Electric vehicle regenerative braking control method based on fuzzy neural network

Abstract

An electric vehicle regenerative braking control method based on a fuzzy neural network belongs to the technical field of new energy vehicle control, and comprises the following steps: collecting vehicle operation state parameters, and preprocessing input variables; a fuzzy controller is constructed, the vehicle state parameters serve as input, a preliminary regenerative braking target value is output, and fuzzy decision of the braking strength is achieved; a neural network optimization module is designed, fuzzy controller output and vehicle response parameters serve as input, a control result is optimized in real time through an error back propagation algorithm, and dynamic self-adaptive adjustment of regenerative braking force is achieved; according to the optimized control quantity, a motor braking instruction is generated and executed, and needed regenerative braking output is achieved; meanwhile, feedback information is collected and used for training a neural network model, and a closed-loop learning mechanism is formed; and through continuous iterative optimization, comprehensive optimization of regenerative braking control in the aspects of energy recovery efficiency, driving comfort and safety is finally achieved, and a stable and effective regenerative braking adjustment scheme is formed.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy vehicle control, and particularly relates to a regenerative braking control method for electric vehicles based on a fuzzy neural network. Background Art

[0002] New energy vehicles have significant advantages such as zero emissions, high energy efficiency, and low noise, which contribute to reducing carbon emissions in the transportation field and optimizing the energy structure. Therefore, they are regarded as an important means to achieve this goal. As an important part of electric vehicles to improve energy utilization efficiency and extend the driving range, the regenerative braking system can recover kinetic energy during braking and convert it into electrical energy to be stored in the power battery. However, traditional regenerative braking control methods generally have problems such as low control accuracy, slow response, and unstable energy recovery effect. Especially in complex and variable actual working conditions, such as low-speed congestion, ramp sliding, or frequent starts and stops, the braking control is prone to mutations, hysteresis, or a decrease in energy recovery efficiency, which even affects driving safety and comfort.

[0003] In the prior art, a fuzzy controller based on fixed rules or a PID control algorithm based on a linear model is often used to achieve regenerative braking control. Although it has a certain practicality, its control strategy depends on preset models and manual experience, making it difficult to adapt to dynamic environmental changes, and its control robustness and adaptive ability are poor. In addition, some studies have proposed introducing a neural network model for intelligent optimization, but most use an offline training method and cannot update control parameters in real time, so their ability to handle sudden working conditions is limited.

[0004] Therefore, there is an urgent need for a regenerative braking control method with adaptive learning ability, which can fuse multi-source input information and achieve dynamic optimization, so as to improve the energy recovery efficiency and braking smoothness of electric vehicles under various typical working conditions. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention proposes a regenerative braking control method for electric vehicles based on a fuzzy neural network. This method combines the strong robustness of fuzzy control and the self-learning ability of the neural network, and can realize intelligent adjustment of the braking process under various actual driving conditions, effectively improving the energy recovery efficiency and ensuring the safety and comfort of vehicle braking.

[0006] To achieve the above object, the present invention is realized through the following technical solutions:

[0007] A regenerative braking control method for electric vehicles based on a fuzzy neural network, comprising the following steps:

[0008] Step 1, collect vehicle operation parameters, including vehicle speed, motor speed, state of charge SOC of the battery, braking pedal opening, vehicle acceleration and deceleration;

[0009] Step 2: Input the above parameters into the fuzzy controller, and output the preliminary regenerative braking target value based on the membership function and fuzzy inference rules;

[0010] Step 3: Input the output of the fuzzy controller and the actual operating state parameters into the neural network control module, optimize the control quantity based on the error backpropagation algorithm, and output the final regenerative braking instruction;

[0011] Step 4: Transmit the regenerative braking instruction to the drive motor controller to control the output of the motor braking torque;

[0012] Step 5: Update the neural network parameters based on the feedback information of vehicle deceleration, current, and voltage to achieve closed-loop self-learning control.

[0013] The beneficial effects achieved by the present invention are as follows: (1) Aiming at the problems existing in the regenerative braking control of existing electric vehicles, such as low energy recovery efficiency under low-speed conditions, obvious control hysteresis, and poor braking smoothness, a regenerative braking control method based on fuzzy neural network is proposed, which integrates the non-linear reasoning ability of fuzzy control and the self-learning characteristics of neural network, constructs an adaptive braking control strategy, and can achieve precise adjustment of regenerative braking torque, improving the energy recovery efficiency; (2) The proposed control method considers the multi-parameter coupling characteristics such as vehicle speed, motor speed, battery SOC, and braking intention, and optimizes the fuzzy control output in real time through the neural network, enhancing the adaptability of the system to complex working conditions and ensuring the safety and smoothness of the braking process; (3) Through the online training mechanism of the neural network, the control parameters can be dynamically adjusted according to the actual operation data, with the ability of continuous self-learning and model self-update, improving the intelligent level of the control strategy; (4) This method can realize the dynamic coordination of electric motor braking and mechanical braking, improve the energy recovery ability of the whole vehicle under typical working conditions such as low speed, slope, and congestion, effectively extend the driving range of electric vehicles, and improve driving comfort and braking safety; (5) This method has good engineering realizability and platform versatility, and can provide effective reference and technical support for the optimal design of energy management and intelligent control systems of new energy vehicles. Brief Description of the Drawings

[0014] Figure 1 It is the flow chart of the fuzzy neural network method in the embodiment of the present invention.

[0015] Figure 2 It is the structure diagram of the fuzzy neural network in the embodiment of the present invention.

[0016] Figure 3 It is the typical NEDC working condition diagram in the embodiment of the present invention.

[0017] Figure 4 It is the comparison curve graph of the motor torque change in the embodiment of the present invention.

[0018] Figure 5 It is the comparison curve graph of the battery current change in the embodiment of the present invention.

[0019] Figure 6 It is the SOC comparison graph under the NEDC cycle condition in the embodiment of the present invention.

[0020] Figure 7 It is the comparison graph of the recovered energy under the NEDC cycle condition in the embodiment of the present invention.

[0021] Figure 8 It is the comparison table of the recovery effects of each method in the embodiment of the present invention. Detailed implementation manners

[0022] The technical solution of the present invention will be further described in detail below in conjunction with the accompanying drawings of the specification.

[0023] The present invention is a regenerative braking control method for electric vehicles based on a fuzzy neural network, and the method includes the following steps.

[0024] Step 1: Collect vehicle operation parameters, including vehicle speed, motor speed, state of charge (SOC) of the battery, brake pedal opening, vehicle acceleration and deceleration, etc.

[0025] (1) Real-time collect key operation state parameters of the whole vehicle through a vehicle control unit (VCU).

[0026] (2) Filter and preprocess the above original signals, eliminate outliers and interference noises, and use median filtering or moving average algorithm to ensure data stability and real-time performance.

[0027] (3) Normalize the collected continuous data so that the subsequent fuzzy controller and neural network model can perform unified dimension input, improving the processing efficiency and accuracy of the algorithm.

[0028] (4) Combine the processed multi-source vehicle operation parameters into an input vector, which is used as the joint input of the fuzzy control system and the neural network controller, providing a complete and accurate working condition information basis for braking control decisions.

[0029] Step 2: Input the above parameters into the fuzzy controller, and output a preliminary regenerative braking target value based on the membership function and fuzzy inference rules.

[0030] (1) Analyze the maximum energy recovery efficiency. Assume that when the vehicle completes longitudinal dynamics and the road surface adhesion coefficient is the same, there is:

[0031] Jω f = r e F xf - ∑T f (1)

[0032] where, ω f represents the wheel angular velocity; r e represents the radius of the vehicle tire; F xf represents the longitudinal force of the vehicle tire; T f is the braking force.

[0033] After discretizing the above formula, we can get:

[0034]

[0035] In order to maximize the energy recovery efficiency while ensuring the braking safety of the electric vehicle, it is necessary to make the power generated by the motor participating in regenerative braking reach the maximum. Therefore, the function J opt is introduced for optimization:

[0036]

[0037] The constraint conditions of the system are:

[0038]

[0039] where, T re is the actual torque generated by the motor; T lim is the maximum braking torque allowed by the motor; SOC max is the maximum state of charge allowed for the battery to perform regenerative braking; z max is the maximum braking intensity to ensure braking safety and stability.

[0040] In summary, in order to maximize the energy recovery efficiency during the braking process of the electric vehicle, it is necessary to optimize the recovery ratio of regenerative braking, make its coefficient K as close as possible to the maximum value, so as to improve the overall energy recovery effect and the system operation efficiency.

[0041] (2) Construct a fuzzy inference rule database

[0042] (a) The braking intensity reflects the urgency of the vehicle during braking. According to the general braking principle, the smaller the braking intensity of the electric vehicle, the safer and more reliable the vehicle operation. At this time, full electric braking can be performed to improve the energy recovery efficiency during braking, but the overall braking force is relatively low and the recovery degree is limited. The greater the braking intensity of the electric vehicle, the higher the intervention degree of electric braking usually is, but the braking safety and stability also need to be considered at this time. The fuzzy threshold of the braking intensity z is set to (0, 1), and the fuzzy sets are (L, M, H).

[0043] (b) The vehicle speed directly reflects the rotational speed of the motor and simultaneously affects the regenerative braking torque of the motor. During the braking process, although the braking torque output by the motor is small at low vehicle speeds, the overall operation is more stable and safe; while at high speeds, the regenerative braking torque that can be recovered increases, and more energy can be obtained, but the corresponding safety risks will also increase. The fuzzy threshold of v is set to (0, 120), and the fuzzy set is (L, M, H).

[0044] (c) The battery SOC is an important factor affecting the recovery of regenerative braking energy. To avoid affecting the battery life, when the SOC level is high, charging should be minimized. While in the case of low SOC, the battery voltage is relatively low, and charging is safer and the energy recovery efficiency is higher at this time. The fuzzy threshold of SOC is set to (0, 1), and the fuzzy set is (L, M, H).

[0045] Step 3: Input the output of the fuzzy controller and the actual operating state parameters into the neural network control module, optimize the control quantity based on the error backpropagation algorithm, and output the final regenerative braking instruction. The structure of the fuzzy neural network controller is as Figure 2 shown.

[0046] (1) Construct the front-end network for fuzzy rule matching. The first layer of the front-end network is the input layer. It is used to input the accurate control variables in the system to the next layer as the input samples for the training of the fuzzy neural network controller. There are 3 neurons, which are:

[0047] x = [x1, x2, x3] T (5)

[0048] Among them, x1, x2, and x3 respectively represent the true input values of the braking intensity z, vehicle speed v, and battery SOC of the current vehicle.

[0049] The second layer is the membership function assignment layer, which is used to convert the accurate variables input in the first layer into fuzzy control variables. The Gaussian function is selected for the fuzzy processing method. The three input variables in the first layer are divided into 9 neurons in the second layer, which can be expressed as:

[0050]

[0051] Among them, i represents the serial number of the value generated by the nodes in the first layer, and j represents the number of fuzzy sets.

[0052] The third layer is the fuzzy inference layer, and its nodes respectively correspond to a certain fuzzy set in the previous layer, with a total of 27 nodes. According to the setting of the matching rules, the fitness values of the above 27 nodes are calculated:

[0053]

[0054] Among them, αj It represents the fitness of the j-th layer. i1, i2, and i3 respectively represent the numbers under the membership degree distribution standard of the second layer, and j = 27.

[0055] The fourth layer is responsible for normalization. To ensure the unified standard of the output variables, the number of generated nodes needs to be the same as that of the previous layer, as follows:

[0056]

[0057] (2) Construct a backend network for standardizing the output. The backend network is highly consistent with the neural network in structure and is used to process the normalization of parameters. It is divided into three layers in total.

[0058] The first layer is the input layer. A total of 4 processing nodes are designed in this layer, including the braking intensity z, vehicle speed v, battery SOC, and a constant 1.

[0059] The second layer is the hidden layer. Its nodes correspond to the number of fuzzy rules and there are also 27 nodes, which are used to calculate the backend values of the conditional rules, as:

[0060] y j = q j0 + q j1 x1 + q j2 x2 + q j3 x3 (9)

[0061] The third layer is the output layer. The output value after the fuzzy neural network calculation is the regenerative braking force distribution coefficient K.

[0062]

[0063] Step 4: Transmit the regenerative braking instruction to the drive motor controller to control its output of the motor braking torque.

[0064] (1) Generate a regenerative braking instruction according to the regenerative braking target value output by the fuzzy control module and the neural network optimization module. The instruction parameters include the target braking torque, response time, desired deceleration, etc.

[0065] (2) Send this regenerative braking instruction to the drive motor controller through the CAN bus as the input of the control command.

[0066] (3) After receiving the instruction, the drive motor controller dynamically adjusts the output of the motor braking torque according to the current motor operating state and the charging power limit allowed by the battery system to make it as close as possible to the target braking torque.

[0067] Step 5: Update the neural network parameters based on the feedback information such as vehicle deceleration, current, and voltage to achieve closed-loop self-learning control.

[0068] (1) Adaptive adjustment is performed on the center parameter c for describing the horizontal axis position of the membership function ij , the scale parameter σ representing the width of the fuzzy set ij , and the neuron connection weight q in the backend network ij . Figure 2 In ij , the center parameter c ij and the scale parameter σ ij are used to determine the shape of the membership function, controlling its position and width on the horizontal axis respectively, so they belong to the adaptive adjustment of membership parameters; while the connection weight q

[0069]

[0070] appears at the backend of the network and is used to weight the output rule results, belonging to the adaptive adjustment of weight parameters. These three types of parameters are continuously optimized through the learning process to improve the accuracy and generalization ability of the model. Further optimization is carried out for the self-tuning parameters of the fuzzy neural network, and the error function is defined as:

[0071] When calibrating the self-learning algorithm for the center parameter c ij of the front-end network and the scale parameter σ ij representing the width of the fuzzy set, the neuron connection weight q ij of the backend network can be fixed first, and only the front-end network is analyzed.

[0072] When the system transmits information forward, the information transmission process of each layer is analyzed as follows:

[0073] The first layer receives the input of the system:

[0074]

[0075] The second layer selects the Gaussian function for fuzzy processing of variables:

[0076]

[0077] where i = 1, 2, 3...n, j = 1, 2, 3...m.

[0078] The third layer performs the fuzzy inference process:

[0079]

[0080] where i1 = 1, 2, 3, 4, 5, i2 = 1, 2, 3, 4, 5, i3 = 1, 2, 3, j = 1...36.

[0081] The fourth layer performs defuzzification normalization:

[0082]

[0083] Among them, i1 = 1, 2, 3, 4, 5, i2 = 1, 2, 3, 4, 5, i3 = 1, 2, 3, and j = 1…36.

[0084] The fifth layer, after acting together with the backend network, outputs:

[0085]

[0086] When error backpropagation occurs, the correction formula for the weight coefficients of the fifth layer is:

[0087]

[0088] Among them, β is the learning rate of the system.

[0089] Therefore, the error feedback signal of the fifth layer is:

[0090]

[0091] At this time, the weight coefficients of the fifth layer with respect to the fourth layer satisfy:

[0092]

[0093] By sequentially passing forward in the same manner, we can obtain:

[0094]

[0095] When calculating the partial derivative of the intermediate deviation using multiplication for f 3 There is:

[0096]

[0097] Otherwise, in another case:

[0098]

[0099] The gradient of the membership degree layer parameters can be obtained from the above formula as:

[0100]

[0101] Considering the influence of the learning rate β, there is:

[0102]

[0103] Therefore, the self-tuning center parameter c ij in the front-end network and the scale parameter σ ij representing the width of the fuzzy set

[0104]

[0105] (2) Backend network q ij The self-tuning algorithm can be expressed as:

[0106]

[0107] Among them, i=1,2,3…n; j=1,2,3…m.

[0108] During the control process, the neural network continuously receives feedback data from the actual operation of the vehicle (such as actual braking deceleration, motor force, pedal response deviation, etc.), and uses it as a supervision signal to perform online corrections on the network parameters based on the above.

[0109] In this embodiment, the typical braking conditions of electric vehicles are taken as the research object, and the analysis is carried out in combination with the actual vehicle operating parameters. The typical NEDC driving conditions are selected, and the vehicle power system model, motor braking characteristics and battery charging and discharging capabilities are used for joint simulation. By building a regenerative braking control system model based on a fuzzy neural network in the MATLAB / Simulink platform, inputting the actual vehicle speed, motor speed, battery state of charge (SOC), brake pedal opening and other parameters, the energy recovery efficiency of the control method is quantitatively analyzed and compared and verified in the CRUISE platform.

[0110] The NEDC (New European Driving Cycle) operating condition selected in this embodiment is a standard test method widely used in Europe, China and Australia, and is often used to evaluate the fuel consumption of traditional fuel vehicles and the range of electric vehicles. The relationship between the vehicle speed and time of the NEDC operating condition is as follows: Figure 3 As shown in the figure, the test process is divided into five stages, including four urban cycle conditions (UDC) and one extra-urban cycle condition (EUDC). The urban cycle condition is designed to simulate a city environment with heavy traffic flow, with an average speed of 18.5km / h, a maximum speed of 50km / h, and a single cycle lasting 195 seconds; the extra-urban cycle condition reflects relatively smooth road conditions, with an average speed of 62km / h, a maximum speed of 120km / h, and a single cycle lasting 400 seconds. The entire NEDC test cycle totals 1180 seconds.

[0111] In this embodiment, two comparison models are set, namely, CRUISE built-in control method and traditional fuzzy control method. Figure 4 and Figure 5It can be seen that for the motor torque, in the braking condition, especially when the braking intensity is relatively large, this method can more accurately adjust the regenerative braking distribution ratio. On the premise of being within the maximum torque range of the vehicle and ensuring braking safety, a larger proportion of the braking force is distributed to the motor for energy recovery. For the battery current, the current fluctuation of this method is more significant during the energy recovery stage, and the maximum recovery current exceeds 50A, which is significantly higher than the other two control methods. This indicates that under the same working conditions, the fuzzy neural network control method can make full use of the braking ability of the motor, more effectively improve the current recovery ability, and contribute to increasing the cruising range of electric vehicles.

[0112] Comparison of SOC changes and energy recovery during the NEDC cycle conditions is as Figure 6 、 Figure 7 shown. Figure 8 It is a comparison table of the recovery effects. In a single NEDC cycle, based on the initial set SOC starting value of 75%, for the regenerative braking control method of electric vehicles based on fuzzy neural network, the SOC decreases from 75% to 66.62%, and the consumption decreases to 8.38%. Compared with the previous two methods, it realizes a power saving of 1.34% and 0.57% respectively, which is equivalent to an improvement of 13.79% and 6.37% respectively. In terms of energy recovery, the regenerative braking control method of electric vehicles based on fuzzy neural network increases the recovered energy to 1159.73 kJ, recovering 624.1 kJ and 269.7 kJ more energy respectively compared with the others.

[0113] The analysis results show that the regenerative braking control method based on fuzzy neural network proposed in this embodiment exhibits excellent control performance under the complex working conditions of typical electric vehicles. This method achieves a higher energy recovery rate and effectively extends the cruising range of the whole vehicle. At the same time, by finely adjusting the braking torque of the motor, it significantly improves the smoothness during the braking process of the whole vehicle, avoids problems such as braking mutations and torque jitters, and improves driving comfort and safety. In addition, the neural network module has the ability of online learning and dynamic adjustment, enabling the system to adaptively optimize the control strategy according to different driving behaviors and environmental changes, further enhancing the intelligence and robustness of the control system.

[0114] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. Any equivalent modification or change made by those of ordinary skill in the art according to the disclosure of the present invention shall be included in the protection scope recorded in the claims.

Claims

1. A regenerative braking control method for electric vehicles based on a fuzzy neural network, characterized in that: It includes the following steps: Step 1: Collect vehicle operation parameters, including vehicle speed, motor speed, state of charge (SOC) of the battery, brake pedal opening, vehicle acceleration and deceleration; Step 2: Input the above parameters into a fuzzy controller, and output a preliminary regenerative braking target value based on membership functions and fuzzy inference rules; Step 3: Input the output of the fuzzy controller and the actual operation state parameters into a neural network control module, optimize the control quantity based on the error backpropagation algorithm, and output a final regenerative braking instruction; Step 4: Transmit the regenerative braking instruction to the drive motor controller to control the output of its motor braking torque; Step 5: Update the neural network parameters based on the feedback information of vehicle deceleration, current and voltage to achieve closed-loop self-learning control.

2. The regenerative braking control method for an electric vehicle based on a fuzzy neural network according to claim 1, characterized in that: In Step 1, the vehicle-mounted controller (VCU) is used to collect key operation state parameters of the whole vehicle in real time, filter and preprocess the above original parameters, eliminate outliers and interference noise, and use median filtering or moving average algorithm to ensure data stability and real-time performance; Normalize the collected continuous data so that subsequent fuzzy controllers and neural network models can perform unified dimension input; compose the processed multi-source vehicle operation parameters into an input vector, which serves as the joint input of the fuzzy control system and the neural network controller, providing a complete and accurate working condition information basis for braking control decisions.

3. A regenerative braking control method for an electric vehicle based on a fuzzy neural network according to claim 1, characterized in that: In Step 2, the fuzzy controller adopts a three-input and one-output structure. The inputs are vehicle speed, braking intensity, and state of charge (SOC) of the battery respectively, and the output is the expected regenerative braking distribution ratio coefficient K value; the calculation and analysis of the maximum energy recovery efficiency are as follows: Assume that the vehicle completes longitudinal dynamics and the road surface adhesion coefficient is the same, then there is: Jω f = r e F xf -∑T f (1) Among them, J represents the moment of inertia of the wheel; ω f represents the angular velocity of the wheel; r e represents the radius of the vehicle tire; F xf represents the longitudinal force of the vehicle tire; T f is the braking force; After discretizing the above formula, we get: where k represents the discrete sampling point; △t represents the sampling time step; To maximize the energy recovery efficiency while ensuring the braking safety of electric vehicles and make the power generated by the motor participating in regenerative braking reach the maximum, the function J is introduced opt for optimization: The constraint conditions of the system are: Among them, T re is the actual torque generated by the motor; T lim is the maximum braking torque allowed for the motor; SOC max is the maximum state of charge that allows the battery to perform regenerative braking; z max is the maximum braking intensity to ensure braking safety and stability, and z is the braking intensity; In order to maximize the energy recovery efficiency during the braking process of an electric vehicle, it is necessary to optimize the recovery ratio of regenerative braking to make its coefficient K as close to the maximum value as possible.

4. A regenerative braking control method for an electric vehicle based on a fuzzy neural network according to claim 1, characterized in that: In Step 2, the fuzzy rules of the fuzzy controller are as follows: (a) The braking intensity reflects the emergency degree of the vehicle during braking. Set the fuzzy threshold of the braking intensity z to (0, 1), and the fuzzy sets are (L, M, H); (b) The vehicle speed reflects the speed of the motor and at the same time affects the regenerative braking torque of the motor. Set the fuzzy threshold of the vehicle speed v to (0, 120), and the fuzzy sets are (L, M, H); (c) The battery SOC affects the regenerative braking energy recovery. Set the fuzzy threshold of the SOC to (0, 1), and the fuzzy sets are (L, M, H).

5. A regenerative braking control method for an electric vehicle based on a fuzzy neural network according to claim 1, characterized in that: In Step 3, the neural network is a feedforward multi-layer perceptron neural network, which is divided into a front-end network for fuzzy rule matching and a back-end network for normalizing the output, and at least includes an input layer, a hidden layer and an output layer. The structure is as follows: The first layer of the front-end network is the input layer; it is used to input accurate control variables in the system to the next layer as input samples for training the fuzzy neural network controller. There are 3 neurons, which are: x = [x1, x2, x3] T (5) where x1, x2, and x3 respectively represent the true input values of the current vehicle's braking intensity z, vehicle speed v, and battery SOC; The second layer is the membership function distribution layer, which is used to convert the precise variables input by the first layer into fuzzy control variables; the Gaussian function is selected as the fuzzy processing method; the three input variables of the first layer are divided into 9 neurons in the second layer, which are expressed as: Among them, i represents the serial number of the value generated by the first-layer nodes, and j represents the number of fuzzy sets; represents the input variable x i For the fuzzy set A ij the membership function value; The third layer is the fuzzy inference layer, and its nodes respectively correspond to a certain fuzzy set in the previous layer, with a total of 27 nodes; according to the setting of the matching rules, the fitness values of the above 27 nodes are calculated: Among them, α j represents the fitness of the j-th layer. i1, i2, and i3 respectively represent the numbers under the membership degree distribution standard of the second layer, and j = 27; The fourth layer is responsible for normalization processing. To ensure the unified standard of the output variables, the number of generated nodes needs to be the same as that of the previous layer, as follows: The backend network is highly consistent with the neural network in structure and is used to process the normalization of parameters. It is divided into three layers in total; The first layer is the input layer; a total of 4 processing nodes are designed in this layer, including the braking intensity z, vehicle speed v, battery SOC, and a constant 1; The second layer is the hidden layer; its nodes correspond to the number of fuzzy rules, and there are also 27 nodes, which are used to calculate the backend values of the conditional rules, as: y j = q j0 + q j1 x1 + q j2 x2 + q j3 x3 (9) where q j0 represents the constant term, i.e., the bias term, of the backend function of the j-th fuzzy rule; q j1 , q j2 , q j3 are respectively the linear influence coefficients of the input variables x1, x2, and x3 on the output in the j-th fuzzy rule; The third layer is the output layer; the output value after the fuzzy neural network calculation is the regenerative braking force distribution coefficient K:

6. The regenerative braking control method for an electric vehicle based on a fuzzy neural network according to claim 1, characterized in that: In step 4, the neural network control module has the ability of online training. The training algorithm adopts the error backpropagation BP algorithm, and the weight parameters are dynamically updated according to the actual operation data. The calculation process is as follows: The fuzzy neural network relies on the error backpropagation algorithm to dynamically adjust the weights, thereby realizing the self-learning ability of the controller; the parameters for adaptive adjustment include the center parameter c used to describe the horizontal axis position of the membership function ij , the scale parameter σ representing the width of the fuzzy set ij , and the neuron connection weights q in the backend network ij ; Optimize the self-tuning parameters of the fuzzy neural network, and define the error function as: (What is the definition of the parameter r) (Reply: r represents the r-th training sample or data point, and the description has been added) Among them, t is the target output value expected by the system, y is the actual output value of the system; r represents the rth training sample; When calibrating the self-learning algorithm for the central parameter c ij of the front-end network and the scale parameter σ ij representing the width of the fuzzy set, first fix the neuron connection weights q ij of the back-end network and only analyze the front-end network; When the system transmits information forward, the information transmission process of each layer is analyzed as follows: The first layer, receiving the input of the system: (What are the definitions of the various parameters in the following formula) (Reply: f i 1 represents the output of the i-th neuron in the first layer, that is, the input data it receives, which is passed to the second layer for use; x i 0 represents the output of the i-th input variable in the 0th layer (input layer), indicating the initial input; xi represents the i-th original input variable of the system, such as braking intensity, vehicle speed, battery SOC, etc.; description added) where, f i 1 represents the output of the i-th neuron in the first layer; represents the output of the i-th input variable in the input layer; x i represents the i-th original input variable of the system; In the second layer, the Gaussian function is selected to perform the fuzzy processing of variables: (What are the definitions of each parameter in the following formula) (Reply: The central parameter cij of the front-end network and the scale parameter σij of the width of the fuzzy set are mentioned above, and the description of ij has also been added) Among them, i = 1, 2, 3...n, j = 1, 2, 3...m; respectively represent that the ith input node generates j fuzzy sets after fuzzy processing; In the third layer, the fuzzy inference process is carried out: where, i1 = 1, 2, 3, i2 = 1, 2, 3, i3 = 1, 2, 3, j = 1…27; represents the membership degree value of the first input variable x1 under the i1-th membership function in the second layer; x j represents the activation strength of the j-th fuzzy rule; In the fourth layer, defuzzification normalization is carried out: Among them, i1 = 1, 2, 3, i2 = 1, 2, 3, i3 = 1, 2, 3, j = 1...27, m is the upper limit of summation, that is, the number of fuzzy rules, m = 27; In the fifth layer, after acting together with the backend network, the output is: Among them, ω ij represents the weight coefficient between the i-th neuron in the fifth layer and the j-th neuron in the fourth layer; When error backpropagation occurs, the weight coefficient correction formula for the fifth layer is: Among them, β is the learning rate of the system; Therefore, the error feedback signal of the fifth layer is: At this time, the weight coefficient of the fifth layer for the fourth layer satisfies: By sequentially forwarding in the same way, we can get: where f j represents the weighted input of the j-th neuron in this layer; g j represents the activation output of the j-th neuron in this layer; When calculating the partial derivative of the intermediate deviation using multiplication for f 3 we have: where s ij represents the partial derivative of the output of the k-th node in the third layer with respect to the membership degree inputs of the i-th and j-th in the second time; represents the output of the membership function of the j-th fuzzy subset corresponding to the i-th input variable in the second layer; represents the output of the k-th node in the third layer of the fuzzy neural network; Otherwise, in another case: The gradient of the membership layer parameters can be obtained through the above formula, as: Considering the influence of the learning rate β, then there is: The central parameter c for self-tuning in the front-end network ij and the scale parameter σ representing the width of the fuzzy set ij The self-learning algorithm is expressed as: Backend network q ij The self-tuning algorithm of Among them, i = 1, 2, 3...n, j = 1, 2, 3...m; During the control process, the neural network continuously receives the feedback data from the actual operation of the vehicle and uses it as a supervision signal to online correct the network parameters according to the above.

7. A regenerative braking control method for an electric vehicle based on a fuzzy neural network according to claim 1, characterized in that: This control method is applicable to a variety of working conditions, including low-speed congestion conditions, high-speed cruise conditions, and frequent braking conditions, specifically as follows: (a) Low-speed congestion condition: In this condition, the vehicle starts and stops frequently, follows the vehicle in front at low speed, and brakes over short distances. At this time, the fuzzy neural network controller performs fine control according to the slightly changing vehicle speed and braking opening. The neural network optimizes the response curve of low-speed braking by learning historical data to achieve smooth regenerative braking at low speeds; (b) High-speed cruise condition: In this condition, the vehicle speed is high, braking is less frequent but the kinetic energy of each braking is large. At this time, the neural network can adaptively optimize the regenerative braking control curve at high rotational speed states. On the premise of not affecting the smoothness of high-speed driving, it improves the participation degree of regenerative braking to achieve long-distance energy recovery and increase the cruising range; (c) Frequent braking condition: In this condition, the vehicle needs to brake continuously multiple times. At this time, fuzzy control combines with the neural network to adjust the regenerative braking intensity and participation degree in real time, and dynamically adjusts the energy recovery ratio in combination with the battery SOC state to avoid overcharging or energy waste, and realizes a regenerative control strategy of alternating between strong and weak to improve the energy efficiency management ability under continuous braking.

Images