Generate a flowchart for the general control strategy. In some embodiments, a general control strategy is generated based on occupant cabin environmental parameters, occupant biometric data, and air conditioning regulation behavior data, including: S121: Generate a state vector based on crew cabin environmental parameters, crew biometric data, and current air conditioning status parameters; S122: Generate action vectors based on air conditioning adjustment behavior data; S123: Reward function for generating power based on action vectors and air conditioner; S124: Reinforcement learning algorithm is used to perform reinforcement learning with reward function as the optimization objective, and a general control policy is generated based on state vector.

[0027] The state vector includes environmental parameters (temperature T, humidity H, light intensity L), biometric data (human activity intensity B, body temperature Tb, clothing thickness C), and current air conditioning state parameters. The current air conditioning state parameters include the current temperature setting Ts, fan speed F, and airflow mode M. The state vector can be represented as: .

[0028] Action vectors include adjustments to the air conditioning system, specifically the temperature adjustment range. Wind speed adjustment range Air outlet mode switching The action vector is represented as: .

[0029] The reward function, which takes into account both comfort and energy consumption, can be expressed as: ; in, This indicates the degree of manual temperature adjustment by the user. This indicates the degree of manual adjustment the user makes to the wind speed. This indicates the amount of manual adjustment by the user to the airflow mode; P represents the power consumption of the air conditioner. This represents a weighting coefficient used to adjust the balance between comfort and energy consumption.

[0030] Figure 2 The shared master model can be optimized through offline training and online updates. The reward function is the core optimization objective for the reinforcement learning algorithm during offline / online reinforcement learning training, determining the quality of the general control strategy and serving as the underlying basis for its generation. Furthermore, the reward function, while optimizing comfort, also considers energy consumption control, achieving a balance between comfort and energy saving, improving the energy efficiency of new energy vehicles, and extending their driving range.

[0031] Please refer to Figure 4 , Figure 4 This is a flowchart illustrating the specific process of generating a general control policy. In some embodiments, a reinforcement learning algorithm is used with the reward function as the optimization objective to generate a general control policy based on the state vector, including: S125: Obtain control parameters for similar users based on cloud-based collaborative or clustering algorithms; S126: Based on control parameters and state vectors, perform evolutionary self-learning to obtain a general control strategy.

[0032] The general control strategy is expressed as: ; in, Representation based on state vector S General control strategy, This represents the action vector corresponding to the general control strategy. Indicates the amount of temperature adjustment. Indicates the amount of wind speed adjustment. This indicates the amount of change in the airflow mode switching.

[0033] Reinforcement learning algorithms can employ a shared master model (User-Agnostic Backbone), comprising two phases: Offline Reinforcement Learning (Offline RL) and Online Reinforcement Learning (Online RL). Offline RL uses IQL (Independent Q-Learning) or CQL (Conservative Q-Learning) algorithms for pre-training to generate a general control policy. Online RL utilizes SAC (Soft Actor-Critic) or PPO (Proximal Policy Optimization) algorithms for online learning to adapt to dynamic environmental changes. Online RL does not directly participate in the fine-tuning process from a single instance of "general control policy" to "personalized adaptation," but rather serves the long-term optimization of the shared master model. It provides the underlying support for the accuracy of the general control policy and optimizes it by continuously iterating and updating the algorithm parameters of the shared master model using real-time collected full data (including policy execution feedback from all users, dynamic environmental changes, etc.), ensuring the general control policy generated by the shared master model is optimized. (Right now It better meets the general needs of all users and provides a better "benchmark strategy" for personalized adaptation of all users in the future.

[0034] A shared master model is employed for reinforcement learning to generate a general control strategy. In the initial stage, a large amount of empirical data is accumulated through calibration to build a basic model, addressing the cold start problem in reinforcement learning. When a new passenger enters, cloud-based collaborative filtering or clustering algorithms are used to find users with similar comfort levels, directly utilizing their control parameters (temperature, gear, and mode). Then, personalized data is used for evolutionary self-learning. Specifically, a user-level lightweight adapter is fine-tuned based on each user's personalized data to generate a personalized control strategy, achieving a "personalized" self-learning evolutionary effect. The shared master model, based on the current state vector S and historical data, performs full de-personalization and aggregation of data to learn group commonalities and generate a general control strategy.

[0035] As another implementation method, the shared master model can be replaced by the DQN (Deep Q-Network) or A3C (Asynchronous Advantage Actor-Critic) algorithm.

[0036] In the initial stage, a basic model is built by accumulating experience data through expert calibration, and cloud-based collaborative filtering or clustering algorithms are used to quickly match new passengers with similar user parameters. This effectively solves the problem of poor user experience in the cold start stage of reinforcement learning and achieves the intelligent control effect of "ready to use".

[0037] Please refer to Figure 5 , Figure 5 A flowchart is generated for the personalized control strategy. In some embodiments, an action vector increment is obtained based on user-personalized data, and the general control strategy is adjusted based on the action vector increment to obtain a personalized control strategy, including: S141: Quantifying user behavior based on personalized data; S142: Determine the personalized adapter based on user behavior data; S143: Utilize personalized data and personalized adapters to obtain action vector increments; S144: Based on the incremental motion vector, the general control strategy is adjusted in a personalized manner to obtain a personalized control strategy.

[0038] Personalized control strategies are represented as follows: ; ; in, Indicates passenger p Personalized action vectors, This indicates a personalized control strategy; Indicates the increment of the action vector. Indicates passenger p Temperature regulation amount, Indicates passenger p Wind speed adjustment amount, Indicates passenger p The amount of change in the air outlet mode switching.

[0039] Personalized data includes, but is not limited to, one or more combinations of passenger body temperature data, clothing thickness data, and historical preference data.

[0040] It can dynamically adjust the air conditioning control parameters according to each passenger's individual characteristics (such as body shape, age, clothing, etc.) and real-time status (such as body temperature, movement, etc.), so as to achieve personalized comfort service for each passenger and significantly improve the passenger's comfort experience.

[0041] In some embodiments, determining a personalized adapter based on the user behavior data includes: If the amount of user behavior is less than the first threshold, then the number of model parameters that need to be adjusted is 0. If the amount of user behavior is between the first threshold and the second threshold, then the model parameters are adjusted in a fine-tuning manner. If the user behavior volume is between the second and third thresholds, then the model parameters are adjusted based on the low-rank matrix combined with small-scale replay data. If the amount of user behavior exceeds the third threshold, then adjust all model parameters.

[0042] A personalized adapter, also known as a user-specific adapter, is used to determine the amount of model parameters that need to be adjusted based on the amount of user behavior, thereby obtaining a personalized control strategy using the selected personalized adapter.

[0043] Among them, such as Figure 2 As shown, the behavior threshold in the personalized adapter can be set as needed, for example: When the number of actions is less than 10: the zero-shot method is adopted, the number of parameters is 0, no update delay is required, and the general control strategy can be directly used as the personalized control strategy; When the number of actions is greater than or equal to 10 and less than 100: fine-tuning is adopted, with a parameter count of 1k-10k and an update delay in the minute range.

[0044] When the number of actions is between 100 and 1000, LoRA (Low-Rank Adaptation) combined with a small replay buffer (experience replay buffer) is used for fine-tuning, with a parameter count of about 1% and an update delay of 10 minutes. When the number of actions exceeds 1000 and the user value is high, the underlying layer is completely fine-tuned with 100% of the parameters and an update delay of hours.

[0045] LoRA technology significantly reduces memory overhead (less than 1%) by splitting the increment matrix into two smaller matrices, while improving CTR (Click-Through Rate) by 3% to 8%.

[0046] Among them, behavioral volume refers to the number or frequency of user interactions with the air conditioning system, specifically including the following behaviors: the number of times the user manually adjusts the temperature, fan speed, airflow mode, etc.; the frequency and duration of user use of the air conditioning system; and the user's adjustment behavior of the air conditioning system under different environmental conditions. The magnitude of behavioral volume reflects the user's personalized needs and preferences for the air conditioning system and is an important basis for selecting the adaptation method of the personalized adapter.

[0047] The number of model parameters refers to the number of model parameters that need to be adjusted during fine-tuning, specifically including the following situations: Zero-shot method: The number of parameters is 0, and no parameters need to be adjusted.

[0048] Fine-tuning method: The number of parameters is 1k-10k, which means that 1,000 to 10,000 model parameters need to be adjusted.

[0049] LoRA combined with small playback buffer fine-tuning: the parameter amount is about 1%, which means that 1% of the model parameters need to be adjusted.

[0050] Fully fine-tune the underlying layer: The parameter count is 100%, which means that all model parameters need to be adjusted.

[0051] The number of model parameters directly affects the update latency and resource consumption of models such as shared master models. The smaller the number of parameters, the faster the update and the less resource consumption; the larger the number of parameters, the slower the update and the more resource consumption.

[0052] Fine-tuning using a personalized adapter involves using user-specific data to add a "personalized correction amount (action vector increment)" to the general control strategy of the shared master model. For example, if a user likes warmth, the temperature will be increased. Model parameter adjustment involves selecting the "correction tool configuration" based on the amount of behavior (e.g., using zero samples for fewer behaviors and LoRA for more behaviors). The former adjusts the results, while the latter selects the tools.

[0053] A combined architecture of a shared master model and a user-level lightweight adapter, along with their data interaction and collaboration, is used to obtain personalized control strategies.

[0054] In some embodiments, obtaining action vector increments using personalized data and a personalized adapter includes: The increment of the action vector is quantified based on the data type and preference weight of the personalized data.

[0055] Action vector increment here It is determined by a user-level lightweight adapter based on user-personalized data and adaptation strategies. The personalization adapter first obtains passenger information. p The personalized data includes biometrics (such as body temperature, clothing thickness, BMI), historical adjustment behaviors (such as the magnitude and frequency of past manual temperature / fan speed adjustments), and real-time feedback (such as whether the air conditioning parameters are currently being manually corrected). This data forms the basis for calculating the action vector increment. The personalized adapter selects the corresponding adaptation algorithm (such as zero-shot, LoRA fine-tuning, etc.) based on the user's behavior data, compares and analyzes the aforementioned personalized data with the general control strategy, and calculates the adjustment magnitude that best suits the user's preferences. For example, if the user prefers a temperature 1°C lower and a fan speed 1 level higher than the general setting, the final output will be... It will contain "Δ T =-1℃、Δ F Specific increments such as "=+1 level".

[0056] Data types, such as temperature regulation and biometrics, can influence the increment of action vectors. Quantifying the increment of action vectors by data type and preference weights makes the obtained increment of action vectors more closely reflect user preferences.

[0057] In some embodiments, quantifying the action vector increment based on the data type and preference weights of the personalized data includes: The single-dimensional action vector increment is calculated based on the data type and corresponding preference weight of the personalized data, so as to obtain the action vector increment based on the single-dimensional action vector increment; If there is a dynamic adjustment, the mapping coefficients are updated, and the increment of the one-dimensional dynamic action vector is obtained based on the updated mapping coefficients. The initial value of the mapping coefficients is the corresponding preference weight.

[0058] Personalized data types and preference weights are quantified, such as direct mapping of behavioral data: a user's historical adjustment range (e.g., frequently adjusting from 24℃ to 26℃) is directly converted into a correction amount in the same direction. (℃); there is also biometric-based adaptation: thicker clothing corresponds to a temperature correction of +0.5℃, higher body temperature corresponds to a fan speed of +1 level / temperature of -0.3℃; there is also real-time feedback and dynamic adjustment: if the user still manually adjusts after correction, the mapping coefficient is updated according to the new adjustment direction (e.g., if the user still adjusts to warmer after correction, the next time...). (Add +0.2℃).

[0059] In the case of static adjustment, the formula for calculating the increment of a single-dimensional action vector can be expressed as: ; in, This indicates the core weights corresponding to air conditioning adjustment behavior data, such as historical adjustment amounts. This indicates the base weights corresponding to occupant biometric data, such as clothing thickness. This indicates the base weights for occupant biometric data, such as high body temperature. This represents the baseline coefficient.

[0060] Preference weight, or preference intensity coefficient, is a coefficient that quantifies the degree of user preference.

[0061] The baseline coefficient is an adjustable calibration factor. If deviation correction is needed later, there is no need to change the core formula; it is only necessary when adjusting the hyperparameters. That's all.

[0062] For example: if the baseline coefficient is 1, the weight of air conditioning adjustment behavior is 2, and the weight of clothing thickness is 0.5, then... ℃, the final .

[0063] For wind speed adjustment And the range of air outlet mode switching The calculation method is the same as the temperature adjustment amount mentioned above. The calculation method is the same, so it will not be repeated here.

[0064] For the baseline coefficients, during the offline phase, they are calibrated based on full user data and expert experience (e.g., if the baseline coefficient for the temperature dimension is set to 1, and the weight is 1, the corresponding single-dimensional action vector increment is 1℃; if the baseline coefficient for the wind speed dimension is set to 1, and the weight is 1, the corresponding single-dimensional action vector increment is 1 level), ensuring uniformity of dimensions.

[0065] In the case of dynamic adjustment, that is, after the air conditioner has been adjusted using the action vector increment, if the user still manually adjusts it, the mapping coefficients need to be updated first: ; in, This represents the current mapping coefficient, with the initial value being the corresponding preference weight. Indicates the updated mapping coefficients; This represents the learning rate, which, if preset to 0.2, can be used to control the extent of manual adjustment by the user. It is used to balance update speed and stability and can be adaptively fine-tuned according to the user's preference for stability. For example, it can be reduced to 0.1 after the user's preference has stabilized. This indicates the user's manual adjustment level, such as + for heating and - for cooling.

[0066] Secondly, the increment of the one-dimensional dynamic action vector is obtained based on the mapping coefficient: ; For example, initially (Clothing thickness mapping coefficient) The user manually adjusts the temperature by 1°C, that is... The updated mapping coefficients are The increment of the new dynamic action vector is then: .

[0067] The system quantifies the user's adjustment range by calculating the increment of the motion vector and dynamically adjusts the increment based on the user's manual adjustments. By collecting user comfort feedback and incorporating it into the control strategy optimization process, a complete "setting-feedback-optimization" closed loop is established, enriching the human-computer interaction dimension and making the interaction between the user and the air conditioning system more natural and intelligent.

[0068] Please refer to Figure 6 , Figure 6 A flowchart for partition control provided in this application. The method specifically includes the following steps: S210: Acquire facial image data for each occupant; S220: Determine the occupant's location coordinates based on facial image data; S230: Determine the corresponding air conditioning vent zone and the corresponding air conditioning vent based on the location coordinates; S240: Generate personalized control strategies based on occupant personalized data; S250: Adjusts the air conditioner vents based on personalized control strategies.

[0069] Please refer to Figure 7 , Figure 7 The flowchart for personalized control is as follows: Real-time acquisition of personalized occupant data generates a state vector S, and a general control strategy is generated using a shared master model. Next, user behavior data is statistically analyzed to determine the model parameters that need adjustment and the adjustment methods in the personalized adapter, generating the final personalized control strategy. This strategy enables precise zone-based control. During subsequent operation, user behavior is continuously monitored, and user behavior records are updated accordingly. Specifically, each identified passenger can be assigned a unique identification tag. Personalized data for each passenger can be personalized comfort parameters, which are core data directly affecting passenger comfort and exhibiting individual differences; their acquisition relies on multi-source sensing devices and historical data accumulation. Personalized comfort parameters include: Biometric parameters: the passenger's own physiological attributes, such as body temperature, BMI (body mass index), clothing thickness (to judge the degree of warmth), and age (different ages have different sensitivities to temperature); historical preference parameters: the passenger's past habits of adjusting the air conditioning, such as commonly used temperature range, preferred fan speed, and commonly used airflow mode (such as face airflow / foot airflow); real-time status parameters: the passenger's current dynamic status, such as activity intensity (such as prolonged sitting / getting up and moving around), and whether the current air conditioning parameters have been manually adjusted (providing feedback on real-time body sensation).

[0070] Personalized comfort parameters can be obtained through real-time sensing devices, such as cameras in the passenger compartment (to identify clothing thickness and activity intensity), infrared sensors (to measure body temperature), and BMI estimation devices (combined with height and weight data). Real-time physiological and status parameters, as well as historical data, can be collected. From the system database, based on the passenger's unique identification (such as a facial recognition tag), the system can retrieve their past air conditioning adjustment records to obtain historical preference parameters.

[0071] A state vector S is generated based on the collected data, and a general control strategy is generated using the shared master model. Next, the user behavior data is statistically analyzed to determine the model parameters that need to be adjusted in the personalized adapter and the adjustment method based on the user behavior data: if the behavior data is less than 10, a general control strategy can be directly adopted. As the final personalized control strategy; if the behavior quantity is between 10 and 100, LoRA fine-tuning is used to obtain the action vector increment. Similarly, if the number of actions is between 100 and 1000, then LoRA with a small buffer is used to obtain the action vector increment. If the number of seated movements is not less than 1000, then perform full fine-tuning to obtain the increment of the motion vector. And generate the final personalized control strategy. .

[0072] Personalized control strategies are generated based on individual comfort parameters, and the air outlet parameters of the corresponding zones are independently adjusted according to the personalized air supply control strategies to achieve precise zone control.

[0073] During subsequent operation, user behavior can be continuously monitored to update user behavior records and continuously optimize control strategies. It has the adaptive evolutionary characteristic of "optimization as it is used", which solves the problem of rigidity in traditional control strategies.

[0074] By identifying each occupant and generating personalized control strategies for each occupant, precise zoned adjustment of the air conditioning can be achieved to meet the individual needs of each occupant.

[0075] Please refer to Figure 8 , Figure 8 This is a schematic diagram of the temperature change curve after adaptive adjustment using an AI reinforcement learning algorithm. The figure shows the temperature change over 180 seconds. The upper limit temperature (28℃) and lower limit temperature (22℃) in the figure represent the adjustment range of the AI ​​reinforcement learning algorithm. If the AI ​​reinforcement learning algorithm is applied when the cabin temperature is 28℃, the cabin temperature will eventually stabilize at a personalized temperature of around 23.5℃. Using the method described in this application, the temperature can be quickly reduced to the target temperature range (25℃), and then adjusted to the personalized comfort temperature (23.5℃) and maintained stably.

[0076] Please refer to Figure 9 , Figure 9 The structural block diagram of the personalized air conditioning control device for the passenger cabin provided in this application should be understood to be related to... Figure 1 The method embodiment executed in this document corresponds to the method described above, and is capable of performing the steps involved in the aforementioned method. The specific functions of this device can be found in the description above; to avoid repetition, detailed descriptions are appropriately omitted here. This device includes, but is not limited to: Data acquisition module 110 is used to collect passenger cabin environmental parameters, passenger biometric data, and air conditioning adjustment behavior data; The general control strategy generation module 120 is used to generate a general control strategy based on occupant cabin environmental parameters, occupant biometric data, and air conditioning regulation behavior data. Personalized data acquisition module 130 is used to acquire personalized user data based on occupant cabin environmental parameters, occupant biometric data, and air conditioning adjustment behavior data. The personalized adjustment module 140 is used to adjust the general control strategy based on user personalized data to obtain a personalized control strategy. The execution module 150 is used to control the operation of the passenger cabin air conditioning based on a personalized control strategy.

[0077] In the technical solution of this application embodiment, a general control strategy is generated based on the commonalities of personalized data, and personalized adjustments are made based on the personalized data to generate a personalized control strategy to meet the personalized needs of passengers.

[0078] According to some embodiments of this application, the data acquisition module 110 is used to acquire environmental parameters such as temperature, humidity and light, occupant biometric data such as clothing thickness and body temperature, and air conditioning regulation behavior data such as temperature setting, fan speed adjustment and air outlet mode switching.

[0079] According to some embodiments of this application, the general control strategy generation module 120 is specifically used to: generate a state vector based on occupant cabin environmental parameters, occupant biometric data, and current air conditioning state parameters; generate an action vector based on air conditioning adjustment behavior data; generate a reward function based on the action vector and the power of the air conditioning; and perform reinforcement learning using a reinforcement learning algorithm with the reward function as the optimization objective to generate a general control strategy based on the state vector.

[0080] According to some embodiments of this application, the specific process of generating a general control strategy includes: obtaining control parameters of similar users based on cloud collaboration or clustering algorithms; and performing evolutionary self-learning based on the control parameters and state vectors to obtain a general control strategy.

[0081] According to some embodiments of this application, the personalized adjustment module 140 is specifically used for: quantifying user behavior volume based on personalized data; determining a personalized adapter based on user behavior volume; obtaining action vector increments using personalized data and personalized adapters; and performing personalized adjustments on the general control strategy based on the action vector increments to obtain a personalized control strategy.

[0082] According to some embodiments of this application, the specific process of determining the personalized adapter based on user behavior volume includes: if the user behavior volume is less than a first threshold, the number of model parameters to be adjusted is 0; if the user behavior volume is between the first threshold and the second threshold, the number of model parameters is adjusted in a fine-tuning manner; if the user behavior volume is between the second threshold and the third threshold, the number of model parameters is adjusted based on a low-rank matrix combined with small-scale replay data; if the user behavior volume is greater than the third threshold, all model parameters are adjusted.

[0083] According to some embodiments of this application, the process of obtaining the action vector increment includes: quantifying the action vector increment based on the data type and preference weight of the personalized data. Specifically: calculating the single-dimensional action vector increment based on the data type and corresponding preference weight of the personalized data, and obtaining the action vector increment based on the single-dimensional action vector increment; if there is dynamic adjustment, updating the mapping coefficient, and obtaining the single-dimensional dynamic action vector increment based on the updated mapping coefficient, wherein the initial value of the mapping coefficient is the corresponding preference weight.

[0084] This application provides an electronic device including a memory and a processor. The memory stores a computer program, and the processor runs the computer program to cause the electronic device to perform the methods in any of the aforementioned optional implementations.

[0085] This application provides a readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the methods in any of the aforementioned optional implementations.

[0086] The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0087] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0088] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0089] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0090] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0091] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0092] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for personalized control of passenger cabin air conditioning, characterized in that, The method includes: Collect passenger cabin environmental parameters, passenger biometric data, and air conditioning behavior data; A general control strategy is generated based on the aforementioned cabin environmental parameters, occupant biometric data, and air conditioning regulation behavior data. Personalized user data is obtained based on the passenger cabin environmental parameters, passenger biometric data, and air conditioning adjustment behavior data. Based on the user's personalized data, an action vector increment is obtained, and the general control strategy is adjusted in a personalized manner based on the action vector increment to obtain a personalized control strategy. The personalized control strategy is used to control the operation of the passenger cabin air conditioning.

2. The personalized control method for passenger cabin air conditioning according to claim 1, characterized in that, The general control strategy generated based on the passenger cabin environmental parameters, passenger biometric data, and air conditioning regulation behavior data includes: A state vector is generated based on the passenger cabin environmental parameters, passenger biometric data, and current air conditioning status parameters. An action vector is generated based on the air conditioning adjustment behavior data; A reward function is generated based on the action vector and the power of the air conditioner; The general control policy is generated based on the state vector by using a reinforcement learning algorithm with the reward function as the optimization objective.

3. The personalized control method for passenger cabin air conditioning according to claim 2, characterized in that, The step of using a reinforcement learning algorithm with the reward function as the optimization objective to perform reinforcement learning, and generating the general control policy based on the state vector, includes: Control parameters of similar users are obtained based on cloud-based collaborative or clustering algorithms; Based on the control parameters and the state vector, evolutionary self-learning is performed to obtain the general control strategy.

4. The personalized control method for passenger cabin air conditioning according to claim 1, characterized in that, The step of obtaining action vector increments based on the user's personalized data, and then adjusting the general control strategy based on the action vector increments to obtain a personalized control strategy, includes: Quantify user behavior based on the personalized data; A personalized adapter is determined based on the user behavior data. The action vector increment is obtained using the personalized data and the personalized adapter; Based on the increment of the action vector, the general control strategy is adjusted in a personalized manner to obtain a personalized control strategy.

5. The personalized control method for passenger cabin air conditioning according to claim 4, characterized in that, The process of determining a personalized adapter based on the user behavior data includes: If the amount of user behavior is less than the first threshold, then the amount of model parameters that need to be adjusted is 0. If the amount of user behavior is between the first threshold and the second threshold, then the amount of model parameters is adjusted in a fine-tuning manner; If the amount of user behavior is between the second threshold and the third threshold, then the model parameters are adjusted based on the low-rank matrix combined with small-scale replay data. If the amount of user behavior exceeds the third threshold, then all model parameters are adjusted.

6. The personalized control method for passenger cabin air conditioning according to claim 4, characterized in that, The step of obtaining the action vector increment using the personalized data and the personalized adapter includes: The increment of the action vector is quantified based on the data type and preference weight of the personalized data.

7. The personalized control method for passenger cabin air conditioning according to claim 6, characterized in that, The step of quantifying the action vector increment based on the data type and preference weight of the personalized data includes: The single-dimensional action vector increment is calculated based on the data type and corresponding preference weight of the personalized data, so as to obtain the action vector increment based on the single-dimensional action vector increment; If there is a dynamic adjustment, the mapping coefficients are updated, and the increment of the one-dimensional dynamic action vector is obtained based on the updated mapping coefficients. The initial value of the mapping coefficients is the corresponding preference weight.

8. A personalized control device for passenger cabin air conditioning, characterized in that, The device includes: The data acquisition module is used to collect environmental parameters of the occupant cabin, occupant biometric data, and air conditioning adjustment behavior data. A general control strategy generation module is used to generate a general control strategy based on the occupant cabin environmental parameters, occupant biometric data, and air conditioning adjustment behavior data. The personalized data acquisition module is used to acquire personalized user data based on the occupant cabin environmental parameters, occupant biometric data, and air conditioning adjustment behavior data. The personalized adjustment module is used to obtain the action vector increment based on the user's personalized data, and to make personalized adjustments to the general control strategy based on the action vector increment to obtain a personalized control strategy. The execution module is used to control the operation of the passenger cabin air conditioning based on the personalized control strategy.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the personalized control method for passenger cabin air conditioning as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores computer program instructions, which, when read and executed by a processor, perform the personalized control method for passenger cabin air conditioning as described in any one of claims 1 to 7.

Images