Intelligent network connection vehicle cooperative self-adaptive cruise control method for mixed traffic

Abstract

The invention discloses an intelligent network connection vehicle cooperative adaptive cruise control method for mixed traffic, and relates to the technical field of intelligent network connection automatic driving. The method comprises the following steps: receiving vehicle state data through a vehicle-mounted V2X wireless communication network; constructing a vehicle following model of the manually driven vehicle; and constructing a physical information neural network (PINN), using the physical information neural network (PINN) to predict the state of the manual driving vehicle in the prediction time domain N, inputting a prediction result as a reference trajectory into the model prediction control MPC, and solving an optimal control sequence. According to the method, a manual driving vehicle behavior prediction model is established, an optimization problem with differential equation dynamic constraints is converted into a parameter fitting problem of a neural network, the acceleration prediction precision is improved, meanwhile, the state prediction result of the manual driving vehicle serves as reference input, and the acceleration prediction accuracy is improved. A predictive cooperative adaptive cruise controller PINN-MPC under mixed traffic is designed by adopting a model predictive control method with soft constraint.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent network connection automatic driving, in particular to a cooperative adaptive cruise control method for intelligent network connection vehicles in mixed traffic. BACKGROUND

[0002] With the continuous progress of science and technology, intelligent network connection vehicles (CAVs) have gradually become an important development direction in the field of transportation. They show great potential in improving traffic efficiency, driving safety and environmental sustainability. Intelligent network connection vehicles can realize information exchange between vehicles and perception of road environment through advanced sensors, communication equipment and automatic control systems, thereby optimizing driving decisions and control strategies.

[0003] With the development of short-range communication technology, cooperative adaptive cruise control (CACC) can obtain the surrounding traffic state through vehicle-mounted equipment, realize more stable spacing control and speed control, and thus improve the safety of the traffic system. However, automatic driving technology has not been fully popularized, and society will be in a transition period from pure human-driven vehicles (HDVs) to autonomous vehicles for a long time. In mixed traffic environment, the behavior intention of human-driven vehicles has strong time-varying and randomness, which increases the safety hazards in decision-making and control of intelligent network connection vehicles. Therefore, it is necessary to consider the behavior characteristics of human-driven vehicles in the environment when designing the control strategy of network connection autonomous vehicles to improve vehicle driving safety and driving comfort. However, the current optimization control method such as MPC usually uses a simplified driver model when predicting the behavior of human-driven vehicles, ignoring the complexity of human driving behavior, which limits its adaptability and robustness in mixed traffic environment. Therefore, the present application proposes a cooperative adaptive cruise control method for intelligent network connection vehicles in mixed traffic. SUMMARY

[0004] The purpose of the present application is to provide a cooperative adaptive cruise control method for intelligent network connection vehicles in mixed traffic, which converts the optimization problem with differential equation dynamics constraints into a parameter fitting problem of neural network to improve the prediction accuracy of acceleration.

[0005] According to the first aspect of the present application, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a cooperative adaptive cruise control method for intelligent network connection vehicles in mixed traffic, applied to a mixed traffic following scene involving a lead vehicle, a human-driven vehicle and a network connection autonomous vehicle, characterized in that it comprises the following steps: The network-connected automatic driving vehicle is taken as a control object, vehicle state data is received through a vehicle-mounted V2X wireless communication network, wherein the vehicle state data includes self-vehicle speed , self-vehicle acceleration , front-vehicle speed , front-vehicle acceleration , following-vehicle distance , and relative vehicle speed ; Based on the received vehicle state data, a following model of a manually driven vehicle is constructed, and the following model parameters include a feedback coefficient of a driver to an ideal vehicle speed of the self-vehicle , a feedback coefficient of the driver to the front-vehicle speed ; A physical information neural network (PINN) is constructed, the physical information neural network (PINN) is used to predict the state of the manually driven vehicle in a prediction time domain N, and the prediction result is taken as a reference trajectory input into model predictive control (MPC) to solve an optimal control sequence; wherein the input of the physical information neural network (PINN) is the vehicle state data and the following model parameters, and the output is the prediction result, which is a vehicle longitudinal acceleration; Based on the solved optimal control sequence, a predictive cooperative adaptive cruise control strategy is designed, the first parameter in the optimal control sequence is taken as a current control change, and rolling optimization control is realized.

[0006] Further, based on the received vehicle state data, a following model of a manually driven vehicle is constructed, and the following model is as follows: The following model of the manually driven vehicle is as follows: , (1) wherein and respectively represent a distance between the manually driven vehicle and a leading vehicle and a longitudinal vehicle speed of the manually driven vehicle, and respectively represent characteristic parameters of a driver to ideal speed and relative speed sensitivity; represent an expected speed of the driver, which is related to the distance between the self-vehicle and the front-vehicle , and is determined by a segmented function: (2) wherein and respectively represent minimum and maximum vehicle distances, and the vehicle reaches a maximum ideal speed of the driver when the vehicle distance is greater than .

[0007] Furthermore, a Physical Information Neural Network (PINN) is constructed. PINN is used to predict the state of the manually driven vehicle within the prediction time domain N. The prediction results are then used as a reference trajectory input to the Model Predictive Control (MPC) to solve for the optimal control sequence, as detailed below: Speed ​​of the vehicle Vehicle acceleration Speed ​​of the vehicle in front acceleration of the vehicle in front Following distance and relative speed As input to a physical information neural network: (3) In the formula Represents the observation state matrix, Indicates the time step of historical data. The length of the time series representing the historical trajectory; The output of the physical information neural network is the predicted longitudinal acceleration of the vehicle: (4) In the formula, Indicates predicted acceleration. Indicates the time step of the forecast data. Indicates the predicted time series length.

[0008] Furthermore, the physical information neural network PINN includes an encoder and a decoder, wherein the encoder comprises a fully connected layer and an LSTM network layer, as shown below: (5) In the formula, and Represents the coding features, and Indicates trainable parameters; The decoder consists of an LSTM decoding layer and a fully connected layer. The sequence decoding process is represented as follows: (6) In the formula, This represents the hidden layer output features of an LSTM network. The initial value is the output feature of the encoder at the last time step. and Indicates trainable parameters, This represents the predicted acceleration; This indicates a normalization operation.

[0009] Furthermore, the physical information neural network PINN is trained using a loss function, which is constructed as follows: (51) The constructed loss function includes the loss of the equation of motion. Supervision loss of predictive data Boundary conditions The loss consists of three parts, specifically represented as follows: (7); (52) The PINN network predicts the behavior of manually driven vehicles by predicting acceleration. The differential equation of the longitudinal motion model of the manually driven vehicle is written in a unified form: (8) In the formula, and These represent the distance between the manually driven vehicle and the lead vehicle, and the longitudinal speed of the manually driven vehicle, respectively. and These are characteristic parameters representing the driver's sensitivity to ideal speed and relative speed, respectively. The speed of the lead car is indicated by its logo. Indicates the driver's desired speed; (53) Assuming that the manually driven vehicle can only perceive the current speed of the lead vehicle, that is, the speed of the lead vehicle is constant, in order to satisfy the differential equation of motion, the speed and following distance are further predicted by Euler method based on the predicted acceleration: (9) Substituting the prediction result into the motion differential equation (8), we get the differential equation loss as: (10) In the formula The weights represent the weights of the loss function in the differential equation. and These represent the predicted speed and following distance, respectively. This represents the predicted acceleration. Indicates the discrete time step; (54) The supervision loss of the prediction data is represented by the mean square error between the predicted acceleration and the actual acceleration: (11) In the formula, Indicates the weights of the data supervision loss function; (55) Furthermore, considering the constraint that the speed is non-negative during driving and the safety distance constraint of the vehicle in front, the loss function of the boundary conditions is expressed as: (12) In the formula, Indicates the weights of the boundary condition loss function. Indicates the activation function; and The speed and following distance are predicted based on acceleration, respectively, and the lead vehicle is set to move at a constant speed during this period.

[0010] Furthermore, based on the solved optimal control sequence, a predictive cooperative adaptive cruise control strategy is designed, using the first parameter in the optimal control sequence as the current control change variable to achieve rolling optimization control, as detailed below: (61) The relative distance between vehicles is defined as the difference in position between the vehicle in front and the vehicle behind. A first-order inertial dynamics model is used instead of a complex nonlinear dynamics model. The vehicle dynamics equations are discretized to obtain: (13) in and These represent the speed and acceleration of the manually driven vehicle ahead, respectively. and This indicates the vehicle's speed and acceleration. Indicates the distance between the vehicle and the vehicle in front. Indicates the discrete time step. This indicates a sluggish response from the vehicle. (62) The model is further written as a state-space equation: (14) in In the formula, and These represent the speed and acceleration of the manually driven vehicle ahead, respectively. and This indicates the vehicle's speed and acceleration. Indicates the distance between the vehicle and the vehicle in front. Indicates the discrete time step. This indicates a sluggish response from the vehicle. (63) In the prediction time domain N, the system state variables are obtained through step-by-step iteration: (15); (64) Acceleration of manually driven vehicles and speed All are predicted by the PINN network, and the controlled output is further obtained: (16) Where m represents the control time domain; (65) The objective function of the model predictive controller is as follows: (17) In the formula, and Indicates the weighting coefficient. The system's target output includes: (18) In the formula, Indicates the distance between the target vehicle and the following vehicle. Indicates the target speed. Indicates the target acceleration; (66) To ensure tracking performance, the target speed and target acceleration The target following distance is designed using the predicted speed and acceleration of manually driven vehicles, respectively, and a fixed-time-distance following strategy. : (19) Design the model predictive controller in an incremental form, let: (20) Substituting (18) into (14) yields: (twenty one) Substituting (19) into (15) and omitting the constant term, we obtain the objective function of the standard quadratic form as follows: (twenty two) in: In the formula, For control sequences; (67) Constraints are imposed on some outputs and control variables. To prevent the MPC from encountering a situation where no feasible solution is found, a relaxation factor is introduced to extend the conditions, thereby increasing the feasible region of the solution. The extended constraint conditions are expressed as follows: (twenty three) In the formula, and These represent the upper and lower bounds of the control quantity, respectively. and These represent the upper and lower bounds of the control variable increment, respectively. and These represent the upper and lower bounds of the system output, respectively. , , , , , These represent the upper and lower bounds of the corresponding parameters, respectively, representing the relaxation amounts. , , Indicates the relaxation factor; Combining formulas (19) and (20), the model predictive control is transformed into a standard quadratic programming problem.

[0011] (68) Solve the standard quadratic programming problem to obtain the optimal control sequence. The first value of the optimal control sequence is used for the control of the current time step. Repeating steps (61) to (67) will achieve rolling control.

[0012] According to a second aspect of the present invention, the present invention provides a cooperative adaptive cruise control system for intelligent connected vehicles in mixed traffic, for implementing a cooperative adaptive cruise control method for intelligent connected vehicles in mixed traffic as described in Embodiment 1, comprising: The receiving module is used to receive vehicle status data via an onboard V2X wireless communication network, with the connected autonomous vehicle as the controlled object. This vehicle status data includes the vehicle's speed. Vehicle acceleration Speed ​​of the vehicle in front acceleration of the vehicle in front Following distance and relative speed ; The model building module is used to construct a following model of the manually driven vehicle based on the received vehicle state data. The parameters of the following model include the driver's feedback coefficient for the ideal speed of the vehicle. Driver's feedback coefficient to the speed of the vehicle in front ; The prediction module is used to construct a physical information neural network (PINN). The PINN is used to predict the state of the manually driven vehicle in the prediction time domain N. The prediction results are used as reference trajectories to input the model predictive control (MPC) to solve for the optimal control sequence. The input to the Physical Information Neural Network (PINN) is vehicle state data and following model parameters, and the output is the predicted longitudinal acceleration of the vehicle. The calculation output module is used to design a predictive cooperative adaptive cruise control strategy based on the solved optimal control sequence. It uses the first parameter in the optimal control sequence as the current control change to achieve rolling optimization control.

[0013] The present invention has at least the following beneficial effects: 1. This invention combines a long short-term memory network with an optimal velocity model to construct a behavior prediction model, transforming the dynamic optimization problem with differential equation constraints into a parameter fitting problem of a neural network, thereby significantly improving the prediction accuracy of acceleration. On the HighD dataset, with a prediction time domain of 3 seconds, the prediction accuracy is improved by 28.9%.

[0014] 2. Compared with the ordinary MPC algorithm, the proposed predictive cooperative adaptive cruise control strategy increases the minimum collision time by 7.6% and reduces the average acceleration change rate by 19.1% in the scenario of continuous change of vehicle speed in LV, effectively improving vehicle safety and driving comfort, and showing good application potential in mixed traffic scenarios.

[0015] 3. The control method of the present invention can adapt to the complexity and diversity of human-driven vehicle behavior in mixed traffic environments. By introducing a physical information-based neural network (PINN), the model can better capture the dynamic characteristics of human driving behavior, thereby maintaining good predictive performance and control effect in different traffic scenarios. Even when faced with sudden behavior of human-driven vehicles, it can quickly make reasonable responses and demonstrate good robustness.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the control method described in this invention; Figure 2 This is a schematic diagram illustrating the framework principle of the basic vehicle-following scenario in this invention; Figure 3 This is a comparison chart of test set losses under the control method of this invention; Figure 4 An acceleration rolling prediction diagram using the control method of this application; Figure 5 The following diagram shows the simulation results of a scenario where the vehicle speed changes continuously under the control method of this application. (a) represents the vehicle speed curve; (b) represents the relative speed curve; (c) represents the following distance curve; (d) represents the following distance error curve; (e) represents the acceleration curve; and (f) represents the acceleration rate curve. Detailed Implementation

[0018] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0019] Example 1: The specific terms used in this embodiment are explained as follows: "Autonomous vehicle" refers to the vehicle actually being controlled. In this embodiment, the connected autonomous vehicle (Ego) is the controlled object.

[0020] The vehicle in front: During the course of travel, the vehicle located in front of the vehicle under actual control.

[0021] Human-driven vehicles (HDVs): Traditional vehicles driven by human drivers in traffic scenarios.

[0022] Lead Vehicle (LV): In a traffic scenario, the vehicle selected by the driver as the target to follow.

[0023] Please see Figures 1-5 This invention provides a technical solution: a cooperative adaptive cruise control method for intelligent connected vehicles in mixed traffic, comprising the following steps: S1: In a basic vehicle-following scenario involving a lead vehicle (LV), a manually driven vehicle (HDV), and a connected autonomous vehicle (Ego), the connected autonomous vehicle is the controlled object, and its speed is obtained through the onboard V2X wireless communication network. Vehicle acceleration Speed ​​of the vehicle in front acceleration of the vehicle in front Following distance and relative speed ; S2: Establish a vehicle-following model for manually driven HDV vehicles. The parameters of the vehicle-following model include the driver's feedback coefficient for their ideal vehicle speed. Driver's feedback coefficient to the speed of the vehicle in front ; S21. The following model for manually driven vehicles is: (1) in and These represent the distance between the manually driven vehicle and the lead vehicle, and the longitudinal speed of the manually driven vehicle, respectively. and These are characteristic parameters representing the driver's sensitivity to ideal speed and relative speed, respectively. This indicates the driver's desired speed, which is relative to the distance between the vehicle and the vehicle in front. This is related to a piecewise function: (2) In the formula, and These represent the minimum and maximum vehicle spacing, respectively. When the vehicle spacing is greater than... When the vehicle reaches the driver's maximum ideal speed ; S3: Construct a physical information neural network PINN, use the physical information neural network PINN to predict the state of the manually driven vehicle in the prediction time domain N, and use the prediction result as a reference trajectory input to the model predictive control MPC to solve for the optimal control sequence. S31. Adjust vehicle speed Vehicle acceleration Speed ​​of the vehicle in front acceleration of the vehicle in front Following distance and relative speed As network input: (3) In the formula Represents the observation state matrix, Indicates the time step of historical data. The length of the time series representing the historical trajectory; The network output is the predicted longitudinal acceleration of the vehicle: (4) In the formula, Indicates predicted acceleration. Indicates the time step of the forecast data. Indicates the predicted time series length; The data-driven part of S32.PINN is based on a sequence-to-sequence long short-term memory neural network (Seq2SeqLSTM) design, such as... Figure 2 As shown, the encoder of the network contains a fully connected layer (FC) and an LSTM network layer, represented as follows: (5) In the formula, and Represents the coding features, and Indicates trainable parameters; The decoding network contains an LSTM decoding layer and a fully connected layer. The sequence decoding process can be represented as follows: (6) In the formula, This represents the hidden layer output features of an LSTM network. The initial value is the output feature of the encoder at the last time step. and Indicates trainable parameters, This represents the predicted acceleration; This represents a normalization operation, the purpose of which is to improve the stability and expressive power of network training; S33. In step S3, the constructed loss function includes the loss of the motion differential equation. Supervision loss of predictive data Boundary conditions The loss consists of three parts, represented as follows: (7) During the longitudinal following process of a vehicle, the longitudinal acceleration of the vehicle is an important parameter reflecting the driving style and intention. Therefore, the PINN network can predict the behavior of the HDV by predicting the acceleration. As shown in formula (1), the differential equation of the longitudinal motion model of the HDV can be written as a unified expression: (8) In the formula, and These represent the distance between the manually driven vehicle and the lead vehicle, and the longitudinal speed of the manually driven vehicle, respectively. and These are characteristic parameters representing the driver's sensitivity to ideal speed and relative speed, respectively. The speed of the lead car is indicated by its logo. Indicates the driver's desired speed; Assuming the manually driven vehicle can only perceive the current speed of the lead vehicle, i.e., the speed of the lead vehicle (LV) is constant, to satisfy the equation of motion, based on the predicted acceleration, the Euler method is used to further predict the speed and following distance: (9) Substituting the prediction result into the motion differential equation (8), we can obtain the differential equation loss as: (10) In the formula The weights represent the weights of the loss function in the differential equation. and These represent the predicted speed and following distance, respectively. This represents the predicted acceleration. The discrete time step is represented; the supervision loss of the prediction data is represented by the mean square error between the predicted acceleration and the actual acceleration. (11) In the formula, Indicates the weights of the data supervision loss function; Furthermore, the prediction results need to meet certain boundary conditions. Considering the constraint that the speed is non-negative during driving and the constraint of the safe distance from the vehicle in front, the loss function under the boundary conditions is expressed as: (12) In the formula, Indicates the weights of the boundary condition loss function. This represents the activation function, ensuring that the loss function only takes effect when the state violates the boundary conditions; and These are the speed and following distance predicted based on acceleration, respectively, during which the preceding vehicle (LV) is assumed to be moving at a constant speed. S4: Based on the solved optimal control sequence, a predictive cooperative adaptive cruise control strategy is designed, using the first parameter in the optimal control sequence as the current control change variable to achieve rolling optimization control; S41: The relative distance between vehicles is defined as the difference in position between the preceding and following vehicles. A first-order inertial dynamics model is used instead of a complex nonlinear dynamics model. The vehicle dynamics equations are discretized to obtain: (13) in and These represent the speed and acceleration of the manually driven vehicle ahead, respectively. and This indicates the vehicle's speed and acceleration. Indicates the distance between the vehicle and the vehicle in front. Indicates the discrete time step. This indicates a sluggish response from the vehicle. S42: The model is further written as a state-space equation: (14) in: In the formula, and These represent the speed and acceleration of the manually driven vehicle ahead, respectively. and This indicates the vehicle's speed and acceleration. Indicates the distance between the vehicle and the vehicle in front. Indicates the discrete time step. This indicates a sluggish response from the vehicle. S43: Within the prediction time domain N, the system state variables can be obtained through step-by-step iteration: (15) S44: HDV vehicle acceleration and speed All are predicted by PINN, and the controlled output can be further obtained: (16) Where m represents the control time domain; S45: The objective function for the model predictive controller is designed as follows: (17) In the formula, and Indicates the weighting coefficient. The system's target output includes: (18) S46: To ensure tracking performance, the target speed... and target acceleration The target following distance is designed using the predicted speed and acceleration of HDV, respectively, and a fixed-time-distance following strategy. : (19) To reduce static error, the model predictive controller is designed as an incremental form, let: (20) Substituting (18) into (14) yields: (twenty one) Substituting (19) into (15) and omitting the constant term, we can obtain the objective function of the standard quadratic form as follows: (twenty two) in In the formula, ∆U For control sequences; S47: Considering safety and comfort, constraints need to be imposed on some outputs and control variables. To prevent the MPC from encountering infeasible solutions, a relaxation factor is introduced to extend the conditions, thereby increasing the feasible region of the solution. The extended constraints are expressed as follows: (twenty three) In the formula, and These represent the upper and lower bounds of the control quantity, respectively. and These represent the upper and lower bounds of the control variable increment, respectively. and These represent the upper and lower bounds of the system output, respectively. , , , , , These represent the upper and lower bounds of the corresponding parameters, respectively, representing the relaxation amounts. , , Indicates the relaxation factor; Combining (19) and (20), model predictive control can be transformed into a standard quadratic programming problem.

[0024] In MATLAB, the quadprog toolbox can be used to solve standard quadratic programming problems and obtain optimal control sequences. Use the first value of the sequence to control the current time step; repeat the above steps to achieve rolling control.

[0025] The technical solution of the present invention will be further described below with reference to specific embodiments: To verify the effectiveness of the Physical Information Neural Network (PINN) prediction model proposed in this embodiment, experimental evaluation was conducted on the HighD dataset. The pure data-driven method Seq2Seq LSTM, which does not contain physical information, was selected as the baseline model. This baseline model consists of an LSTM encoder, an LSTM decoder, and two fully connected layers. Its network structure and parameter scale are consistent with the PINN model to ensure fair comparison. The average Euclidean error (ADE) was used as the evaluation metric. (25) In the formula, B represents the number of samples, and N represents the prediction time domain.

[0026] Figure 3 The graph shows the loss distribution of the model on the test set. As can be seen, the test error of Seq2Seq LSTM fluctuates around 0.06, while the test loss of PINN is generally below 0.05, with its peak and mean values ​​significantly lower than those of Seq2Seq LSTM.

[0027] Meanwhile, this application also verifies the predictive model control algorithm based on the Matlab / Simulink platform. The scenario is set up as shown in Figure 2, where the vehicle's LV and HDV data are derived from a 30-second following segment in the HighD dataset. Simulation scenarios include a scenario with continuously changing LV vehicle speed and a scenario with LV cut-in. For a more complete driving condition, PINN is used for rolling time-domain prediction, and the speed is predicted synchronously based on the predicted acceleration. The acceleration prediction results obtained in the scenario with continuously changing LV vehicle speed are shown below. Figure 4 As shown.

[0028] The general MPC algorithm and the CACC algorithm are selected as comparison algorithms. The MPC algorithm uses a constant acceleration prediction method for HDV state prediction. The average relative velocity is used. Mean following distance error Minimum collision time Maximum acceleration and the mean rate of change of acceleration To compare the performance of various algorithms in terms of safety and user experience, the metrics are defined as follows: (26) Simulation results for the scenario of continuously changing vehicle speed (LV) are as follows: Figure 5 As shown, PINN-MPC effectively predicts the acceleration and deceleration states of the HDV, enabling it to respond to speed changes faster than MPC and CACC, thereby reducing speed fluctuations and improving following accuracy. Compared to MPC, PINN-MPC reduces the average following distance by 13.5%, but increases the minimum collision time by 7.6%, effectively improving the safety of CAV following. The maximum acceleration is reduced by 18.2% and 27.7% compared to CACC and MPC, respectively, effectively improving comfort during following.

[0029] The average single-step solution time of the PINN-MPC control algorithm is 0.89 ms, and the maximum solution time is 1.79 ms, which is less than the cycle of a typical vehicle control system and meets the real-time requirements.

[0030] In summary, PINN-MPC, while accurately predicting HDV driving behavior, achieves more precise vehicle speed tracking, reduces collision risk, and minimizes acceleration fluctuations, thereby improving driving safety and comfort. Comparing the performance of the MPC algorithm reveals that inaccurate acceleration predictions can mislead the MPC controller's desired path, leading to decreased control performance—even worse than the CACC control method based on feedforward and feedback control. Therefore, the PINN architecture proposed in this invention, which combines physics and data-driven prediction, not only improves trajectory prediction accuracy but also enhances the MPC controller's adaptability to human driving behavior.

[0031] Example 2: This embodiment provides an intelligent connected vehicle cooperative adaptive cruise control system for mixed traffic, used to implement the intelligent connected vehicle cooperative adaptive cruise control method for mixed traffic described in Embodiment 1, including: The receiving module is used to receive vehicle status data via an onboard V2X wireless communication network, with the connected autonomous vehicle as the controlled object. This vehicle status data includes the vehicle's speed v. i Vehicle acceleration a i Speed ​​of the vehicle in front, v LV,t The acceleration of the vehicle in front, a LV,t Following distance dt and relative speed ∆v t ; The model building module is used to build a following model of the manually driven vehicle based on the received vehicle status data. The following model parameters include the driver's feedback coefficient α for the ideal speed of the vehicle and the driver's feedback coefficient β for the speed of the vehicle in front. The prediction module is used to construct a physical information neural network (PINN). The PINN is used to predict the state of the manually driven vehicle in the prediction time domain N. The prediction results are used as reference trajectories to input the model predictive control (MPC) to solve for the optimal control sequence. The input to the Physical Information Neural Network (PINN) is vehicle state data and following model parameters, and the output is the predicted longitudinal acceleration of the vehicle. The calculation output module is used to design a predictive cooperative adaptive cruise control strategy based on the solved optimal control sequence. It uses the first parameter in the optimal control sequence as the current control change to achieve rolling optimization control.

[0032] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0033] For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances. When an element is referred to as being "assembled on," "mounted on," "fixed to," or "set on" another element, it may be directly on the other element or there may be an intermediate element present. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element present. The terms "vertical," "horizontal," "upper," "lower," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible embodiments.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0035] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A cooperative adaptive cruise control method for intelligent connected vehicles in mixed traffic scenarios, applied to a mixed traffic following scenario involving a lead vehicle, a manually driven vehicle, and a connected autonomous vehicle, characterized in that... Includes the following steps: Using connected autonomous vehicles as the controlled object, vehicle status data is received through the onboard V2X wireless communication network, including the vehicle's speed. Vehicle acceleration Speed ​​of the vehicle in front acceleration of the vehicle in front Following distance and relative speed ; Based on the received vehicle status data, a following model for the manually driven vehicle is constructed. The parameters of the following model include the driver's feedback coefficient for the ideal speed of the vehicle. Driver's feedback coefficient to the speed of the vehicle in front ; A physical information neural network (PINN) is constructed. The PINN is used to predict the state of the manually driven vehicle in the prediction time domain N. The prediction results are used as reference trajectories to input the model predictive control (MPC) to solve for the optimal control sequence. The input to the Physical Information Neural Network (PINN) is vehicle state data and following model parameters, and the output is the predicted longitudinal acceleration of the vehicle. Based on the solved optimal control sequence, a predictive cooperative adaptive cruise control strategy is designed, which uses the first parameter in the optimal control sequence as the current control change variable to achieve rolling optimization control.

2. The intelligent connected vehicle cooperative adaptive cruise control method for mixed traffic according to claim 1, characterized in that: Based on the received vehicle status data, a following model for manually driven vehicles is constructed, as follows: The following model for manually driven vehicles is as follows: ，（1） in and These represent the distance between the manually driven vehicle and the lead vehicle, and the longitudinal speed of the manually driven vehicle, respectively. and These are characteristic parameters representing the driver's sensitivity to ideal speed and relative speed, respectively. This indicates the driver's desired speed, which is relative to the distance between the vehicle and the vehicle in front. This is related to a piecewise function: （2） In the formula, and These represent the minimum and maximum vehicle spacing, respectively. When the vehicle spacing is greater than... When the vehicle reaches the driver's maximum ideal speed .

3. The intelligent connected vehicle cooperative adaptive cruise control method for mixed traffic according to claim 2, characterized in that: A Physical Information Neural Network (PINN) is constructed to predict the state of the manually driven vehicle within the prediction time domain N. The prediction results are then used as a reference trajectory input to the Model Predictive Control (MPC) to solve for the optimal control sequence, as detailed below: Speed ​​of the vehicle Vehicle acceleration Speed ​​of the vehicle in front acceleration of the vehicle in front Following distance and relative speed As input to a physical information neural network: （3） In the formula Represents the observation state matrix, Indicates the time step of historical data. The length of the time series representing the historical trajectory; The output of the physical information neural network is the predicted longitudinal acceleration of the vehicle: （4） In the formula, Indicates predicted acceleration. Indicates the time step of the forecast data. Indicates the predicted time series length.

4. The intelligent connected vehicle cooperative adaptive cruise control method for mixed traffic according to claim 3, characterized in that: The Physical Information Neural Network (PINN) includes an encoder and a decoder, wherein the encoder comprises a fully connected layer and an LSTM network layer, as shown below: （5） In the formula, and Represents the coding features, and Indicates trainable parameters; The decoder consists of an LSTM decoding layer and a fully connected layer. The sequence decoding process is represented as follows: （6） In the formula, This represents the hidden layer output features of an LSTM network. The initial value is the output feature of the encoder at the last time step. and Indicates trainable parameters, This represents the predicted acceleration; This indicates a normalization operation.

5. The intelligent connected vehicle cooperative adaptive cruise control method for mixed traffic according to claim 4, characterized in that: The physical information neural network PINN is trained using a loss function, which is constructed as follows: (51) The constructed loss function includes the loss of the equation of motion. Supervision loss of predictive data Boundary conditions The loss consists of three parts, specifically represented as follows: （7）； (52) The PINN network predicts the behavior of manually driven vehicles by predicting acceleration. The differential equation of the longitudinal motion model of the manually driven vehicle is written in a unified form: （8） In the formula, and These represent the distance between the manually driven vehicle and the lead vehicle, and the longitudinal speed of the manually driven vehicle, respectively. and These are characteristic parameters representing the driver's sensitivity to ideal speed and relative speed, respectively. The speed of the lead car is indicated by its logo. Indicates the driver's desired speed; (53) Assuming that the manually driven vehicle can only perceive the current speed of the lead vehicle, that is, the speed of the lead vehicle is constant, in order to satisfy the differential equation of motion, the speed and following distance are further predicted by Euler method based on the predicted acceleration: （9） Substituting the prediction result into the motion differential equation (8), we get the differential equation loss as: （10） In the formula The weights represent the weights of the loss function in the differential equation. and These represent the predicted speed and following distance, respectively. This represents the predicted acceleration. Indicates the discrete time step; (54) The supervision loss of the prediction data is represented by the mean square error between the predicted acceleration and the actual acceleration: （11） In the formula, Indicates the weights of the data supervision loss function; (55) Furthermore, considering the constraint that the speed is non-negative during driving and the safety distance constraint of the vehicle in front, the loss function of the boundary conditions is expressed as: （12） In the formula, Indicates the weights of the boundary condition loss function. Indicates the activation function; and The speed and following distance are predicted based on acceleration, respectively, and the lead vehicle is set to move at a constant speed during this period.

6. The intelligent connected vehicle cooperative adaptive cruise control method for mixed traffic according to claim 5, characterized in that: Based on the solved optimal control sequence, a predictive cooperative adaptive cruise control strategy is designed, using the first parameter in the optimal control sequence as the current control change variable to achieve rolling optimization control, as detailed below: (61) The relative distance between vehicles is defined as the difference in position between the vehicle in front and the vehicle behind. A first-order inertial dynamics model is used instead of a complex nonlinear dynamics model. The vehicle dynamics equations are discretized to obtain: （13） in and These represent the speed and acceleration of the manually driven vehicle ahead, respectively. and This indicates the vehicle's speed and acceleration. Indicates the distance between the vehicle and the vehicle in front. Indicates the discrete time step. This indicates a sluggish response from the vehicle. (62) The model is further written as a state-space equation: （14） in In the formula, and These represent the speed and acceleration of the manually driven vehicle ahead, respectively. and This indicates the vehicle's speed and acceleration. Indicates the distance between the vehicle and the vehicle in front. Indicates the discrete time step. This indicates a sluggish response from the vehicle. (63) In the prediction time domain N, the system state variables are obtained through step-by-step iteration: （15）； (64) Acceleration of manually driven vehicles and speed All are predicted by the PINN network, and the controlled output is further obtained: （16） Where m represents the control time domain; (65) The objective function of the model predictive controller is as follows: （17） In the formula, and Indicates the weighting coefficient. The system's target output includes: （18） In the formula, Indicates the distance between the target vehicle and the following vehicle. Indicates the target speed. Indicates the target acceleration; (66) To ensure tracking performance, the target speed and target acceleration The target following distance is designed using the predicted speed and acceleration of manually driven vehicles, respectively, and a fixed-time-distance following strategy. : (19) Design the model predictive controller in an incremental form, let: （20） Substituting (18) into (14) yields: （21） Substituting (19) into (15) and omitting the constant term, we obtain the objective function of the standard quadratic form as follows: （22） in: In the formula, For control sequences; (67) Constraints are imposed on some outputs and control variables. To prevent the MPC from encountering a situation where no feasible solution is found, a relaxation factor is introduced to extend the conditions, thereby increasing the feasible region of the solution. The extended constraint conditions are expressed as follows: （23） In the formula, and These represent the upper and lower bounds of the control quantity, respectively. and These represent the upper and lower bounds of the control variable increment, respectively. and These represent the upper and lower bounds of the system output, respectively. , , , , , These represent the upper and lower bounds of the corresponding parameters, respectively, representing the relaxation amounts. , , Indicates the relaxation factor; Combining formulas (19) and (20), the model predictive control is transformed into a standard quadratic programming problem. (68) Solve the standard quadratic programming problem to obtain the optimal control sequence. The first value of the optimal control sequence is used for the control of the current time step. Repeating steps (61) to (67) will achieve rolling control.

7. A cooperative adaptive cruise control system for intelligent connected vehicles in mixed traffic, used to implement the cooperative adaptive cruise control method for intelligent connected vehicles in mixed traffic as described in any one of claims 1 to 6, characterized in that, include: The receiving module is used to receive vehicle status data via an onboard V2X wireless communication network, with the connected autonomous vehicle as the controlled object. This vehicle status data includes the vehicle's speed. Vehicle acceleration Speed ​​of the vehicle in front acceleration of the vehicle in front Following distance and relative speed ; The model building module is used to construct a following model of the manually driven vehicle based on the received vehicle state data. The parameters of the following model include the driver's feedback coefficient for the ideal speed of the vehicle. Driver's feedback coefficient to the speed of the vehicle in front ; The prediction module is used to construct a physical information neural network (PINN). The PINN is used to predict the state of the manually driven vehicle in the prediction time domain N. The prediction results are used as reference trajectories to input the model predictive control (MPC) to solve for the optimal control sequence. The input to the Physical Information Neural Network (PINN) is vehicle state data and following model parameters, and the output is the predicted longitudinal acceleration of the vehicle. The calculation output module is used to design a predictive cooperative adaptive cruise control strategy based on the solved optimal control sequence. It uses the first parameter in the optimal control sequence as the current control change to achieve rolling optimization control.

Images