Multi-objective optimization method and system for brake-by-wire system based on digital twinning

Abstract

The invention relates to the technical field of automobile braking, in particular to a multi-objective optimization method and system for a brake-by-wire system based on digital twinning, and the method comprises the steps: building a virtual platform which comprises a brake-by-wire system physical model and a digital twinning model, and enabling the virtual platform to integrate a calculation model, simulation data and a physical entity on the same platform; constructing a virtual-real interaction interface on the virtual platform to realize real-time interaction between physical model data and digital twin model data; performing multi-target optimization on the brake-by-wire system based on a depth deterministic strategy gradient algorithm, and coordinating friction braking and motor braking force distribution on the premise of ensuring braking safety according to the driving intention of a driver; a physical engine is adopted to simulate pressure fluctuation of a hydraulic pipeline in real time, the brake energy recovery amount and system response characteristics are optimized under the ABS activation working condition, a depth deterministic strategy gradient algorithm is adopted to perform multi-target optimization on a brake-by-wire system, and multiple targets such as energy recovery efficiency can be better balanced.

Description

Technical Field

[0001] The present invention relates to the technical field of automotive braking, and particularly to a multi-objective optimization method and system for a wire-controlled braking system based on digital twin, which is applicable to the braking energy recovery and braking control of electric vehicles, hybrid electric vehicles, and range-extended electric vehicles. Background Art

[0002] In recent years, with the rapid development of new energy vehicle technology, wire-controlled braking systems have been widely used in the field of electric vehicles due to their highly integrated and controllable characteristics. The wire-controlled braking system controls the motor and hydraulic system through an electronic control unit (ECU) to achieve intelligent distribution of braking force, and has higher energy recovery efficiency and more precise braking force control ability compared with traditional braking systems.

[0003] However, the existing wire-controlled braking systems still face the following technical challenges in practical applications: First, there is a contradiction between braking energy recovery efficiency and braking safety, and maximizing energy recovery while ensuring braking safety is a complex multi-objective optimization problem; Second, the dynamic characteristics of wire-controlled braking systems are complex under different working conditions, especially in the ABS (antilock braking system) activation condition, and the braking force distribution and control are more difficult; Third, traditional linear control strategies are difficult to adapt to complex and variable driving scenarios and lack the ability of adaptive adjustment.

[0004] At present, conventional wire-controlled braking systems mostly adopt rule-based control strategies or simple PID controllers, lacking accurate modeling and optimization of the system's dynamic characteristics. At the same time, the braking energy recovery strategy often adopts a fixed distribution ratio and cannot be dynamically adjusted according to real-time working conditions, resulting in low energy recovery efficiency. In addition, the existing systems have insufficient research on the braking force coordination control under the ABS activation condition and are difficult to maintain good energy recovery efficiency while ensuring safety.

[0005] Therefore, how to construct an accurate model of the wire-controlled braking system and achieve multi-objective optimization of braking energy recovery and braking safety based on this is a technical problem that needs to be solved urgently. Summary of the Invention

[0006] The object of the present invention is to provide a multi-objective optimization method and system for a wire-controlled braking system based on digital twin, aiming to solve the contradiction between braking energy recovery efficiency and braking safety of the existing wire-controlled braking system, construct a high-precision system model through digital twin technology, and achieve multi-objective optimization by combining the deep deterministic policy gradient algorithm, maximizing the energy recovery efficiency while ensuring braking safety and maintaining good braking comfort at the same time.

[0007] The present invention discloses a multi-objective optimization method for a wire-controlled braking system based on digital twin, including: Build a virtual platform that includes a physical model and a digital twin model of the electronic stability program (ESP) system, integrating a computational model, simulation data, and physical entities on the same platform; Construct a virtual-real interaction interface on the virtual platform to achieve real-time interaction between the physical model data and the digital twin model data; Perform multi-objective optimization on the electronic stability program (ESP) system based on the Deep Deterministic Policy Gradient (DDPG) algorithm, and coordinate the distribution of frictional braking and motor braking force according to the driver's driving intention while ensuring braking safety; Use a physical engine to real-time simulate the hydraulic pipeline pressure fluctuation, and optimize the braking energy recovery amount and system response characteristics under the condition of ABS activation.

[0008] Preferably, the building of the virtual platform that includes a physical model and a digital twin model of the electronic stability program (ESP) system includes: Build a physical model of the electronic stability program (ESP) system, which includes a vehicle kinematics model, a hydraulic model, and a wheel braking dynamics model; Build a digital twin model of the electronic stability program (ESP) system, which includes a digital environment, a virtual hydraulic system, and a virtual motor system; Import the physical model into a virtual laboratory built based on MATLAB and Simulink, and obtain the brake pedal displacement and hydraulic pipeline pressure data according to the physical model; Store the brake pedal displacement and hydraulic pipeline pressure data into the physical engine to obtain a training sample set for updating the digital twin model parameters.

[0009] Preferably, the vehicle kinematics model is used to obtain the total deceleration of the target vehicle; the hydraulic model is used to obtain the pressure value of the hydraulic cylinder; the wheel braking dynamics model is used to obtain the slip ratio of the wheels of the target vehicle and determine whether the wheels enter the ABS safety area.

[0010] Preferably, the construction of the virtual-real interaction interface includes: Configure a data acquisition interface from physical to digital to real-time collect sensing data such as brake pedal displacement, wheel speed, hydraulic pressure, motor speed, motor current, and vehicle acceleration; Implement a data preprocessing module to perform noise filtering, outlier detection, data compensation, and data synchronization on the collected sensing data; Configure a control execution interface from digital to physical to convert control instructions into motor torque instructions, brake master cylinder pressure instructions, and ABS adjustment valve control instructions, and implement the control through corresponding execution units.

[0011] Preferably, the multi-objective optimization of the electronic stability program (ESP) system based on the Deep Deterministic Policy Gradient (DDPG) algorithm includes: Design the control state space, including states related to braking demand, vehicle states, system states, and environmental states; Design the control action space, including motor braking torque commands, hydraulic braking pressure commands, and state switching flags; Design a multi-objective reward function, including terms for energy recovery efficiency, braking performance, comfort, safety, and system response; Construct an experience replay pool to store control experiences and improve the sample utilization efficiency based on a priority sampling strategy; Separate policy evaluation and improvement through a dual-network architecture to ensure the stability of the learning process.

[0012] Preferably, the weights of the multi-objective reward function adopt an adaptive adjustment mechanism to dynamically adjust the weight ratios of various indicators according to different braking conditions: In the case of emergency braking, increase the safety weight and reduce the energy recovery weight; In normal braking conditions, increase the energy recovery weight and maintain a moderate comfort weight; In the case of ABS activation, set the safety weight to the highest and the energy recovery weight to the lowest; In low-speed conditions, increase the comfort weight and the system response weight.

[0013] Preferably, the experience replay pool adopts a multi-level structure design, including: A short-term pool for storing recently generated experience samples, with a capacity of 100; A medium-term pool for storing experience samples with a medium time span, with a capacity of 500; A long-term pool for storing long-term valid experience samples, with a capacity of 2000; A priority calculation strategy based on temporal difference error to increase the sampling probability of rare condition samples and samples with rich gradient information.

[0014] Preferably, the real-time simulation of hydraulic pipeline pressure fluctuations using a physical engine includes: Use the PX-OSG plugin for simulation and implement real-time data exchange between the physical engine and the digital twin model based on the WebSocket protocol; Real-time simulate hydraulic pipeline pressure fluctuations according to physical model data and digital twin model data; The physical model data includes ABS pressure, friction braking wheel cylinder pressure, motor braking force, and system temperature; The digital twin data includes brake pedal displacement, motor torque, regenerative current, and pipeline pressure.

[0015] Preferably, the optimization of the braking energy recovery amount and the system response characteristics under the ABS activation condition includes: Coordinating the distribution of frictional braking and electric motor braking forces under the ABS activation condition, and adjusting the solenoid valve state and the hydraulic cylinder drive according to the change of the wheel cylinder pressure; When the wheel cylinder pressure rises and approaches the target value, open the solenoid valve to connect the oil circuit of the hydraulic cylinder control reservoir and the master cylinder; When the wheel cylinder pressure drops and the braking pressure is low, close the solenoid valve to disconnect the oil circuit between the reservoir and the master cylinder, and at the same time drive the hydraulic cylinder to adjust the oil pressure; Through the above control strategy, the braking energy recovery amount per 100 kilometers is increased by 19%, while keeping the deceleration fluctuation rate not exceeding 0.3g.

[0016] A multi-objective optimization system for a wire-controlled braking system based on digital twin, used to implement the above method, includes: A physical model establishment module, used to establish a physical model of the wire-controlled braking system, including a vehicle kinematics model, a hydraulic model, and a wheel braking dynamics model; A digital model building module, used to establish a digital twin model of the wire-controlled braking system, including a virtual laboratory construction unit, a training sample collection unit, a neural network model training unit, and a digital twin mapping unit; A virtual-real interaction interface module, used to realize the real-time interaction between the physical model data and the digital twin model data; An optimization target setting module, used to set and manage multiple optimization indexes for braking energy recovery and braking safety; A weight optimization module, used to determine and optimize the weight ratio of each optimization index to achieve multi-objective balance; A real-time optimization module, used to search for the optimal braking pedal displacement and braking motor torque in real time under actual working conditions; A braking control module, used to control the braking energy recovery and braking safety of the vehicle wire-controlled braking system according to the optimization results, and coordinate the distribution of frictional braking and electric motor braking forces under the ABS activation condition.

[0017] The present invention has the following beneficial effects: 1. By using digital twin technology, a high-precision virtual platform for the wire-controlled braking system is constructed, realizing the real-time mapping between the physical entity and the digital model, providing a reliable basis for system optimization; 2. Using the deep deterministic policy gradient algorithm to perform multi-objective optimization on the wire-controlled braking system, compared with traditional control methods, it can better balance multiple objectives such as energy recovery efficiency, braking safety, and comfort; 3. A dedicated braking force distribution strategy is designed for the ABS activation condition, solving the problem of insufficient braking force coordination in the existing system under this condition, achieving a 19% increase in the braking energy recovery per 100 kilometers, while keeping the deceleration fluctuation rate not exceeding 0.3g; 4. An efficient sample learning strategy is designed based on the experience replay mechanism, significantly improving the convergence speed and control effect of the algorithm; 5. Through the adaptive weight adjustment mechanism, the system realizes intelligent adaptation to different braking conditions, further improving the overall performance. Description of the Drawings

[0018] Figure 1 It is a flowchart of the multi-objective optimization method for the by-wire braking system based on digital twin of the present invention; Figure 2 It is a schematic structural diagram of the physical model of the by-wire braking system of the present invention; Figure 3 It is a schematic diagram of data interaction between the digital twin model and the physical model of the present invention; Figure 4 It is a schematic structural diagram of the deep deterministic policy gradient algorithm network of the present invention; Figure 5 It is a schematic diagram of the braking force distribution strategy under the ABS activation condition of the present invention; Figure 6 It is a schematic diagram of the system architecture of the present invention. Detailed Embodiments

[0019] The preferred embodiments of the present invention will be described in detail below with reference to the drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0020] As Figure 1 shown, the present invention provides a multi-objective optimization method for the by-wire braking system based on digital twin, including the following steps: Step S1: Establish a virtual platform 30 including the physical model 10 and the digital twin model 20 of the by-wire braking system, and the virtual platform 30 integrates the computational model, simulation data, and physical entities on the same platform.

[0021] Step S2: Construct a virtual-real interaction interface 40 on the virtual platform 30 to realize real-time interaction between the physical model data and the digital twin model data.

[0022] Step S3: Perform multi-objective optimization on the by-wire braking system based on the deep deterministic policy gradient algorithm, and coordinate the friction braking and electric motor braking force distribution according to the driver's driving intention while ensuring braking safety.

[0023] Step S4: Use a physical engine to simulate the hydraulic pipeline pressure fluctuations in real time and optimize the braking energy recovery amount and system response characteristics under the ABS activation condition.

[0024] As Figure 2 shown, in step S1, establishing a virtual platform 30 including a by-wire braking system physical model 10 and a digital twin model 20 includes: establishing a by-wire braking system physical model 10, the physical model 10 including a vehicle kinematics model 11, a hydraulic model 12, and a wheel braking dynamics model 13; establishing a by-wire braking system digital twin model 20, the digital twin model 20 including a digital environment 21, a virtual hydraulic system 22, and a virtual motor system 23; importing the physical model 10 into a virtual laboratory built based on MATLAB and Simulink, and obtaining brake pedal displacement and hydraulic pipeline pressure data according to the physical model 10; storing the brake pedal displacement and hydraulic pipeline pressure data into the physical engine to obtain a training sample set for updating the parameters of the digital twin model 20.

[0025] In an embodiment of the present invention, the above vehicle kinematics model 11 is used to obtain the total deceleration of the target vehicle being collected; the hydraulic model 12 is used to obtain the pressure value of the hydraulic cylinder; the wheel braking dynamics model 13 is used to obtain the slip rate of the collected wheels of the target vehicle and determine whether the collected wheels enter the ABS safety area.

[0026] Preferably, the vehicle kinematics model 11 can be expressed as: , wherein, is the vehicle speed, in m / s; is the time, in s; is the braking force, in N; is the rolling resistance, in N; is the air resistance, in N; is the ramp resistance, in N; is the vehicle mass, in kg; is the rotating mass conversion coefficient, dimensionless, usually 1.1 - 1.5, and for ordinary passenger cars, it can be taken as 1.2.

[0027] In an actual by-wire braking system, taking a certain type of pure electric SUV as an example, its mass is 1850 kg, the rotating mass conversion coefficient is 1.25, when the vehicle speed is 60 km / h, when braking on a dry asphalt road surface, the total braking force is about 6000 N, and a deceleration of about 0.3g can be achieved, meeting the daily comfortable braking requirements. The rolling resistance Can be calculated as: , Wherein, is the rolling resistance coefficient, dimensionless, related to tire type, tire pressure and road surface type, usually taking 0.013 - 0.018; g is the acceleration due to gravity, taking 9.8 m / s²; is the road inclination angle, in units of rad. For a normal passenger car driving on a standard asphalt road, the rolling resistance coefficient f generally takes 0.015. When the vehicle mass m is 1850 kg, on a flat road surface ( = 0), the rolling resistance is approximately 270 N.

[0028] Air resistance Can be calculated as: , Wherein, is the air resistance coefficient, dimensionless, determined by the vehicle shape, usually taking 0.3 - 0.4; is the frontal area of the vehicle, in units of m 2 is the air density, usually taking 1.225 kg / m 3 is the vehicle speed, in units of m / s. For the electric SUV involved in the optimization of the wire control braking system, its air resistance coefficient is 0.35, and the frontal area is 2.4 m 2 . When the vehicle speed is 60 km / h (16.7 m / s), the air resistance is approximately 290 N.

[0029] Gradient resistance Can be calculated as: , Wherein, is the vehicle mass, in units of kg; is the acceleration due to gravity, taking 9.8 m / s 2 , is the road inclination angle, in units of rad. When driving on a 3% gradient (about 1.7°), the gradient resistance is approximately 545 N. This has an important impact on the braking energy recovery strategy. The energy recovery potential is higher in the downhill condition and needs to be considered in the control strategy.

[0030] The hydraulic model 12 is used to describe the dynamic characteristics of the braking hydraulic system and can be expressed as: , Wherein, is the hydraulic cylinder pressure, with the unit of Pa; is the bulk modulus of the hydraulic fluid, with the unit of Pa, which depends on the type and temperature of the brake fluid. Usually, for DOT4 brake fluid at room temperature, take Pa; is the pipeline volume, with the unit of ; is the flow rate entering the pipeline, with the unit of ; is the flow rate leaving the pipeline, with the unit of ; is the volumetric change rate, with the unit of , considering factors such as pipeline expansion. In an electronic stability control system, the hydraulic model is crucial for accurately controlling the braking pressure. For a typical medium-sized passenger car braking system, the master cylinder diameter is 20 mm, the stroke is 30 mm, the inner diameter of the brake pipeline is 3.2 mm, and the total length is about 8 m. These parameters directly affect the system response characteristics. When the brake pedal is depressed at a speed of 20 mm / s, the flow rate generated by the master cylinder is approximately , considering the pipeline volume is , and the system pressure rise rate is approximately Pa / s (1.17 bar / s). This parameter has a direct impact on the responsiveness and comfort of the braking system. The wheel braking dynamics model 13 mainly describes the tire slip characteristics and can be expressed as: , where is the tire slip ratio, dimensionless; is the vehicle speed, with the unit of m / s; is the wheel angular velocity, with the unit of rad / s; is the wheel radius, with the unit of m.

[0031] There is an important relationship between the slip ratio and the braking force and road surface adhesion, which is usually expressed as the adhesion coefficient-slip ratio curve (μ-λ curve). For a dry asphalt road surface, the maximum adhesion coefficient usually reaches its peak at around the slip ratio λ = 0.15, approximately 0.9; for a wet road surface, the peak adhesion coefficient decreases to around 0.6 and appears at around the slip ratio λ = 0.1. In ABS control, the slip ratio is usually controlled between 0.1 - 0.3 to obtain the best braking effect and directional stability. When the slip ratio exceeds 0.3, the wheel enters the unstable region and is prone to wheel lock-up. At this time, it is determined that the wheel needs to enter the ABS safety control.

[0032] In the optimization of the online control braking system, when the vehicle is in an emergency braking situation under high-speed conditions (such as 100 km / h), the control accuracy of the slip ratio has a significant impact on the braking distance. Experimental data shows that controlling the slip ratio at the optimal point (λ = 0.15) can shorten the braking distance by about 15% compared to improper control (λ = 0.4), and at the same time increase the energy recovery efficiency by about 12%.

[0033] As Figure 3 shown, in step S2, the construction of the virtual-real interaction interface 40 includes: configuring a physical-to-digital data acquisition interface 41 to collect sensing data such as brake pedal displacement, wheel speed, hydraulic pressure, motor speed, motor current, and vehicle acceleration in real time; implementing a data preprocessing module 42 to perform noise filtering, outlier detection, data compensation, and data synchronization on the collected sensing data; configuring a digital-to-physical control execution interface 43 to convert control instructions into motor torque instructions, brake master cylinder pressure instructions, and ABS adjustment valve control instructions, and implement control through corresponding execution units.

[0034] In the embodiment of the present invention, the physical-to-digital data acquisition interface 41 is implemented using a high-speed data acquisition card, and the sampling frequency is designed according to the signal characteristics of different sensors. Among them, the sampling frequency of the brake pedal displacement sensor is 100 Hz, and the signal bandwidth is 0 - 20 Hz; the sampling frequency of the wheel speed sensor is 200 Hz, and the signal bandwidth is 0 - 50 Hz; the sampling frequency of the hydraulic pressure sensor is 200 Hz, and the signal bandwidth is 0 - 80 Hz; the sampling frequency of the motor speed sensor is 100 Hz, and the signal bandwidth is 0 - 30 Hz; the sampling frequency of the motor current sensor is 200 Hz, and the signal bandwidth is 0 - 60 Hz; the sampling frequency of the vehicle acceleration sensor is 100 Hz, and the signal bandwidth is 0 - 20 Hz. The sampling frequency design follows the Nyquist sampling theorem to ensure that the sampling frequency is at least twice the signal bandwidth to avoid signal aliasing.

[0035] The data preprocessing module 42 uses an adaptive window sliding average filtering algorithm, and the window size is automatically adjusted according to the data change rate. Specifically, when the signal change rate is small (such as in the initial stage of braking and the uniform braking stage), a larger window (7 sampling points) is used to improve the filtering effect; when the signal change rate is large (such as in the emergency braking and ABS activation stages), a smaller window (3 sampling points) is used to ensure the response speed. The filtering algorithm can be expressed as: , where is the filtered data at the nth moment; is the original data; is the filtering window size, which is adaptively adjusted according to the signal change rate, and the range is 3 - 7. The signal change rate is calculated as: , wherein, is the normalized signal change rate; is the sampling period, with the unit of s; is the signal full-scale value. The window size is adaptively adjusted according to the change rate: when , When , When , . Outlier detection adopts the 3 principle based on the historical data distribution, that is, when the data deviates from the mean by more than 3 times the standard deviation, it is determined as an outlier. At this time, the adjacent valid value is used for substitution. For the braking pressure signal, the normal working range is 0 - 8 MPa. When a mutation value such as 12 MPa is detected, the system automatically identifies it as an outlier and corrects it to ensure the stability of the control algorithm.

[0036] Data compensation mainly performs time compensation for sensor delay. Different sensors have different response times. The response time of the brake pedal displacement sensor is about 0.8 ms, the response time of the hydraulic pressure sensor is about 15 ms, and the response time of the wheel speed sensor is about 5 ms. Time compensation is used to ensure the time consistency of the data. The compensation formula is: , wherein, is the compensated data; is the original data; is the sensor response delay, which can be obtained through system identification. For the hydraulic pressure sensor, its compensation value is about -15 ms, and the Lagrange interpolation method is used to obtain the value at non-sampling points.

[0037] Data synchronization is achieved based on timestamps. The master clock is used to trigger sampling, and the synchronization accuracy is controlled within ±2 ms. A unified timestamp mark is added to all sensing data to ensure the time consistency of data fusion processing. This is crucial for correctly evaluating the system state and formulating control strategies. Especially during emergency braking under high-speed conditions, a 2-ms time synchronization error is only equivalent to the vehicle moving about 6 cm, ensuring high-precision system state evaluation and control.

[0038] The digital-to-physical control execution interface 43 converts control instructions into signals recognizable by hardware according to the specified format. The motor torque instruction range is ±300 Nm, with a resolution of 0.5 Nm. A positive value represents driving torque, and a negative value represents braking torque; the brake master cylinder pressure instruction range is 0 - 8 MPa, with a resolution of 0.05 MPa; the ABS adjustment valve control instructions include three states: opening, closing, and adjusting. The solenoid valve opening is controlled by a PWM signal, with a PWM frequency of 100 Hz and a duty cycle range of 0% - 100%. In practical applications, to avoid system shocks caused by sudden changes in control instructions, a ramp transition strategy is adopted. The motor torque change rate is limited within 1000 Nm / s, and the brake pressure change rate is limited within 20 MPa / s to ensure the smooth and rapid response of the braking system.

[0039] As Figure 4 shown, in step S3, the multi-objective optimization of the electronic braking system based on the deep deterministic policy gradient algorithm includes: designing a control state space, including states related to braking demand, vehicle states, system states, and environmental states; designing a control action space, including motor braking torque instructions, hydraulic brake pressure instructions, and state transition flags; designing a multi-objective reward function, including terms for energy recovery efficiency, braking performance, comfort, safety, and system response; constructing an experience replay pool to store control experiences and improving the sample utilization efficiency based on a priority sampling strategy; separating policy evaluation and improvement through a dual-network architecture to ensure the stability of the learning process.

[0040] In a preferred embodiment of the present invention, the control state space is designed as follows: States related to braking demand include brake pedal displacement and the rate of change of pedal displacement . Among them, the range is 0 - 100 mm, representing the intensity of the driver's braking intention; the range is typically -200 mm / s to 200 mm / s, representing the changing trend of the braking intention, which is of great significance for distinguishing situations such as sudden braking and slow pedal release. Vehicle state-related states include vehicle speed , wheel angular velocity and slip ratio . Among them, the range is 0 - 180 km / h, which is a key input for braking control; is the set of angular velocities of the four wheels , with the unit of rad / s. The subscripts , , , represent the left front wheel, right front wheel, left rear wheel, and right rear wheel respectively; is the set of slip ratios of the four wheels , which is the core parameter for ABS control.

[0041] The system state-related states include the motor speed 、 and , among which usually ranges from 0 to 12000 rpm, affecting the motor braking force and energy recovery efficiency; ranges from -20°C to 120°C. Under high-temperature conditions, the motor braking force needs to be limited to protect the motor; is the set of braking pressures of the four wheels , with the unit of MPa, directly determining the magnitude of the mechanical braking force. The environment-related states include the estimated value of the road surface adhesion coefficient and the estimated value of the road slope . Among them ranges from 0.1 to 1.0 and is estimated from the slip ratio and the wheel angular acceleration; ranges from -20° to 20° and is estimated from the vehicle longitudinal acceleration and speed changes. These parameters have an important impact on the selection of braking strategies. For example, a more conservative braking strategy is required on low-adhesion road surfaces, and the energy recovery ratio can be increased on downhill sections.

[0042] The historical state memory includes the key state variables of the previous N steps (N = 10), forming a time-series state vector, providing information on the dynamic changes of the system, and helping to improve the stability and predictability of the control strategy.

[0043] To sum up, the complete state vector can be expressed as: , where represents the historical state sequence, containing the key state information of the previous N steps. The dimension of the state vector is 10 + 4×4 + 2 + 10 = 38, which is an acceptable input scale for deep reinforcement learning. The design of the control action space includes: the motor braking torque command , a continuous value, ranging from [-300 Nm, 300 Nm], where a negative value represents braking and a positive value represents driving; the hydraulic braking pressure command , a continuous value, ranging from [0 MPa, 8 MPa], uniformly controlling the pressure of the four-wheel brakes, and then distributing it to each wheel according to the vehicle state through the braking force distributor; the state switching flag , a discrete value, indicating the ABS activation / deactivation control, with the value of 0 representing deactivation and the value of 1 representing activation. The action vector can be expressed as: , The dimension of the action vector is 3. This low-dimensional action space design is beneficial to improving the learning efficiency and control accuracy. In actual control, and The coordinated cooperation is the key to ensuring the energy recovery efficiency and braking safety. For example, in the high-speed light braking condition (such as 80 km / h, deceleration 0.1g), the electric motor braking is preferentially used, is about -120 Nm, is close to 0; in the medium-speed moderate braking condition (such as 50 km / h, deceleration 0.3g), the electric motor braking and hydraulic braking work together, is about -280 Nm, is about 2 MPa; in the low-speed emergency braking condition (such as 30 km / h, deceleration 0.8g), mainly rely on hydraulic braking, is about -100 Nm, is about 6 MPa.

[0044] The core of the Deep Deterministic Policy Gradient algorithm is to approximate the policy function and value function through a deep neural network, and combine the deterministic policy gradient and experience replay mechanism to achieve efficient optimization of the continuous action space. This algorithm includes four networks: Actor network, Critic network, target Actor network and target Critic network.

[0045] The Actor network is used to generate deterministic actions according to the current state. Its structure is a 4-layer neural network, including: input layer, two hidden layers and output layer. The number of neurons in the input layer is the same as the state dimension, which is 38 neurons; the first hidden layer contains 32 neurons, and the second hidden layer contains 16 neurons, both using the ReLU activation function; the number of neurons in the output layer is the same as the action dimension, which is 3 neurons. The first two use the tanh activation function to map the output to the [-1, 1] interval, and then scale it to the actual control range. The last one uses the sigmoid activation function to map the output to the [0, 1] interval for ABS state control.

[0046] The ReLU activation function is defined as: , where, is the neuron input; is the activation function output. The ReLU function has the advantages of simple calculation and stable gradient, and is suitable for deep network training. The tanh activation function is defined as: , where, is the neuron input; is the output of the activation function, with a range of [-1, 1]. The zero-centered property of the tanh function is beneficial for gradient update. The sigmoid activation function is defined as: , where is the neuron input; is the output of the activation function, with a range of [0, 1]. The sigmoid function is suitable for binary classification problems and is used in this case for the activation judgment of the ABS state.

[0047] The Actor network can be expressed as: , where is the action vector, including , and ; is the state vector; is the Actor network with parameters . The parameter contains all network weights and biases, with a total number of parameters of approximately . The Critic network is used to evaluate the value of the state-action pair, and its structure is a 4-layer neural network: the input layer receives the concatenated vector of the state and action, with the number of neurons being the sum of the state dimension and the action dimension, i.e., 41 neurons; the first hidden layer contains 64 neurons, and the second hidden layer contains 32 neurons, both using the ReLU activation function; the output layer is 1 neuron, using the linear activation function, representing the Q value of the state-action pair. The Critic network can be expressed as: , where is the value of the state-action pair; is the state vector; is the action vector; is the Critic network with parameters . The parameter contains all network weights and biases, with a total number of parameters of approximately . The target Actor network and the target Critic network have the same structure as the corresponding main network, but the parameter update frequency is lower, which is used to provide a stable learning target and prevent unstable training. The parameter update adopts a soft update strategy, with an update frequency of once every 20 steps, and the update coefficient .

[0048] The multi-objective reward function is designed as the weighted sum of each performance index: , where is the total reward; is the energy recovery efficiency reward, which is directly proportional to the motor recovery current; is the braking performance reward, which is inversely proportional to the deviation between the actual deceleration and the target deceleration; is the comfort reward, which is inversely proportional to the deceleration fluctuation rate; is the safety reward, which is inversely proportional to the degree of deviation of the tire slip rate from the optimal slip rate; is the system response reward, which is inversely proportional to the change rate of the control command; to are the weight coefficients of each reward, and satisfy . The energy recovery efficiency reward can be calculated as: , where: is the recovery current, with the unit of A; is the motor terminal voltage, with the unit of V; is the total braking power, with the unit of W; is the energy recovery efficiency coefficient, usually taken as 2 - 5, which is used to adjust the weight of this reward. Taking a certain type of pure electric vehicle as an example, when braking with a deceleration of 0.2g at a vehicle speed of 50 km / h, the motor speed is about 3000 rpm, and the motor terminal voltage is about 350 V, the recovery current is about 80 A, the total braking power is about 38 kW, the energy recovery power is about 28 kW, and the recovery efficiency is about 74%. At this time the calculated value is 3.7 (assuming ), which is a positive incentive.

[0049] The braking performance reward can be calculated as: , where, is the actual deceleration, with the unit of is the target deceleration, with the unit of is the braking performance coefficient, usually taken as 1 - 3, which is used to adjust the weight of this reward. Taking the comfort braking condition as an example, when the driver steps on the brake pedal by 40 mm, the corresponding target deceleration is about , if the deceleration achieved by the actual system control is 0.32g (3.14 ), then the deviation is , assuming , the braking performance reward the calculated value is -0.4, which is a slight penalty. The comfort reward can be calculated as: , where, is the standard deviation of deceleration, representing the degree of deceleration fluctuation, with the unit of m / s is the comfort coefficient, usually taken as 2 - 4, and is used to adjust the weight of this reward item.

[0050] Standard deviation of deceleration It is calculated as: , where, is the deceleration at the i-th moment, with the unit of is the average deceleration at N moments, with the unit of is the number of sampling points, usually taking data at 10 - 20 moments. Under ideal braking conditions, the deceleration fluctuation is small, is approximately 0.05g (0.49 m / s²), assuming , then the comfort reward The calculated value is -1.47, which is a medium penalty; under the ABS activation condition, the deceleration fluctuation is large, can reach , under the same conditions, the calculated value of the comfort reward is -5.88, which is a strong penalty, and this will prompt the control strategy to seek a smoother braking method.

[0051] The safety reward can be calculated as: , where, is the actual slip ratio of the i-th wheel, dimensionless; is the optimal slip ratio, usually taken as 0.15 - 0.2, dimensionless; is the safety coefficient, usually taken as 3 - 5, and is used to adjust the weight of this reward item. The summation symbol represents the accumulation of the slip ratio deviations of the four wheels.

[0052] Under normal braking conditions, the slip ratios of the four wheels are close to the optimal value. For example, are all 0.16, is 0.18, then the total sum of the slip ratio deviations is 0.08. Assuming , the safety reward The calculated value is -0.32, which is a slight penalty; under the emergency braking condition on a wet road surface, if the slip ratio of a certain wheel reaches 0.4, far from the optimal value of 0.18, the safety reward will be significantly reduced, prompting the control strategy to activate the ABS for intervention.

[0053] The system response reward can be calculated as: , where, is the change in motor torque between two adjacent control cycles, with the unit of Nm; is the change in hydraulic pressure between two adjacent control cycles, with the unit of MPa; is the response coefficient, usually taken as 0.5 - 1.5, used to adjust the weight of this item of reward; is the pressure change weight coefficient, usually taken as 2 - 5, considering that the impact of pressure change on comfort is usually greater than that of motor torque change. During the smooth braking transition stage, if the change in motor torque is 10 Nm, and the change in hydraulic pressure is 0.2 MPa, assuming , the system response reward calculated value is -10 - 3 , which is a medium penalty; in the case of sudden braking force change, such as is 50 Nm, is 1 MPa, and under the same conditions, the reward value is -53, which is a strong penalty, which will prompt the control strategy to seek a smoother braking transition method.

[0054] In the embodiments of the present invention, the weights of the reward function adopt an adaptive adjustment mechanism, and dynamically adjust the weight ratios of various indicators according to different braking conditions. Specifically, in the case of emergency braking, the safety weight is increased and the energy recovery weight is reduced; in the case of general braking, the energy recovery weight is increased and a moderate comfort weight is maintained; in the case of ABS activation, the safety weight is set to the highest and the energy recovery weight is the lowest; in the case of low-speed conditions, the comfort weight and the system response weight are increased.

[0055] For example, the basic condition weight configuration is: energy recovery , safety , comfort , responsiveness = 0.1); the emergency braking weight configuration is: safety ( , responsiveness , comfort , energy recovery ( The general braking weight configuration is: energy recovery , comfort , safety , responsiveness The low-speed braking weight configuration is: comfort , energy recovery , responsiveness , safety . The condition judgment criterion is: when the brake pedal displacement or the pedal speed , it is determined as the emergency braking condition; when the vehicle speed , it is determined as the low-speed condition; when the slip ratio of any wheel When it is 0.3, it is determined as the ABS activation condition; in other cases, it is determined as the general braking condition.

[0056] To improve the convergence efficiency and stability of the algorithm, the present invention adopts an experience replay mechanism to store and utilize control experiences. In a preferred embodiment of the present invention, the experience replay pool adopts a multi-level structure design, including: a short-term pool for storing the recently generated experience samples with a capacity of 100; a medium-term pool for storing the experience samples with a medium time span with a capacity of 500; and a long-term pool for storing the long-term effective experience samples with a capacity of 2000.

[0057] The structure of the experience sample is <current state, executed action, obtained reward, next state, condition label>, where the condition label is used to identify the condition type to which the sample belongs, such as emergency braking, general braking, low adhesion road surface, ABS activation, etc. In the initial stage of training, to ensure the balance of various condition samples, a condition balance strategy is adopted. For example, in the constructed training scenarios, 30% of the general braking conditions, 30% of the emergency braking conditions, 20% of the low adhesion road surface conditions, and 20% of the ABS activation conditions are set to ensure that the algorithm can learn the optimal control strategies under different conditions.

[0058] To improve the sample utilization efficiency, the present invention adopts a priority calculation strategy based on the temporal difference error to increase the sampling probability of rare condition samples and samples with rich gradient information. The temporal difference error is calculated as: , Where, is the temporal difference error; is the immediate reward; is the discount factor, usually taking 0.95 - 0.99; is the target Critic network; is the target Actor network; is the main Critic network; is the current state; is the executed action; is the next state. In the ABS activation condition, due to the fast state change and large immediate reward change, the temporal difference error is usually large, such as up to 3 - 5, indicating that these samples contain richer gradient information; while in the steady-state braking, the state and reward change less, and the temporal difference error is usually small, such as 0.1 - 0.5. The priority sampling strategy will make the ABS activation condition samples obtain a higher sampling probability and accelerate the learning process. The priority of the sample is proportional to the temporal difference error and is calculated as: , Where, is the sample priority; is a small positive number, usually taken as 0.01 to prevent the priority from being zero; is the priority exponent, usually taken as 0.6 - 0.8, which is used to control the influence degree of the priority. When the value is small (such as 0.6), the priority differences of each sample are small and the sampling is more uniform; When the value is large (such as 0.8), high - error samples are more likely to be sampled, but this may lead to over - attention to some samples.

[0059] At the same time, for rare working condition samples, such as ABS activation condition, low - adhesion road surface condition, etc., an additional priority coefficient is added, usually taken as 1.5 - 3, to further increase the sampling probability of these samples. For example, for the ABS activation sample under a wet road surface, its original priority may be 2.5, and after applying the rare working condition coefficient , the priority is increased to 5.0, significantly increasing the probability of being sampled. Based on the above priorities, a priority sampling strategy is used to sample training samples from the experience replay pool. To prevent high - priority samples from overly dominating the training process, importance sampling weights are used for correction: , where, is the importance sampling weight; is the size of the experience replay pool; is the probability that sample i is sampled, calculated as is the importance sampling exponent, and its initial value is usually taken as 0.4 and then linearly grows to 1. When the value is small, the correction effect of importance sampling is weak; When the value is close to 1, the correction effect is strong and it is closer to uniform sampling. The network parameters are updated using temporal difference (TD) learning and policy gradient algorithms. Specifically, the loss function of the Critic network is: , where: is the loss function of the Critic network; is the batch size, usually taken as 32 - 64; is the importance sampling weight of sample i; is the TD target of sample i, calculated as ; is the predicted value of the Critic network for sample i. The summation symbol represents the accumulation of all samples in the batch, and then the average value is taken as the loss function. The parameters of the Actor network are updated using the deterministic policy gradient (DPG) algorithm: , where: is the policy gradient of the Actor network; is the gradient of the Q value with respect to the action a, calculated at the action location; is the policy function with respect to the parameter gradient. The summation symbol indicates that the policy gradient is calculated for all samples in the batch and then averaged for parameter update. The intuitive understanding of the policy gradient is: adjust the policy parameter so that the policy at the state produces the action that can maximize the Q value . Taking motor torque control as an example, if increasing the motor braking torque can improve the energy recovery efficiency and thus increase the Q value, the policy gradient will adjust the parameters so that the network outputs a larger motor braking torque in similar states. To ensure the stability of learning, the target network parameters adopt a soft update strategy: , , where, and are the parameters of the target Actor network and the target Critic network respectively; and are the parameters of the main Actor network and the main Critic network respectively; is the soft update coefficient, usually taking 0.001 - 0.01, which controls the smoothness of the update. When the value is small (such as 0.001), the target network parameters change slowly, and the learning stability is good, but the convergence speed is slow; When the value is large (such as 0.01), the target network parameters are updated quickly, and the convergence speed is fast, but it may lead to unstable learning.

[0060] In practical applications, dynamically adjust the value according to the system convergence situation: take a smaller value such as 0.001 in the initial stage, and gradually increase it to 0.01 as the training progresses to balance stability and convergence speed. The target network update frequency is set to update once every 20 steps instead of every step to further enhance the learning stability.

[0061] For example Figure 5As shown, in step S4, the real-time simulation of the hydraulic pipeline pressure fluctuation by using the physical engine includes: using the PX-OSG plug-in for simulation, and realizing the real-time data exchange between the physical engine and the digital twin model based on the WebSocket protocol; real-time simulating the hydraulic pipeline pressure fluctuation according to the physical model data and the digital twin model data; the physical model data includes the ABS pressure, the friction brake wheel cylinder pressure, the electric motor braking force, and the system temperature; the digital twin data includes the brake pedal displacement, the motor torque, the regenerative current, and the pipeline pressure.

[0062] In an embodiment of the present invention, the physical engine uses the open-source physical simulation engine PhysX combined with the OSG (OpenSceneGraph) visualization library to construct the physical model and the visualization model of the virtual hydraulic system. The PX-OSG plug-in is based on the WebSocket protocol and uses the TCP / IP communication method to realize the real-time data exchange between the physical engine and the digital twin model. The communication frequency is 100Hz, and the data delay is controlled within 10ms. The communication protocol includes two parts: a request message and a response message.

[0063] The simulation of the hydraulic pipeline pressure fluctuation is based on a one-dimensional fluid dynamics model, considering factors such as pipeline friction, oil compressibility, and pipe wall elasticity. The fluctuation equation can be expressed as: , , where P is the pressure, with the unit of Pa; v is the flow velocity, with the unit of m / s; t is the time, with the unit of s; x is the position, with the unit of m; is the oil density, usually taken as 850 - 900 kg / m³; c is the pressure wave speed, usually taken as 1200 - 1500 m / s; f is the friction coefficient, dimensionless, usually taken as 0.02 - 0.04; D is the pipe diameter, with the unit of m. represents the partial derivative of the pressure with respect to time, characterizing the rate of change of the pressure with time; represents the partial derivative of the pressure with respect to the position, characterizing the distribution of the pressure along the pipeline; represents the partial derivative of the flow velocity with respect to time, characterizing the rate of change of the flow velocity with time; represents the partial derivative of the flow velocity with respect to the position, characterizing the distribution of the flow velocity along the pipeline. For a typical hydraulic pipeline of a braking system, the pipe diameter is 3.2mm, the length is 1 - 2m, the brake fluid density is 880 kg / m³, the pressure wave speed is 1350 m / s, and the friction coefficient It is 0.025. Under emergency braking conditions, the master cylinder pressure can rise from 0 to 8 MPa within 0.1 s, and the resulting hydraulic shock can cause pipeline pressure fluctuations. The fluctuation amplitude can reach 1 - 2 MPa, and the period is about 2 - 5 ms. This pressure fluctuation propagates through the hydraulic pipeline, ultimately leading to brake caliper pressure fluctuations, affecting braking force stability and comfort.

[0064] Optimizing the braking energy recovery amount and system response characteristics under ABS activation conditions is an important innovation point of the present invention. Specifically, it includes: coordinating the distribution of frictional braking and electric motor braking force under ABS activation conditions, adjusting the solenoid valve state and hydraulic cylinder drive according to the wheel cylinder pressure change; when the wheel cylinder pressure rises and approaches the target value, opening the solenoid valve to connect the oil circuit between the hydraulic cylinder control reservoir and the master cylinder; when the wheel cylinder pressure drops and the braking pressure is low, closing the solenoid valve to disconnect the oil circuit between the reservoir and the master cylinder, and at the same time driving the hydraulic cylinder to adjust the oil pressure; through the above control strategy, the braking energy recovery amount per 100 kilometers is increased by 19%, while keeping the deceleration fluctuation rate not exceeding 0.3g.

[0065] The judgment criterion for ABS activation conditions is: when the slip ratio \(\lambda_i\) of any wheel exceeds 0.3, trigger ABS control. The goal of ABS control is to maintain the slip ratio within the optimal slip range (0.1 - 0.2), ensuring both high braking force and maintaining the directional stability of the wheel.

[0066] Under normal conditions, the hydraulic cylinder controls the on - off of the oil circuit between the reservoir and the master cylinder, and the hydraulic cylinder drives the secondary master cylinder piston and the master cylinder piston to move and output braking pressure. Under ABS activation conditions, according to the braking system parameters, the hydraulic cylinder closes the oil circuit between the reservoir and the master cylinder, and the pressure control motor adjusts the braking pressure. Introduce the braking pressure into the wheel cylinder. The specific control strategy is as follows: 1. When the wheel cylinder pressure rises and reaches 90% of the target value (the threshold is set according to the actual system response characteristics), open the solenoid valve to connect the oil circuit between the hydraulic cylinder control reservoir and the master cylinder to prevent the wheel from locking due to excessive wheel cylinder pressure; 2. When the wheel cylinder pressure drops and is lower than 60% of the target value (the threshold is set considering system stability and response speed), close the solenoid valve to disconnect the oil circuit between the reservoir and the master cylinder, and at the same time drive the hydraulic cylinder to adjust the oil pressure to prevent the braking distance from increasing due to insufficient braking force; 3. When the vehicle speed is lower than 10 km / h, gradually exit the ABS control mode and return the control right to the normal braking control mode to prevent instability caused by frequent system switching at low speeds.

[0067] When the traditional ABS system is activated, it usually significantly reduces or completely shuts off the electric motor braking, which is not conducive to energy recovery. The innovation of the present invention lies in that, even under the ABS activation condition, an appropriate electric motor braking force is still maintained to achieve energy recovery. The specific strategy is as follows: 1. When the ABS is not activated, the proportion of the electric motor braking force can reach 90%, maximizing energy recovery; 2. At the initial stage of ABS activation, the electric motor braking force rapidly decreases to 40% to ensure that the system can quickly respond to the slip condition; 3. As the ABS control stabilizes, the electric motor braking force gradually recovers to 60% - 70% to improve energy recovery on the premise of ensuring braking safety; 4. When the ABS is frequently activated under special road conditions (such as ice and snow roads), the electric motor braking force is further reduced to less than 20% to give priority to ensuring vehicle safety.

[0068] The dynamic distribution of the electric motor braking force and the hydraulic braking force is achieved through the following formula: , where: is the proportion of the electric motor braking force in the total braking force; is the slip ratio of the i-th wheel; is the time after ABS activation, with the unit of s; is the time coefficient, usually taking 0.5 - 1.5, which controls the recovery speed of the electric motor braking force; is the estimated road surface adhesion coefficient. In the conditional judgment statement, "if" represents the condition, "max" represents taking the maximum value, represents the negative exponential decay function, which gradually approaches 0 as time increases.

[0069] For a certain type of pure electric SUV, when performing an emergency braking at a speed of 100 km / h on a dry asphalt road surface (adhesion coefficient μ = 0.8), adopting the above control strategy, the proportion of the electric motor braking force at the initial stage of ABS activation is 0.4. As the system stabilizes, the proportion of the electric motor braking force recovers to 0.67 after 3 seconds, and the braking energy recovery amount reaches 0.65 kWh; while under the traditional control strategy, the electric motor braking is completely shut off after ABS activation, and the braking energy recovery is only 0.24 kWh. Comprehensive tests under various working conditions show that the control strategy of the present invention increases the braking energy recovery per 100 kilometers from 82 kWh of the traditional system to 97.6 kWh, with a promotion rate of 19%.

[0070] Meanwhile, by precisely controlling the hydraulic pressure and the motor torque, the deceleration fluctuation rate is controlled within 0.3g, greatly improving the braking comfort. The deceleration fluctuation rate is defined as the ratio of the deceleration standard deviation to the average deceleration, and the calculation formula is: Among them, is the deceleration volatility, dimensionless; is the average deceleration, with the unit of .

[0071] Under the emergency braking condition (the average deceleration is about 0.8g or 7.84m / s²), the standard deviation of deceleration under the traditional control strategy is about 0.4g (3.92m / s²), and the deceleration volatility is 0.5; while adopting the control strategy of the present invention, the standard deviation of deceleration is reduced to 0.24g (2.35m / s²), and the deceleration volatility is 0.3, with the fluctuation reduced by 40%, significantly improving the braking comfort.

[0072] As Figure 6 shown, the present invention also provides a multi-objective optimization system for a wire-controlled braking system based on digital twin, including: a physical model establishment module 100 for establishing a physical model of the wire-controlled braking system, including a vehicle kinematic model, a hydraulic model, and a wheel braking dynamics model; a digital model construction module 200 for establishing a digital twin model of the wire-controlled braking system, including a virtual laboratory construction unit, a training sample acquisition unit, a neural network model training unit, and a digital twin mapping unit; a virtual-real interaction interface module 300 for realizing real-time interaction between the physical model data and the digital twin model data; an optimization target setting module 400 for setting and managing multiple optimization indexes of braking energy recovery and braking safety; a weight optimization module 500 for determining and optimizing the weight ratio of each optimization index to achieve multi-objective balance; a real-time optimization module 600 for searching for the optimal braking pedal displacement and braking motor torque in real time under actual working conditions; and a braking control module 700 for controlling the braking energy recovery and braking safety of the vehicle wire-controlled braking system according to the optimization results, coordinating the friction braking and the electric motor braking force distribution under the ABS activation condition, increasing the braking energy recovery per 100 kilometers by 19%, and at the same time keeping the deceleration volatility not exceeding 0.3g.

[0073] The implementation manner of the physical model establishment module 100 is the same as the above physical model establishment steps, including establishing a vehicle kinematic model, a hydraulic model, and a wheel braking dynamics model. These models together constitute the physical basis of the wire-controlled braking system and provide a basis for subsequent optimization.

[0074] The digital model building module 200 is responsible for establishing the digital twin model of the electronic brake system, including a virtual laboratory construction unit, a training sample collection unit, a neural network model training unit, and a digital twin mapping unit. The virtual laboratory construction unit builds a virtual simulation platform based on MATLAB and Simulink, constructs a virtual vehicle model, introduces real driving scenario data, and simulates vehicle behaviors under different working conditions; the training sample collection unit is responsible for collecting data such as brake pedal displacement, vehicle speed, wheel speed, and braking pressure, with a sampling frequency of 100 Hz and a collection duration of 30 - 60 seconds for each test, covering various working conditions such as normal braking, emergency braking, and braking on low-adhesion road surfaces; the neural network model training unit trains a deep deterministic policy gradient model based on the collected data, using the aforementioned network structure and training parameters; the digital twin mapping unit establishes the mapping relationship between the physical model and the digital model, realizes the real-time conversion from the physical state to the digital state, with the error controlled within 5%.

[0075] The virtual-real interaction interface module 300 realizes the real-time interaction between the physical model data and the digital twin model data. Its implementation method is the same as the above-mentioned virtual-real interaction interface construction steps, mainly including a data acquisition interface from physical to digital, a data preprocessing module, and a control execution interface from digital to physical.

[0076] The optimization objective setting module 400 is responsible for setting and managing multiple optimization indicators for brake energy recovery and braking safety, including a control variable setting unit, an objective function construction unit, a constraint condition setting unit, and an indicator evaluation unit. The control variable setting unit sets the optimization indicators by the control variable method, such as setting conditions like fixed vehicle speed and slope, and analyzes the influence of different braking strategies on the energy recovery efficiency; the objective function construction unit constructs an objective function for comprehensively evaluating various indicators, such as the aforementioned multi-objective reward function; the constraint condition setting unit sets hard constraint conditions such as safety and comfort, such as the slip ratio not exceeding 0.4 and the braking distance not exceeding 110% of the reference value; the indicator evaluation unit evaluates the achievement of the optimization indicators in real time, calculates indicators such as energy recovery efficiency, braking distance, and comfort, and provides feedback.

[0077] The weight optimization module 500 determines and optimizes the weight ratios of various optimization metrics to achieve multi-objective balance, including an initial weight setting unit, a differential evolution algorithm unit, a weight evaluation unit, and a weight adaptation unit. The initial weight setting unit sets the initial weights of each optimization metric, such as the weight configuration of the aforementioned base operating conditions; the differential evolution algorithm unit adjusts the weight ratios based on the differential evolution algorithm, with the population size set to 30, the maximum number of iterations to 50, the mutation factor F taken as 0.5, and the crossover probability CR taken as 0.7; the weight evaluation unit evaluates the impact of weight adjustment on system performance and calculates the comprehensive performance metrics under different weight combinations; the weight adaptation unit adaptively adjusts the weight ratios according to different operating conditions, such as dynamically increasing the safety weight under the ABS activation condition.

[0078] The real-time optimization module 600 searches for the optimal brake pedal displacement and brake motor torque in real time under actual operating conditions, including a target speed setting unit, an optimization algorithm execution unit, a fitness evaluation unit, and a digital twin update unit. The target speed setting unit sets the target speed for acquisition, such as setting the target deceleration to 0.2g in the smooth deceleration condition and setting the target deceleration to 0.8g in the emergency braking condition; the optimization algorithm execution unit executes the optimization algorithm based on deep deterministic policy gradients, with a control period of 10 ms and real-time updating of the output of the policy network; the fitness evaluation unit evaluates the fitness value of the optimization result and calculates metrics such as the deviation between the actual deceleration and the target deceleration and the energy recovery efficiency; the digital twin update unit updates the digital twin model based on the fitness value and adjusts the model parameters to make the prediction results consistent with the actual performance.

[0079] The brake control module 700 controls the brake energy recovery and brake safety of the vehicle's by-wire braking system according to the optimization results, including a control strategy application unit, an electric motor brake control unit, a hydraulic brake control unit, and a coordinated control unit. The control strategy application unit applies the optimized control strategy and converts the output of the policy network into specific control instructions; the electric motor brake control unit controls the electric motor to output brake torque, with a response time not exceeding 30 ms and a torque control accuracy of ±5 Nm; the hydraulic brake control unit controls the hydraulic system to output brake pressure, with a response time not exceeding 50 ms and a pressure control accuracy of ±0.2 MPa; the coordinated control unit coordinates the smooth transition between electric motor braking and hydraulic braking and dynamically adjusts the outputs of each actuator according to the aforementioned braking force distribution strategy to ensure braking smoothness.

[0080] Based on the above system architecture, the present invention realizes the multi-objective optimization of the by-wire braking system, maximizes the energy recovery efficiency while ensuring brake safety, and at the same time maintains good braking comfort. Especially under the ABS activation condition, by precisely coordinating the friction braking and electric motor braking force distribution, the braking energy recovery per 100 kilometers is increased by 19%, while keeping the deceleration fluctuation rate not exceeding 0.3g, significantly improving the overall performance of the system.

[0081] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various obvious changes and deformations can be made to the present invention without departing from the scope of the present invention, and these changes and deformations all fall within the protection scope of the present invention.

Claims

1. A multi-objective optimization method for a wire control braking system based on digital twin, characterized in that, Including: Establish a virtual platform that includes a physical model and a digital twin model of the electronic brake system, where the virtual platform integrates a computational model, simulation data, and physical entities on the same platform; Construct a virtual-real interaction interface on the virtual platform to achieve real-time interaction between the physical model data and the digital twin model data; Perform multi-objective optimization on the electronic brake system based on the Deep Deterministic Policy Gradient algorithm, and coordinate the distribution of friction braking and electric motor braking force according to the driver's driving intention while ensuring braking safety; Use a physical engine to real-time simulate the hydraulic pipeline pressure fluctuation, and optimize the braking energy recovery amount and system response characteristics under the ABS activation condition.

2. The multi-objective optimization method of the wire control braking system based on digital twin according to claim 1, wherein The establishment of the virtual platform that includes a physical model and a digital twin model of the electronic brake system includes: Establish a physical model of the electronic brake system, where the physical model includes a vehicle kinematics model, a hydraulic model, and a wheel braking dynamics model; Establish a digital twin model of the electronic brake system, where the digital twin model includes a digital environment, a virtual hydraulic system, and a virtual motor system; Import the physical model into a virtual laboratory built based on MATLAB and Simulink, and obtain the brake pedal displacement and hydraulic pipeline pressure data according to the physical model; Store the brake pedal displacement and hydraulic pipeline pressure data in the physical engine to obtain a training sample set for updating the digital twin model parameters.

3. The multi-objective optimization method for the wire control braking system based on digital twin according to claim 2, wherein, The vehicle kinematics model is used to obtain the total deceleration of the target vehicle being collected; the hydraulic model is used to obtain the pressure value of the hydraulic cylinder; the wheel braking dynamics model is used to obtain the slip ratio of the collected wheels of the target vehicle and determine whether the collected wheels enter the ABS safety area.

4. The multi-objective optimization method of the wire control braking system based on digital twin according to claim 1, wherein, The construction of the virtual-real interaction interface includes: Configure a data acquisition interface from physical to digital to real-time collect sensing data such as brake pedal displacement, wheel speed, hydraulic pressure, motor speed, motor current, and vehicle acceleration; Implement a data preprocessing module to perform noise filtering, outlier detection, data compensation, and data synchronization on the collected sensing data; Configure a control execution interface from digital to physical to convert the control instruction into a motor torque instruction, a brake master cylinder pressure instruction, and an ABS adjustment valve control instruction, and implement the control through the corresponding execution unit.

5. The multi-objective optimization method for a wire control braking system based on digital twin according to claim 1, wherein The multi-objective optimization of the electronic brake system based on the Deep Deterministic Policy Gradient algorithm includes: Design a control state space, including states related to braking demand, vehicle states, system states, and environmental states; Design a control action space, including an electric motor braking torque instruction, a hydraulic braking pressure instruction, and a state switching flag; Design a multi-objective reward function, including an energy recovery efficiency term, a braking performance term, a comfort term, a safety term, and a system response term; Construct an experience replay pool to store control experiences, and improve the sample utilization efficiency based on the priority sampling strategy; Realize the separation of policy evaluation and improvement through a dual-network architecture to ensure the stability of the learning process.

6. The multi-objective optimization method of the wire control braking system based on digital twin according to claim 5, characterized in that, The weights of the multi-objective reward function adopt an adaptive adjustment mechanism to dynamically adjust the weight ratios of various indicators according to different braking conditions: In the case of emergency braking, increase the safety weight and reduce the energy recovery weight; Under normal braking conditions, increase the energy recovery weight and maintain a moderate comfort weight; Under ABS activation conditions, set the safety weight to the highest and the energy recovery weight to the lowest; Under low-speed conditions, increase the comfort weight and the system response weight.

7. The multi-objective optimization method for a wire control braking system based on digital twin according to claim 5, wherein The experience replay pool adopts a multi-level structure design, including: A short-term pool for storing recently generated experience samples, with a capacity of 100; A medium-term pool for storing experience samples with a medium time span, with a capacity of 500; A long-term pool for storing long-term valid experience samples, with a capacity of 2000; Based on the priority calculation strategy of temporal difference error, increase the sampling probability of rare condition samples and samples rich in gradient information.

8. The multi-objective optimization method for a wire control braking system based on digital twin according to claim 1, characterized in that The real-time simulation of hydraulic pipeline pressure fluctuations using a physical engine includes: Using the PX-OSG plugin for simulation and realizing real-time data exchange between the physical engine and the digital twin model based on the WebSocket protocol; Real-time simulating the hydraulic pipeline pressure fluctuations according to the physical model data and the digital twin model data; The physical model data includes ABS pressure, friction brake wheel cylinder pressure, motor braking force, and system temperature; The digital twin data includes brake pedal displacement, motor torque, regenerative current, and pipeline pressure.

9. The multi-objective optimization method of the wire control braking system based on digital twin according to claim 1, wherein, The optimization of the brake energy recovery amount and the system response characteristics under ABS activation conditions includes: Under ABS activation conditions, coordinate the distribution of friction braking and motor braking force, and adjust the solenoid valve state and hydraulic cylinder drive according to the change of wheel cylinder pressure; When the wheel cylinder pressure rises and approaches the target value, open the solenoid valve to connect the oil circuit between the hydraulic cylinder control reservoir and the master brake cylinder; When the wheel cylinder pressure drops and the braking pressure is low, close the solenoid valve to disconnect the oil circuit between the reservoir and the master brake cylinder, and at the same time drive the hydraulic cylinder to adjust the oil pressure; Through the above control strategy, the brake energy recovery amount per 100 kilometers is increased by 19%, while keeping the deceleration volatility not exceeding 0.3g.

10. A multi-objective optimization system for a wire control braking system based on digital twin, which is used to implement the method according to any one of claims 1-9, characterized in that Including: A physical model establishment module for establishing the physical model of the electronic brake system, including a vehicle kinematics model, a hydraulic model, and a wheel braking dynamics model; A digital model construction module for establishing the digital twin model of the electronic brake system, including a virtual laboratory construction unit, a training sample collection unit, a neural network model training unit, and a digital twin mapping unit; A virtual-real interaction interface module for realizing the real-time interaction between the physical model data and the digital twin model data; An optimization target setting module for setting and managing multiple optimization indicators for brake energy recovery and brake safety; A weight optimization module for determining and optimizing the weight ratio of each optimization indicator to achieve multi-objective balance; A real-time optimization module for searching for the optimal brake pedal displacement and brake motor torque in real time under actual conditions; A brake control module for controlling the brake energy recovery and brake safety of the vehicle electronic brake system according to the optimization results, and coordinating the distribution of friction braking and motor braking force under ABS activation conditions.

Images