Braking control method, system and equipment for distributed driving electric vehicle and medium

Abstract

The invention relates to the technical field of automobile braking, and discloses a braking control method, system and equipment for a distributed driving electric vehicle and a medium. The method comprises the steps that for the distributed driving electric vehicle, state data of the vehicle during braking are obtained, and whether the braking intention of a driver during braking is conventional braking or emergency braking is recognized according to the state data; the current longitudinal speed and the current side slip angle of the vehicle are obtained; the braking intention of a driver, the current longitudinal speed of the vehicle and the difference between the current side slip angle of the vehicle and the expected side slip angle of the vehicle in the absolute stable state serve as input data, the input data are input into the trained braking control model, and the braking torque used for stably braking the vehicle is obtained; and the vehicle is stably braked according to the braking intention of the driver and the obtained braking torque. According to the braking control method of the distributed driving electric vehicle, the control effect of the braking stability of the vehicle can be improved.

Description

Technical Field

[0001] The present invention relates to the field of automobile braking technology, and in particular to a braking control method, system, equipment and medium for a distributed drive electric vehicle. Background Art

[0002] In recent years, automobile intelligence has become the main development direction of automobile technology, and advanced driver assistance functions above level L2 have been mass-produced and applied in mid-to-high-end models. As people's requirements for automobile quality become higher and higher, the safety and ride comfort of vehicles have attracted more and more attention from consumers. Among them, vehicle stability during braking is one of the key indicators affecting safety. Under urban conditions, vehicles need to decelerate and brake frequently, which is easy to cause traffic accidents. The existing braking system control method is far from meeting the high-standard control requirements of the rapidly developing intelligent and networked electric vehicles. The braking performance of the vehicle can be improved by using high-performance braking devices, but they are expensive and cannot be actually applied to mid-range mass-produced models. Therefore, it is of great significance to study the key issues of braking efficiency, stability and economy in the braking process of distributed drive electric vehicles and give full play to the advantages of wire control braking, which is of great significance to improving vehicle stability.

[0003] Traditional automobile braking systems use brakes to achieve vehicle braking. Most of the energy generated by the friction of the brake pads is lost in the form of heat energy, which consumes a lot of energy and affects the braking performance. In the distributed drive electric vehicle motor braking system, the kinetic energy generated during braking can be converted into electrical energy and stored in the battery through electro-hydraulic coordinated control of braking, realizing energy recycling and improving economic performance. Under urban conditions, frequent starting and stopping are required during driving, and most braking is low-intensity braking, so energy recovery in electro-hydraulic braking can play a greater role. The rotation of the four wheels of the distributed drive electric vehicle is precisely controlled by its wheel hub motor, and the control system is more flexible, making the vehicle safer and more stable. Therefore, the coordinated distribution of the braking torque of the four wheel hub motors meets the braking control requirements, which is the most critical technical issue of the braking system.

[0004] However, in order to achieve stable braking of distributed drive electric vehicles, the total braking force of the distributed drive electric vehicles is mainly adjusted, which does not take into account the actual situation of the distributed drive electric vehicles and the driver's braking intention, and the braking stability control effect is poor. Summary of the invention

[0005] The object of the present invention is to provide a brake control method, system, device and medium for a distributed drive electric vehicle, which can improve the brake stability control effect of the distributed drive electric vehicle.

[0006] To solve the above technical problems, an embodiment of the present invention provides a brake control method for a distributed drive electric vehicle, comprising the following steps:

[0007] Acquire the state data of the distributed drive electric vehicle during braking, and identify the driver's braking intention as regular braking or emergency braking according to the state data;

[0008] Obtain the current longitudinal speed and the current center of mass sideslip angle of the distributed drive electric vehicle;

[0009] The driver's braking intention, the current longitudinal speed of the distributed drive electric vehicle, and the difference between the current center of mass slip angle of the distributed drive electric vehicle and the expected center of mass slip angle when the distributed drive electric vehicle is in an absolutely stable state are input as input data to the trained braking control model; wherein the braking control model is obtained by training a convolutional neural network based on a fusion channel attention mechanism and a spatial attention mechanism;

[0010] A convolutional neural network is used to extract features from each input data, and a channel attention mechanism and a spatial attention mechanism are used to apply different channel attention weights and different spatial attention weights to the features of each extracted input data from the channel and space aspects respectively, and a braking torque for stable braking of a distributed drive electric vehicle is obtained by applying the channel attention weights and the spatial attention weights to the features of each input data;

[0011] According to the driver's braking intention and the obtained braking torque, the distributed drive electric vehicle is braked stably.

[0012] In some optional embodiments, the step of acquiring the state data of the distributed drive electric vehicle during braking, and identifying whether the driver's braking intention during braking is conventional braking or emergency braking according to the state data, includes:

[0013] Obtain various state data of the distributed drive electric vehicle during braking, and input all the state data into the trained intention recognition model; wherein the intention recognition model is trained based on the GRU neural network with fusion channel attention mechanism;

[0014] The GRU neural network is used to extract features of the state data, and according to the importance of each state data to the stable braking of the distributed drive electric vehicle, a channel attention mechanism is used to apply different channel attention weights to the features extracted from each state data from the channel aspect. The driver's braking intention is obtained through the features after applying the channel attention weights.

[0015] In some optional embodiments, the state data is feature extracted by the intention recognition model, and according to the importance of each state data to the stable braking of the distributed drive electric vehicle, a channel attention mechanism is used to apply different channel attention weights to the features extracted from each state data from the channel aspect, including:

[0016] For each state data, feature extraction is performed on the state data through the GRU neural network;

[0017] The features of the extracted state data are compressed through global average pooling to obtain features of state data of multiple different dimensions;

[0018] Based on the channel-dependent gate mechanism, the Sigmoid excitation function is used to excite the characteristics of the state data of each dimension;

[0019] The features of the multi-dimensional state data after excitation are restored to the features of the original dimensions through two fully connected layers to obtain the features of the multi-dimensional state data after applying the channel attention weights.

[0020] In some optional embodiments, obtaining the current longitudinal speed of the distributed drive electric vehicle includes:

[0021] Obtaining the longitudinal speed of the distributed drive electric vehicle on a horizontal road;

[0022] Obtain the slope of the road on which the distributed drive electric vehicle is currently traveling;

[0023] According to the slope of the current road surface on which the distributed drive electric vehicle is traveling, a Kalman filter is used to correct the longitudinal speed of the distributed drive electric vehicle on the horizontal road surface to obtain the current longitudinal speed of the distributed drive electric vehicle;

[0024] Among them, the gain matrix of the Kalman filter includes the gain values ​​of the four wheels of the distributed drive electric vehicle. When the Kalman filter is used to correct the longitudinal speed of the distributed drive electric vehicle on a horizontal road, if the difference between the actual longitudinal speed and the estimated longitudinal speed of the distributed drive electric vehicle exceeds the preset speed difference, or the difference between the wheel acceleration and the body acceleration of the distributed drive electric vehicle exceeds the preset acceleration difference, it is determined that the wheel of the distributed drive electric vehicle is slipping, and the gain value of the slipping wheel in the gain matrix of the Kalman filter is set to zero.

[0025] In some optional embodiments, the channel attention mechanism and the spatial attention mechanism are used to apply different channel attention weights and different spatial attention weights to the features extracted from each input data from the channel and space aspects respectively, including:

[0026] For each feature of input data, the global weight vector is initialized by global pooling;

[0027] The Sigmoid excitation function is used to excite the initialized global weight vector;

[0028] Expand the weight vector obtained by stimulating the initialized global weight vector so that the dimension of the weight vector is consistent with the dimension of the features of the input data;

[0029] The expanded weight vector is multiplied and added to the features of the input data to obtain the features of each input data after applying the channel attention weight;

[0030] For each feature of the input data, sum the features channel by channel to get the initialized global weight matrix;

[0031] The initialized global weight matrix is ​​passed through the L1 norm and Softmax function, and then the weighted average is taken to obtain the spatial attention weight matrix;

[0032] Expand the dimension of the spatial attention weight matrix by copying it to be consistent with the dimension of the input data features;

[0033] The expanded spatial attention weight matrix and the features of the input data are multiplied and added respectively to obtain the features of each input data after applying the spatial attention weight.

[0034] In some optional embodiments, the step of obtaining a braking torque for achieving stable braking of a distributed drive electric vehicle by applying a channel attention weight and a spatial attention weight to the features of each input data includes:

[0035] According to the gradient information in the features of each input data after the channel attention weights and the spatial attention weights are applied, different gradient attention weights are applied to the features of each input data after the channel attention weights and the spatial attention weights are applied;

[0036] The features of all input data after applying the gradient attention weights are fused to obtain a braking torque for stable braking of the distributed drive electric vehicle according to the fused features.

[0037] In some optional embodiments, the braking of the distributed drive electric vehicle includes motor braking and hydraulic braking according to the driver's braking intention and the obtained braking torque, and the stable braking of the distributed drive electric vehicle according to the driver's braking intention and the obtained braking torque includes:

[0038] If the driver's braking intention is conventional braking, when braking the distributed drive electric vehicle according to the obtained braking torque, motor braking is used first, and when the motor braking force reaches the limit, hydraulic braking is used to achieve stable braking of the distributed drive electric vehicle;

[0039] If the driver's braking intention is emergency braking, then when the distributed drive electric vehicle is braked according to the obtained braking torque, the braking force in the high frequency band is allocated to the motor braking, and the braking force in the low frequency band is allocated to the hydraulic braking, so that the distributed drive electric vehicle can be braked stably; or, when the distributed drive electric vehicle is braked according to the obtained braking torque, the motor braking and the hydraulic braking are used to coordinately control the braking of the distributed drive electric vehicle, and a model with the wheel speed as the control variable is used to track the control of the wheel slip rate during the braking process to prevent wheel locking and ensure stable braking of the distributed drive electric vehicle.

[0040] An embodiment of the present invention further provides a brake control system for a distributed drive electric vehicle, comprising:

[0041] An intention recognition module is used to obtain the state data of the distributed drive electric vehicle during braking, and to recognize whether the driver's braking intention during braking is regular braking or emergency braking according to the state data;

[0042] A parameter acquisition module, used to acquire the current longitudinal speed and the current center of mass sideslip angle of the distributed drive electric vehicle;

[0043] A model input module is used to input the driver's braking intention, the current longitudinal speed of the distributed drive electric vehicle, and the difference between the current center of mass sideslip angle of the distributed drive electric vehicle and the expected center of mass sideslip angle when the distributed drive electric vehicle is in an absolutely stable state as input data into a trained braking control model; wherein the braking control model is obtained by training a convolutional neural network based on a fusion channel attention mechanism and a spatial attention mechanism;

[0044] A model recognition module is used to extract features of each input data through a convolutional neural network, and use a channel attention mechanism and a spatial attention mechanism to apply different channel attention weights and different spatial attention weights to the features of each extracted input data from the channel and space aspects respectively, and obtain a braking torque for stable braking of a distributed drive electric vehicle by applying the channel attention weights and the spatial attention weights to the features of each input data;

[0045] The distributed drive electric vehicle braking module is used to perform stable braking on the distributed drive electric vehicle according to the driver's braking intention and the obtained braking torque.

[0046] An embodiment of the present invention also provides a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned distributed drive electric vehicle braking control method.

[0047] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-mentioned braking control method for a distributed drive electric vehicle.

[0048] The brake control method of the distributed drive electric vehicle provided by the present invention has at least the following beneficial effects:

[0049] The present invention not only takes into account the driver's braking intention (i.e., conventional braking or emergency braking), but also takes into account the current state of the distributed drive electric vehicle (i.e., the current longitudinal speed and the current center of mass slip angle), and when implementing the braking of the distributed drive electric vehicle, the driver's braking intention, the current state of the distributed drive electric vehicle, and the expected center of mass slip angle of the distributed drive electric vehicle in an absolutely stable state are combined to determine the braking torque that can enable the distributed drive electric vehicle to brake stably, and then brake by the determined braking torque, so that the distributed drive electric vehicle can be braked stably. In addition, since a variety of factors affecting the braking effect are taken into account, the stable braking effect of the distributed drive electric vehicle can be improved.

[0050] In addition, for each of the above factors affecting the braking effect, different channel attention weights and different spatial attention weights are applied to the characteristics of each data from the channel and space aspects respectively through the channel attention mechanism and the spatial attention mechanism. The characteristics of each data can be refined, making the output braking torque more accurate, further improving the stable braking effect of the distributed drive electric vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] One or more embodiments are exemplarily described by the pictures in the corresponding drawings, and these exemplary descriptions do not constitute limitations on the embodiments.

[0052] Figure 1 is a flow chart of a brake control method of a distributed drive electric vehicle provided according to an embodiment of the present invention;

[0053] Figure 2 is a schematic diagram of state data collection of a distributed drive electric vehicle provided according to an embodiment of the present invention;

[0054] Figure 3is a structural schematic diagram of an intent recognition model provided according to an embodiment of the present invention;

[0055] Figure 4 is a structural schematic diagram of a SE attention module provided according to an embodiment of the present invention;

[0056] Figure 5 is a schematic diagram of a vehicle speed estimation algorithm on a sloped road surface provided according to an embodiment of the present invention;

[0057] Figure 6 is a structural schematic diagram of a braking control model provided according to an embodiment of the present invention;

[0058] Figure 7 is a schematic diagram of a structure of a fusion multi-attention mechanism module provided according to an embodiment of the present invention;

[0059] Figure 8 is a schematic diagram of a braking force distribution correction provided according to an embodiment of the present invention;

[0060] Fig. 9 is a schematic diagram of an electro-hydraulic brake coordinated control provided according to an embodiment of the present invention;

[0061] Fig.10 is a schematic diagram of a frequency separation method provided according to an embodiment of the present invention;

[0062] Fig.11 It is a schematic diagram of an electro-hydraulic anti-lock braking control provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in the embodiments of the present invention, many technical details are proposed in order to enable the reader to better understand the present invention. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present invention can be implemented. The division of the following embodiments is for the convenience of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined and referenced with each other without contradiction.

[0064] In order to improve the braking safety of distributed drive electric vehicles, a large number of distributed drive electric vehicle active safety control technologies have been developed and applied. For example: a distributed drive electric vehicle stability control strategy based on adaptive robust control and hierarchical sliding mode control, which combines the advantages of adaptive robust control and hierarchical sliding mode methods, can achieve excellent tracking performance, but this method has poor ability to overcome external disturbances; through real-time control of the braking force distribution of distributed drive electric vehicles and the control of the lateral movement of distributed drive electric vehicles, a control method for improving the stability of distributed drive electric vehicles is proposed, called enhanced basic braking control, and the stability and handling performance of distributed drive electric vehicles are verified by actual vehicles, but this method has a single structure and is quite different from the actual driving conditions of distributed drive electric vehicles; a distributed drive electric vehicle lateral dynamics optimization control strategy based on driver modeling is used to avoid traffic accidents, using sensors and controllers to detect the driving conditions and driver behavior of distributed drive electric vehicles to predict potential accidents The paper proposes a sliding mode control strategy to improve the stability of distributed drive electric vehicles based on the kinematic model of distributed drive electric vehicles. The strategy improves the stability of distributed drive electric vehicles under different road conditions by real-time monitoring of the state of distributed drive electric vehicles and controlling the lateral force of distributed drive electric vehicles, but ignores the consideration of the dynamic model of distributed drive electric vehicles. Regarding two controllers: tracking controller and torque distribution controller, the tracking controller realizes the optimal control of the lateral motion of distributed drive electric vehicles through high-order matrices, while the torque distribution controller reduces slip and sideslip by balancing the lateral friction of the tires. A strategy to ensure the stability control of distributed drive electric vehicles is proposed. A five-degree-of-freedom distributed drive electric vehicle model is used for modeling, and a model predictive control (Model Practical Control) is designed based on this model. A distributed drive electric vehicle stability control strategy based on a linear model predictive control algorithm is proposed. The controller takes into account the lateral tire force, front and rear and left and right loads, and the front wheel angle, and optimizes the control input to improve the stability of the distributed drive electric vehicle. In order to improve the stability and smoothness of braking, a coordinated control strategy based on the motor and the electric power steering system (Electric Power Steering, EPS) is designed. A new type of automotive anti-lock braking method is adopted to combine the motor and hydraulic braking to achieve an efficient braking effect. When the distributed drive electric vehicle is driving on a road with a low road adhesion coefficient, the motor is used to brake. When the road adhesion coefficient increases, the braking torque of the motor will decrease accordingly. A braking control strategy based on fuzzy logic control is designed, and the distribution ratio of the motor regenerative braking torque and the hydraulic braking torque of each wheel is obtained by using fuzzy logic reasoning. Most of them use hub motor braking.The electric motor torque and hydraulic braking torque of the four wheels of the distributed drive electric vehicle can be operated separately, and the motor and hydraulic braking force distribution control method is studied to achieve the braking efficiency and stability of the distributed drive electric vehicle; a control method for the brake system of a distributed drive electric vehicle is designed, and a hierarchical control brake system is designed. In the upper control, a method based on a sliding model is applied to solve the coordination problem of the longitudinal force and lateral force generated by the car during braking. In the lower control, the least squares method is used to optimize the motor and hydraulic braking torque of the four wheels respectively. However, in practical applications, the real-time and stability of such high-order dynamic models need to be verified. ;

[0065] Therefore, current research mainly improves the braking stability of distributed drive electric vehicles by adjusting the total braking force, but pays less attention to the driver's actual braking intention and the actual operating conditions of distributed drive electric vehicles. It is unable to give full play to the advantages of human-machine co-driving of intelligent distributed drive electric vehicles and the advantages of decoupling of braking force of wire control brake system. At the same time, most of the existing methods only consider the control of conventional braking, and ignore the braking effect under complex and special working conditions. The effectiveness and actual effect of the method still need to be further verified.

[0066] With the advancement of automobile intelligence and networking, the requirements for the safety and stability of distributed drive electric vehicles are further improved, and the existing control methods cannot adapt to the new distributed drive electric vehicle braking scenarios. In view of the limitations of current braking technology, considering the braking process of distributed drive electric vehicles, the driver's braking intention and the state of distributed drive electric vehicles, a hierarchical electric vehicle braking stability control method considering the driver's braking intention and the state of distributed drive electric vehicles is proposed to improve the driving safety of distributed drive electric vehicles.

[0067] An embodiment of the present invention relates to a braking control method for a distributed drive electric vehicle. The implementation details of the braking control method for the distributed drive electric vehicle of this embodiment are specifically described below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for implementing this solution.

[0068] The specific process of the brake control method of the distributed drive electric vehicle of this embodiment can be as follows: Figure 1 As shown, including:

[0069] Step 101 , obtaining state data of the distributed drive electric vehicle during braking, and identifying whether the driver's braking intention during braking is conventional braking or emergency braking according to the state data.

[0070] Specifically, if Figure 2As shown, first, driving data is collected in real time through a driver-in-the-loop experiment: a distributed drive electric vehicle dynamics model (i.e., a Simulink distributed drive electric vehicle model) is built in the control host, a virtual driving scene is built in the PreScan software, and a display (i.e., a scene display module) is used to transmit driving images to the driver. The driver transmits the control signal to the control host in real time through the built control simulator, thereby realizing closed-loop control of the distributed drive electric vehicle in the virtual driving scene.

[0071] Through the above method, the state data of the distributed drive electric vehicle in a unified time series is collected, including the longitudinal speed of the distributed drive electric vehicle, the lateral speed of the distributed drive electric vehicle, the sideslip angle of the center of mass of the distributed drive electric vehicle, the yaw angular velocity of the distributed drive electric vehicle, the longitudinal acceleration of the distributed drive electric vehicle, the lateral acceleration of the distributed drive electric vehicle, the torque of the distributed drive electric vehicle, the steering wheel angle and the yaw angle of the distributed drive electric vehicle. Since the accuracy of the sensors used in the handling simulator varies in magnitude, if they are directly input into the model training, the convergence speed of the recognition model training will be reduced, and the training proportion of high-level sensor data will be affected. The collected data needs to be mapped to a unified scale, and the data needs to be calibrated after signal smoothing, that is, the data set is divided. Assuming that the brake is touched 2266 times in the state data collected above, it is determined through artificial distinction that there are 555 emergency brakes, 583 ordinary brakes, and 1128 no brakes.

[0072] After obtaining the status data of the distributed drive electric vehicle, Figure 3 The intention recognition model shown in the figure obtains the driver's braking intention; wherein the intention recognition model is obtained by training the gated recurrent unit (GRU) neural network integrated with the channel attention mechanism. Specifically, a variety of state data of the distributed drive electric vehicle during braking are obtained, and all state data are input into the trained intention recognition model; the state data are feature extracted through the GRU neural network, and according to the importance of each state data to the stable braking of the distributed drive electric vehicle, the channel attention mechanism is used to apply different channel attention weights to the features extracted from each state data from the channel aspect, and the driver's braking intention is obtained through the features after applying the channel attention weights.

[0073] The intent recognition model is constructed by the channel (Squeeze-and-Excitation, SE) attention mechanism and the gated recurrent unit (GRU) in the recurrent neural network. GRU can use historical input to extract information and has good performance in time series classification tasks. At the same time, in order to improve the efficiency of intent recognition, this embodiment introduces an attention mechanism. The essence of the attention mechanism is to achieve efficient allocation of resources, give high weights to important information, give low weights to invalid information, and modify the weights in real time during the training process. The SE attention mechanism is a channel attention module that has the advantage of strengthening the channel features of the input feature parameters and not changing the values ​​of the input feature parameters. Therefore, the GRU that integrates the SE attention mechanism has the following advantages: First, it effectively avoids information loss; second, the weight allocation of each input information is completed through the SE attention mechanism, and intent recognition can be performed more accurately.

[0074] In the specific implementation, for each state data, the GRU neural network is used to extract the features of the state data; the features of the extracted state data are compressed through global average pooling to obtain the features of state data of multiple dimensions; based on the channel-dependent gate mechanism, the Sigmoid excitation function is used to excite the features of the state data of each dimension; through two fully connected layers, the features of the state data of multiple dimensions after excitation are restored to the features of the original dimensions to obtain the features of the multiple state data after applying channel attention weights.

[0075] Among them, the structure of the SE attention module is as follows Figure 4 As shown in Figure 1, the feature U is first compressed. Since the filter processes the local receptive field and cannot use the context information, to avoid this problem, the SE attention module uses global average pooling to reduce the Q×V spatial dimension of feature U. The reduced information is represented as z c , and its calculation formula is:

[0076]

[0077] In the formula, F sq is the global average pooling, Q is the query, V is the value, u C is a low-dimensional global embedding.

[0078] After feature U is compressed, its dimensions change from Q×V×C to 1×V / 16×C, 1×V / 8×C, 1×V / 4×C, and 1×1×C, and then an excitation operation is performed on it. The excitation process uses a gate mechanism based on channel dependency, and uses the Sigmoid activation function to capture channel dependency. The activation function is as follows:

[0079]

[0080] In the formula, is the ReLU function, W1 is the dimension reduction layer parameter, W2 is the dimension increase layer parameter, r is the dimension reduction ratio, and z is the dimension increase and decrease layer coefficient.

[0081] In order to limit the complexity of the model and assist in generalization, two fully connected layers are used to restore the feature dimension to the original input feature size, and the activation function is used to obtain the final output. The activation function is as follows:

[0082]

[0083] In the formula, represents the output obtained by the activation function, s C and u C Both represent characteristic parameters, F scale (u C ,s C ) means to convert s C and u C Multiply channel by channel.

[0084] Among them, the GRU network simplifies the input gate, forget gate, and output gate into a reset gate and an update gate, and uses the candidate hidden layer state to replace the LSTM internal unit state and output unit state, reducing the parameters required to remember the current state from two to one, thereby reducing the required training parameters and speeding up the training speed.

[0085] h transmitted by the neuron at the previous moment t-1 and the input x of the current node t Take two gate states as two inputs.

[0086] Reset Gate t Used to calculate the amount of memory that needs to be retained before, update gate z t Used to calculate the updated content at the current moment, the expression is:

[0087]

[0088] In the formula, σ is the sigmoid function, w is the weight matrix, b is the bias, and h is t-1 is the neuron output at the previous moment, x t The input of the current node.

[0089] The output is calculated based on the current input and internal state. The expression is:

[0090]

[0091] In the formula, tanh represents the double tangent function, and the calculation symbol * represents element-wise multiplication. is the current output.

[0092] After the feature vector is obtained through neural network processing, the Softmax classification predictor calculates the classification probability of each category and finally obtains the classification result:

[0093] y=Soffmax(W SE +b SE )

[0094] Where W SE is the output weight, b SE is the compensation parameter, and y is the output result of the intention recognition model.

[0095] Step 102, obtaining the current longitudinal speed and the current center of mass sideslip angle of the distributed drive electric vehicle.

[0096] In one example, for the current longitudinal speed of the distributed drive electric vehicle, the longitudinal speed of the distributed drive electric vehicle on the horizontal road is first obtained, and then the slope of the road on which the distributed drive electric vehicle is currently traveling is obtained. The longitudinal speed of the distributed drive electric vehicle on the horizontal road is corrected by using the Kalman filter according to the slope of the road on which the distributed drive electric vehicle is currently traveling, so as to obtain the current longitudinal speed of the distributed drive electric vehicle. Wherein, the gain matrix of the Kalman filter includes the gain values ​​of the four wheels of the distributed drive electric vehicle. When the longitudinal speed of the distributed drive electric vehicle on the horizontal road is corrected by using the Kalman filter, if the difference between the actual longitudinal speed of the distributed drive electric vehicle and the estimated longitudinal speed exceeds the preset speed difference, or the difference between the wheel acceleration of the distributed drive electric vehicle and the body acceleration exceeds the preset acceleration difference, it is determined that the wheel of the distributed drive electric vehicle is slipping, and the gain value of the slipping wheel in the gain matrix of the Kalman filter is set to zero.

[0097] Specifically, the longitudinal speed of traditional distributed drive electric vehicles used for braking control can be obtained through the speed of the driven wheels, but distributed drive electric vehicles do not have driven wheels, so it is necessary to observe and estimate the longitudinal speed of the center of mass. Generally, the method based on kinematics and dynamic models is used to estimate the longitudinal speed of the center of mass, mainly using wheel speed and body acceleration information. When estimating the wheel speed of a certain wheel during braking, the braking pressure of the wheel is first reduced and maintained at a constant value within the same time period, but this may affect the overall braking effect. The longitudinal speed can be calculated by deriving the braking force, wheel vertical force and tire longitudinal slip stiffness during braking. The derivation process is as follows:

[0098] The road surface adhesion coefficient is:

[0099]

[0100] In the formula, F B is the braking force, FN is the vertical load on the tire.

[0101] Since the road adhesion coefficient is proportional to the slip rate, the slip rate expression can be obtained as:

[0102]

[0103] Where μ is the road surface adhesion coefficient, C λ is the tire longitudinal slip stiffness of the μ-λ curve segment.

[0104] Then according to the wheel speed v obtained by the wheel speed sensor w , we can get the speed of the vehicle when driving in a straight line, and its expression is:

[0105]

[0106] In the formula, v w is the wheel speed, C λ is the tire longitudinal slip stiffness of the μ-λ curve segment, F B is the braking force, F N is the tire vertical load

[0107] However, due to the deviation and noise in the sensor measurement, there is a cumulative error in the vehicle speed obtained in long-term operation. The observer based on the dynamic model has relatively low requirements for the sensor, but it is heavily dependent on the dynamic model of the distributed drive electric vehicle. In order to avoid the error of dynamic modeling, the wheel speed and body acceleration information of the four wheels of the distributed drive electric vehicle are used. In this embodiment, a longitudinal speed estimation method based on slope recognition is designed to compensate for the longitudinal speed in order to improve the accuracy of the longitudinal speed estimation.

[0108] Specifically, the relationship between the longitudinal speed and velocity of a distributed drive electric vehicle can be expressed by the following measurement equation:

[0109] [v ij ]=[1]v(x)+[e ij (k)]

[0110] In the formula, v(x) is the vehicle speed at the previous moment, v lf k, v lr k, v rf k, v rr k are the centroid speeds of the four wheels, which are approximated as the product of the wheel speed and the rolling radius, e ij kij=lf,lr,rf,rr, represents the measurement error of the corresponding speed.

[0111] The state equation is set as:

[0112] x(k+1)=A(k)x(k)+B(k)u(k)

[0113] In the formula, the vehicle speed prediction estimate x(k+1) and the vehicle speed estimate x(k) are both one-dimensional data, so the state transfer matrix A(k)=[1] is taken, and the input is the product of the acceleration value u(k) and the sampling time t, so B(k)=[t].

[0114] The model parameters are:

[0115] v x (k+1)=v x (k)+t×a(k)

[0116] In the formula, v x (k+1) is the estimated value of vehicle speed after the parameters are introduced, v x (k) is the estimated value of the vehicle speed with the input parameters, t is the sampling time, and a(k) is the acceleration value with the input parameters.

[0117] The center-of-mass velocities of the four wheels are taken as observation values. In addition, an acceleration sensor is added to the vehicle body to measure the body acceleration, and the measured data is used as the input vector.

[0118] In the center of mass longitudinal velocity estimation model, the measurement equation can be expressed as:

[0119] Y(k)=H(k)x(k)+E(k)

[0120] Where H(k) is the measurement matrix and E(k) is the estimation error.

[0121] The linear Kalman filter equation is:

[0122] K k =P k H k (H k R k H k +R k )

[0123] X k =X k +K k (Z K -H k X k )

[0124] X k =X k +K k (Z K -H k X k )

[0125] In the formula, Kk is the Kalman gain matrix, H k is the observation matrix, R k is the observation error covariance matrix; P k and X k are the estimation error covariance and prediction error covariance matrices, respectively, is the Kalman gain error covariance.

[0126] Therefore, the recursive equations can be expressed as:

[0127] P - (k) = P (k-1) + Q

[0128] K(k)=P - (k)H T (HP - (k)H T +R)

[0129]

[0130]

[0131]

[0132] Where Kk is the Kalman gain matrix at time t, P k and P - k is the estimation error covariance matrix and prediction error covariance matrix at time t, Q is the system model error covariance matrix, τ is the sensor signal sampling time, v x k+1 and They are the estimated speed and predicted speed at time t+1, is the actual vehicle speed at time t+1.

[0133] By using the above Kalman filter, the vehicle speed can be well estimated when the adhesion is good. If one wheel suddenly slips or several wheels slip at the same time, if the recursive relationship is not adjusted, there will be a large error between the estimated vehicle speed and the actual vehicle speed. Therefore, this embodiment adds the adjustment of the gain matrix K on the basis of the conventional Kalman filter method.

[0134] The expression of K is:

[0135] K=[k lf k lr k rf k rr ]

[0136] In the formula, k lf , k lr , k rf , k rrRespectively represent the gain values ​​corresponding to the four wheels.

[0137] In the specific implementation, when the wheels of the distributed drive electric vehicle slip, the gain matrix K should be adjusted in real time so that the corresponding element values ​​of each distributed drive electric vehicle are calculated accordingly. When the wheel does not slip, the analysis and calculation are performed according to the Kalman filter recursive formula. The rule for judging whether the wheel is slipping is: the difference between the center of mass speed and the estimated speed exceeds the maximum speed difference; the difference between the wheel acceleration and the body acceleration exceeds the maximum acceleration difference. If any of the standard conditions is met, the K matrix element corresponding to the wheel is set to zero.

[0138] Based on the above speed estimation results on horizontal roads, a speed estimation algorithm on sloped roads is proposed. The algorithm flow is as follows: Figure 5 As shown in the figure: When estimating the slope using a single-axis deceleration sensor and four wheel speed sensors, the single-axis deceleration sensor is only sensitive to the inertial force in one direction. When driving on a sloped road, the deceleration sensor measurement signal contains motion deceleration information and slope information. In theory, the state quantity measured by the deceleration sensor is:

[0139] a m = ag × sinθ

[0140] In the formula, a m The sensor measures deceleration, a is the deceleration, and g is the acceleration due to gravity.

[0141] When performing slope identification, the slope information in the deceleration sensor must be separated. First, the actual deceleration a must be estimated. When the wheels are not completely slipping, the first-order derivative is calculated to estimate the vehicle body deceleration:

[0142] a=ω×r

[0143] Where a is the estimated deceleration, ω is the first-order derivative of the wheel speed, i.e., the wheel acceleration, and r is the wheel rolling radius.

[0144] Therefore, the estimated deceleration is obtained by differentiation, and compared with the actual deceleration, the two are highly consistent and can be used as a reference for slope estimation. Then, the slope is obtained using the deceleration sensor information.

[0145] In summary, the slope angle can be obtained as:

[0146]

[0147] Finally, based on the above slope result, the longitudinal speed of the distributed drive electric vehicle on the horizontal road is corrected to obtain the current longitudinal speed of the distributed drive electric vehicle.

[0148] Step 103, the driver's braking intention, the current longitudinal speed of the distributed drive electric vehicle, and the difference between the current center of mass sideslip angle of the distributed drive electric vehicle and the expected center of mass sideslip angle when the distributed drive electric vehicle is in an absolutely stable state are all used as input data and input into the trained braking control model; wherein the braking control model is obtained based on the convolutional neural network training that integrates the channel attention mechanism and the spatial attention mechanism.

[0149] Step 104, extracting features of each input data through a convolutional neural network, and applying different channel attention weights and different spatial attention weights to the features of each extracted input data from the channel and space aspects respectively using a channel attention mechanism and a spatial attention mechanism, and obtaining a braking torque for stable braking of a distributed drive electric vehicle by applying the channel attention weights and the spatial attention weights to the features of each input data.

[0150] Specifically, for each feature of input data, an initialized global weight vector is obtained through global pooling; the initialized global weight vector is excited by the Sigmoid excitation function; the weight vector obtained by exciting the initialized global weight vector is expanded so that the dimension of the weight vector is consistent with the dimension of the feature of the input data; the expanded weight vector is multiplied and added with the feature of the input data respectively to obtain the feature of each input data after applying the channel attention weight. At the same time, for each feature of the input data, the features are summed channel by channel to obtain the initialized global weight matrix; the initialized global weight matrix is ​​subjected to the L1 norm and the Softmax function and then the weighted average is taken to obtain the spatial attention weight matrix; the dimension of the spatial attention weight matrix is ​​expanded to be consistent with the dimension of the feature of the input data by copying; the expanded spatial attention weight matrix is ​​multiplied and added with the feature of the input data respectively to obtain the feature of each input data after applying the spatial attention weight.

[0151] In one example, after applying channel attention weights and spatial attention weights to the features of each input data, different gradient attention weights are applied to the features of each input data after applying the channel attention weights and spatial attention weights according to the gradient information in the features of each input data after applying the channel attention weights and spatial attention weights, and then the features of all the input data after applying the gradient attention weights are fused to obtain the braking torque for stable braking of the distributed drive electric vehicle based on the fused features.

[0152] In the specific implementation, the structure of the braking control model is as follows Figure 6As shown in the figure, the framework mainly includes three parts, namely the encoding network, the multi-attention mechanism module and the decoding network, in which the encoding network and the decoding network constitute a self-encoder-decoder network. First, the driver's braking intention, the longitudinal speed of the distributed drive electric vehicle, and the difference between the actual distributed drive electric vehicle's center of mass slip angle and the expected center of mass slip angle are increased from 3 to 64 through a convolutional layer, and then sent to an encoding network with four cascaded self-encoder blocks to extract features respectively; then, the extracted features are matched in the multi-attention mechanism module, and the matching information is fed to the decoding network containing six decoding blocks for optimal braking torque solution; finally, the braking torque that enables the distributed drive electric vehicle to brake stably is obtained.

[0153] Among them, since the braking system control algorithm belongs to feature-level fusion, the multi-attention mechanism module is the most important. In the framework of the braking system control algorithm integrating the multi-attention mechanism, this embodiment designs a fusion multi-attention mechanism integrating the channel attention mechanism and the spatial attention mechanism. This strategy cleverly uses the channel attention module and the spatial attention module to pay attention to the spatial and logical information in the braking characteristics of the distributed drive electric vehicle, so that the output braking torque contains the driver's real braking intention, the current longitudinal speed of the distributed drive electric vehicle, and the difference between the actual center of mass sideslip angle and the expected center of mass sideslip angle. In addition, the average fusion of the input features lacks pertinence. In the braking system control algorithm integrating the multi-attention mechanism, this embodiment also uses an adaptive distribution weight based on gradient information to calculate the gradient information richness according to the gradient information in the input features, and then the richness of the input features is calculated by weighted average to obtain the comprehensive weight. Based on the input feature gradient information, the output braking torque is more reasonable.

[0154] Below Figure 6 The brake control model shown is explained in detail:

[0155] (1) Self-encoder-decoder network

[0156] The self-encoder-decoder network consists of four parts: preprocessing layer, encoding network, decoding network and post-processing layer. The encoding network contains four serially connected encoding blocks. Each encoding block has two convolutional layers. First, the input features are padded, then a convolution with a convolution kernel size of 3×3 is performed, and finally the neurons in the convolution are randomly inactivated to obtain the first convolutional layer model; the difference between the second convolutional layer and the first convolutional layer is that the convolution kernel of the latter is 1×1. There are 6 decoding blocks in the decoding network. In terms of overall structure, the decoding block is basically the same as the encoding block, except that it replaces the Dropout function in the convolutional layer with the Relu activation function. Among them, the six decoding blocks belong to three levels respectively. The bottom layer receives the most feature channels. The decoding blocks in the same layer are connected from left to right, and the decoding blocks in the previous layer receive the output of the next layer.

[0157] (2) Fusion of multiple attention mechanism modules

[0158] The input features are divided into multiple channels and sent to the fusion multi-attention mechanism module. Then the attention mechanism can help the decision-making system filter out key information from complex channels according to different tasks. Based on this feature, an effective attention module is proposed to prompt the convolutional neural network to apply different attention weights to features from both channel and space aspects, thereby refining the input features. Based on the advantages of the attention mechanism in input feature processing, this embodiment proposes a channel space attention feature fusion strategy based on information retention, such as Figure 7 As shown, it consists of three parts, namely, the channel attention mechanism module, the spatial attention mechanism module and the information retention weight.

[0159] Among them, when the driver's braking intention, the longitudinal speed of the distributed drive electric vehicle, and the difference between the actual center of mass sideslip angle and the expected center of mass sideslip angle are integrated, the intrinsic relationship between different channels in the three features, namely the channel attention weight matrix, is calculated by the channel attention mechanism module.

[0160] The input features are passed through a convolution layer that combines convolution and normalization, and then global pooling is used to initialize the global weight vector W. c (I k ), the calculation method is as follows:

[0161] W c (I k )=G(Φ i (I k ))

[0162] In the formula, I k(k∈{int, spe, dif}, i∈{1, 2, 3, 4}) int, spe, dif represent the features extracted by the four encoding blocks from the driver's braking intention, the longitudinal speed of the distributed drive electric vehicle, and the difference between the actual center of mass sideslip angle and the expected center of mass sideslip angle, the corresponding feature initialization global weights and source data.

[0163] Feature Φ i The dimension is C i ×V i ×Q i , C, W and H represent the number of channels, size and direction respectively. G(·) represents channel-by-channel average global pooling, and the size of the pooling kernel is v×q. The initial global weight vector is then passed through the Softmax function to obtain the weight vector corresponding to the three input features. In order to multiply the input features point by point, the weight vector is expanded to the same size as the input features, that is, v×q. Finally, the input features and the expanded weights are multiplied and added channel by channel to obtain the fused output features of the channel attention module, which is calculated as follows:

[0164]

[0165] Where W c (I int ) is the number of driver’s braking intention channels, Φ i (I int ) is the dimension of driver characteristics, W c (I spe ) is the channel number of vehicle speed, Φ i (I spe ) is the dimension of vehicle speed, W c (I dif ) is the number of vehicle center of mass side slip angle channels, Φ i (I dif ) is the dimension of the vehicle's center of mass sideslip angle, W c (I k )

[0166] To initialize the global weight vector, ε=0.0001 to prevent the denominator from being 0 and causing meaningless data.

[0167] Among them, the spatial attention mechanism module uses the spatial relationship of the features to generate a spatial attention weight matrix, and applying a pooling operation along the channel axis can effectively highlight the information area. i , the channel attention mechanism module focuses on each feature vector C× x i ×y i The spatial attention mechanism module focuses on the relationship between C i The relationship within ×V×Q.

[0168] Different from the channel attention mechanism module, the spatial attention mechanism module first sums the features channel by channel as the initialization global weight matrix Calculation method:

[0169]

[0170] In the formula, To initialize the global weight matrix, is the sum of feature channels, and C is the dimension.

[0171] Then, the initialized global weight matrix is ​​passed through the L1 norm and Softmax function, and then the weighted average is taken to obtain the final spatial attention weight matrix. The dimension of the weight matrix is ​​expanded to C dimension by copying. Finally, the expanded weights and input features are multiplied and added to obtain the output features of the spatial attention module. The calculation method is as follows:

[0172]

[0173] In the formula, is the driver's braking intention attention weight matrix, is the speed attention weight matrix of the distributed driving electric vehicle, is the attention weight matrix of the sideslip angle of the center of mass of the distributed drive electric vehicle, Sum the weight vector.

[0174] The greater the information retention of one of the input features, the greater the impact of the feature on the final fused feature. Therefore, in order to fuse the output features of the channel attention mechanism module and the spatial attention mechanism module again, this embodiment proposes to estimate the information retention of the three fused features based on the gradient information, and then calculate the weight of the fusion of the three fused features based on the information retention. Compared with the entity derived from general information theory, the gradient is a measure based on the local spatial structure and is easy to calculate and store. Information retention G k The calculation method is as follows:

[0175]

[0176] In the formula, represents the features output by the kth attention mechanism module, k represents the channel or spatial attention mechanism module, represents the Laplace operator.

[0177] The fusion weight calculation formula of the output features of the two attention mechanism modules is as follows:

[0178] W c ,W s =SoftmaxG c ,G s

[0179] In the formula, the role of the Softmax function is to limit the two weights between 0 and 1, W c and W s They represent the weights of the channel attention mechanism module and the spatial attention mechanism module respectively.

[0180] Finally, the output relationship of the multi-attention mechanism fusion strategy is as follows:

[0181]

[0182] In the formula, is the output result of the multi-attention mechanism fusion strategy, W k is the weight matrix, Φ i O k It is the feature output by the kth attention mechanism module.

[0183] Step 105 : Stable braking is performed on the distributed drive electric vehicle according to the driver's braking intention and the acquired braking torque.

[0184] (1) Conventional braking scenario

[0185] Based on the proportion of hydraulic braking torque and electric motor braking torque in the braking system, this scenario selects a structure based on electro-hydraulic series braking. In the electro-hydraulic series braking structure, motor braking is used first. When the electric motor braking force reaches the limit, the distributed drive electric vehicle still cannot achieve the braking effect, and then the hydraulic braking force is used. Figure 8 The figure shows the electro-hydraulic brake coordinated control strategy, which inputs the brake control demand signal, distributes and corrects the braking force in time, and then compensates with the motor force to obtain the actual braking force.

[0186] When implementing the brake force distribution correction, in the control strategy that prioritizes motor braking, when the hydraulic braking force intervenes, the motor braking force is saturated and has no compensation capability. In order to improve the dynamic response of electro-hydraulic braking, the brake force distribution correction is performed to ensure that the motor has a certain ability to compensate for the hydraulic braking force. The specific process of this solution is as follows: Fig. 9 As shown. When the target braking intensity changes within a slightly smaller range, the wheel hub motor torque is used to effectively compensate the hydraulic brake actuator. However, when the rate of change is slightly larger, even if the front and rear motors can fully brake, the actual deceleration process cannot continue to track the expected deceleration due to the lag of the hydraulic brake, causing the hydraulic control system to quickly rise to the expected level after the delay, so further control instructions are required.

[0187] Specifically, as the target braking intensity increases, the motor input signal changes accordingly, approaching the ideal braking force distribution. When the hydraulic braking force is withdrawn, there is no motor braking torque saturation. When the desired braking deceleration increases, the signal input to the hydraulic brake actuator is three times that of conventional braking, and the input signal of the hub motor is changed accordingly, so that the hydraulic braking force is close to the required distribution ratio. The braking force is distributed when the target braking deceleration does not change. When the hydraulic brake is turned off, the motor braking torque will not be saturated.

[0188] When the regenerative braking force is withdrawn at low speed, the hydraulic braking force should be applied first. Assuming the wheel speed is ω and the delay time of the hydraulic system is T hd , the rise time is T r =2.2T h , the estimated time is T = T hd +T r When the wheel speed is lower than the critical region speed of the motor speed, the required braking force signal is designated and distributed to the hydraulic brake system to ensure that the hydraulic brake system performs effective braking before the motor power is reduced.

[0189] When performing motor force compensation, the actual motor and hydraulic braking forces on the front and rear axles are:

[0190] F bmreal =G m (s)F bm

[0191] F bhreal =G h (s)F bh

[0192] In the formula, F bmreal , F bhreal The actual motor and hydraulic braking forces on the front and rear axles, G m s, G h s are the transfer functions of the motor and hydraulic brake system, F bm , F bh They are the input signals for the motor and hydraulic brake system respectively.

[0193] The total braking force is:

[0194] F b =F bmreal +F bhreal

[0195] In the formula, F b is the required braking force on the axis, and its error is:

[0196] E=F b -(F bmreal +Fbhreal )

[0197] Combining them gives:

[0198] E=F b -G m (s)F bm -G h (s)F bh

[0199] Then, the motor brake system responds quickly, taking the input signal of the motor brake actuator as the control quantity. The input hydraulic braking force signal will interfere with the brake control system. The process of motor force compensation is as follows: Fig.10 shown.

[0200] set up:

[0201] E=F b -G m (s)F bm -G h (s)F bh

[0202] Substituting into E, we get

[0203]

[0204] To make the steady-state error equal to 0, let F b and F bh The coefficient of is 0, so:

[0205]

[0206]

[0207] (2) Emergency braking scenario

[0208] Under emergency braking conditions, the motor needs to have fast response and precise control to improve the performance of the anti-lock braking system. When the hydraulic brake is fully controllable, an electro-hydraulic brake coordination control method based on the frequency separation method is designed to achieve electro-hydraulic brake anti-lock control; in addition, the traditional hydraulic ABS system of the car is retained, and a motor regenerative braking and hydraulic brake coordination control method is also designed. Both of the above methods can effectively improve the anti-lock braking performance of distributed drive electric vehicles.

[0209] Among them, regarding the coordinated control of electro-hydraulic brakes based on the frequency separation method:

[0210] Although the hydraulic brake system can output a large braking torque, the response speed is relatively slow. The wheel hub motor has the opposite characteristics, with a fast response speed but a small output braking torque. Therefore, through the reasonable design of the controller, the brake system can use the motor and the hydraulic actuator to make up for each other's shortcomings and achieve the purpose of quickly tracking the target braking force. Fig.10 The schematic diagram of the frequency separation method is shown in Figure 2. Based on the frequency separation method, an electro-hydraulic braking force coordination controller is established to distribute the braking force in the low frequency band to the hydraulic braking system and the braking force in the high frequency band to the motor braking system, thus achieving coordinated distribution of the electro-hydraulic braking force.

[0211] Ignore the delay link of the motor and hydraulic system and simplify them into the corresponding inertia link model:

[0212]

[0213]

[0214] In the formula, τ m is the time constant of the motor characteristic transfer function, τ h is the time constant of the hydraulic brake characteristic transfer function, which can be selected as τ m =0.03, τ h =0.1.

[0215] In order to realize the function of tracking the target braking torque, G SISO The method for calculating s is:

[0216]

[0217] In the formula, τ h is the time constant of the hydraulic brake characteristic transfer function.

[0218] C motor s and C hyd s are designed into high-pass filter and low-pass filter according to the different frequency characteristics of the brake actuator, and the corresponding C motor s to achieve the tracking of target braking torque.

[0219] Substitute the inertia link model formulas of the motor and hydraulic system into G SISO s formula, we can get:

[0220]

[0221]

[0222] In the formula, τ m is the time constant of the motor characteristic transfer function, τ h is the time constant of the hydraulic brake characteristic transfer function, Cmotor s is the motor high-pass filter, C hyd s Hydraulic system low pass filter.

[0223] Among them, regarding the coordinated control of electro-hydraulic braking under emergency braking conditions:

[0224] The coordinated control of the motor brake and hydraulic brake system under emergency braking conditions can adopt the model tracking control method, which is essentially the control of the wheel slip rate by the anti-lock braking control system, specifically the control of the slip rate during braking by model tracking with the wheel speed as the control variable.

[0225] Assume that the equivalent mass of the wheel in the distributed drive electric vehicle model is m. When the slip rate is 0, the running speed of the distributed drive electric vehicle is equal to the longitudinal speed of the center of mass of the wheel. In the single-wheel model, it is also proportional to the wheel speed, that is:

[0226]

[0227] In the formula, τ m is the time constant of the motor characteristic transfer function, τ h is the time constant of the hydraulic brake characteristic transfer function, C motor s is the motor high-pass filter, C hyd s Hydraulic system low pass filter.

[0228] The dynamic equation of the two-degree-of-freedom single-wheel distributed drive electric vehicle model can be obtained:

[0229] mv x =-F d

[0230] J w ω=F d r b

[0231] In the formula, v x is the driving speed of the distributed drive electric vehicle, F d The braking force provided to the ground, J w is the wheel moment of inertia, ω is the wheel angular velocity, T b is the braking torque, and r is the rolling radius of the wheel.

[0232] Thus we can get:

[0233]

[0234] Based on the equivalent car model, the feedback control method is applied. When the distributed drive electric vehicle is in emergency braking, its output speed is much lower than the speed required by the standard mode, and the difference between the actual wheel speed and the ideal wheel speed is used as the control parameter to provide the motor with the corresponding braking torque control signal to prevent the wheel from locking. The block diagram of the coordinated control of the model tracking control and the hydraulic brake anti-lock system is shown in the figure. Fig.11 As shown in the figure. Since the hydraulic anti-lock braking system based on logic gate limitation is difficult to describe with differential equations, the expression of the feedback control system cannot be derived. Therefore, the feedback control system is set as a differential link by referring to the control method used for driving anti-skid, and an inertial filter link is added to the feedback loop to suppress high-frequency oscillation.

[0235] In this embodiment, not only the driver's braking intention (i.e., conventional braking or emergency braking) is taken into account, but also the current state of the distributed drive electric vehicle (i.e., the current longitudinal speed and the current center of mass slip angle) is taken into account. When implementing the braking of the distributed drive electric vehicle, the driver's braking intention, the current state of the distributed drive electric vehicle, and the expected center of mass slip angle of the distributed drive electric vehicle in an absolutely stable state are combined to determine the braking torque that can enable the distributed drive electric vehicle to brake stably. Then, the distributed drive electric vehicle can be braked by the determined braking torque. Stable braking of the distributed drive electric vehicle can be achieved. Moreover, since a variety of factors affecting the braking effect are taken into account, the stable braking effect of the distributed drive electric vehicle can be improved. In addition, for each of the above factors affecting the braking effect, different channel attention weights and different spatial attention weights are applied to the features of each data from the channel and space aspects respectively through the channel attention mechanism and the spatial attention mechanism, so that the features of each data can be refined, so that the output braking torque is more accurate, and the stable braking effect of the distributed drive electric vehicle is further improved.

[0236] The step division of the various methods above is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the protection scope of the present invention. Adding insignificant modifications or introducing insignificant designs to the algorithm or process without changing the core design of the algorithm and process are all within the protection scope of the invention.

[0237] Another embodiment of the present invention relates to a brake control system for a distributed drive electric vehicle. The implementation details of the brake control system for the distributed drive electric vehicle of this embodiment are described in detail below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this solution. The brake control system for the distributed drive electric vehicle of this embodiment includes: an intention recognition module, a parameter acquisition module, a model input module, a model recognition module and a distributed drive electric vehicle brake module.

[0238] Specifically, the intention recognition module is used to obtain the state data of the distributed drive electric vehicle during braking, and to recognize whether the driver's braking intention during braking is regular braking or emergency braking based on the state data.

[0239] The parameter acquisition module is used to obtain the current longitudinal speed and the current center of mass sideslip angle of the distributed drive electric vehicle.

[0240] The model input module is used to input the driver's braking intention, the current longitudinal speed of the distributed drive electric vehicle, and the difference between the current center of mass sideslip angle of the distributed drive electric vehicle and the expected center of mass sideslip angle when the distributed drive electric vehicle is in an absolutely stable state as input data into the trained braking control model; wherein the braking control model is obtained based on the convolutional neural network training that integrates the channel attention mechanism and the spatial attention mechanism.

[0241] The model recognition module is used to extract features of each input data through a convolutional neural network, and use a channel attention mechanism and a spatial attention mechanism to apply different channel attention weights and different spatial attention weights to the features of each extracted input data from the channel and space aspects respectively, and obtain the braking torque used to stabilize the braking of the distributed drive electric vehicle by applying the channel attention weights and the spatial attention weights to the features of each input data.

[0242] The distributed drive electric vehicle braking module is used to perform stable braking on the distributed drive electric vehicle according to the driver's braking intention and the obtained braking torque.

[0243] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment, and this embodiment can be implemented in conjunction with the above method embodiment. The relevant technical details and technical effects mentioned in the above embodiment are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiment.

[0244] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of the present invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by the present invention, but this does not mean that there are no other units in this embodiment.

[0245] Another embodiment of the present invention relates to a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the braking control method of the distributed drive electric vehicle in the above-mentioned embodiments.

[0246] Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium via an antenna, and further, the antenna also receives data and transmits the data to the processor.

[0247] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0248] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program, which implements the above method embodiment when executed by a processor.

[0249] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions for a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as: ROM), random access memory (Random Access Memory, referred to as: RAM), disk or optical disk and other media that can store program codes.

[0250] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present invention, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A brake control method for a distributed drive electric vehicle, characterized in that: include: Acquire the state data of the distributed drive electric vehicle during braking, and identify the driver's braking intention as regular braking or emergency braking according to the state data; Obtain the current longitudinal speed and the current center of mass sideslip angle of the distributed drive electric vehicle; The driver's braking intention, the current longitudinal speed of the distributed drive electric vehicle, and the difference between the current center of mass slip angle of the distributed drive electric vehicle and the expected center of mass slip angle when the distributed drive electric vehicle is in an absolutely stable state are input as input data to the trained braking control model; wherein the braking control model is obtained by training a convolutional neural network based on a fusion channel attention mechanism and a spatial attention mechanism; A convolutional neural network is used to extract features from each input data, and a channel attention mechanism and a spatial attention mechanism are used to apply different channel attention weights and different spatial attention weights to the features of each extracted input data from the channel and space aspects respectively, and a braking torque for stable braking of a distributed drive electric vehicle is obtained by applying the channel attention weights and the spatial attention weights to the features of each input data; The distributed drive electric vehicle is braked according to the driver's braking intention and the obtained braking torque.

2. The brake control method of a distributed drive electric vehicle according to claim 1, characterized in that: The step of obtaining the state data of the distributed drive electric vehicle during braking and identifying whether the driver's braking intention during braking is conventional braking or emergency braking according to the state data includes: Obtain various state data of the distributed drive electric vehicle during braking, and input all the state data into the trained intention recognition model; wherein the intention recognition model is trained based on the GRU neural network with fusion channel attention mechanism; The GRU neural network is used to extract features of the state data, and according to the importance of each state data to the stable braking of the distributed drive electric vehicle, a channel attention mechanism is used to apply different channel attention weights to the features extracted from each state data from the channel aspect. The driver's braking intention is obtained through the features after applying the channel attention weights.

3. The brake control method of a distributed drive electric vehicle according to claim 2, characterized in that: The state data is feature extracted by the GRU neural network, and according to the importance of each state data to the stable braking of the distributed drive electric vehicle, a channel attention mechanism is used to apply different channel attention weights to the features extracted from each state data from the channel aspect, including: For each state data, feature extraction is performed on the state data through the GRU neural network; The features of the extracted state data are compressed through global average pooling to obtain features of state data of multiple dimensions; Based on the channel-dependent gate mechanism, the Sigmoid excitation function is used to excite the characteristics of the state data of each dimension; The features of the multi-dimensional state data after excitation are restored to the features of the original dimensions through two fully connected layers to obtain the features of the multi-dimensional state data after applying the channel attention weights.

4. The brake control method of a distributed drive electric vehicle according to claim 1, characterized in that: The obtaining of the current longitudinal speed of the distributed drive electric vehicle comprises: Obtaining the longitudinal speed of the distributed drive electric vehicle on a horizontal road; Obtain the slope of the road on which the distributed drive electric vehicle is currently traveling; According to the slope of the current road surface on which the distributed drive electric vehicle is traveling, a Kalman filter is used to correct the longitudinal speed of the distributed drive electric vehicle on the horizontal road surface to obtain the current longitudinal speed of the distributed drive electric vehicle; Among them, the gain matrix of the Kalman filter includes the gain values ​​of the four wheels of the distributed drive electric vehicle. When the Kalman filter is used to correct the longitudinal speed of the distributed drive electric vehicle on a horizontal road, if the difference between the actual longitudinal speed and the estimated longitudinal speed of the distributed drive electric vehicle exceeds the preset speed difference, or the difference between the wheel acceleration and the body acceleration of the distributed drive electric vehicle exceeds the preset acceleration difference, it is determined that the wheel of the distributed drive electric vehicle is slipping, and the gain value of the slipping wheel in the gain matrix of the Kalman filter is set to zero.

5. The brake control method of a distributed drive electric vehicle according to claim 1, characterized in that: The channel attention mechanism and the spatial attention mechanism are used to apply different channel attention weights and different spatial attention weights to the features extracted from each input data from the channel and space aspects respectively, including: For each feature of input data, the global weight vector is initialized by global pooling; The Sigmoid excitation function is used to excite the initialized global weight vector; The weight vector obtained by stimulating the initialized global weight vector is expanded so that the dimension of the weight vector is consistent with the dimension of the feature of the input data; The expanded weight vector is multiplied and added to the features of the input data respectively to obtain the features of each input data after applying the channel attention weight; For each feature of the input data, sum the features channel by channel to get the initialized global weight matrix; The initialized global weight matrix is ​​passed through the L1 norm and Softmax function, and then the weighted average is taken to obtain the spatial attention weight matrix; Expand the dimension of the spatial attention weight matrix by copying it to be consistent with the dimension of the input data features; The expanded spatial attention weight matrix and the features of the input data are multiplied and added respectively to obtain the features of each input data after applying the spatial attention weight.

6. The brake control method of a distributed drive electric vehicle according to claim 5, characterized in that: The method of obtaining a braking torque for stable braking of a distributed drive electric vehicle by applying the characteristics of each input data after the channel attention weight and the spatial attention weight are applied comprises: According to the gradient information in the features of each input data after the channel attention weights and the spatial attention weights are applied, different gradient attention weights are applied to the features of each input data after the channel attention weights and the spatial attention weights are applied; The features of all input data after applying the gradient attention weights are fused to obtain a braking torque for stable braking of the distributed drive electric vehicle according to the fused features.

7. The brake control method of a distributed drive electric vehicle according to claim 1, characterized in that: The braking of the distributed drive electric vehicle includes motor braking and hydraulic braking. The stable braking of the distributed drive electric vehicle according to the braking intention of the driver and the obtained braking torque includes: If the driver's braking intention is conventional braking, when braking the distributed drive electric vehicle according to the obtained braking torque, motor braking is used first, and when the motor braking force reaches the limit, hydraulic braking is used to achieve stable braking of the distributed drive electric vehicle; If the driver's braking intention is emergency braking, then when the distributed drive electric vehicle is braked according to the obtained braking torque, the braking force in the high frequency band is allocated to the motor braking, and the braking force in the low frequency band is allocated to the hydraulic braking, so that the distributed drive electric vehicle can be braked stably; or, when the distributed drive electric vehicle is braked according to the obtained braking torque, the motor braking and the hydraulic braking are used to coordinately control the braking of the distributed drive electric vehicle, and a model with the wheel speed as the control variable is used to track the control of the wheel slip rate during the braking process to prevent wheel locking and ensure stable braking of the distributed drive electric vehicle.

8. A brake control system for a distributed drive electric vehicle, characterized in that: include: An intention recognition module is used to obtain the state data of the distributed drive electric vehicle during braking, and to recognize whether the driver's braking intention during braking is regular braking or emergency braking according to the state data; A parameter acquisition module, used to acquire the current longitudinal speed and the current center of mass sideslip angle of the distributed drive electric vehicle; A model input module is used to input the driver's braking intention, the current longitudinal speed of the distributed drive electric vehicle, and the difference between the current center of mass sideslip angle of the distributed drive electric vehicle and the expected center of mass sideslip angle when the distributed drive electric vehicle is in an absolutely stable state as input data into a trained braking control model; wherein the braking control model is obtained by training a convolutional neural network based on a fusion channel attention mechanism and a spatial attention mechanism; A model recognition module is used to extract features of each input data through a convolutional neural network, and use a channel attention mechanism and a spatial attention mechanism to apply different channel attention weights and different spatial attention weights to the features of each extracted input data from the channel and space aspects respectively, and obtain a braking torque for stable braking of a distributed drive electric vehicle by applying the channel attention weights and the spatial attention weights to the features of each input data; The distributed drive electric vehicle braking module is used to perform stable braking on the distributed drive electric vehicle according to the driver's braking intention and the obtained braking torque.

9. A computer device, characterized in that: include: at least one processor; And, a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the braking control method for a distributed drive electric vehicle as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the braking control method for a distributed drive electric vehicle as claimed in any one of claims 1 to 7 is implemented.

Images