Energy recovery control method and device and vehicle

Abstract

The invention provides an energy recovery control method and device and a vehicle, and is applied to the technical field of new energy vehicles, and the method comprises the steps: obtaining real-time state data, real-time road condition information and real-time environment information of the vehicle, inputting the information into a pre-trained multi-modal sensor data fusion model, carrying out the feature fusion, obtaining a plurality of fusion feature vectors, and carrying out the recognition of the fusion feature vectors; inputting the fusion feature vector into a pre-trained road slope and resistance estimation model for calculation to obtain an estimated road slope and an estimated driving resistance value of a road where the vehicle is located, and inputting the estimated road slope, the estimated driving resistance value and the vehicle state data into a pre-trained recovery gain regulation and control model for calculation, and a first regenerative braking control strategy is determined, and energy recovery is conducted on the vehicle according to the first regenerative braking control strategy. According to the method, the vehicle state and external environment factors are fully considered, the data are analyzed in combination with the deep learning algorithm, a more accurate energy recovery decision is made, and the energy recovery efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of new energy vehicle technology, specifically to an energy recovery control method, device, and vehicle. Background Technology

[0002] With the increasing prominence of global energy shortages and environmental pollution, the automotive industry is accelerating its transformation towards new energy sources. However, the driving range of new energy vehicles remains one of the main obstacles to their widespread adoption. Energy recovery, as a key technology for improving the driving range of new energy vehicles, can recover some energy during braking, reducing reliance on batteries. Furthermore, by reducing energy waste during braking, it further reduces vehicle energy consumption and carbon emissions. Therefore, energy recovery for new energy vehicles is not only an important technological means to address energy and environmental challenges, but also a key technology for improving vehicle performance and user experience, possessing significant practical importance and broad development prospects.

[0003] Existing energy recovery control methods often rely solely on simple vehicle status information, such as vehicle speed and acceleration, without fully considering external environmental factors. As a result, existing energy recovery methods cannot adapt to complex and ever-changing road conditions in real time, leading to unsatisfactory energy recovery results. Summary of the Invention

[0004] In view of this, this application provides an energy recovery control method that solves or improves the technical problem that existing energy recovery methods cannot adapt to complex and ever-changing road conditions in real time, resulting in unsatisfactory energy recovery effects.

[0005] As a first aspect of this application, this application provides an energy recovery control method, comprising:

[0006] Acquire real-time vehicle status data, real-time road condition information of the road where the vehicle is located, and real-time environmental information of the environment in which the vehicle is located.

[0007] The real-time status data, the real-time road condition information, and the real-time environmental information are input into a pre-trained multimodal sensor data fusion model to perform feature fusion, thereby obtaining multiple fused feature vectors. The fused feature vectors are used to represent the visual information around the vehicle, the three-dimensional spatial information around the vehicle, and the motion state of the vehicle.

[0008] The fused feature vector is input into a pre-trained road slope and resistance estimation model for calculation to obtain the estimated road slope and estimated driving resistance value of the road where the vehicle is located.

[0009] inputting the estimated road slope, the estimated driving resistance value and real-time state data into a pre-trained recycling gain regulation model for calculation to determine a first regenerative braking control strategy;

[0010] recovering energy of the vehicle according to the first regenerative braking control strategy.

[0011] Optionally, the real-time state data comprises real-time IMU data of the vehicle and real-time vehicle speed of the vehicle; the real-time road condition information comprises a real-time road image; and the real-time environmental information comprises real-time laser radar point cloud data.

[0012] After obtaining the real-time state data of the vehicle, the real-time road condition information of the road where the vehicle is located and the real-time environmental information of the environment where the vehicle is located, the method further comprises:

[0013] converting the real-time road image into preset format image data;

[0014] projecting and converting the real-time laser radar point cloud data to obtain a two-dimensional image;

[0015] time-synchronizing the real-time inertial sensor data and the real-time vehicle speed.

[0016] Optionally, the multi-modal sensor data fusion model comprises a convolutional neural network layer and a long short-term memory network layer.

[0017] The real-time state data, the real-time road condition information and the real-time environmental information are input into a pre-trained multi-modal sensor data fusion model for feature fusion, comprising:

[0018] inputting the preset format image data and the two-dimensional image into the convolutional neural network layer for spatial feature extraction to obtain a first feature vector, the first feature vector being used to represent visual information around the vehicle and three-dimensional spatial information around the vehicle;

[0019] inputting the first feature vector, the time-synchronized real-time inertial sensor data and the real-time vehicle speed into the long short-term memory network layer for calculation to obtain a fusion feature vector.

[0020] Optionally, after inputting the fusion feature vector into a pre-trained road slope and resistance estimation model for calculation to obtain an estimated road slope of the road where the vehicle is located and an estimated driving resistance value, the method further comprises:

[0021] adjusting model parameters of the road slope and resistance estimation model when a difference between the estimated road slope and an actual road slope of the road where the vehicle is located is greater than a preset slope threshold.

[0022] Optionally, after energy recovery of the vehicle is performed according to the first regenerative braking control strategy, the method further comprises:

[0023] calculating an actual recovery efficiency when the vehicle recovers energy in the first regenerative braking control strategy;

[0024] adjusting a model parameter of the recovery gain regulation model when the actual recovery efficiency is less than a preset recovery efficiency threshold.

[0025] Optionally, the energy recovery control method provided in the first aspect of the application further comprises:

[0026] obtaining real-time driving behavior data and real-time driving preference data of a user;

[0027] determining a real-time driving style of the user according to the real-time driving behavior data and the real-time driving preference data;

[0028] finding a second regenerative braking control strategy matching the real-time driving style in a pre-constructed driving style type library; wherein the pre-constructed driving style type library comprises a preset driving style and a preset regenerative braking control strategy corresponding to the preset driving style;

[0029] determining a target braking control strategy of the user according to the first regenerative braking control strategy and the second regenerative braking control strategy;

[0030] performing energy recovery of the vehicle according to the target braking control strategy.

[0031] Optionally, determining the target braking control strategy of the user according to the first regenerative braking control strategy and the second regenerative braking control strategy comprises:

[0032] when a difference between the first regenerative braking control strategy and the second regenerative braking control strategy is less than a preset difference, correcting the first regenerative braking control strategy based on the second regenerative braking control strategy to determine the target braking control strategy of the user.

[0033] Optionally, determining the target braking control strategy of the user according to the first regenerative braking control strategy and the second regenerative braking control strategy comprises:

[0034] when the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is greater than or equal to the preset difference, determining one of the first regenerative braking control strategy and the second regenerative braking control strategy as the target braking control strategy of the user based on a preset priority rule.

[0035] As a second aspect of the present application, the present application provides an energy recovery control device, comprising:

[0036] a data acquisition module, configured to acquire real-time state data of the vehicle, real-time road condition information of a road where the vehicle is located, and real-time environmental information of an environment where the vehicle is located;

[0037] a multi-modal sensor data fusion model, configured to perform feature fusion on the real-time state data, the real-time road condition information, and the real-time environmental information, to obtain a plurality of fusion feature vectors, the fusion feature vectors being used to represent visual information around the vehicle, three-dimensional space information around the vehicle, and a motion state of the vehicle;

[0038] a road slope and resistance estimation model, configured to input the fusion feature vectors into a pre-trained road slope and resistance estimation model to perform calculation, to obtain an estimated road slope and an estimated driving resistance value of the road where the vehicle is located;

[0039] a recovery gain regulation model, configured to perform calculation on the estimated road slope, the estimated driving resistance value, and vehicle state data, to determine a first regenerative braking control strategy;

[0040] an energy recovery module, configured to perform energy recovery on the vehicle according to the first regenerative braking control strategy.

[0041] As a third aspect of the present application, the present application provides a vehicle comprising the energy recovery control device of the second aspect.

[0042] Based on the above, the energy recovery control method provided by the present application can fully consider the vehicle state and external environmental factors and achieve efficient data fusion and time series analysis by using the multi-modal sensor data fusion model to perform feature fusion on the acquired real-time state data of the vehicle, real-time road condition information of a road where the vehicle is located, and real-time environmental information of an environment where the vehicle is located, thereby providing data support for subsequent estimation of road slope and driving resistance value. By inputting the fusion feature vectors into the road slope and resistance estimation model to perform calculation, the estimated road slope and the estimated driving resistance value of the road where the vehicle is located under a dynamic environment can be obtained, and accurate basis can be provided for dynamic adjustment of braking force, i.e., determination of energy recovery decision. Through comprehensive evaluation of the estimated road slope, the estimated driving resistance value, and the vehicle state data by the recovery gain regulation model, the regenerative braking control strategy obtained is more suitable for the current vehicle state, road condition, and environment where the vehicle is located, so that the energy recovery decision made is more accurate and the recovery efficiency is higher. BRIEF DESCRIPTION OF DRAWINGS

[0043] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description thereof taken in conjunction with the accompanying drawings, in which like reference characters designate like elements in the figures. The accompanying drawings are intended to provide a further understanding of the present application, and are incorporated and constitute a part of this specification, illustrate embodiments of the present application and serve to explain the present application, and are not intended to limit the present application in any manner. In the drawings:

[0044] Figure 1 Fig. 1 shows a flowchart of an energy recovery control method according to an embodiment of the present application.

[0045] Figure 2 Fig. 2 shows a flowchart of another energy recovery control method according to an embodiment of the present application.

[0046] Figure 3 Fig. 3 shows a flowchart of another energy recovery control method according to an embodiment of the present application.

[0047] Figure 4 Fig. 4 shows a flowchart of another energy recovery control method according to an embodiment of the present application.

[0048] Figure 5 Fig. 5 shows a flowchart of another energy recovery control method according to an embodiment of the present application.

[0049] Figure 6 Fig. 6 shows a flowchart of another energy recovery control method according to an embodiment of the present application.

[0050] Figure 7 Fig. 7 shows a structural block diagram of an energy recovery control device according to an embodiment of the present application.

[0051] Figure 8 Fig. 8 shows a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0052] In the description of the present application, the meaning of "a plurality of" is at least two, such as two, three, etc., unless otherwise explicitly and specifically limited. All directional indications (such as upper, lower, left, right, front, back, top, bottom, etc.) in the embodiments of the present application are only used to explain the relative position relationship, movement condition, etc. between components, and if the specific posture (as shown in the drawings) changes, the directional indications will also change accordingly. In addition, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0053] In addition, the reference to "embodiments" herein means that the specific features, structures or characteristics described in connection with the embodiments can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean that it refers to the same embodiment, nor is it independent or alternative to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0054] To solve the above technical problems, the present application provides an energy recovery control method, device and vehicle, which aims to fully consider the vehicle state and external environmental factors, analyze these data combined with deep learning algorithm, make more accurate energy recovery decision, and improve energy recovery efficiency. The specific implementation scheme is described in detail through embodiments and examples below.

[0055] Exemplary method

[0056] In the new energy vehicle system, an electronic device can be added, which can be a gateway or an independent network control device of the gateway. The energy recovery control method of the present application is applied to the electronic device, such as Figure 1 As shown in the figure, in the present exemplary embodiment, an energy recovery control method is provided, which can include the following steps:

[0057] S10: acquiring real-time state data of the vehicle, real-time road condition information of the road where the vehicle is located, and real-time environmental information of the environment where the vehicle is located.

[0058] The vehicle can be a pure electric vehicle or a hybrid vehicle, and a plurality of sensor assemblies can be arranged on the vehicle, including a camera, a laser radar, an IMU (Inertial Measurement Unit), a vehicle speed sensor, etc. Real-time state data of the vehicle, real-time road condition information of the road where the vehicle is located, and real-time environmental information of the environment where the vehicle is located can be collected through the sensor assemblies. The control module can obtain the real-time state data of the vehicle and the real-time environmental information of the vehicle collected by the plurality of sensor assemblies. It can be understood that the real-time state data of the vehicle, the real-time road condition information of the road where the vehicle is located, and the real-time environmental information of the environment where the vehicle is located can also be obtained by other means, such as vehicle state information can be obtained by CAN bus, which is not limited.

[0059] The real-time state data of the vehicle can include real-time IMU data of the vehicle and real-time vehicle speed. The real-time IMU data of the vehicle can include acceleration and angular velocity of the vehicle. By monitoring the acceleration and angular velocity of the vehicle, the dynamic state information of the vehicle can be obtained. The sampling frequency of the IMU can be set to 200Hz to ensure the real-time nature of the data. The vehicle speed is fed back in real time by the speed sensor to provide basic data for dynamic adjustment.

[0060] The sensor assembly can also include a voltage sensor, and the real-time state data of the vehicle can also include voltage sensor data. The current power of the vehicle is determined according to the voltage sensor data, which provides a basis for the development of the regenerative braking control strategy. Of course, the current power of the vehicle can also be measured by other sensor assemblies, which will not be described in detail here.

[0061] The real-time road condition information of the road where the vehicle is located can include real-time road images. The real-time road images can be obtained by a camera or other image acquisition device. The real-time road images can include road information, traffic signs, pedestrians, etc. Specifically, to ensure the clarity and detail capture of the images, the resolution of the image acquisition device can be set. For example, when the image acquisition device is a camera, the resolution of the camera can be set to 1920x1080.

[0062] The real-time environmental information of the environment where the vehicle is located can include real-time laser radar point cloud data. Through the laser radar, the point cloud data can be updated in real time to form a three-dimensional image of the environment around the vehicle, providing high-precision three-dimensional environmental information and accurately measuring the distance and position of surrounding objects.

[0063] S20: inputting the real-time state data, the real-time road condition information, and the real-time environmental information into a pre-trained multi-modal sensor data fusion model to perform feature fusion to obtain a plurality of fusion feature vectors.

[0064] The multi-modal sensor data fusion model is pre-trained, and the combination of the CNN (Convolutional Neural Network) and the LSTM (Long Short-Term Memory) of the application forms a composite model, i.e., a multi-modal sensor data fusion model.

[0065] The specific training process of the multi-modal sensor data fusion model includes the following stages: data preparation, model construction, model training, and model verification and evaluation. Specifically, the data preparation stage can include data collection and data processing.

[0066] Data collection: Collect the state data of vehicles in different driving environments and different weather conditions, the road information and environmental information, the driving environment such as city, countryside, highway, etc., the weather condition such as day and night, sunny and rainy, etc., the road information can include the type, length, width, number of lanes, road condition, traffic flow, surrounding facilities, etc., then label the collected information to form a data set, and divide the data set into training set, validation set and test set for supervised learning of the model.

[0067] Data processing: Use data augmentation techniques such as rotation, flipping, noise addition, etc. to expand the training set and enhance the generalization ability of the model.

[0068] Model construction: It is divided into CNN part and LSTM part. For the CNN part, set the input layer according to the data dimension, add multiple convolution layers to extract the spatial features of the data through the convolution kernel, add the pooling layer to reduce and compress the features to obtain the final feature vector; for the LSTM part, add the LSTM layer and set the appropriate number of units and output mode.

[0069] Model training: First, set the training parameters, and use the cross-entropy loss function to optimize the parameters of CNN and LSTM. Specifically, adjust the number of convolution layers, the size of the convolution kernel of CNN, and the number of hidden layer units of LSTM to optimize the performance of the model. Then use the Adam optimizer to set the initial learning rate and learning rate decay strategy to dynamically adjust the learning rate to improve the training efficiency. The specific training process is divided into multiple stages. First, train CNN and LSTM separately, input the training data into the model, extract the features through CNN, and then input the LSTM for time series modeling. Then, combine the two parts for end-to-end training to ensure the effective fusion of the two models.

[0070] Model validation and evaluation: Cross-validation method is used to evaluate the effect of different super parameter combinations, to ensure that the model performs well in various environments. If the output feature vector does not meet the requirements, such as missing features, excessive noise or abnormal distribution, leading to unreliable fusion results, the weight or structure of the fusion model can be adjusted, such as the number of hidden layers or neurons, or the model can be retrained or fine-tuned to optimize feature extraction and fusion effect. Through testing on the validation set, the accuracy and robustness of the model are evaluated. By comparing the performance of multi-modal fusion model with traditional single sensor model, it is found that the multi-modal sensor data fusion model significantly improves the decision accuracy in complex environment, and the response time of multi-modal sensor data fusion model is reduced, and the real-time data processing capability is significantly enhanced.

[0071] The multi-modal sensor data fusion model outputs the fused feature vectors, which are used to represent the visual information around the vehicle, the three-dimensional spatial information around the vehicle, and the motion state of the vehicle, such as road information, environmental information, vehicle speed, IMU data, etc. These fused feature vectors are used as input data for the subsequent road slope and resistance estimation model.

[0072] In some embodiments, the acceleration data, angular velocity data, real-time vehicle speed, road information, traffic signs, pedestrian information, and distance and position information of surrounding objects of the vehicle are input into a multi-modal sensor data fusion model formed by combining convolutional neural network and long short-term memory network, and feature fusion is performed to obtain a plurality of fused feature vectors.

[0073] The multi-modal sensor data fusion model in this embodiment can effectively process data with spatio-temporal characteristics by combining the spatial feature extraction capability of CNN and the time series modeling capability of LSTM. By fully considering the vehicle state and external environmental factors, and realizing efficient data fusion and time series analysis, it can provide data support for subsequent estimation of road slope and driving resistance value.

[0074] S30: Input the fused feature vector into the pre-trained road slope and resistance estimation model for calculation to obtain the estimated road slope and estimated driving resistance value of the road where the vehicle is located.

[0075] When optimizing energy recovery, new energy vehicles need to consider multiple factors, among which road slope and driving resistance are key factors affecting energy recovery efficiency. Traditional methods usually use simple linear models to estimate these parameters, which fail to fully consider dynamic changes and nonlinear relationships. Therefore, building an efficient model that can estimate road slope and driving resistance in real time is crucial for obtaining the estimated road slope and estimated driving resistance value of the road where the vehicle is located in dynamic environment.

[0076] In some embodiments, an RNN (Recurrent Neural Network) can be used as the main model for road slope and resistance estimation. RNN can effectively process time series data and is suitable for state estimation in dynamic environments. The recurrent neural network includes an input layer, a hidden layer and an output layer, the fusion feature vector is input into the pre-trained road slope and resistance estimation model, the input layer of the recurrent neural network accepts the fused feature vector, the hidden layer of the recurrent neural network performs calculation, the hidden layer contains multiple RNN units to capture the time dependence in the data. And through LSTM or GRU (Gated Recurrent Unit) to solve the gradient disappearance problem of traditional RNN in long sequence data, the output layer outputs the estimated road slope and estimated driving resistance value of the road where the vehicle is located. Among them, the estimated road slope represents the degree of inclination of the current road section, with a unit of percentage (%), and the estimated driving resistance value is an estimated value considering factors such as air resistance and rolling resistance, with a unit of Newton (N).

[0077] The specific training process of the road slope and resistance estimation model includes several stages such as data collection and preprocessing, model training, and model verification and evaluation.

[0078] Data collection and preprocessing: Collect historical driving data under different slopes, road conditions and environmental conditions to form a data set. The driving data includes the speed, acceleration and environmental sensor data of the vehicle. Through the actual driving data and GPS information, the road slope and driving resistance at each time point are calculated as the training labels of the model.

[0079] Model training: The data set is divided into a training set and a test set, for example, the ratio of the training set and the test set is 8:2, the test set uses MSE (Mean Squared Error) as the loss function, the model is trained using the training set data, and regression training is performed for road slope and resistance.

[0080] Model verification and evaluation: According to the training results, adjust the number of layers, the number of units, the learning rate and other hyperparameters of the RNN, and evaluate the influence of different hyperparameters on the performance of the model through cross-validation method. Introduce early stopping mechanism to prevent overfitting and ensure that the model performs well on the validation set. Specifically, when underfitting occurs, increase the number of hidden layers to improve the model's expression ability, and when overfitting occurs, reduce the number of hidden layers to reduce the complexity of the model.

[0081] The present application estimates the road slope and driving resistance in real time through the recurrent neural network (RNN), overcoming the limitations of traditional linear models. The model can adapt to changing driving conditions, dynamically adjust parameters, and ensure that the road slope and resistance estimation model can run efficiently in various environments.

[0082] S40: inputting the estimated road slope, the estimated driving resistance value and the real-time state data into a pre-trained recovery gain regulation model to perform calculation, so as to determine a first regenerative braking control strategy.

[0083] In the energy recovery process of a new energy vehicle, optimization of a regenerative braking control strategy is a key to improving the endurance capability. The traditional regenerative braking control strategy depends on fixed rules, which is difficult to adapt to different driving environments. According to the estimated road slope, the estimated driving resistance value and the vehicle state data, the regenerative braking strategy is dynamically adjusted, so as to maximize the energy recovery efficiency. The output of the model is the optimal regenerative braking strategy under the current environment and state. This strategy will directly guide the vehicle to adjust the intensity of regenerative braking during driving, so as to achieve maximum energy recovery, i.e. output the optimal braking intensity. Specifically, the output optimal braking intensity is the regenerative braking intensity suggestion given according to the current state, which is in units of Newton (N).

[0084] In some embodiments, according to the current speed, acceleration, angular velocity of the vehicle, and the estimated road slope and driving resistance value output by the road slope and driving resistance estimation model, the required braking force of the current vehicle is calculated, and according to the calculated braking force requirement and the real-time state data of the vehicle, a suitable braking strategy is selected. For example, when driving on a downhill road and the driving resistance is large, the braking force is appropriately increased to maintain the stable speed of the vehicle; when driving on a flat road and the resistance is small, the braking force is appropriately reduced to improve the energy recovery efficiency.

[0085] The application adopts DQN (Deep Q-Network) in deep reinforcement learning as a regenerative braking strategy optimization model. DQN can make intelligent decisions in complex environments by learning state-action value functions.

[0086] DQN includes a state space, an action space and a reward function. The estimated road slope, the estimated driving resistance value and the vehicle state data are input into a pre-trained recovery gain regulation model. The state space of DQN receives the input estimated road slope, estimated driving resistance value and vehicle state data. The action space of DQN performs calculation, wherein a plurality of regenerative braking strategy options can be defined in the action space, such as increasing regenerative braking force, reducing regenerative braking force, maintaining current braking force, etc., so as to flexibly adjust the recovery gain regulation model according to environmental changes. The reward function sets a reasonable reward mechanism, for example, gives a high reward when more energy is successfully recovered, and gives a penalty when the recovery efficiency is reduced, so as to guide the model to optimize the decision.

[0087] The specific training process of the recovery gain regulation model includes several stages of data preparation, model training and model optimization.

[0088] Data preparation: First, build a simulation environment based on real roads and traffic conditions to train in multiple scenarios. This environment can simulate different driving conditions, including highways, urban roads, complex intersections, etc. During the simulation, record the state, action and reward of the vehicle under different strategies to form a training dataset.

[0089] Model training: Use the experience replay mechanism to randomly select training samples from the training dataset to avoid the influence of sample correlation on model performance during training. Use the combination of target network and main network to improve training stability, and update the parameters of the target network regularly to ensure continuous optimization of the learning process.

[0090] Model optimization: Adjust the hyperparameters in DQN, such as learning rate, discount factor, exploration rate, etc., to optimize the learning effect of the model. Regularly evaluate the performance of the model under different strategies, compare the rewards and convergence speed under different parameter combinations, and select the best parameter configuration.

[0091] This application uses DQN to optimize the regenerative braking strategy. Through real-time learning and adjustment of the strategy, the model can automatically select the best regenerative braking scheme in different driving environments to maximize energy recovery efficiency. This method has high adaptability and can learn and optimize the behavior patterns of drivers.

[0092] S50: Energy recovery is performed on the vehicle according to the first regenerative braking control strategy.

[0093] When the first regenerative braking control strategy is determined, the motor is controlled to brake and recover energy.

[0094] In some embodiments of the present application, real-time state data of the vehicle, real-time road condition information of the road where the vehicle is located, and real-time environmental information of the environment where the vehicle is located are obtained, and the real-time state data, the real-time road condition information, and the real-time environmental information are input into a pre-trained multi-modal sensor data fusion model for feature fusion to obtain a plurality of fusion feature vectors, the fusion feature vectors being used to represent visual information around the vehicle, three-dimensional space information around the vehicle, and a motion state of the vehicle. The fusion feature vectors are input into a pre-trained road slope and resistance estimation model for calculation to obtain an estimated road slope and an estimated driving resistance value of the road where the vehicle is located. The estimated road slope, the estimated driving resistance value, and the vehicle state data are input into a pre-trained recovery gain regulation model for calculation to determine a first regenerative braking control strategy, and the vehicle is energy recovered according to the first regenerative braking control strategy. The present application fully considers the vehicle state and external environmental factors, and improves the energy recovery rate and the cruising range of the new energy vehicle in the driving process through the fusion and innovative application of the above models, thereby providing a longer cruising range for the new energy vehicle user.

[0095] In some embodiments, in order to improve the data quality and the effectiveness of subsequent analysis, after obtaining the real-time state data of the vehicle, the real-time road condition information of the road where the vehicle is located, and the real-time environmental information of the environment where the vehicle is located, a series of preprocessing steps are required. The acquired real-time road images are denoised, color corrected, and edge detected using Photoshop, OpenCV, or other image editing software to ensure the accuracy of the acquired image information, and the images are segmented to extract important information such as roads, vehicles, and pedestrians. The point cloud data generated by the laser radar is filtered to remove noise points and segmented on the ground to identify static and dynamic objects, and the processed point cloud data is converted into a unified coordinate system for easy fusion with other sensor data. The acceleration and angular velocity data in the acquired IMU data are low-pass filtered to eliminate high-frequency noise, ensure signal smoothness, and combined with GPS data, the time synchronization of the sensors is performed to ensure the consistency between the data sources. Specifically, the GPS time is used as the master clock, and other sensors are synchronized to this time through hardware or software methods, and NTP (Network Time Protocol) or PTP (Precision Time Protocol) is used to realize time synchronization within the local area network.

[0096] The pre-processed data will be output as structured information, including: processed real-time road image data, which identifies the type, length, width, number of lanes, road surface condition, traffic flow, surrounding facilities, traffic signs on the road, and pedestrian position on the road according to the real-time road image data; point cloud data generated by the laser radar, information after ground segmentation and object recognition; smooth acceleration, angular velocity and position information output by the IMU; real-time vehicle speed (hereinafter referred to as vehicle speed) data output by the vehicle speed sensor.

[0097] In some embodiments of the present application, as shown in Figure 2 After obtaining the real-time state data of the vehicle, the real-time road condition information of the road where the vehicle is located, and the real-time environmental information of the environment where the vehicle is located, the energy recovery control method further includes the following steps:

[0098] S11: Convert the real-time road image into a preset format image data.

[0099] Convert the pre-processed real-time road image into a preset format suitable for multi-modal sensor data fusion model input, such as RGB three channels.

[0100] S12: Project and convert the real-time laser radar point cloud data to obtain a two-dimensional image.

[0101] Similarly, project and convert the pre-processed real-time laser radar point cloud data to obtain a two-dimensional image.

[0102] S13: Time synchronize the real-time inertial sensor data and the real-time vehicle speed.

[0103] Similarly, time synchronize the pre-processed real-time IMU data and the real-time vehicle speed.

[0104] In some embodiments, after obtaining the real-time IMU data of the vehicle, the real-time vehicle speed, the real-time road image, and the real-time laser radar point cloud data, the real-time road image can be converted into RGB three channel image data through image processing software such as Photoshop, GIMP, or using programming languages such as Python, etc. The real-time laser radar point cloud data can be projected and converted to obtain a two-dimensional image using Python and Open3D library. The inertial sensor and vehicle speed sensor can be time synchronized by combining GPS data.

[0105] This embodiment can make the multi-dimensional data input into the multi-modal sensor data fusion model identifiable and processable by processing the real-time road image, laser radar point cloud data, and real-time IMU data, and real-time vehicle speed respectively.

[0106] In some embodiments of the present application, the multi-modal sensor data fusion model comprises a convolutional neural network layer and a long short-term memory network layer. As shown in Figure 3 The real-time state data, real-time traffic information and real-time environment information are input into the pre-trained multi-modal sensor data fusion model in S20 for feature fusion, which can include the following steps:

[0107] S21: The pre-set format image data and the two-dimensional image are input into the convolutional neural network layer for spatial feature extraction to obtain a first feature vector.

[0108] The pre-processed camera image and the point cloud data of the laser radar, i.e., the RGB three-channel image data and the two-dimensional image, are input into the convolutional neural network layer, the spatial features of the image are extracted through multiple convolutional layers, various features such as edges and corner points are captured through convolutional kernels of different sizes, the maximum pooling layer is used in the pooling layer to reduce the feature dimension while retaining important features and improve the calculation efficiency, and the feature vector output by the convolutional neural network layer is named as the first feature vector for distinction. The first feature vector is used to represent the visual information around the vehicle and the three-dimensional spatial information around the vehicle.

[0109] S22: The first feature vector, the time-synchronized real-time inertial sensor data and the real-time vehicle speed are input into the long short-term memory network layer for calculation to obtain a fusion feature vector.

[0110] The first feature vector output by the convolutional neural network layer, the time-synchronized real-time IMU data and the real-time vehicle speed are combined to form a comprehensive input, which is input into the long short-term memory network layer. The long short-term memory network layer effectively processes time series data and captures dynamic changing information such as the relationship between speed change and environmental state through memory cells and gating mechanisms, and calculates to obtain a fusion feature vector. These fusion feature vectors are used to represent the visual information around the vehicle, the three-dimensional spatial information around the vehicle and the motion state of the vehicle.

[0111] This embodiment uses convolutional neural network (CNN) and long short-term memory network (LSTM) for deep fusion of multi-modal sensor data, fully utilizes multi-source data such as camera, laser radar and IMU, and improves the accuracy of environmental perception. This fusion method enables the system to perceive the surrounding environment in real time, realizes efficient data fusion and time series analysis, and provides a comprehensive information basis for subsequent decision-making.

[0112] On the basis of integrating multiple models, a feedback mechanism is established to enable the above models to be continuously optimized according to actual effects.

[0113] In some embodiments of the present application, as shown in Figure 4As shown, after the fusion feature vector is input into the pre-trained road slope and resistance estimation model for calculation to obtain the estimated road slope and estimated driving resistance value of the road where the vehicle is located, the energy recovery control method can further include the following steps:

[0114] S31: When the difference between the estimated road slope and the actual road slope of the road where the vehicle is located is greater than a preset slope threshold, the model parameters of the road slope and resistance estimation model are adjusted.

[0115] The preset slope threshold is set in advance. After the estimated road slope output by the road slope and resistance estimation model is obtained, the estimated road slope is compared with the actual road slope of the road where the vehicle is located. When the difference between the estimated road slope and the actual road slope of the road where the vehicle is located is greater than the preset slope threshold, it indicates that the road slope and resistance estimation model is not accurate enough at this time, and the model parameters of the road slope and resistance estimation model need to be adjusted, such as adjusting the weights and biases of the RNN through the back propagation algorithm to optimize the prediction accuracy, and repeating the cycle until the difference between the estimated road slope and the actual road slope of the road where the vehicle is located is less than or equal to the preset slope threshold.

[0116] Similarly, a preset resistance threshold is set in advance. When the difference between the estimated driving resistance value and the actual driving resistance of the road where the vehicle is located is greater than the preset resistance threshold, the model parameters of the road slope and resistance estimation model are adjusted.

[0117] In this embodiment, the estimated road slope output by the model in real time is used to adjust the parameters related to the road slope section in the road slope and resistance estimation model, a reasonable preset slope threshold is set to evaluate the performance of the road slope and resistance estimation model, and the road slope and resistance estimation model can be continuously optimized, thereby avoiding the situation that the model performs well on the training data but performs poorly in actual application.

[0118] In some embodiments of the present application, as shown in Figure 5 As shown, after the vehicle is subjected to energy recovery according to the first regenerative braking control strategy, the energy recovery control method can further include the following steps:

[0119] S51: Calculate the actual recovery efficiency when the vehicle is subjected to energy recovery according to the first regenerative braking control strategy.

[0120] A preset recovery efficiency threshold is set in advance. After the first regenerative braking control strategy is determined, the actual recovery efficiency when the vehicle is subjected to energy recovery according to the first regenerative braking control strategy is calculated, and the actual recovery efficiency is compared with the preset recovery efficiency threshold. When the actual recovery efficiency is greater than or equal to the preset recovery efficiency threshold, the recovery gain control model meets the expected requirements and can be used continuously.

[0121] S52: Adjusting the model parameters of the recovery gain regulation model when the actual recovery efficiency is less than the preset recovery efficiency threshold.

[0122] When the actual recovery efficiency is less than the preset recovery efficiency threshold, the recovery gain regulation model does not meet the expected requirements at this time, and the policy parameters need to be optimized through the Actor-Critic structure to maximize the energy recovery efficiency. Specifically, Actor update: adjusting the policy network parameters to optimize the regenerative braking control strategy. Critic update: adjusting the value network parameters to evaluate the pros and cons of the current policy and provide feedback.

[0123] After determining the first regenerative braking control strategy, the embodiment sets a preset recovery efficiency threshold to calculate the size relationship between the actual recovery efficiency when the vehicle recovers energy with the first regenerative braking control strategy and the preset recovery efficiency threshold, which serves as the basis for adjusting the recovery gain regulation model. This provides a clear target for the recovery gain regulation model optimization and can quickly determine whether the model meets the actual demand.

[0124] The application also establishes performance index tests for model adjustment and update, such as energy recovery efficiency, response time, user satisfaction, etc. to evaluate the effect of the model. A reasonable threshold is set to ensure that the performance improvement after model update meets the expected requirements.

[0125] In addition, the application also establishes an automated model test process, based on real data for simulation testing, and compares the performance of different versions of the model to ensure the stability and consistency of the updated recovery gain regulation model.

[0126] In some embodiments of the application, as shown in Figure 6 After recovering energy from the vehicle according to the first regenerative braking control strategy, the above energy recovery control method can further include the following steps:

[0127] S60: Obtaining real-time driving behavior data and real-time driving preference data of the user.

[0128] Collect driving behavior related data through vehicle sensors and user interface, such as acceleration, braking frequency, speed change, etc. Record the user's driving preference data under different road conditions, such as expected value of regenerative braking force, braking frequency, etc. These data will be used to analyze the user's individualized needs under different working conditions.

[0129] S70: Determining the real-time driving style of the user according to the real-time driving behavior data and the real-time driving preference data.

[0130] The driving behavior data is labeled in combination with driving behavior data and user feedback. For example, the driver's braking preference is classified as gentle braking or sudden stop braking, and the acceleration preference is divided into gentle acceleration or fast acceleration, etc. Through machine learning algorithm, the user is divided into different driving style types, such as smooth driver and aggressive driver, so as to optimize the personalized recycling gain control model.

[0131] During the user driving, the user's driving behavior data changes are monitored in real time, for example, the user changes from habitually driving at low speed to driving at high speed, that is, the real-time driving style changes from smooth to aggressive. Through reinforcement learning algorithm, the system can continuously adapt to the adjustment strategy during driving, and enhance the personalized response ability. Each time the user preference changes, the recycling gain control model will automatically fine-tune to ensure that it is consistent with the user's expectations.

[0132] S80: Find a second regenerative braking control strategy that matches the real-time driving style in the pre-constructed driving style type library.

[0133] The pre-constructed driving style type library includes preset driving styles and preset regenerative braking control strategies corresponding to the preset driving styles. According to the user's driving style, the regenerative braking strategy is customized. For example, for users who are used to driving at low speed, the recycling gain control model will automatically reduce the intensity of sudden braking, thereby bringing a smoother driving experience.

[0134] In addition, according to different driving scenes, the user can be provided with multiple regenerative braking strategy options such as energy saving mode and sports mode, and the user can select the appropriate mode according to his own needs to realize a more personalized energy recycling scheme.

[0135] S90: Determine the target braking control strategy of the user according to the first regenerative braking control strategy and the second regenerative braking control strategy.

[0136] The second regenerative braking control strategy is a supplement and optimization of the first regenerative braking control strategy, rather than a completely independent strategy. When the first regenerative braking control strategy and the second regenerative braking control strategy are inconsistent, the target braking control strategy of the user is determined according to the difference between the first regenerative braking control strategy and the second regenerative braking control strategy.

[0137] S100: Energy recycling is performed on the vehicle according to the target braking control strategy.

[0138] The embodiment determines the real-time driving style of the user according to the real-time driving behavior data and the real-time driving preference data of the user, matches the second regenerative braking control strategy for the user according to the real-time driving style, and ensures that the final target braking control strategy is the regenerative braking control strategy generated by combining the real-time driving style and the current vehicle state, road information and environmental information, which can not only provide personalized driving experience, but also significantly improve the energy recovery efficiency, enhance the braking safety, reduce the mechanical wear, and improve the overall robustness and comprehensive performance of the recovery gain regulation model.

[0139] In some embodiments of the present application, the driving experience feedback of the user is regularly obtained through the user interface of the vehicle or the mobile phone application. For example, the satisfaction of the user with the smoothness, response speed and energy recovery efficiency of the regenerative braking strategy is investigated. The user feedback data is integrated into the feedback loop of the model optimization, so that the recovery gain regulation model can continuously improve the user experience while maintaining high energy recovery efficiency.

[0140] In some embodiments, the energy recovery efficiency, response time and user preference when different regenerative braking regulation strategies are used for energy recovery can also be regularly checked. Analysis of different data dimensions helps to find the factors affecting the performance of the regenerative braking regulation strategy, provides basis for model optimization, identifies the user behavior change trend and the performance of the recovery gain regulation model in extreme environments, and ensures that the recovery gain regulation model can maintain stable and efficient performance in various situations. In addition, by continuously collecting new data, the existing model is retrained, so that the multi-modal sensor data fusion model, the road slope and resistance estimation model, and the recovery gain regulation model are iteratively updated based on new data to maintain adaptability to new environments and user behaviors. In addition, the transfer learning technology can be combined to migrate the experience in the historical training model to new data, so that when facing new driving scenarios, each model can adapt and optimize more quickly.

[0141] In some embodiments of the present application, the target braking control strategy of the user is determined according to the first regenerative braking control strategy and the second regenerative braking control strategy, which can specifically include the following steps:

[0142] When the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is less than the preset difference, the first regenerative braking control strategy is corrected based on the second regenerative braking control strategy to determine the target braking control strategy of the user.

[0143] The first regenerative braking control strategy focuses on the safety and energy recovery efficiency of the vehicle, and the second regenerative braking control strategy focuses more on meeting the personalized needs of the user. A preset difference is set in advance. When the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is less than the preset difference, the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is not large, and the first regenerative braking control strategy is fine-tuned based on the driving behavior data and driving preference data of the user corresponding to the second regenerative braking control strategy, such as acceleration, deceleration habit, comfort demand, etc., so as to meet the actual needs of the user.

[0144] In some embodiments of the present application, the target braking control strategy of the user is determined according to the first regenerative braking control strategy and the second regenerative braking control strategy, and the specific steps can further include the following steps:

[0145] When the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is greater than or equal to the preset difference, one of the first regenerative braking control strategy and the second regenerative braking control strategy is determined as the target braking control strategy of the user based on a preset priority rule.

[0146] The preset priority rule is set in advance, and the preset priority rule can include priority conditions of energy recovery efficiency, user personalized needs, and vehicle safety. The strategy that ensures driving safety is preferentially selected, for example, if the user's personalized needs result in insufficient braking force, which may affect driving safety, the driving safety of the vehicle is preferentially ensured; if the safety and user needs are met, the energy recovery efficiency is optimized as much as possible; under the premise of safety, the personalized needs of the user are preferentially met, for example, if the user prefers a more gentle braking experience, the energy recovery strength can be appropriately reduced. When the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is greater than or equal to the preset difference, the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is large, and the fine-tuning of the first regenerative braking control strategy cannot meet the actual needs, and one of the first regenerative braking control strategy and the second regenerative braking control strategy is selected according to the preset priority rule set in advance.

[0147] In this embodiment, the preset priority rule is set, and when multiple options are faced, the optimal selection can be quickly determined. When multiple targets or needs conflict with each other, the priority rule can help solve the conflict and ensure that the recovery gain control model makes a decision on the higher priority target.

[0148] Exemplary device

[0149] As a second aspect of the present application, an energy recovery control device is also provided, as shown in Figure 7 The energy recovery control device comprises:

[0150] The data acquisition module 701 is configured to acquire real-time state data of the vehicle, real-time road condition information of a road where the vehicle is located, and real-time environmental information of an environment where the vehicle is located.

[0151] A plurality of sensor assemblies can be arranged on the vehicle, including a camera, a laser radar, an IMU, a vehicle speed sensor, and the like. The real-time state data of the vehicle, the real-time road condition information of the road where the vehicle is located, and the real-time environmental information of the environment where the vehicle is located can be detected through the sensor assemblies. The data acquisition module 701 is connected with the plurality of sensor assemblies on the vehicle, and can acquire the real-time state data of the vehicle, the real-time road condition information of the road where the vehicle is located, and the real-time environmental information of the environment where the vehicle is located.

[0152] The multi-modal sensor data fusion model 702 is configured to perform feature fusion on the real-time state data, the real-time road condition information, and the real-time environmental information, to obtain a plurality of fusion feature vectors. The fusion feature vectors are used to represent visual information around the vehicle, three-dimensional space information around the vehicle, and a motion state of the vehicle.

[0153] The multi-modal sensor data fusion model 702 processes data from different sensor assemblies, such as real-time road images, laser radars, IMU data, and the like, and generates feature vectors with visual information around the vehicle, three-dimensional space information around the vehicle, and a motion state of the vehicle. The multi-modal sensor data fusion model 702 provides input for the subsequent road slope and resistance estimation model 703.

[0154] The road slope and resistance estimation model 703 is configured to calculate the fusion feature vectors input into a pre-trained road slope and resistance estimation model, to obtain an estimated road slope and an estimated driving resistance value of the road where the vehicle is located.

[0155] The road slope and resistance estimation model 703 inputs the fusion feature vectors output from the multi-modal sensor data fusion model 702, processes the dependency of time series data through a plurality of hidden layers and gate units, ensures continuity estimation in a dynamic environment, and outputs the estimated road slope and the estimated driving resistance value of the road where the vehicle is located. The road slope and resistance estimation model 703 provides input for the subsequent regenerative gain regulation model 704 and provides a decision basis for optimal strategy selection.

[0156] The regenerative gain regulation model 704 is configured to calculate the estimated road slope, the estimated driving resistance value, and the real-time state data, to determine a first regenerative braking control strategy.

[0157] The recovery gain regulation model 704 is a decision layer of the present application, responsible for controlling the regenerative braking strategy in a real-time dynamic environment to maximize the energy recovery gain. The recovery gain regulation model 704 model makes the optimal regenerative braking adjustment strategy according to the estimated road slope and estimated driving resistance values of the road where the vehicle is located obtained by the road slope and resistance estimation model 703 and the real-time state data of the vehicle.

[0158] The energy recovery module 705 is configured to perform energy recovery on the vehicle according to the first regenerative braking control strategy.

[0159] The energy recovery control device provided in the embodiment belongs to the same application concept as the energy recovery control method provided in the above-mentioned embodiments of the present application, can execute the energy recovery control method provided in any of the above-mentioned embodiments of the present application, and has the corresponding functional units and beneficial effects of the energy recovery control method. The technical details not described in detail in the embodiment can refer to the specific processing content of the energy recovery control method provided in the above-mentioned embodiments of the present application, which will not be described here.

[0160] Exemplary vehicle

[0161] Next, as a third aspect of the present application, the present application also provides a vehicle, which comprises the energy recovery control device described above.

[0162] In the present application, the vehicle is a new energy vehicle, such as a pure electric vehicle or a hybrid electric vehicle.

[0163] Exemplary electronic device

[0164] Next, as a fourth aspect of the present application, the present application also provides an electronic device. Reference Figure 8 is made to the electronic device according to the embodiments of the present application.

[0165] Figure 8 FIG. 1 illustrates a structural block diagram of an electronic device according to an embodiment of the present application.

[0166] As Figure 8 shown, the electronic device 80 includes one or more processors 801 and a memory 802.

[0167] The processor 801 can be a central processing unit (CPU) or other form of processing unit having data processing and / or instruction execution capabilities, and can control other components in the electronic device 80 to perform desired functions.

[0168] The memory 802 can include one or more computer program products that can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory, for example, can include random access memory (RAM), cache memory, and / or the like. The non-volatile memory, for example, can include read only memory (ROM), hard disk, flash memory, and / or the like. The computer-readable storage media can store one or more computer program instructions implementing the energy recovery control method of various embodiments of the present application described above and / or other desired functions.

[0169] In one example, the electronic device 80 can further include an input device 803 and an output device 804, which are interconnected through a bus system and / or other forms of connection mechanisms (not shown).

[0170] When the electronic device is a stand-alone device, the input device 803 can be a communication network connector for receiving the acquired input signals from the first device and the second device.

[0171] In addition, the input device 803 can further include, for example, a keyboard, a mouse, and / or the like.

[0172] The output device 804 can output various information, including the determined distance information, direction information, and / or the like, to the outside. The output device 804 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and / or the like.

[0173] Of course, in order to simplify, Figure 8 Only some of the components in the electronic device 80 related to the present application are shown in the block diagram of FIG. 8, and components such as buses, input / output interfaces, and / or the like are omitted. In addition, the electronic device 80 can further include any other appropriate components according to the specific application.

[0174] Exemplary computer-readable storage medium

[0175] As a fifth aspect of the present application, a computer-readable storage medium is provided, which stores a computer program for executing the steps in the energy recovery control method of various embodiments described above.

[0176] The computer readable storage medium can be a combination of one or more computer readable media. The computer readable media can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium can include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include the following: an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0177] In addition to the methods and devices described above, embodiments of the present application can also be a computer program product including a computer program information, which, when executed by a processor, causes the processor to perform the steps of the energy recovery control method of various embodiments of the present application.

[0178] The computer program product can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++, etc., and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device or entirely on the remote computing device or server.

[0179] The basic principles of the present application are described above in conjunction with specific embodiments, but it should be noted that the advantages, advantages, effects, etc. mentioned in the present application are only examples and are not limiting, and these advantages, advantages, effects, etc. cannot be considered as the must-have of each embodiment of the present application. In addition, the above-mentioned specific details are only for the purpose of example and for the purpose of understanding, and are not limited to the above-mentioned specific details, and the above-mentioned details do not limit the present application to the must-use of the above-mentioned specific details.

[0180] The block diagrams of the devices, apparatuses, equipment, systems referred to in this application are only illustrative examples and are not intended to require or imply that the connection, arrangement, configuration must be as shown in the block diagrams. These devices, apparatuses, equipment, systems can be connected, arranged, configured in any way as those skilled in the art will recognize. Words such as "include", "contain", "have", and the like are open-ended words, mean "including but not limited to", and can be used interchangeably. The words "or" and "and" as used herein mean the word "and / or", and can be used interchangeably unless the context clearly indicates otherwise. The word "such as" as used herein means the phrase "such as but not limited to", and can be used interchangeably.

[0181] It is also to be noted that in the devices, apparatuses, and methods of the present application, each component or step can be decomposed and / or recombined. These decompositions and / or recombinations are to be considered as equivalents of the present application.

Claims

1. An energy recovery control method, characterized in that, The method includes: Acquire real-time vehicle status data, real-time road condition information of the road where the vehicle is located, and real-time environmental information of the environment in which the vehicle is located. The real-time status data, the real-time road condition information, and the real-time environmental information are input into a pre-trained multimodal sensor data fusion model to perform feature fusion, thereby obtaining multiple fused feature vectors. The fused feature vectors are used to represent the visual information around the vehicle, the three-dimensional spatial information around the vehicle, and the motion state of the vehicle. The fused feature vector is input into a pre-trained road slope and resistance estimation model for calculation to obtain the estimated road slope and estimated driving resistance value of the road where the vehicle is located. The estimated road gradient, the estimated driving resistance value, and the real-time status data are input into a pre-trained regenerative braking control model for calculation to determine the first regenerative braking control strategy. The vehicle recovers energy according to the first regenerative braking control strategy.

2. The energy recovery control method according to claim 1, characterized in that, The real-time status data includes the vehicle's real-time inertial sensor data and the vehicle's real-time speed; the real-time road condition information includes real-time road images; and the real-time environmental information includes real-time lidar point cloud data. After acquiring real-time vehicle status data, real-time road condition information of the road where the vehicle is located, and real-time environmental information of the environment in which the vehicle is located, the method further includes: Convert the real-time road images into image data in a preset format; The real-time lidar point cloud data is projected and transformed to obtain a two-dimensional image; The real-time inertial sensor data and the real-time vehicle speed are synchronized in time.

3. The energy recovery control method according to claim 2, characterized in that, The multimodal sensor data fusion model includes a convolutional neural network layer and a long short-term memory network layer; Specifically, the real-time status data, real-time traffic information, and real-time surrounding environment information are input into a pre-trained multimodal sensor data fusion model for feature fusion, including: The preset format image data and the two-dimensional image are input into the convolutional neural network layer for spatial feature extraction to obtain a first feature vector. The first feature vector is used to represent the visual information around the vehicle and the three-dimensional spatial information around the vehicle. The first feature vector, the real-time inertial sensor data after time synchronization, and the real-time vehicle speed are input into the long short-term memory network layer for calculation to obtain the fused feature vector.

4. The energy recovery control method according to claim 2, characterized in that, After inputting the fused feature vector into a pre-trained road slope and resistance estimation model for calculation to obtain the estimated road slope and estimated driving resistance values ​​of the road where the vehicle is located, the method further includes: When the difference between the estimated road slope and the actual road slope where the vehicle is located is greater than a preset slope threshold, the model parameters of the road slope and resistance estimation model are adjusted.

5. The energy recovery control method according to claim 1, characterized in that, After recovering energy from the vehicle according to the first regenerative braking control strategy, the method further includes: Calculate the actual energy recovery efficiency of the vehicle when performing energy recovery using the first regenerative braking control strategy; When the actual recovery efficiency is less than the preset recovery efficiency threshold, the model parameters of the recovery gain control model are adjusted.

6. The energy recovery control method according to claim 1, characterized in that, The method further includes: Acquire users' real-time driving behavior data and real-time driving preference data; The user's real-time driving style is determined based on the real-time driving behavior data and real-time driving preference data; The system searches a pre-built driving style type library for a second regenerative braking control strategy that matches the real-time driving style. The pre-built driving style type library includes preset driving styles and preset regenerative braking control strategies corresponding to the preset driving styles. The user's target braking control strategy is determined based on the first regenerative braking control strategy and the second regenerative braking control strategy. Energy recovery is performed on the vehicle according to the target braking control strategy.

7. The energy recovery control method according to claim 6, characterized in that, Determining the user's target braking control strategy based on the first regenerative braking control strategy and the second regenerative braking control strategy includes: When the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is less than a preset difference, the first regenerative braking control strategy is modified based on the second regenerative braking control strategy to determine the user's target braking control strategy.

8. The energy recovery control method according to claim 6, characterized in that, Determining the user's target braking control strategy based on the first regenerative braking control strategy and the second regenerative braking control strategy includes: When the difference between the first regenerative braking control strategy and the second regenerative braking control strategy is greater than or equal to a preset difference, one of the first regenerative braking control strategy and the second regenerative braking control strategy is determined as the user's target braking control strategy based on a preset priority rule.

9. An energy recovery control device, characterized in that, The device includes: The data acquisition module is used to acquire real-time vehicle status data, real-time road condition information of the road where the vehicle is located, and real-time environmental information of the environment in which the vehicle is located. A multimodal sensor data fusion model is used to fuse real-time status data, real-time road condition information, and real-time environmental information to obtain multiple fused feature vectors. The fused feature vectors are used to represent the visual information around the vehicle, the three-dimensional spatial information around the vehicle, and the motion state of the vehicle. A road slope and resistance estimation model is used to calculate the estimated road slope and estimated driving resistance value of the road where the vehicle is located by inputting the fused feature vector into the pre-trained road slope and resistance estimation model. A regenerative braking control model is used to calculate the estimated road gradient, the estimated driving resistance value, and vehicle state data to determine the first regenerative braking control strategy. An energy recovery module is used to recover energy from the vehicle according to the first regenerative braking control strategy.

10. A vehicle, characterized in that, include: The energy recovery control device according to claim 9.

Images