Intelligent cabin environment adaptive adjustment technology based on AI large model

Abstract

The invention relates to the technical field of automobile cabin environment control, and discloses an intelligent cabin environment adaptive adjustment system based on an AI large model, which comprises a sensing module, an AI large model decision module, an execution module and a feedback module. According to the intelligent cabin environment self-adaptive adjustment technology based on the AI large model, a mixed decision sub-model of an AI large model decision-making module adopts an MPC-TD3 fusion algorithm, and a user behavior learning sub-model of the AI large model decision-making module learns user physiological parameters and operation records collected by a sensing module; the multi-physics field prediction sub-model predicts a cabin environment and a power battery state, and corrects a decision result in combination with comfort parameters fed back by a PMV-PPD evaluation model and an SET evaluation model of the feedback module in real time, so that the problems that a single MPC model is insufficient in robustness under complex working conditions and a single reinforcement learning model is difficult to quantify the comfort feeling of a user are solved, and the user experience is improved. And the dynamic response speed and the adjustment precision are considered.

Description

Technical Field

[0001] This invention relates to the field of automotive cabin environment control technology, specifically to intelligent cabin environment adaptive adjustment technology based on AI large model. Background Technology

[0002] Electric vehicles have become the core carrier of the green energy industry, and the cabin environment regulation system, as a key component that directly affects the user's driving experience and the energy economy of the whole vehicle, has become an inevitable trend in the industry to upgrade to intelligent and precise systems. In particular, with the increasing demand for personalized comfort from users and the higher requirements for the thermal safety of power batteries, traditional cabin environment regulation technology is facing the dual challenges of intelligent adaptation and multi-objective collaborative control.

[0003] Existing cabin environment regulation technologies suffer from two core defects: First, the application of AI control algorithms is limited, often employing single models such as MPC or reinforcement learning. MPC models rely on precise system modeling, resulting in insufficient robustness under complex operating conditions. Reinforcement learning models, lacking deep coupling with human thermal regulation models, struggle to accurately quantify user comfort. Neither can balance dynamic response speed and regulation accuracy. Second, cabin temperature control and power battery temperature control are separate, failing to consider their energy consumption coupling relationship. In high-temperature environments, a vicious cycle can easily occur where "high-load air conditioning exacerbates battery discharge heat generation, and strong battery cooling consumes additional energy," failing to balance cabin comfort, battery safety, and range economy. These two major problems severely restrict the development of intelligent cabin environment regulation technology. Therefore, an intelligent cabin environment adaptive regulation technology based on a large AI model is proposed. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an intelligent cockpit environment adaptive adjustment technology based on an AI large model. This technology offers comprehensive support for adjustment decisions by integrating multi-dimensional sensory information; it achieves a balance between dynamic response and adjustment accuracy through an AI large model decision module that integrates multiple sub-models; it balances comfort, battery safety, and range economy by linking cockpit and power battery temperature control; and it precisely adapts to user comfort needs through real-time decision adjustments via a feedback module. These advantages solve the problems of insufficient robustness and adjustment accuracy caused by the single AI control algorithm in existing cockpit environment adjustment technologies, as well as the vicious cycle of energy consumption and the inability to balance multiple objectives caused by the separation of cockpit and power battery temperature control.

[0005] (II) Technical Solution To achieve the above-mentioned goals of providing comprehensive support for adjustment decisions by integrating multi-dimensional sensing information; balancing dynamic response and adjustment accuracy through an AI large-model decision module that integrates multiple sub-models; achieving a balance between comfort, battery safety, and range economy by linking cabin and power battery temperature control; and precisely adapting to user comfort needs through a feedback module that adjusts decisions in real time, this invention provides the following technical solution: an intelligent cockpit environment adaptive adjustment system based on an AI large-model, comprising a sensing module, an AI large-model decision module, an execution module, and a feedback module; The sensing module includes an environmental parameter sensor group, a user physiological parameter sensor group, and a vehicle operating condition sensor group, which respectively collect cabin environmental parameters, user physiological parameters, and vehicle operating condition parameters. The AI ​​large model decision module is communicatively connected to the perception module, including a user behavior learning sub-model, a multi-physics prediction sub-model and a hybrid decision sub-model. The user behavior learning sub-model adopts an LSTM neural network, the multi-physics prediction sub-model is built on Simulink and integrates the cabin fluid-heat transfer coupling model and the power battery electrothermal coupling model, and the hybrid decision sub-model adopts the MPC-TD3 fusion algorithm. The execution module is communicatively connected to the AI ​​large model decision-making module and includes an air conditioning execution unit, a TEC regulation unit, and a battery cooling execution unit, which respectively adjust the cooling capacity, TEC air outlet parameters, and battery cooling intensity. The air conditioning execution unit and the battery cooling execution unit are equipped with control logic and communication interfaces adapted to the AI ​​large model decision-making requirements. The TEC device of the TEC regulation unit is a general thermoelectric refrigeration module and adopts a multi-position distributed installation layout and an adjustable jet angle structure, which is adapted to the cabin zone temperature control requirements. The feedback module is communicatively connected to the perception module, the AI ​​large model decision module, and the execution module. It uses the PMV-PPD evaluation model and the SET evaluation model to calculate real-time thermal comfort parameters. The PMV-PPD evaluation model is used to quantify the user's overall thermal comfort experience, while the SET evaluation model is used to accurately reflect the impact of the local cabin environment on the user's thermal perception. The feedback module transmits the real-time PMV and SET values ​​to the hybrid decision sub-model. When the SET value deviates from the user's thermal preference temperature output by the user behavior learning sub-model by ±1℃, the weight ω1 of the MPC cost function in the hybrid decision sub-model is adjusted synchronously, increasing ω1 by 10% to prioritize cabin thermal comfort.

[0006] Preferably, the environmental parameters collected by the environmental parameter sensor group include cabin air temperature, cabin relative humidity, cabin CO2 concentration, solar radiation intensity, and cabin wall temperature. The cabin air temperature is measured in the range of -10℃ to 60℃, with an accuracy of ±0.1℃; the cabin relative humidity is measured in the range of 20%RH to 80%RH, with an accuracy of ±1%RH; the cabin CO2 concentration is measured in the range of 400ppm to 5000ppm, with an accuracy of ±5ppm; the solar radiation intensity is measured in the range of 0W / m² to 1200W / m², with an accuracy of ±10W / m²; and the cabin wall temperature is measured in the range of -10℃ to 80℃, with an accuracy of ±0.2℃.

[0007] Preferably, the training samples for the user behavior learning sub-model are operation records and physiological data collected by the user's physiological parameter sensor group over the past 30 days, with a sample size of not less than 5000 groups; the training process uses the Adam optimizer with a learning rate of 1×10⁻³ and an iteration count of not less than 200 rounds, and training stops when the validation set loss function MSE is not greater than 0.01; the model parameters are updated every 7 days; during training, the collected user operation records are preprocessed to remove abnormal operations with adjustment values ​​greater than 5℃ or less than 0.5℃, and then the preprocessed operation records are time-series aligned with the physiological data of the same period to construct an input feature vector with a dimension of 12.

[0008] Preferably, the cabin fluid-heat coupling model of the multiphysics prediction sub-model adopts the Realizable k-ε turbulence model and the DO radiation model; the wall functions of the Realizable k-ε turbulence model are Non-Equilibrium wall functions, and the Theta Divisions and Phi Divisions of the DO radiation model are 3. The electrothermal coupling model of the power battery adopts the Rint equivalent circuit model and the lumped parameter thermal model. The ohmic internal resistance of the Rint equivalent circuit model ranges from 1.0 mΩ to 5.0 mΩ, and the specific heat capacity of the lumped parameter thermal model is 3422 J / (kg・K). The multiphysics prediction sub-model outputs the predicted temperature values ​​of each area of ​​the cabin and the power battery within the next 600 seconds. The accuracy of the predicted temperature values ​​of each area of ​​the cabin is ±0.3℃, and the accuracy of the predicted temperature value of the power battery is ±0.2℃.

[0009] Preferably, the MPC prediction time domain of the hybrid decision sub-model is 20 steps and the control time domain is 10 steps, and the cost function expression is: In the formula, For cabin temperature, For users' preferred thermal temperatures, For the temperature of the power battery, The target temperature for the power battery, This refers to the compressor speed. This refers to the evaporator fan speed. =4000、 =2000、 =1000、 =800; The experience pool sample size for the TD3 algorithm is 1×102. 6 The soft update parameter is 1×10⁻³, the discount factor is 0.995, the learning rate of the Actor network is 1×10⁻³, and the learning rate of the Critic network is 1×10⁻⁻⁶. 4 .

[0010] Preferably, the TD3 algorithm of the hybrid decision sub-model adopts a main-auxiliary reward function, and the total reward expression is as follows: The main reward function expression is: The expression for the auxiliary line reward function is: In the formula, =4000、 =2000、 =1000, PMV is the predicted average vote value. For the temperature of the power battery, =25℃, This represents the total energy consumption for air conditioning and battery cooling. This refers to the evaporator wall temperature. For cabin temperature, The user's preferred temperature.

[0011] Preferably, the air conditioning execution unit of the execution module includes a DC compressor and an evaporator fan. The DC compressor has a displacement of 34cc, and the evaporator fan has a maximum airflow of 300m³ / h. The air conditioning execution unit integrates a dedicated AI control submodule. This control submodule receives speed commands from the AI ​​large-scale model decision module via a CAN bus, with a baud rate set to 500kbps. An airflow distribution valve is added to the connection section between the evaporator fan and the existing cabin air duct. Based on the cabin area temperature deviation output by the decision module, the airflow ratio of each air duct branch is adjusted. When the temperature deviation is greater than 1℃, the airflow of the corresponding air duct is increased by 20%. The TEC regulating unit includes five TEC devices, each with a cooling capacity ranging from 33W to 43W and a COP ranging from 0.27 to 1.12. Each TEC device is fixed by an L-shaped aluminum alloy bracket with a thickness of 2mm. High-temperature resistant silicone pads with a temperature resistance range of -40℃ to 120℃ are adhered between the bracket and the cabin components (windshield and instrument panel). TEC1 and TEC2 are fixed to the upper part of the windshield at a vertical distance of 670mm from the user, while TEC3 and TEC4 are fixed to the upper part of the instrument panel at a horizontal distance of 600mm from the user. mm, TEC5 is fixed directly above the human body; the cross angle of the TEC jets is adjusted by a stepper motor, with gear transmission and an adjustment step of 0.5°. At the same time, the angle sensor feeds back the angle information to the decision module in real time to form a closed-loop control. The cross angle range of TEC1 and TEC2 is 0° to 19.5°, and the cross angle range of TEC3 and TEC4 is 0.2° to 21.5°. The air outlets of each TEC device are connected to the existing windshield duct, instrument panel duct and ceiling duct of the cabin through flexible air ducts. The air duct interface adopts a snap-on design, which does not require modification of the original air duct body. The battery cooling unit includes an electric water pump and a plate heat exchanger. The electric water pump has a head of 6m and a maximum flow rate of 60L / min. It also integrates a flow sensor with an accuracy of ±2%FS, which can feed the flow information back to the decision module in real time. The plate heat exchanger has a brazed structure, with plates made of 316 stainless steel, a corrugation angle of 30°, and a heat exchange area of ​​0.8m². The coolant inlet of the plate heat exchanger is connected to the outlet of the existing cooling circuit of the power battery pack via a quick-connect coupling. The outlet of the plate heat exchanger is connected to the inlet of the electric water pump, forming a series circulation path of "battery pack → plate heat exchanger → electric water pump → battery pack".

[0012] Preferably, the expression for calculating the PMV value of the feedback module is: In the formula, This refers to the human metabolic rate, with values ​​ranging from 1.0 MET to 1.5 MET. It is the partial pressure of water vapor. This is the surface area coefficient of clothing. The surface temperature of the clothing. The mean radiation temperature. The heat transfer coefficient of the human body surface; the expression for calculating the SET value is: In the formula, The airflow velocity inside the cabin. The relative humidity inside the cabin; when When the value deviates from the user's preferred temperature by ±1℃, it is fed back to the hybrid decision sub-model and included in the cost function. Increase by 10%.

[0013] Preferably, the arrangement of the environmental parameter sensor group satisfies: The cabin air temperature sensors are located at the front, middle and rear of the cabin, and are all 1.2m above the ground. The relative humidity sensor and CO2 concentration sensor in the cabin are integrated in the central armrest, which is 0.8m above the ground; The solar radiation intensity sensor is located on the exterior of the vehicle roof and faces directly upwards. The cabin wall temperature sensors are affixed to the windshield, instrument panel, and seat back, with two sensors affixed to each component.

[0014] Preferably, the coolant flow rate regulation logic of the battery cooling execution unit is as follows: The average temperature of the power battery is collected every 10 seconds. ,when At temperatures above 27°C, the coolant flow rate is set to 15 L / min. When 25℃≤ When the temperature is ≤27℃, the coolant flow rate is set to 12L / min; when When the temperature is below 25℃, the coolant flow rate is set to 8L / min.

[0015] (III) Beneficial Effects Compared with existing technologies, this invention provides an intelligent cockpit environment adaptive adjustment technology based on an AI large model, which has the following beneficial effects: 1. This AI-based intelligent cockpit environment adaptive adjustment technology utilizes the MPC-TD3 fusion algorithm in the hybrid decision-making sub-model of the AI ​​large-scale model decision-making module. First, the user behavior learning sub-model of this module learns the user's physiological parameters and operation records collected by the perception module. The multi-physics prediction sub-model predicts the cockpit environment and power battery status. Then, it combines the comfort parameters fed back in real time by the PMV-PPD evaluation model and SET evaluation model of the feedback module. Revise decision results This addresses the shortcomings of a single MPC model in robustness under complex operating conditions and the difficulty of a single reinforcement learning model in quantifying user comfort, achieving a balance between dynamic response speed and adjustment accuracy.

[0016] 2. This AI-based intelligent cockpit environment adaptive adjustment technology uses an AI big data model decision module to receive cabin environment parameters and vehicle operating condition parameters collected by the perception module. The vehicle operating condition parameters include power battery-related parameters. The decision module then synchronously sends decision commands to the air conditioning execution unit, TEC adjustment unit, and battery cooling execution unit of the execution module. Link cabin temperature regulation with power battery temperature regulation They are no longer separated, avoiding the vicious cycle of high-load air conditioning cooling exacerbating battery discharge heat generation and strong battery cooling consuming additional energy under high-temperature conditions, thus balancing cabin comfort, battery safety, and range economy.

[0017] 3. This AI-based intelligent cockpit environment adaptive adjustment technology continuously feeds real-time PMV and SET values ​​back to the hybrid decision-making sub-model of the AI-based big data model decision-making module. The hybrid decision-making sub-model then adjusts the functions of each unit in the execution module accordingly. Adjustment movement Meanwhile, user behavior learning sub-model Continuously update user habit data Multiphysics prediction sub-model Anticipate changes in cabin and battery temperature in advance This allows decisions to better align with actual needs, addressing the shortcomings of a single AI algorithm while enhancing the synergy between cabin and battery temperature control, thereby improving the accuracy of adaptive adjustment and the overall vehicle energy efficiency. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the intelligent cockpit environment adaptive adjustment system of the present invention; Figure 2 This is a flowchart of the collaborative decision-making and feedback correction process of the three modules of the AI ​​large model decision-making module of the present invention; Figure 3 This is a flowchart illustrating the generation and execution of cabin-power battery temperature control linkage commands according to the present invention. Figure 4 This is a flowchart of the adaptive adjustment three-action closed-loop optimization process of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1-4 The intelligent cockpit environment adaptive adjustment system based on AI big data model includes a perception module, an AI big data model decision module, an execution module and a feedback module; The sensing module includes an environmental parameter sensor group, a user physiological parameter sensor group, and a vehicle operating condition sensor group, which respectively collect cabin environmental parameters, user physiological parameters, and vehicle operating condition parameters. The AI ​​large model decision module is communicatively connected to the perception module, including a user behavior learning sub-model, a multi-physics prediction sub-model and a hybrid decision sub-model. The user behavior learning sub-model adopts an LSTM neural network, the multi-physics prediction sub-model is built on Simulink and integrates the cabin fluid-heat transfer coupling model and the power battery electrothermal coupling model, and the hybrid decision sub-model adopts the MPC-TD3 fusion algorithm. The execution module is communicatively connected to the AI ​​large model decision-making module and includes an air conditioning execution unit, a TEC regulation unit, and a battery cooling execution unit, which respectively adjust the cooling capacity, TEC air outlet parameters, and battery cooling intensity. The air conditioning execution unit and the battery cooling execution unit are equipped with control logic and communication interfaces adapted to the AI ​​large model decision-making requirements. The TEC device of the TEC regulation unit is a general thermoelectric refrigeration module and adopts a multi-position distributed installation layout and an adjustable jet angle structure, which is adapted to the cabin zone temperature control requirements. The feedback module is communicatively connected to the perception module, the AI ​​large model decision module, and the execution module. It uses the PMV-PPD evaluation model and the SET evaluation model to calculate real-time thermal comfort parameters. The PMV-PPD evaluation model is used to quantify the user's overall thermal comfort experience, while the SET evaluation model is used to accurately reflect the impact of the local cabin environment on the user's thermal perception. The feedback module transmits the real-time PMV and SET values ​​to the hybrid decision sub-model. When the SET value deviates from the user's thermal preference temperature output by the user behavior learning sub-model by ±1℃, the weight ω1 of the MPC cost function in the hybrid decision sub-model is adjusted synchronously, increasing ω1 by 10% to prioritize cabin thermal comfort.

[0021] Example 1: This embodiment realizes the hardware deployment, software module initialization, and communication link establishment of the intelligent cockpit environment adaptive adjustment system, providing a foundation for the implementation of the adaptive adjustment function.

[0022] First, the hardware selection and parameter configuration of the sensing and execution modules were carried out, with the selection and placement of sensors for cabin temperature, humidity, CO2 concentration, and solar radiation intensity based on the design. Among the key execution components, the TEC device selected a thermoelectric cooling module with specifications of TEC1-16013FX, with a maximum cooling capacity of a single unit. This value is based on a hot-end temperature of 27°C, and its cooling capacity is calculated using a formula based on a simplified energy balance model: In the formula, This refers to the cooling capacity of the TEC cold end, expressed in W. Ii is the Seebeck coefficient, with a value of 0.0309 V / K; II is the input current, in amperes (A). RR is the cold junction temperature in Kelvin; RR is the TEC internal resistance, with a value of 1.1068Ω. The thermal conductivity of TEC is taken as 1.3959 W / K; This refers to the hot end temperature, expressed in Kelvin (K). This represents the heat output of the TEC hot end, expressed in W.

[0023] The air conditioner compressor uses a DC inverter type, and its mass flow rate and outlet specific enthalpy formula is as follows: In the formula, This refers to the compressor's mass flow rate, expressed in kg / s. For volumetric efficiency; This refers to the density of the refrigerant at the compressor inlet, expressed in kg / m³. This refers to the compressor speed, measured in rpm. This refers to the compressor displacement, taken as 34cc. Specific enthalpy at compressor outlet, in kJ / kg; Specific enthalpy at compressor inlet, in kJ / kg; This is the refrigerant outlet specific enthalpy under isentropic compression, expressed in kJ / kg. It is isentropic efficiency.

[0024] Next, the core sub-models of the AI ​​large-scale model decision-making module are constructed. Each sub-model is optimized for the intelligent cockpit scenario, and the rules for the targeted allocation of perception data are clearly defined, as follows: The user behavior learning sub-model adopts an LSTM neural network. Two optimization designs were added in the training data preprocessing stage: First, a "physiological-operation time sequence alignment algorithm" was proposed, which accurately matches the user's physiological parameters (body temperature, heart rate) collected by the perception module with the operation records of temperature adjustment and airflow adjustment in the past 30 days (sample size not less than 5000 groups) according to 1 second timestamps to construct an input feature vector with a dimension of 12. Second, a "dynamic anomaly removal rule" was designed, which removes abnormal operations with adjustment amounts greater than 5℃ or less than 0.5℃ based on the reasonable range of cabin temperature adjustment from -10℃ to 60℃, and then completes the missing data through neighborhood interpolation.

[0025] Training employs the Adam optimizer with a learning rate of 1×10⁻³ and at least 200 iterations. Training stops when the validation set loss function MSE is no greater than 0.01. Model parameters are updated every 7 days based on feedback module data to ensure the accuracy of user-preferred temperature data. The prediction accuracy is controlled within ±0.3℃.

[0026] In terms of data distribution, only user physiological parameter sensor group data and user historical operation records are received from the sensing module, and environmental parameters or vehicle operating condition parameters are not connected to avoid data interference.

[0027] The multiphysics prediction sub-model is built on Simulink, integrating the cabin fluid-heat transfer coupling model and the power battery electrothermal coupling model. Core optimization designs include: The "cabin-battery energy consumption coupling equation" uses the cabin air conditioning cooling capacity collected by the sensing module as the external heat load input to the power battery thermal model, while also incorporating the power battery's Joule heat... As the internal heat source of the cabin fluid model (where The battery current is collected by the sensing module. The internal resistance of the Rint equivalent circuit model is ohmic (ranging from 1.0mΩ to 5.0mΩ), enabling temperature-linked prediction between the two. Second, the model parameter adaptability was optimized. The cabin fluid-heat transfer coupling model adopts the Realizable k-ε turbulence model (with wall functions set to Non-Equilibrium wall functions) and the DO radiation model (ThetaDivisions are 3, PhiDivisions are 3). Its turbulence equation is: In the formula, This refers to the fluid density, expressed in kg / m³. Time, in seconds; It is turbulent kinetic energy, with units of m² / s²; This is a velocity vector, with units of m / s; ν is the molecular viscosity coefficient, with units of Pa·s; Turbulent viscosity, in Pa·s; The Prandtl number represents the turbulent kinetic energy, with a value of 1.0. The Prandtl number represents the turbulent dissipation rate, with a value of 1.3. This is the turbulent dissipation rate, with units of m² / s³. This is an empirical constant with a value of 1.44; This is an empirical constant with a value of 1.92; This is an empirical constant with a value of 0.09; This is the turbulent kinetic energy generation term, with units of W / m³.

[0028] The electrothermal coupling model of the power battery adopts the Rint equivalent circuit and the lumped parameter thermal model, with a specific heat capacity of 3422 J / (kg·K). The formula for convective heat transfer is: In the formula, This refers to convective heat transfer, expressed in W. The convective heat transfer coefficient is taken as 15 W / (m²·K); The battery heat dissipation area is set to 0.5 m². This refers to the surface temperature of the battery, in °C. The ambient air temperature is expressed in °C.

[0029] Ultimately, the accuracy of temperature prediction for all areas of the cabin within the next 600 seconds will be ±0.3℃, and the accuracy of temperature prediction for the power battery will be ±0.2℃.

[0030] In terms of data allocation, only the environmental parameter sensor group data and the vehicle operating condition sensor group data of the perception module are received as the initial boundary conditions of the model.

[0031] The hybrid decision-making sub-model adopts the MPC-TD3 fusion algorithm, and its core optimization design is a "multi-objective cost function". In addition to the comfort and economy objectives, a power battery safety penalty term is added. The cost function formula is as follows: In the formula, The cost function value; To predict the step size; For the first Cockpit temperature, in °C; This refers to the user's preferred temperature, expressed in °C. For the first Step power battery temperature, in °C; For the first Step compressor speed, in rpm; For the first Step evaporator fan speed, in rpm.

[0032] The TD3 algorithm uses a mainline-secondaryline reward function, with a total reward of... The number of samples in the experience pool is 1×10. 6 The soft update parameter is 1×10⁻³, the discount factor is 0.995, the learning rate of the Actor network is 1×10⁻³, and the learning rate of the Critic network is 1×10⁻⁻⁶. 4 .

[0033] Regarding data input sources, the data received is the output of the user behavior learning sub-model. The multiphysics prediction sub-model outputs predicted cabin and battery temperatures for the next 600 seconds, and the feedback module outputs real-time PMV and SET values.

[0034] The TD3 algorithm's reward function introduces the relationship between TEC cooling capacity and energy consumption, as shown in the formula: In the formula, Total reward value; To predict the average thermal sensation index; This is the real-time temperature of the power battery, in °C. Total energy consumption, in kWh; This is the bonus value for evaporator wall temperature. It is set to 2000 when the evaporator wall temperature is greater than 0℃, and -2000 otherwise. This is the cabin temperature deviation bonus value. It is set to 2000 when the absolute value of the difference between the cabin temperature and the user's preferred temperature does not exceed 1℃, and otherwise it is set to -2000.

[0035] Finally, the feedback module formula deployment and communication link establishment were completed. PMV calculation adopted the Fanger formula. In the formula, To predict the average thermal sensation index; The rate of heat production from human metabolism is expressed in W / m². The amount of work done by the human body, expressed in W / m². This refers to the heat exchanged between the human body and the environment via convection, expressed in W / m². The heat exchanged between the human body and the environment is expressed in W / m². This refers to the amount of heat dissipated through human skin by evaporation, expressed in W / m². This refers to the heat dissipation caused by human respiration, expressed in W / m². This refers to the amount of heat dissipated through evaporation during human respiration, expressed in W / m².

[0036] SET calculations use the standard effective temperature formula: In the formula, Standard effective temperature, in °C; The cabin air temperature is expressed in °C. The average radiation temperature is expressed in °C. Air velocity, measured in m / s; This represents relative humidity, expressed as a percentage (%).

[0037] In the communication link, the sensing module and the decision module transmit data via Ethernet with a baud rate of 100Mbps. The IP addresses are fixed as 192.168.1.10 for the sensing module and 192.168.1.20 for the decision module. The feedback module transmits the PMV and SET values ​​to the hybrid decision sub-model every 0.5s. When the SET deviates from the user's thermal preference temperature by ±1℃, the weight ω1 of the MPC cost function is adjusted from 4000 to 4400.

[0038] Example 2: This embodiment enables the training and optimization of user behavior learning and multiphysics prediction models to ensure decision-making accuracy.

[0039] First, a user behavior learning model is trained. Training data is collected according to user operation and physiological data collection requirements. In the preprocessing stage, outlier data is removed, and a 12-dimensional feature vector is constructed. Feature normalization uses the Min-Max normalization formula. In the formula, These are the normalized eigenvalues; These are the original eigenvalues; The minimum characteristic value is the temperature regulation characteristic. The value is 0.5℃; The maximum characteristic value is the temperature regulation characteristic. The value is 5℃.

[0040] The model was trained using the Adam optimizer with a learning rate of 1×10⁻³ and a loss function of mean squared error. In the formula, Mean square error; The number of samples; For the first Predicted values ​​for each sample; For the first The model calculates the true values ​​of each sample. It undergoes 200 training iterations, with an early stopping condition of validation set MSE ≤ 0.01. After training, the model's prediction deviation for user thermal preference temperatures should not exceed 0.5℃.

[0041] The multiphysics prediction model was validated using three typical operating conditions: high temperature (35℃), normal temperature (25℃), and low temperature (10℃), each lasting 600 seconds. Data was collected from nine temperature sensors in the cockpit and six temperature sensors in the battery, at a frequency of 1 Hz. The TEC jet characteristic analysis in the model employed the cross-jet fusion zone formula. In the formula, This is the starting distance of the fusion zone, in meters. The jet intersection angle is expressed in degrees (°). This represents the distance at the end of the fusion zone, in meters.

[0042] Based on this formula, the cross angle range of TEC1 and TEC2 is determined to be 0°~19.5°, and the cross angle range of TEC3 and TEC4 is determined to be 0.2°~21.5°.

[0043] During the model optimization phase, adjustments were made to the Realizable model to address the temperature deviation at the rear of the cockpit under high-temperature conditions. The wall roughness of the model was increased from 0.01 mm to 0.02 mm; the calculation parameters of the TEC cooling capacity were corrected based on the recorded fitting formula between the TEC outlet air temperature drop and the input power. In the formula, The temperature drop of the TEC inlet and outlet airflow is expressed in °C. This represents the TEC input power, measured in W. After optimization, the predicted temperature deviation of the power battery is stabilized within ±0.2℃.

[0044] Example 3: This embodiment achieves collaboration between the decision-making and execution modules, balancing cabin comfort, power battery safety, and overall vehicle energy efficiency.

[0045] In the hybrid decision sub-model, the MPC controller has 20 prediction steps corresponding to 600 seconds in the prediction time domain and 10 control steps in the control time domain. The constraints refer to the parameters of the air conditioning components, specifically the compressor speed of 1000-6000 rpm, the evaporator fan speed of 500-2000 rpm, and the TEC outlet air temperature of 15-30℃. The optimization solution uses the interior point method. After obtaining the control quantity, it is weighted and fused with the output of the TD3 algorithm. In the formula, The final compressor speed, in rpm; This refers to the compressor speed output by the MPC, in rpm. This represents the compressor speed output by the TD3 algorithm, in rpm. This refers to the final evaporator fan speed, in rpm. This refers to the evaporator fan speed output by the MPC, in rpm. This is the evaporator fan speed output by the TD3 algorithm, in rpm.

[0046] The TEC actuator module control employs a differentiated adjustment strategy, adjusting the outlet air temperature based on the difference in heat gain between the left and right sides of the body. The outlet air temperature calculation is based on the TEC cooling capacity formula. In the formula, This refers to the TEC cooling capacity, expressed in watts (W). The density of air is taken as 1.184 kg / m³. The specific heat capacity of air is taken as 1.005 kJ / (kg·K); The outlet air velocity of the TEC is expressed in m / s. The value of the TEC air outlet area is 0.00105 m². The temperature of the TEC inlet air is in °C. This represents the TEC outlet air temperature, in °C. The required outlet air temperature can be calculated using this formula. The final air outlet temperatures for TEC1 and TEC3 were determined to be 24.14℃, for TEC2 28.46℃, and for TEC4 27.61℃.

[0047] The battery cooling actuator uses flow control logic based on the average battery temperature. Adjusting the PWM duty cycle of the electronic water pump, the flow rate calculation formula is as follows: In the formula, This refers to the coolant flow rate, expressed in L / min. The flow coefficient is 0.25 L / (min·%). This represents the PWM duty cycle of the electronic water pump, in units of %.

[0048] The specific control logic is as follows: >27℃ =60 corresponds to a flow rate of 15L / min; 25℃≤ ≤27℃ =48 corresponds to a flow rate of 12L / min; <25℃ =32 corresponds to a flow rate of 8L / min.

[0049] Example 4: This embodiment implements the deployment of the feedback module adjustment logic and the overall system verification to ensure the reliability of adaptive adjustment.

[0050] In the feedback module, PMV calculation strictly follows the Fanger formula, where convective heat transfer... Radiative heat exchange Heat flux calculation method: In the formula, This refers to convective heat transfer, expressed in W. The convective heat transfer coefficient is taken as 10 W / (m²·K); This refers to the surface area of ​​the human body, expressed in m². Human skin temperature, expressed in °C; The ambient air temperature is expressed in °C. This refers to radiative heat transfer, measured in W. The skin emissivity is set to 0.97. Here is the Boltzmann constant, with a value of 5.67 × 10⁻ 8 W / (m²·K 4 ); The average radiant temperature is expressed in Kelvin. The PMV value is calculated every 0.5 seconds. If PMV > 0.5, the compressor speed is increased by 5%; if PMV < -0.5, the compressor speed is decreased by 5%.

[0051] SET feedback adjustment employs a deviation correction strategy. When SET deviates from the user's preferred temperature by 25℃±1℃, in addition to adjusting the MPC weights... In addition, the TEC outlet air temperature is corrected using a heat flux difference formula: In the formula, This represents the difference in heat flux between the left and right sides of the human body, expressed in W / m². This represents the heat flux on the left side of the human body, expressed in W / m². This represents the heat flux on the right side of the human body, expressed in W / m². The corrected difference in heat flux between the left and right sides of the human body should not exceed 5 W / m².

[0052] In summary, this AI-based intelligent cockpit environment adaptive adjustment technology achieves precise adjustment primarily through a three-module collaborative decision-making mechanism within the AI ​​large-scale model decision-making module. First, the user behavior learning sub-model is based on an LSTM neural network. It learns user physiological parameters such as body temperature and heart rate collected by the perception module, as well as historical operation records such as temperature adjustment operations in the past 30 days, and outputs the user's personalized thermal preference temperature. ; Secondly, the multiphysics prediction sub-model integrates the cabin fluid-heat transfer coupling model and the power battery electrothermal coupling model, receives cabin environmental parameters such as air temperature and solar radiation from the sensing module and vehicle operating parameters such as battery current and voltage, and predicts the temperature of each area of ​​the cabin with an accuracy of ±0.3℃ and the power battery temperature with an accuracy of ±0.2℃ within the next 600s. Finally, the hybrid decision sub-model adopts the MPC-TD3 fusion algorithm, which uses the user's thermal preference temperature plus the temperature prediction value as the basis for decision-making, and combines the PMV value output by the feedback module in real time to quantify the overall thermal comfort and the SET value to quantify the local thermal perception correction decision. If the SET value deviates If the temperature deviation is ±1℃, the cabin temperature deviation weight ω1 in the MPC cost function will be increased by 10%. At the same time, the reward value of the TD3 algorithm will be dynamically adjusted according to the following rules, and finally, the adapted control command will be output to make up for the shortcomings of the single model and balance dynamic response and adjustment accuracy: PMV-based reward tier adjustment: When PMV is in the comfortable range of -0.5 to 0.5, a basic positive reward is given, specifically calculated as -4000 multiplied by the absolute value of PMV plus 1000; when the absolute value of PMV is greater than 0.5 and enters the uncomfortable range, the positive reward is canceled and a penalty is added, calculated as -4000 multiplied by the absolute value of PMV. If the absolute value of PMV is greater than 1.0, an additional reward of 2000 will be deducted. Co-verification with SET: If SET deviates from the user's preferred thermal temperature If the temperature reaches ±1℃, even if the PMV is within the comfortable range, the reward will still be deducted. The deduction amount is determined according to the degree of deviation, ranging from 1000 to 3000. The greater the deviation, the more reward will be deducted. Reward smoothing constraint: To avoid sudden changes in reward, a quadratic function is used instead of linear penalty, specifically calculated as -4000 multiplied by the square of PMV; at the same time, the boundary of the reward value is set to -10000 to 2000 to prevent oscillations during the training process.

[0053] This AI-based intelligent cockpit environment adaptive adjustment technology achieves coordinated temperature control between the cockpit and the power battery through data linkage and command synchronization. The first step is for the sensing module to collect cabin environmental parameters such as air temperature and relative humidity, and vehicle operating parameters including the power battery temperature. Current is synchronously transmitted to the AI ​​large-scale model decision-making module; The second step involves analyzing the data using a multiphysics prediction sub-model. If the prediction... If the battery temperature exceeds the upper limit of the safety threshold of 27℃, the high temperature warning signal will be transmitted to the hybrid decision sub-model. The third step involves the hybrid decision sub-model generating linkage commands: on the one hand, sending commands to the air conditioning execution unit to reduce the cooling load, such as reducing the compressor speed from 6000rpm to 4500rpm, to avoid high load power consumption and increased battery heat generation; on the other hand, sending commands to the battery cooling execution unit to increase the coolant flow rate from 12L / min to 15L / min, while controlling the TEC adjustment unit to focus on cooling the core area of ​​the cabin, such as adjusting the air outlet angle of TEC1-TEC2 to 15°, prioritizing cooling of the front passenger area; Commands are simultaneously sent to three execution units, breaking away from the traditional separate temperature control, avoiding a vicious cycle of energy consumption, and balancing comfort, safety, and battery economy.

[0054] This AI-based intelligent cockpit environment adaptive adjustment technology achieves adaptive optimization through three core adjustment actions: real-time feedback, dynamic updates, and advance prediction. ① Adjustment actions of the hybrid decision sub-model: Receive the PMV / SET value transmitted by the feedback module every 0.5s. If PMV > 0.5, the user is too hot. Adjust the air conditioning execution unit to increase the air volume and the evaporator fan speed from 1500rpm to 1800rpm. Adjust the TEC adjustment unit to reduce the outlet air temperature from 25℃ to 23℃. At the same time, fine-tune the flow rate of the battery cooling execution unit to avoid excessive power consumption. ② Adjustment actions of the user behavior learning sub-model: Every 7 days, the model parameters are updated based on the user's physiological data and operation records added by the perception module, after removing abnormal operations. For example, if a user has frequently adjusted the temperature from 24℃ to 26℃ recently, then... Updated to 25.5℃ to ensure decisions align with the latest practices; ③ Adjustment actions of the multiphysics prediction sub-model: Based on real-time environmental parameters such as a sudden increase in solar radiation intensity from 800W / m² to 1200W / m², it predicts 200 seconds in advance that the temperature at the front of the cabin will rise to over 28°C. Send a pre-adjustment signal to the hybrid decision sub-model in advance so that the execution unit can start fine-tuning 100 seconds in advance, such as pre-increasing the compressor speed by 500 rpm, to avoid overheating. The three actions work in a cyclical and coordinated manner to continuously improve the accuracy of regulation and energy efficiency.

[0055] It solves the problems of insufficient robustness and adjustment accuracy caused by the single AI control algorithm in existing cabin environment regulation technology, as well as the vicious cycle of energy consumption caused by the separation of cabin and power battery temperature control and the inability to balance multiple objectives.

[0056] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the prior art. The computer software programs or protocols involved in these functional modules are technologies known to those skilled in the art and are not improvements to this system. The improvement of this system lies in the interaction or connection between the modules, that is, in improving the overall structure of the system to solve the corresponding technical problems that this system aims to address.

[0057] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent cockpit environment adaptive adjustment system based on an AI large model, characterized in that, It includes a perception module, an AI large-scale model decision-making module, an execution module, and a feedback module; The sensing module includes an environmental parameter sensor group, a user physiological parameter sensor group, and a vehicle operating condition sensor group, which respectively collect cabin environmental parameters, user physiological parameters, and vehicle operating condition parameters. The AI ​​large model decision module is communicatively connected to the perception module, including a user behavior learning sub-model, a multi-physics prediction sub-model and a hybrid decision sub-model. The user behavior learning sub-model adopts an LSTM neural network, the multi-physics prediction sub-model is built on Simulink and integrates the cabin fluid-heat transfer coupling model and the power battery electrothermal coupling model, and the hybrid decision sub-model adopts the MPC-TD3 fusion algorithm. The execution module is communicatively connected to the AI ​​large model decision-making module and includes an air conditioning execution unit, a TEC regulation unit, and a battery cooling execution unit, which respectively adjust the cooling capacity, TEC air outlet parameters, and battery cooling intensity. The air conditioning execution unit and the battery cooling execution unit are equipped with control logic and communication interfaces adapted to the AI ​​large model decision-making requirements. The TEC device of the TEC regulation unit is a general thermoelectric refrigeration module and adopts a multi-position distributed installation layout and an adjustable jet angle structure, which is adapted to the cabin zone temperature control requirements. The feedback module is communicatively connected to the perception module, the AI ​​big model decision module and the execution module respectively. It uses the PMV-PPD evaluation model and the SET evaluation model to calculate real-time thermal comfort parameters. The PMV-PPD evaluation model is used to quantify the user's overall thermal comfort, while the SET evaluation model is used to accurately reflect the impact of the local cabin environment on the user's thermal perception. The feedback module transmits real-time PMV and SET values ​​to the hybrid decision sub-model. When the SET value deviates from the user thermal preference temperature output by the user behavior learning sub-model by ±1℃, the weight ω1 of the MPC cost function in the hybrid decision sub-model is adjusted synchronously, increasing ω1 by 10% to prioritize cabin thermal comfort.

2. The intelligent cockpit environment adaptive adjustment system based on an AI large model according to claim 1, characterized in that, The environmental parameters collected by the environmental parameter sensor group include cabin air temperature, cabin relative humidity, cabin CO2 concentration, solar radiation intensity, and cabin wall temperature. The cabin air temperature is measured in the range of -10℃ to 60℃, with an accuracy of ±0.1℃; the cabin relative humidity is measured in the range of 20%RH to 80%RH, with an accuracy of ±1%RH; the cabin CO2 concentration is measured in the range of 400ppm to 5000ppm, with an accuracy of ±5ppm; the solar radiation intensity is measured in the range of 0W / m² to 1200W / m², with an accuracy of ±10W / m²; and the cabin wall temperature is measured in the range of -10℃ to 80℃, with an accuracy of ±0.2℃.

3. The intelligent cockpit environment adaptive adjustment system based on an AI large model according to claim 1, characterized in that, The training samples for the user behavior learning sub-model are the operation records and physiological data collected by the user's physiological parameter sensor group over the past 30 days, with a sample size of no less than 5000 groups. The training process uses the Adam optimizer with a learning rate of 1×10⁻³ and an iteration count of no less than 200 rounds. Training stops when the validation set loss function MSE is no greater than 0.

01. The model parameters are updated every 7 days. During training, the collected user operation records are first preprocessed to remove abnormal operations with adjustment values ​​greater than 5℃ or less than 0.5℃. Then, the preprocessed operation records are time-series aligned with the physiological data of the same period to construct an input feature vector with a dimension of 12.

4. The intelligent cockpit environment adaptive adjustment system based on an AI large model according to claim 1, characterized in that, The cabin fluid-heat coupling model of the multiphysics prediction sub-model adopts the Realizable k-ε turbulence model and the DO radiation model; the wall functions of the Realizable k-ε turbulence model are Non-Equilibrium wall functions, and the Theta Divisions and Phi Divisions of the DO radiation model are 3. The electrothermal coupling model of the power battery adopts the Rint equivalent circuit model and the lumped parameter thermal model. The ohmic internal resistance of the Rint equivalent circuit model ranges from 1.0 mΩ to 5.0 mΩ, and the specific heat capacity of the lumped parameter thermal model is 3422 J / (kg・K). The multiphysics prediction sub-model outputs the predicted temperature values ​​of each area of ​​the cabin and the power battery within the next 600 seconds. The accuracy of the predicted temperature values ​​of each area of ​​the cabin is ±0.3℃, and the accuracy of the predicted temperature value of the power battery is ±0.2℃.

5. The intelligent cockpit environment adaptive adjustment system based on an AI large model according to claim 1, characterized in that, The hybrid decision sub-model has a 20-step prediction time domain and a 10-step control time domain for its MPC (Model Predictive Control) function, and the cost function expression is as follows: In the formula, For cabin temperature, For users' preferred thermal temperatures, For the temperature of the power battery, The target temperature for the power battery, This refers to the compressor speed. This refers to the evaporator fan speed. =4000、 =2000、 =1000、 =800; The experience pool sample size for the TD3 algorithm is 1×102. 6 The soft update parameter is 1×10⁻³, the discount factor is 0.995, the learning rate of the Actor network is 1×10⁻³, and the learning rate of the Critic network is 1×10⁻⁻⁶. 4 .

6. The intelligent cockpit environment adaptive adjustment system based on an AI large model according to claim 1, characterized in that, The TD3 algorithm of the hybrid decision sub-model adopts a main-auxiliary reward function, and the total reward expression is as follows: The main reward function expression is: The expression for the auxiliary line reward function is: In the formula, =4000、 =2000、 =1000, PMV is the predicted average vote value. For the temperature of the power battery, =25℃, This represents the total energy consumption for air conditioning and battery cooling. This refers to the evaporator wall temperature. For cabin temperature, The user's preferred temperature.

7. The intelligent cockpit environment adaptive adjustment system based on an AI large model according to claim 1, characterized in that, The air conditioning execution unit of the execution module includes a DC compressor and an evaporator fan. The DC compressor has a displacement of 34cc, and the evaporator fan has a maximum airflow of 300m³ / h. The air conditioning execution unit integrates a dedicated AI control submodule. This control submodule receives speed commands from the AI ​​large-scale model decision module via a CAN bus, with a baud rate set to 500kbps. An airflow distribution valve is added to the connection section between the evaporator fan and the existing cabin air duct. Based on the cabin area temperature deviation output by the decision module, the airflow ratio of each air duct branch is adjusted. When the temperature deviation is greater than 1℃, the airflow of the corresponding air duct is increased by 20%. The TEC regulating unit includes five TEC devices, each with a cooling capacity ranging from 33W to 43W and a COP ranging from 0.27 to 1.

12. Each TEC device is fixed by an L-shaped aluminum alloy bracket with a thickness of 2mm. High-temperature resistant silicone pads with a temperature resistance range of -40℃ to 120℃ are adhered between the bracket and the cabin components (windshield and instrument panel). TEC1 and TEC2 are fixed to the upper part of the windshield at a vertical distance of 670mm from the user, while TEC3 and TEC4 are fixed to the upper part of the instrument panel at a horizontal distance of 600mm from the user. mm, TEC5 is fixed directly above the human body; the cross angle of the TEC jets is adjusted by a stepper motor, with gear transmission and an adjustment step of 0.5°. At the same time, the angle sensor feeds back the angle information to the decision module in real time to form a closed-loop control. The cross angle range of TEC1 and TEC2 is 0° to 19.5°, and the cross angle range of TEC3 and TEC4 is 0.2° to 21.5°. The air outlets of each TEC device are connected to the existing windshield duct, instrument panel duct and ceiling duct of the cabin through flexible air ducts. The air duct interface adopts a snap-on design, which does not require modification of the original air duct body. The battery cooling unit includes an electric water pump and a plate heat exchanger. The electric water pump has a head of 6m and a maximum flow rate of 60L / min. It also integrates a flow sensor with an accuracy of ±2%FS, which can feed the flow information back to the decision module in real time. The plate heat exchanger has a brazed structure, with plates made of 316 stainless steel, a corrugation angle of 30°, and a heat exchange area of ​​0.8m². The coolant inlet of the plate heat exchanger is connected to the outlet of the existing cooling circuit of the power battery pack via a quick-connect coupling. The outlet of the plate heat exchanger is connected to the inlet of the electric water pump, forming a series circulation path of "battery pack → plate heat exchanger → electric water pump → battery pack".

8. The intelligent cockpit environment adaptive adjustment system based on an AI large model according to claim 1, characterized in that, The expression for calculating the PMV value of the feedback module is as follows: In the formula, This refers to the human metabolic rate, with values ​​ranging from 1.0 MET to 1.5 MET. It is the partial pressure of water vapor. This is the surface area coefficient of clothing. The surface temperature of the clothing. The mean radiation temperature. The heat transfer coefficient of the human body surface; the expression for calculating the SET value is: In the formula, The airflow velocity inside the cabin. The relative humidity inside the cabin; when When the value deviates from the user's preferred temperature by ±1℃, it is fed back to the hybrid decision sub-model and included in the cost function. Increase by 10%.

9. The intelligent cockpit environment adaptive adjustment system based on an AI large model according to claim 1, characterized in that, The arrangement of the environmental parameter sensor group satisfies: The cabin air temperature sensors are located at the front, middle and rear of the cabin, and are all 1.2m above the ground. The relative humidity sensor and CO2 concentration sensor in the cabin are integrated in the central armrest, which is 0.8m above the ground; The solar radiation intensity sensor is located on the exterior of the vehicle roof and faces directly upwards. The cabin wall temperature sensors are affixed to the windshield, instrument panel, and seat back, with two sensors affixed to each component.

10. The AI-based large-scale intelligent cockpit environment adaptive adjustment system according to claim 1, characterized in that, The coolant flow regulation logic of the battery cooling actuator is as follows: The average temperature of the power battery is collected every 10 seconds. ,when At temperatures above 27°C, the coolant flow rate is set to 15 L / min. When 25℃≤ When the temperature is ≤27℃, the coolant flow rate is set to 12L / min; when When the temperature is below 25℃, the coolant flow rate is set to 8L / min.

Images