Energy management method and device for extended-range hybrid vehicle

Abstract

The application relates to the technical field of energy management, in particular to an energy management method and device for a range-extender hybrid vehicle, wherein the method comprises the following steps: extracting static features according to running information and environmental information of a plurality of range-extender hybrid vehicles at at least one historical moment, extracting time sequence features according to road condition information and predicted average vehicle speed of a target driving section after the at least one historical moment, and constructing a training data set according to the static features, the time sequence features and actual energy output information; based on the training data set, generating a supervision learning label of an initial energy management model; and using the training data set, the supervision learning label, a target loss function and a target optimizer to train the initial energy management model in an end-to-end mode, so as to construct a target energy management model. Therefore, the problems that an energy management method in the related art has certain limitations under complex and changeable road conditions, information transmission efficiency is low, and optimization is difficult to be integrated are solved.

Description

Technical Field

[0001] This application relates to the field of energy management technology, and in particular to an energy management method and device for a range-extended hybrid vehicle. Background Technology

[0002] In related technologies, the energy management strategies of range-extended hybrid vehicles mostly adopt rule-based or model-based prediction methods. These methods usually implement the energy consumption prediction model and the energy management control strategy separately. That is, the energy consumption of the range-extended hybrid vehicle is predicted first, and then the energy of the range-extended hybrid vehicle is managed on the basis of this prediction.

[0003] However, among related technologies, rule-based methods are difficult to adapt to dynamic environments, while model-based prediction control accuracy is limited by model quality and has certain limitations under complex and ever-changing road conditions. Furthermore, the separate implementation of energy consumption prediction models and energy management control strategies can easily lead to low information transmission efficiency and make it difficult to integrate the entire process optimization, which urgently needs to be addressed. Summary of the Invention

[0004] This application provides an energy management method and device for range-extended hybrid vehicles to solve the problems in related technologies, such as rule-based methods being difficult to adapt to dynamic environments, model-based prediction control accuracy being limited by model quality and having certain limitations under complex and changing road conditions, and the separate implementation of energy consumption prediction models and energy management control strategies leading to low information transmission efficiency and difficulty in integrating the whole process optimization.

[0005] The first aspect of this application provides an energy management method for a range-extended hybrid electric vehicle (REEV), applied in the model building stage, comprising the following steps: collecting operating information, environmental information, and actual energy output information of multiple REEVs at at least one historical moment, as well as road condition information and predicted average vehicle speed of the multiple REEVs on a target driving segment after the at least one historical moment; extracting static features corresponding to the multiple REEVs at the at least one historical moment based on the operating information and the environmental information, extracting temporal features corresponding to the multiple REEVs on the target driving segment after the at least one historical moment based on the road condition information and the predicted average vehicle speed, and constructing a training dataset based on the static features, the temporal features, and the actual energy output information; generating motor torque distribution results, engine torque distribution results, and engine start-stop control results of the multiple REEVs at the at least one historical moment based on the training dataset, to generate supervised learning labels for an initial energy management model; and training the initial energy management model end-to-end using the training dataset, the supervised learning labels, a target loss function, and a target optimizer to construct a target energy management model for managing the energy of a target REEV.

[0006] Through the above-mentioned technical means, the embodiments of this application can use a large dataset, supervised learning labels, target loss function and target optimizer to train and optimize the initial energy management model end-to-end to obtain the target energy management model. The target energy management model can directly output the optimal power source torque distribution and engine start-stop control from information input to output, eliminating the complex modular design and information transmission process in traditional methods, reducing the error and computational burden in intermediate links, greatly improving the efficiency of energy management, and helping to achieve full-process optimization.

[0007] Optionally, in one embodiment of this application, before training the initial energy management model end-to-end using the training dataset, the supervised learning labels, the target loss function, and the target optimizer, the method further includes: determining the energy management optimization target corresponding to the plurality of range-extended hybrid vehicles based on the energy management requirements corresponding to the plurality of range-extended hybrid vehicles; obtaining the optimization result according to the energy management optimization target; and determining the target loss function according to the optimization result.

[0008] Through the above-mentioned technical means, the embodiments of this application can determine the energy management optimization target according to the energy management needs of the vehicle, and then determine the target loss function of the model. This can transform the energy management target into a quantifiable and optimizable mathematical form, providing a clear optimization direction for model training, enabling the model to optimize energy management by minimizing the loss function, and improving performance indicators such as energy utilization efficiency.

[0009] Optionally, in one embodiment of this application, before training the initial energy management model end-to-end using the training dataset, the supervised learning labels, the target loss function, and the target optimizer, the method further includes: constructing an energy consumption loss function, a power battery balance loss function, and a smoothness loss function for the target loss function; and determining the target loss function based on the energy consumption loss function, the power battery balance loss function, and the smoothness loss function.

[0010] Through the above-mentioned technical means, the embodiments of this application can generate a target loss function by comprehensively considering multiple factors, which can simultaneously take into account different dimensions of objectives, such as ensuring driving smoothness while minimizing vehicle energy consumption. Moreover, it helps different loss functions to constrain the target energy management model from different angles, reducing the risk of overfitting, making the target energy management model more stable on complex and variable data, and enhancing the robustness and generalization ability of the target energy management model.

[0011] Optionally, in one embodiment of this application, the method further includes: optimizing a preset initial energy management model using a learning rate decay strategy and an early stopping mechanism to construct the initial energy management model.

[0012] Through the above-mentioned technical means, the embodiments of this application can introduce a learning rate decay strategy and an early stopping mechanism to optimize the preset initial energy management model, avoid overfitting of the initial energy management model, and ensure that the target energy management model obtained after training can adapt to new and unseen test data, thereby ensuring the performance of the target energy management model in actual application scenarios.

[0013] A second aspect of this application provides an energy management method for a range-extended hybrid electric vehicle (REEV), applied in the model application stage. The method includes the following steps: collecting real-time operating information and real-time environmental information of the target REEV, and collecting road condition information and predicted average speed of the target REEV on a future target driving segment; inputting the real-time operating information, real-time environmental information, road condition information, and predicted average speed into a pre-constructed target energy management model; extracting the real-time static features and temporal features of the target REEV on the future target driving segment; fusing the static features and temporal features to obtain fused features; outputting real-time motor torque distribution results, real-time engine torque distribution results, and real-time engine start-stop control results of the target REEV based on the fused features; and generating corresponding control commands based on the real-time motor torque distribution results, real-time engine torque distribution results, and real-time engine start-stop control results to control the target REEV for energy management.

[0014] Through the above-mentioned technical means, the embodiments of this application can utilize a trained target energy management model to achieve optimal power source torque distribution and engine start-stop control directly from information input to output. This eliminates the complex modular design and information transmission process in traditional methods, reduces errors and computational burdens in intermediate links, significantly improves the efficiency of energy management, and helps to achieve full-process optimization.

[0015] A third aspect of this application provides an energy management device for a range-extended hybrid electric vehicle (REEV), applied in the model building stage, comprising: a data acquisition module for acquiring operating information, environmental information, and actual energy output information of multiple REEVs at at least one historical moment, as well as road condition information and predicted average vehicle speed of the multiple REEVs on a target driving segment after the at least one historical moment; and an extraction module for extracting static features of the multiple REEVs corresponding to the at least one historical moment based on the operating information and the environmental information, and extracting static features of the multiple REEVs corresponding to the at least one historical moment based on the road condition information and the predicted average vehicle speed. The system uses the following modules to construct a training dataset: a static feature corresponding to the target driving segment, a time-series feature, and actual energy output information; a generation module generates motor torque distribution results, engine torque distribution results, and engine start-stop control results for the multiple range-extended hybrid vehicles at at least one historical moment, based on the training dataset, to generate supervised learning labels for the initial energy management model; and a first construction module trains the initial energy management model end-to-end using the training dataset, the supervised learning labels, the objective loss function, and the objective optimizer, to construct a target energy management model for managing the energy of the target range-extended hybrid vehicles.

[0016] Through the above-mentioned technical means, the embodiments of this application can use a large dataset, supervised learning labels, target loss function and target optimizer to train and optimize the initial energy management model end-to-end to obtain the target energy management model. The target energy management model can directly output the optimal power source torque distribution and engine start-stop control from information input to output, eliminating the complex modular design and information transmission process in traditional methods, reducing the error and computational burden in intermediate links, greatly improving the efficiency of energy management, and helping to achieve full-process optimization.

[0017] Optionally, in one embodiment of this application, it further includes: a first determining module, configured to determine the energy management optimization objective of the plurality of range-extended hybrid vehicles based on the energy management requirements of the plurality of range-extended hybrid vehicles before training the initial energy management model end-to-end using the training dataset, the supervised learning labels, the target loss function, and the target optimizer; and a second determining module, configured to obtain an optimization result based on the energy management optimization objective, and determine the target loss function based on the optimization result.

[0018] Through the above-mentioned technical means, the embodiments of this application can determine the energy management optimization target according to the energy management needs of the vehicle, and then determine the target loss function of the model. This can transform the energy management target into a quantifiable and optimizable mathematical form, providing a clear optimization direction for model training, enabling the model to optimize energy management by minimizing the loss function, and improving performance indicators such as energy utilization efficiency.

[0019] Optionally, in one embodiment of this application, it further includes: a second construction module, configured to construct an energy consumption loss function, a power battery balance loss function, and a smoothness loss function of the target loss function before training the initial energy management model end-to-end using the training dataset, the supervised learning labels, the target loss function, and the target optimizer; and a third determination module, configured to determine the target loss function based on the energy consumption loss function, the power battery balance loss function, and the smoothness loss function.

[0020] Through the above-mentioned technical means, the embodiments of this application can generate a target loss function by comprehensively considering multiple factors, which can simultaneously take into account different dimensions of objectives, such as ensuring driving smoothness while minimizing vehicle energy consumption. Moreover, it helps different loss functions to constrain the target energy management model from different angles, reducing the risk of overfitting, making the target energy management model more stable on complex and variable data, and enhancing the robustness and generalization ability of the target energy management model.

[0021] Optionally, in one embodiment of this application, it further includes: a third construction module, used to optimize a preset initial energy management model using a learning rate decay strategy and an early stopping mechanism, so as to construct the initial energy management model.

[0022] Through the above-mentioned technical means, the embodiments of this application can introduce a learning rate decay strategy and an early stopping mechanism to optimize the preset initial energy management model, avoid overfitting of the initial energy management model, and ensure that the target energy management model obtained after training can adapt to new and unseen test data, thereby ensuring the performance of the target energy management model in actual application scenarios.

[0023] This application provides a fourth aspect of an energy management device for a range-extended hybrid electric vehicle (REEV), applied in the model application stage. The device comprises: a data acquisition module for acquiring real-time operating information and real-time environmental information of the target REEV, and acquiring road condition information and predicted average speed of the target REEV on a future target driving segment; a processing module for inputting the real-time operating information, real-time environmental information, road condition information, and predicted average speed into a pre-constructed target energy management model, extracting real-time static features and temporal features of the target REEV on the future target driving segment, and fusing the static features and temporal features to obtain fused features, and outputting real-time motor torque distribution results, real-time engine torque distribution results, and real-time engine start-stop control results of the target REEV based on the fused features; and a management module for generating corresponding control commands based on the real-time motor torque distribution results, real-time engine torque distribution results, and real-time engine start-stop control results to control the target REEV for energy management.

[0024] Through the above-mentioned technical means, the embodiments of this application can utilize a trained target energy management model to achieve optimal power source torque distribution and engine start-stop control directly from information input to output. This eliminates the complex modular design and information transmission process in traditional methods, reduces errors and computational burdens in intermediate links, significantly improves the efficiency of energy management, and helps to achieve full-process optimization.

[0025] A fifth aspect of this application provides a vehicle, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the energy management method for a range-extended hybrid vehicle as described in the above embodiments.

[0026] A sixth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the energy management method for a range-extended hybrid vehicle as described above.

[0027] A seventh aspect of this application provides a computer program product, including a computer program that, when executed, is used to implement the energy management method for a range-extended hybrid vehicle as described above.

[0028] This application embodiment can collect various current vehicle information and future target driving segment information, as well as predicted average vehicle speed, and directly generate torque distribution results and engine start-stop control results through a trained target energy management model. Then, it generates corresponding control commands to control the target range-extended hybrid vehicle to achieve energy management. Thus, it achieves end-to-end training and optimization of the initial energy management model using a large dataset, supervised learning labels, a target loss function, and a target optimizer to obtain the target energy management model. This allows the target energy management model in this application to directly output the optimal power source torque distribution and engine start-stop control from information input, eliminating the complex modular design and information transmission process of traditional methods, reducing errors and computational burdens in intermediate links, thereby significantly improving the efficiency of energy management and facilitating full-process optimization. Furthermore, while ensuring that the obtained power source torque distribution result is the globally optimal energy management strategy, the target energy management model in this application possesses reasoning capabilities that enable real-time energy management of the target range-extended hybrid vehicle. Faced with complex and changing road conditions and environmental conditions, it can quickly respond and adjust the energy management strategy, demonstrating high practicality while ensuring the robustness and generalization of the target energy management model. This solves the problems in related technologies, such as the difficulty of rule-based methods to adapt to dynamic environments, the limitation of model-based prediction control accuracy due to model quality, the limitations of energy consumption prediction models and energy management control strategies, and the difficulty of integrating the whole process optimization.

[0029] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0030] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0031] Figure 1 This is a flowchart illustrating an embodiment of an end-to-end range-extended hybrid vehicle energy management method according to this application.

[0032] Figure 2 This is a flowchart illustrating an energy management method for a range-extended hybrid vehicle applied in the model building stage, according to an embodiment of this application.

[0033] Figure 3 This is a schematic diagram of a dataset construction and annotation framework according to an embodiment of this application;

[0034] Figure 4 This is a schematic diagram illustrating model training and optimization according to an embodiment of this application;

[0035] Figure 5 This is a schematic diagram of the energy management method device for a range-extended hybrid vehicle applied in the model building stage according to an embodiment of this application;

[0036] Figure 6 A flowchart illustrating an energy management method for a range-extended hybrid vehicle applied in the model application stage, as provided in this application embodiment;

[0037] Figure 7 This is a flowchart illustrating the practical application of a model according to one embodiment of this application;

[0038] Figure 8 This is a schematic diagram of the energy management method device for range-extended hybrid vehicles applied in the model application stage according to an embodiment of this application;

[0039] Figure 9 This is a structural schematic diagram of a vehicle provided according to an embodiment of this application.

[0040] Figure label:

[0041] 10 - Energy management method device for range-extended hybrid vehicles applied in the model building stage: 100 - Acquisition module, 200 - Extraction module, 300 - Generation module and 400 - Construction module; 10 - Energy management method device for range-extended hybrid vehicles applied in the model application stage: 500 - Acquisition module, 600 - Processing module, 700 - Management module; 901 - Memory, 902 - Processor and 903 - Communication interface. Detailed Implementation

[0042] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0043] The energy management method and apparatus for a range-extended hybrid electric vehicle according to embodiments of this application are described below with reference to the accompanying drawings. In view of the related technologies mentioned in the background section, rule-based methods are difficult to adapt to dynamic environments, while model-based prediction control accuracy is limited by model quality and has certain limitations in complex and changing road conditions. Furthermore, implementing the energy consumption prediction model and energy management control strategy separately can lead to low information transmission efficiency and difficulty in integrating the entire process optimization. This application provides an energy management method for a range-extended hybrid electric vehicle. In this method, various current vehicle information and various information from the future target driving segment, as well as the predicted average vehicle speed, can be collected. A trained target energy management model can then directly generate torque distribution results and engine start-stop control results, and corresponding control commands can be generated to control the target range-extended hybrid electric vehicle to achieve energy management. Therefore, an end-to-end method for training the target energy management model is realized. This method utilizes a large dataset, supervised learning labels, a target loss function, and a target optimizer to train and optimize the initial energy management model to obtain the target energy management model. This allows the target energy management model in this application to directly output the optimal power source torque allocation and engine start-stop control from information input to output. It eliminates the complex modular design and information transmission process in traditional methods, reduces errors and computational burdens in intermediate links, and thus significantly improves the efficiency of energy management. This facilitates the optimization of the entire process. Furthermore, while ensuring that the obtained power source torque allocation result is the globally optimal energy management strategy, the target energy management model in this application has the reasoning ability to perform real-time energy management of the target range-extended hybrid vehicle. When faced with complex and changing road conditions and environmental conditions, it can quickly respond and adjust the energy management strategy. This demonstrates high practicality while ensuring the robustness and generalization of the target energy management model. This solves the problems in related technologies, such as the difficulty of rule-based methods to adapt to dynamic environments, the limitation of model-based prediction control accuracy due to model quality, the limitations of energy consumption prediction models and energy management control strategies, and the difficulty of integrating the whole process optimization.

[0044] Before explaining the energy management method for range-extended hybrid vehicles in the embodiments of this application, the overall architecture of the energy management model involved in the embodiments of this application will be explained first.

[0045] In recent years, the rapid rise of artificial intelligence technology has brought revolutionary changes to the field of energy management, making it possible to build more intelligent and efficient energy management systems. With the development of AI technology, deep learning and reinforcement learning have shown great potential in data-driven prediction and control optimization, and can be used to solve complex optimization problems in the energy management of range-extended hybrid vehicles.

[0046] Therefore, this application can realize energy management of range-extended hybrid vehicles based on AI deep learning models, that is, directly outputting power source torque distribution and engine start-stop control from environmental and vehicle status information input, thereby realizing global optimization and real-time control.

[0047] It should be noted that, in the embodiments of this application, the preset initial energy management model, the initial energy management model and the target energy management model involved in the following description steps are essentially end-to-end deep learning models, and each includes, but is not limited to, three parts: a feature extraction module, a feature fusion module and an output module.

[0048] The feature extraction module can be used to extract temporal features (time series features) and static features (static global features):

[0049] Time series feature extraction: Dynamic information of a range-extended hybrid vehicle on a future road segment after a certain moment, including but not limited to average vehicle speed v. avg The information, including slope (grade), road segment length (L), and congestion level (cong), is input into the LSTM network in time series format. Let the input sequence be X. seq ={x1,x2,…,x T}, where the data dimension for each time step is d. seq It can be expressed, but is not limited to, as follows:

[0050] h t =LSTM(h t-1 ,x t ;θ LSTM )

[0051] Among them, h t θ represents the hidden state at time step t, containing dynamic features of the current time step and past road segment information; LSTM These are the parameters of the LSTM network, including the weight matrix and biases.

[0052] Finally, embodiments of this application may, but are not limited to, take the last hidden state of the LSTM as the time series feature output:

[0053]

[0054] Static global feature extraction: The current static state of the vehicle includes accelerator pedal opening, brake pedal opening, battery SOC, engine speed, motor speed, actual vehicle speed, and ambient temperature. This information is then extracted into a feature vector X. stat The input is given in the form of a fully connected network to extract its global features. Its output can be, but is not limited to, represented as:

[0055]

[0056] Among them, W stat and b stat For the weights and biases of the fully connected layer; F stat It is the global feature representation after static feature extraction.

[0057] Feature fusion module:

[0058] In this embodiment of the application, the time series feature F seq and static features F stat Global feature representations can be generated by combining features through a feature fusion module, but not limited to this. This is achieved by combining F... seq and F stat After concatenation, the data is input into a multi-layer fully connected network. The fusion process can be represented, but is not limited to, as follows:

[0059]

[0060] Among them, [F seq ,F stat ] is for feature splicing; W fusion and b fusion It refers to the weights and biases of the fusion network; F global It is the fused global feature representation, which serves as the final feature representation of the model.

[0061] Output module:

[0062] In the output module, embodiments of this application may, but are not limited to, be based on global feature F. global The engine torque T is generated through three fully connected networks. engine Motor torque T motor and engine start / stop state engine Allocation strategy:

[0063] T engine =W eng F global +b eng

[0064] T motor =W motor F global +b motor

[0065] S engine =σ(W state F global +b state ).

[0066] Among them, W eng W motor and W state It is the weight matrix of the output layer; b engb motor and b state It is the bias of the output layer; T engine and T motor S represents the torque output of the engine and the electric motor, respectively. engine This represents the engine start / stop state, and σ represents the sigmoid activation function.

[0067] By leveraging the reasoning capabilities of deep learning models, the embodiments of this application can achieve real-time management of vehicle energy. In particular, when facing complex and ever-changing road conditions and environmental conditions, it can quickly respond and adjust strategies, effectively solving the shortcomings of traditional methods such as low transmission efficiency and poor real-time performance.

[0068] Figure 1 This is a flowchart of an embodiment of the end-to-end range-extended hybrid vehicle energy management method according to this application, as shown below. Figure 1 As shown, the end-to-end range-extended hybrid vehicle energy management method in this application embodiment is divided into a model building stage and a model application stage.

[0069] Specifically, Figure 2 This is a flowchart illustrating an energy management method for a range-extended hybrid vehicle applied during the model building stage, as provided in an embodiment of this application.

[0070] like Figure 2 As shown, the energy management method for range-extended hybrid vehicles is applied in the model building phase, including the following steps:

[0071] In step S201, the operating information, environmental information and actual energy output information of multiple range-extended hybrid vehicles at at least one historical moment are collected, as well as the road condition information and predicted average speed of multiple range-extended hybrid vehicles on the target driving segment after at least one historical moment.

[0072] In some embodiments, this application can construct a unified deep learning model that can directly obtain and output the optimal power source torque distribution and engine start-stop control from the input of the vehicle's future road condition information and other vehicle information. Therefore, before constructing this model, embodiments of this application can first design the information that needs to be input into the model. It should be noted that, in the embodiments of this application and in the subsequent description, "vehicle" refers to a range-extended hybrid electric vehicle.

[0073] For example, embodiments of this application can collect operating and environmental information of multiple range-extended hybrid vehicles at at least one historical moment. It should be noted that the at least one historical moment selected for each range-extended hybrid vehicle can be chosen or adjusted according to actual circumstances; that is, the at least one historical moment selected for each range-extended hybrid vehicle can be the same or different. This is merely an illustrative example and does not impose any specific limitations.

[0074] In this application embodiment, the operating information includes, but is not limited to, the accelerator pedal opening, brake pedal opening, battery SOC, engine speed, motor speed, and actual vehicle speed at a certain historical moment. The environmental information includes, but is not limited to, the ambient temperature of the vehicle at a certain historical moment.

[0075] Furthermore, to ensure the accuracy of the model's output results, this application can design the input data more comprehensively. Therefore, embodiments of this application can also collect road condition information and predicted average vehicle speed for a target driving segment after a certain historical moment. Here, the target driving segment can be understood as the future driving segment of the vehicle after a certain historical moment. The specific selection rules for the target driving segment can be set by those skilled in the art according to actual conditions. For example, it can be set as a driving segment within a certain time period (e.g., 30 seconds, 1 minute) after a certain historical moment, or a driving segment within a certain length (e.g., 500 meters, 1 kilometer) after a certain historical moment, etc. This embodiment is only illustrative and does not impose specific limitations.

[0076] The predicted average vehicle speed here refers to the average speed of the vehicle within the target driving segment predicted at this historical moment. The specific calculation rules can be selected or adjusted by those skilled in the art according to the actual situation of the vehicle or actual needs. For example, it can be calculated based on the vehicle's driving situation over a period of time before a certain historical moment (such as the average speed 1 minute or 1 km before a certain historical moment), or based on the average driving speed of other vehicles in the network system, or based on the vehicle's driving situation over a period of time before a certain historical moment combined with the road condition information of the target driving segment (such as calculating based on the average speed 1 minute or 1 km before a certain historical moment combined with the road condition information of the target driving segment). The embodiments in this application are only illustrative and do not impose specific limitations.

[0077] Furthermore, in the embodiments of this application, the road condition information collected includes, but is not limited to, information such as the slope, length, and congestion level of the target driving road segment.

[0078] Additionally, to facilitate later determination that the model can adapt to the actual driving scenarios of the vehicle, embodiments of this application can also collect the actual energy output information of the vehicle at a certain historical moment, thereby providing comprehensive input data for the model.

[0079] The embodiments of this application can collect a large amount of historical vehicle data to prepare the dataset required for building the model, and effectively ensure the effectiveness of model training through a large amount of comprehensive data.

[0080] Step S202: Extract static features of multiple range-extended hybrid vehicles at at least one historical moment based on operation information and environmental information; extract temporal features of multiple range-extended hybrid vehicles at at least one historical moment for the target driving road segment based on road condition information and predicted average vehicle speed; and construct a training dataset based on static features, temporal features and actual energy output information.

[0081] In other embodiments, this application can also perform feature extraction on the collected data, which can save computing power of the model while effectively improving the accuracy and precision of the output results.

[0082] For example, Figure 3 This is a schematic diagram illustrating a dataset construction and annotation framework according to an embodiment of this application. Figure 3 As shown, in this embodiment of the application, static features of multiple range-extended hybrid vehicles at at least one historical moment can be extracted based on operating information and environmental information. The network architecture used when extracting static features can be, but is not limited to, a fully connected network.

[0083] Furthermore, in this embodiment of the application, the temporal features corresponding to the target driving road segment of multiple range-extended hybrid vehicles after at least one historical moment can be extracted based on road condition information and predicted average vehicle speed. The network architecture used when extracting the temporal features may be, but is not limited to, LSTM (Long Short-Term Memory).

[0084] The embodiments of this application can construct a dataset by extracting time-series features, static features, and actual energy output information from a large amount of actual vehicle historical data. This enables the final trained model to dynamically adjust the energy management strategy according to different input features, giving it strong adaptability. Whether in urban congested sections, highways, steep slopes, or other extreme conditions, the model can reasonably allocate power sources to ensure stable system operation, effectively improving the practicality and applicability of the model in this application.

[0085] It should be noted that, in addition to constructing the training dataset, this embodiment may also construct a certain amount of validation and test datasets. For example, 70% of all data may be divided into a training dataset, 20% into a validation dataset, and 10% into a test dataset. The training dataset may be used, but is not limited to, for optimizing model parameters; the validation dataset may be used, but is not limited to, for tuning hyperparameters and preventing overfitting; and the test dataset may be used, but is not limited to, for evaluating the performance of the trained model on unknown data.

[0086] Step S203: Based on the training dataset, generate motor torque distribution results, engine torque distribution results, and engine start-stop control results for multiple range-extended hybrid vehicles at at least one historical moment to generate supervised learning labels for the initial energy management model.

[0087] As one possible approach, the model in this embodiment uses supervised learning. To provide optimal labels for the model's supervised learning, this embodiment can calculate the optimal power source allocation strategy and engine start-stop control strategy for multiple range-extended hybrid vehicles at at least one historical moment based on the training dataset. This means calculating the optimal motor torque allocation result, the optimal engine torque allocation result, and the optimal engine start-stop control result, and using these as the supervised learning labels for the initial energy management model. Here, the initial energy management model can be understood as an end-to-end deep learning model for managing vehicle energy that has not been trained or tested.

[0088] For example, still as Figure 3 As shown, this application may, but is not limited to, use a dynamic programming (DP) algorithm to generate the optimal power source allocation strategy for each sample. Each sample can be understood as a sample of static features, temporal features and actual energy output information corresponding to each time step after the road segment is divided into multiple discrete time steps.

[0089] By using the DP algorithm, this application can achieve a strategy that minimizes vehicle energy consumption through global optimization, thereby providing a reliable supervision signal for deep learning models.

[0090] Suppose the objective function of the system is as follows:

[0091] J = ∫(fuel consumption(T) engine )+λ·|ΔSOC|+μ·ΔT smooth )dt

[0092] Among them, fuel consumption (T) engine ΔSOC represents engine fuel consumption, ΔT represents SOC deviation, used to constrain battery state to avoid overcharging or over-discharging. smooth The power source switching smoothness index, where λ and μ are weighting coefficients.

[0093] For each sample, this embodiment of the application can use dynamic programming to search for the optimal power allocation scheme of the engine and motor under the current operating conditions. Finally, the optimal engine torque, motor torque and engine start-stop state at each moment are recorded as supervised learning labels.

[0094] The embodiments of this application can use dynamic programming algorithm to generate the optimal power source allocation strategy for each sample as the supervised learning label of the model. This can provide the optimal label for the supervised learning of the model, and ensure that the model can learn an energy management strategy close to the global optimum during the model training process. This ensures the global optimum of the final model output without having to run a complex optimization process in real time.

[0095] Step S204: Using the training dataset, supervised learning labels, objective loss function, and objective optimizer, train the initial energy management model end-to-end to construct a target energy management model for managing the energy of the target range-extended hybrid vehicle.

[0096] In some embodiments, after obtaining the training dataset, the embodiments of this application can use the training dataset and supervised learning labels, combined with the target loss function and the target optimizer, to train the initial energy management model end-to-end, so as to construct a target energy management model for managing the energy of the target range-extended hybrid vehicle.

[0097] Here, the objective loss function can be understood as a function that measures the difference between the model output and the supervised learning label; the objective optimizer can be understood as a variety of stochastic optimization algorithms that can adaptively adjust the learning rate for each training parameter in the objective energy management model.

[0098] For example, Figure 4 This is a schematic diagram illustrating model training and optimization according to an embodiment of this application. Figure 4 As shown, this application can input the training dataset and supervised learning labels into the initial energy management model, and then use the Adam optimizer for end-to-end training, designing an objective loss function to minimize energy consumption, maintain SOC balance, and improve driving smoothness, etc.

[0099] For example, the Adam algorithm can adjust the learning rate of the training parameters in the initial energy management model based on historical information about the first moment (mean) and second moment (variance) of the gradient, thereby adapting to the sparsity or variability of different training parameters. Its parameter update formula can be, but is not limited to, expressed as follows:

[0100]

[0101] Among them, g t For gradient, and To estimate the momentum and variance of the gradient at time t, θ t+1 Here are the updated parameters, η is the learning rate, and ∈ is introduced to avoid division by zero or near zero. β1 and β2 are hyperparameters that control the decay rates of the first and second moments.

[0102] This application embodiment can directly generate and output the optimal power source torque distribution strategy and engine start-stop control strategy from various data inputs of the vehicle by constructing a unified deep learning model. This achieves end-to-end optimization, eliminating the complex modular design and information transmission processes of traditional methods, reducing errors and computational burdens in intermediate stages, and thus significantly improving the efficiency of energy management.

[0103] Optionally, in one embodiment of this application, before training the initial energy management model end-to-end using the training dataset, supervised learning labels, target loss function, and target optimizer, the method further includes: determining the energy management optimization target corresponding to the target range-extended hybrid vehicle based on the energy management requirements corresponding to multiple range-extended hybrid vehicles; obtaining the optimization result according to the energy management optimization target; and determining the target loss function based on the optimization result.

[0104] In some embodiments, in order to achieve efficient energy management of range-extended hybrid vehicles, this application may also determine the energy management optimization target based on the energy management requirements corresponding to multiple range-extended hybrid vehicles, thereby obtaining the optimization result according to the energy management optimization target, and then determining the target loss function.

[0105] Among them, energy management requirements can be understood as the requirements of users or vehicle designers for vehicle energy management. For example, is it to ensure the vehicle's energy consumption is minimized, or to prioritize the user's driving experience? The energy management optimization goal is to optimize the management of the vehicle's energy while meeting these energy management requirements.

[0106] For example, if the energy management requirements of multiple range-extended hybrid vehicles in this application are all to reduce energy consumption while ensuring the user's driving experience, then the optimization objectives can be set as two parts: minimizing the total energy consumption of the vehicle and optimizing driving smoothness.

[0107] Minimizing total vehicle energy consumption can be achieved by optimizing fuel and electricity consumption. This involves dynamically adjusting the use of power sources based on factors such as engine and motor power output, vehicle status, road conditions, and driving demands. This can be expressed, but is not limited to, the following formula:

[0108] J=∫(C fuel +λ·C battery )dt

[0109] Among them, C fuel Indicates fuel consumption, C battery This indicates the amount of electrical energy consumed.

[0110] Furthermore, the driving smoothness optimization in this application embodiment can be achieved by reducing the frequent switching between the engine and the electric motor, that is, by reducing vehicle vibration and load changes through smooth torque distribution, which can be expressed as a formula, but is not limited to, as follows:

[0111] min∑(|T engine,t+1 -T engine,t |+|T motor,t+1 -T motor,t |)

[0112] It should be noted that if the energy management requirement is only to minimize the vehicle's total energy consumption or to prioritize driving smoothness, then different energy management optimization objectives can be determined based on the corresponding energy management requirements.

[0113] After determining the energy management optimization objective, the embodiments of this application can construct the objective loss function of the energy management model.

[0114] The embodiments of this application can determine the energy management optimization target based on the energy management needs of the vehicle, and then determine the target loss function of the model. This can transform the energy management target into a quantifiable and optimizable mathematical form, providing a clear optimization direction for model training, enabling the model to optimize energy management by minimizing the loss function, and improving performance indicators such as energy utilization efficiency.

[0115] Optionally, in one embodiment of this application, before training the initial energy management model end-to-end using the training dataset, supervised learning labels, target loss function, and target optimizer, the method further includes: constructing an energy consumption loss function, a power battery balance loss function, and a smoothness loss function for the target loss function; and determining the target loss function based on the energy consumption loss function, the power battery balance loss function, and the smoothness loss function.

[0116] Based on the descriptions of other embodiments, it will be understood that the embodiments of this application can construct a target loss function according to the energy management optimization objective.

[0117] In actual implementation, when training the initial energy management model, this application constructs a target loss function L to ensure that the energy management optimization objective can be achieved. total It can be composed of, but is not limited to, an energy consumption loss function, a power battery balance loss function, and a smoothness loss function, and its formula can be, but is not limited to, expressed as:

[0118] L total =L energy +L SOC-balance +L smoothness

[0119] Among them, L energy L represents energy loss. SOC-balanceL represents the SOC balance loss (power battery balance loss). smoothness This indicates a loss of smoothness.

[0120] The construction expression and meaning of each loss term can be expressed as follows:

[0121] (1) Energy loss L energy The model is constructed to encourage the model to minimize the total energy consumption of the range-extended hybrid vehicle during driving, including but not limited to fuel consumption and battery energy consumption, and its expression can be, but is not limited to, as follows:

[0122] L energy =λ fuel ·C fuel +λ battery ·C battery

[0123] Among them, C fuel Indicates fuel consumption, C battery Indicates energy consumption, λ fuel , λ battery This is a weighting factor used to balance the contributions of fuel consumption and electricity consumption to total energy consumption.

[0124] (2) SOC balance loss L SOC The structure is designed to ensure that the battery operates within a reasonable range, avoiding over-discharge or over-charging, and preventing inefficiency caused by frequent start-stop of the range extender. Its expression can be, but is not limited to, the following:

[0125] L SOC =α SoC ·(ΔSOC) 2

[0126] In fact, α SOC Here, ΔSOC represents the SOC deviation, and its formula can be, but is not limited to, expressed as:

[0127] ΔSOC=SOC current -SOC target

[0128] Among them, SOC current This indicates the current battery SOC. target Indicates the target SOC.

[0129] To further constrain SOC changes and avoid drastic fluctuations in battery SOC, embodiments of this application may add a penalty term to constrain the change in SOC between two consecutive time points. The formula for this penalty term may, but is not limited to, be expressed as:

[0130] L SOC_change =β SOC ·(SOCcurrent,t+1 -SOC current,t ) 2

[0131] Where, β SOC It is the weighting coefficient of the SOC change penalty term.

[0132] The total SOC balance loss can be expressed, but is not limited to, as:

[0133] L SoC_balance =L SOC +L SOC_change

[0134] (3) The smoothness loss is constructed to reduce the switching frequency of the power source and avoid switching the working state between the engine and the motor too frequently, thereby improving the smoothness of driving. Its expression can be, but is not limited to, the following:

[0135]

[0136] Where, γ engine and γ motor ΔT is the weighting coefficient. engine and ΔT motor The formulas representing the changes in engine torque and motor torque, respectively, can be expressed as, but are not limited to, the following:

[0137] ΔT engine =|T engine,t+1 -T engine,t |

[0138] ΔT motor =|T motor,t+1 -T motor,t |

[0139] The embodiments of this application can generate a target loss function by comprehensively considering multiple factors, which can simultaneously take into account different dimensions of objectives, such as ensuring driving smoothness while minimizing vehicle energy consumption. Moreover, it helps different loss functions to constrain the target energy management model from different angles, reducing the risk of overfitting, making the target energy management model more stable on complex and variable data, and enhancing the robustness and generalization ability of the target energy management model.

[0140] Optionally, in one embodiment of this application, the method further includes: optimizing a preset initial energy management model using a learning rate decay strategy and an early stopping mechanism to construct an initial energy management model.

[0141] Based on the descriptions of other embodiments, it is understood that the embodiments of this application can use a target optimizer such as the Adam optimizer for model training. Furthermore, considering that in deep learning, the target energy management model may have the risk of overfitting—that is, it may perform very well on training data, fitting various features and patterns in the training data well, but perform poorly on new, unseen test data or in real-world application scenarios—the embodiments of this application can also introduce a learning rate decay strategy and an early stopping mechanism to optimize the preset initial energy management model, avoiding overfitting of the initial energy management model. Here, the preset initial energy management model can be understood as an initial energy management model without the introduction of the learning rate decay strategy and the early stopping mechanism.

[0142] Specifically, it remains as follows Figure 4 As shown, to avoid escaping local optima, this embodiment of the application can set a learning rate decay strategy in the energy management model, gradually reducing the learning rate as iterations proceed. The expression for this strategy is as follows:

[0143]

[0144] Where, η t η is the learning rate at the t-th iteration, η0 is the initial learning rate, γ is the decay rate, and step is the decay period.

[0145] Next, this embodiment of the application can monitor the loss on the validation set, set patience parameters and thresholds, and introduce an early stopping mechanism. When the loss on the validation set no longer decreases or begins to increase, training is stopped in advance to prevent the initial energy management model from overfitting the training data.

[0146] Furthermore, Table 1 is a training parameter setting table for one embodiment of this application, which can be as follows:

[0147] Table 1

[0148]

[0149]

[0150] The embodiments of this application can introduce a learning rate decay strategy and an early stopping mechanism to optimize the preset initial energy management model, avoid overfitting of the initial energy management model, and ensure that the target energy management model obtained after training can adapt to new and unseen test data, thereby ensuring the performance of the target energy management model in real application scenarios.

[0151] According to the energy management method for range-extended hybrid vehicles proposed in the embodiments of this application, which is applied to the model building stage, the method can collect various current vehicle information and various information of the future target driving segment and predict the average vehicle speed. Through the trained target energy management model, the method can directly generate torque distribution results and engine start-stop control results, and then generate corresponding control commands to control the target range-extended hybrid vehicle to achieve energy management. Therefore, this method enables the end-to-end training and optimization of an initial energy management model using a large dataset, supervised learning labels, a target loss function, and a target optimizer to obtain a target energy management model. This allows the target energy management model in this application to directly output the optimal power source torque allocation and engine start-stop control from information input to output, eliminating the complex modular design and information transmission process in traditional methods, reducing errors and computational burdens in intermediate links, and thus significantly improving the efficiency of energy management. This facilitates optimization of the entire process. Furthermore, while ensuring that the obtained power source torque allocation result is the globally optimal energy management strategy, the target energy management model in this application has the reasoning ability to perform real-time energy management of the target range-extended hybrid vehicle. When faced with complex and changing road conditions and environmental conditions, it can quickly respond and adjust the energy management strategy, demonstrating high practicality while ensuring the robustness and generalization of the target energy management model. This solves the problems in related technologies, such as the difficulty of rule-based methods to adapt to dynamic environments, the limitation of model-based prediction control accuracy due to model quality, the limitations of energy consumption prediction models and energy management control strategies, and the difficulty of integrating the whole process optimization.

[0152] Next, referring to the accompanying drawings, an energy management device for a range-extended hybrid vehicle applied in the model building stage according to an embodiment of this application is described.

[0153] Figure 5 This is a schematic diagram of the energy management device for a range-extended hybrid vehicle according to an embodiment of this application.

[0154] like Figure 5 As shown, the energy management device 10 of the range-extended hybrid vehicle includes: a data acquisition module 100, an extraction module 200, a generation module 300, and a first construction module 400.

[0155] Among them, the acquisition module 100 is used to acquire the operating information, environmental information and actual energy output information of multiple range-extended hybrid vehicles at at least one historical moment, as well as the road condition information and predicted average vehicle speed of multiple range-extended hybrid vehicles on the target driving section after at least one historical moment.

[0156] The extraction module 200 is used to extract static features of multiple range-extended hybrid vehicles at at least one historical moment based on operating information and environmental information, extract temporal features of multiple range-extended hybrid vehicles at at least one historical moment after the target driving road segment based on road condition information and predicted average vehicle speed, and construct a training dataset based on static features, temporal features and actual energy output information.

[0157] The generation module 300 is used to generate motor torque distribution results, engine torque distribution results, and engine start-stop control results for multiple range-extended hybrid vehicles at at least one historical moment based on the training dataset, so as to generate supervised learning labels for the initial energy management model.

[0158] The first building module 400 is used to train an initial energy management model end-to-end using a training dataset, supervised learning labels, a target loss function, and a target optimizer to build a target energy management model for managing the energy of a target range-extended hybrid vehicle.

[0159] Optionally, in one embodiment of this application, it further includes: a first determining module and a second determining module.

[0160] The first determining module is used to determine the energy management optimization target of the target range-extended hybrid vehicle based on the energy management requirements of the target range-extended hybrid vehicle before training the initial energy management model end-to-end using the training dataset, supervised learning labels, target loss function and target optimizer.

[0161] The second determination module is used to obtain the optimization results based on the energy management optimization objectives, and to determine the target loss function based on the optimization results.

[0162] Optionally, in one embodiment of this application, it further includes: a second building module and a third determining module.

[0163] The second building module is used to construct the energy consumption loss function, power battery balance loss function, and smoothness loss function of the target loss function before training the initial energy management model end-to-end using the training dataset, supervised learning labels, target loss function, and target optimizer.

[0164] The third determination module is used to determine the target loss function based on the energy consumption loss function, the power battery balance loss function, and the smoothness loss function.

[0165] Optionally, in one embodiment of this application, it further includes: a third construction module, used to optimize a preset initial energy management model using a learning rate decay strategy and an early stopping mechanism, so as to construct an initial energy management model.

[0166] It should be noted that the foregoing explanation of the energy management method embodiment applied to the model building stage of the range-extended hybrid vehicle also applies to the energy management device of the range-extended hybrid vehicle applied to the model building stage of the same embodiment, and will not be repeated here.

[0167] According to the embodiments of this application, the energy management device for range-extended hybrid vehicles applied in the model building stage can collect various current vehicle information and various information of the future target driving segment and predict the average vehicle speed. It can directly generate torque distribution results and engine start-stop control results through the trained target energy management model, and then generate corresponding control commands to control the target range-extended hybrid vehicle to achieve energy management. Therefore, this method enables the end-to-end training and optimization of an initial energy management model using a large dataset, supervised learning labels, a target loss function, and a target optimizer to obtain a target energy management model. This allows the target energy management model in this application to directly output the optimal power source torque allocation and engine start-stop control from information input to output, eliminating the complex modular design and information transmission process in traditional methods, reducing errors and computational burdens in intermediate links, and thus significantly improving the efficiency of energy management. This facilitates optimization of the entire process. Furthermore, while ensuring that the obtained power source torque allocation result is the globally optimal energy management strategy, the target energy management model in this application has the reasoning ability to perform real-time energy management of the target range-extended hybrid vehicle. When faced with complex and changing road conditions and environmental conditions, it can quickly respond and adjust the energy management strategy, demonstrating high practicality while ensuring the robustness and generalization of the target energy management model. This solves the problems in related technologies, such as the difficulty of rule-based methods to adapt to dynamic environments, the limitation of model-based prediction control accuracy due to model quality, the limitations of energy consumption prediction models and energy management control strategies, and the difficulty of integrating the whole process optimization.

[0168] The above embodiments describe the model building stage. The following describes embodiments of the model application stage.

[0169] Specifically, Figure 6 This is a flowchart illustrating an energy management method for a range-extended hybrid vehicle applied in the model application stage, as provided in an embodiment of this application.

[0170] like Figure 6 As shown, the energy management method for range-extended hybrid vehicles is applied in the model building phase, including the following steps:

[0171] In step S601, real-time operating information and real-time environmental information of the target range-extended hybrid vehicle are collected, as well as road condition information and predicted average speed of the target range-extended hybrid vehicle in the future target driving segment.

[0172] In actual implementation, the embodiments of this application can collect the input information required by the target energy management model, namely the real-time operating information and real-time environmental information of the target range-extended hybrid vehicle, and collect the road condition information and predicted average speed of the target range-extended hybrid vehicle on the target driving segment in the future. Here, the target range-extended hybrid vehicle can be understood as a range-extended hybrid vehicle that uses the energy management method of the range-extended hybrid vehicle in this application to realize vehicle energy management.

[0173] For example, this application can collect real-time accelerator pedal opening, real-time brake pedal opening, real-time battery SOC, real-time engine speed, real-time motor speed, real-time vehicle speed, and real-time ambient temperature of the target range-extended hybrid vehicle; as well as the predicted average vehicle speed of the target range-extended hybrid vehicle on the future target driving segment and information such as the slope, length, and congestion level of the future target driving segment.

[0174] Here, the future target driving segment refers to a driving segment of a certain duration or length in the future, based on the current time. It should be noted that the selection rules for the future target driving segment and the calculation rules for the predicted average speed are the same as those for the target driving segment selection rules and the calculation rules for the predicted average speed mentioned above.

[0175] In step S602, real-time operating information, real-time environmental information, road condition information, and predicted average vehicle speed are input into the pre-built target energy management model. The real-time static features of the target range-extended hybrid vehicle and the temporal features of the target range-extended hybrid vehicle in the future target driving segment are extracted. The static features and temporal features are then fused to obtain fused features. Based on the fused features, the real-time motor torque distribution results, real-time engine torque distribution results, and real-time engine start-stop control results of the target range-extended hybrid vehicle are output.

[0176] In some embodiments, after collecting the information required by the target energy management model, the embodiments of this application can input this real-time operating information, real-time environmental information, road condition information and predicted average vehicle speed into the pre-built target energy management model.

[0177] The feature extraction module in the target energy management model can extract the static features of the target range-extended hybrid vehicle at the current moment and the temporal features of the target driving segment in the future based on this information. Then, the feature fusion module in the target energy management model can fuse the static features and temporal features to obtain fused features. Based on the fused features, the model can generate the real-time motor torque distribution results, real-time engine torque distribution results, and real-time engine start-stop control results of the target range-extended hybrid vehicle at the current moment. Finally, the model can output the results using the output module in the target energy management model.

[0178] In step S603, corresponding control commands are generated based on the real-time motor torque distribution results, the real-time engine torque distribution results, and the real-time engine start-stop control results to control the target range-extended hybrid vehicle for energy management.

[0179] In some embodiments, Figure 7 This is a flowchart illustrating the practical application of a model according to one embodiment of this application. Figure 7 As shown:

[0180] Step S701: Collect information on vehicles and future target driving routes in real time;

[0181] Step S702: Input the collected information into the trained target energy management model for torque allocation, and output the real-time motor torque allocation result, the real-time engine torque allocation result, and the real-time engine start-stop control result.

[0182] Step S703: Generate corresponding control commands based on these results;

[0183] Step S704: These control commands are sent to the corresponding components of the target range-extended hybrid vehicle. The components execute the control commands to ultimately realize the energy management of the target range-extended hybrid vehicle.

[0184] According to the energy management method for range-extended hybrid vehicles proposed in the embodiments of this application, which is applied to the model application stage, the method can collect various current vehicle information and various information of the future target driving segment and predict the average vehicle speed. Through the trained target energy management model, the method can directly generate torque distribution results and engine start-stop control results, and then generate corresponding control commands to control the target range-extended hybrid vehicle to achieve energy management. Therefore, this method enables the end-to-end training and optimization of an initial energy management model using a large dataset, supervised learning labels, a target loss function, and a target optimizer to obtain a target energy management model. This allows the target energy management model in this application to directly output the optimal power source torque allocation and engine start-stop control from information input to output, eliminating the complex modular design and information transmission process in traditional methods, reducing errors and computational burdens in intermediate links, and thus significantly improving the efficiency of energy management. This facilitates optimization of the entire process. Furthermore, while ensuring that the obtained power source torque allocation result is the globally optimal energy management strategy, the target energy management model in this application has the reasoning ability to perform real-time energy management of the target range-extended hybrid vehicle. When faced with complex and changing road conditions and environmental conditions, it can quickly respond and adjust the energy management strategy, demonstrating high practicality while ensuring the robustness and generalization of the target energy management model. This solves the problems in related technologies, such as the difficulty of rule-based methods to adapt to dynamic environments, the limitation of model-based prediction control accuracy due to model quality, the limitations of energy consumption prediction models and energy management control strategies, and the difficulty of integrating the whole process optimization.

[0185] Next, referring to the accompanying drawings, an energy management device for a range-extended hybrid vehicle proposed according to an embodiment of this application and applied in the model application stage is described.

[0186] Figure 8 This is a schematic diagram of the energy management device for a range-extended hybrid vehicle applied in the model application stage according to an embodiment of this application.

[0187] like Figure 8 As shown, the energy management device 20 for range-extended hybrid vehicles applied in the model application stage includes: a data acquisition module 500, a processing module 600, and a management module 700.

[0188] The acquisition module 500 is used to collect real-time operating information and real-time environmental information of the target range-extended hybrid vehicle, as well as road condition information and predicted average speed of the target range-extended hybrid vehicle on the target driving route in the future.

[0189] The processing module 600 is used to input real-time operating information, real-time environmental information, road condition information and predicted average vehicle speed into a pre-built target energy management model, extract the real-time static features of the target range-extended hybrid vehicle and the temporal features of the target range-extended hybrid vehicle in the future target driving segment, and fuse the static features and temporal features to obtain fused features, so as to output the real-time motor torque distribution results, real-time engine torque distribution results and real-time engine start-stop control results of the target range-extended hybrid vehicle based on the fused features.

[0190] The management module 700 is used to generate corresponding control commands based on the real-time motor torque distribution results, the real-time engine torque distribution results, and the real-time engine start-stop control results, so as to control the target range-extended hybrid vehicle for energy management.

[0191] It should be noted that the foregoing explanation of the energy management method embodiment for range-extended hybrid vehicles applied in the model application stage also applies to the energy management device for range-extended hybrid vehicles applied in the model application stage of this embodiment, and will not be repeated here.

[0192] According to the embodiments of this application, the energy management device for range-extended hybrid vehicles applied in the model application stage can collect various current vehicle information and various information of the future target driving segment and predict the average vehicle speed. Through the trained target energy management model, it can directly generate torque distribution results and engine start-stop control results, and then generate corresponding control commands to control the target range-extended hybrid vehicle to achieve energy management. Therefore, this method enables the end-to-end training and optimization of an initial energy management model using a large dataset, supervised learning labels, a target loss function, and a target optimizer to obtain a target energy management model. This allows the target energy management model in this application to directly output the optimal power source torque allocation and engine start-stop control from information input to output, eliminating the complex modular design and information transmission process in traditional methods, reducing errors and computational burdens in intermediate links, and thus significantly improving the efficiency of energy management. This facilitates optimization of the entire process. Furthermore, while ensuring that the obtained power source torque allocation result is the globally optimal energy management strategy, the target energy management model in this application has the reasoning ability to perform real-time energy management of the target range-extended hybrid vehicle. When faced with complex and changing road conditions and environmental conditions, it can quickly respond and adjust the energy management strategy, demonstrating high practicality while ensuring the robustness and generalization of the target energy management model. This solves the problems in related technologies, such as the difficulty of rule-based methods to adapt to dynamic environments, the limitation of model-based prediction control accuracy due to model quality, the limitations of energy consumption prediction models and energy management control strategies, and the difficulty of integrating the whole process optimization.

[0193] Figure 9A schematic diagram of the structure of a vehicle provided in an embodiment of this application. The vehicle may include:

[0194] The memory 901, the processor 902, and the computer program stored on the memory 901 and capable of running on the processor 902.

[0195] When the processor 902 executes the program, it implements the energy management method for range-extended hybrid vehicles provided in the above embodiments.

[0196] Furthermore, the vehicle also includes:

[0197] Communication interface 903 is used for communication between memory 901 and processor 902.

[0198] The memory 901 is used to store computer programs that can run on the processor 902.

[0199] The memory 901 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0200] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0201] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0202] The processor 902 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0203] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the energy management method for the range-extended hybrid vehicle described above.

[0204] This application also provides a computer program product, including a computer program that can run computer instructions. When the computer instructions are executed by a processor, they implement the energy management method for range-extended hybrid vehicles provided in this application.

[0205] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0206] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0207] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0208] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0209] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0210] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0211] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0212] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. An energy management method for a range extended hybrid vehicle, characterized in that, The method is applied to a model construction stage, and the method comprises the following steps: Collecting running information, environment information and actual energy output information of a plurality of extended-range hybrid vehicles at at least one historical moment, and road condition information and predicted average speed of a target driving section after the at least one historical moment of the plurality of extended-range hybrid vehicles, wherein the running information comprises a throttle pedal opening degree, a brake pedal opening degree, a battery SOC, an engine speed, a motor speed and an actual vehicle speed of a vehicle at a certain historical moment, the environment information comprises an environment temperature of the vehicle at a certain historical moment, and the road condition information comprises a slope, a length and a congestion degree of the target driving section, and the vehicle is an extended-range hybrid vehicle; Extracting static features corresponding to the plurality of extended-range hybrid vehicles at the at least one historical moment according to the running information and the environment information, extracting time sequence features corresponding to the target driving section after the at least one historical moment of the plurality of extended-range hybrid vehicles according to the road condition information and the predicted average speed, and constructing a training data set according to the static features, the time sequence features and the actual energy output information; Based on the training data set, generating motor torque distribution results, engine torque distribution results and engine start-stop control results of the plurality of extended-range hybrid vehicles at the at least one historical moment to generate a supervised learning label of an initial energy management model; Using the training data set, the supervised learning label, a target loss function and a target optimizer, the initial energy management model is trained end to end to construct a target energy management model for managing energy of a target extended-range hybrid vehicle.

2. The method of claim 1, wherein, Before the initial energy management model is trained end to end using the training data set, the supervised learning label, the target loss function and the target optimizer, the method further comprises: Based on energy management requirements corresponding to the plurality of extended-range hybrid vehicles, determining energy management optimization objectives corresponding to the plurality of extended-range hybrid vehicles; According to the energy management optimization objectives, obtaining an optimization result, and determining the target loss function according to the optimization result.

3. The method of claim 1, wherein, Before the initial energy management model is trained end to end using the training data set, the supervised learning label, the target loss function and the target optimizer, the method further comprises: Constructing an energy consumption loss function, a power battery balance loss function and a smoothness loss function of the target loss function; Determining the target loss function based on the energy consumption loss function, the power battery balance loss function and the smoothness loss function.

4. The method of claim 1, wherein, The method further comprises: Optimizing a preset initial energy management model using a learning rate decay strategy and an early stopping mechanism to construct the initial energy management model.

5. An energy management method for a range extended hybrid vehicle, characterized in that, The method is applied to a model application stage, and the energy management method of the extended-range hybrid vehicle according to any one of claims 1-4 comprises the following steps: Collecting real-time running information and real-time environment information of a target extended-range hybrid vehicle, and collecting road condition information and predicted average speed of a future target driving section of the target extended-range hybrid vehicle; input the real-time running information, the real-time environment information, the road condition information and the predicted average vehicle speed into a pre-constructed target energy management model, extract real-time static characteristics of the target extended-range hybrid vehicle and time sequence characteristics of the target extended-range hybrid vehicle on a future target driving road section, and fuse the static characteristics and the time sequence characteristics to obtain fused characteristics, so as to output real-time motor torque distribution results, real-time engine torque distribution results and real-time engine start-stop control results of the target extended-range hybrid vehicle according to the fused characteristics; generate corresponding control instructions according to the real-time motor torque distribution results, the real-time engine torque distribution results and the real-time engine start-stop control results, so as to control the target extended-range hybrid vehicle to perform energy management.

6. An energy management device of a range extended hybrid vehicle, characterized in that, application in a model construction stage, wherein the device comprises: a collection module configured to collect running information, environment information and actual energy output information of a plurality of extended-range hybrid vehicles at at least one historical time, and road condition information and predicted average vehicle speed of the plurality of extended-range hybrid vehicles on a target driving road section after the at least one historical time, wherein the running information comprises a throttle pedal opening degree, a brake pedal opening degree, a battery SOC, an engine speed, a motor speed and an actual vehicle speed of a vehicle at a certain historical time, the environment information comprises an environment temperature of the vehicle at a certain historical time, the road condition information comprises a slope, a length and a congestion degree of the target driving road section, and the vehicle is an extended-range hybrid vehicle; an extraction module configured to extract static characteristics of the plurality of extended-range hybrid vehicles corresponding to the at least one historical time according to the running information and the environment information, extract time sequence characteristics of the plurality of extended-range hybrid vehicles corresponding to the target driving road section after the at least one historical time according to the road condition information and the predicted average vehicle speed, and construct a training data set according to the static characteristics, the time sequence characteristics and the actual energy output information; a generation module configured to generate motor torque distribution results, engine torque distribution results and engine start-stop control results of the plurality of extended-range hybrid vehicles at the at least one historical time based on the training data set, so as to generate a supervised learning label of an initial energy management model; a construction module configured to end-to-end train the initial energy management model by using the training data set, the supervised learning label, a target loss function and a target optimizer, so as to construct a target energy management model for managing energy of a target extended-range hybrid vehicle.

7. An energy management device of a range extended hybrid vehicle, characterized in that, application in a model application stage, the energy management device of the extended-range hybrid vehicle according to claim 5, wherein the device comprises: a collection module configured to collect real-time running information and real-time environment information of a target extended-range hybrid vehicle, and collect road condition information and predicted average vehicle speed of the target extended-range hybrid vehicle on a future target driving road section; The processing module is configured to input the real-time running information, the real-time environment information, the road condition information and the predicted average vehicle speed into a pre-constructed target energy management model, extract real-time static characteristics of the target extended-range hybrid vehicle and time sequence characteristics of the target extended-range hybrid vehicle in a future target driving route, and fuse the static characteristics and the time sequence characteristics to obtain fused characteristics, so as to output real-time motor torque distribution results, real-time engine torque distribution results and real-time engine start-stop control results of the target extended-range hybrid vehicle according to the fused characteristics. The management module is configured to generate corresponding control instructions according to the real-time motor torque distribution results, the real-time engine torque distribution results and the real-time engine start-stop control results, so as to control the target extended-range hybrid vehicle to perform energy management.

8. A vehicle characterized by comprising: The computer program is executed by the processor to implement the energy management method of the extended-range hybrid vehicle according to any one of claims 1-4 or 5. The computer program is executed by the processor to implement the energy management method of the extended-range hybrid vehicle according to any one of claims 1-4 or 5.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the energy management method of the extended-range hybrid vehicle according to any one of claims 1-4 or 5.

10. A computer program product comprising a computer program, characterized in that, ​

Images