Air suspension and ESC coordination control method and system based on deep learning

Abstract

The invention discloses an air suspension and ESC coordination control method and system based on deep learning, and belongs to the field of automobile braking, and the method comprises the following steps: S1, collecting historical data of vehicle driving, generating extreme working condition data by using vehicle dynamics simulation software, and forming a training data set; s2, constructing a double-branch fusion control model, and training the double-branch fusion control model; s3, vehicle state parameters and road surface condition data are collected in real time; s4, after feature extraction and fusion are carried out on the processed multi-source parameters, the trained double-branch fusion control model is input, and an optimal control strategy is generated; and S5, the ESC executes the optimal control strategy and collects vehicle state data after execution to form a control closed loop. By the adoption of the air suspension and ESC coordinated control method and system based on deep learning, the deep learning and the vehicle dynamics model are combined, the vehicle motion state can be monitored in real time, accurate control is conducted according to real-time data, and the potential out-of-control situation is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of automotive braking, and particularly to a coordinated control method and system for an air suspension and an ESC based on deep learning. Background Art

[0002] In the wave of automotive intelligence and electrification, vehicle chassis control technology is undergoing profound changes, gradually moving from the traditional independent control mode of a single subsystem to the direction of multi-system collaborative optimization. In this development process, the air suspension system and the electronic stability control system (ESC) have emerged as the core components for enhancing vehicle safety, comfort, and handling performance. The above-mentioned technology integration and intelligent upgrade have become the research focus in the current automotive industry.

[0003] In recent years, academia and industry have actively engaged in the exploration of coordinated control of air suspension and ESC. Many studies have attempted to optimize suspension stiffness and ESC braking force distribution by means of model predictive control (MPC) or fuzzy logic algorithms, combined with vehicle body attitude and tire force information. These efforts have achieved certain results to provide new ideas and methods for improving vehicle performance.

[0004] However, it cannot be ignored that the existing coordinated control methods still face severe challenges. On the one hand, the complexity of data fusion is prominent. During the vehicle operation, data from multi-source heterogeneous sensors such as inertial measurement units, wheel speed sensors, and pressure sensors are of various types and characteristics. When processing these data in real time, not only efficient data processing algorithms are required, but also high-precision dynamic state estimation needs to be ensured. At present, there is still a large room for improvement in this regard, and the real-time performance of data processing and the accuracy of dynamic state estimation are difficult to reach the ideal level.

[0005] On the other hand, the adaptability of the model has limitations. Traditional coordinated control is mostly based on physical models, and these models show obvious deficiencies when facing complex and changeable actual working conditions. Especially in extreme scenarios of strong nonlinearity and coupling, such as when the vehicle is cornering at high speed, braking emergently, and the road surface is wet and slippery, traditional physical models are often difficult to accurately describe the dynamic characteristics of the vehicle, resulting in the failure of the coordinated control strategy and the inability to effectively ensure the safety and stability of the vehicle. Summary of the Invention

[0006] The purpose of the present invention is to provide a coordinated control method and system for an air suspension and an ESC based on deep learning to solve the above technical problems.

[0007] To achieve the above object, the present invention provides a coordinated control method for an air suspension and an ESC based on deep learning, comprising the following steps: S1. Collect historical data of vehicle driving, and generate extreme condition data by using vehicle dynamics simulation software to form a training data set; S2. Build a dual-branch fusion control model based on a deep learning model, define the control objectives of the air suspension and the ESC actuator at the same time, and then train the dual-branch fusion control model by using the training data set; S3. Multi-source parameter acquisition: Real-time collect vehicle state parameters and road surface condition data, perform data cleaning and time alignment processing by using linear interpolation; S4. After feature extraction and fusion of the processed multi-source parameters, input them into the trained dual-branch fusion control model to generate an optimal control strategy; S5. The ESC executes the optimal control strategy and collects the vehicle state data after execution to form a control closed loop.

[0008] Preferably, the historical data of vehicle driving described in step S1 includes historical vehicle state parameters and road surface condition data; Moreover, the historical vehicle state parameters described in step S1 and the vehicle state parameters described in step S3 both include vehicle dynamics parameters, air suspension and ESC actuator state parameters; wherein, the vehicle dynamics parameters include vehicle body three-axis acceleration, vehicle speed, steering wheel angle, yaw rate, pitch angle, roll angle and wheel speed; the air suspension and ESC actuator state parameters include air suspension stiffness and execution feedback signals, and the execution feedback signals include air suspension height adjustment amount, braking torque distribution result, engine torque; the road surface condition data includes road surface friction coefficient estimation value and road surface unevenness.

[0009] Preferably, an LSTM model is used to learn the time series features and static features to estimate the real-time road surface friction coefficient; The input layer of the LSTM model is used to input time series features and static features, wherein the time series features include slip rate, lateral force, vertical load, and the static features include road surface type coding and road surface humidity; the LSTM layer of the LSTM model is used to capture time series dependence relationships; the output layer of the LSTM model is used to output the current friction coefficient estimation value; Moreover, the training objective of the LSTM model is set to minimize the mean square error between the predicted friction coefficient estimation value and the true value of the friction coefficient: (1); In the formula, represents the mean square error loss function; represents the number of samples; represents the th sample's friction coefficient estimation value; Represents the true value of the friction coefficient of the th sample.

[0010] Preferably, the double-branch fusion control model described in step S2 includes an ESC control sub-model and an air suspension control sub-model, and the ESC control sub-model communicates with the air suspension control sub-model via a bus to form a joint observation , represents the state space of the air suspension control sub-model, represents the state space of the ESC control sub-model; Among them, the state space of the air suspension control sub-model , represents the body vertical acceleration; represents the dynamic deflection of the air suspension; represents the body vertical displacement; represents the road surface roughness spectrum; the action space is the active force of the suspension or the damping adjustment coefficient , and its action value , represents the array; The state space of the ESC control sub-model , represents the steering wheel angle; represents the yaw angular velocity, and ; represents the lateral acceleration; represents the wheel slip ratio; the action space is the braking torque distribution coefficient or the stability control torque , and its action value ; Set the multi-objective weighted loss function of the ESC control sub-model as follows: (2); In the formula, represents the total loss of the ESC control sub-model; , and both represent weight coefficients; represents the safety loss, and its optimization goal is to minimize the sideslip angle and the yaw angular velocity error ; represents the first comfort loss, and its optimization goal is to smooth the braking pressure change by punishing the braking pressure change rate under the premise of safety, represents the pressure change value within the time period ; represents the compliance loss; and minimize the sideslip angle and the yaw rate error The expressions are as follows: (3); (4); In the formula, represents the vehicle speed, represents the wheelbase; represents the reference yaw rate; Set the multi-objective weighted loss function of the air suspension control sub-model as follows: (5); In the formula, represents the total loss of the air suspension control sub-model; represents the second comfort loss, and its optimization objective is to minimize the root mean square value of the vehicle body vertical acceleration and the pitch angular velocity of the root mean square value ; represents the loss of stepping on empty, and its penalty term is , represents the displacement of the wheel in the vertical direction, represents the displacement of the road surface in the vertical direction; , and both represent the weight coefficients; represents the energy consumption loss; and minimize the vehicle body vertical acceleration and the pitch angular velocity of the root mean square value The expressions are as follows: (6); In the formula, represents the vehicle body vertical acceleration of the th sample; represents the pitch angular velocity of the th sample; represents the number of samples.

[0011] Preferably, in step S2, set the dual-branch fusion control model to adopt a centralized training-distributed execution strategy: using the MADDPG framework, the central Critic network collects the state space information and action space information of the ESC control sub-model and the air suspension control sub-model; the Actor network generates actions based on the state space information of the ESC control sub-model or the air suspension control sub-model. Set the optimizer to AdamW or RMSprop, and use a sliding window or sequence generator to implement real-time data stream training.

[0012] Preferably, the linear interpolation expression described in step S3 is as follows: (7); In the formula, represents the data after time alignment; and respectively represent two original data at adjacent times; and respectively represent two timestamps at adjacent times; represents the target timestamp.

[0013] Preferably, step S4 specifically includes the following steps: S41. Feature extraction: For vehicle state parameters, use LSTM or Transformer for time series modeling to extract vehicle state time series features; For road surface condition data, extract local features through a CNN network; S42. Input the vehicle state time series features and local features into Transformer to capture global time series correlations using Transformer; S421. Set the feature sequence , respectively represent the features at the th moment in the feature sequence . Linearly transform the feature sequence input into Transformer into a query matrix , a key matrix and a value matrix : (8); In the formula, , and all represent attention weights, and , represents the model dimension; represents the attention head dimension; S422. Dynamically allocate attention weights through scaled dot-product attention: (9); S423. Use the multi-head attention mechanism to splice the results of multiple attention heads and then perform a linear transformation: (10); In the formula, represents the output result after splicing and linear transformation by the multi-head attention mechanism; respectively represent the output results of the -th attention head, represents the number of attention heads; represents the output weight, and , represents the dimension of the value; S43. Feature fusion: Concatenate the global temporal correlation captured by the Transformer with the vehicle state time series features and local features to obtain fused features; S44. Input the fused features into the dual-branch fusion control model to generate an optimal control strategy.

[0014] Preferably, in step S41, for the road surface condition data, first use multiple convolutional kernels of the CNN network to extract local features: (11); In the formula, represents the output local feature; represents the activation function; represents the convolutional kernel weight; represents the data point of the input sequence at ; represents the convolutional kernel size; represents the bias term; Then perform dimensionality reduction through max pooling: (12); In the formula, represents the result after max pooling dimensionality reduction; respectively represent the features at the moment within the pooling window; represents the pooling window size.

[0015] Preferably, the optimal control strategy generated in step S44 includes a suspension control strategy and an ESC control strategy. Among them, the suspension control strategy includes a body height control strategy and a suspension stiffness adjustment strategy. The expression of the body height control strategy is as follows: (13); In the formula, represents the driving mode, and , when it represents the comfort mode, when it represents the sports mode, and when it represents the off-road mode; represents the road surface unevenness level; , , , , and All represent fitting coefficients; represents the dynamic compensation term, and , and respectively represent the lateral acceleration and longitudinal acceleration at the centroid, and respectively represent the wheelbase and track width, represents the gravitational acceleration; The expression of the suspension stiffness adjustment strategy is as follows: (14); In the formula, represents the adjusted suspension stiffness; represents the basic stiffness; and All represent empirical coefficients; The ESC control strategy is to adjust the braking force distribution and torque, where The formula for adjusting the braking force distribution is as follows: (15); In the formula, represents the braking force on the front axle; and All represent distribution coefficients, and ; represents the total braking force; represents the braking force on the rear axle; The formula for adjusting the torque is as follows: (16); In the formula, represents the torque transmitted to the driving wheels; represents controlling the torque of the engine by adjusting the throttle opening; represents the transmission ratio; represents the efficiency.

[0016] A system for implementing a coordinated control method of an air suspension and ESC based on deep learning includes: A historical data collection module for collecting historical data of vehicle driving and generating extreme condition data using vehicle dynamics simulation software to form a training data set; A model construction module for constructing a dual-branch fusion control model based on a deep learning model, defining the control objectives of the air suspension and ESC actuators, and then training the dual-branch fusion control model using the training data set; The real-time data acquisition module is used to acquire vehicle state parameters and road surface condition data in real time, perform data cleaning, and conduct time alignment processing using linear interpolation. It includes a brake pressure sensor, a wheel speed sensor, an inertial measurement unit, a gyroscope, a yaw rate sensor, a road surface perception sensor, and a suspension state sensor; The optimal strategy generation module is used to extract and fuse features from the processed multi-source parameters, and then input them into the trained dual-branch fusion control model to generate an optimal control strategy; The execution module, the ESC executes the optimal control strategy and acquires the vehicle state data after execution to form a control closed-loop.

[0017] Therefore, the present invention adopts the above-mentioned air suspension and ESC coordinated control method and system based on deep learning, and the beneficial effects are as follows: 1. In terms of comfort, the system automatically adjusts the suspension stiffness and damping according to road surface excitation and driving mode, and combines with the braking torque optimization of the ESC, significantly reducing vehicle body vibration and enhancing steering sensitivity; 2. In terms of safety, the system realizes rapid response to extreme conditions such as sideslip and tire blowout through active adjustment of suspension stiffness-damping and predictive braking force distribution of the ESC, and supports universality across vehicle types (sedan / SUV / truck) and environments (desert / snow / mud); 3. It breaks through the limitations of traditional single subsystem control, and for the first time realizes deep coordination between the air suspension (mechanical system) and the ESC (electronic control system), and solves the performance conflicts under complex working conditions through multi-physical field coupling control (such as joint optimization of stiffness-damping-braking torque); 4. The online incremental learning and reinforcement learning capabilities based on deep learning enable the system to have the potential for lifelong optimization, and can dynamically adapt to driver habits and environmental changes, and its performance and scalability lay a foundation for future intelligent chassis technologies (such as autonomous driving scenarios).

[0018] In summary, the present invention realizes a comprehensive improvement in vehicle performance through real-time data fusion and dynamic decision-making.

[0019] The technical solution of the present invention will be further described in detail below through the accompanying drawings and embodiments. Description of the Drawings

[0020] Figure 1 It is a flow block diagram of the air suspension and ESC coordinated control method based on deep learning of the present invention. Detailed Embodiment

[0021] In order to make the objectives, technical solutions, and advantages of the embodiments of the present invention more clearly understood, the following further details the embodiments of the present invention in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not used to limit the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout.

[0022] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0023] The following further details the embodiments of the present invention in conjunction with the accompanying drawings.

[0024] As Figure 1 shown, a coordinated control method for an air suspension and ESC based on deep learning includes the following steps: S1. Collect historical data of vehicle driving and generate extreme condition data using vehicle dynamics simulation software (such as CarSim / Simulink co-simulation) to form a training data set; The historical data of vehicle driving described in step S1 includes historical vehicle state parameters and road surface condition data; Moreover, both the historical vehicle state parameters described in step S1 and the vehicle state parameters described in step S3 include vehicle dynamics parameters, air suspension and ESC actuator state parameters; among them, the vehicle dynamics parameters include vehicle body three-axis acceleration, vehicle speed, steering wheel angle, yaw rate, pitch angle, roll angle, and wheel speed; the air suspension and ESC actuator state parameters include air suspension stiffness and execution feedback signals, and the execution feedback signals include air suspension height adjustment amount, braking torque distribution result, and engine torque; the road surface condition data includes road surface friction coefficient estimate and road surface unevenness.

[0025] Use an LSTM model to learn temporal features and static features to estimate the real-time road surface friction coefficient; The input layer of the LSTM model is used to input temporal features and static features, where the temporal features include slip ratio, lateral force, and vertical load, and the static features include road surface type encoding and road surface humidity; the LSTM layer of the LSTM model is used to capture temporal dependence relationships; the output layer of the LSTM model is used to output the current friction coefficient estimate value; Moreover, the training objective of the LSTM model is set to minimize the mean square error between the predicted friction coefficient estimate and the true friction coefficient: (1); In the formula, represents the mean square error loss function; represents the number of samples; represents the th estimated friction coefficient of the sample; represents the th true friction coefficient of the sample.

[0026] S2. Build a dual-branch fusion control model based on the deep learning model, define the control objectives of the air suspension and the ESC actuator, and then train the dual-branch fusion control model using the training dataset; The dual-branch fusion control model described in step S2 includes an ESC control sub-model and an air suspension control sub-model, and the ESC control sub-model and the air suspension control sub-model communicate via a bus to form a joint observation , represents the state space of the air suspension control sub-model, represents the state space of the ESC control sub-model; Among them, the state space of the air suspension control sub-model , represents the body vertical acceleration; represents the dynamic deflection of the air suspension; represents the body vertical displacement; represents the road surface roughness spectrum; the action space is the suspension active force or the damping adjustment coefficient , and its action value , represents the array; The state space of the ESC control sub-model , represents the steering wheel angle; represents the yaw angular velocity, and ; represents the lateral acceleration; represents the wheel slip ratio; the action space is the braking torque distribution coefficient or the stability control torque , and its action value ; Set the multi-objective weighted loss function of the ESC control sub-model as follows: (2); In the formula, Represents the total loss of the ESC control sub-model; , and both represent weight coefficients; Represents the safety loss, and its optimization objective is to minimize the sideslip angle and the yaw rate error ; Represents the first comfort loss, and its optimization objective is to smooth the braking pressure change by penalizing the rate of change of braking pressure under the premise of safety, Represents the pressure change value within the time period ; Represents the compliance loss; In this embodiment, the weight coefficients of the multi-objective weighted loss function are dynamically adjusted according to the driving mode or the real-time state of the vehicle. When the sideslip angle ( represents the maximum sideslip angle value) is detected, the safety weight is automatically increased to more than 0.8; when in the comfort mode, the comfort weight is preferentially allocated to the range of 0.6 - 0.7, and the driving mode signal and the sideslip detection flag are received in real time through the vehicle bus to trigger the weight adjustment.

[0027] And the expressions for minimizing the sideslip angle and the yaw rate error are as follows: (3); (4); In the formula, represents the vehicle speed, represents the wheelbase; represents the reference yaw rate; The multi-objective weighted loss function of the air suspension control sub-model is set as follows: (5); In the formula, represents the total loss of the air suspension control sub-model; Represents the second comfort loss, and its optimization objective is to minimize the root mean square value of the body vertical acceleration and the pitch angular velocity ; ; Represents the loss of stepping on empty, and its penalty term is , represents the displacement of the wheel in the vertical direction, represents the displacement of the road surface in the vertical direction; , and All represent weight coefficients; represents the energy consumption loss; and minimize the vertical acceleration of the vehicle body and the pitch angular velocity of the root mean square value The expression is as follows: (6); In the formula, represents the vertical acceleration of the vehicle body of the th sample; represents the th sample of the pitch angular velocity; represents the number of samples.

[0028] In step S2, it is set that the dual-branch fusion control model adopts a centralized training-distributed execution strategy: using the MADDPG framework, the central Critic network collects the state space information and action space information of the ESC control sub-model and the air suspension control sub-model; the Actor network generates actions based on the state space information of the ESC control sub-model or the air suspension control sub-model; Set the optimizer to AdamW or RMSprop, and use a sliding window or a sequence generator to implement real-time data stream training.

[0029] In this embodiment, the trained model is deployed to the in-vehicle computing unit (DCU), and low-latency inference (≤10 ms) is achieved through technologies such as TensorRT quantization and pruning.

[0030] S3. Multi-source parameter acquisition: Real-time acquisition of vehicle state parameters and road surface condition data, and data cleaning and time alignment processing using linear interpolation are performed; The linear interpolation expression described in step S3 is as follows: (7); In the formula, represents the data after time alignment; and respectively represent two original data at adjacent times; and respectively represent two timestamps at adjacent times; represents the target timestamp.

[0031] In this embodiment, the data cleaning includes using Kalman filtering to eliminate the drift of the inertial measurement unit (IMU), and using moving average filtering to smooth the high-frequency vibration signal; the time alignment processing ensures that the multi-sensor data synchronization accuracy is within ±1 ms, and virtual frame data matching the target timestamp is generated through linear interpolation.

[0032] S4. After extracting and fusing the features of the processed multi-source parameters, input them into the trained dual-branch fusion control model to generate the optimal control strategy; Step S4 specifically includes the following steps: S41. Feature extraction: For vehicle state parameters, use LSTM or Transformer for time series modeling to extract vehicle state time series features; For road surface condition data, extract local features through a CNN network; S42. Input the vehicle state time series features and local features into Transformer to capture global time series correlations using Transformer; S421. Set the feature sequence composed of time series features and local features , respectively represent the features at the th moment in the feature sequence Linearly transform the feature sequence input into Transformer into a query matrix , a key matrix and a value matrix : In the formula, , and all represent attention weights, and , represents the model dimension; represents the attention head dimension; S422. Dynamically allocate attention weights through scaled dot-product attention: : S423. Use the multi-head attention mechanism to splice the results of multiple attention heads and then perform a linear transformation: : In the formula, represents the output result after splicing and linear transformation by the multi-head attention mechanism; respectively represent the output results of the th attention head, represents the number of attention heads; represents the output weight, and , represents the dimension of the value; S43. Feature fusion: Splice the global time series correlations captured by Transformer with the vehicle state time series features and local features to obtain the fused features; S44. Input the fusion features into the dual-branch fusion control model to generate the optimal control strategy.

[0033] Preferably, in step S41, for the road surface condition data, first use multiple convolutional kernels of the CNN network to extract local features: (11); In the formula, represents the output local features; represents the activation function; represents the convolutional kernel weight; represents the input sequence at data points; represents the convolutional kernel size; represents the bias term; Then perform dimensionality reduction through max pooling: (12); In the formula, represents the result after max pooling dimensionality reduction; respectively represent the features at time in the pooling window; represents the pooling window size.

[0034] Preferably, the optimal control strategy generated in step S44 includes a suspension control strategy and an ESC control strategy. Among them, the suspension control strategy includes a vehicle body height control strategy and a suspension stiffness adjustment strategy. The expression of the vehicle body height control strategy is as follows: (13); In the formula, represents the driving mode, and , when it represents the comfort mode, when it represents the sport mode, when it represents the off-road mode; represents the road surface unevenness level; , , , , and all represent fitting coefficients; represents the dynamic compensation term, and , and respectively represent the lateral acceleration and longitudinal acceleration at the centroid, and respectively represent the wheelbase and track width, represents the gravitational acceleration; The expression of the suspension stiffness adjustment strategy is as follows: (14); Wherein, represents the adjusted suspension stiffness, and ; represents the basic stiffness; and both represent empirical coefficients; The ESC control strategy is to adjust the braking force distribution and torque adjustment, where The formula for adjusting the braking force distribution is as follows: (15); Wherein, represents the braking force on the front axle; and both represent distribution coefficients, and , ; represents the total braking force; represents the braking force on the rear axle; The formula for torque adjustment is as follows: (16); Wherein, represents the torque transmitted to the drive wheel; represents the torque of the engine controlled by adjusting the throttle opening; represents the transmission ratio; represents the efficiency, and .

[0035] In this embodiment, safety redundancy is also set in the rule layer of the deep learning model. The safety redundancy includes ESC control redundancy and suspension control redundancy. Among them, the ESC control redundancy adjusts the braking pressure of a single wheel through PID to make the actual yaw rate track the reference value: (17); Wherein, represents the additional yaw moment; , and respectively represent the proportional, integral, and differential coefficients.

[0036] The suspension control redundancy includes passive suspension control and semi-active suspension control. The passive suspension control adjusts the shock absorber damping based on the PID algorithm (such as formula (15)), and the response delay is relatively large (>50ms)); the active suspension control adjusts the damping through the linear mapping of the solenoid valve opening, and it includes the following steps: Step 1. Determine the damping adjustment variable: According to the system requirements, select the physical quantity directly related to the damping (such as the solenoid valve opening, coil current, spool displacement, etc.) as the control variable. For example, establish the solenoid valve opening through experiments or simulations. Linear relationship with damping force : (18); In the formula,[[]]END]] represents the proportionality coefficient,[[]]END]] represents the offset.[[]]END]]

[0037] Step 2: Calibrate the opening-damping curve: On the test bench, by changing the opening of the solenoid valve (such as the displacement of the regulating valve core or the coil current), record the corresponding damping force data.[[]]END]]

[0038] Step 3: Use the least squares method to fit the data to obtain the linear mapping relationship between the opening and the damping.[[]]END]]

[0039] S5. The ESC executes the optimal control strategy and collects the vehicle state data after execution to form a control closed-loop.[[]]END]]

[0040] A system for implementing the coordinated control method of air suspension and ESC based on deep learning, including:[[]]END]] A historical data collection module, which is used to collect the historical data of vehicle driving and generate extreme condition data using vehicle dynamics simulation software to form a training data set;[[]]END]] A model construction module, which is used to construct a double-branch fusion control model based on a deep learning model, define the control objectives of the air suspension and the ESC actuator at the same time, and then use the training data set to train the double-branch fusion control model;[[]]END]] A real-time data acquisition module, which is used to collect vehicle state parameters and road surface condition data in real time, perform data cleaning and time alignment processing using linear interpolation. It includes a brake pressure sensor, a wheel speed sensor, an inertial measurement unit, a gyroscope, a yaw rate sensor, a road surface perception sensor and a suspension state sensor. The road surface perception sensor includes a lidar for identifying road surface unevenness and a camera for visually perceiving road surface texture / water accumulation. The suspension state sensor includes an air spring pressure sensor, a solenoid valve displacement sensor and a damper current feedback sensor for monitoring the damper action[[]]END]] An optimal strategy generation module, which is used to perform feature extraction and fusion on the processed multi-source parameters, input the trained double-branch fusion control model, and generate an optimal control strategy;[[]]END]] An execution module, the ESC executes the optimal control strategy and collects the vehicle state data after execution to form a control closed-loop.[[]]END]]

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions of the present invention or make equivalent replacements, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A coordinated control method for air suspension and ESC based on deep learning, characterized in that: It includes the following steps: S1. Collect historical data of vehicle driving, and generate extreme condition data by using vehicle dynamics simulation software to form a training data set; S2. Build a dual-branch fusion control model based on a deep learning model, define the control objectives of the air suspension and the ESC actuator at the same time, and then use the training data set to train the dual-branch fusion control model; S3. Multi-source parameter acquisition: Real-time collect vehicle state parameters and road surface condition data, perform data cleaning and time alignment processing using linear interpolation; S4. After feature extraction and fusion of the processed multi-source parameters, input them into the trained dual-branch fusion control model to generate an optimal control strategy; S5. The ESC executes the optimal control strategy and collects the vehicle state data after execution to form a control closed-loop.

2. The air suspension and ESC coordinated control method based on deep learning according to claim 1, characterized in that: The historical data of vehicle driving described in step S1 includes historical vehicle state parameters and road surface condition data; Moreover, the historical vehicle state parameters described in step S1 and the vehicle state parameters described in step S3 both include vehicle dynamics parameters, air suspension and ESC actuator state parameters; among them, the vehicle dynamics parameters include vehicle body three-axis acceleration, vehicle speed, steering wheel angle, yaw angular velocity, pitch angle, roll angle and wheel speed; the air suspension and ESC actuator state parameters include air suspension stiffness and execution feedback signals, and the execution feedback signals include air suspension height adjustment amount, braking torque distribution result, engine torque; the road surface condition data includes road surface friction coefficient estimation value and road surface unevenness.

3. The coordinated control method of air suspension and ESC based on deep learning according to claim 2, characterized in that: Use the LSTM model to learn the time series features and static features to estimate the real-time road surface friction coefficient; The input layer of the LSTM model is used to input time series features and static features, where the time series features include slip ratio, lateral force, and vertical load, and the static features include road surface type coding and road surface humidity; the LSTM layer of the LSTM model is used to capture time series dependencies; the output layer of the LSTM model is used to output the current friction coefficient estimation value; Moreover, the training objective of the LSTM model is set to minimize the mean square error between the predicted friction coefficient estimation value and the true value of the friction coefficient: （1）； In the formula, represents the mean square error loss function; represents the number of samples; represents the estimated value of the friction coefficient of the th sample; represents the true value of the friction coefficient of the 4. The coordinated control method of air suspension and ESC based on deep learning according to claim 2, characterized in that: The dual-branch fusion control model described in step S2 includes an ESC control sub-model and an air suspension control sub-model, and the ESC control sub-model and the air suspension control sub-model communicate via a bus to form a joint observation , represents the state space of the air suspension control sub-model, represents the state space of the ESC control sub-model; Among them, the state space of the air suspension control sub-model , represents the vertical acceleration of the vehicle body; represents the dynamic deflection of the air suspension; represents the vertical displacement of the vehicle body; represents the road surface roughness spectrum; the action space is the active force of the suspension or the damping adjustment coefficient , and its action value , represents an array; State space of the ESC control sub-model , represents the steering wheel angle; represents the yaw rate, and ; represents the lateral acceleration; represents the wheel slip ratio; the action space is the braking torque distribution coefficient or the stability control torque and its action value ; Set the multi-objective weighted loss function of the ESC control sub-model as follows: （2）； In the formula, represents the total loss of the ESC control sub-model; , and both represent weight coefficients; represents the safety loss, and its optimization objective is to minimize the sideslip angle and the yaw rate error ; represents the first comfort loss, and its optimization objective is to smooth the braking pressure change by penalizing the braking pressure change rate under the premise of safety, represents the pressure change value within the time period ; represents the compliance loss; and minimize the sideslip angle and the yaw rate error The expressions are as follows: （3）； （4）； In the formula, represents the vehicle speed, represents the wheelbase; represents the reference yaw rate; Set the multi-objective weighted loss function of the air suspension control sub-model as follows: （5）； In the formula, represents the total loss of the air suspension control sub-model; represents the second comfort loss, and its optimization objective is to minimize the root mean square value of the vertical acceleration of the vehicle body and the pitch angular velocity ; represents the loss of stepping on empty, and its penalty term is , represents the displacement of the wheel in the vertical direction, and , and all represent weight coefficients; represents the energy consumption loss; and minimize the vertical body acceleration and the pitch angular velocity of the root mean square value The expression is as follows: （6）； In the formula, represents the vertical body acceleration of the th sample; represents the pitch angular velocity of the th sample; represents the number of samples.

5. The coordinated control method of the air suspension and ESC based on deep learning according to claim 4, characterized in that: In step S2, set the dual-branch fusion control model to adopt a centralized training-distributed execution strategy: Use the MADDPG framework, and the central Critic network collects the state space information and action space information of the ESC control sub-model and the air suspension control sub-model; the Actor network generates actions based on the state space information of the ESC control sub-model or the air suspension control sub-model; Set the optimizer as AdamW or RMSprop, and use a sliding window or a sequence generator to implement real-time data stream training.

6. The coordinated control method of air suspension and ESC based on deep learning according to claim 4, characterized in that: The linear interpolation expression described in step S3 is as follows: （7）； In the formula, represents the data after time alignment; and respectively represent two original data at adjacent times; and respectively represent two timestamps at adjacent times; represents the target timestamp.

7. The coordinated control method of air suspension and ESC based on deep learning according to claim 4, characterized in that: Step S4 specifically includes the following steps: S41. Feature extraction: For vehicle state parameters, use LSTM or Transformer for time series modeling to extract vehicle state time series features; For road surface condition data, extract local features through the CNN network; S42. Input the vehicle state time series features and local features into the Transformer to capture the global temporal correlations using the Transformer; S421. Set a feature sequence composed of time series features and local features , respectively represent the features at the th moment in the feature sequence . Linearly transform the feature sequence input to the Transformer into a query matrix , a key matrix and a value matrix : （8）； In the formula, , and all represent attention weights, and , represents the model dimension; represents the attention head dimension; S422. Dynamically allocate attention weights through scaled dot-product attention: （9）； S423. Use the multi-head attention mechanism to splice the results of multiple attention heads and then perform a linear transformation: （10）； In the formula, represents the output result after the concatenation and linear transformation of the multi-head attention mechanism; respectively represent the output results of the -th attention head, represents the number of attention heads; represents the output weight, and , represents the dimension of the value; S43. Feature fusion: Splice the global temporal correlations captured by the Transformer with the vehicle state time series features and local features to obtain fused features; S44. Input the fused features into the dual-branch fusion control model to generate the optimal control strategy.

8. The coordinated control method of air suspension and ESC based on deep learning according to claim 7, characterized in that: In step S41, for the road surface condition data, first use multiple convolutional kernels of the CNN network to extract local features: （11）； Wherein, represents the local feature of the output; represents the activation function; represents the convolutional kernel weight; represents the input sequence at data points; represents the convolutional kernel size; represents the bias term; Then perform dimensionality reduction through max pooling: （12）； In the formula, represents the result after maximum pooling dimensionality reduction; respectively represent the features at moments within the pooling window; represents the pooling window size.

9. The coordinated control method of air suspension and ESC based on deep learning according to claim 7, characterized in that: The optimal control strategy generated in step S44 includes a suspension control strategy and an ESC control strategy. Among them, the suspension control strategy includes a vehicle body height control strategy and a suspension stiffness adjustment strategy. The expression of the vehicle body height control strategy is as follows: （13）； Wherein, represents the driving mode, and when it represents the comfort mode, when it represents the sport mode, when it represents the off-road mode; represents the road surface roughness level; 、 、 、 、 and all represent fitting coefficients; represents the dynamic compensation term, and , and respectively represent the lateral acceleration and longitudinal acceleration at the centroid, and respectively represent the wheelbase and track width, represents the acceleration due to gravity; The expression of the suspension stiffness adjustment strategy is as follows: （14）； In the formula, represents the adjusted suspension stiffness; represents the basic stiffness; and both represent empirical coefficients; The ESC control strategy is to adjust the braking force distribution and torque, where The formula for adjusting the braking force distribution is as follows: （15）； Wherein, represents the front axle braking force; and both represent the distribution coefficients, and ; represents the total braking force; represents the rear axle braking force; The formula for adjusting the torque is as follows: （16）； wherein, represents the torque transmitted to the drive wheel; represents controlling the torque of the engine by adjusting the throttle opening; represents the transmission ratio; represents the efficiency.

10. A system for implementing the deep learning-based coordinated control method of air suspension and ESC according to any one of claims 1-9 above, characterized in that: Includes: A historical data collection module for collecting the historical data of vehicle driving and generating extreme condition data using vehicle dynamics simulation software to form a training data set; A model construction module for constructing a dual-branch fusion control model based on a deep learning model, defining the control objectives of the air suspension and ESC actuators at the same time, and then training the dual-branch fusion control model using the training data set; A real-time data acquisition module for real-time collecting vehicle state parameters and road surface condition data, performing data cleaning and time alignment processing using linear interpolation. It includes a brake pressure sensor, a wheel speed sensor, an inertial measurement unit, a gyroscope, a yaw rate sensor, a road surface perception sensor, and a suspension state sensor; An optimal strategy generation module for performing feature extraction and fusion on the processed multi-source parameters and then inputting them into the trained dual-branch fusion control model to generate the optimal control strategy; An execution module where the ESC executes the optimal control strategy and collects the vehicle state data after execution to form a control loop.

Images