A multi-scenario adaptive thermal management control method, device, medium and product

Abstract

This application discloses a multi-scenario adaptive thermal management control method, device, medium, and product, relating to the field of electric vehicle thermal system management. The method includes: constructing a thermal system model library and obtaining dynamic response characteristics; obtaining a thermal system controller based on a hierarchical control strategy using a dynamic programming algorithm; determining personalized control parameters using an asymmetric Gaussian process regression model and a perceived temperature prediction model; integrating multi-mode switching logic to dynamically adjust personalized control parameters, resulting in a panel controller; determining thermal scenario prediction data using an online identification model and clustering thermal system sensitive features using an improved K-Means++ algorithm to generate a typical scenario feature vector library; and determining a multi-scenario adaptive thermal management control strategy using a hierarchical controller constructed based on the thermal system controller and the panel controller. This application can determine a dynamic control strategy based on real-time data, achieving thermal system adaptation under real-time conditions.

Description

Technical Field

[0001] This application relates to the field of automotive thermal system management, and in particular to a multi-scenario adaptive thermal management control method, device, medium, and product. Background Technology

[0002] Current thermal management technology for new energy electric vehicles is rapidly transforming from a traditional "single, independent, and passive" model to an "integrated, efficient, intelligent, and refined" one. The core objective is to minimize energy consumption and extend driving range while ensuring battery safety and passenger comfort. The most prominent technical feature is the application of intelligent control algorithms, which significantly optimizes system integration and control precision.

[0003] Current thermal management technologies for new energy vehicles still have several limitations. The core contradiction of traditional fixed threshold control stems from the mismatch between the static nature of the threshold setting and the dynamic nature of actual operating conditions. Essentially, it triggers temperature control actions by pre-setting a fixed temperature critical point. This rigid logic inevitably makes it difficult to balance energy efficiency and response speed. For example, lowering the threshold to pursue high response speed, while enabling rapid responses to temperature fluctuations, leads to frequent start-stop cycles of the temperature control system. This results in mechanical components frequently experiencing instantaneous mechanical stress and current surges during start-stop cycles, sacrificing the lifespan of the thermal system to some extent. Furthermore, this "black and white" triggering logic of fixed thresholds lacks the ability to predict and dynamically adjust for temperature change trends. However, current control strategies based on dynamic thresholds are still in their early stages. The mapping relationship between thermal management control parameters and strategy responses is not yet clear, and the related model construction and system adaptation are still imperfect, hindering their widespread application and performance optimization in real vehicles.

[0004] Therefore, based on the above problems, there is an urgent need to provide a multi-scenario adaptive thermal management control method that can determine dynamic control strategies based on real-time data and realize the self-adaptation of the thermal system under real-time conditions. Summary of the Invention

[0005] The purpose of this application is to provide a multi-scenario adaptive thermal management control method, device, medium and product that can determine dynamic control strategies based on real-time data and realize the self-adaptation of the thermal system under real-time conditions.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a multi-scenario adaptive thermal management control method, including: A thermal system model library was built based on Modelica; and the dynamic response characteristics of the actuators were obtained through parametric modeling; the dynamic response characteristics include: battery SOC, electric water pump speed and compressor speed; Based on the dynamic response characteristics, a thermal system controller is obtained using a dynamic programming algorithm based on a hierarchical control strategy. The bottom layer of the hierarchical control strategy uses a hybrid algorithm combining PID and fuzzy control to dynamically adjust PID parameters and regulate the actuator. The upper layer of the hierarchical control strategy uses a predictive controller based on deep reinforcement learning, taking ambient temperature, battery SOC, and occupant thermal comfort as state inputs to optimize energy consumption. A human-computer interaction feature database is constructed, and an asymmetric Gaussian process regression model combined with a body temperature prediction model is used to determine personalized control parameters. The human-computer interaction feature database includes the driver's temperature setting deviation and fan speed adjustment frequency on the air conditioning panel. The personalized control parameters include: battery SOC, electric water pump speed, and compressor speed. Based on the personalized control parameters, multi-mode switching logic is integrated into the AUTOSAR application layer software architecture to dynamically adjust the personalized control parameters, thereby obtaining a panel controller. Based on dynamic response characteristics and vehicle historical operating data, thermal system sensitive features are determined, online identification models are used to determine thermal scene prediction data, and the improved K-Means++ algorithm is used to cluster the thermal system sensitive features to generate a typical scene feature vector library. Based on a feature vector library of typical scenarios, a hierarchical controller constructed based on a thermal system controller and a panel controller is used to determine a multi-scenario adaptive thermal management control strategy. The multi-scenario adaptive thermal management control strategy is used to dynamically adjust the control threshold and actuator parameters to achieve multi-scenario adaptive thermal management.

[0007] Optionally, the bottom layer of the hierarchical control strategy uses a hybrid algorithm combining PID and fuzzy control to dynamically adjust PID parameters and regulate the actuator, specifically including: Based on the error and error change rate of the thermal system, a fuzzy control algorithm and the IF-THEN fuzzy rule base are used for logical reasoning to obtain the increment of the PID control parameters; the PID control parameters include: proportional gain, integral gain and derivative gain. The PID control parameters are updated using increments in the PID control parameters, and the actuator is adjusted accordingly.

[0008] Optionally, updating the PID control parameters using the increments of the PID control parameters and adjusting the actuator specifically includes: Using formula Update the proportional gain; Using formula Update the integral gain; Using formula Update the differential gain; in, For the updated proportional gain, The ratio gain before the update. For the proportional gain increment, For the updated points gain, This is the points gain before the update. For the integral gain increment, For the updated differential gain, The differential gain before the update. This is the differential gain increment.

[0009] Optionally, the step of determining thermal system sensitive features based on dynamic response characteristics and vehicle historical operating data, determining thermal scene prediction data using an online identification model, and clustering the thermal system sensitive features using an improved K-Means++ algorithm to generate a typical scene feature vector library specifically includes: The correlation between the dynamic response characteristics and the vehicle's historical operating data is analyzed, and dynamic response characteristics with a correlation threshold are selected as sensitive features of the thermal system. Based on historical vehicle operation data, the thermal system sensitive features are labeled and filtered sequentially to obtain preliminary prediction data; the historical vehicle operation data includes historical environmental data and historical driving data. Based on the preliminary prediction data, an initial prediction model is constructed using the Prophet time series prediction algorithm and the LightGBM machine learning algorithm; and the initial prediction model is used to determine historical hot scene prediction data. The historical thermal scene prediction data is denoised using Fourier transform to obtain denoised historical thermal scene prediction data. The initial prediction model is trained using denoised historical thermal scene prediction data to obtain an online identification model. The online identification model is used to determine real-time thermal scene prediction data based on current environmental data and driving data, and integrates historical vehicle operation data and real-time vehicle operation data. Based on the thermal scene prediction data, the improved K-Means++ algorithm is used to cluster the sensitive features of the thermal system to generate a typical scene feature vector library.

[0010] Optionally, the step of constructing an initial prediction model based on the preliminary prediction data, using the Prophet time series prediction algorithm and the LightGBM machine learning algorithm, specifically includes: The Prophet time series forecasting algorithm is used to remove time series data from the preliminary forecast data, resulting in the preliminary forecast data after removal. Based on the preliminary prediction data after elimination, an initial prediction model is constructed using the LightGBM machine learning algorithm.

[0011] Optionally, the step of denoising the historical thermal scene prediction data using Fourier transform to obtain denoised historical thermal scene prediction data specifically includes: Using formula The historical thermal scene prediction data is subjected to a positive Fourier transform to obtain a frequency domain signal; Using formula The frequency domain signal is filtered to obtain the filtered frequency domain signal; Using formula The filtered frequency domain signal is inversely transformed to obtain the denoised historical thermal scene prediction data; in, For historical hot scene prediction data, To perform weighted summation of sine or cosine signals of different frequencies, It is a frequency domain signal. This is the filtered frequency domain signal. For the filter frequency response, For the denoised historical thermal scene prediction data, For frequency.

[0012] Optionally, the step of clustering the sensitive features of the thermal system using the improved K-Means++ algorithm based on the thermal scene prediction data to generate a typical scene feature vector library specifically includes: Real-time vehicle operation data is cleaned to obtain static feature vectors; The static feature vector is standardized, and the feature weights and the number of clusters are determined. Clustering of the thermal system's sensitive features based on feature weights and the number of clusters yields multiple cluster center vectors; The multiple cluster center vectors are stored to generate a typical scenario feature vector library.

[0013] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described multi-scenario adaptive thermal management control method.

[0014] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described multi-scenario adaptive thermal management control method.

[0015] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described multi-scenario adaptive thermal management control method.

[0016] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a multi-scenario adaptive thermal management control method, device, medium, and product. A thermal system model library is built based on Modelica. Based on dynamic response characteristics and a hierarchical control strategy, a dynamic programming algorithm is used to obtain the thermal system controller. A deep reinforcement learning predictive controller provides data-driven adaptability and decision-making flexibility, while a hybrid algorithm provides stability and long-term planning capabilities for deep reinforcement learning, compensating for its short-sightedness and security deficiencies. A human-computer interaction feature database is constructed, and an asymmetric Gaussian process regression model combined with a perceived temperature prediction model is used to determine personalized control parameters. The asymmetric Gaussian process regression model can specifically characterize uncertainties in different directions, thereby improving the practicality of prediction. Based on the personalized control parameters, an AUTOSAR application layer is applied. The software architecture integrates multi-mode switching logic and dynamically adjusts personalized control parameters to obtain a panel controller. This enables the AUTOSAR application layer software architecture to not only execute a single function but also smoothly and safely switch between different functional modes based on system status and driver input. Based on dynamic response characteristics and historical vehicle operating data, thermal system sensitive features are determined. An online identification model is used to determine thermal scenario prediction data, and an improved K-Means++ algorithm is used to cluster these thermal system sensitive features, generating a typical scenario feature vector library. This allows for more accurate prediction of future loads. Based on the typical scenario feature vector library, a hierarchical controller built on the thermal system controller and panel controller is used to determine multi-scenario adaptive thermal management control strategies, achieving multi-scenario adaptive thermal management. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a multi-scenario adaptive thermal management control method according to an embodiment of this application; Figure 2 This is a schematic diagram of a hierarchical controller module in one embodiment of this application; Figure 3 This is a schematic diagram of a multi-scenario adaptive thermal management control strategy in one embodiment of this application; Figure 4 This is a schematic diagram illustrating the prediction data of various algorithms for changes in environmental wind speed in one embodiment of this application; Figure 5This is a schematic diagram of the Q-value evaluation process of the dual-delay deep deterministic strategy gradient algorithm in one embodiment of this application. Detailed Implementation

[0019] 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, and 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.

[0020] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] In one exemplary embodiment, such as Figure 1 As shown, a multi-scenario adaptive thermal management control method is provided, including the following S1 to S6. Wherein: S1: Build a thermal system model library based on Modelica; and obtain the dynamic response characteristics of the actuator through parametric modeling.

[0022] A high-precision thermal system model library has been built based on Modelica, covering key components such as battery liquid cooling circuits and air conditioning systems. Parametric modeling accurately characterizes the dynamic response characteristics of actuators, including compressor speed, expansion valve opening, and three-way valve flow distribution. Dynamic response characteristics include battery SOC, electric water pump speed, and compressor speed, encompassing the core components of the entire thermal system, as well as blower speed, radiator power, condenser power, and three-way valve flow. The core idea behind using a modeling language like Modelica to build a high-precision thermal system model library is to describe the physical laws of the system rather than the solution steps, allowing simulation tools to automatically solve the equations. Its advantage lies in its ability to seamlessly describe systems across multiple physical domains, including mechanical, electrical, thermal, hydraulic, and control systems, making it ideal for complex mechatronic products such as the new energy electric vehicles involved in this application. The high-precision thermal system model library employs object-oriented design, using `partial` to define abstract interfaces and `replaceable` to implement model variants. Furthermore, to achieve high-precision physical modeling, nonlinear factors such as friction and gaps should be incorporated during construction, and piecewise functions should be used to improve the calculation accuracy of each interval.

[0023] The physical processes of actuators are highly complex, involving multiple coupled effects such as electromagnetics, thermal, and mechanical processes. Therefore, directly studying the specific motion of atoms within them is impractical. However, by parametrically modeling the core components of the thermal system, the abstract relationship between the controller input and actuator output can be transformed into an intuitive mathematical expression. This allows us to obtain the dynamic response characteristics of the thermal system actuator and determine subsequent optimization methods based on these characteristics.

[0024] S2: Based on the dynamic response characteristics, a dynamic programming algorithm is used to obtain the thermal system controller according to the hierarchical control strategy.

[0025] The bottom layer of the hierarchical control strategy uses a hybrid algorithm combining PID and fuzzy control to dynamically adjust PID parameters and regulate the actuator. The upper layer of the hierarchical control strategy employs a predictive controller based on deep reinforcement learning, using ambient temperature, battery SOC, and occupant thermal comfort as state inputs to optimize energy consumption.

[0026] S2 specifically includes: S21: Based on the error and error change rate of the thermal system, the fuzzy control algorithm and the IF-THEN fuzzy rule base are used for logical reasoning to obtain the increment of the PID control parameters.

[0027] Before parametric modeling, only the relationship between actuator input and output is often obtained, making the physical result difficult to analyze. However, after parametric modeling, the physical entity can be abstracted into a parameter-driven, computable, and analyzable mathematical object through parametric equations. Furthermore, parametric modeling provides a solid foundation for determining the thermal system controller. When designing a thermal system controller, the designer can systematically calculate the proportional gain of the PID controller. Integral gain Differential gain Instead of blindly experimenting, we can adjust model parameters to analyze whether the performance of the control system meets the standards under different operating conditions. This upgrades the "experience-driven" trial-and-error method to a "model-driven" systematic and scientific design method, greatly improving the performance and reliability of the control algorithm.

[0028] The output of a traditional PID controller It is composed of three parts: ; in, For the controller output, For proportional gain, The error of the thermal system is defined as the difference between the setpoint and the actual value, i.e., the difference between the target temperature / setpoint temperature and the actual temperature. For integral gain, For differential gain, For a moment.

[0029] in, , , These are the three core parameters of PID control. These parameters need precise tuning and, once set, remain constant throughout the system's operation. However, for complex systems with strong nonlinearity, time-varying characteristics, and uncertain models, fixed PID parameters are unlikely to achieve good results under all operating conditions. In vehicle models, due to the nonlinearity of suspension elastic elements and damping elements, the vehicle must be analyzed as a nonlinear system. Therefore, a single set of PID parameters often cannot account for the different operating conditions of the vehicle in various scenarios.

[0030] In various scenarios, actuator parameters can change significantly. To adapt to the dynamic response characteristics of these changes and improve the response adjustment speed of the actuator, the thermal system controller in this application adopts a hybrid algorithm that combines PID and fuzzy control at its underlying level.

[0031] The hybrid algorithm in this application dynamically adjusts the three core parameters of the PID controller using fuzzy logic, thereby enabling the controller to achieve optimal performance under various operating conditions. The fuzzy PID controller constructed using the hybrid algorithm mainly consists of two parts: a standard PID controller and a fuzzy logic inferencer. Its workflow is as follows: First, the fuzzy logic inference engine receives the errors of the thermal system in real time. and error change rate .

[0032] Secondly, the precise error value and error change rate Converting to fuzzy linguistic variables essentially involves calculating the membership degree of a precise value to a certain fuzzy set (the value ranges from 0 to 1, and the closer the value is to 1, the stronger the membership degree).

[0033] ; in, For fuzzy linguistic variables, The left endpoint is the minimum value with a membership degree of 0. The midpoint is the peak point with a membership degree of 1. The right endpoint is the maximum value with a membership degree of 0. This is the error.

[0034] Furthermore, logical reasoning is performed based on an "IF-THEN" fuzzy rule base. If the error is large and still increasing in the positive direction, it indicates that the system response is too slow and the proportional gain needs to be significantly increased. To speed up the response, slightly reduce the integral gain. To avoid overshoot, differential gain The result remains unchanged. Then, the fuzzy output derived from fuzzy inference is converted back to a precise value. This precise value is the increment by which the PID parameters should be changed.

[0035] S22: Update the PID control parameters using the increment of the PID control parameters and adjust the actuator.

[0036] The process of updating PID control parameters is as follows: ; ; ; in, For the updated proportional gain, The ratio gain before the update. For the proportional gain increment, For the updated points gain, This is the points gain before the update. For the integral gain increment, For the updated differential gain, The differential gain before the update. This is the differential gain increment.

[0037] The output is calculated based on the updated PID coefficients. The output is then sent to the actuator, thereby enabling real-time, dynamic, and intelligent tuning of the PID parameters.

[0038] The upper layer of the thermal system controller employs a predictive controller based on deep reinforcement learning. The combination of the two reduces system energy consumption through superior decision-making capabilities. Deep reinforcement learning provides predictive control with data-driven adaptability and decision-making flexibility, addressing its strong dependence on models and computational bottlenecks in high-dimensional scenarios. Conversely, predictive control provides deep reinforcement learning with model-based stability guarantees and long-term planning capabilities, compensating for its short-sightedness and insufficient security.

[0039] The thermal system controller uses a hierarchical dynamic programming algorithm to decompose the complex problem of optimizing system performance coefficients and energy consumption into multiple overlapping subproblems. By solving the subproblems and storing their solutions, the original problem can be solved efficiently.

[0040] The actuator received The system then dynamically responds, producing an actuator output corresponding to the output of the thermal system controller. Therefore, the change in the PID coefficients during this process means a change in the proportional-integral coefficients, resulting in changes in both the thermal system controller output and the actuator output. Thus, under this control method, the thermal system (thermal system controller + actuator) can be regulated by altering the PID coefficients. For example, an increase in the actuator output, such as the speed of an electric water pump, can lower the actual temperature, leading to errors. As the temperature decreases, the output of the PID controller decreases accordingly, meaning the input to the actuator decreases. This results in a reduction in the speed of the electric water pump, and ultimately, the temperature gradually approaches the setpoint. The significance of fuzzy control lies in controlling the rate at which this temperature approaches the setpoint.

[0041] S3: Construct a database of human-computer interaction features and use an asymmetric Gaussian process regression model combined with a body temperature prediction model to determine personalized control parameters.

[0042] The human-machine interaction feature database contains operational data from the driver's air conditioning panel, such as temperature setting deviations and fan speed adjustment frequency. This database systematically stores and manages multi-dimensional feature data from the human-machine interaction process, providing fundamental support for algorithm optimization and scenario adaptation. Based on the driver's air conditioning panel operation data, an asymmetric Gaussian process regression model is used to analyze occupant operating habits, and combined with a body temperature prediction model, personalized control parameters are generated. These personalized control parameters include battery SOC, electric water pump speed, and compressor speed.

[0043] Dynamic response characteristics refer to how the actuator output changes in response to variations in the actuator input. The actuator input is the control command, which aims to achieve a specific value in the actuator output. For example, dynamic response characteristics describe the relationship between battery state of charge (SOC) and temperature, while personalized control parameters specify the exact value (percentage) the battery SOC should reach.

[0044] Asymmetric Gaussian process regression models are an important extension of classical Gaussian process regression. Their core lies in breaking through the traditional classical Gaussian process regression's assumption of "symmetric error distribution" by introducing asymmetric kernel functions or asymmetric likelihood functions to more accurately model the asymmetric uncertainties prevalent in real-world scenarios. In the thermal management systems of new energy vehicles, numerous direction-sensitive asymmetric features exist. For example, the side of the battery cell closer to the motor often has a higher operating temperature. Using a classical Gaussian process regression model in this case would lead to over-cooling of the cold part and insufficient heat dissipation of the hot part. For instance, the probability of a driver not adjusting the air conditioning temperature at a suitable temperature (normal temperature) is different from the probability of lowering the temperature when it is 5°C above the suitable temperature versus raising it when it is 5°C below the suitable temperature. These significant asymmetric features lead to a decrease in the accuracy of classical Gaussian process regression models and an underestimation of potential risks. Therefore, it is necessary to use asymmetric Gaussian process regression models to allow the model to specifically characterize uncertainties in different directions, thereby improving the practicality of predictions.

[0045] To further enhance the asymmetry, an asymmetric Gaussian distribution is used as the likelihood function: ; in, Let y (the regulating operation) be the probability of occurring under the operating conditions (wind speed / temperature). It is an asymmetric Gaussian density function, used to adapt to operating habits, such as whether there is a preference for high or low temperature. For the i-th operating condition, wind speed or temperature regulation. For working conditions The general values ​​for the following conditions, such as the general values ​​for ambient temperature and ambient wind speed. The variance represents the conditional fluctuations under the corresponding operating conditions. This represents the total number of operating conditions.

[0046] The probability density function of the asymmetric Gaussian distribution is: ; in, For wind speed or temperature adjustment, This refers to a typical value for the outside temperature or outside wind speed. This is the skewness coefficient. , It is asymmetrical at times. Overestimation , Emphasis on underestimation .

[0047] The probability density function adjusts the weights of high and low deviations to suit users' operating habits of preferring to adjust to high or low temperatures.

[0048] Given new input The formulas for calculating the mean and variance of the predicted distribution of the adaptive Gaussian process regression output operation are as follows: ; in, To represent the operating conditions based on the user's preference (high or low) for newly input values, This refers to the general values ​​(wind speed or temperature) for the corresponding operating conditions that are newly entered. For variance, The kernel matrix of the training data, for Kernel phasor of training data, .

[0049] The asymmetric Gaussian process regression model can flexibly control which side of the error is more sensitive by adjusting the parameters of the kernel function or likelihood function, adapting to the risk preferences of different scenarios. It also inherits the advantages of classical Gaussian process regression, retaining its good performance in small samples and the ability to quantify uncertainty. This makes it suitable for diverse scenarios with significant asymmetric errors as described in this application. The data signals transmitted from the human-computer interaction feature database are processed using the asymmetric Gaussian process regression model, and combined with a body temperature prediction model to obtain personalized control parameters for the panel controller.

[0050] S4: Based on personalized control parameters, multi-mode switching logic is integrated into the AUTOSAR application layer software architecture to dynamically adjust personalized control parameters and obtain a panel controller.

[0051] Based on the generated personalized control parameters, an application layer software architecture based on the AUTOSAR standard is built, integrating multi-mode switching logic. AUTOSAR stands for Automotive Open System Architecture. As a core standardized framework in the field of automotive electronics, its application layer is the final carrier for realizing vehicle functions, responsible for defining and executing specific automotive business functions, such as adaptive cruise control, lighting control, and battery management. Mode switching thresholds are defined, and different modes affect the opening and closing degree of the thermal management system's shut-off valves. When the state of charge (SOC) is lower than the set SOC threshold, the system enters energy-saving mode; when the ambient temperature is higher than the set maximum temperature threshold, lower than the set minimum temperature threshold, or the ambient temperature rises or falls too rapidly, the system enters rapid temperature control mode; under normal circumstances, it operates in comfort mode. The multi-mode switching logic can dynamically adjust the thresholds of personalized control parameters between energy-saving mode, comfort mode, and rapid temperature control mode, and learns implicit user needs online through a shadow mode. The multi-mode switching logic is implemented using C code compliant with the ASAM XIL standard generated by Matlab.

[0052] Integrating multi-mode switching logic into the AUTOSAR application layer software architecture means that the architecture can not only perform a single function, but also smoothly and safely switch between different functional modes based on system status and driver input. However, using a large number of if-else logic statements to handle mode switching in a large software component leads to bloated code that is difficult to maintain and test. Therefore, a mode manager is used to act as the brain of the system, responsible for making decisions on all modes. It receives input signals such as driver requests and vehicle state parameters (vehicle speed, tire steering angle, battery SOC, motor power) through an internal finite state machine. After the mode manager analyzes the input signals and outputs a decision to determine the current active mode, it broadcasts this signal to various software components of AUTOSAR, thereby realizing mode switching and dynamically adjusting control parameter thresholds.

[0053] S5: Based on the dynamic response characteristics and historical vehicle operation data, determine the thermal system sensitive features, use an online identification model to determine thermal scene prediction data, and use an improved K-Means++ algorithm to cluster the thermal system sensitive features to generate a typical scene feature vector library.

[0054] S5 specifically includes: S51: Perform correlation analysis between dynamic response characteristics and historical vehicle operation data, and select dynamic response characteristics that are greater than the correlation threshold as sensitive features of the thermal system.

[0055] Simulating the parameterized model can yield performance parameters such as energy consumption of the thermal management system under the corresponding simulation conditions, as well as actuator parameters such as electric water pump speed and cooling fan speed (these parameters are obtained from the controller signal input to the actuator and are also called control parameters). Based on environmental data analysis, the correlation between control parameters and performance parameters, as well as the correlation between the performance of the thermal management system and scenario parameters, can be used to extract the sensitive features of the thermal system.

[0056] S52: Based on the vehicle's historical environmental data and historical driving data, the thermal system sensitive features are labeled and filtered in sequence to obtain preliminary prediction data.

[0057] S53: Based on the preliminary prediction data, construct an initial prediction model using the Prophet time series prediction algorithm and the LightGBM machine learning algorithm; and use the initial prediction model to determine the historical hot scene prediction data.

[0058] S53 specifically includes: S531: Use the Prophet time series forecasting algorithm to remove time series data from the preliminary forecast data to obtain the preliminary forecast data after removal.

[0059] Prophet is used to remove purely time-dependent components from the initial forecast data, resulting in baseline forecasts dominated by trend and seasonal factors. At its core, Prophet is an additivity decomposition model that transforms a time series... It is decomposed into three main components and one error term, and the calculation formula is as follows: ; in, The trend term is used to capture the non-periodic, long-term overall growth or decline component in a time series. Seasonality is a component used to capture periodic changes (such as recurring patterns weekly or yearly). The "Holidays" category is used to capture components of sudden events that may not follow cyclical patterns and have potentially huge impacts. This is the error term, used to represent random fluctuations that Prophet cannot explain, and is usually assumed to follow a normal distribution. .

[0060] From raw data Subtract the model fitted What remains is the error term. These are also known as residuals. These residuals can be analyzed and represent changes that cannot be explained by time trends and seasonal factors. They are often fluctuations caused by accidental damage to parts or sudden weather changes that affect the vehicle.

[0061] S532: Based on the preliminary prediction data after elimination, construct an initial prediction model using the LightGBM machine learning algorithm.

[0062] Historical thermal scenario prediction data (also known as forecast data) are determined using the initial prediction model.

[0063] S54: Use Fourier transform to denoise the historical thermal scene prediction data to obtain the denoised historical thermal scene prediction data.

[0064] To better understand the specific frequencies that make up historical thermal scene prediction data, Fourier transform is used to denoise and standardize the data. The core idea of ​​Fourier transform is that any complex signal can be viewed as a superposition of countless sine waves with different frequencies and amplitudes. The time domain is the dimension we typically observe in data. The x-axis represents time, and the y-axis represents the signal amplitude. In this dimension, signal and noise are mixed together and difficult to distinguish directly. The frequency domain, however, is the dimension of the data after the Fourier transform. The x-axis represents frequency, and the y-axis represents the amplitude of that frequency component. In this dimension, signal and noise are usually distributed in different frequency regions, making them easier to separate.

[0065] S54 specifically includes: S541: Perform a positive Fourier transform on the historical thermal scene prediction data to obtain the frequency domain signal.

[0066] First, import historical thermal scene prediction data into the library. At this point, the data is a noisy signal. Perform a positive Fourier transform on the historical thermal scene prediction data to obtain a complex array, i.e., a frequency domain signal, which contains the amplitude and phase information of each frequency component. The calculation formula is as follows: ; in, Let be a frequency domain function, representing the historical thermal scene prediction data as a function of frequency. The change in frequency domain signal is also called the frequency domain signal. This is a time-domain function representing the changes in historical thermal scene prediction data over time, i.e., historical thermal scene prediction data. In Euler's formula form, Essentially, it involves weighted summation of sine or cosine signals of different frequencies.

[0067] S542: Filter the frequency domain signal to obtain the filtered frequency domain signal.

[0068] A frequency domain plot is drawn, and filtering is performed by setting a threshold. The inverse Fourier transform of the frequency domain signal yields the denoised data signal. With filter frequency response Multiply to obtain the filtered frequency domain signal. : .

[0069] S543: Perform inverse transform on the filtered frequency domain signal to obtain denoised historical thermal scene prediction data.

[0070] For the filtered frequency domain signal Performing an inverse Fourier transform yields the final time-domain output signal, which is the denoised historical thermal scene prediction data. The calculation formula is as follows: .

[0071] Since the mean and standard deviation of the data can be known through the frequency domain plot, the data can be easily standardized, thereby facilitating subsequent optimization of the initial prediction model.

[0072] S55: The initial prediction model is trained using the denoised historical thermal scene prediction data to obtain an online identification model.

[0073] The initial prediction model is trained using denoised historical thermal scene prediction data to obtain an online identification model. The online identification model is used to determine real-time thermal scene prediction data based on current environmental data and driving data.

[0074] Environmental wind speed changes were predicted using porphet, lightGBM, and a hybrid algorithm combining porphet and lightGBM, respectively, and compared with real thermal scene data. Figure 4 As shown, the horizontal axis represents the predicted wind speed, and the vertical axis represents the number of iterations. The error between the denoised historical thermal scene prediction data and the actual thermal scene data (real data), i.e., the residual, is determined. The complex relationship between the residual and abnormal external changes is learned to establish a connection between the two, thus predicting the residual based on external changes. The initial prediction model is iteratively optimized based on the residual. As the indicators (temperature fluctuation amplitude, system thermal efficiency) converge, the result is the optimal iteration round of the initial prediction model. The evaluation functions are used to query the various evaluation indicators of the prediction model validation set. Bayesian optimization is used to construct a probabilistic model based on prior results to guide the next search, maximizing tuning efficiency, resulting in the LightGBM model to be used, i.e., the online identification model.

[0075] By acquiring the thermal system's sensitive features based on real-time vehicle operating data and dynamic response characteristics, real-time thermal scenario prediction data can be obtained through an online identification model. In other words, the online identification model can integrate historical thermal scenario prediction data, real-time thermal scenario prediction data, and historical and real-time vehicle operating data.

[0076] In summary, this application classifies data using categorical features and then encapsulates it using lgb.Dataset. After parameter configuration, lgb.train is used as the interface for training to prevent overfitting and control the frequency of log output to avoid excessive information output. By inputting the thermal system's sensitive features, it is possible to predict future environmental change trends, thereby identifying and switching to appropriate thermal management control strategies before changes in the thermal environment are imminent.

[0077] Parameters of different scenarios, such as thermal systems and passenger compartments (integrating data such as altitude gradient, solar radiation intensity, and humidity range to characterize the current thermal scenario of the electric vehicle. The online identification model determines the current thermal scenario by collecting input thermal system sensitivity data) are used as prediction variables. Sensitive features of typical thermal system scenarios are used as input variables. The hyperparameters of the LightGBM machine learning algorithm, such as learning rate, number of iterations, tree depth, and feature ratio of iterations, are adjusted to continuously improve the accuracy of the algorithm. By fusing environmental prediction information, inputting thermal system sensitive features, and outputting scenario parameters, the model achieves online vehicle identification.

[0078] This application utilizes historical vehicle operation data to construct an initial prediction model and optimizes it into an online identification model to process real-time vehicle operation data, achieving a combination of model prediction and dynamic parameters, thus breaking through the traditional fixed threshold control.

[0079] S55: Based on the thermal scene prediction data, the improved K-Means++ algorithm is used to cluster the sensitive features of the thermal system to generate a typical scene feature vector library.

[0080] By integrating current thermal scenario prediction data, historical thermal scenario prediction data, historical vehicle operation data, and current vehicle operation data through an online identification model, the thermal scenario prediction data includes environmental parameters, thermal management status, air conditioning usage logs, and the performance degradation curves of the three electric components, etc. An improved K-Means++ algorithm is used to dynamically cluster the sensitive features of the thermal system to generate a typical scenario feature vector library.

[0081] Using the improved K-Means++ algorithm to dynamically cluster the sensitive features of thermal systems and generate a feature vector library for typical scenarios is essentially a process of automatically discovering and refining models or templates from large amounts of high-dimensional, dynamic runtime data. The improved K-Means++ algorithm, based on K-Means++, incorporates optimizations such as introducing weights and adaptively determining the K value specifically for the characteristics of thermal system data.

[0082] S55 specifically includes: S551: Clean the real-time vehicle operation data to obtain static feature vectors.

[0083] The real-time vehicle operation data is cleaned, and then the continuous time series data is converted into static feature vectors.

[0084] S552: Standardize the static feature vectors and determine the feature weights and the number of clusters.

[0085] All features are standardized to determine the feature weights and the optimal number of clusters.

[0086] S553: ​​Cluster the thermal system's sensitive features based on feature weights and the number of clusters to obtain multiple cluster center vectors.

[0087] Select the most suitable feature weights and the K value with the highest silhouette coefficient to perform K-Means++ clustering, dividing all preprocessed samples into K clusters.

[0088] If the temporal dispersion of the k-th cluster obtained after clustering the sensitive features of the thermal system exceeds the threshold, it is divided into two sub-clusters. When the time span within a cluster is too large, it means that the data in it do not belong to the same operating scenario, so it needs to be split. The formula is as follows: ; in, The temporal dispersion of the cluster. cluster The The time corresponding to each sample For clusters The number of contained samples, For clusters Average time, This is the time-series dispersion threshold.

[0089] When the centers of two clusters are too close together, they are merged into one. A new cluster is formed around the center, meaning that when the weighted distance between the centers of two clusters is less than a distance threshold, they are merged into one cluster: ; in, For clusters center with cluster center The weighted distance, Distance threshold For clusters The weighting coefficients correspond to the number of samples in the cluster. For clusters The weighting coefficients correspond to the number of samples in the cluster. It serves as the center of the new cluster.

[0090] Based on the dynamic equilibrium characteristics of the thermal system, the iteration termination condition is: For clusters The change in energy consumption deviation , Average energy consumption within the cluster For clusters The Energy consumption per sample The threshold for cluster center change, This is the threshold for changes in cluster energy consumption.

[0091] The iterative termination condition formula means that the change in cluster center is small enough and the change in cluster energy consumption is also small enough (meaning that the iterative result converges), at which point the iteration stops.

[0092] After the K-Means++ algorithm finishes, it generates K cluster centers. Each cluster center is itself a feature vector, which represents the average state of all samples in that cluster.

[0093] S554: Store multiple cluster center vectors to generate a typical scenario feature vector library.

[0094] These K named cluster center vectors are stored to form a typical scenario feature vector library. Using this library, the scenario in which the thermal system operates can be identified. After scenario identification and matching, the optimal control strategy for that scenario is invoked to achieve adaptive control.

[0095] This application constructs a feature vector library for typical scenarios under multiple scenarios, building a platform for more efficient and rapid scenario matching and parameter adjustment, and laying the foundation for the future development of the field of new energy vehicle battery thermal management under multiple scenarios. S6: Based on the feature vector library of typical scenarios, a hierarchical controller built on the thermal system controller and panel controller is used to determine the multi-scenario adaptive thermal management control strategy.

[0096] A schematic diagram of the hierarchical controller module is shown below. Figure 2 As shown in the diagram, the multi-scenario adaptive thermal management control strategy is illustrated below. Figure 3 As shown, the multi-scenario adaptive thermal management control strategy is used to dynamically adjust the control threshold and actuator parameters to achieve multi-scenario adaptive thermal management.

[0097] The essence of thermal system controllers is to trade space for time, significantly improving efficiency by avoiding redundant calculations. Its core implementation steps involve abstracting the physical system into a dynamic programming (DP) mathematical model with a decision-making mechanism and reward function, specifically as follows: First, determine the required action space and state space of the system, and set a reward function that includes tracking error and control energy consumption. Then, construct a policy network and a value network using a dual-delay deep deterministic policy gradient algorithm. Continuously sample and train the network from the experience replay buffer. Finally, integrate model predictive control with deep deterministic policy gradient to continuously optimize the control sequence. Specifically: Policy Network The input is the state. The output is a deterministic threshold adjustment amount. Value Network Two Critics networks are used to evaluate in Perform an action in a state The expected cumulative reward.

[0098] The process of determining the target Q value is as follows: Figure 5 As shown, the calculation formula is as follows: ; in, For the target action in the next moment, for The output of the target policy network under the given state, For the noise term of the target policy in the next time step, , Let be the noise standard deviation of the target strategy.

[0099] ; in, For the minimum range of motion, For maximum range of motion, This is a shearing operation, which truncates the action within a valid range. In other words, the shearing operation restricts the target action at the next moment to this minimum to maximum range.

[0100] When updating, the Critic network uses the next action chosen by the current policy to calculate the target Q-value. If the Critic network is overly optimistic in its evaluation of the value of certain state-action pairs, and this bias accumulates with training iterations, it will eventually lead the Actor network to learn a suboptimal policy. This is a manifestation of relying on erroneous high Q-value guidance. The Dual-Delay Deep Deterministic Policy Gradient (DDP) algorithm, however, uses two independent Critic networks to evaluate Q-values ​​and takes the minimum as the target Q-value, suppressing overestimation. When calculating the target Q-value, the DDP algorithm adds a small amount of noise to the next action, enabling the Critic to learn a more robust value function and avoiding oversensitivity to a single action. Through the four-layer core network of the DDP algorithm, explicit division of labor is optimized to achieve threshold adaptation and dynamically adjust the threshold relaxation coefficient.

[0101] To optimize the dynamic allocation of weight coefficients, a multi-objective reward design was adopted in the algorithm. A hierarchical reward function was introduced, defining comfort reward, economic reward, and safety reward. The hierarchical reward function was optimized by Bayesian optimization using a probabilistic surrogate model and a data collection function. A probabilistic model was fitted with existing historical data to approximate the unknown objective function and quantify the prediction uncertainty of each location. Based on the prediction results of the surrogate model, the sampling value of each historical data point was calculated, and the point with the highest value was selected as the location for the next evaluation.

[0102] Multi-objective reward function express Under the weighted combination, the overall performance of the system in terms of comfort, economy, and safety follows a Gaussian process distribution: ; Among them, in this formula This represents the weighted combination of comfort, economy, and safety for the decision variables, for example... The weighting coefficients are 0.2, 0.3, and 0.5. The weighting coefficients are 0.3, 0.3, and 0.4. It follows a Gaussian distribution. Let be the prior mean of the Gaussian process. This is the kernel function for a Gaussian process.

[0103] Based on the historical dataset D, for any new candidate point objective function value The posterior distribution is still a Gaussian distribution, and the formula is: ; in, To adopt the new weight allocation combination The predicted mean at that time To adopt the new weight allocation combination The prediction variance represents the uncertainty of the prediction.

[0104] Acquisition function comprehensive Predict the mean and variance, and calculate candidate points. The sampling value.

[0105] The sampling value is determined by the expected improvement of the reward. The formula is as follows: ; As expected.

[0106] Current best historical reward value Expanded as: ; in, To adopt the new weight allocation combination The standard deviation of the forecast at that time , Let be the cumulative distribution function of the standard normal distribution. It is the probability density function of the standard normal distribution.

[0107] Based on the acquisition function, the optimal decision variables for the next evaluation are found by optimizing the acquisition function. This means the optimal combination of decision weight coefficients for the next evaluation. The one that maximizes the expected reward improvement across all possible domains. The value of is calculated using the following formula: ; in, For weighted coefficient combination All possible domains.

[0108] The algorithm is further optimized by feasible region constraints. Using the feature vector library of typical scenarios built above, the feasible region boundary of the control parameters is constructed by the support vector data description algorithm. The SAC output parameters are mapped to a high-dimensional feature space to eliminate abnormal strategies that exceed the thermodynamic stability boundary.

[0109] During simulation, extreme disturbances are rehearsed in advance using digital twins to assess the inadequacy of existing thresholds. After online optimization, the results are deployed to the physical system, preventing the physical system from directly bearing the risk. After injecting a disturbance, the digital twin must be used to evaluate whether the existing thresholds can handle the disturbance. The core is to analyze the timeliness of threshold triggering and the effectiveness of control actions. Simultaneously, the adjustment of dynamic thresholds must emphasize both scientific optimization and safety, forming a virtuous cycle of update-verification-optimization.

[0110] When the reward function converges, it means that during continuous iterations, the cumulative reward obtained by the algorithm gradually stabilizes around a fixed value and no longer fluctuates significantly with iterations. This indicates that the deep reinforcement learning model has found the optimal policy or is close to the optimal policy. The final output includes a dynamic threshold adjustment rule base and a feasible region feature set of control parameters, enabling adaptive control strategies for the thermal system in multiple scenarios.

[0111] This application designs a hierarchical reward function and a dynamic threshold coordination mechanism for the control strategy to dynamically adjust the threshold relaxation coefficient and the weight coefficient. Based on a typical scenario feature vector library, the SAC output parameters are mapped, outlier strategies are eliminated, and the strategy threshold parameters are updated online to achieve a multi-scenario adaptive thermal system control strategy. Compared with traditional fuel cell thermal management technologies, this application has adaptability to multiple scenarios, can meet the user's needs in different thermal scenarios, and significantly improves the user experience and energy efficiency ratio.

[0112] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores multi-scenario adaptive thermal management control data. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a multi-scenario adaptive thermal management control method.

[0113] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0114] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0115] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0116] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0117] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0118] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0119] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0120] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A multi-scenario adaptive thermal management control method, characterized in that, The multi-scenario adaptive thermal management control method includes: A thermal system model library was built based on Modelica; and the dynamic response characteristics of the actuators were obtained through parametric modeling; the dynamic response characteristics include: battery SOC, electric water pump speed and compressor speed; Based on the dynamic response characteristics, a thermal system controller is obtained using a dynamic programming algorithm based on a hierarchical control strategy. The bottom layer of the hierarchical control strategy uses a hybrid algorithm combining PID and fuzzy control to dynamically adjust PID parameters and regulate the actuator. The upper layer of the hierarchical control strategy uses a predictive controller based on deep reinforcement learning, taking ambient temperature, battery SOC, and occupant thermal comfort as state inputs to optimize energy consumption. A human-computer interaction feature database is constructed, and an asymmetric Gaussian process regression model combined with a body temperature prediction model is used to determine personalized control parameters. The human-computer interaction feature database includes the driver's temperature setting deviation and fan speed adjustment frequency on the air conditioning panel. The personalized control parameters include: battery SOC, electric water pump speed, and compressor speed. Based on the personalized control parameters, multi-mode switching logic is integrated into the AUTOSAR application layer software architecture to dynamically adjust the personalized control parameters, thereby obtaining a panel controller. Based on dynamic response characteristics and vehicle historical operating data, thermal system sensitive features are determined, online identification models are used to determine thermal scene prediction data, and the improved K-Means++ algorithm is used to cluster the thermal system sensitive features to generate a typical scene feature vector library. Based on a feature vector library of typical scenarios, a hierarchical controller constructed based on a thermal system controller and a panel controller is used to determine a multi-scenario adaptive thermal management control strategy. The multi-scenario adaptive thermal management control strategy is used to dynamically adjust the control threshold and actuator parameters to achieve multi-scenario adaptive thermal management.

2. The multi-scenario adaptive thermal management control method of claim 1, wherein, The hierarchical control strategy employs a hybrid algorithm combining PID and fuzzy control to dynamically adjust PID parameters and regulate the actuator at its lowest level. Specifically, this includes: Based on the error and error change rate of the thermal system, a fuzzy control algorithm and the IF-THEN fuzzy rule base are used for logical reasoning to obtain the increment of the PID control parameters; the PID control parameters include: proportional gain, integral gain and derivative gain. The PID control parameters are updated using increments in the PID control parameters, and the actuator is adjusted accordingly.

3. The multi-scenario adaptive thermal management control method of claim 2, wherein, The step of updating the PID control parameters using the increments of the PID control parameters and adjusting the actuator specifically includes: Using the formula The proportional gain is updated; Using the formula updating the integral gain; Using the formula updating the differential gain; wherein, Kp is the updated proportional gain, Kp is the updated proportional gain, Kp is the proportional gain increment, Ki is the updated integral gain, Ki is the updated integral gain, Ki is the integral gain increment, Kd is the updated derivative gain, Kd is the updated derivative gain, Kd is the derivative gain increment.

4. The multi-scenario adaptive thermal management control method according to claim 1, characterized in that, The process involves determining thermal system sensitivity features based on dynamic response characteristics and historical vehicle operating data, using an online identification model to determine thermal scene prediction data, and then using an improved K-Means++ algorithm to cluster the thermal system sensitivity features to generate a typical scene feature vector library. Specifically, this includes: The correlation between the dynamic response characteristics and the vehicle's historical operating data is analyzed, and dynamic response characteristics with a correlation threshold are selected as sensitive features of the thermal system. Based on historical vehicle operation data, the thermal system sensitive features are labeled and filtered sequentially to obtain preliminary prediction data; the historical vehicle operation data includes historical environmental data and historical driving data. Based on the preliminary prediction data, an initial prediction model is constructed using the Prophet time series prediction algorithm and the LightGBM machine learning algorithm; and the initial prediction model is used to determine historical hot scene prediction data. The historical thermal scene prediction data is denoised using Fourier transform to obtain denoised historical thermal scene prediction data. The initial prediction model is trained using denoised historical thermal scene prediction data to obtain an online identification model. The online identification model is used to determine real-time thermal scene prediction data based on current environmental data and driving data, and integrates historical vehicle operation data and real-time vehicle operation data. Based on the predicted thermal scene data, the improved K-Means++ algorithm is used to cluster the sensitive features of the thermal system to generate a typical scene feature vector library.

5. The multi-scenario adaptive thermal management control method according to claim 4, characterized in that, The step of constructing an initial prediction model based on the preliminary prediction data, using the Prophet time series prediction algorithm and the LightGBM machine learning algorithm, specifically includes: The Prophet time series forecasting algorithm is used to remove time series data from the preliminary forecast data, resulting in the preliminary forecast data after removal. Based on the preliminary prediction data after elimination, an initial prediction model is constructed using the LightGBM machine learning algorithm.

6. The multi-scenario adaptive thermal management control method according to claim 4, characterized in that, The step of denoising the historical thermal scene prediction data using Fourier transform to obtain denoised historical thermal scene prediction data specifically includes: Using formula The historical thermal scene prediction data is subjected to a positive Fourier transform to obtain a frequency domain signal; Using formula The frequency domain signal is filtered to obtain the filtered frequency domain signal; Using formula The filtered frequency domain signal is inversely transformed to obtain the denoised historical thermal scene prediction data; in, For historical hot scene prediction data, To perform weighted summation of sine or cosine signals of different frequencies, For frequency domain signals, This is the filtered frequency domain signal. For the filter frequency response, For the denoised historical thermal scene prediction data, For frequency.

7. The multi-scenario adaptive thermal management control method according to claim 4, characterized in that, The step of clustering the sensitive features of the thermal system using the improved K-Means++ algorithm based on the thermal scene prediction data to generate a typical scene feature vector library specifically includes: Real-time vehicle operation data is cleaned to obtain static feature vectors; The static feature vector is standardized, and the feature weights and the number of clusters are determined. Clustering of the thermal system's sensitive features based on feature weights and the number of clusters yields multiple cluster center vectors; The multiple cluster center vectors are stored to generate a typical scenario feature vector library.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the multi-scenario adaptive thermal management control method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-scenario adaptive thermal management control method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-scenario adaptive thermal management control method according to any one of claims 1-7.

Images