A thermal management method for a vehicle and the vehicle

Abstract

This application discloses a vehicle thermal management method and a vehicle. The method includes: acquiring the vehicle's comprehensive state characteristics and route characteristic sequence; inputting the comprehensive state characteristics into an operation phase identification model for identification to obtain the vehicle's operation phase category; inputting the comprehensive state characteristics, operation phase category, and route characteristic sequence into an operation parameter prediction model for prediction to obtain the vehicle's future operation parameters; inputting several candidate thermal management parameters and the future operation parameters into a thermal management benefit prediction model for prediction to obtain the comprehensive benefit value corresponding to each candidate thermal management parameter; and inputting the comprehensive state characteristics, operation phase category, future operation parameters, and the comprehensive benefit value corresponding to each candidate thermal management parameter into a thermal management parameter inference model for inference to obtain the vehicle's target thermal management parameters and perform thermal management on the vehicle accordingly. This method avoids solving complex optimization problems online, ensuring the real-time performance and effectiveness of vehicle thermal management.

Description

Technical Field

[0001] This application belongs to the field of vehicle thermal management, specifically relating to a vehicle thermal management method and a vehicle. Background Technology

[0002] With the popularization of new energy vehicles and the development of fast charging networks, battery thermal management has become a key system affecting range, charging capacity, power, and lifespan. Some mass-produced models already have a navigation-triggered battery pre-treatment function, which can adjust the battery temperature in advance to optimize vehicle driving performance and fast charging performance at charging stations.

[0003] However, existing solutions struggle to achieve a dynamic trade-off between energy consumption, fast charging efficiency, and power / regeneration availability under uncertain operating conditions. There is a lack of a unified evaluation standard between energy input and output. Furthermore, existing MPC (Model Predictive Control)-based methods require online solution of thermal management control variables, which involves large computational loads and high dependence on model accuracy. This makes it difficult to deploy stably on vehicle platforms, affecting the real-time performance and effectiveness of vehicle thermal management. Summary of the Invention

[0004] The purpose of this application is to provide a thermal management method and a vehicle, which is achieved as follows: In a first aspect, embodiments of this application provide a thermal management method for a vehicle, the method comprising: Acquire the comprehensive state characteristics of the vehicle and the sequence of route characteristics ahead of the vehicle within a preset time period; The comprehensive state features are input into the trained operation phase identification model for identification to obtain the operation phase category of the vehicle; The comprehensive state features, the operation stage category, and the route feature sequence are input into a trained operation parameter prediction model to predict the future operation parameters of the vehicle. Several candidate thermal management parameters are obtained, and each candidate thermal management parameter and the future operating parameters are input into the thermal management benefit prediction model for prediction to obtain the comprehensive benefit value corresponding to each candidate thermal management parameter. The comprehensive state characteristics, the operational stage category, the future operational parameters, and the comprehensive benefit value corresponding to each of the candidate thermal management parameters are input into the trained thermal management parameter inference model for inference to obtain the target thermal management parameters of the vehicle. Thermal management is performed on the vehicle according to the target thermal management parameters.

[0005] Secondly, embodiments of this application provide a vehicle including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method described above.

[0006] The embodiments of this application have the following advantages: This application acquires the comprehensive state characteristics of the vehicle within a preset time period and the route feature sequence ahead of the vehicle. The comprehensive state characteristics are input into a trained operation phase recognition model for identification, resulting in the vehicle's operation phase category. This allows for the determination of the vehicle's actual operation phase, providing reference information for subsequent determination of thermal management parameters. This application also inputs the comprehensive state characteristics, operation phase category, and route feature sequence into a trained operation parameter prediction model for prediction, obtaining the vehicle's future operation parameters. This enables the prediction of operation parameters during future driving and charging processes, providing sufficient predictive information for determining thermal management parameters. Furthermore, this application acquires several candidate thermal management parameters and inputs each candidate thermal management parameter, along with future operation parameters, into a thermal management benefit prediction model for prediction, obtaining the comprehensive benefit value corresponding to each candidate thermal management parameter. This allows for a unified quantitative evaluation of the benefits corresponding to thermal management parameters. Finally, this application inputs the comprehensive state characteristics, operation phase category, future operation parameters, and the comprehensive benefit value corresponding to each candidate thermal management parameter into a trained thermal management parameter inference model for inference, obtaining the vehicle's target thermal management parameters. Thermal management of the vehicle is then performed based on these target thermal management parameters. It can avoid solving complex optimization problems online, can be stably deployed on the vehicle platform, and ensures the real-time performance and effectiveness of vehicle thermal management. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0008] Figure 1 This is a flowchart of the steps of a vehicle thermal management method provided in an embodiment of this application. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been presented in the various embodiments of this application to enable readers to better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and updates based on the following embodiments. The division of the various embodiments below is for the convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.

[0010] With the continuous increase in the penetration rate of new energy vehicles and the accelerated popularization of high-power DC fast charging networks, the battery thermal management system has expanded from traditional safety temperature control to a key system affecting range energy consumption, charging efficiency, power output capability, braking energy recovery capability, and battery life.

[0011] In low-temperature environments, high-load driving, or continuous journeys involving fast charging and subsequent driving, the battery thermal state exhibits significant time lag and stage-coupling characteristics. The thermal state and state of charge during the driving phase directly determine the charging power platform and required charging time that the vehicle can achieve during the fast charging phase upon arrival at a charging station. The high thermal load generated during the fast charging phase, in turn, affects the vehicle's available drive power after charging, the probability of thermal protection functions being triggered, and energy consumption performance in subsequent driving phases. Related research and reviews indicate that battery thermal management requires a multi-objective trade-off between temperature safety constraints, energy consumption, and vehicle performance, and necessitates coordinated optimization for different operating phases.

[0012] To improve the user experience of fast charging, some mass-produced vehicles are equipped with navigation-triggered battery pretreatment or battery temperature control functions. Some mass-produced vehicles allow users to set a fast charging station as their navigation destination to trigger high-voltage battery pretreatment, resulting in better charging conditions. Other mass-produced vehicles activate battery temperature control when a user sets a fast charging station as their destination or stop in the in-vehicle navigation system, establishing a suitable battery temperature for charging before the estimated arrival time. Still other mass-produced vehicles integrate their charging planner into the navigation and application, enabling them to control battery pre-temperature adjustment before parking at a charging station to optimize charging performance and reduce travel time.

[0013] However, the pre-processing functions in the aforementioned mass-produced vehicles are primarily implemented in engineering by combining single-scenario triggering with empirical thresholds or fixed target ranges. They typically focus on the single objective of achieving the correct temperature for fast charging upon arrival, making it difficult to cover more complex end-to-end collaborative needs such as queuing scenarios, continued driving after fast charging, and pre-cooling scenarios before high-load road sections. When faced with external uncertainties such as fluctuations in estimated arrival time, uncertainties in queuing time, differences in actual charging station power, and changes in driving style, insufficient or excessive pre-processing can easily occur, leading to energy loss and fluctuations in user experience.

[0014] Existing battery thermal management control and pretreatment strategies mainly fall into three categories: rule-based methods, model-based predictive control or optimization methods, and data-driven machine learning methods.

[0015] Rule-based methods execute preheating, precooling, or temperature maintenance strategies by presetting battery temperature thresholds, state of charge thresholds, or event triggering conditions. These methods are simple to implement, offer strong real-time performance, are easy to mass-produce, and are highly compatible with existing vehicle control architectures. However, this approach relies on human experience and calibrated thresholds, making it difficult to achieve multi-objective adaptive trade-offs under complex operating conditions. In scenarios where driving and fast charging are continuously coupled, rule-based strategies struggle to simultaneously consider driving energy consumption, fast charging temperature hit at the destination, thermal management energy consumption during charging, and driving performance after charging. This can easily lead to issues such as power drop during charging or peak energy consumption and frequent actuator switching caused by emergency strong cooling or heating.

[0016] Model predictive control (MRC) or optimization methods utilize thermal system models and future disturbance predictions to solve constrained optimization problems within each control cycle, minimizing energy consumption or temperature deviations while satisfying safety constraints. These methods have a solid theoretical foundation for handling constrained continuous control problems and can be extended to address comprehensive costs such as energy consumption and lifetime. However, when considering multi-actuator coupling in the thermal management system, switching between various operating modes, and cross-stage objective changes where the objective functions differ between the driving and charging stages, the complexity of online solutions increases significantly. Furthermore, they are highly dependent on model accuracy and the accuracy of operational condition predictions, posing challenges to real-time performance, robustness, and engineering maintainability for mass production deployment. Existing research has begun to incorporate the driving and fast-charging stages into a unified predictive optimization framework. For example, MRC can simultaneously minimize thermal management energy consumption and estimate charging time while handling state of charge, power, and temperature constraints. Other studies have employed dynamic programming combined with route and charging information to integrate and optimize the driving and fast-charging stop stages, demonstrating the potential of full-stage synergy in reducing energy consumption and improving performance through experimental verification.

[0017] Data-driven machine learning methods utilize historical data to establish mapping relationships between temperature, thermal load, and control variables, or construct predictive models to assist control, and even employ deep reinforcement learning to directly learn control strategies, thereby reducing energy consumption while meeting temperature requirements. Related reviews indicate that data-driven methods have potential in nonlinear modeling, generalization to complex operating conditions, and predictive capabilities, and can be combined with technologies such as digital twins and model reduction. However, in automotive mass production scenarios, these methods still face challenges such as insufficient interpretability, difficulty in proving safety boundaries, difficulties in cross-vehicle and cross-regional migration, and insufficient policy stability and diagnosability under uncertain operating conditions, making it difficult to directly replace existing deterministic control chains.

[0018] In summary, existing rule-based methods, model predictive control or optimization methods, and data-driven methods each have their advantages in engineering implementation, constraint control, and predictive modeling. Mass-produced vehicles already have basic functions such as navigation-triggered battery preprocessing, but overall, the following shortcomings still exist.

[0019] First, the design goal of existing mass-produced pre-treatment functions focuses on achieving the optimal charging temperature upon arrival at the station, rather than optimizing system-level benefits for the entire journey. When a fast-charging station is set as a navigation destination or stop along the way, the system activates battery temperature regulation to ensure a suitable charging temperature upon arrival. This mechanism is an event-triggered pre-treatment, lacking coordinated consideration of the queuing waiting phase, the continued driving phase after fast charging, the pre-cooling phase before high-load road sections, and the unlocking phase of the recycling function under low-temperature conditions.

[0020] Second, existing strategies make decisions based on fixed thresholds, empirical rules, or single-stage objectives, lacking a unified benefit assessment model that integrates estimated arrival time of navigation, road gradient, real-time traffic conditions, user charging intentions, and battery status.

[0021] Third, existing model-based optimization methods rely on high-precision thermal models, load prediction models, and actuator models. When considering multi-mode switching, multi-actuator coupling, and multi-objective constraints, the complexity of online solution increases significantly.

[0022] Therefore, this application provides a vehicle thermal management method and a vehicle, which acquires comprehensive state characteristics and route characteristic sequences, obtains the operating stage category through an operating stage identification model, obtains future operating parameters through an operating parameter prediction model, inputs each candidate thermal management parameter and the future operating parameters into a thermal management benefit prediction model to obtain the corresponding comprehensive benefit value, and inputs the above information into a thermal management parameter inference model to obtain the target thermal management parameters and execute them. This method avoids solving complex optimization problems online, can be stably deployed on an on-vehicle platform, and ensures the real-time performance and effectiveness of vehicle thermal management.

[0023] Reference Figure 1 The diagram illustrates a flowchart of the steps of a vehicle thermal management method according to an embodiment of this application.

[0024] This method may specifically include the following steps: Step 101: Obtain the comprehensive state characteristics of the vehicle and the route characteristic sequence in front of the vehicle within a preset time period.

[0025] In this embodiment of the application, the comprehensive state characteristics of the vehicle and the route feature sequence ahead of the vehicle within a preset time period can be obtained. The comprehensive state characteristics can refer to feature data encompassing four dimensions: vehicle operating state, battery state, navigation and recharging events, and environmental conditions.

[0026] The preset time refers to a continuous time window ending at the current moment. The length of this time window can be set according to the actual application scenario.

[0027] In practical implementation, the comprehensive state feature is a feature vector composed of multiple sub-features, which can be represented as follows:

[0028] in, Indicates the overall state characteristics. Indicates the characteristics of the operating status. Indicates battery characteristics, Indicates the characteristics of navigation power replenishment events. Indicates environmental characteristics.

[0029] Operating status characteristics It can be represented in the following form:

[0030] in, It's the vehicle speed. It is acceleration. It is the cumulative mileage. It is the power recovery. It is the drive power. It refers to the road slope.

[0031] Battery characteristics It can be represented in the following form:

[0032] in, The current state of charge, For battery temperature, This is the current coolant temperature. Battery current, and These are the upper limits for charging power and discharging power, respectively.

[0033] Navigation replenishment event characteristics It can be represented in the following form:

[0034] in, Distance from charging station For the estimated arrival time, For charging pile power, For the estimated waiting time, Encode user intent.

[0035] Environmental characteristics It can be represented in the following form:

[0036] in, It is the ambient temperature. It is ambient humidity. Encode the weather status.

[0037] In this embodiment of the application, the route feature sequence is a sequence composed of road attribute features extracted from each road segment after dividing the future path ahead of the vehicle into multiple road segments at fixed distance intervals. It records the static attribute information of the road that the vehicle will travel on in the future at each location.

[0038] In practice, route feature sequences can be obtained from in-vehicle navigation systems or map data.

[0039] The future path ahead of the vehicle is planned by the navigation system based on the destination set by the user or the current driving direction. The vehicle controller reads the path information from the navigation system and divides the path into multiple segments at preset distance intervals, such as every 100 meters or every 500 meters. Then, for each segment, the corresponding road attribute features are extracted from the map data, including the slope, speed limit, road type and traffic congestion status of the segment. These features are arranged in order according to the driving direction to obtain the route feature sequence.

[0040] In practice, the collected data can be preprocessed.

[0041] Data preprocessing can be performed as follows: outlier removal, interpolation to complete missing data, and normalization of all features. Outliers refer to data points that significantly deviate from the normal range; these are identified and removed using statistical methods. Missing data refers to data points that were not recorded during the acquisition process due to sensor malfunctions or communication interruptions; these are filled in using valid data from adjacent time points using interpolation methods. Normalization is used to transform feature data with different dimensions to a uniform order of magnitude, eliminating the impact of dimensional differences on data processing.

[0042] Step 102: Input the comprehensive state features into the trained operation phase recognition model for recognition to obtain the vehicle's operation phase category.

[0043] In this embodiment of the application, the acquired comprehensive state features can be input into a trained operation phase identification model. The operation phase identification model processes and analyzes the comprehensive state features and outputs the probability that the vehicle belongs to each preset operation phase. The operation phase with the highest probability is taken as the vehicle's operation phase category.

[0044] The operation phase category is used to characterize which operation mode the vehicle belongs to at the current moment. Different operation phases correspond to different thermal management requirements and control objectives, providing a phase basis for the formulation of subsequent thermal management strategies.

[0045] In practice, the operation phases include normal driving phase, high load phase, recovery sensitive phase, pre-charging preparation phase, charging waiting phase, fast charging phase, and charging before driving phase.

[0046] These phases cover the entire journey of a vehicle from driving to charging and then back to driving, with different thermal management objectives and strategy preferences at each stage. For example, in the pre-charging preparation phase, the thermal management objective is to establish suitable battery temperature conditions for the upcoming fast charging. During the fast charging phase, the thermal management objective is to maintain a high charging power while ensuring safety. In the post-charging driving phase, the thermal management objective is to reduce thermal energy consumption while ensuring power availability. By identifying the current operating phase of the vehicle, subsequent thermal management decisions can be made more targeted and adaptive.

[0047] Step 103: Input the comprehensive state features, operation stage category and route feature sequence into the trained operation parameter prediction model to predict the future operation parameters of the vehicle.

[0048] In this embodiment of the application, the acquired comprehensive state features and route feature sequences, as well as the identified operating stage categories, can be input into a trained operating parameter prediction model. This operating parameter prediction model can then predict the key operating parameters of the vehicle during future driving and charging processes, and output the future operating parameters of the vehicle.

[0049] Future operating parameters are used to characterize the operating state that a vehicle may reach in the future, providing predictive information for subsequent thermal management decisions.

[0050] In its implementation, the operational parameter prediction model makes forward predictions from two dimensions: distance axis and charge state axis, based on vehicle state information provided by the current comprehensive state characteristics, future road static attribute information provided by the route feature sequence, and stage information provided by the operational stage category.

[0051] The distance axis refers to the prediction dimension based on the spatial position along the vehicle's travel path. Along the distance axis, the future path ahead of the vehicle is divided into multiple segments at fixed distance intervals, each segment corresponding to a spatial position. The distance axis prediction sub-model in the operational parameter prediction model predicts the speed, driving power, and regenerative braking power of the vehicle at each segment. The distance axis reflects load changes at different spatial positions during future travel, combining the prediction results with the actual geographical information of the road ahead. In the distance axis dimension, the distance axis prediction sub-model in the operational parameter prediction model can predict the speed, driving power, and regenerative braking power corresponding to each position of the vehicle during travel.

[0052] The state of charge (POC) axis refers to the prediction dimension based on the battery's state of charge (SOC). Along the POC axis, the charging process is divided into multiple SOC intervals at fixed intervals. Each interval corresponds to a charge stage during the charging process. The POC axis prediction sub-model in the operational parameter prediction model makes predictions for each SOC interval separately, obtaining the predicted charging power at the end of each SOC interval and the cumulative charging time from the start of charging to the end of that interval. The POC axis reflects the trend of charging power changes at different charge stages during the charging process, combining the prediction results with the actual changes in the battery's SOC. Along the POC axis, the POC axis prediction sub-model in the operational parameter prediction model can predict the predicted charging power and cumulative charging time for each SOC interval during the charging process.

[0053] Meanwhile, the thermal state prediction sub-model in the operating parameter prediction model can combine the above prediction results with the thermal management actuator state in the comprehensive state characteristics to further predict the battery temperature at each location during future driving. By using the predicted vehicle speed, predicted drive power, predicted regenerative power, predicted charging power, cumulative charging time, and future battery temperature together as future operating parameters, the future operating parameters encompass forward-looking information across three dimensions: driving load, charging power, and battery temperature, providing a comprehensive predictive basis for subsequent thermal management decisions.

[0054] Step 104: Obtain several candidate thermal management parameters, input each candidate thermal management parameter and future operating parameters into the thermal management benefit prediction model for prediction, and obtain the comprehensive benefit value corresponding to each candidate thermal management parameter.

[0055] In this embodiment of the application, multiple candidate thermal management parameters can be obtained. Each candidate thermal management parameter and the obtained future operating parameters are input into a trained thermal management benefit prediction model. The thermal management benefit prediction model evaluates the effect of each candidate thermal management parameter under future operating conditions and outputs the comprehensive benefit value corresponding to each candidate thermal management parameter.

[0056] Candidate thermal management parameters are selectable thermal management control schemes, each containing a specific set of control commands. Future operating parameters provide predictive information for future driving and charging processes, enabling benefit assessments to consider the actual performance of candidate schemes under future operating conditions.

[0057] The comprehensive benefit value is used to quantitatively evaluate the overall performance of candidate thermal management parameters across multiple target dimensions, providing an evaluation basis for subsequent selection of thermal management parameters.

[0058] In its implementation, the thermal management benefit prediction model incorporates predicted charging power, predicted recovery power, and predicted driving power from future operating parameters into the calculations of fast charging time benefit, recovery benefit, and power benefit, respectively. Simultaneously, it includes heating power, cooling power, and pump flow rate from candidate thermal management parameters into the calculation of thermal management energy consumption cost. Finally, a weighted combination yields the comprehensive benefit value. A higher comprehensive benefit value indicates better performance of the candidate thermal management parameter under multi-objective comprehensive evaluation. By calculating the corresponding comprehensive benefit value for each candidate thermal management parameter, the merits of multiple candidate schemes can be quantitatively compared in a unified manner, providing a basis for subsequent scheme selection.

[0059] Step 105: Input the comprehensive state characteristics, operation stage category, future operation parameters, and comprehensive benefit values ​​corresponding to each candidate thermal management parameter into the trained thermal management parameter inference model to obtain the target thermal management parameters of the vehicle.

[0060] In this embodiment, the acquired comprehensive state features, the identified operating stage category, the predicted future operating parameters, and the calculated comprehensive benefit values ​​corresponding to each candidate thermal management parameter are input into the trained thermal management parameter inference model. The thermal management parameter inference model processes the above information comprehensively and outputs the target thermal management parameters of the vehicle.

[0061] The target thermal management parameters are the specific control commands that are ultimately determined and used to control the vehicle's thermal management system.

[0062] In its implementation, the thermal management parameter inference model integrates multiple pieces of information, including comprehensive state characteristics, operating stage category, future operating parameters, and comprehensive benefit values ​​corresponding to each candidate thermal management parameter. After forward propagation of the model, it outputs the target thermal management parameters for the current moment, including thermal management mode, lower limit of battery target temperature, upper limit of battery target temperature, upper limit of heating power, upper limit of cooling power, and upper limit of pump flow rate.

[0063] The target thermal management parameters can be obtained through a single forward propagation of the thermal management parameter inference model, without the need for online iterative solutions to the optimization problem, thus ensuring the planning effect while meeting the real-time requirements of the vehicle platform.

[0064] Step 106: Perform thermal management on the vehicle according to the target thermal management parameters.

[0065] In this embodiment, the obtained target thermal management parameters can be sent to the vehicle's thermal management actuator. The thermal management actuator then operates according to the control commands in the target thermal management parameters to adjust and control the vehicle's battery temperature, ensuring that the battery temperature is within the temperature range set by the target thermal management parameters. Simultaneously, the heating power, cooling power, and pump flow rate are controlled to not exceed the upper power limit set by the target thermal management parameters.

[0066] In specific implementations, thermal management actuators may include one or more of the following: compressor, water pump, heater, electronic expansion valve, cooling fan, three-way valve, or four-way valve.

[0067] The thermal management actuator determines the current control strategy based on the thermal management mode in the target thermal management parameters, determines the allowable fluctuation range of the battery temperature based on the lower and upper limits of the battery target temperature, and constrains the output power and flow rate of the actuator based on the upper limits of heating power, cooling power, and pump flow rate.

[0068] The thermal management actuator monitors the current battery temperature in real time and maintains the battery temperature within the target temperature range by adjusting the compressor, heater, water pump and valves, while ensuring that the heating power, cooling power and pump flow do not exceed the corresponding upper limit values.

[0069] For example, when the battery temperature is below the target lower limit, the thermal management actuator activates the heating circuit to heat the battery. When the battery temperature is above the target upper limit, the thermal management actuator activates the cooling circuit to cool the battery. When the battery temperature is within the target temperature range, the thermal management actuator maintains its current operating state or makes only minor adjustments. Through this closed-loop control, the battery temperature is always kept within the target temperature range, thus achieving full-journey thermal management control.

[0070] This application acquires the comprehensive state characteristics of the vehicle within a preset time period and the route feature sequence ahead of the vehicle. The comprehensive state characteristics are input into a trained operation phase recognition model for identification, resulting in the vehicle's operation phase category. This allows for the determination of the vehicle's actual operation phase, providing reference information for subsequent determination of thermal management parameters. This application also inputs the comprehensive state characteristics, operation phase category, and route feature sequence into a trained operation parameter prediction model for prediction, obtaining the vehicle's future operation parameters. This enables the prediction of operation parameters during future driving and charging processes, providing sufficient predictive information for determining thermal management parameters. Furthermore, this application acquires several candidate thermal management parameters and inputs each candidate thermal management parameter, along with future operation parameters, into a thermal management benefit prediction model for prediction, obtaining the comprehensive benefit value corresponding to each candidate thermal management parameter. This allows for a unified quantitative evaluation of the benefits corresponding to thermal management parameters. Finally, this application inputs the comprehensive state characteristics, operation phase category, future operation parameters, and the comprehensive benefit value corresponding to each candidate thermal management parameter into a trained thermal management parameter inference model for inference, obtaining the vehicle's target thermal management parameters. Thermal management of the vehicle is then performed based on these target thermal management parameters. It can avoid solving complex optimization problems online, can be stably deployed on the vehicle platform, and ensures the real-time performance and effectiveness of vehicle thermal management.

[0071] Optionally, the runtime identification model includes a first feature extraction module, a bidirectional loop module, and a cross-attention fusion module.

[0072] In this embodiment, the first feature extraction module may be a temporal convolutional network, used to extract features in the time dimension from the input comprehensive state features, capture the changing trends of each physical quantity within a short time window, and output state change features.

[0073] The bidirectional loop module can be a bidirectional gated loop unit used to perform bidirectional time series modeling of state change characteristics, and simultaneously capture the state evolution patterns in both directions from historical time to current time and from future time to current time, and output state time series characteristics.

[0074] The cross-attention fusion module is used to perform cross-attention calculation by using navigation and power replenishment event features as query vectors and state temporal features as key-value pairs. This enables the model to adaptively weight the temporal features according to navigation and power replenishment intentions and output state fusion features.

[0075] The steps of inputting comprehensive state features into a trained operation phase identification model to identify the vehicle's operation phase category include: S1011, Input the comprehensive state features into the first feature extraction module for feature extraction to obtain state change features; S1012, input the state change characteristics into the bidirectional loop module for time series modeling to obtain the state time series characteristics; S1013, Obtain navigation power replenishment event features from comprehensive state features; Input navigation power replenishment event features and state temporal features into cross-attention fusion module for fusion to obtain state fusion features; S1014, determine the vehicle's operating phase category based on state fusion characteristics.

[0076] In this embodiment, the acquired comprehensive state features can be input into a first feature extraction module for feature extraction to obtain state change features. The first feature extraction module can be a temporal convolutional network used to extract the changing trends of each physical quantity in the comprehensive state features within a short time window.

[0077] State change characteristics can include information on the direction and magnitude of changes in the vehicle's multi-dimensional states, such as vehicle speed, acceleration, state of charge, and battery temperature, within a short time window, providing local dynamic information for subsequent time series modeling.

[0078] In this embodiment, state change features can be input into a bidirectional loop module for time-series modeling to obtain state time-series features. The bidirectional loop module can be a bidirectional gated loop unit, used to simultaneously perform time-series modeling of state change features from both forward and backward directions, capturing the evolution pattern of state change features over a longer time range.

[0079] State temporal features can include information on the evolution trend of vehicle state over a long time dimension, providing a temporal information basis for subsequent fusion with navigation recharge event features.

[0080] In this embodiment of the application, navigation power replenishment event features can be extracted from comprehensive state features.

[0081] Navigation charging event characteristics may include distance to charging station, estimated arrival time, charging pile power, estimated waiting time, and user intent encoding.

[0082] The navigation and power replenishment event features and state temporal features are input into the cross-attention fusion module for fusion to obtain the state fusion features. The cross-attention fusion module can use the navigation and power replenishment event features as query vectors and the state temporal features as key-value pairs to perform cross-attention calculation. This is used to adaptively weight the temporal features according to the navigation and power replenishment intentions, highlighting the temporal information related to the current navigation and power replenishment intentions and suppressing irrelevant information.

[0083] State fusion features can include the temporal evolution of vehicle state and navigation recharge intention information, providing a more comprehensive feature representation for stage recognition.

[0084] In this embodiment, the state fusion features can be mapped to the probability that the vehicle belongs to each preset operating stage, and the operating stage with the highest probability is taken as the vehicle's operating stage category. The operating stage category is used to characterize which operating mode the vehicle belongs to at the current moment.

[0085] In the specific implementation, the identification process of the model during the runtime phase is as follows: To achieve multi-stage coordination of driving, fast charging, and recharging before driving, this application constructs an operation stage identification model. The operation stage identification model can use an Intent-EnhancedStage Recognition Network (IE-SRN) to identify the current operation stage of the vehicle and provide stage labels for subsequent prediction and planning.

[0086] The input to the identification model during the runtime phase is a comprehensive set of state features:

[0087] in, Indicates the overall state characteristics. Indicates the characteristics of the operating status. Indicates battery characteristics, Indicates navigation power replenishment characteristics, Indicates environmental characteristics.

[0088] The IE-SRN network for the recognition model during the runtime phase consists of three parts: a first feature extraction module, a bidirectional recurrent module, and a cross-attention fusion module.

[0089] The first feature extraction module uses a one-dimensional temporal convolution network (TCN) to extract dynamic features within a short time window:

[0090] in, This represents the state change feature output by the first feature extraction module. This feature encodes the changing trend and dynamic pattern of the vehicle's multidimensional state within a short time window. This represents the time series consisting of the comprehensive state features of L consecutive times from time t-L+1 to time t, where L represents the length of the local history window, i.e., the number of time steps covered by the temporal convolutional network in the time dimension. TCN() represents a one-dimensional temporal convolutional network.

[0091] This formula means that the comprehensive state features of the most recent L time steps are used as input, and the features are extracted through a temporal convolutional network to obtain the state change features at time t.

[0092] The bidirectional loop module uses a bidirectional gated loop unit (BiGRU) to extract long-term time dependencies:

[0093]

[0094]

[0095] in, This represents the state change features output by the first feature extraction module. This represents the forward hidden state output by the forward GRU in the bidirectional loop module at time t. This state encodes the unidirectional timing information from the historical time to the current time. This represents the forward hidden state output at time t-1. This represents the backward hidden state output by the backward GRU in the bidirectional loop module at time t. This state encodes unidirectional timing information from a future time to the current time. This represents the backward hidden state output at time t-1. This represents the state-time feature obtained by concatenating the forward hidden state and the backward hidden state. This feature integrates temporal information from both the historical time to the current time and the future time to the current time. GRU(·) represents a gated cyclic unit.

[0096] This set of formulas indicates that the state change features are input into the forward GRU and backward GRU respectively for bidirectional temporal modeling, resulting in forward hidden states and backward hidden states. The forward hidden states and backward hidden states are then concatenated to obtain the state temporal features.

[0097] The cross-attention fusion module performs cross-attention fusion on navigation path features and state temporal features:

[0098]

[0099] in, This represents the navigation power replenishment event features extracted from the comprehensive state features. , , These are the projection weight matrices of the query vector, key vector, and value vector in the cross-attention calculation, respectively. This indicates the state timing characteristics of the output of the bidirectional loop module. This represents the state fusion feature output by the cross-attention fusion module. Attn(·) represents the cross-attention function, which is used to calculate the attention weight between the query vector and the key vector, and then apply the attention weight to the value vector to obtain the weighted feature representation.

[0100] This is the weight matrix of the fully connected layer. This represents the bias vector of the fully connected layer. Softmax(·) denotes the normalization exponential function, used to map the output of the fully connected layer to a probability distribution. This represents the probability distribution of each vehicle belonging to a different preset operating phase.

[0101] This set of formulas maps navigation recharge event features to query vectors and state temporal features to key and value vectors, respectively. State fusion features are then calculated using cross-attention. The state temporal features and state fusion features are concatenated and input into a fully connected layer. A normalized exponential function maps these features to the probability of the vehicle belonging to each preset operating stage. The operating stage with the highest probability is then designated as the vehicle's operating stage category.

[0102] In practical implementation, the runtime identification model can be trained in the following way: Acquire training samples, which include the comprehensive state features of the samples and the categories of the sample operation stages; The comprehensive state features of the samples are input into the operational phase identification model to be trained to obtain the predicted operational phase category; Calculate the loss function value based on the predicted operational phase category and the sample operational phase category; The parameters of the running phase recognition model to be trained are adjusted based on the loss function value until a preset training stopping condition is met, resulting in a trained running phase recognition model. The preset training stopping condition can include any of the following: the loss function value is less than a preset threshold, the number of training epochs reaches a preset maximum, or the validation set accuracy no longer improves after a preset number of consecutive training epochs.

[0103] The loss function value of the identification model during the runtime phase can be calculated as follows: The cross-entropy loss method is used to train the runtime identification model.

[0104] in, This represents the loss function value of the identification model during the runtime phase, where K represents the total number of categories during the runtime phase. This represents the running stage category of the sample at time t in the training samples. This represents the probability of the i-th stage category predicted by the model at time t. This loss function measures the difference between the probability distribution predicted by the model and the stage category of the sample. During training, minimizing this loss function value optimizes the model parameters, making the stage category predicted by the model closer to the stage category of the sample.

[0105] This application achieves multi-dimensional perception of vehicle operation stages by setting up a first feature extraction module to extract short-term change trends of comprehensive state features, a bidirectional loop module to perform bidirectional temporal modeling of state change features, and a cross-attention fusion module to fuse navigation power replenishment event features with temporal features. This solves the problem in the prior art that stage recognition relies on a single threshold trigger and cannot perceive navigation power replenishment intentions, improves the accuracy of stage recognition and adaptability to complex driving scenarios, and provides a reliable decision-making basis for the subsequent staged collaborative optimization of thermal management strategies.

[0106] Optionally, the operating parameter prediction model includes a distance axis prediction sub-model, a charge state axis prediction sub-model, and a thermal state prediction sub-model.

[0107] In this embodiment, the distance axis prediction sub-model is used to predict the vehicle speed, driving power, and regenerative power at each position during the future driving process, based on the future road static attribute information provided by the route feature sequence, combined with comprehensive state features and operating stage categories.

[0108] The state-of-charge (POC) prediction sub-model is used to predict the charging power and cumulative charging time for each POC interval during the future charging process, based on the current POC, battery temperature, and charging pile power, combined with the operating phase category.

[0109] The thermal state prediction sub-model is used to predict the battery temperature at each location during future driving, based on the predicted vehicle speed, predicted driving power, and predicted charging power output by the distance axis prediction sub-model and the charge state axis prediction sub-model, combined with comprehensive state characteristics.

[0110] The steps involved in inputting the comprehensive state characteristics, operational phase category, and route feature sequence into a trained operational parameter prediction model to predict the future operational parameters of the vehicle include: S1021, the route feature sequence, comprehensive state features and operation stage category are input into the distance axis prediction sub-model for prediction, to obtain the predicted vehicle speed, predicted driving power and predicted recovery power corresponding to each position in the future driving process; wherein, each position in the future driving process is obtained by discretizing the future path in front of the vehicle according to a preset distance interval; S1022, obtain the current state of charge, battery temperature, coolant temperature, charging pile power and ambient temperature from the comprehensive state characteristics; S1023, Input the current state of charge, battery temperature, coolant temperature, charging pile power, ambient temperature and operation stage category into the state of charge axis prediction sub-model for prediction, and obtain the predicted charging power and cumulative charging time corresponding to each state of charge interval in the future charging process; wherein, each state of charge interval in the future charging process is obtained by discretizing the future charging process according to the preset state of charge interval. S1024, the predicted vehicle speed, predicted driving power, predicted charging power and comprehensive state characteristics are input into the thermal state prediction sub-model for prediction, and the future battery temperature corresponding to each position during the future driving process is obtained. S1025 uses predicted vehicle speed, predicted drive power, predicted regenerative power, predicted charging power, cumulative charging time, and future battery temperature as future operating parameters.

[0111] In this embodiment, the route feature sequence, comprehensive state features, and operation stage category can be input into the distance axis prediction sub-model for prediction to obtain the predicted vehicle speed, predicted driving power, and predicted recovery power corresponding to each position during future driving.

[0112] The route feature sequence is obtained by discretizing the future path ahead of the vehicle at preset distance intervals. Each location corresponds to a road segment, and each road segment contains road attribute information such as gradient, speed limit, road type, and traffic congestion status.

[0113] The distance axis prediction sub-model uses this road attribute information, combined with the current vehicle status and operating phase category, to predict the vehicle speed, required driving power, and recyclable power when the vehicle travels to each road segment.

[0114] In this embodiment of the application, the current state of charge, battery temperature and charging pile power in the comprehensive state characteristics can be obtained. The current state of charge, battery temperature, charging pile power and operation stage category are input into the state of charge axis prediction sub-model for prediction, so as to obtain the predicted charging power and cumulative charging time corresponding to each state of charge interval in the future charging process.

[0115] The state of charge intervals in the future charging process are obtained by discretizing the future charging process according to the preset state of charge intervals, and each state of charge interval corresponds to a power interval in the charging process.

[0116] The state-of-charge (POC) prediction sub-model predicts the charging power and the cumulative time consumed from the start of charging to the end of each POC interval based on the current POC, current battery temperature, and charging pile power, combined with the operation stage category.

[0117] In this embodiment, the current state of charge, battery temperature, coolant temperature, charging pile power, and ambient temperature can be obtained from the comprehensive state characteristics. The current state of charge, battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category are input into the state of charge axis prediction sub-model for prediction, so as to obtain the predicted charging power and cumulative charging time corresponding to each state of charge interval in the future charging process.

[0118] The state of charge intervals in the future charging process are obtained by discretizing the future charging process according to the preset state of charge intervals, and each state of charge interval corresponds to a power interval in the charging process.

[0119] The state-of-charge (POC) prediction sub-model predicts the charging power and the cumulative time consumed from the start of charging to the end of each POC interval based on the current POC, current battery temperature, coolant temperature, charging pile power, and ambient temperature, combined with the operating phase category.

[0120] In this embodiment, the predicted vehicle speed, predicted driving power, predicted charging power, and comprehensive state characteristics can be input into the thermal state prediction sub-model for prediction to obtain the future battery temperature corresponding to each position during future driving.

[0121] The thermal state prediction sub-model determines the battery heat generation during driving based on the predicted vehicle speed and predicted drive power at each future location, and determines the impact of the charging stage on the battery temperature based on the predicted charging power during the future charging process. Combining the current battery temperature, current coolant temperature, and thermal management actuator status in the comprehensive state characteristics, the model predicts the battery temperature at each future location through the thermal balance equation and residual correction.

[0122] In this embodiment, predicted vehicle speed, predicted driving power, predicted regenerative power, predicted charging power, cumulative charging time, and future battery temperature can be used together as future operating parameters. These future operating parameters encompass predicted information across three dimensions: driving load, charging load, and battery temperature, providing a comprehensive predictive basis for subsequent thermal management decisions.

[0123] This application achieves forward-looking prediction of driving load by setting up a distance axis prediction sub-model to predict vehicle speed, driving power, and regenerative power during future driving based on route feature sequences; it achieves forward-looking prediction of charging load by setting up a state of charge axis prediction sub-model to predict charging power and charging time during future charging based on current state of charge, battery temperature, and charging pile power; and it achieves forward-looking prediction of battery thermal state evolution by setting up a thermal state prediction sub-model to predict future battery temperature based on driving load prediction results and charging load prediction results. This solves the problem in existing technologies that cannot simultaneously predict operating parameters during driving and charging phases, enabling subsequent thermal management decisions to simultaneously consider the needs of both driving and charging phases, and providing comprehensive predictive information for collaborative optimization throughout the entire journey.

[0124] Optionally, the distance axis prediction sub-model includes a second feature extraction module, a path feature extraction module, and a first state gating module.

[0125] In this embodiment of the application, the second feature extraction module is used to extract local features of the road attribute features of each road segment in the route feature sequence and capture the local correlation information between adjacent road segments.

[0126] The path feature extraction module is used to extract global features from the local features of the route output by the second feature extraction module, capturing the dependencies between long-distance road segments along the entire path.

[0127] The first state gating module generates state weights based on the current vehicle state and the category of the operating stage. These weights are used to weight the global features of the path, enabling the model to assign different levels of attention to features at different locations along the path based on the current vehicle state.

[0128] The steps of inputting route feature sequences, comprehensive state features, and operational phase categories into the distance axis prediction sub-model for prediction, and obtaining the predicted vehicle speed, predicted driving power, and predicted regenerative power for each position during future driving, include: S1031, Input the route feature sequence into the second feature extraction module for feature extraction to obtain local route features; S1032, Input the local features of the route into the path feature extraction module for feature extraction to obtain the global features of the path; S1033, input the comprehensive state characteristics and operation stage category into the first state gating module for weight calculation to obtain the path state weight; S1034, The global features of the path are weighted according to the path state weight to obtain the target path state features; S1035, Determine the predicted vehicle speed at each location during future driving based on the target path state characteristics; S1036, Determine the road gradient at each location during future travel based on the route feature sequence; S1037, based on the predicted vehicle speed and road gradient at each location, determine the predicted driving power at each location during future driving; S1038, calculates the predicted recovery power at each location during future driving based on the predicted driving power.

[0129] In this embodiment, the route feature sequence can be input into a second feature extraction module for feature extraction to obtain local route features. The second feature extraction module can be a temporal convolutional network, used to extract local road attribute changes between adjacent road segments in the route feature sequence and output local route features.

[0130] Local features of a route can include information such as changes in slope, speed limit, and road type between each road segment and its adjacent road segments, providing a local basis for subsequent global feature extraction.

[0131] In this embodiment, local route features can be input into the path feature extraction module for feature extraction to obtain global path features. The path feature extraction module can be a Transformer network (a neural network based on a self-attention mechanism) used to perform global dependency modeling on local route features, capture the mutual influence relationships between long-distance road segments along the entire path, and output global path features.

[0132] Global path features can include global context information for the entire path, enabling the model to predict future power demand changes based on road attributes of distant road segments ahead.

[0133] In this embodiment, the comprehensive state features and operating stage category can be input into the first state gating module for weight calculation to obtain the path state weight. The first state gating module calculates the weight coefficient corresponding to each position based on vehicle state information such as current vehicle speed, state of charge, and battery temperature, as well as the current operating stage category.

[0134] Path state weights are used to characterize the degree of attention paid to path features at different locations under the current vehicle state. For example, when the state of charge is low, higher attention is given to the drive power demand of the uphill section ahead.

[0135] In this embodiment of the application, the global features of the path can be weighted according to the path state weight to obtain the target path state features.

[0136] Weighted processing can be done by element-wise multiplication, multiplying the path state weights with the feature vectors corresponding to each position in the global features of the path. This allows the model to adaptively adjust the features at different positions in the path according to the current vehicle state, thereby enhancing the parts of the path features that are related to the current state and suppressing the parts that are unrelated to the current state.

[0137] In this embodiment, the predicted vehicle speed at each location during future travel can be determined based on the target path state features. The target path state features are mapped to the predicted vehicle speed at each location through a fully connected layer, and the predicted vehicle speed reflects the possible travel speed of the vehicle on each road segment in the future.

[0138] In this embodiment, the road gradient at each location during future travel can be determined based on the route feature sequence. The road gradient is the gradient value of each road segment directly extracted from the route feature sequence, which reflects the degree of inclination of each road segment in the vertical direction.

[0139] In the embodiments of this application, the predicted driving power at each location during future driving can be determined based on the predicted vehicle speed and road gradient at each location.

[0140] Predicted driving power can be calculated based on parameters such as vehicle mass, air resistance coefficient, rolling resistance coefficient, road gradient, and predicted vehicle speed, reflecting the driving power required for the vehicle to overcome driving resistance on various road sections.

[0141] In the embodiments of this application, the predicted recovery power at each location during future driving can be calculated based on the predicted driving power.

[0142] The predicted regenerative power can be calculated based on the predicted driving power and the preset braking regeneration efficiency. When the predicted driving power is negative, it indicates that the vehicle is decelerating or going downhill, and braking energy can be recovered at this time.

[0143] In the specific implementation, the prediction process of the distance axis prediction sub-model is as follows: A route-aware distance-axis driving load prediction sub-model is constructed to predict the vehicle speed, driving power, and regenerative braking power at each position during driving.

[0144] The input to the distance axis prediction sub-model is:

[0145] in, Indicates from the first The continuous path from the p-th segment Route feature sequence of each road segment This indicates the number of road segments covered during route feature extraction, i.e., the number of road segment steps covered by the second feature extraction module in the spatial dimension. This represents the overall state characteristics at time t. This indicates the stage category at time t.

[0146] The distance axis prediction sub-model includes a second feature extraction module, a path feature extraction module, and a first state gating module.

[0147] The second feature extraction module uses a one-dimensional temporal convolution network (TCN) to extract dynamic features within a short time window:

[0148] in, Indicates from the first The continuous path from the p-th segment The route feature sequence of each road segment, where TCN() represents a one-dimensional temporal convolutional network. This represents the local features of the route output by the second feature extraction module.

[0149] Then, the path feature extraction module, which is the Transformer network, is used to encode the global path dependencies:

[0150] in, This represents the local route features output by the second feature extraction module. Transformer(·) represents the Transformer network. This represents the global path features output by the path feature extraction module.

[0151] To reflect the impact of the current vehicle state on future power demand, a state gating mechanism is introduced, namely the first state gating module:

[0152]

[0153] in, This represents the path state weight output by the first-state gating module. This weight is used to characterize the degree of attention the current vehicle state pays to path features at different locations. σ(·) represents the sigmoid activation function, used to map the output value to a weight value between 0 and 1. This represents the weight matrix of the first-state gating module. This represents the bias vector of the first-state gating module. ⊙ represents the element-wise multiplication operation. This represents the target path state features obtained after weighting the global path features according to the path state weights. The meanings of the other parameters can be found in the preceding parameter definitions.

[0154] The final outputs include future speed, drive power, and regenerative power.

[0155] in, Indicates the target path state characteristics. This represents the weight matrix of the fully connected layer for vehicle speed prediction. This represents the bias vector of the fully connected layer for vehicle speed prediction. This represents the predicted vehicle speed at the p-th position during the future driving process, as output by the distance axis prediction sub-model.

[0156]

[0157] in, This represents the predicted vehicle speed at the p-th position during future driving, as output by the distance axis prediction sub-model. Let m represent the predicted acceleration at position p, and m represent the total vehicle mass. Indicates the air drag coefficient. This represents the rolling resistance coefficient, and g represents the acceleration due to gravity. This represents the road slope at the p-th location, a value obtained from the route feature sequence. This represents the predicted driving power at the p-th position during the future driving process, as output by the distance axis prediction sub-model.

[0158] This formula represents the calculation of predicted drive power based on predicted vehicle speed and road gradient. The first term on the right-hand side of the equation... The second item represents acceleration resistance power, reflecting the power required for a vehicle to accelerate or decelerate; The third item represents air resistance power, reflecting the power required to overcome air resistance during vehicle operation; This represents rolling resistance power, reflecting the power required to overcome rolling friction between the tires and the ground during vehicle operation; the fourth item... This represents the slope resistance power, which reflects the power required for a vehicle to overcome the component of gravity when going uphill or downhill. When the road slope is positive, this power is positive, indicating that additional driving power is needed. When the road slope is negative, this power is negative, indicating that gravity can be used to assist driving.

[0159]

[0160] in, This represents the predicted recovery power at the p-th position during future driving, as output by the distance-axis prediction sub-model. Indicates the braking energy recovery efficiency. This represents the predicted driving power at the p-th position during the future driving process, as output by the distance axis prediction sub-model. This means that when the predicted drive power is negative, the absolute value of the drive power is taken as the reference value of the recovered power; when the predicted drive power is positive, the recovered power is zero.

[0161] This formula calculates the predicted regenerative braking power based on the predicted driving power. When the predicted driving power is negative, it indicates that the vehicle is decelerating or going downhill, in which case braking energy can be recovered, and the recovered power equals the absolute value of the predicted driving power multiplied by the braking energy recovery efficiency. When the predicted driving power is positive, it indicates that the vehicle is accelerating or going uphill, in which case there is no regenerative braking energy, and the recovered power is zero.

[0162] In practice, the distance axis prediction sub-model can be trained in the following way: Acquire training samples, which include sample route feature sequences, sample comprehensive state features, sample operation stage categories, sample predicted vehicle speed, sample predicted driving power, and sample predicted recovery power. Input the sample route feature sequence, sample comprehensive state features and sample operation stage category into the distance axis prediction sub-model to be trained to obtain the predicted vehicle speed, predicted driving power and predicted recovery power; The loss function value is calculated based on the predicted vehicle speed and the sample predicted vehicle speed, the predicted driving power and the sample predicted driving power, and the predicted recovery power and the sample predicted recovery power. The parameters of the distance axis prediction sub-model to be trained are adjusted based on the loss function value until a preset training stopping condition is met, resulting in a trained distance axis prediction sub-model. The preset training stopping condition can include any of the following: the loss function value is less than a preset threshold, the number of training epochs reaches a preset maximum, or the validation set accuracy no longer improves after a preset number of consecutive training epochs.

[0163] Among them, the training samples of the distance axis prediction sub-model are used for driving phase modeling, and the sample route feature sequence can be used to predict the future path at fixed distance intervals. Discretized Each road segment will be constructed as follows:

[0164] Where R represents the product of A sample route feature sequence composed of the features of each road segment. This represents the route characteristics corresponding to the p-th road segment, which includes information such as the slope value, speed limit value, road type, and traffic congestion status of the road segment.

[0165] The loss function value of the distance axis prediction sub-model can be calculated as follows:

[0166] in, This represents the loss function value of the distance axis prediction sub-model. denoted as the predicted vehicle speed sequence at each position during the future driving process predicted by the distance axis prediction sub-model, and v represents the corresponding sample predicted vehicle speed sequence in the training samples. This represents the predicted drive power sequence at each position during future driving, as predicted by the distance-axis prediction sub-model. This represents the predicted driving power sequence of the corresponding samples in the training samples. This represents the predicted recoverable power sequence at each position during the future driving process, as predicted by the distance-axis prediction sub-model. This represents the sequence of predicted recovery power for the corresponding samples in the training samples. , , These represent the weighting coefficients for vehicle speed prediction loss, drive power prediction loss, and regenerative power prediction loss, respectively, used to balance the contribution ratio of the three prediction tasks to the total loss. This represents the square of the L2 norm, used to calculate the squared Euclidean distance between the predicted value and the sample value.

[0167] This application achieves comprehensive perception and adaptive adjustment of the path information ahead by setting a second feature extraction module to extract local features from the route feature sequence, setting a path feature extraction module to extract global features from the local features of the route, and setting a first state gating module to adaptively weight the global features of the path according to the current vehicle state. This enables the prediction of vehicle speed, driving power and regenerative power to integrate road static attribute information and current vehicle dynamic state information, thereby improving the accuracy and adaptability of the prediction.

[0168] Optionally, the charged state axis prediction sub-model includes a first feature mapping module and a second state gating module.

[0169] In this embodiment, the first feature mapping module is used to map the current state of charge to a high-dimensional feature space to obtain the state of charge mapping feature.

[0170] The second state gating module is used to generate charging state weights based on battery temperature, charging pile power, and operating stage category. These weights are used to weight the state of charge mapping features, enabling the model to adaptively adjust the state of charge features according to the current battery temperature, charging pile power, and operating stage.

[0171] The steps of inputting the current state of charge (SOC), battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category into the SOC prediction sub-model to predict the predicted charging power and cumulative charging time for each SOC interval during the future charging process include: S1041, Input the current state of charge into the first feature mapping module to perform feature mapping, and obtain the state of charge mapping feature; S1042, input the battery temperature, coolant temperature, charging pile power, ambient temperature and operating stage category into the second state gating module for weight calculation to obtain the charging state weight; S1043, The state of charge mapping features are weighted according to the state of charge weights to obtain the target state of charge features. S1044, Based on the target charging state characteristics, determine the predicted charging power corresponding to each state of charge interval during the future charging process. S1045, Based on the predicted charging power corresponding to each state of charge interval, determine the interval charging time corresponding to each state of charge interval. S1046, Based on the interval charging time corresponding to each state of charge interval, determine the cumulative charging time when reaching the end of each state of charge interval.

[0172] In this embodiment, the current state of charge can be input into a first feature mapping module for feature mapping. The first feature mapping module can be a fully connected layer, used to map the scalar value of the current state of charge into a high-dimensional feature vector.

[0173] State of charge mapping features are feature representations of the current state of charge in a high-dimensional feature space, enabling the model to capture the complex mapping relationship between the state of charge and charging power in a high-dimensional space.

[0174] In this embodiment, battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category can be input into the second state gating module for weight calculation. The second state gating module calculates the weight coefficient corresponding to each state of charge interval based on the above input information. The charging state weight is used to characterize the degree of attention paid to each state of charge interval under the current battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage conditions.

[0175] In this embodiment, the state of charge mapping features can be weighted according to the charging state weights. The weighting process can be element-wise multiplication, multiplying the charging state weights by the feature vectors corresponding to each state of charge interval in the state of charge mapping features, allowing the model to adaptively adjust the state of charge interval features based on the current charging conditions. The target charging state feature is the weighted feature representation used for subsequent charging power prediction.

[0176] In this embodiment, the predicted charging power corresponding to each state of charge interval during the future charging process is determined based on the target charging state characteristics. The target charging state characteristics are mapped to the predicted charging power for each state of charge interval through a fully connected layer.

[0177] The predicted charging power reflects the charging power that the battery can accept when charging to each state of charge range under the current battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage conditions.

[0178] In this embodiment, the interval charging time corresponding to each state of charge interval is determined based on the predicted charging power corresponding to each state of charge interval. The interval charging time is the time length obtained by dividing the increase in the amount of electricity in the state of charge interval by the corresponding predicted charging power, reflecting the time required to complete the charging of the state of charge interval.

[0179] In this embodiment, the cumulative charging time to reach the end of each state of charge interval is determined based on the interval charging time corresponding to each state of charge interval. The cumulative charging time is the sum of the interval charging times from the start of charging to the end of the state of charge interval, reflecting the total time consumed from the start of charging to reaching the end of the state of charge interval.

[0180] In the specific implementation, the prediction process of the charge state axis prediction sub-model is as follows: The input to the charge state axis prediction sub-model is:

[0181] in, This represents the current charge state corresponding to the k-th charge state interval. This represents the battery temperature corresponding to the k-th state of charge interval. This represents the coolant temperature corresponding to the k-th state of charge interval. Indicates the charging pile's power. Indicates ambient temperature. This indicates the operational phase category corresponding to the k-th charge state interval.

[0182] The charged state axis prediction sub-model includes a first feature mapping module and a second state gating module.

[0183] First, the first feature mapping module is used to map the scalar value of the current state of charge into a high-dimensional feature vector:

[0184] in, This represents the state-of-charge mapping characteristic, derived from the current state of charge. It is obtained by mapping through the first feature mapping module. This represents the weight matrix of the first feature mapping module. This represents the bias vector of the first feature mapping module, which is used to map the scalar value of the state of charge into a high-dimensional feature vector.

[0185] Then, the battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category are input into the second state gating module for weight calculation:

[0186] in, This represents the charging state weight output by the second-state gating module, used to characterize the charging state at the current battery temperature. Coolant temperature Charging pile power Ambient temperature and operational phase categories Under these conditions, the degree of attention paid to each charge state range. This represents the weight matrix of the second-state gating module. This represents the bias vector of the second-state gating module. σ(·) represents the sigmoid activation function, used to map the gating output value to weights between 0 and 1.

[0187] Therefore, we can predict the future... The charging power for each SOC range is:

[0188] in, This represents the predicted charging power corresponding to the k-th charge state interval during the future charging process, as predicted by the charge state axis prediction sub-model. This represents the weight matrix of the fully connected layer for predicting charging power. This represents the bias vector of the fully connected layer for predicting charging power. ⊙ represents the element-wise multiplication operation used to map the state of charge to features. With charging state weight After performing element-wise multiplication to obtain the target charging state features, they are mapped to the predicted charging power.

[0189] To ensure the monotonicity of the cumulative charging time, the charging time for each SOC increment is defined as:

[0190] in, This represents the interval charging time corresponding to the k-th state of charge interval. softplus(·) represents the softplus activation function, used to ensure that the output interval charging time is a positive value. This represents the weight matrix of the fully connected layer for predicting interval charging time. This represents the bias vector for predicting the interval charging time of the fully connected layer. ⊙ represents the element-wise multiplication operation, used to map the state of charge to features. With charging state weight Element-wise multiplication is performed. The softplus activation function ensures that the charging time within a given interval is positive, thus guaranteeing that the cumulative charging time monotonically increases with the state of charge, which aligns with the physical law that time accumulates with the increase of charge during actual charging.

[0191] The total charging time is:

[0192] in, This represents the cumulative charging time to reach the end of the k-th state of charge interval, which is equal to the charging time of all intervals from the 1st state of charge interval to the k-th state of charge interval. The sum of .

[0193] In practical implementation, the charge state axis prediction sub-model can be trained in the following way: Acquire training samples, which include the current state of charge of the sample, the sample battery temperature, the sample coolant temperature, the sample charging pile power, the sample ambient temperature, the sample operating stage category, the sample predicted charging power, and the sample cumulative charging time. Input the current state of charge of the sample, the sample battery temperature, the sample coolant temperature, the sample charging pile power, the sample ambient temperature, and the sample operation stage category into the state of charge axis prediction sub-model to be trained to obtain the predicted charging power and the predicted cumulative charging time. The loss function value is calculated based on the predicted charging power, the sample predicted charging power, the predicted cumulative charging time, and the sample cumulative charging time. The parameters of the charged state axis prediction sub-model to be trained are adjusted based on the loss function value until a preset training stopping condition is met, resulting in a trained charged state axis prediction sub-model. The preset training stopping condition can include any of the following: the loss function value is less than a preset threshold, the number of training epochs reaches a preset maximum, and the accuracy of the validation set no longer improves after a preset number of consecutive training epochs.

[0194] Among them, in the training samples of the State of Charge (SOC) axis prediction sub-model, the SOC axis samples are used for modeling the charging phase, and the charging process is divided into fixed SOC intervals. Discretized Intervals, construct

[0195] Where C represents by A sequence of charging state samples consisting of a range of charging state intervals. This represents the charging state sample corresponding to the k-th state of charge interval. The sample includes information such as the state of charge value, battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category for that interval.

[0196] The loss function value of the charged state axis prediction sub-model can be calculated as follows:

[0197] in, This represents the loss function value of the charge state axis prediction sub-model. This represents the predicted charging power sequence corresponding to each state of charge interval during the future charging process, as predicted by the state of charge axis prediction sub-model. This represents the predicted charging power sequence of the corresponding sample in the training samples. This represents the cumulative charging time series predicted by the charge state axis prediction sub-model when each charge state interval ends. This represents the cumulative charging time series of the corresponding samples in the training samples. and These represent the weighting coefficients for the predicted charging power loss and the predicted cumulative charging time loss, respectively, used to balance the contribution ratio of the two prediction tasks to the total loss.

[0198] This represents the square of the L2 norm, used to calculate the squared Euclidean distance between the predicted value and the sample value.

[0199] This loss function measures the difference between the charging power sequence and the cumulative charging time sequence predicted by the state-of-charge prediction sub-model and the corresponding sample sequence. During training, the model parameters are optimized by minimizing the value of this loss function, so that the charging power and cumulative charging time predicted by the model are closer to the sample values.

[0200] This application achieves multi-condition perception and adaptive adjustment of the charging process by setting a first feature mapping module to map the state of charge and a second state gating module to generate charging state weights based on battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category, and adaptively weighting the state of charge mapping features. This enables the prediction of charging power and charging time to integrate the influence of multiple factors such as battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage, thereby improving the accuracy of charging process prediction and adaptability to different charging conditions.

[0201] Optionally, the thermal state prediction sub-model includes a residual correction module.

[0202] In this embodiment, the residual correction module is used to compensate for the prediction bias of the heat balance equation caused by factors such as model simplification, unmodeled disturbances, and system hysteresis.

[0203] The steps involved in predicting the future battery temperature at various locations during future driving include inputting predicted vehicle speed, predicted drive power, predicted charging power, and comprehensive state characteristics into the thermal state prediction sub-model. S1051, obtain the current battery temperature and current coolant temperature from the thermal management actuator status and comprehensive status characteristics; S1052, based on the predicted driving power and predicted charging power corresponding to each position during future driving, determine the battery heat generation corresponding to each position during future driving; S1053, based on the current battery temperature, the current coolant temperature, the battery heat generation at each location during future driving, the thermal management actuator status, and the preset thermal balance equation, determine the theoretical battery temperature and theoretical coolant temperature at each location during future driving. S1054 inputs the comprehensive state characteristics, operating stage category, predicted vehicle speed, predicted driving power, predicted charging power, theoretical battery temperature and theoretical coolant temperature corresponding to each position during future driving into the residual correction module to obtain the battery temperature correction amount corresponding to each position during future driving. S1055 determines the future battery temperature at each location during future driving based on the theoretical battery temperature and battery temperature correction amount at each location during future driving.

[0204] In this embodiment of the application, the current battery temperature and current coolant temperature can be obtained from the thermal management actuator status and the comprehensive status characteristics.

[0205] The current battery temperature reflects the battery's actual temperature at the current moment and serves as the initial condition for the thermal balance equation. The current coolant temperature reflects the actual temperature of the coolant in the thermal management system at the current moment and affects the heat exchange efficiency between the battery and the coolant.

[0206] The status of the thermal management actuator includes information such as compressor speed, pump flow rate, valve opening, heating power, and cooling power, which reflects the actual working status of the thermal management system at the current moment. This information determines the thermal management system's ability to actively regulate battery temperature.

[0207] In a practical implementation, the state of the thermal management actuator can be represented in the following form:

[0208] in, This is the status of the thermal management actuator. It refers to the compressor speed. It is the pump flow rate. For valve opening, It is the heating power. It refers to cooling capacity.

[0209] In this embodiment of the application, the battery heat generation at each location during future driving can be determined based on the predicted driving power and predicted charging power corresponding to each location.

[0210] Battery heat generation is jointly determined by the driving power demand during driving and the charging power demand during charging. The driving power corresponds to the heat generation during battery discharge, and the charging power corresponds to the heat generation during battery charging. Battery heat generation is the heat source term in the heat balance equation, directly affecting the battery temperature change trend.

[0211] In this embodiment, the theoretical battery temperature and theoretical coolant temperature at each location during future driving can be determined based on the current battery temperature, the current coolant temperature, the battery heat generation at each location during future driving, the thermal management actuator status, and a preset thermal balance equation.

[0212] The heat balance equation describes the heat exchange process between the battery and the coolant, as well as between the coolant and the environment. This includes the heat generated by the battery, the heat exchange between the battery and the coolant, the heating provided by the heater, the cooling provided by the cooler, and the heat dissipation between the coolant and the external environment. The theoretical battery temperature and the theoretical coolant temperature are temperature values ​​calculated under ideal conditions based on the heat balance equation.

[0213] In this embodiment, the comprehensive state characteristics, operating stage category, predicted vehicle speed, predicted driving power, predicted charging power, theoretical battery temperature and theoretical coolant temperature corresponding to each position during future driving can be input into the residual correction module to obtain the battery temperature correction amount corresponding to each position during future driving.

[0214] The residual correction module can be a multilayer perceptron network used to learn the residual mapping relationship between the theoretical predictions of the heat balance equation and the actual temperature. The battery temperature correction is used to compensate for prediction biases caused by factors such as model simplification, unmodeled disturbances, and system hysteresis.

[0215] In this embodiment, the future battery temperature at each location during future driving can be determined based on the theoretical battery temperature and the battery temperature correction amount. The theoretical battery temperature is then added to the battery temperature correction amount to obtain the corrected future battery temperature. This temperature serves as the final prediction result of the thermal state prediction sub-model, providing forward-looking information on the battery's thermal state for subsequent thermal management decisions.

[0216] In the specific implementation, the prediction process of the thermal state prediction sub-model is as follows: First, establish the basic thermal balance equations for the battery and liquid circuit:

[0217]

[0218] in, This represents the theoretical battery temperature at the next moment, calculated using the thermal balance equation. This represents the battery temperature at the current moment. Δt represents the time step. It represents the battery's equivalent heat capacity, used to characterize the relationship between battery temperature changes and the heat it absorbs or releases. This represents the heat generated by the battery at time t. This represents the heat exchange between the battery and the coolant at time t. This represents the theoretical coolant temperature at the next moment, calculated using the heat balance equation. This indicates the current coolant temperature. This indicates the equivalent heat capacity of the coolant. This represents the amount of heat provided by the heater at time t. This indicates the amount of cooling provided by the cooler at time t. This represents the amount of heat dissipated between the coolant and the external environment at time t.

[0219] This set of formulas describes the evolution of battery temperature and coolant temperature over time. The change in battery temperature is determined by the difference between the heat generated by the battery and the heat exchanged by the battery coolant, while the change in coolant temperature is jointly determined by the heat exchanged by the battery coolant, the heating capacity of the heater, the cooling capacity of the cooler, and the heat dissipation between the coolant and the environment.

[0220] The heat generated by the battery is defined as follows:

[0221] This represents the heat generated by the battery at time t, calculated from the battery's equivalent internal resistance. With battery current I t It is calculated by multiplying the squares of the products. This indicates that the battery's equivalent internal resistance is a function of its state of charge and battery temperature; the equivalent internal resistance takes different values ​​at different states of charge and battery temperatures. t The value represents the battery current at time t. A positive value indicates the battery's output power during discharge, while a negative value indicates the battery absorbs external power during charging. The calculation of battery heat generation reflects the internal resistive heating effect during charging and discharging, and is one of the key inputs to the heat balance equation.

[0222] Then, the residual correction module was constructed. Due to the hysteresis, model simplification error, and unmodeled disturbances in the real system, this application further constructed the residual correction module:

[0223] in, This represents the residual vector output by the residual correction module, which is composed of the battery temperature correction amount. and coolant temperature correction Composition. MLP(·) represents a multilayer perceptron network. This represents the overall state characteristics at time t. This indicates the stage category at time t. This represents the predicted driving power at the k-th position predicted by the distance axis prediction sub-model. This represents the predicted charging power predicted by the state-of-charge prediction sub-model. This represents the theoretical battery temperature calculated from the thermal balance equation. This represents the theoretical coolant temperature calculated from the heat balance equation.

[0224] The future battery temperature and future coolant temperature are as follows:

[0225]

[0226] in, This indicates that the final predicted future battery temperature is equal to the theoretical battery temperature. With battery temperature correction sum. This indicates that the final predicted future coolant temperature is equal to the theoretical coolant temperature. With coolant temperature correction sum.

[0227] This set of formulas indicates that by inputting comprehensive state characteristics, operating phase category, predicted drive power, predicted charging power, theoretical battery temperature, and theoretical coolant temperature into the residual correction module, the battery temperature correction amount and the coolant temperature correction amount are obtained. The theoretical battery temperature is added to the battery temperature correction amount to obtain the final predicted future battery temperature; the theoretical coolant temperature is added to the coolant temperature correction amount to obtain the final predicted future coolant temperature. This data-driven residual correction method compensates for prediction biases caused by model simplification, unmodeled disturbances, and system hysteresis while preserving the interpretability of the thermal balance equations.

[0228] In practical implementation, the thermal state prediction sub-model can be trained in the following way: Acquire training samples, which include comprehensive state features of the samples, category of the sample's operating stage, predicted vehicle speed of the samples, predicted driving power of the samples, predicted charging power of the samples, theoretical battery temperature of the samples, theoretical coolant temperature of the samples, future battery temperature of the samples, and future coolant temperature of the samples. Input the comprehensive state characteristics of the sample, the category of the sample's operating stage, the predicted driving power of the sample, the predicted charging power of the sample, the theoretical battery temperature of the sample, and the theoretical coolant temperature of the sample into the thermal state prediction sub-model to be trained, and obtain the predicted future battery temperature and the predicted future coolant temperature. The loss function value is calculated based on the predicted future battery temperature and the sample future battery temperature, the predicted future coolant temperature and the sample future coolant temperature. The parameters of the hot state prediction sub-model to be trained are adjusted based on the loss function value until a preset training stopping condition is met, resulting in a trained hot state prediction sub-model. The preset training stopping condition can include any of the following: the loss function value is less than a preset threshold, the number of training epochs reaches a preset maximum, or the validation set accuracy no longer improves after a preset number of consecutive training epochs.

[0229] The loss function value of the thermal state prediction sub-model can be calculated as follows:

[0230] in, This represents the loss function value of the thermal state prediction sub-model. This represents the predicted future battery temperature sequence for each location predicted by the thermal state prediction sub-model. This represents the future battery temperature sequence of the corresponding sample in the training samples. This represents the predicted future coolant temperature sequence for each location predicted by the thermal state prediction sub-model. This represents the future coolant temperature sequence of the corresponding sample in the training samples. and These represent the weighting coefficients for battery temperature prediction loss and coolant temperature prediction loss, respectively, used to balance the contribution ratio of the two prediction tasks to the total loss.

[0231] This represents the square of the L2 norm, used to calculate the squared Euclidean distance between the predicted value and the sample value.

[0232] This loss function measures the difference between the battery temperature sequence and coolant temperature sequence predicted by the thermal state prediction sub-model and the corresponding sample sequence. During training, the model parameters are optimized by minimizing the value of this loss function, so that the battery temperature and coolant temperature predicted by the model are closer to the sample values.

[0233] This application corrects the theoretical prediction results of the thermal balance equation by setting a residual correction module, realizing the combination of physical model and data-driven approach. While preserving the interpretability of the thermal balance equation, it uses a data-driven approach to compensate for model simplification errors and unmodeled disturbances, thereby improving the accuracy of battery temperature prediction.

[0234] Optionally, the thermal management revenue prediction model includes a revenue calculation module.

[0235] In this embodiment, the benefit calculation module is used to calculate the performance of each candidate thermal management parameter in four dimensions—fast charging time benefit, recycling benefit, power benefit, and thermal management energy consumption cost—in conjunction with future operating parameters. The benefits and costs of the four dimensions are then weighted and combined to obtain the comprehensive benefit value corresponding to the candidate solution. This comprehensive benefit value is used to quantitatively evaluate the performance of the candidate thermal management parameter under a multi-objective comprehensive evaluation, providing an evaluation basis for the subsequent selection of thermal management parameters.

[0236] The steps of inputting each candidate thermal management parameter and future operating parameters into the thermal management benefit prediction model to obtain the comprehensive benefit value corresponding to each candidate thermal management parameter include: S1061, based on each candidate thermal management parameter and future operating parameters, calculate the fast charging time benefit, recycling benefit, power benefit and thermal management energy consumption cost corresponding to each candidate thermal management parameter through the benefit calculation module; S1062, based on the fast charging time benefit, recycling benefit, power benefit and thermal management energy consumption cost corresponding to each candidate thermal management parameter, determine the comprehensive benefit value corresponding to each candidate thermal management parameter.

[0237] In this embodiment of the application, the fast charging time benefit, recycling benefit, power benefit and thermal management energy consumption cost corresponding to each candidate thermal management parameter can be calculated by the benefit calculation module based on each candidate thermal management parameter and future operating parameters.

[0238] Fast charging time benefits are calculated based on the predicted charging power in future operating parameters, reflecting the contribution of the candidate scheme to shortening charging time. When the candidate scheme enables the battery to reach a higher charging power during charging, the fast charging time benefits increase.

[0239] The recovery benefit is calculated based on the predicted recovery power and predicted battery temperature in the future operating parameters, reflecting the contribution of the candidate scheme to improving the braking energy recovery efficiency. When the candidate scheme can put the battery temperature in a state more conducive to recovery, the upper limit of braking energy recovery power is increased, and the recovery benefit increases.

[0240] The power benefit is calculated based on the predicted drive power and predicted battery temperature in the future operating parameters, reflecting the contribution of the candidate scheme to ensuring the availability of vehicle drive power. When the candidate scheme can put the battery temperature in a state that is more conducive to high-power discharge, the upper limit of drive power is increased and the power benefit increases.

[0241] The energy cost of thermal management is calculated based on the heating power, cooling power and pump flow rate of the candidate scheme, reflecting the energy consumed in the heating, cooling and pumping process. The higher the energy cost, the more electrical energy the candidate scheme consumes in the thermal management process.

[0242] In this embodiment, the comprehensive benefit value corresponding to each candidate thermal management parameter can be determined based on the fast charging time benefit, recycling benefit, power benefit, and thermal management energy consumption cost corresponding to each candidate thermal management parameter. The comprehensive benefit value is equal to the weighted sum of the fast charging time benefit, recycling benefit, and power benefit, minus the thermal management energy consumption cost, so that the comprehensive benefit value numerically reflects the result of the multi-objective comprehensive evaluation.

[0243] A higher overall benefit value indicates a better performance of the candidate thermal management parameter under multi-objective comprehensive evaluation. By calculating the corresponding overall benefit value for each candidate thermal management parameter, the merits and demerits of multiple candidate schemes can be quantitatively compared in a unified manner, providing a basis for subsequent scheme selection.

[0244] In practice, the prediction process of the thermal management benefit prediction model is as follows: Define candidate thermal management parameters:

[0245] in, This represents the candidate thermal management parameters at the current moment. This indicates a reference value for battery-side heating power, used to provide heat when the battery temperature is below the target temperature. This indicates a reference value for battery-side cooling power, used to remove heat when the battery temperature is higher than the target temperature. This is a reference value indicating the pump flow rate, used to control the circulation speed of coolant in the thermal management system. This represents a reference value for the thermal management mode, used to determine the control strategy that the thermal management system should currently employ.

[0246] The revenue calculation module is as follows, including the revenue from fast charging time:

[0247] in, Indicates the adoption of candidate thermal management parameters The benefits of fast charging time gained during the process This indicates the predicted charging time under the baseline strategy, which can be a preset default thermal management strategy or a traditional rule-based strategy. Indicates the adoption of candidate thermal management parameters The charging time is then predicted. When a candidate solution can shorten the charging time, the reward is positive, and the greater the reduction, the higher the reward.

[0248] The future recovery losses are as follows:

[0249] in, Indicates the adoption of candidate thermal management parameters Future recovery losses at that time. This represents the total number of positions obtained by discretization during the future driving process. This represents the predicted recovery power at the i-th position during the future driving process, as predicted by the distance axis prediction sub-model. This represents the predicted battery temperature at position i. and predicting the state of charge Under certain conditions, the maximum allowable charging power of the battery is the upper limit of the maximum regenerative power that the battery can accept. This means that when the predicted recovery power exceeds the battery's maximum allowable charging power, the excess cannot be absorbed by the battery, resulting in a recovery loss; when the predicted recovery power does not exceed the maximum charging power, the recovery loss is zero.

[0250] The corresponding recovery benefit is:

[0251] This represents the recovery loss under the baseline strategy. Indicates the adoption of candidate thermal management parameters The recovery gain obtained at that time is equal to the recovery loss under the baseline strategy minus the recovery loss after adopting the candidate thermal management parameters. This gain reflects the degree to which the candidate scheme contributes to enabling the battery to absorb more recovered power by adjusting the battery temperature.

[0252] Define future insufficient power loss as

[0253] in, Indicates the adoption of candidate thermal management parameters The future momentum is insufficient and losses are incurred. This represents the predicted driving power at the i-th position during the future driving process, as predicted by the distance axis prediction sub-model. This represents the predicted battery temperature at position i. and predicting the state of charge Under certain conditions, the maximum allowable discharge power of the battery is the upper limit of the maximum driving power that the battery can provide. This means that when the predicted drive power exceeds the battery's maximum allowable discharge power, the excess cannot be provided by the battery, resulting in insufficient power loss; when the predicted drive power does not exceed the maximum discharge power, the insufficient power loss is zero.

[0254] The corresponding power benefits are:

[0255] This indicates the loss due to insufficient momentum under the baseline strategy. Indicates the adoption of candidate thermal management parameters The power gain obtained is equal to the power loss under the baseline strategy minus the power loss after adopting the candidate thermal management parameters. This gain reflects the contribution of the candidate scheme to enable the battery to provide more driving power by adjusting the battery temperature.

[0256] Define the energy consumption cost of the thermal management system as

[0257] in, Indicates the adoption of candidate thermal management parameters The energy cost of thermal management at that time. This represents the heating power at time τ. This represents the cooling power at time τ. Let τ represent the pump energy consumption at time τ. H represents the planning time domain length. Δt represents the sampling interval. The thermal management energy cost equals the sum of heating power, cooling power, and pump energy consumption at all times within the planning time domain multiplied by the sampling interval, reflecting the total electrical energy consumed in implementing the candidate thermal management parameter throughout the entire planning time domain.

[0258] Overall benefit value corresponding to candidate thermal management parameters:

[0259] Indicates the adoption of candidate thermal management parameters The overall return value at that time. , , , These parameters represent the weights of fast charging time benefits, recycling benefits, power benefits, and thermal management energy consumption costs, respectively, and are used to adjust the relative importance of each benefit and cost in the overall benefits.

[0260] This unified benefit function integrates four objectives—fast charging time benefit, recycling benefit, power benefit, and thermal management energy consumption cost—into a single evaluation framework, achieving a comprehensive multi-objective assessment. A higher comprehensive benefit value indicates better performance of the candidate thermal management parameter under this multi-objective comprehensive evaluation.

[0261] In its implementation, the thermal management benefit prediction model is a calculation module based on a preset formula. It directly calculates the fast charging time benefit, recycling benefit, power benefit, and thermal management energy consumption cost corresponding to each candidate scheme based on future operating parameters and candidate thermal management parameters, and obtains the comprehensive benefit value through weighted summation.

[0262] This application sets up a benefit calculation module to comprehensively evaluate each candidate thermal management parameter in four dimensions: fast charging time benefit, recycling benefit, power benefit, and thermal management energy consumption cost. It integrates multiple objectives such as energy consumption, fast charging efficiency, power availability, and recycling availability into a single evaluation framework, solving the problem of the lack of a unified evaluation standard between energy input and benefit in the prior art, and enabling thermal management parameters to be selected from the perspective of multi-objective comprehensive optimization.

[0263] Optionally, the thermal management parameter inference model includes a second feature mapping module, a stage gating module, a pattern recognition module, and a parameter generation module.

[0264] The second feature mapping module is used to map the spliced ​​target state parameters to a high-dimensional feature space and extract the fused state features.

[0265] The stage gating module is used to adaptively weight the fusion state features according to the category of the running stage, so that the model pays different attention to the fusion state features in different running stages.

[0266] The pattern recognition module is used to identify the thermal management mode that should be adopted at the current stage based on the stage fusion characteristics.

[0267] The parameter generation module is used to generate specific control parameter values ​​based on the stage fusion characteristics.

[0268] The steps for obtaining the vehicle's target thermal management parameters by inputting the comprehensive state characteristics, operating stage category, future operating parameters, and the comprehensive benefit value corresponding to each candidate thermal management parameter into the trained thermal management parameter inference model include: S1071, the comprehensive state characteristics, operation stage category, future operation parameters and the comprehensive benefit value corresponding to each candidate thermal management parameter are concatenated into the target state parameter; S1072, Input the target state parameters into the second feature mapping module for feature mapping to obtain the fused state features; S1073, input the fused state features and the operational stage category into the stage gating module for weight calculation to obtain the stage state weight; S1074, The fusion state features are weighted according to the stage state weights to obtain the stage fusion features; S1075, input the stage fusion features into the pattern recognition module to obtain the vehicle's thermal management mode; S1076, input the stage fusion feature into the parameter generation module to obtain the vehicle's target temperature lower limit, target temperature upper limit, heating power upper limit, cooling power upper limit and pump flow rate upper limit; S1077 uses the thermal management mode, the lower limit of the target temperature of the battery, the upper limit of the target temperature of the battery, the upper limit of the heating power, the upper limit of the cooling power, and the upper limit of the pump flow rate as the target thermal management parameters at the current moment.

[0269] In this embodiment, the comprehensive state characteristics, operational stage categories, future operational parameters, and the comprehensive benefit values ​​corresponding to each candidate thermal management parameter can be concatenated into a target state parameter. The concatenation operation connects the above four types of information sequentially into a one-dimensional feature vector, enabling subsequent modules to simultaneously obtain current state information, stage information, future prediction information, and benefit assessment information, providing a complete information foundation for comprehensive decision-making.

[0270] In this embodiment, the target state parameters can be input into a second feature mapping module for feature mapping to obtain fused state features. The second feature mapping module can be a neural network composed of fully connected layers, used to map the target state parameters to a high-dimensional feature space and extract deep feature representations of each dimension of the target state parameters.

[0271] Fusion state features are the feature representations of target state parameters in a high-dimensional space, which include a fusion representation of current state information, stage information, future prediction information, and benefit assessment information.

[0272] In this embodiment, the fused state features and the operational phase category can be input into the phase gating module for weight calculation to obtain the phase state weight. The phase gating module generates a corresponding weight coefficient based on the operational phase category, which is used to adaptively adjust the fused state features.

[0273] Stage state weights are used to characterize the degree of attention paid to each dimension of the fused state features under different operating stages. For example, temperature-related features are given more attention in the pre-charging preparation stage, and power-related features are given more attention in the fast charging stage.

[0274] In this embodiment, the fused state features can be weighted according to the stage state weights to obtain stage fused features. The weighting process can be element-wise multiplication, multiplying the stage state weights and fused state features element-wise, so that the model can adaptively adjust the fused state features according to the current running stage.

[0275] Stage fusion features are the result of stage adaptation based on fusion state features. The same fusion state features under different operating stages will result in different stage fusion features after stage gating, enabling subsequent pattern recognition and parameter generation to make differentiated decisions based on the operating stage.

[0276] In this embodiment, the stage fusion features can be input into the pattern recognition module to obtain the vehicle's thermal management modes. The pattern recognition module can be a classifier composed of a fully connected layer and a softmax function, used to map the stage fusion features to a probability distribution of each thermal management mode, and select the thermal management mode with the highest probability as the output.

[0277] Thermal management modes include normal mode, preheating mode, precooling mode and maintenance mode, and different modes correspond to different thermal management control strategies.

[0278] In this embodiment, the stage fusion features can be input into the parameter generation module to obtain the vehicle's target temperature lower limit, target temperature upper limit, heating power upper limit, cooling power upper limit, and pump flow rate upper limit. The parameter generation module can be a regression network composed of fully connected layers, used to map the stage fusion features into specific control parameter values.

[0279] The lower and upper limits of the battery target temperature together constitute the allowable fluctuation range of the battery temperature. The upper limit of heating power constrains the maximum output power of the heater. The upper limit of cooling power constrains the maximum output power of the cooling system. The upper limit of pump flow rate constrains the target flow rate of the coolant circulation pump. These parameters together constitute a complete thermal management control scheme.

[0280] In this embodiment, the thermal management mode, the lower limit of the target temperature of the battery, the upper limit of the target temperature of the battery, the upper limit of the heating power, the upper limit of the cooling power, and the upper limit of the pump flow rate can be used as the target thermal management parameters at the current moment and sent to the thermal management actuator for execution.

[0281] The target thermal management parameters can be obtained through a single forward propagation of the thermal management parameter inference model, without the need for online iterative solutions to the optimization problem, thus ensuring the planning effect while meeting the real-time requirements of the vehicle platform.

[0282] In practical implementation, the reasoning process of the thermal management parameter inference model is as follows: The input to the thermal management parameter inference model is:

[0283] in, The input to the thermal management parameter inference model is represented by the comprehensive state characteristics. Operational phase categories Future operating parameters The comprehensive benefit value corresponding to each candidate thermal management parameter It was pieced together.

[0284] The thermal management parameter inference model adopts a shared backbone layer and a dual-output head structure. The shared backbone layer maps the input and output to a high-level feature space. The second feature mapping module is as follows:

[0285]

[0286] This represents the first intermediate feature output by the first fully connected layer. φ() represents the activation function. This represents the weight matrix of the first fully connected layer. This represents the bias vector of the first fully connected layer.

[0287] This represents the second intermediate feature output by the second fully connected layer, which is the fused state feature. This represents the weight matrix of the second fully connected layer. This represents the bias vector of the second fully connected layer.

[0288] This set of formulas indicates that the input will be... The second intermediate feature is obtained by sequentially performing feature mapping through two fully connected layers. .

[0289] The thermal management parameter inference model includes a second feature mapping module, a stage gating module, a pattern recognition module, and a parameter generation module.

[0290] To enhance feature representation at different stages, a stage gating module is introduced:

[0291]

[0292] in, This represents the stage state weights output by the stage gating module. σ() represents the sigmoid activation function. This represents the weight matrix of the stage gating module. This represents the bias vector of the stage gating module. Indicates the category of the running phase. The stage fusion feature output by the stage gating module is equal to the stage state weight. Features of fusion state The result of element-wise multiplication.

[0293] Through a stage-gating mechanism, the model adaptively weights the second intermediate feature according to the current running stage. The same second intermediate feature will result in different stage-fused features after gating in different running stages.

[0294] The output of the discrete pattern recognition module is:

[0295]

[0296] in, This represents the pattern logic value vector output by the pattern recognition module. This indicates the stage fusion feature output by the stage gating module. This represents the weight matrix of the pattern recognition module. This represents the bias vector of the pattern recognition module. This represents the probability distribution of thermal management modes output by the pattern recognition module. The softmax function maps the pattern logic value vector to the probability value of each mode. This represents the probability that a vehicle belongs to one of M different thermal management modes, where M is the total number of mode categories. The mode with the highest probability is selected as the output thermal management mode.

[0297] The parameter generation module for continuous programming outputs:

[0298] in, This represents the continuous planning quantity output by the parameter generation module. This represents the weight matrix of the parameter generation module. This represents the bias vector of the parameter generation module. This represents the stage fusion feature output by the stage gating module. Continuous planning quantity This includes the lower limit of the target temperature for the battery. upper limit of target temperature for battery Upper limit of heating power Cooling power upper limit and pump flow rate upper limit .

[0299] This application achieves comprehensive processing of multi-source information and stage-adaptive generation of thermal management parameters by setting a second feature mapping module to perform feature mapping on target state parameters, setting a stage gating module to adaptively weight the fused state features according to the operation stage category, and setting a pattern recognition module and a parameter generation module to output thermal management mode and specific control parameters respectively. This enables the target thermal management parameters to make comprehensive decisions based on the current vehicle state, operation stage, future prediction and benefit assessment, thereby improving the accuracy and stage adaptability of thermal management decisions.

[0300] Optionally, the thermal management parameter inference model is trained in the following manner: S1081, Obtain training samples. The training samples include the comprehensive state features of the samples, the category of the sample's operating stage, the future operating parameters of the samples, the comprehensive benefit value of the samples corresponding to the candidate thermal management parameters, and the target thermal management parameters of the samples. S1082, input the comprehensive state characteristics of the sample, the category of the sample's operational stage, the sample's future operational parameters, and the comprehensive benefit value of the sample's candidate thermal management parameters into the thermal management parameter inference model to be trained, and obtain the predicted target thermal management parameters; S1083, calculate the loss function value based on the predicted target thermal management parameters and the sample target thermal management parameters; S1084, adjust the parameters of the thermal management parameter inference model to be trained based on the loss function value until the preset training stopping condition is met, and obtain the trained thermal management parameter inference model.

[0301] In this embodiment of the application, training samples can be obtained. The training samples include comprehensive state features of the samples, category of sample operation stage, future operation parameters of the samples, comprehensive benefit value of the samples corresponding to candidate thermal management parameters, and target thermal management parameters of the samples.

[0302] The sample comprehensive state features are the vehicle's comprehensive state features extracted from historical driving data. The sample operation stage category is the actual operation stage category at the corresponding moment in the historical data. The sample future operation parameters are future operation parameters obtained from historical data. The sample comprehensive benefit value corresponding to the sample candidate thermal management parameters is the comprehensive benefit value of each candidate scheme calculated by the thermal management benefit prediction model. The sample target thermal management parameters are the target thermal management parameters actually executed by the vehicle in the historical data, serving as supervision labels during the training process.

[0303] In this embodiment, the comprehensive state characteristics of the sample, the sample's operational stage category, the sample's future operational parameters, and the comprehensive benefit value corresponding to the sample's candidate thermal management parameters can be input into the thermal management parameter inference model to be trained to obtain the predicted target thermal management parameters. The thermal management parameter inference model to be trained performs forward calculations based on the input information and outputs the predicted target thermal management parameters.

[0304] In this embodiment, a loss function value can be calculated based on the predicted target thermal management parameters and the sample target thermal management parameters. The loss function value is used to measure the degree of difference between the predicted target thermal management parameters and the sample target thermal management parameters. The specific calculation method of the loss function value can refer to the loss function defined for the thermal management parameter inference model in this application.

[0305] In this embodiment, the parameters of the thermal management parameter inference model to be trained can be adjusted based on the loss function value until the preset training stop condition is met, thus obtaining the trained thermal management parameter inference model.

[0306] The training process can employ gradient descent and its variants as optimization algorithms. Gradients are calculated and model parameters are updated via backpropagation, gradually reducing the loss function value. Preset training stopping conditions can include any of the following: the loss function value is less than a preset threshold, the number of training epochs reaches a preset maximum, or the validation set accuracy no longer improves after a preset number of consecutive training iterations.

[0307] The training process ends when the preset training termination condition is met, and the current model parameters are fixed to the parameters of the trained thermal management parameter inference model. This trained thermal management parameter inference model is used for forward inference during vehicle online operation to output the target thermal management parameters.

[0308] In practical implementation, the loss function value of the thermal management parameter inference model can be set in the following form: During the training phase, a teacher planner is used to generate supervised labels, and a student policy network learns to fit the labels.

[0309] For the discrete mode portion, cross-entropy loss is used:

[0310] in, This represents the cross-entropy loss of the discrete mode portion. This represents the thermal management mode of the k-th sample at time t, and its value is 0 or 1. Let M represent the probability of the k-th thermal management mode at time t predicted by the pattern recognition module. M represents the total number of thermal management mode categories.

[0311] This loss function is used to measure the difference between the pattern probability distribution predicted by the pattern recognition module and the sample thermal management pattern. During training, minimizing the value of this loss function makes the pattern distribution predicted by the pattern recognition module closer to the sample thermal management pattern.

[0312] For the continuous budget portion, the mean squared error loss is used:

[0313] in, This represents the mean squared error loss of the parameter generation module. and These represent the lower limit of the target temperature of the sample battery at time t and the lower limit of the target temperature of the battery predicted by the parameter generation module at time t, respectively. and and represent the upper limit of the target battery temperature at time t and the upper limit of the target battery temperature at time t predicted by the parameter generation module, respectively. and These represent the upper limit of the sample heating power at time t and the upper limit of the heating power predicted by the parameter generation module at time t, respectively. and These represent the upper limit of the sample cooling power at time t and the upper limit of the cooling power predicted by the parameter generation module at time t, respectively. and These represent the upper limit of the sample pump flow rate at time t and the upper limit of the pump flow rate predicted by the parameter generation module at time t, respectively.

[0314] This loss function is used to measure the difference between the continuous programming quantity predicted by the parameter generation module and the sample continuous programming quantity. During training, minimizing the value of this loss function makes the continuous parameters predicted by the parameter generation module closer to the sample continuous programming quantity.

[0315] Furthermore, to prevent abrupt changes in output between adjacent time steps, a smoothing regularization term is introduced:

[0316] in, This represents a smoothing regularization term, used to suppress abrupt changes in output parameters between adjacent time steps. This represents the continuous planning quantity predicted by the parameter generation module at time t. This represents the continuous planning quantity predicted by the parameter generation module at time t-1. This regularization term is used to constrain the continuous planning quantity outputs at adjacent times from changing drastically, ensuring that the output results remain smooth and continuous in the time dimension.

[0317] The total loss function value of the thermal management parameter inference model is defined as follows:

[0318] in, This represents the total loss function value of the thermal management parameter inference model. This represents the weighting coefficient for the mean squared error loss in continuous planning. These represent the weight coefficients of the smoothing regularization term. The total loss function value is equal to the cross-entropy loss of the pattern recognition module. plus weighting coefficients Multiply by the mean squared error loss of the parameter generation module plus weighting coefficients Multiply by smoothing regularization term .

[0319] During training, the model parameters are optimized by minimizing the total loss function value, so that the model can accurately fit the sample labels while maintaining the smoothness of the output in the time dimension.

[0320] This application constructs a target state parameter by concatenating comprehensive state characteristics, operating stage categories, future operating parameters, and the comprehensive benefit values ​​corresponding to each candidate thermal management parameter. Through feature extraction, stage-gated weighting, and parameter generation using a thermal management parameter inference model, it achieves comprehensive processing of multi-source information and stage-adaptive thermal management parameter generation. This solves the problem in existing technologies where thermal management decisions rely on a single threshold trigger and cannot integrate multi-source information for comprehensive decision-making. It enables the target thermal management parameter to make multi-objective comprehensive decisions based on vehicle status, operating stage, future predictions, and benefit assessment, thereby improving the accuracy and adaptability of thermal management decisions.

[0321] Optionally, the step of obtaining training samples includes: S1091, Obtain historical driving data and historical charging data of the sample vehicle; S1092, For any historical moment, determine the comprehensive state characteristics of the sample, the category of the sample's operating stage, the sample's future operating parameters, and the comprehensive benefit value of the sample's candidate thermal management parameters based on historical driving data and historical charging data. S1093, input the comprehensive state characteristics of the sample corresponding to the historical moment, the sample operation stage category, the sample future operation parameters, and the sample comprehensive benefit value corresponding to the sample candidate thermal management parameters into the thermal management label prediction model to obtain the sample target thermal management parameters corresponding to the historical moment. S1094 uses the comprehensive state characteristics of samples corresponding to historical moments, the category of sample operation stages, the future operation parameters of samples, the comprehensive benefit value of samples corresponding to candidate thermal management parameters, and the target thermal management parameters of samples corresponding to historical moments as a set of training samples.

[0322] In this embodiment of the application, historical driving data and historical charging data of the sample vehicle can be obtained.

[0323] Historical driving data includes vehicle speed, acceleration, cumulative mileage, drive power, regenerative braking power, road gradient and other operating status information during past driving, as well as battery status information such as battery state of charge, battery temperature, coolant temperature and other battery status information.

[0324] Historical charging data includes information about the charging process, such as charging power, charging time, charging station power, and ambient temperature.

[0325] Historical driving data and historical charging data together form the data foundation of the training samples, covering complete information on both the driving and charging phases.

[0326] In the embodiments of this application, for any historical moment, the comprehensive state characteristics of the sample, the category of the sample operation stage, the sample future operation parameters, and the sample candidate thermal management parameters corresponding to that historical moment can be determined based on historical driving data and historical charging data.

[0327] The sample comprehensive state features are the comprehensive state features of the vehicles at that moment extracted from historical data.

[0328] The sample operation stage category is the actual operation stage category of the vehicle at that moment.

[0329] The sample's future operating parameters are obtained from historical data, taking the current moment as the starting point for future operating parameters.

[0330] The sample comprehensive benefit value corresponding to the sample candidate thermal management parameters is the comprehensive benefit value of each candidate scheme calculated by the thermal management benefit prediction model.

[0331] In this embodiment, the comprehensive state characteristics of the sample corresponding to a historical moment, the sample operation stage category, the sample future operation parameters, and the sample comprehensive benefit value corresponding to the sample candidate thermal management parameters can be input into the thermal management label prediction model to obtain the sample target thermal management parameters corresponding to the historical moment.

[0332] The thermal management label prediction model calculates the comprehensive cost value of each candidate thermal management parameter based on the input information, and selects the candidate thermal management parameter with the smallest comprehensive cost value as the sample target thermal management parameter for that historical moment. The sample target thermal management parameter is the optimal thermal management decision result output by the thermal management label prediction model, and serves as the supervision label during the training process.

[0333] In this embodiment of the application, the comprehensive state features of the sample corresponding to the historical moment, the sample operation stage category, the sample future operation parameters, the sample comprehensive benefit value corresponding to the sample candidate thermal management parameters, and the sample target thermal management parameters corresponding to the historical moment can be used as a set of training samples.

[0334] The input to the training samples includes the sample's comprehensive state features, the sample's operational stage category, the sample's future operational parameters, and the sample's comprehensive benefit value corresponding to the candidate thermal management parameters. The label part is the sample's target thermal management parameter. Multiple sets of training samples corresponding to multiple historical moments constitute the training sample set, which is used to train the thermal management parameter inference model.

[0335] This application obtains historical driving and charging data of sample vehicles, determines the comprehensive state characteristics, operational stage category, future operational parameters, and comprehensive benefit value of the sample for any given historical moment, and obtains the target thermal management parameters of the sample through a thermal management label prediction model to form training samples. This achieves automatic generation of training samples and solves the problem that the training samples of thermal management parameter inference models in the prior art rely on manual annotation and have high acquisition costs. It provides supervision labels covering the entire driving and charging process for model training.

[0336] Optionally, the training samples for the thermal management parameter inference model include the comprehensive state features of the samples at each historical moment, the sample operation stage category, the sample future operation parameters, the comprehensive benefit value of the sample corresponding to the candidate thermal management parameters, and the sample target thermal management parameters as training labels. The sample target thermal management parameters are predicted by a thermal management label prediction model. The thermal management label prediction model includes a cost calculation module.

[0337] In this embodiment, the thermal management label prediction model is used to select the optimal solution from several candidate thermal management parameters as the sample target thermal management parameter.

[0338] The cost calculation module is used to calculate the comprehensive cost value of each candidate scheme based on the comprehensive state characteristics of the sample, the category of the sample operation stage, the future operation parameters of the sample, the sample candidate thermal management parameters, and the comprehensive benefit value of the sample candidate thermal management parameters. The smaller the comprehensive cost value, the better the candidate scheme performs under the multi-objective comprehensive evaluation.

[0339] For any given historical moment, the target thermal management parameters for the sample are predicted using the following method: S1101, input the sample comprehensive state characteristics corresponding to the historical moment, the sample operation stage category, the sample future operation parameters, the sample candidate thermal management parameters and the sample comprehensive benefit value corresponding to the sample candidate thermal management parameters into the cost calculation module to calculate the cost and obtain the comprehensive cost value corresponding to each sample candidate thermal management parameter; S1102, From a number of sample candidate thermal management parameters, select the sample candidate thermal management parameter whose comprehensive cost value is less than the comprehensive cost value of other sample candidate thermal management parameters, and use it as the sample target thermal management parameter for the historical moment.

[0340] In this embodiment, the comprehensive state characteristics of the sample corresponding to a historical moment, the sample operation stage category, the sample future operation parameters, the sample candidate thermal management parameters, and the sample comprehensive benefit value corresponding to the sample candidate thermal management parameters can be input into the cost calculation module for cost calculation to obtain the comprehensive cost value corresponding to each sample candidate thermal management parameter.

[0341] The comprehensive cost is composed of the negative value of the comprehensive benefit plus the temperature over-limit penalty, the actuator change rate penalty, and the safety constraint penalty. The temperature over-limit penalty is used to constrain the battery temperature from not exceeding the safe range, the actuator change rate penalty is used to constrain the actuator output from drastic changes, and the safety constraint penalty is used to ensure the safe operation of the system.

[0342] The smaller the overall cost, the better the candidate solution performs in terms of multi-objective comprehensive performance under the premise of meeting the constraints.

[0343] In this embodiment, a sample candidate thermal management parameter whose comprehensive cost value is less than that of other sample candidate thermal management parameters can be selected from a number of sample candidate thermal management parameters as the sample target thermal management parameter for the historical moment.

[0344] By comparing the comprehensive cost of all candidate solutions, the candidate solution with the smallest comprehensive cost is selected as the optimal solution. This optimal solution is the output supervision label, which is the sample target thermal management parameter.

[0345] In its implementation, the prediction process of the thermal management label prediction model is as follows: The instantaneous cost function can be defined from the preceding comprehensive benefit value:

[0346] in, This represents the immediate cost function, which is equal to the total return value. The negative value indicates that the higher the overall benefit, the lower the cost.

[0347] To facilitate constraint processing, constraint penalties are further introduced.

[0348] in, This represents the overall cost after introducing constraints and penalties. This represents the overall return value. Indicates temperature exceeding the limit penalty item The weighting coefficients, Indicates the penalty term for the rate of change of the actuator. The weighting coefficients, Indicates safety constraint penalty items The three weighting coefficients are used to balance the relative importance of each penalty in the overall cost.

[0349] Define the battery temperature out-of-bounds penalty as follows:

[0350] This indicates a penalty for exceeding the temperature limit. Indicates the adoption of candidate thermal management parameters The battery temperature was then predicted by the thermal state prediction sub-model. This indicates the upper limit of the battery's safe temperature range. This indicates the lower limit of the battery's safe temperature range. This means that when the predicted battery temperature exceeds the upper limit of the safe temperature, the excess is treated as a temperature over-limit penalty; when the predicted battery temperature does not exceed the upper limit of the safe temperature, the penalty is zero. This means that when the predicted battery temperature is below the lower limit of the safe lower temperature, the portion below this level is treated as a temperature over-limit penalty; when the predicted battery temperature is not below the lower limit of the safe lower temperature, the penalty is zero.

[0351] The temperature over-limit penalty is equal to the sum of the square of the portion of the battery temperature exceeding the upper safety limit and the square of the portion falling below the lower safety limit. It is used to constrain thermal management parameters so that the battery temperature does not exceed the safe range.

[0352] Define the actuator change rate penalty term as

[0353] This indicates the executor smoothing penalty term. The penalty coefficient represents the rate of change of heating power, used to suppress drastic changes in the upper limit of heating power between adjacent time periods. The penalty coefficient represents the rate of change of cooling power, used to suppress drastic changes in the upper limit of cooling power between adjacent time periods. The penalty coefficient represents the rate of change of pump flow rate, used to suppress drastic changes in the upper limit of pump flow rate between adjacent time points. and These represent the upper limit values ​​of heating power at the current moment and the previous moment, respectively. and These represent the maximum cooling power values ​​at the current and previous moments, respectively. and These represent the upper limit values ​​of the pump flow rate at the current time and the previous time, respectively.

[0354] The actuator smoothing penalty term is used to reduce drastic fluctuations in the flow of compressors, heaters, and pumps, and to avoid increased energy consumption and mechanical wear caused by frequent start-ups and shutdowns and large-scale adjustments of the actuators.

[0355] Safety constraint penalties are used to ensure that thermal management parameters do not lead the system into an unsafe state.

[0356] When the battery temperature exceeds the upper limit of the safe temperature range or falls below the lower limit of the safe temperature range, the safety constraint penalty item increases the penalty value according to the degree of deviation. The greater the deviation between the battery temperature and the safe boundary, the greater the penalty value.

[0357] When the upper limit of heating power, upper limit of cooling power, or upper limit of pump flow exceeds the physical limit of the actuator, the safety constraint penalty item increases the penalty value according to the degree of exceedance. The greater the exceedance of the actuator's limit, the greater the penalty value.

[0358] When the battery state of charge is lower than the preset discharge cutoff threshold, the safety constraint penalty item increases the penalty value according to the degree of lower charge. The greater the deviation between the state of charge and the discharge cutoff threshold, the greater the penalty value.

[0359] The weighted sum of the above penalties constitutes the total value of the security constraint penalty item. The weight coefficients of each penalty are preset according to the security level and the severity of the constraint. When the system state is within the security boundary, the total value of the security constraint penalty item is zero.

[0360] The aforementioned dual-principal-axis prediction model, i.e., the runtime parameter prediction model, is used as the state transition approximation function:

[0361] It represents the overall state characteristics at the next moment. ( ) represents the state transition function, which is based on the operating parameter prediction model to determine the comprehensive state characteristics at the current moment. and candidate thermal management parameters Mapped to the comprehensive state at the next moment .

[0362] Then, soft dynamic programming is used to solve the problem, and the action-value function is defined as follows:

[0363] in, The state-action value function is equal to the total cost at the current moment. With the soft-value function at the next time step The sum of expected values.

[0364] in For soft-value functions, the soft-value function is defined as follows:

[0365] in, This represents a soft-valued function. The softening temperature parameter controls the smoothness of candidate thermal management parameter selection in soft dynamic programming. U represents the set of candidate thermal management parameters, containing all candidate thermal management parameters that satisfy actuator physical constraints and battery safety boundary constraints. Represents the state action value function. This represents the overall state characteristics at the current moment. This represents the candidate thermal management parameters.

[0366] when When the values ​​are small, soft dynamic programming approaches standard optimal dynamic programming, and the selection of candidate thermal management parameters tends to choose the candidate thermal management parameter with the smallest state-action value function. When When the value is large, soft dynamic programming smooths the differences in the state-action value functions of each candidate thermal management parameter, which helps to obtain a smoother distribution of teacher strategies and makes the thermal management parameter inference model easier to learn.

[0367] The target thermal management parameters for the sample can be expressed as:

[0368] This represents the target thermal management parameters for the sample. `argmin()` represents the function that selects the state action value. The smallest candidate thermal management parameter is selected as the optimal thermal management parameter.

[0369] Therefore, the offline teacher planner outputs a set of target thermal management parameters:

[0370] in, This indicates the target thermal management parameters for the sample. This indicates the sample thermal management mode. This indicates the lower limit of the target temperature for the sample battery. This indicates the upper limit of the target temperature for the sample battery. This indicates the upper limit of the sample heating power. This indicates the upper limit of the sample cooling power. This represents the upper limit of the sample pump flow rate. The teacher planner selects the candidate scheme with the smallest state-action value function from the set of candidate thermal management parameters and uses it as the sample target thermal management parameter at that moment, i.e., the supervision label in distillation training. This supervision label is used to train the thermal management parameter inference model, making the target thermal management parameter output by the model closer to the sample target thermal management parameter output by the teacher planner.

[0371] In practice, the teacher planner, also known as the hot management label prediction model, is used to generate the optimal labels during the offline phase.

[0372] The teacher planner uses a dual-principal prediction model as the state transition approximation function and a unified reward function as the optimization objective. It solves the optimal sequence decision problem with temperature, actuator, and safety constraints using soft dynamic programming within the long programming time domain. The teacher planner enumerates all candidate thermal management parameters, calculates the comprehensive cost of each candidate scheme through a cost calculation module, and selects the candidate scheme with the minimum comprehensive cost as the optimal planning result at that moment. It outputs the optimal thermal management mode, the optimal lower and upper limits of the optimal battery target temperature, the optimal upper limit of heating power, the optimal upper limit of cooling power, and the optimal upper limit of pump flow rate. These optimal planning results serve as the sample target thermal management parameters, constituting the supervision labels for the training samples. Due to its high computational cost, the teacher planner does not run on the onboard controller; it is only used to generate labels for training samples during the offline training phase.

[0373] The student policy network, also known as the thermal management parameter inference model, learns to fit the output of the teacher planner through distillation during the offline training phase.

[0374] During offline training, the student policy network is trained using the sample target thermal management parameters output by the teacher planner as supervised labels. The student policy network takes the sample's comprehensive state features, sample's operational stage category, sample's future operational parameters, and the sample's comprehensive reward value corresponding to the candidate thermal management parameters as input. Through forward propagation, it obtains the predicted target thermal management parameters. The network parameters are optimized by calculating the loss function value between the predicted target thermal management parameters and the sample target thermal management parameters. The loss function includes the cross-entropy loss of the pattern recognition module, the mean squared error loss of the parameter generation module, and a smoothing regularization term for adjacent time steps, ensuring that the network accurately fits the teacher labels while maintaining the smoothness of the output in the time dimension. After training, the student policy network is deployed on the vehicle controller. During online operation, it takes the current comprehensive state features, operational stage category, future operational parameters, and the comprehensive reward value corresponding to each candidate thermal management parameter as input. Through a single forward propagation, it directly outputs the target thermal management parameters without needing to iteratively solve the optimization problem online, thus meeting the real-time requirements of the vehicle platform while ensuring planning effectiveness.

[0375] One embodiment of this application also provides a vehicle that may include a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the method described above.

[0376] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0377] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, embodiments of this application can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of this application can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0378] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0379] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0380] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0381] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other modifications and updates to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all modifications and updates falling within the scope of the embodiments of the present application.

[0382] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the aforementioned element.

[0383] The above provides a detailed description of a vehicle thermal management method and a vehicle. Specific examples have been used 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 method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A thermal management method for a vehicle, characterized in that, The method includes: Acquire the comprehensive state characteristics of the vehicle and the sequence of route characteristics ahead of the vehicle within a preset time period; The comprehensive state features are input into the trained operation phase identification model for identification to obtain the operation phase category of the vehicle; The comprehensive state features, the operation stage category, and the route feature sequence are input into a trained operation parameter prediction model to predict the future operation parameters of the vehicle. Several candidate thermal management parameters are obtained, and each candidate thermal management parameter and the future operating parameters are input into the thermal management benefit prediction model for prediction to obtain the comprehensive benefit value corresponding to each candidate thermal management parameter. The comprehensive state characteristics, the operational stage category, the future operational parameters, and the comprehensive benefit value corresponding to each of the candidate thermal management parameters are input into the trained thermal management parameter inference model for inference to obtain the target thermal management parameters of the vehicle. Thermal management is performed on the vehicle according to the target thermal management parameters.

2. The method according to claim 1, characterized in that, The operational phase identification model includes a first feature extraction module, a bidirectional loop module, and a cross-attention fusion module; The step of inputting the comprehensive state features into a trained operation phase identification model to identify the vehicle's operation phase category includes: The comprehensive state features are input into the first feature extraction module for feature extraction to obtain state change features; The state change features are input into the bidirectional loop module for time series modeling to obtain state time series features; Obtain the navigation power replenishment event features from the comprehensive state features; input the navigation power replenishment event features and the state temporal features into the cross-attention fusion module for fusion to obtain the state fusion features; The vehicle's operating phase category is determined based on the state fusion characteristics.

3. The method according to claim 1, characterized in that, The operating parameter prediction model includes a distance axis prediction sub-model, a charge state axis prediction sub-model, and a thermal state prediction model; The steps of inputting the comprehensive state features, the operational phase category, and the route feature sequence into a trained operational parameter prediction model to predict the future operational parameters of the vehicle include: The route feature sequence, the comprehensive state features, and the operation stage category are input into the distance axis prediction sub-model for prediction to obtain the predicted vehicle speed, predicted driving power, and predicted regenerative power corresponding to each position in the future driving process; wherein, each position in the future driving process is obtained by discretizing the future path in front of the vehicle according to a preset distance interval; The current state of charge, battery temperature, coolant temperature, charging pile power, and ambient temperature are obtained from the comprehensive state characteristics. The current state of charge, battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category are input into the state of charge axis prediction sub-model for prediction, to obtain the predicted charging power and cumulative charging time corresponding to each state of charge interval in the future charging process; wherein, each state of charge interval in the future charging process is obtained by discretizing the future charging process according to a preset state of charge interval; The predicted vehicle speed, the predicted driving power, the predicted charging power, and the comprehensive state features are input into the thermal state prediction sub-model for prediction, so as to obtain the future battery temperature corresponding to each position during the future driving process. The predicted vehicle speed, the predicted drive power, the predicted regenerative braking power, the predicted charging power, the cumulative charging time, and the future battery temperature are used as the future operating parameters.

4. The method according to claim 3, characterized in that, The distance axis prediction sub-model includes a second feature extraction module, a path feature extraction module, and a first state gating module; The steps of inputting the route feature sequence, the comprehensive state features, and the operation phase category into the distance axis prediction sub-model for prediction to obtain the predicted vehicle speed, predicted driving power, and predicted regenerative power corresponding to each position during future driving include: The route feature sequence is input into the second feature extraction module for feature extraction to obtain local route features; The local features of the route are input into the path feature extraction module for feature extraction to obtain the global features of the path. The comprehensive state features and the operation stage category are input into the first state gating module for weight calculation to obtain the path state weight; The global features of the path are weighted according to the path state weights to obtain the target path state features; The predicted vehicle speed at each position during the future driving process is determined based on the target path state characteristics. The road gradient at each location during future travel is determined based on the route feature sequence. Based on the predicted vehicle speed and road gradient at each location, the predicted driving power at each location during future driving is determined. Based on the predicted driving power, the predicted recovery power at each location during future driving is calculated.

5. The method according to claim 3, characterized in that, The charge state axis prediction sub-model includes a first feature mapping module and a second state gating module; The steps of inputting the current state of charge, battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category into the state of charge axis prediction sub-model to predict the predicted charging power and cumulative charging time for each state of charge interval during future charging include: The current state of charge is input into the first feature mapping module for feature mapping to obtain the state of charge mapping feature. The battery temperature, coolant temperature, charging pile power, ambient temperature, and operating stage category are input into the second state gating module for weight calculation to obtain the charging state weight; The state of charge mapping features are weighted according to the state of charge weights to obtain the target state of charge features. Based on the target charging state characteristics, the predicted charging power corresponding to each state of charge interval during the future charging process is determined. Based on the predicted charging power corresponding to each state of charge interval, the interval charging time corresponding to each state of charge interval is determined. Based on the charging time corresponding to each state of charge interval, the cumulative charging time at the end of each state of charge interval is determined.

6. The method according to claim 3, characterized in that, The thermal state prediction sub-model includes a residual correction module; The steps of inputting the predicted vehicle speed, the predicted driving power, the predicted charging power, and the comprehensive state features into the thermal state prediction sub-model to predict the future battery temperature at each location during future driving include: Obtain the thermal management actuator status and the current battery temperature and current coolant temperature from the comprehensive status characteristics; Based on the predicted driving power and the predicted charging power corresponding to each position during future driving, determine the battery heat generation corresponding to each position during future driving. Based on the current battery temperature, the current coolant temperature, the heat generated by the battery at each location during future driving, the state of the thermal management actuator, and the preset thermal balance equation, the theoretical battery temperature and theoretical coolant temperature at each location during future driving are determined. The comprehensive state characteristics, the operating stage category, the predicted vehicle speed, the predicted driving power, the predicted charging power, the theoretical battery temperature and the theoretical coolant temperature corresponding to each position during future driving are input into the residual correction module to obtain the battery temperature correction amount corresponding to each position during future driving. The future battery temperature at each location during future driving is determined based on the theoretical battery temperature and the battery temperature correction amount.

7. The method according to claim 1, characterized in that, The thermal management revenue prediction model includes a revenue calculation module; The step of inputting each of the candidate thermal management parameters and the future operating parameters into the thermal management benefit prediction model for prediction, and obtaining the comprehensive benefit value corresponding to each of the candidate thermal management parameters, includes: Based on the candidate thermal management parameters and the future operating parameters, the revenue calculation module calculates the fast charging time revenue, recycling revenue, power revenue and thermal management energy consumption cost corresponding to each candidate thermal management parameter. Based on the fast charging time benefit, the recycling benefit, the power benefit, and the thermal management energy consumption cost corresponding to each candidate thermal management parameter, the comprehensive benefit value corresponding to each candidate thermal management parameter is determined.

8. The method according to claim 1, characterized in that, The thermal management parameter inference model includes a second feature mapping module, a stage gating module, a pattern recognition module, and a parameter generation module; The step of inputting the comprehensive state characteristics, the operational stage category, the future operational parameters, and the comprehensive benefit value corresponding to each of the candidate thermal management parameters into a trained thermal management parameter inference model to obtain the target thermal management parameters of the vehicle includes: The comprehensive state characteristics, the operational stage category, the future operational parameters, and the comprehensive benefit values ​​corresponding to each of the candidate thermal management parameters are concatenated to form the target state parameters. The target state parameters are input into the second feature mapping module for feature mapping to obtain fused state features; The fusion state features and the operation stage category are input into the stage gating module for weight calculation to obtain the stage state weight; The fused state features are weighted according to the stage state weights to obtain the stage fusion features; The stage fusion features are input into the pattern recognition module to obtain the thermal management mode of the vehicle; The stage fusion features are input into the parameter generation module to obtain the vehicle's target temperature lower limit, target temperature upper limit, heating power upper limit, cooling power upper limit, and pump flow rate upper limit. The thermal management mode, the lower limit of the target battery temperature, the upper limit of the target battery temperature, the upper limit of the heating power, the upper limit of the cooling power, and the upper limit of the pump flow rate are used as the target thermal management parameters at the current moment.

9. The method according to claim 8, characterized in that, The training samples of the thermal management parameter inference model include the comprehensive state features of the samples at each historical moment, the sample operation stage category, the sample future operation parameters, the comprehensive benefit value of the sample corresponding to the candidate thermal management parameters, and the sample target thermal management parameters as training labels; the sample target thermal management parameters are predicted by a thermal management label prediction model; the thermal management label prediction model includes a cost calculation module: For any of the aforementioned historical moments, the target thermal management parameters for the sample are predicted using the following method: The sample comprehensive state characteristics corresponding to the historical moment, the sample operation stage category, the sample future operation parameters, the sample candidate thermal management parameters, and the sample comprehensive benefit value corresponding to the sample candidate thermal management parameters are input into the cost calculation module to calculate the cost and obtain the comprehensive cost value corresponding to each sample candidate thermal management parameter. From a number of candidate thermal management parameters, the candidate thermal management parameter whose comprehensive cost value is less than that of other candidate thermal management parameters is selected as the target thermal management parameter for the historical moment.

10. A vehicle, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method as described in any one of claims 1-9.

Images