A Thermal Comfort Evaluation Method and System Based on AI Agent

Abstract

The application provides an AI Agent-based thermal comfort evaluation method and system, which fuses driver physiological data and in-vehicle environment data, carries out multi-modal analysis, and realizes accurate evaluation and individualized adjustment of the driver's thermal comfort state. The above multi-source data can be processed in real time, and a customized comfort adjustment scheme is generated according to the physiological feedback of the driver and the change of the environmental parameters. In terms of data processing, the application completes the data processing and comfort adjustment operation locally on the vehicle terminal, avoids uploading data to the cloud, and fundamentally guarantees the privacy and security of user data. At the same time, the application has dynamic learning ability, can continuously optimize the adjustment strategy based on the feedback data of the driver, and deeply excavates the individualized preferences of the user.

Description

[0001] This application is a divisional application of the following original application: Application date of the original application: April 30, 2025; Application number of the original application: 202510566060.8; Title of the original application: Thermal comfort evaluation method and system based on AI Agent. Technical Field

[0002] This invention relates to the field of intelligent management technology, specifically to a thermal comfort evaluation method and system based on AI Agent. Background Technology

[0003] With the in-depth development of intelligent cockpit technology and environmental comfort research, thermal comfort evaluation systems are being used more and more widely in the automotive, construction and other fields. Although existing thermal comfort evaluation technologies have achieved certain results, they still have many limitations and cannot meet users' needs for a comfortable, intelligent and personalized experience.

[0004] Currently, most thermal comfort assessment technologies rely primarily on single-dimensional data collection and analysis. For example, some technologies only collect environmental data (such as temperature, humidity, and wind speed) and perform comfort assessments using pre-set standard models. However, this approach ignores the influence of individual user physiological characteristics (such as heart rate and body temperature) on thermal comfort perception, failing to accurately reflect the user's actual thermal comfort state. Different users, even in the same environment, experience vastly different levels of thermal comfort due to differences in their physical constitutions; therefore, the analysis of single-environmental data is insufficient to provide an accurate assessment.

[0005] In terms of personalized adjustment, existing systems lack in-depth understanding and dynamic response to users' individual preferences. Most systems adjust the environment based solely on general comfort standards, failing to make flexible adjustments based on users' historical feedback and real-time physiological data. For example, when adjusting air conditioning temperature, they cannot accurately adapt to different users' temperature preferences, nor can they optimize settings in real time based on changes in users' physical sensations, resulting in users not obtaining a truly comfortable experience that meets their individual needs.

[0006] Insufficient real-time performance and adaptability are also major shortcomings of existing technologies. Faced with complex and ever-changing driving or indoor environments, traditional thermal comfort assessment systems respond slowly and cannot promptly capture changes in user comfort needs. When environmental factors change rapidly, the system cannot quickly adjust environmental parameters, causing users to remain in an uncomfortable state for extended periods, impacting user experience and safety.

[0007] Furthermore, with increasing awareness of user privacy, existing thermal comfort assessment technologies pose privacy risks in data processing. Many systems upload user data to the cloud for analysis, increasing the possibility of data leaks and failing to meet users' requirements for data privacy protection.

[0008] A patent search revealed invention patent CN119761169A, which discloses a method and system for optimizing indoor thermal comfort in buildings based on machine learning. This patent establishes a BIM 3D information model of the building using laser scanning and IoT sensors, and then utilizes machine learning technology to optimize indoor thermal comfort. By collecting, analyzing, and predicting environmental data in real time, the system can automatically adjust temperature control equipment to ensure indoor comfort. However, this patent relies on high-precision sensors and BIM models, requiring extremely accurate data collection. It lacks sufficient tolerance for missing data or sensor malfunctions, and it lacks in-depth analysis of personalized user needs, making it unable to flexibly adjust to adapt to different users and environments. Furthermore, it has limitations in real-time performance and high computational complexity.

[0009] In summary, given the problems of the existing technologies, researching a thermal comfort evaluation method and system based on AI Agents has become a critical task that urgently needs to be addressed. Summary of the Invention

[0010] To address the shortcomings of existing technologies, the purpose of this invention is to provide a thermal comfort evaluation method and system based on AI Agent.

[0011] A thermal comfort evaluation method based on an AI agent, provided by the present invention, includes the following steps: Step S1: Acquire multimodal data including environmental and physiological data; Step S2: Construct a thermal comfort evaluation model by using multimodal data to form a dataset; Step S3: Input the real-time acquired multimodal data into the thermal comfort evaluation model, output thermal comfort score and environmental adjustment suggestions, and dynamically adjust the in-vehicle environment or model parameters based on the feedback data from car owners. Step S4: Encrypt and store multimodal data and feedback data locally on the vehicle terminal; Step S5: The thermal comfort evaluation model interacts collaboratively with the in-vehicle environment control system.

[0012] Preferably, step S1 includes the following sub-steps: Step S1.1: Collect environmental data inside the vehicle using the vehicle temperature sensor, vehicle humidity sensor, and vehicle air velocity sensor. The environmental data includes temperature data, humidity data, and air velocity data. Step S1.2: Collect the driver's physiological data through a heart rate monitor and a skin temperature sensor. The physiological data includes heart rate data, body temperature data, and skin conductance response data. Step S1.3 involves preprocessing the environmental and physiological data to obtain preprocessed environmental and physiological data, i.e., multimodal data.

[0013] Preferably, in step S1.3, the preprocessing includes data denoising, normalization, and outlier removal.

[0014] Preferably, step S2 includes the following sub-steps: Step S2.1: Receive the car owner's evaluation of the current thermal comfort through the in-vehicle interactive interface, generate thermal comfort evaluation labels, and associate and label the multimodal data with the thermal comfort evaluation labels to form a labeled dataset for subsequent model training. Step S2.2: Based on the dataset, construct a deep learning-based thermal comfort evaluation model and output thermal comfort scores and environmental adjustment suggestions.

[0015] Preferably, step S2.2 includes the following sub-steps: Step S2.2.1: Use a convolutional neural network to extract spatial features of environmental data, use a long short-term memory network to process the temporal dependence of physiological data, and obtain joint features. The training data is the dataset constructed in step S2.1. Step S2.2.2: Input the joint features into the fully connected regression layer, and after linear transformation and activation function processing, output the theoretical optimal value under the current environment; Step S2.2.3: Based on the theoretical optimal value, integrate a multi-objective optimization algorithm to generate adjustment schemes for air conditioning temperature parameters, seat heating parameters, and air conditioning fan speed parameters; In step S2.2.4, when the car owner corrects the comfort score through the interactive interface, the model parameters are updated through the incremental learning algorithm so that the subsequent output thermal comfort score is close to the car owner's corrected value.

[0016] Preferably, step S3 includes the following sub-steps: Step S3.1: Acquire multimodal data in real time and input it into the trained thermal comfort evaluation model. Display the output thermal comfort score and environmental adjustment suggestions through the in-vehicle interface. Step S3.2: If the car owner provides feedback data within a preset time, then control the in-vehicle environmental parameters or optimize the personalized adaptation capability of the thermal comfort evaluation model. Step S3.4: If the vehicle owner does not provide feedback data within the preset time, step S3.1 will be executed again at the preset frequency.

[0017] Preferably, step S3.2 includes the following sub-steps: Step S3.2.1: If the feedback data is an adjustment command, send the adjustment command to the air conditioning controller, seat controller and air flow controller via the CAN bus; Step S3.2.1: When the feedback data is a corrected value of the comfort score, update the parameters of the thermal comfort evaluation model to adapt to individual needs.

[0018] Preferably, in step S4: an independent data partition is divided in the encrypted storage area of ​​the vehicle MCU chip, and the multimodal data and feedback data are encrypted and stored using the AES-256 algorithm, with the encryption key bound to the vehicle VIN code.

[0019] Preferably, in step S5, the thermal comfort evaluation model is used to interact with the intelligent navigation system and the in-vehicle intelligent seat via an in-vehicle Ethernet network.

[0020] This invention also provides a thermal comfort evaluation system based on an AI agent, comprising: Module M1 acquires multimodal data, including environmental and physiological data. Module M2 uses multimodal data to construct a dataset and build a thermal comfort evaluation model. Module M3 inputs real-time acquired multimodal data into the thermal comfort evaluation model, outputs a comfort score and environmental adjustment suggestions, and dynamically adjusts the in-vehicle environment or model parameters based on feedback data from car owners. Module M4 encrypts and stores multimodal data and feedback data locally on the vehicle terminal; Module M5 enables the thermal comfort evaluation model to interact collaboratively with the in-vehicle environmental control system.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention innovatively integrates the driver's physiological data (such as heart rate and body temperature) with in-vehicle environmental data. Leveraging the powerful multi-dimensional analysis capabilities of the AI ​​Agent, it achieves accurate assessment and real-time adjustment of the driver's thermal comfort status. This invention not only efficiently processes environmental data such as in-vehicle temperature, humidity, and wind speed, but also closely integrates the driver's physiological responses and incorporates personalized preference factors to dynamically adjust in-vehicle environmental parameters. Through this comprehensive analysis and adjustment mechanism, it changes the traditional single-dimensional evaluation model, providing drivers with a more accurate and personalized assessment of their thermal comfort.

[0022] 2. The personalized thermal comfort adjustment mechanism introduced in this invention relies heavily on driver feedback data and real-time physiological data to achieve dynamic, real-time adjustment of the in-vehicle environment. Specifically, this invention can finely adjust settings such as air conditioning temperature, fan speed, and seat heating based on the driver's long-term temperature preferences and current sensory data. This ensures the driver is always in the most comfortable environment, effectively solving the problem that traditional systems struggle to meet users' personalized needs.

[0023] 3. This invention possesses powerful real-time monitoring and dynamic response capabilities. By continuously monitoring the physiological changes of the car owner (such as heart rate fluctuations, body temperature rises and falls) and real-time changes in in-vehicle environmental data, it can keenly capture subtle changes in the car owner's thermal comfort needs. The system can promptly optimize the in-vehicle environment, allowing the car owner to enjoy the best thermal comfort experience under any circumstances. Attached Figure Description

[0024] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart of an AI Agent-based thermal comfort evaluation method in an embodiment of the present invention. Detailed Implementation

[0025] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0026] This patent provides an AI Agent-based method and system for evaluating thermal comfort. By integrating the driver's physiological data (such as heart rate and body temperature) with in-vehicle environmental data (such as temperature, humidity, and airflow), it conducts multimodal analysis to achieve accurate assessment and personalized adjustment of the driver's thermal comfort state. It can process the aforementioned multi-source data in real time and generate customized comfort adjustment plans based on the driver's physiological feedback and changes in environmental parameters. In terms of data processing, this invention performs data processing and comfort adjustment operations locally on the in-vehicle terminal, avoiding data uploads to the cloud and fundamentally protecting user data privacy and security. Simultaneously, this invention possesses dynamic learning capabilities, continuously optimizing adjustment strategies based on driver feedback data and deeply mining personalized user preferences.

[0027] Example 1: Figure 1 This is a flowchart of an AI Agent-based thermal comfort evaluation method in an embodiment of the present invention.

[0028] like Figure 1 As shown, this embodiment provides a thermal comfort evaluation method based on an AI agent, including the following steps: Step S1: Acquire multimodal data, including environmental and physiological data.

[0029] Specifically, step S1 includes the following sub-steps: Step S1.1: Collect environmental data inside the vehicle using the vehicle temperature sensor, vehicle humidity sensor, and vehicle air velocity sensor. The environmental data includes temperature data, humidity data, and air velocity data. Step S1.2: Collect the driver's physiological data through a heart rate monitor and a skin temperature sensor. The physiological data includes heart rate data, body temperature data, and skin conductance response data. Step S1.3 involves preprocessing the environmental and physiological data to obtain preprocessed environmental and physiological data, i.e., multimodal data.

[0030] Furthermore, in step S1.3, preprocessing includes data denoising, normalization, and outlier removal, transforming all data to the same standardized range to ensure data consistency and accuracy.

[0031] Step S2: Construct a thermal comfort evaluation model by using multimodal data to form a dataset.

[0032] Specifically, step S2 includes the following sub-steps: Step S2.1: Receive the car owner's evaluation of the current thermal comfort through the in-vehicle interactive interface, generate thermal comfort evaluation labels, and associate and label the multimodal data with the thermal comfort evaluation labels to form a labeled dataset for subsequent model training. Step S2.2: Based on the dataset, construct a deep learning-based thermal comfort evaluation model and output thermal comfort scores and environmental adjustment suggestions.

[0033] Specifically, deep learning models process environmental and physiological data, identify patterns and features in the data, assess thermal comfort in real time, and provide suggestions for environmental regulation.

[0034] Further, step S2.2 includes the following sub-steps: Step S2.2.1: A convolutional neural network (CNN) is used to extract spatial features of the environmental data, and a long short-term memory network (LSTM) is used to process the temporal dependencies of the physiological data to obtain joint features. The training data is the dataset constructed in step S2.1.

[0035] Step S2.2.2, combine the features The input is fed into a fully connected regression layer and undergoes a linear transformation. , Processed with the ReLU activation function, it outputs the theoretically optimal value under the current environment. , , The formula is as follows: , in, , and , These are the weight matrix and bias of the regression layer, respectively; The activation function is ReLU; this process involves one forward propagation, requiring no backpropagation. The model output is compared with the labels provided by the car owner. For comparison, the joint loss function is:

[0036] Part One The comfort score regression error measures the comfort score predicted by the model. With real labels Mean square error; Part Two For the regression error of each environmental regulation parameter, calculate its squared error relative to the corresponding label, where, and , which is a weighting coefficient used to balance the importance ratio between the comfort score regression and the environmental parameter regression.

[0037] Step S2.2.3: Based on the theoretical optimal value, integrate a multi-objective optimization algorithm to generate adjustment schemes for air conditioning temperature parameters, seat heating parameters, and air conditioning fan speed parameters.

[0038] Multi-objective optimization algorithms achieve parameter tuning by establishing an objective function. Represented as:

[0039] in, , , These are the current settings for air conditioning temperature, seat heating power, and air conditioning fan speed, respectively. , , This represents the theoretical optimal value output by the model. , , The weighting coefficients of the target are used to balance the adjustment priorities of different parameters.

[0040] To minimize the objective function To achieve the goal, we apply gradient descent to iterate over the current parameters. , , until convergence to the optimal solution. , , These parameters serve as recommended values ​​for air conditioning temperature, seat heating power, and air conditioning fan speed, respectively, and are used as suggestions for environmental adjustment.

[0041] In step S2.2.4, when the car owner corrects the comfort score through the interactive interface, the model parameters are updated through the incremental learning algorithm so that the subsequent output thermal comfort score is close to the car owner's corrected value.

[0042] Step S3: Input the real-time acquired multimodal data into the thermal comfort evaluation model, output thermal comfort score and environmental adjustment suggestions, and dynamically adjust the in-vehicle environment or model parameters based on the feedback data from car owners.

[0043] Specifically, step S3 includes the following sub-steps: Step S3.1: Acquire multimodal data in real time and input it into the trained thermal comfort evaluation model. Display the output thermal comfort score and environmental adjustment suggestions through the in-vehicle interface. Step S3.2: If the car owner provides feedback data (including settings for environmental parameters such as temperature, humidity, and wind speed) within a preset time, then control the in-vehicle environmental parameters or optimize the personalized adaptation capability of the thermal comfort evaluation model. Further, step S3.2 includes the following sub-steps: Step S3.2.1: If the feedback data is an adjustment command, send the adjustment command to the air conditioning controller, seat controller and air flow controller via the CAN bus.

[0044] In this embodiment, the in-vehicle environment is adjusted in real time according to the adjustment commands given by the car owner through the in-vehicle touch screen or voice assistant.

[0045] Step S3.2.1: When the feedback data is a corrected value of the comfort score, update the parameters of the thermal comfort evaluation model to adapt to individual needs.

[0046] Step S3.4: If the vehicle owner does not provide feedback data within the preset time, step S3.1 will be executed again at the preset frequency.

[0047] Step S4: Encrypt and store multimodal data and feedback data locally on the vehicle terminal.

[0048] Specifically, in step S4: an independent data partition is divided in the encrypted storage area of ​​the vehicle MCU chip, and the multimodal data and feedback data are encrypted and stored using the AES-256 algorithm. The encryption key is bound to the vehicle VIN code.

[0049] In this embodiment, all collected environmental and physiological data are processed locally within the in-vehicle system to ensure user data privacy. The owner's physiological data, comfort feedback data, etc., are not uploaded to the cloud or external servers; they are only stored and used on the in-vehicle terminal. All data stored on the in-vehicle terminal is encrypted, especially sensitive physiological data. Data can only be accessed and used with the owner's authorization. The owner can view, manage, and delete personal data at any time, ensuring data controllability and transparency.

[0050] Step S5: The thermal comfort evaluation model interacts collaboratively with the in-vehicle environment control system.

[0051] Specifically, in step S5, the thermal comfort evaluation model is used to interact with the intelligent navigation system and the in-vehicle intelligent seat via the vehicle Ethernet.

[0052] In this embodiment, the thermal comfort evaluation model works in conjunction with other systems such as in-vehicle smart seats and smart navigation. For example, when the in-vehicle smart seat reports a temperature adjustment signal, it triggers the thermal comfort evaluation model to reassess the thermal comfort score and environmental adjustment recommendations.

[0053] Example 2: The present invention also provides an AI Agent-based thermal comfort evaluation system. The AI ​​Agent-based thermal comfort evaluation system can be implemented by executing the process steps of the AI ​​Agent-based thermal comfort evaluation method. That is, those skilled in the art can understand the AI ​​Agent-based thermal comfort evaluation method as a preferred embodiment of the AI ​​Agent-based thermal comfort evaluation system.

[0054] Specifically, the AI ​​Agent-based thermal comfort evaluation system includes: Module M1 acquires multimodal data, including environmental and physiological data. Module M2 uses multimodal data to construct a dataset, builds and trains a thermal comfort evaluation model, and obtains the trained thermal comfort evaluation model. Module M3 inputs real-time acquired multimodal data into the trained thermal comfort evaluation model, outputs comfort scores and environmental adjustment suggestions, and dynamically adjusts the in-vehicle environment or model parameters based on feedback data from car owners. Module M4 encrypts and stores multimodal data and feedback data locally on the vehicle terminal; Module M5 enables the thermal comfort evaluation model to interact collaboratively with the in-vehicle environmental control system.

[0055] Specifically, module M1 includes the following sub-steps: Module M1.1 collects environmental data inside the vehicle through an onboard temperature sensor, an onboard humidity sensor, and an onboard airflow sensor. The environmental data includes temperature data, humidity data, and airflow data. Module M1.2 collects the driver's physiological data through a heart rate monitor and a skin temperature sensor. The physiological data includes heart rate data, body temperature data, and skin conductance data. Module M1.3 preprocesses environmental and physiological data to obtain preprocessed environmental and physiological data, i.e., multimodal data.

[0056] Furthermore, in module M1.3, preprocessing includes data denoising, normalization, and outlier removal, uniformly transforming all data to the same standardized range to ensure data consistency and accuracy.

[0057] Specifically, module M2 includes the following sub-steps: Module M2.1 receives the car owner's evaluation of the current thermal comfort through the in-vehicle interactive interface, generates thermal comfort evaluation labels, and associates and annotates the multimodal data with the thermal comfort evaluation labels to form a labeled dataset for subsequent model training. Module M2.2, based on the dataset, constructs a deep learning-based thermal comfort evaluation model, outputting thermal comfort scores and environmental adjustment suggestions.

[0058] Specifically, deep learning models process environmental and physiological data, identify patterns and features in the data, assess thermal comfort in real time, and provide suggestions for environmental regulation.

[0059] Furthermore, module M2.2 includes the following sub-steps: Module M2.2.1 uses a convolutional neural network (CNN) to extract spatial features from environmental data and a long short-term memory network (LSTM) to process the temporal dependencies of physiological data, thus obtaining joint features. The training data is a dataset consisting of module M2.1.

[0060] Module M2.2.2 will combine features The input is fed into a fully connected regression layer and undergoes a linear transformation. , Processed with the ReLU activation function, it outputs the theoretically optimal value under the current environment. , , The formula is as follows: , in, , and , These are the weight matrix and bias of the regression layer, respectively; The activation function is ReLU; this process involves one forward propagation, requiring no backpropagation. The model output is compared with the labels provided by the car owner. For comparison, the joint loss function is:

[0061] Part One The comfort score regression error measures the comfort score predicted by the model. With real labels Mean square error; Part Two For the regression error of each environmental regulation parameter, calculate its squared error relative to the corresponding label, where, and , which is a weighting coefficient used to balance the importance ratio between the comfort score regression and the environmental parameter regression.

[0062] Module M2.2.3, based on theoretical optimal values, integrates a multi-objective optimization algorithm to generate adjustment schemes for air conditioning temperature parameters, seat heating parameters, and air conditioning fan speed parameters.

[0063] Multi-objective optimization algorithms achieve parameter tuning by establishing an objective function. Represented as:

[0064] in, , , These are the current settings for air conditioning temperature, seat heating power, and air conditioning fan speed, respectively. , , This represents the theoretical optimal value output by the model. , , The weighting coefficients of the target are used to balance the adjustment priorities of different parameters.

[0065] To minimize the objective function To achieve the goal, we apply gradient descent to iterate over the current parameters. , , until convergence to the optimal solution. , , These parameters serve as recommended values ​​for air conditioning temperature, seat heating power, and air conditioning fan speed, respectively, and are used as suggestions for environmental adjustment.

[0066] Module M2.2.4 updates the model parameters through an incremental learning algorithm when the car owner corrects the comfort rating through the interactive interface, so that the subsequent output thermal comfort rating is close to the car owner's corrected value.

[0067] Specifically, module M3 includes the following sub-steps: Module M3.1 acquires multimodal data in real time and inputs it into the trained thermal comfort evaluation model, displaying the output thermal comfort score and environmental adjustment suggestions through the in-vehicle interface; Module M3.2, if the car owner provides feedback data (including settings for environmental parameters such as temperature, humidity, and wind speed) within a preset time, then controls the in-vehicle environmental parameters or optimizes the personalized adaptation capability of the thermal comfort evaluation model. Furthermore, module M3.2 includes the following sub-steps: Module M3.2.1, if the feedback data is an adjustment command, sends the adjustment command to the air conditioning controller, seat controller and air flow controller via the CAN bus.

[0068] In this embodiment, the in-vehicle environment is adjusted in real time according to the adjustment commands given by the car owner through the in-vehicle touch screen or voice assistant.

[0069] Module M3.2.1 updates the thermal comfort evaluation model parameters when the feedback data is a corrected value of the comfort score, in order to adapt to individual needs.

[0070] If the vehicle owner does not provide feedback data within a preset time, module M3.4 will re-execute module M3.1 at a preset frequency.

[0071] Specifically, in module M4: an independent data partition is divided in the encrypted storage area of ​​the vehicle MCU chip, and multimodal data and feedback data are encrypted and stored using the AES-256 algorithm. The encryption key is bound to the vehicle VIN code.

[0072] In this embodiment, all collected environmental and physiological data are processed locally within the in-vehicle system to ensure user data privacy. The owner's physiological data, comfort feedback data, etc., are not uploaded to the cloud or external servers; they are only stored and used on the in-vehicle terminal. All data stored on the in-vehicle terminal is encrypted, especially sensitive physiological data. Data can only be accessed and used with the owner's authorization. The owner can view, manage, and delete personal data at any time, ensuring data controllability and transparency.

[0073] Specifically, in module M5, data interaction between the thermal comfort evaluation model and intelligent navigation and in-vehicle intelligent seats is achieved through in-vehicle Ethernet.

[0074] In this embodiment, the thermal comfort evaluation model works in conjunction with other systems such as in-vehicle smart seats and smart navigation. For example, when the in-vehicle smart seat reports a temperature adjustment signal, it triggers the thermal comfort evaluation model to reassess the thermal comfort score and environmental adjustment recommendations.

[0075] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0076] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A thermal comfort evaluation method based on AI Agent, characterized in that, Includes the following steps: Step S1: Acquire multimodal data including environmental and physiological data; Step S2: Construct a thermal comfort evaluation model by using the multimodal data to form a dataset; Step S3: Input the real-time acquired multimodal data into the thermal comfort evaluation model, output thermal comfort score and environmental adjustment suggestions, and dynamically adjust the in-vehicle environment or model parameters based on the feedback data from the car owner. Step S4: Encrypt and store the multimodal data and the feedback data locally on the vehicle terminal; Step S5: The thermal comfort evaluation model interacts collaboratively with the in-vehicle environment control system. Step S2 includes the following sub-steps: Step S2.1: Receive the car owner's evaluation of the current thermal comfort through the in-vehicle interactive interface, generate thermal comfort evaluation labels, and associate and label the multimodal data with the thermal comfort evaluation labels to form a labeled dataset for subsequent model training. Step S2.2: Based on the dataset, construct a deep learning-based thermal comfort evaluation model and output a thermal comfort score and environmental adjustment suggestions; Step S2.2 includes the following sub-steps: Step S2.2.1: Use a convolutional neural network to extract spatial features of environmental data, use a long short-term memory network to process the temporal dependence of physiological data, and obtain joint features. The training data is the dataset constructed in step S2.

1. Step S2.2.2: Input the joint features into the fully connected regression layer, and after linear transformation and activation function processing, output the theoretical optimal value under the current environment; Step S2.2.3: Based on the theoretical optimal value, integrate a multi-objective optimization algorithm to generate adjustment schemes for air conditioning temperature parameters, seat heating parameters, and air conditioning fan speed parameters; In step S2.2.4, when the car owner corrects the comfort score through the interactive interface, the model parameters are updated through the incremental learning algorithm so that the subsequent output thermal comfort score is close to the car owner's corrected value.

2. The thermal comfort evaluation method based on AI Agent according to claim 1, characterized in that, Step S1 includes the following sub-steps: Step S1.1: Collect environmental data inside the vehicle using an onboard temperature sensor, an onboard humidity sensor, and an onboard airflow sensor. The environmental data includes temperature data, humidity data, and airflow data. Step S1.2: Collect the driver's physiological data through a heart rate monitor and a skin temperature sensor. The physiological data includes heart rate data, body temperature data, and skin conductance data. Step S1.3: Preprocess the environmental data and the physiological data to obtain preprocessed environmental data and preprocessed physiological data, i.e., multimodal data.

3. The thermal comfort evaluation method based on AI Agent according to claim 2, characterized in that, In step S1.3, the preprocessing includes data denoising, normalization, and outlier removal.

4. The thermal comfort evaluation method based on AI Agent according to claim 1, characterized in that, In step S4: an independent data partition is divided in the encrypted storage area of ​​the vehicle MCU chip, and the multimodal data and the feedback data are encrypted and stored using the AES-256 algorithm. The encryption key is bound to the vehicle VIN code.

5. The thermal comfort evaluation method based on AI Agent according to claim 1, characterized in that, In step S5, the thermal comfort evaluation model is used to interact with the intelligent navigation system and the in-vehicle intelligent seat via the vehicle Ethernet.

6. A thermal comfort evaluation system based on AI Agent, characterized in that, include: Module M1 acquires multimodal data, including environmental and physiological data. Module M2 uses multimodal data to construct a dataset and build a thermal comfort evaluation model. Module M3 inputs real-time acquired multimodal data into the thermal comfort evaluation model, outputs thermal comfort scores and environmental adjustment suggestions, and dynamically adjusts the in-vehicle environment or model parameters based on feedback data from car owners. Module M4 encrypts and stores the multimodal data and the feedback data locally on the vehicle terminal; Module M5, in which the thermal comfort evaluation model interacts collaboratively with the in-vehicle environment control system; The module M2 includes the following sub-steps: Module M2.1 receives the car owner's evaluation of the current thermal comfort through the in-vehicle interactive interface, generates thermal comfort evaluation labels, and associates and annotates the multimodal data with the thermal comfort evaluation labels to form a labeled dataset for subsequent model training. Module M2.2, based on the dataset, constructs a deep learning-based thermal comfort evaluation model, and outputs a thermal comfort score and environmental adjustment suggestions; The module M2.2 includes the following sub-steps: Module M2.2.1 uses a convolutional neural network to extract spatial features of environmental data and a long short-term memory network to process the temporal dependencies of physiological data to obtain joint features. The training data is the dataset composed of module M2.

1. Module M2.2.2 inputs the joint features into a fully connected regression layer, processes them through linear transformation and activation function, and outputs the theoretical optimal value under the current environment; Module M2.2.3, based on the theoretical optimal value, integrates a multi-objective optimization algorithm to generate adjustment schemes for air conditioning temperature parameters, seat heating parameters, and air conditioning fan speed parameters; Module M2.2.4 updates the model parameters through an incremental learning algorithm when the car owner corrects the comfort rating through the interactive interface, so that the subsequent output thermal comfort rating is close to the car owner's corrected value.

Images