Automobile thermal management intelligent control method and domain controller

Abstract

The invention provides an intelligent control method for automobile thermal management and a domain controller, belongs to the field of automobile thermal management, and is used for solving the problems of high energy consumption and insufficient cooperative control capability of a thermal management system in related technologies. According to the method, hierarchical intelligent cooperation is realized through three control stages: in the first control stage, a whole vehicle power distribution strategy is solved based on a multi-objective optimization algorithm; in the second control stage, the heat management loop is cooperatively adjusted based on model prediction control and neural network prediction; and in the third control stage, adaptive tuning is carried out on the execution component based on particle swarm optimization and reinforcement learning. In each stage, information interaction and overall optimization are realized through key parameter transmission. A corresponding domain controller comprises a processing unit for executing an algorithm, a signal acquisition module, a driving output module and a vehicle network communication module. The global optimization of energy consumption, safety and comfort of the thermal management system is realized, and the endurance mileage and environmental adaptability of the whole vehicle are improved.

Description

Technical Field

[0001] This application relates to the field of automotive thermal management, and more particularly to an intelligent control method and domain controller for automotive thermal management. Background Technology

[0002] With the rapid development of the automotive industry, the vehicle thermal management system, as a key component in ensuring vehicle safety, range, and driving comfort, has received widespread attention regarding its energy consumption and intelligence level. Currently, replacing traditional distributed control units with integrated domain controllers and introducing advanced algorithms to improve energy efficiency has become an important development direction for the industry.

[0003] In existing technologies, some solutions attempt to optimize thermal management through hardware integration or the application of multiple algorithms. For example, some technologies achieve multi-mode management through the integration of eight-way valves, but their core chips and architecture rely on foreign technologies; other solutions achieve heat exchange between the three electrical systems, but the control accuracy and intelligence level need to be improved; still other solutions focus on intelligent control algorithms, but lack an independent and controllable hardware platform to support them.

[0004] However, existing technologies still have shortcomings. They either rely on non-autonomous hardware or lack systematic coordination at the algorithm level, failing to form a complete, adaptive closed-loop control system that ranges from global energy planning for the entire vehicle to dynamic coordination of multiple heat source circuits and fine-tuning of actuators. This makes it difficult to achieve a globally optimal balance between energy consumption, safety, and comfort when facing complex and variable operating conditions. Summary of the Invention

[0005] This application provides an intelligent control method and domain controller for automotive thermal management, which can solve the problems of high energy consumption and insufficient collaborative control capability of existing thermal management systems through a hierarchical, integrated, and collaboratively optimized intelligent control system.

[0006] Firstly, this application provides an intelligent control method for automotive thermal management. The method includes: a first control stage, based on a multi-objective optimization algorithm, using vehicle energy consumption, battery state of charge, and battery temperature as optimization objectives, to solve for a power allocation strategy between the range extender and the power battery, and to determine the operating mode of the vehicle thermal management system; a second control stage, based on the operating mode and the power allocation strategy, dynamically adjusting the battery thermal management loop and the air conditioning thermal management loop using a model predictive control algorithm, wherein the constraints of the model predictive control algorithm incorporate the passenger compartment thermal demand level predicted based on a neural network; a third control stage, responding to at least one environmental disturbance parameter among ambient temperature, solar radiation intensity, and altitude, solving for the control parameter combination of each energy-consuming component in the thermal management system using a particle swarm optimization algorithm, and dynamically adjusting the operating parameters of the particle swarm optimization algorithm based on a reinforcement learning algorithm; wherein the power allocation strategy output by the first control stage provides input constraints or objective function parameters for the model predictive control algorithm in the second control stage, and the cabin temperature control objective output by the second control stage provides the basis for the fitness function of the particle swarm optimization algorithm in the third control stage.

[0007] By adopting the above technical solution, this application organically integrates and cascades three-layer algorithms—multi-objective optimization, model predictive control and neural network prediction, and particle swarm optimization and reinforcement learning—according to the logic of "macro-power planning - meso-loop coordination - micro-component tuning." The output of the upper-level algorithm serves as the input or constraint of the lower-level algorithm, thereby forming synergy in terms of time scale and control granularity. This achieves global, dynamic, and adaptive optimization control of the vehicle's thermal management system, effectively reducing the overall energy consumption of the system.

[0008] Furthermore, the first control stage includes: establishing a multi-objective optimization problem with the optimization objectives of reducing fuel consumption of the range extender, maintaining the battery state of charge within a preset high-efficiency range, and constraining the battery temperature within a safe operating window; using a multi-objective optimization algorithm to iteratively solve the multi-objective optimization problem, and outputting the power allocation sequence of the range extender and the power battery.

[0009] By adopting the above technical solutions, the optimization goals of vehicle energy management become more specific and comprehensive, taking into account economy (fuel consumption), battery life (SOC maintenance), and safety (temperature constraints), thus laying a reasonable energy distribution foundation for subsequent thermal management coordinated control.

[0010] Furthermore, the objective function of the multi-objective optimization problem includes a first sub-objective function for evaluating fuel consumption, a second sub-objective function for evaluating the deviation of the battery state of charge from the desired range, and a third sub-objective function for evaluating the battery temperature exceeding the safe operating range. By adopting the above technical solution, the specific composition of each sub-objective in the multi-objective optimization is clarified, making the mathematical description of the optimization problem clearer and ensuring that the algorithm solution process directly serves the key performance indicators of vehicle operation.

[0011] Furthermore, the second control phase includes: establishing a thermodynamically coupled state-space model characterizing the dynamic changes in battery temperature and cabin temperature; in each control cycle, based on the thermodynamically coupled state-space model, the current system state, and the predicted cabin thermal demand level in the time domain, solving a rolling optimization problem with constraints, the rolling optimization problem aiming to reduce battery temperature tracking error and actuator control cost; and outputting control commands for adjusting at least one actuator in the battery thermal management loop and the air conditioning thermal management loop according to the solution results.

[0012] By adopting the above technical solution and utilizing the forward-looking optimization capabilities of model predictive control, the air conditioning and battery cooling circuits are coordinated while meeting the battery temperature safety control requirements. The system also responds in advance to changes in the thermal demand of the passenger compartment, achieving dynamic and efficient collaboration among multiple heat sources.

[0013] Furthermore, the constraint conditions are integrated based on the passenger compartment thermal demand level predicted by the neural network, including: inputting the ambient temperature, current passenger compartment temperature, vehicle speed and historical thermal demand data into the trained neural network model to predict the passenger compartment thermal demand level in the future time domain; when the predicted passenger compartment thermal demand level is higher than a preset threshold, the constraint condition on the passenger compartment temperature rise rate is activated in the rolling optimization problem.

[0014] By adopting the above technical solution, intelligent prediction and optimized control are closely integrated. The neural network accurately predicts the thermal demand of the passenger cabin and directly translates it into constraints for the controller, enabling the system to proactively adjust its strategy in advance to meet comfort requirements such as rapid warming, thereby improving the intelligence and comfort of the control.

[0015] Furthermore, in the third control stage, the combination of control parameters for each energy-consuming component in the thermal management system is solved based on the particle swarm optimization algorithm. This includes: using the control parameters of at least one thermal management system execution component that affects the cabin temperature and system energy consumption as the particle position vector; using the weighted sum of the time it takes for the cabin to reach the target temperature and the total system energy consumption as the fitness function; and obtaining a set of control parameter combinations for the energy-consuming components through iterative search of the particle swarm optimization algorithm.

[0016] By adopting the above technical solution, under the given upper-level instructions, a set of locally optimal operating points that can balance the heating rate and energy consumption can be quickly found for the actuators such as fans, PTC, and water valves, thus realizing refined energy efficiency management at the execution layer.

[0017] Furthermore, the step of dynamically adjusting the operating parameters of the particle swarm optimization algorithm based on reinforcement learning includes: using the environmental perturbation parameters as the state of the reinforcement learning agent; using the action of adjusting the inertia weight or learning factor of the particle swarm optimization algorithm as the action of the agent; constructing a reward function based on the degree of reduction in total system energy consumption relative to the benchmark value and the degree of deviation of battery temperature from the target value; and outputting the adjustment amount of the operating parameters of the particle swarm optimization algorithm through the interactive learning between the agent and the environment.

[0018] By adopting the above technical solutions, the system is endowed with the ability to adapt to complex environmental disturbances. Reinforcement learning, through online learning, dynamically optimizes the parameters of the PSO algorithm itself, enabling PSO to converge to the optimal solution adapted to the new environment more quickly and accurately, thereby improving the robustness and environmental adaptability of the entire control system.

[0019] Secondly, this application provides a domain controller for executing the intelligent control method for automotive thermal management described in the first aspect. The domain controller includes a hardware platform comprising: a processing unit configured to execute algorithms for the first control stage, the second control stage, and the third control stage; a signal acquisition module configured to acquire analog signals from a temperature sensor, a pressure sensor, and a humidity sensor; a drive output module configured to output digital control signals for driving a water pump, valves, a heater, and a compressor; and a vehicle network communication module configured to interact with a vehicle controller, a battery management system, and an air conditioning controller.

[0020] By adopting the above technical solution, a dedicated hardware platform is provided capable of supporting and efficiently running the aforementioned complex hierarchical fusion algorithm. This controller integrates rich signal acquisition, drive output, and communication interfaces, enabling centralized and precise control of all key sensors and actuators in the thermal management system. It forms the physical basis for the physical realization of the aforementioned intelligent control method.

[0021] Furthermore, the vehicle network communication module includes at least two Controller Area Network (CAN) bus interfaces, at least one of which supports a specified frame wake-up function under low power consumption. By adopting the above technical solution, stable and efficient data exchange between the controller and other vehicle systems is ensured. At the same time, the low-power wake-up function meets the stringent requirements of automobiles for the static current of domain controllers, which helps to reduce the overall vehicle energy consumption.

[0022] Furthermore, the signal acquisition module includes multiple analog signal acquisition channels, at least one of which is configured to connect to a high-precision battery temperature sensor. By adopting the above technical solution, the accuracy and reliability of key status information (such as battery temperature) acquisition are ensured, providing accurate data input for upper-level algorithms, especially model predictive control and battery temperature safety monitoring, which is the foundation for ensuring the control accuracy of the entire system.

[0023] In summary, this application has at least the following beneficial effects:

[0024] A hierarchical intelligent collaborative automotive thermal management solution is provided. Through the fusion of multiple algorithms, optimization from global to local levels is achieved, significantly reducing system energy consumption and improving vehicle range. By predicting thermal demand through neural networks and incorporating model predictive control constraints, the comfort and foresight of temperature control are improved. By dynamically optimizing particle swarm optimization parameters through reinforcement learning, the system's adaptability and robustness in the face of environmental disturbances are enhanced.

[0025] It should be understood that the description in the Summary Section is not intended to limit the key or essential features of the embodiments of this application, nor is it intended to restrict the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0026] The above and other features, advantages, and aspects of the embodiments of this application will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0027] Figure 1 The schematic diagram of the domain controller in this application is shown.

[0028] Figure 2 A flowchart of an intelligent control method for automotive thermal management according to an embodiment of this application is shown. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0031] This application provides an intelligent control method and domain controller for automotive thermal management. Through the hierarchical integration and synergy of multi-objective optimization, model prediction and neural networks, particle swarm optimization and reinforcement learning, it achieves adaptive closed-loop control from global energy planning of the whole vehicle to fine-tuning of the execution components, thereby significantly reducing the overall energy consumption of the system, increasing the driving range and enhancing environmental adaptability.

[0032] In a first aspect, embodiments of this application disclose a domain controller.

[0033] Figure 1 The schematic diagram of the domain controller in this application is shown.

[0034] Reference Figure 1 The domain controller in this embodiment is an electronic hardware platform specifically designed to execute the aforementioned intelligent control method for automotive thermal management. This controller employs domestically produced components and circuit designs that meet automotive-grade reliability and safety standards. Physically, it integrates all the functional modules necessary for complex algorithm calculations, multi-channel signal acquisition, high-power drive control, and high-speed vehicle network communication. These modules are all housed within a metal casing with electromagnetic shielding and heat dissipation structures, forming a complete, independently operating control unit. The controller is integrated into the vehicle's electrical control center and connects to the vehicle's wiring harness via its standard mounting interfaces and connectors.

[0035] The core of this domain controller is a processing unit, which serves as the computational and control hub of the entire hardware platform. This processing unit can be a domestically produced multi-core microcontroller or system-on-a-chip that meets the ISO 26262 ASIL-D functional safety requirements. For example, the processing unit's clock frequency can be set between 200MHz and 500MHz, preferably 300MHz, and it integrates 2MB to 8MB of Flash memory, preferably 4MB, and at least 512KB of RAM. This configuration provides sufficient computing performance and storage resources to reliably run the entire algorithm flow of the first, second, and third control stages of the intelligent control method in real time. These algorithm flows can be deployed in layers based on the AUTOSAR software architecture, where complex multi-objective genetic algorithms, model predictive control algorithms, and neural network inference algorithms run on the high-performance core of the processing unit, while real-time tasks such as signal processing and drive control run on dedicated cores. The processing unit, through its internal high-speed parallel bus or serial peripheral interface, closely exchanges data and transmits instructions with other functional modules of the hardware platform, forming the physical basis for the entire control logic.

[0036] To accurately sense the real-time status of the thermal management system, the controller includes a signal acquisition module. This module contains multiple analog signal acquisition channels, defined by its function of receiving and converting voltage or current signals from various analog sensors. In one specific embodiment, the signal acquisition module provides a total of 38 analog signal acquisition channels. These channels are designed to accommodate sensors of different ranges and types. For example, eight channels are configured to support active sensor inputs with a voltage range of 0 to 12V, which can be directly connected to devices such as pressure transmitters; the other 30 channels are configured to support passive sensor inputs with a voltage range of 0 to 5V, for connecting temperature sensors such as thermistors. All channels are connected to one or more high-precision analog-to-digital converters (ADCs), with conversion accuracy of 10-bit, 12-bit, or 16-bit, preferably 12-bit, to achieve accurate quantification of key parameters such as coolant temperature, battery pack temperature, ambient temperature, refrigerant pressure, and cabin humidity. Specifically, at least one analog signal acquisition channel is specially optimized and configured, for example, by employing a higher-precision 16-bit ADC, a sophisticated reference voltage source, and filtering circuitry, to connect to a high-precision battery temperature sensor with extremely high measurement accuracy requirements. This sensor's measurement error is required to be less than ±0.5 degrees Celsius to ensure the most accurate battery cell thermal state feedback is provided to the upper-level control algorithm, which is crucial for ensuring battery temperature safety and control accuracy. After converting the acquired analog signals into digital signals, the signal acquisition module transmits them to the processing unit in real time via the internal SPI bus.

[0037] To execute control decisions, the controller is equipped with a drive output module. This module's function is to convert the digital control commands calculated by the processing unit into high-power electrical signals capable of directly driving high-power actuators via power switching devices. Its output is a series of digital control signals, specifically including 14 independent digital output channels. For example, 10 of these are designed as low-side drive channels, each with a peak drive current capability of 1A, used to drive loads such as relays and low-power solenoid valves; the other 4 are designed as high-side drive channels, each with a peak drive current capability of 500mA, suitable for loads requiring direct connection to the positive terminal of a power supply. In addition, the drive output module also integrates 5 H-bridge drive circuits for bidirectional control of DC motors, such as regulating damper actuators, with each H-bridge having a continuous drive current capability of no less than 500mA. These signals, amplified and controlled by internal power metal-oxide-semiconductor field-effect transistors, reliably control the operating status of various actuators in the thermal management system. Examples include adjusting the cooling water pump speed using PWM, controlling the opening and closing of electronic or solenoid valves, adjusting the power of the PTC heater, and managing the operation of the air conditioning compressor. The drive output module receives instructions from the processing unit through an isolated digital I / O interface, serving as a bridge between the intelligent algorithm and the physical actuators, ensuring that optimized instructions are executed accurately and promptly.

[0038] To achieve coordination with other vehicle control systems, the controller integrates a vehicle network communication module. This module acts as a gateway for interaction between the controller and the vehicle network, enabling high-speed and reliable data transmission and reception according to standard vehicle network protocols. Specifically, this module includes at least two independent Controller Area Network (CAN) bus physical layer interfaces with the controller. One CAN bus interface, for example conforming to the CAN FD protocol, serves as the primary data channel, with a communication rate configurable to 500kbps or 1Mbps. It is used for real-time data exchange with key nodes such as the vehicle controller, battery management system, and air conditioning controller, transmitting information such as vehicle speed, battery SOC, and system commands. The other CAN bus interface is specifically designed to support specified frame wake-up functionality in low-power mode, with a static operating current as low as 70 microamps. This means that when the vehicle is in sleep mode, the controller's main processing unit and most circuitry can be powered off to save energy, while specific circuitry on this communication interface can continuously monitor specific wake-up messages on the bus, such as the remote air conditioning start command frame with ID 0x123. Once a valid wake-up frame is detected, the internal wake-up circuit generates an interrupt signal, triggering the power management circuit to power on the entire domain controller, thereby completing the system startup within hundreds of milliseconds. This meets the stringent requirements of automobiles for static power consumption and the user's need for a fast response experience.

[0039] In summary, the domain controller in this embodiment highly integrates a high-performance processing unit, a high-precision multi-channel signal acquisition module, a high-drive output module, and a high-speed dual-channel communication module with low-power wake-up capability into a single hardware platform. These components are interconnected and work collaboratively through a precise circuit board layout, internal bus, and unified power supply and clock system. The processing unit coordinates and schedules operations, the signal acquisition module provides sensory input, the vehicle network communication module acquires collaborative information, and after algorithmic decision-making, the drive output module executes actions. Together, they constitute a physical entity capable of supporting and efficiently executing complex hierarchical fusion intelligent control algorithms, thereby achieving centralized, precise, and adaptive global optimization control of the vehicle's thermal management system.

[0040] Secondly, embodiments of this application disclose an intelligent control method for automotive thermal management.

[0041] Figure 2 A flowchart of an intelligent control method for automotive thermal management according to an embodiment of this application is shown.

[0042] Reference Figure 2 This method is executed periodically on the thermal management domain controller. It aims to achieve global optimization control of the range extender, power battery, air conditioning system and cabin heating and cooling components through the orderly coordination of multi-level intelligent algorithms, and ultimately achieve the comprehensive goal of reducing vehicle energy consumption, ensuring battery safety and improving driving comfort.

[0043] The method consists of three progressive and information-interactive control stages in terms of time scale and optimization granularity. The first control stage (macro-energy planning stage) is responsible for solving the optimal power allocation strategy between the range extender and the power battery within a relatively long time window, based on a multi-objective optimization algorithm, and deciding whether the vehicle's thermal management system should operate in pure electric mode or range-extended mode. The second control stage (meso-loop coordination stage) dynamically coordinates the battery thermal management loop and the air conditioning thermal management loop within a shorter control cycle, based on the output of the first stage, using a model predictive control algorithm. The constraints of its control law incorporate intelligent predictions of future thermal demand in the passenger compartment. The third control stage (micro-component tuning stage) responds to real-time environmental disturbances, using a particle swarm optimization algorithm to quickly solve for the optimal operating point of each actuator, and introduces a reinforcement learning algorithm to make the particle swarm optimization process adaptive. These three stages are not isolated. The power allocation strategy output by the first control stage provides key input constraints or objective function parameters for the model predictive control algorithm of the second control stage, such as limiting the upper limit of auxiliary power that can be used for air conditioning or battery heating. The expected cabin temperature change trajectory calculated by the second control stage serves as an important part of the fitness function in the particle swarm optimization algorithm of the third control stage, and is used to balance the heating rate and energy consumption.

[0044] The implementation of the first control phase includes two core steps: establishing a problem model and solving it. First, a multi-objective optimization problem is established. The optimization variable for this problem is the range extender output power at each discrete time step within a future planning cycle. and power battery output power Its objective function It is a vector function designed to simultaneously minimize three sub-objectives: the first sub-objective function... Used to assess total fuel consumption, where The fuel consumption rate function is obtained by interpolation using the engine universal characteristic map, with parameters... Derived from optimization variables, The planning period length; the second sub-objective function This is used to penalize the battery's state of charge for deviating from the high-efficiency reference range, where Based on the battery model The integral calculation yields the result. The preset value (e.g., 0.5); the third sub-objective function This is used to penalize battery temperatures exceeding the safe operating window, where Based on a simplified battery thermal model, and considering the current temperature, And the ambient temperature is predicted. and A preset safety boundary (e.g., 45°C and -5°C) is set. The constraints that this optimization problem must satisfy include: power balance constraints. ,in The drive power required by the vehicle controller, For the estimated base power of the auxiliary system; battery SOC boundary constraints. ; and range extender power upper and lower limit constraints. Subsequently, a multi-objective genetic algorithm is used to iteratively solve this problem. The algorithm initializes a population consisting of multiple power allocation sequences, evolves it through selection, crossover, and mutation operations, and maintains the diversity and convergence of the population using fast non-dominated sorting and crowding calculation. Finally, from the obtained Pareto optimal solution set, a compromise solution is selected according to a specific strategy (such as fuzzy decision-making or preset weights), and the output is the power allocation sequence of the range extender and the power battery. And based on this, determine the working mode (if in the sequence) If the value is always 0, it is in pure electric mode; otherwise, it is in range-extended mode.

[0045] The core of the second control stage is the fusion of model predictive control and neural network prediction. First, a discretized thermodynamically coupled state-space model is established to describe the temperature dynamics of the battery and the crew compartment. This model can be expressed as:

[0046]

[0047] Wherein, the state vector , and These are the average battery pack temperature and the average passenger compartment temperature at time k, respectively, obtained directly from sensors. Control input vector. , This is the electronic expansion valve opening command (0% to 100%). Power command for PTC heater. Disturbance vector. Including ambient temperature and solar radiation intensity Data is collected by vehicle sensors. System matrix. Input matrix and perturbation matrix These models, derived through system identification or simplification based on physical laws, encapsulate the heat capacity and thermal resistance of the battery and passenger compartment, as well as the possible heat exchange relationships between them. Based on this, rolling optimization is performed within each control cycle (e.g., 1 second). The objective function of the optimization problem is:

[0048]

[0049] in, To predict the time domain, To control the time domain, and It is a positive definite weight matrix. The target battery temperature is set at 25°C. This optimization problem is constrained by state-space model dynamics, control input amplitude constraints, and key coupling constraints. These coupling constraints incorporate the crew cabin thermal demand levels predicted by the neural network. Specifically, a pre-trained multilayer perceptron neural network was used to predict the future. The heat demand level of the step. The input layer of this network includes the current ambient temperature. Current cabin temperature Current vehicle speed In addition to historical heat demand data over a period of time, the output layer is heat demand level. Its value is an integer from 1 to 10, with higher levels indicating more urgent heating needs. When predicting the heating demand level for a future time period... If the temperature consistently exceeds a preset threshold (e.g., 7), then a constraint on the rate of temperature rise in the occupant cabin is activated in the corresponding time-domain rolling optimization problem, for example, by adding... ,in A positive minimum heating rate requirement is required (e.g., 3°C / min). Additionally, if the operating mode is range-extended, the optimization problem will also incorporate the available waste heat power constraint derived from the first-stage power allocation strategy. Solving this constrained quadratic programming problem yields the optimal control input sequence, and its first element is used as the actual control command output to drive actuators such as the electronic expansion valve and PTC.

[0050] The third control stage aims to find the component-level parameter combination that minimizes energy consumption under given upper-level instructions, while also possessing environmental adaptability. First, local optimization is performed based on the particle swarm optimization algorithm. The control parameters of the execution component to be optimized (such as the PTC operating voltage) are then considered. Cooling fan PWM duty cycle Water valve opening ) combined into a particle position vector Define a fitness function. To evaluate the merits of this parameter combination:

[0051]

[0052] in, The estimated time (in seconds) required for the cabin temperature to rise from the current value to the target value under this parameter combination is calculated based on a simplified cabin thermal balance model. The total energy consumption (watt-seconds) is estimated, mainly including PTC energy consumption and fan energy consumption. and As a weighting coefficient, its relative magnitude reflects the trade-off between heating rate and energy consumption, and The value of can be related to the urgency of the cabin temperature control target output in the second stage. The PSO algorithm initializes a swarm of particles, and each particle, based on its individual best historical position and the swarm's best historical position, follows the formula... and Update its velocity and position, where the inertial weights Individual learning factors Social learning factors These are the key parameters of the algorithm. Through iterative search, the final combination of control parameters that minimizes the fitness function is obtained.

[0053] To enable the PSO algorithm to adapt to different environmental disturbances (such as high temperature and humidity, high altitude and low air pressure), this embodiment introduces reinforcement learning to dynamically adjust its operating parameters. A reinforcement learning agent is defined, whose state... For example, the normalized environmental disturbance vector. ,in , , These are the normalized values ​​for the current ambient temperature, humidity, and altitude, respectively. (Action) Defined as parameters of the PSO algorithm , , The adjustment amount, for example The action space can be discrete (e.g., increasing, decreasing, remaining constant) or continuous. Reward function. The design is directly geared towards the overall system objective:

[0054]

[0055] in, This represents the actual energy consumption of the thermal management system during the current control cycle. Energy consumption under a certain baseline strategy; Current battery temperature; As a reward weight; The indicator function incurs a large penalty when the battery temperature exceeds the limit. The agent employs algorithms such as Q-learning or Actor-Critic to continuously explore the environment (trying different actions). Adjusting PSO parameters, observing state transitions, and obtaining rewards. To learn an optimal strategy This strategy can output the most suitable PSO parameters to adjust the action according to the current environmental state, so that the PSO algorithm can always maintain high search performance in a variable environment and output the optimal combination of component control parameters.

[0056] In summary, this embodiment constructs a complete intelligent control system for automotive thermal management through the three clearly defined and tightly coupled control stages described above. The first stage, macroscopic energy planning, sets the tone for both economy and safety for the entire system; the second stage, predictive and coordinated control, achieves dynamic and precise matching among multiple heat sources; and the third stage, adaptive sub-optimization, ensures the optimal response of the execution layer to disturbances. Through the transmission of key parameters such as power budget, temperature target, and weighting coefficients, each stage forms an organic whole, ultimately achieving the core inventive objective of global energy efficiency optimization.

[0057] Based on the stable operation of the three-layer intelligent control system and the generation of continuously optimized data flow, this section introduces an endogenous cognitive evolution framework. This framework is not an external module independent of the aforementioned control stages, but is deeply rooted in its data flow and decision-making logic. By constructing a unified, computable, and interventionist system cognitive model, and on this basis, achieving closed-loop self-examination, reasoning, refinement, and co-evolution, it endows the entire thermal management system with the fundamental ability to understand, interpret, and transcend established rules.

[0058] The framework begins with the synchronous perception and structured representation of the system's overall operational state. This is derived from the power allocation sequence in the first control phase. Operating mode instructions, derived from the model predictive control instruction sequence of the second control stage. Its internal coupled state-space model parameters, derived from the component optimization parameter combination of the third control stage. and environmental disturbance vector Along with the original sensor timing data These data, together, constitute a multi-source heterogeneous input stream. This data is fed in real-time into a dynamic causal graph structure learning engine. The core task of this engine is to construct and continuously update a graph model representing the system's internal operating mechanism. Among them, the node set It includes two types of nodes: entity nodes (such as...) , , ) directly corresponds to physical quantities or actuators; abstract nodes (such as , Edge sets are generated by aggregating underlying entity nodes through an attention mechanism, representing system-level performance and state metrics. Each directed edge in Indicates from node To the node The potential causal influence, with two key attributes: influence strength weight. and influence function These attributes are learned through a differentiable causal discovery process. Specifically, for each node... We assume its observed values From its parent node set The value is determined by an additive noise model: ,in Therefore Small neural networks with parameters, As independent noise. By optimizing a score-based matching objective function over the entire graph structure and utilizing sparsity regularization, the system can simultaneously learn the graph structure (i.e., which...). (non-zero) and influence function The dynamic cause-effect graph generated in this process It is a computable cognitive model of the system, which encapsulates a complex, time-varying causal network between the behavior of micro-components and the performance of the macro-system.

[0059] Based on this dynamic causal graph The system executes three deeply integrated cognitive processes in parallel: interpretive analysis, counterfactual reasoning, and policy causal correction. Interpretive analysis is used to understand the effects of given control decisions. For example, it analyzes the system's energy consumption after a certain control cycle. When the value is abnormally high, the system will follow... middle finger Tracing back along the edges, we calculate the parent node (e.g., ...) The contribution of the variable to the current efficiency decline is determined by its influence path. This contribution can be quantified by calculating the expected change in the efficiency node when all other variables remain constant and only this variable is restored to its typical value; this is essentially calculating an attribution quantity. Counterfactual reasoning is used to explore the possibility of a better decision. The system automatically constructs and answers questions such as, "If the compressor's maximum speed limit were reduced in the past five minutes..." Simultaneously, specific sequence compensation is applied to the PTC power, and the current average battery temperature... Total energy consumption The question, "How will it change?", corresponds mathematically to the concept of cognitive models. Execution Calculate and perform probability inference: ,in These are all other facts actually observed. This reasoning process is entirely completed within the model, requiring no interaction with the real environment, achieving zero-cost policy exploration. The policy causal correction process then directly optimizes the policy generation in the third control stage using the above analysis results. Specifically, the reward function of the reinforcement learning agent... Enhanced to ,in It is a causal consistency reward. This reward is based on the actions taken by the agent. With cognitive models The degree of agreement between the revealed efficient causal models is used for calculation. For example, if Clearly indicate the specific environmental state Given that the causal effect of "water valve opening degree" on "cabin temperature rise rate" is far greater than that of "PTC power," if the intelligent agent takes a corresponding action and prioritizes adjusting the water valve, it will obtain a positive [result]. This guides the agent not only to learn "which actions have high rewards," but also "why this action is effective in this context," thereby improving the interpretability and generalization ability of its strategies.

[0060] With cognitive models With the continuous enrichment of knowledge and the ongoing optimization of strategies, the system faces the challenge of efficiently deploying increasingly complex cognitive capabilities on resource-constrained domain controllers. To address this, the system employs a task-aware model compilation and knowledge condensation mechanism. When a specific control task needs to be addressed (e.g., "when the ambient temperature suddenly drops"), the system will... Within the next ten minutes, the cabin temperature was raised to the highest possible level. And the battery temperature is not lower than The neural compiler begins its work when it generates a minimalist and efficient control strategy. The compiler uses the task description... and hardware constraints As input, for the complete cognitive model Perform functional trimming. It begins with... Running and Tasks Related counterfactual queries and impact analysis identified the smallest subset of nodes that are critical to achieving the mission objectives. and edge subset Then, the compiler will... and Local causal structure and its related influence function Compile into a lightweight, task-specific... Decision function network ,in Only includes with The relevant observation states. This compilation process may involve approximating complex nonlinear relationships as piecewise linear functions, or converting probabilistic inference into deterministic table lookup operations. The final generated... It is a constraint that is satisfied Under the premise of accurately capturing the core causal logic required to complete the task, the efficient strategy module can be directly loaded and replaced in the original control loop to maximize efficiency.

[0061] The evolution of individual systems ultimately merges into a decentralized, collective cognitive network. In this network, each vehicle node periodically learns from its cognitive model. Extract causal relationship fragments that have been verified with high confidence. And related strategy improvement fragments These fragments are transmitted via vehicle-to-everything (V2X) broadcasts or roadside units in encrypted and abstract form (e.g., only logical expressions of causal relationships and direction vectors for policy updates are transmitted, not the original data or complete model parameters). Upon receiving them, other nodes first perform security verification within their local cognitive models, i.e., within their own... The system verifies whether the causal relationship can be replicated in the given environment. Once verified, the causal fragment is integrated into the local model, rapidly gaining new knowledge about rare operating conditions or system aging patterns. For policy improvement fragments, nodes employ an integrated gradient fusion method, organically combining them with local policies rather than directly replacing them. The entire network forms a distributed cognitive system that continuously discovers, verifies, absorbs, and reinvents, with its collective intelligence evolving at a network effect rate, enabling the entire fleet's energy efficiency management capabilities to continuously and autonomously advance towards the Pareto frontier.

[0062] Through the deep integration of the aforementioned cognitive evolution framework, the original three-layer control architecture has been endowed with complete cognitive capabilities, including self-modeling, counterfactual thinking, causal explanation, on-demand compilation, and group collaboration. The system is no longer merely a controller executing preset optimization algorithms, but has evolved into a cognitive agent capable of understanding the consequences of its own actions, exploring unknown optimal solutions, and sharing wisdom with other intelligent agents. This ensures that the thermal management system of new energy vehicles can maintain excellent, robust, and interpretable performance even when facing unprecedentedly complex scenarios, hardware degradation, or individual differences, representing a crucial step towards general artificial intelligence in intelligent control.

[0063] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0064] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A smart control method for automotive thermal management, characterized in that, The method includes: In the first control stage, based on a multi-objective optimization algorithm, with the vehicle energy consumption, battery state of charge and battery temperature as optimization objectives, the power distribution strategy between the range extender and the power battery is solved, and the working mode of the vehicle thermal management system is determined. In the second control phase, based on the working mode and the power distribution strategy, the battery thermal management loop and the air conditioning thermal management loop are coupled and dynamically adjusted according to the model predictive control algorithm. The constraints of the model predictive control algorithm incorporate the passenger cabin thermal demand level predicted based on the neural network. In the third control stage, in response to at least one environmental disturbance parameter among ambient temperature, solar radiation intensity and altitude, the control parameter combination of each energy-consuming component in the thermal management system is solved based on the particle swarm optimization algorithm, and the operating parameters of the particle swarm optimization algorithm are dynamically adjusted based on the reinforcement learning algorithm. The power allocation strategy output by the first control phase provides input constraints or objective function parameters for the model predictive control algorithm of the second control phase, and the cabin temperature control objective output by the second control phase provides the basis for the fitness function of the particle swarm optimization algorithm of the third control phase.

2. The intelligent control method for automotive thermal management according to claim 1, characterized in that, The first control phase includes: Establish a multi-objective optimization problem with the optimization objectives of reducing range extender fuel consumption, maintaining battery state of charge within a preset high-efficiency range, and constraining battery temperature within a safe operating window; A multi-objective optimization algorithm is used to iteratively solve the multi-objective optimization problem, and the power allocation sequence of the range extender and the power battery is output.

3. The intelligent control method for automotive thermal management according to claim 2, characterized in that, The objective function of the multi-objective optimization problem includes a first sub-objective function for evaluating fuel consumption, a second sub-objective function for evaluating the deviation of the battery state of charge from the desired range, and a third sub-objective function for evaluating the battery temperature exceeding the safe operating range.

4. The intelligent control method for automotive thermal management according to claim 1, characterized in that, The second control phase includes: Establish a thermodynamically coupled state-space model to characterize the dynamic changes in battery temperature and crew cabin temperature; In each control cycle, based on the thermodynamic coupled state-space model, the current system state, and the predicted crew cabin thermal demand level in the time domain, a rolling optimization problem with constraints is solved. The rolling optimization problem aims to reduce battery temperature tracking error and actuator control cost. Based on the solution results, output control commands for adjusting at least one actuator in the battery thermal management circuit and the air conditioning thermal management circuit.

5. The intelligent control method for automotive thermal management according to claim 4, characterized in that, The constraints are fused based on the crew cabin thermal demand levels predicted by neural networks, including: By inputting ambient temperature, current passenger cabin temperature, vehicle speed, and historical thermal demand data into a trained neural network model, the future passenger cabin thermal demand level can be predicted. When the predicted thermal demand level of the crew cabin is higher than a preset threshold, a constraint on the rate of temperature rise of the crew cabin is activated in the rolling optimization problem.

6. The intelligent control method for automotive thermal management according to claim 1, characterized in that, The third control stage involves solving for the combination of control parameters for each energy-consuming component in the thermal management system using a particle swarm optimization algorithm, including: The control parameters of at least one thermal management system actuator that affects the temperature of the crew cabin and the energy consumption of the system are used as the particle position vector. The fitness function is the weighted sum of the time it takes for the crew cabin to reach the target temperature and the total system energy consumption. A set of control parameters for the energy-consuming component is obtained through iterative search using the particle swarm optimization algorithm.

7. The intelligent control method for automotive thermal management according to claim 6, characterized in that, The dynamic adjustment of the operating parameters of the particle swarm optimization algorithm based on reinforcement learning includes: The environmental disturbance parameters are used as the state of the reinforcement learning agent; The action of adjusting the inertia weights or learning factors of the particle swarm optimization algorithm is taken as the action of the agent. The reward function is constructed by combining the degree of reduction in total system energy consumption relative to the baseline value and the degree of deviation of battery temperature from the target value. Through the interaction and learning between the agent and the environment, the agent outputs the adjustment amount for the running parameters of the particle swarm optimization algorithm.

8. A domain controller for executing the intelligent control method for automotive thermal management according to any one of claims 1 to 7, characterized in that, Includes a hardware platform, the hardware platform comprising: A processing unit configured to execute the algorithms of the first control phase, the second control phase, and the third control phase; A signal acquisition module, configured to acquire analog signals from a temperature sensor, a pressure sensor, and a humidity sensor; A drive output module, configured to output digital control signals for driving water pumps, valves, heaters and compressors; The vehicle network communication module is configured to interact with the vehicle controller, battery management system and air conditioning controller.

9. The domain controller according to claim 8, characterized in that, The vehicle network communication module includes at least two controller area network (MAN) bus interfaces, of which at least one MAN bus interface supports the specified frame wake-up function under low power consumption.

10. The domain controller according to claim 8, characterized in that, The signal acquisition module includes multiple analog signal acquisition channels, at least one of which is configured to connect to a high-precision battery temperature sensor.

Images