An electro-hydraulic steer-by-wire system based on CPS architecture and its energy prediction and optimization method

Abstract

This invention discloses an electro-hydraulic steer-by-wire system based on a Cyber-Physical System (CPS) architecture and its energy prediction and optimization method. The system includes a physical electro-hydraulic steer-by-wire unit, an energy prediction and optimization unit, and a big data platform. The big data platform communicates with the physical electro-hydraulic steer-by-wire unit and the energy prediction and optimization unit. In this invention, the physical electro-hydraulic steer-by-wire unit serves as the physical layer in the Cyber-Physical System (CPS) architecture, while the energy prediction and optimization unit and the big data platform serve as the information layer in the Cyber-Physical System (CPS) architecture. By integrating the Cyber-Physical System, the energy-saving potential of the electro-hydraulic steer-by-wire chassis is fully utilized.

Description

Technical Field

[0001] This invention belongs to the field of automotive steer-by-wire chassis technology, specifically relating to an electro-hydraulic steer-by-wire system based on CPS architecture and its energy prediction and optimization method. Background Technology

[0002] With the rapid development of the automotive industry, the economy, safety, and comfort of automobiles are receiving increasing attention. The vehicle steering system serves as a bridge and link connecting the human-vehicle-road closed-loop system, affecting not only the driver's handling feel and driving safety but also being a crucial component of vehicle energy flow. As demands for energy conservation, environmental protection, and safety in vehicles continue to rise, energy efficiency and electrification of steering systems have become inevitable trends. Electro-hydraulic hybrid steering systems combine the advantages of electric power steering and hydraulic power steering systems, featuring adjustable steering modes based on operating conditions. This ensures good steering feel while providing sufficient steering assistance and reducing energy consumption.

[0003] In congested urban rush hour conditions, existing electro-hydraulic steering systems based on high-pressure electric steering pumps account for the third largest share of energy consumption in commercial vehicles, after the powertrain and air conditioning. Electro-hydraulic drive-by-wire chassis can not only directly optimize chassis energy consumption by coordinating the electro-hydraulic steering's operating modes, but also integrate intelligent transportation cyber-physical systems (CPS) for energy prediction and optimization, achieving energy conservation and emission reduction throughout the entire lifecycle.

[0004] Chinese utility model patent application number CN202020191955.0, entitled "A steer-by-wire system capable of recovering driver steering energy," proposes a steer-by-wire system that recovers steering energy by adding a power generation mechanism to the existing steering system. However, this patent neglects the steering requirements of vehicles under various driving conditions, and the addition of the power generation mechanism also increases the control complexity and production cost of the steer-by-wire chassis system. Chinese invention patent application number CN202111177986.6, entitled "An energy management method for an electro-hydraulic integrated steering system based on long-term and short-term fusion," proposes a method for energy management based on dynamic programming algorithms. This method uses system switching frequency as an objective to plan the optimal mode sequence in the long-term future. Simultaneously, it embeds an equivalent fuel consumption minimization strategy into the inner layer of the dynamic programming to solve the torque distribution problem in the short-term domain. However, this patent fails to integrate intelligent transportation cyber-physical systems to predict the longitudinal and lateral coupling conditions during the driving of electro-hydraulic steer-by-wire vehicles, and the established model has low adaptability to these coupling conditions.

[0005] Existing chassis energy optimization methods typically ignore the impact of future vehicle driving conditions and driver intentions, failing to leverage the energy-saving potential of electro-hydraulic drive-by-wire chassis by integrating intelligent transportation cyber-physical systems (ITS). For light commercial vehicles primarily used in urban delivery, their driving routes are relatively fixed and follow certain patterns. Therefore, designing chassis energy optimization strategies for commercial vehicles based on predicting demand steering according to driving conditions within an ITS environment holds great promise. Summary of the Invention

[0006] In view of the shortcomings of the prior art, the purpose of this invention is to provide an electro-hydraulic steer-by-wire system based on CPS architecture and its energy prediction and optimization method, so as to solve the problem that the existing technology cannot realize the energy-saving potential of the electro-hydraulic steer-by-wire chassis because it ignores the influence of the vehicle's future driving conditions and driving intentions.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] The present invention discloses an electro-hydraulic steer-by-wire system based on a CPS architecture, comprising: a physical electro-hydraulic steer-by-wire unit, an energy prediction and optimization unit, and a big data platform; the big data platform communicates with the physical electro-hydraulic steer-by-wire unit and the energy prediction and optimization unit respectively.

[0009] The physical electro-hydraulic steer-by-wire unit includes: hydraulic components, electrical components, and mechanical components. The electrical components send control signals to control the mechanical and hydraulic components respectively. The hydraulic components are connected to the mechanical components through hydraulic lines. Based on data from a big data platform, the physical electro-hydraulic steer-by-wire unit controls the mechanical and hydraulic components through the electrical components to achieve electro-hydraulic or electric steering of the vehicle.

[0010] The energy prediction and optimization unit includes: a vehicle driving data acquisition module, a longitudinal and lateral coupling working condition prediction module, and a steering energy optimization module;

[0011] The vehicle driving data acquisition module is used to collect vehicle sensor data and store it in a big data platform.

[0012] The longitudinal and lateral coupling working condition prediction module is used to predict the longitudinal and lateral coupling working conditions of the vehicle by collecting the steering wheel angle and vehicle speed from the vehicle driving data acquisition module.

[0013] The steering energy optimization module is used to optimize the vehicle's steering energy.

[0014] The big data platform includes: a data storage repository and a data analysis module;

[0015] A data repository is used to store data for the physical electro-hydraulic steering unit and the energy prediction and optimization unit;

[0016] The data analysis module is used to perform smoothing filtering and correlation analysis on the data stored in the data repository to screen out key parameters that affect the operating characteristics of electro-hydraulic drive-by-wire chassis vehicles.

[0017] Furthermore, the hydraulic components include: a vane pump, an accumulator, a solenoid valve, and a hydraulic cylinder; the vane pump is connected to the accumulator via a hydraulic line, the solenoid valve is connected to the accumulator via a hydraulic line, and the hydraulic cylinder is connected to the solenoid valve via a hydraulic line.

[0018] Furthermore, the electrical components include: an electronic control unit (ECU), a pump drive motor, and an auxiliary motor; the ECU controls the working state of the pump drive motor and the auxiliary motor through control signals, the pump drive motor drives the vane pump through a mechanical connection, and the ECU controls the opening and closing degree of the solenoid valve through control signals.

[0019] Furthermore, the mechanical components include: a steering wheel, a reducer, a steering shaft, and wheels; the steering wheel and the reducer are connected through the steering shaft, the steering shaft and the wheels are connected through a gear and rack, the power steering motor is mounted on the reducer to provide steering assistance, the hydraulic cylinder is connected to the wheels through a rack, and the electronic control unit (ECU) receives steering signals from the steering wheel through a steering angle sensor.

[0020] Furthermore, the vehicle's sensor data includes: vehicle position coordinates, steering wheel angle, driver torque, vehicle speed, acceleration, yaw rate, and steering wheel angular velocity.

[0021] Furthermore, the longitudinal and transverse coupling condition prediction module uses an encoder decoupling prediction network based on a time-series Transformer model to predict longitudinal and transverse coupling conditions.

[0022] Furthermore, the steering energy optimization module employs an adaptive particle swarm optimization algorithm to optimize the steering energy.

[0023] Furthermore, during electric steering, the electronic control unit (ECU) receives the steering wheel signal and controls the power assist motor to transmit torque to the reducer. The reducer reduces the combined torque of the steering wheel and the power assist motor and then transmits it to the steering shaft. The steering shaft acts on the wheels through a rack and pinion mechanism to complete the electric steering.

[0024] Furthermore, during electro-hydraulic steering, the electronic control unit (ECU) receives the steering wheel steering signal and, while executing electric steering, controls the pump drive motor to drive the vane pump. The vane pump pumps in liquid to store energy in the accumulator. The accumulator releases energy to drive the solenoid valve. The solenoid valve drives the hydraulic cylinder by changing the liquid pressure and flow direction in the hydraulic line. The hydraulic cylinder then drives the rack to act on the wheel to complete the electro-hydraulic steering.

[0025] The present invention provides an energy prediction and optimization method for an electro-hydraulic steer-by-wire system based on a CPS architecture. Based on the aforementioned system, the steps are as follows:

[0026] 1) Collect vehicle sensor data and process it through a big data platform;

[0027] 2) Using the steering wheel angle and vehicle speed from the processed vehicle sensor data as a multivariate time series, and based on the designed working condition prediction network, perform longitudinal and lateral coupled working condition prediction of the vehicle.

[0028] 3) Establish an electro-hydraulic steering optimization allocation model based on the predicted working conditions and optimize the energy of the electro-hydraulic steer-by-wire.

[0029] 4) Set optimization conditions to complete the energy prediction optimization of the electro-hydraulic steering-by-wire system.

[0030] Furthermore, step 1) specifically includes: the vehicle driving data acquisition module collects vehicle position coordinates, steering wheel angle, driver torque, vehicle speed, acceleration, yaw rate, and steering wheel angular velocity data and stores them in the data storage repository of the big data platform.

[0031] Furthermore, in step 1), the data analysis module uses the moving average method to smooth and filter the vehicle position coordinates, steering wheel angle, driver torque, vehicle speed, acceleration, yaw rate, and steering wheel angular velocity data in the data repository, eliminating outliers in the data and obtaining a dataset that meets the requirements.

[0032] Furthermore, the formula for calculating the moving average method is as follows:

[0033]

[0034] Among them, y t-n For the data before smoothing, y t The data is smoothed, and n is the number of filter terms.

[0035] Furthermore, in step 1), the Pearson correlation coefficient is used to perform correlation analysis between the smoothed and filtered data and the steering torque and braking torque, respectively. The closer the Pearson correlation coefficient is to 1, the higher the correlation between the data and the steering torque and braking torque. This shows that vehicle speed and steering wheel angle have a high correlation with steering torque and braking torque, which are key parameters affecting the operating characteristics of electro-hydraulic drive-by-wire chassis vehicles. The correlation analysis calculation formula is as follows:

[0036]

[0037] Where, ρ X，YHere, σ is the Pearson correlation coefficient, X and Y are the two variables used in the correlation analysis, cov(x, Y) is the covariance of X and Y, and σ is the Pearson correlation coefficient. X and σ Y Let μ be the standard deviation of X and Y, respectively. X and μ Y Let X and Y be the average values, respectively, and E be the expected value.

[0038] Furthermore, the working condition prediction network in step 2) is an encoder decoupling prediction network based on the temporal Transformer model, including: an embedding layer, a self-attention mechanism, a normalization layer, a feedforward neural network, and a multilayer perceptron.

[0039] The embedding layer embeds each variable into a variable token according to the method of each variable, and obtains the dependencies between different variables through a multi-head self-attention mechanism. The normalization layer eliminates the distribution differences between the variables input to the feedforward neural network. The feedforward neural network encodes and decodes the multivariate time series. The prediction results after decoding by the feedforward neural network are input into the multilayer perceptron for mapping, and then evaluated by mean square error and root mean square error.

[0040] Furthermore, the specific steps for predicting operating conditions in step 2) are as follows:

[0041] 21) By embedding the variable tokens according to the vehicle speed and steering wheel angle through the embedding layer, feature representations of vehicle speed and steering wheel angle are obtained;

[0042] 22) Employing a multi-head self-attention mechanism, the dependency relationship between vehicle speed and steering wheel angle time series in the information physical traffic environment is obtained by calculating the correlation of different variables;

[0043] 23) Apply the normalization layer to the feature representation of each variable so that the feature sequences of all variables are under a relatively uniform normal distribution, thereby reducing the impact of differences caused by different variable value ranges and inconsistent measurements;

[0044] 24) Feedforward neural networks encode the observed time series through stacking Transformer models and decode the representation of the predicted sequence using dense nonlinear connections;

[0045] 25) Input driving intention, road curvature, and road condition data into the condition prediction network to improve the network's adaptability to longitudinal and lateral coupled conditions. Based on the data from 0 to 800 ms, predict the condition from 800 ms to 1000 ms. The prediction results are evaluated by mean square error and root mean square error.

[0046] Furthermore, in step 21), the key parameters that affect the working characteristics of the electro-hydraulic drive-by-wire chassis vehicle are screened out by the correlation analysis in step 1), namely vehicle speed and steering wheel angle. Vehicle speed reflects the longitudinal working condition during vehicle driving; steering wheel angle reflects the lateral working condition during vehicle driving. Vehicle speed and steering wheel angle are used as multivariate time series to predict the longitudinal and lateral coupled working conditions.

[0047] Furthermore, step 21) is implemented based on a multilayer perceptron, and its specific expression is as follows:

[0048] H 0 =MLP(X)

[0049] Where H 0 X is the variable token, X is the input variable, and MLP is a multilayer perceptron.

[0050] Furthermore, in step 22), the multi-head self-attention mechanism consists of multiple self-attention modules in parallel. Each attention module focuses on different parts of the input sequence and calculates different weighted sums, encoding different aspects of features into different attention modules to form multiple sets of encoding modules.

[0051] Furthermore, the multi-head self-attention mechanism used in step 22) obtains the dependency relationship between vehicle speed and steering wheel time series in the cyber-physical traffic environment by calculating the correlation of different variables; the self-attention module generates a vector representation for each input token, and the input vector is converted into a query vector, key vector, and value vector. The vectors derived from different inputs are packaged into three matrices Q, K, and V. The Softmax function generates a probability distribution based on the pairwise similarity scores between vectors to obtain the dependency relationship between vehicle speed and steering wheel time series, realizing the coupling of longitudinal and lateral prediction conditions. The process expression is as follows:

[0052]

[0053] H l =LayerNorm(H l-1 +Self-Attn(H l-1 ))

[0054] In the formula, d is the scaling factor, Q, K, and V are three matrices, Attention is the attention mechanism function, Softmax is the normalization exponential function, LayerNorm is the normalization processing function, Self-Attn is the self-attention mechanism function, and H... l The input variable H of the l-th encoding module l-1 This is the input variable for the (l-1)th group of encoding modules.

[0055] Furthermore, in step 23), the normalization layer normalizes along the feature dimension. The normalization layer calculates the mean and standard deviation of all features for each data point in a specific layer, and uses the obtained mean and standard deviation of the features to normalize all features in the specific layer.

[0056] The formula for expressing the normalized layer is:

[0057]

[0058]

[0059]

[0060] Output = γ × H norm +β

[0061] Where μ l The mean, Let n be the i-th feature value of the input variable of the l-th encoding module. l Let l represent the number of neuron nodes in the l-th encoding module, where l∈{1,2,...,L} represents the l-th encoding module. Here, H represents the variance, and Output represents the output of the normalization layer. l The input variable H of the l-th encoding module norm The values ​​are normalized values, and γ and β are the learnable scaling and offset parameters, respectively.

[0062] Furthermore, in step 24), the expression for the feedforward neural network encoding the observed time series through stacked Transformer models and decoding the representation of the predicted sequence using dense nonlinear connections is as follows:

[0063] H l =LayerNorm(H l-1 +Feed-Forward(H l-1 ))

[0064] In the formula, LayerNorm is the normalization function, Feed-Forward is the feedforward neural network layer, and H... l-1 H is the input variable for the (l-1)th group of encoding modules. l The input variables of the l-th encoding module.

[0065] Furthermore, in step 24), the feedforward neural network is a two-layer fully connected network, using a Gaussian error linear unit (GELU) as the activation function; the activation function is expressed as:

[0066]

[0067] In the formula, GELU is the Gaussian error linear unit activation function, and x is the function input variable.

[0068] Further, the prediction result of the encoding module in step 24) is input to the multilayer perceptron for mapping, and the expression is:

[0069]

[0070] In the formula, For the mapping of the prediction results, H l The input variables of the l-th encoding module.

[0071] Furthermore, the formulas for evaluating the prediction results in step 25) using mean squared error (MSE) and root mean square error (RMSE) are as follows:

[0072]

[0073]

[0074] In the formula, MSE is the mean square error, RMSE is the root mean square error, and y i As input variables for calculating errors, This is the mean of the input variables.

[0075] Further, the specific process of the rolling time-domain optimization in step 3) is as follows: For the kth sampling time, the rolling time-domain optimization algorithm calculates the optimal control input sequence in the time domain by combining the state information obtained from the working condition prediction, and takes the first item of the sequence as the optimization result at the kth time; at the k+1th time, the above rolling optimization process is repeated cyclically using the updated predicted state information obtained from the working condition prediction, and the ratio of electric steering and electro-hydraulic steering is allocated to achieve energy optimization of electro-hydraulic steer-by-wire; wherein, the length of the rolling sliding window is determined according to the time domain length of 800ms to 1000ms predicted by the working condition.

[0076] Furthermore, the specific steps for optimizing the electro-hydraulic steering-by-wire energy based on the rolling time domain in step 3) are as follows:

[0077] 31) Define the ratio of the target electric power assist torque to the steering power assist torque as the electro-hydraulic steering distribution coefficient;

[0078] 32) Define the average loss over the finite time domain as the steering energy cost function;

[0079] 33) Calculate the target speed of the pump drive motor to represent the smoothness of the electro-hydraulic mode switching;

[0080] 34) Define the cost function for electro-hydraulic switching smoothness;

[0081] 35) Establish an electro-hydraulic steering optimization allocation model and solve for optimization.

[0082] Furthermore, the formula for calculating the electro-hydraulic steering distribution coefficient in step 31) is as follows:

[0083]

[0084] In the formula, x e This is the electro-hydraulic steering distribution coefficient. For the target electric assist torque, This is the steering assist torque.

[0085] Furthermore, in step 32), the average loss over the finite time domain is defined as the steering energy cost function as follows:

[0086]

[0087] In the formula, J1 is the steering energy consumption cost function, P EHCS For the energy consumption of electro-hydraulic combined steering, P total Let t be the total energy consumption, t0 be the start time of the sliding window, and t be the total energy consumption. f This is the end time of the scrolling window.

[0088] Furthermore, in step 33), the target speed of the pump drive motor is related to the total required steering torque, vehicle speed, and electro-hydraulic steering distribution coefficient, and its expression is:

[0089]

[0090] In the formula, The target speed of the pump drive motor is given by V, where v is the vehicle speed and T is the speed of the vehicle. s For the total demand steering torque, θ sw x is the motor rotation angle. e Electro-hydraulic steering distribution coefficient.

[0091] Furthermore, the smoothness of the electro-hydraulic mode switching in step 33) is represented by the rate of change of the target speed of the pump drive motor, expressed as:

[0092]

[0093] In the formula, This represents the rate of change of the target speed of the pump drive motor.

[0094] Further, in step 34), the electro-hydraulic switching smoothness cost function is defined as the time average of the square of the ratio of the target speed change rate of the pump drive motor to its upper limit, and the expression is:

[0095]

[0096] In the formula, J2 is the electro-hydraulic switching smoothness cost function, α maxLet t0 be the rate of change of the upper limit of the target speed of the pump drive motor, and t0 be the start time of the scrolling window. f This is the end time of the scrolling window.

[0097] Furthermore, in step 35), based on the energy consumption and smoothness cost function established in the prediction time domain, the expression for the electro-hydraulic steering optimization allocation model is obtained as follows:

[0098]

[0099] In the formula, w1 and w2 are weighting coefficients; T m1 To assist the motor torque, T m1_max This refers to the upper limit of the assist motor torque; n m1 To increase the motor speed, n m1_max This refers to the upper limit of the motor's speed; P m1 To help reduce motor energy consumption, P m1_max To help the motor's energy consumption limit; T m2 T is the torque of the pump drive motor. m2_max n is the upper limit of the pump drive motor torque. m2 n is the speed of the pump drive motor. m2_max P represents the upper limit of the pump drive motor speed. m2 For the energy consumption of the pump drive motor, P m2_max This represents the upper limit of energy consumption for the pump drive motor.

[0100] Furthermore, the algorithm for solving the electro-hydraulic steering energy optimization problem in step 35) is an adaptive particle swarm optimization algorithm. The specific implementation process is as follows: before the particle swarm optimization algorithm updates the motion inertia of the electro-hydraulic steering optimization allocation model formula, an inertia weight coefficient is added. In the early stage of the search when the particles are in the expansion phase, the global convergence ability is increased. In the later stage of the search, when the particles enter the local mining phase, the local convergence ability is enhanced by decreasing the weight coefficient, thus overcoming the defect of insufficient search ability of the particle swarm optimization algorithm.

[0101] Furthermore, in step 4), the length of the scrolling window is set to 1. Based on the scrolling time domain optimization predicted by the working condition prediction network, the length of the scrolling window is set to the prediction time domain. According to the above settings, the energy prediction optimization of the electro-hydraulic steer-by-wire system is completed, and the optimization results are transmitted to the big data platform. The physical electro-hydraulic steer-by-wire unit completes the electro-hydraulic steering or electric steering of the vehicle according to the optimization results data of the big data platform.

[0102] The beneficial effects of this invention are:

[0103] This invention considers the impact of future vehicle driving conditions and driving intentions, and predicts steering demand based on driving conditions in the context of a cyber-physical system (CPS). The physical electro-hydraulic steering unit serves as the physical layer in the CPS architecture, while the energy prediction and optimization unit and big data platform serve as the information layer. By integrating the CPS, the energy-saving potential of the electro-hydraulic steering chassis is fully utilized.

[0104] This invention uses Pearson correlation coefficient for correlation analysis, which can better measure the strength of the linear association between two variables. Parametric correlation analysis can screen out key parameters that affect the operating characteristics of electro-hydraulic drive-by-wire chassis vehicles, simplifying the calculation of driving condition prediction.

[0105] This invention utilizes an encoder-decoupled prediction network based on a temporal Transformer model. Leveraging the Transformer's ability to capture hidden dependencies between time steps and different time series, it improves prediction accuracy and robustness. Furthermore, addressing the multimodal time series prediction problem under vehicle longitudinal and lateral coupling conditions, this invention embeds each variable into a variable token, treating the time step as a feature of a single variable, thus meeting the prediction requirements for vehicle longitudinal and lateral coupling conditions with relatively low computational power.

[0106] Based on the encoder decoupling prediction network condition prediction based on the temporal Transformer model, this invention adopts an energy prediction optimization method for electro-hydraulic steer-by-wire system based on the rolling time domain. The rolling time domain optimization extends the performance optimization problem at a single point to the overall performance optimization within a certain time domain. The objective function is transformed into the overall performance within a finite time domain starting from the current moment. The time domain window moves forward continuously with the change of time, which has a significant energy-saving characteristic.

[0107] This invention overcomes the shortcomings of traditional PSO search capabilities by using an adaptive particle swarm optimization algorithm during energy prediction optimization, which adds an inertia weight coefficient before the motion inertia in the velocity update formula of the traditional particle swarm optimization algorithm. The inertia weight coefficient is increased in the early stage of the search when the particles are in the expansion phase, thereby improving the global convergence ability. In the later stage of the search, when the particles enter the local mining phase, the inertia weight coefficient is decreased to enhance the local convergence ability. Attached Figure Description

[0108] Figure 1 This is a structural diagram of the system of the present invention.

[0109] Figure 2 This is a flowchart of the method of the present invention.

[0110] Figure 3 This is a diagram of the ED-TFN prediction network architecture of the present invention. Detailed Implementation

[0111] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0112] Reference Figure 1 As shown, an electro-hydraulic steer-by-wire system based on a CPS architecture according to the present invention includes: a physical electro-hydraulic steer-by-wire unit, an energy prediction and optimization unit, and a big data platform; the big data platform communicates with the physical electro-hydraulic steer-by-wire unit and the energy prediction and optimization unit respectively.

[0113] The physical electro-hydraulic steer-by-wire unit includes: hydraulic components, electrical components, and mechanical components. The electrical components send control signals to control the mechanical and hydraulic components respectively. The hydraulic components are connected to the mechanical components through hydraulic lines. Based on data from a big data platform, the physical electro-hydraulic steer-by-wire unit controls the mechanical and hydraulic components through the electrical components to achieve electro-hydraulic or electric steering of the vehicle.

[0114] The hydraulic components include: a vane pump, an accumulator, a solenoid valve, and a hydraulic cylinder; the vane pump is connected to the accumulator via a hydraulic line, the solenoid valve is connected to the accumulator via a hydraulic line, and the hydraulic cylinder is connected to the solenoid valve via a hydraulic line.

[0115] The electrical components include: an electronic control unit (ECU), a pump drive motor, and an auxiliary motor. The ECU controls the working state of the pump drive motor and the auxiliary motor through control signals. The pump drive motor drives the vane pump through a mechanical connection. The ECU controls the opening and closing degree of the solenoid valve through control signals.

[0116] The mechanical components include: a steering wheel, a reducer, a steering shaft, and wheels; the steering wheel and reducer are connected by the steering shaft, the steering shaft and wheels are connected by a gear and rack, the power steering motor is mounted on the reducer to provide steering assistance, the hydraulic cylinder is connected to the wheels by a rack, and the electronic control unit (ECU) receives steering signals from the steering wheel through a steering angle sensor.

[0117] The energy prediction and optimization unit includes: a vehicle driving data acquisition module, a longitudinal and lateral coupling working condition prediction module, and a steering energy optimization module;

[0118] The vehicle driving data acquisition module is used to collect vehicle sensor data and store it in a big data platform.

[0119] The longitudinal and lateral coupling working condition prediction module is used to predict the longitudinal and lateral coupling working conditions of the vehicle by collecting the steering wheel angle and vehicle speed from the vehicle driving data acquisition module.

[0120] The steering energy optimization module is used to optimize the vehicle's steering energy.

[0121] The big data platform includes: a data storage repository and a data analysis module;

[0122] A data repository is used to store data for the physical electro-hydraulic steering unit and the energy prediction and optimization unit;

[0123] The data analysis module is used to perform smoothing filtering and correlation analysis on the data stored in the data repository to screen out key parameters that affect the operating characteristics of electro-hydraulic drive-by-wire chassis vehicles.

[0124] The sensor data of the vehicle includes: vehicle position coordinates, steering wheel angle, driver torque, vehicle speed, acceleration, yaw rate, and steering wheel angular velocity.

[0125] The longitudinal and transverse coupling condition prediction module employs an encoder decoupling transform network (ED-TFN) based on a temporal transform model to predict longitudinal and transverse coupling conditions. (Refer to...) Figure 3 As shown.

[0126] The steering energy optimization module employs an adaptive particle swarm optimization (APSO) algorithm to optimize steering energy.

[0127] When electric steering is applied, the electronic control unit (ECU) receives the steering wheel signal and controls the power assist motor to transmit torque to the reducer. The reducer reduces the combined torque of the steering wheel and the power assist motor and then transmits it to the steering shaft. The steering shaft acts on the wheels through a rack and pinion mechanism to complete the electric steering.

[0128] When electro-hydraulic steering is performed, the electronic control unit (ECU) receives the steering wheel steering signal and executes electric steering. At the same time, it controls the pump drive motor to drive the vane pump. The vane pump pumps in liquid to store energy in the accumulator. The accumulator releases energy to drive the solenoid valve. The solenoid valve drives the hydraulic cylinder by changing the liquid pressure and flow direction in the hydraulic line. The hydraulic cylinder drives the rack and pinion to act on the wheel to complete the electro-hydraulic steering.

[0129] Reference Figure 2 As shown, the present invention provides an energy prediction and optimization method for an electro-hydraulic steer-by-wire system based on a CPS architecture. Based on the aforementioned system, the steps are as follows:

[0130] 1) Collect vehicle sensor data and process it through a big data platform;

[0131] The vehicle driving data acquisition module collects vehicle position coordinates, steering wheel angle, driver torque, vehicle speed, acceleration, yaw rate, and steering wheel angular velocity data and stores them in the data storage repository of the big data platform.

[0132] The data analysis module uses the moving average method to smooth and filter the vehicle position coordinates, steering wheel angle, driver torque, vehicle speed, acceleration, yaw rate, and steering wheel angular velocity data in the data repository, eliminating outliers and obtaining a dataset that meets the requirements.

[0133] The formula for calculating the moving average method is as follows:

[0134]

[0135] Among them, y t-n For the data before smoothing, y t The data is smoothed, and n is the number of filter terms.

[0136] In step 1), the Pearson correlation coefficient is used to perform correlation analysis between the smoothed and filtered data and the steering torque and braking torque. The closer the Pearson correlation coefficient is to 1, the higher the correlation between the data and the steering torque and braking torque. This shows that vehicle speed and steering wheel angle have a high correlation with steering torque and braking torque, which are key parameters affecting the operating characteristics of electro-hydraulic drive-by-wire chassis vehicles. The correlation analysis calculation formula is as follows:

[0137]

[0138] Where, ρ X，Y Here, σ is the Pearson correlation coefficient, X and Y are the two variables used in the correlation analysis, cov(X,Y) is the covariance of X and Y, and σ is the Pearson correlation coefficient. X and σ Y Let μ be the standard deviation of X and Y, respectively. X and μ Y Let X and Y be the average values, respectively, and E be the expected value.

[0139] 2) Using the steering wheel angle and vehicle speed from the processed vehicle sensor data as a multivariate time series, and based on the designed working condition prediction network, perform longitudinal and lateral coupled working condition prediction of the vehicle.

[0140] The working condition prediction network is an encoder-decoupling transformer network (ED-TFN) based on a temporal transformer model, which includes: an embedding layer, a self-attention mechanism, a normalization layer, a feedforward neural network (FNN), and a multi-layer perceptron (MLP).

[0141] The embedding layer embeds each variable into a variable token according to its own method, and obtains the dependencies between different variables through a multi-head self-attention mechanism. The normalization layer eliminates the distribution differences between the variables input to the feedforward neural network. The feedforward neural network encodes and decodes the multivariate time series. The prediction results after decoding by the feedforward neural network are input into the multilayer perceptron for mapping, and then evaluated by mean squared error (MSE) and root mean square error (RMSE).

[0142] The specific steps for performing load condition prediction are as follows:

[0143] 21) By embedding the variable tokens according to the vehicle speed and steering wheel angle through the embedding layer, feature representations of vehicle speed and steering wheel angle are obtained;

[0144] 22) Employing a multi-head self-attention mechanism, the dependency relationship between vehicle speed and steering wheel angle time series in the information physical traffic environment is obtained by calculating the correlation of different variables;

[0145] 23) Apply the normalization layer to the feature representation of each variable so that the feature sequences of all variables are under a relatively uniform normal distribution, thereby reducing the impact of differences caused by different variable value ranges and inconsistent measurements;

[0146] 24) Feedforward neural networks encode the observed time series through stacking Transformer models and decode the representation of the predicted sequence using dense nonlinear connections;

[0147] 25) Input driving intention, road curvature, and road condition data into the condition prediction network to improve the network's adaptability to longitudinal and lateral coupled conditions. Based on the data from 0 to 800 ms, predict the condition from 800 ms to 1000 ms. The prediction results are evaluated by mean square error and root mean square error.

[0148] In step 21), the key parameters that affect the working condition characteristics of the electro-hydraulic drive-by-wire chassis vehicle are screened out by the correlation analysis in step 1), namely vehicle speed and steering wheel angle. Vehicle speed reflects the longitudinal working condition during vehicle driving; steering wheel angle reflects the lateral working condition during vehicle driving. Vehicle speed and steering wheel angle are used as multivariate time series to predict the longitudinal and lateral coupled working conditions.

[0149] Step 21) is implemented based on a multilayer perceptron, and its specific expression is as follows:

[0150] Where H 0 X is the variable token, X is the input variable, and MLP is a multilayer perceptron.

[0151] In step 22), the multi-head self-attention mechanism consists of multiple self-attention modules in parallel. Each attention module focuses on different parts of the input sequence and calculates different weighted sums, encoding different aspects of features into different attention modules to form multiple sets of encoding modules.

[0152] The multi-head self-attention mechanism used in step 22) obtains the dependency relationship between vehicle speed and steering wheel time series in the cyber-physical traffic environment by calculating the correlation of different variables. The self-attention module generates a vector representation for each input token. The input vector is converted into a query vector, a key vector, and a value vector. The vectors derived from different inputs are packaged into three matrices: Q, K, and V. The Softmax function generates a probability distribution based on the pairwise similarity scores between vectors to obtain the dependency relationship between vehicle speed and steering wheel time series, thereby achieving the coupling of longitudinal and lateral prediction conditions. The process expression is as follows:

[0153]

[0154] H l =LayerNorm(H l-1 +Self-Attn(H l-1 ))

[0155] In the formula, d is the scaling factor, Q, K, and V are three matrices, Attention is the attention mechanism function, Softmax is the normalization exponential function, LayerNorm is the normalization processing function, Self-Attn is the self-attention mechanism function, and H... l The input variable H of the l-th encoding module l-1 This is the input variable for the (l-1)th group of encoding modules.

[0156] In step 23), the normalization layer normalizes along the feature dimension. The normalization layer calculates the mean and standard deviation of all features for each data point in the specific layer, and uses the obtained mean and standard deviation of the features to normalize all features in the specific layer.

[0157] The formula for expressing the normalized layer is:

[0158]

[0159]

[0160]

[0161] Outputt = γ × H norm +β

[0162] Where μ l The mean, Let n be the i-th feature value of the input variable of the l-th encoding module. l Let l represent the number of neuron nodes in the l-th encoding module, where l∈{1,2,...,L} represents the l-th encoding module. Here, H represents the variance, and Output represents the output of the normalization layer. l The input variable H of the l-th encoding module norm The values ​​are normalized values, and γ and β are the learnable scaling and offset parameters, respectively.

[0163] In step 24), the expression for how the feedforward neural network encodes the observed time series through stacked Transformer models and decodes the representation of the predicted sequence using dense nonlinear connections is as follows:

[0164] H l =LayerNorm(H l-1 +Feed-Forward(H l-1 ))

[0165] In the formula, LayerNorm is the normalization function, Feed-Forward is the feedforward neural network layer, and H... l-1 H is the input variable for the (l-1)th group of encoding modules. l The input variables of the l-th encoding module.

[0166] In step 24), the feedforward neural network is a two-layer fully connected network that uses a Gaussian error linear unit (GELU) as the activation function; the activation function is expressed as:

[0167]

[0168] In the formula, GELU is the Gaussian error linear unit activation function, and x is the function input variable.

[0169] In step 24), the prediction result of the encoding module is input to the multilayer perceptron for mapping, and the expression is:

[0170]

[0171] In the formula, For the mapping of the prediction results, H l The input variables of the l-th encoding module.

[0172] In step 25), the prediction results are evaluated using the mean square error (MSE) and root mean square error (RMSE) as expressed by the following formulas:

[0173]

[0174]

[0175] In the formula, MSE is the mean square error, RMSE is the root mean square error, and y i As input variables for calculating errors, This is the mean of the input variables.

[0176] 3) Establish an electro-hydraulic steering optimization allocation model based on the predicted working conditions and optimize the energy of the electro-hydraulic steer-by-wire.

[0177] The specific process of rolling time-domain optimization is as follows: For the k-th sampling time, the rolling time-domain optimization algorithm calculates the optimal control input sequence in the time domain by combining the state information obtained from the working condition prediction, and takes the first item of the sequence as the optimization result at the k-th time; at the k+1-th time, the above rolling optimization process is repeated cyclically using the updated predicted state information obtained from the working condition prediction, allocating the ratio of electric steering and electro-hydraulic steering to achieve energy optimization of electro-hydraulic steer-by-wire; wherein, the length of the rolling sliding window is determined according to the time domain length of 800ms to 1000ms predicted by the working condition.

[0178] The specific steps for optimizing the energy of electro-hydraulic steering by steer-by-wire based on the rolling time domain are as follows:

[0179] 31) Define the ratio of the target electric power assist torque to the steering power assist torque as the electro-hydraulic steering distribution coefficient;

[0180] 32) Define the average loss over the finite time domain as the steering energy cost function;

[0181] 33) Calculate the target speed of the pump drive motor to represent the smoothness of the electro-hydraulic mode switching;

[0182] 34) Define the cost function for electro-hydraulic switching smoothness;

[0183] 35) Establish an electro-hydraulic steering optimization allocation model and solve for optimization.

[0184] The formula for calculating the electro-hydraulic steering distribution coefficient in step 31) is as follows:

[0185]

[0186] In the formula, x e This is the electro-hydraulic steering distribution coefficient. For the target electric assist torque, This is the steering assist torque.

[0187] In step 32), the average loss over the finite time domain is defined as the steering energy cost function as follows:

[0188]

[0189] In the formula, J1 is the steering energy consumption cost function, P EHCS For the energy consumption of electro-hydraulic combined steering, P total Let t be the total energy consumption, t0 be the start time of the sliding window, and t be the total energy consumption. f This is the end time of the scrolling window.

[0190] In step 33), the target speed of the pump drive motor is related to the total required steering torque, vehicle speed, and electro-hydraulic steering distribution coefficient, and its expression is:

[0191]

[0192] In the formula, The target speed of the pump drive motor is given by V, where v is the vehicle speed and T is the speed of the vehicle. s For the total demand steering torque, θ sw x is the motor rotation angle. e Electro-hydraulic steering distribution coefficient.

[0193] In step 33), the smoothness of the electro-hydraulic mode switching is represented by the rate of change of the target speed of the pump drive motor, and the expression is:

[0194]

[0195] In the formula, This represents the rate of change of the target speed of the pump drive motor.

[0196] In step 34), the electro-hydraulic switching smoothness cost function is defined as the time average of the square of the ratio of the target speed change rate of the pump drive motor to its upper limit, and its expression is:

[0197]

[0198] In the formula, J2 is the electro-hydraulic switching smoothness cost function, α maxLet t0 be the rate of change of the upper limit of the target speed of the pump drive motor, and t0 be the start time of the scrolling window. f This is the end time of the scrolling window.

[0199] In step 35), based on the energy consumption and smoothness cost function established in the prediction time domain, the expression for the electro-hydraulic steering optimization allocation model is obtained as follows:

[0200]

[0201] In the formula, w1 and w2 are weighting coefficients; T m1 To assist the motor torque, T m1_max This refers to the upper limit of the assist motor torque; n m1 To increase the motor speed, n m1_max This refers to the upper limit of the motor's speed; P m1 To help reduce motor energy consumption, P m1_max To help the motor's energy consumption limit; T m2 T is the torque of the pump drive motor. m2_max n is the upper limit of the pump drive motor torque. m2 n is the speed of the pump drive motor. m2_max P represents the upper limit of the pump drive motor speed. m2 For the energy consumption of the pump drive motor, P m2_max This represents the upper limit of energy consumption for the pump drive motor.

[0202] In step 35), the algorithm for solving the electro-hydraulic steering energy optimization problem is the Adaptive Particle Swarm Optimization (APSO) algorithm. The specific implementation process is as follows: before the Particle Swarm Optimization (PSO) algorithm updates the motion inertia of the electro-hydraulic steering optimization allocation model formula, an inertia weight coefficient is added. In the early stage of the search when the particles are in the expansion phase, the inertia weight coefficient is increased to improve the global convergence ability. In the later stage of the search, when the particles enter the local mining phase, the inertia weight coefficient is decreased to enhance the local convergence ability and overcome the deficiency of insufficient search ability of the Particle Swarm Optimization algorithm.

[0203] 4) Set optimization conditions to complete the energy prediction optimization of the electro-hydraulic steering-by-wire system;

[0204] The length of the scroll window is set to 1. Based on the rolling time domain optimization predicted by the working condition prediction network, the length of the scroll window is set to the prediction time domain. According to the above settings, the energy prediction optimization of the electro-hydraulic steer-by-wire system is completed, and the optimization results are transmitted to the big data platform. The physical electro-hydraulic steer-by-wire unit completes the electro-hydraulic steering or electric steering of the vehicle according to the optimization results data of the big data platform.

[0205] This invention has many specific applications. The above description is only a preferred embodiment of this invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of this invention, and these improvements should also be considered within the scope of protection of this invention.

Claims

1. An energy prediction and optimization method for an electro-hydraulic steer-by-wire system based on a CPS architecture, the system comprising: Physical electro-hydraulic steering unit, energy prediction and optimization unit, and big data platform; The big data platform communicates with the physical electro-hydraulic steering unit and the energy prediction and optimization unit respectively. The physical electro-hydraulic steer-by-wire unit includes: hydraulic components, electrical components, and mechanical components. The electrical components send control signals to control the mechanical and hydraulic components respectively. The hydraulic components are connected to the mechanical components through hydraulic lines. Based on data from a big data platform, the physical electro-hydraulic steer-by-wire unit controls the mechanical and hydraulic components through the electrical components to achieve electro-hydraulic or electric steering of the vehicle. The energy prediction and optimization unit includes: a vehicle driving data acquisition module, a longitudinal and lateral coupling working condition prediction module, and a steering energy optimization module; The vehicle driving data acquisition module is used to collect vehicle sensor data and store it in a big data platform. The longitudinal and lateral coupling working condition prediction module is used to predict the longitudinal and lateral coupling working conditions of the vehicle by collecting the steering wheel angle and vehicle speed from the vehicle driving data acquisition module. The steering energy optimization module is used to optimize the vehicle's steering energy. The big data platform includes: a data storage repository and a data analysis module; A data repository is used to store data for the physical electro-hydraulic steering unit and the energy prediction and optimization unit; The data analysis module is used to perform smoothing filtering and correlation analysis on the data stored in the data repository to screen out key parameters that affect the operating characteristics of the electro-hydraulic drive-by-wire chassis vehicle. The method is characterized by the following steps: 1) Collect vehicle sensor data and process it through a big data platform; 2) Using the steering wheel angle and vehicle speed from the processed vehicle sensor data as a multivariate time series, and based on the designed working condition prediction network, perform longitudinal and lateral coupled working condition prediction of the vehicle. 3) Establish an electro-hydraulic steering optimization allocation model based on the predicted working conditions and optimize the energy of the electro-hydraulic steer-by-wire. 4) Set optimization conditions to complete the energy prediction optimization of the electro-hydraulic steering-by-wire system; The specific steps for optimizing the electro-hydraulic steering-by-wire energy based on the rolling time domain in step 3) are as follows: 31) Define the ratio of the target electric power assist torque to the steering power assist torque as the electro-hydraulic steering distribution coefficient; 32) Define the average loss over the finite time domain as the steering energy cost function; 33) Calculate the target speed of the pump drive motor to indicate the smoothness of the electro-hydraulic mode switching; 34) Define the cost function for electro-hydraulic switching smoothness; 35) Establish an electro-hydraulic steering optimization allocation model and solve for optimization; In step 35), based on the energy consumption and smoothness cost function established in the prediction time domain, the expression for the electro-hydraulic steering optimization allocation model is obtained as follows: In the formula, w1 and w2 are weighting coefficients; T m1 To assist the motor torque, T m1_max This refers to the upper limit of the assist motor torque; n m1 To increase the motor speed, n m1_max This refers to the upper limit of the motor's speed; P m1 To help reduce motor energy consumption, P m1_max To help the motor's energy consumption limit; T m2 T is the torque of the pump drive motor. m2_max n is the upper limit of the pump drive motor torque. m2 n is the speed of the pump drive motor. m2_max P represents the upper limit of the pump drive motor speed. m2 For the energy consumption of the pump drive motor, P m2_max This represents the upper limit of energy consumption for the pump drive motor.

2. The method according to claim 1, characterized in that, The hydraulic components include: a vane pump, an accumulator, a solenoid valve, and a hydraulic cylinder; the vane pump is connected to the accumulator via a hydraulic line, the solenoid valve is connected to the accumulator via a hydraulic line, and the hydraulic cylinder is connected to the solenoid valve via a hydraulic line. The electrical components include: an electronic control unit (ECU), a pump drive motor, and an auxiliary motor; the ECU controls the working state of the pump drive motor and the auxiliary motor through control signals, the pump drive motor drives the vane pump through a mechanical connection, and the ECU controls the opening and closing degree of the solenoid valve through control signals. The mechanical components include: a steering wheel, a reducer, a steering shaft, and wheels; the steering wheel and reducer are connected by the steering shaft, the steering shaft and wheels are connected by a gear and rack, a power steering motor is mounted on the reducer to provide steering assistance, a hydraulic cylinder is connected to the wheels by a rack, and an electronic control unit (ECU) receives steering signals from the steering wheel through a steering angle sensor.

3. The method according to claim 1, characterized in that, In step 1), the data analysis module uses the moving average method to smooth and filter the vehicle position coordinates, steering wheel angle, driver torque, vehicle speed, acceleration, yaw rate, and steering wheel angular velocity data in the data repository, eliminating outliers and obtaining a dataset that meets the requirements.

4. The method according to claim 3, characterized in that, The formula for calculating the moving average method is as follows: Among them, y t-n For the data before smoothing, y t The data is smoothed, and n is the number of filter terms; The Pearson correlation coefficient was used to perform correlation analysis between the smoothed and filtered data and the steering torque and braking torque. The closer the Pearson correlation coefficient is to 1, the higher the correlation between the data and the steering torque and braking torque. This revealed a high correlation between vehicle speed and steering wheel angle and steering torque and braking torque, which are key parameters affecting the operating characteristics of electro-hydraulic drive-by-wire chassis vehicles. The correlation analysis calculation formula is as follows: Where, ρ X，Y Here, σ is the Pearson correlation coefficient, X and Y are the two variables used in the correlation analysis, cov(X,Y) is the covariance of X and Y, and σ is the Pearson correlation coefficient. X and σ Y Let μ be the standard deviation of X and Y, respectively. X and μ Y Let X and Y be the average values, respectively, and E be the expected value.

5. The method according to claim 1, characterized in that, In step 2), the working condition prediction network is an encoder decoupling prediction network based on the temporal Transformer model, which includes: an embedding layer, a self-attention mechanism, a normalization layer, a feedforward neural network, and a multilayer perceptron. The embedding layer embeds each variable into a variable token according to the method of each variable, and obtains the dependencies between different variables through a multi-head self-attention mechanism. The normalization layer eliminates the distribution differences between the variables input to the feedforward neural network. The feedforward neural network encodes and decodes the multivariate time series. The prediction results after decoding by the feedforward neural network are input into the multilayer perceptron for mapping, and then evaluated by mean square error and root mean square error.

6. The method according to claim 5, characterized in that, The specific steps for predicting operating conditions in step 2) are as follows: 21) By embedding the variable tokens according to the vehicle speed and steering wheel angle through the embedding layer, feature representations of vehicle speed and steering wheel angle are obtained; 22) Employing a multi-head self-attention mechanism, the dependency relationship between vehicle speed and steering wheel angle time series in the information physical traffic environment is obtained by calculating the correlation of different variables; 23) Apply the normalization layer to the feature representation of each variable so that the feature sequences of all variables are under a relatively uniform normal distribution, thereby reducing the impact of differences caused by different variable value ranges and inconsistent measurements; 24) Feedforward neural networks encode the observed time series through stacking Transformer models and decode the representation of the predicted sequence using dense nonlinear connections; 25) Input driving intention, road curvature, and road condition data into the condition prediction network to improve the network's adaptability to longitudinal and lateral coupled conditions. Based on the data from 0 to 800 ms, predict the condition from 800 ms to 1000 ms. The prediction results are evaluated by mean square error and root mean square error.

7. The method according to claim 1, characterized in that, The specific process of rolling time-domain optimization in step 3) is as follows: For the kth sampling time, the rolling time-domain optimization algorithm calculates the optimal control input sequence in the time domain by combining the state information obtained from the working condition prediction, and takes the first item of the sequence as the optimization result at the kth time; at the k+1th time, the updated predicted state information obtained from the working condition prediction is used to repeatedly perform the rolling optimization process, allocate the ratio of electric steering and electro-hydraulic steering, and realize the energy optimization of electro-hydraulic steer-by-wire; wherein, the length of the rolling sliding window is determined according to the time domain length of 800ms to 1000ms predicted by the working condition.

Images