A cloud-based intelligent connected vehicle control system and method

Abstract

The application provides a cloud-based intelligent networked vehicle control system and method, which comprises an edge cloud layer, a vehicle end control layer and a vehicle end execution layer; the edge cloud layer collects all road test sensor information and state information of all vehicles in a control area, comprehensively decides and plans an optimal path, and respectively performs lateral control amount fusion and longitudinal control amount fusion; the vehicle end control layer combines vehicle state information to perform real-time vehicle stability judgment, and when vehicle instability occurs, a multi-agent coordinated controller is used to distribute additional yaw moment to an active steering module, a differential braking module and a differential driving module; when control information of the edge cloud layer is not transmitted to the vehicle end in time, the vehicle end calls a vehicle end lateral L-MPC controller and a vehicle end longitudinal double PID controller to perform tracking control; and the vehicle end execution layer controls the work of each actuator to realize execution of the control amount. The application improves the precision and stability of vehicle control, and simultaneously realizes control redundancy of the vehicle.

Description

Technical Field

[0001] This invention belongs to the field of vehicle-road-cloud collaborative control technology, specifically relating to a cloud-based intelligent connected vehicle control system and method. Background Technology

[0002] As my country's intelligent vehicle development enters a critical stage, the long-tail problems of perception, decision-making, and planning, as well as the computing power limitations of onboard computing platforms, are becoming increasingly prominent in traditional single-vehicle intelligence-based development solutions. Therefore, cloud-controlled intelligent connected vehicles based on cloud computing and network interconnection are gradually emerging. Cloud control platforms possess high computing power and comprehensive information perception capabilities, providing more information and greater computing power support for vehicle perception, decision-making, planning, and motion control. Therefore, it is necessary to study cloud-based intelligent connected vehicle control architectures and related control methods.

[0003] In the development of intelligent vehicles, single-vehicle intelligent solutions equip vehicles with numerous expensive sensors and design a large number of complex perception fusion and filtering algorithms to obtain as much comprehensive and accurate driving environment information as possible, significantly increasing the cost per vehicle. Due to the limitations of the computing power of the onboard computing platform, single-vehicle intelligent control algorithms cannot use high-precision nonlinear vehicle dynamics models to calculate control quantities. Moreover, due to the complex and uncertain driving environment, the onboard computing platform may operate in harsh environments, making it difficult to guarantee its safety and stability. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a cloud-based intelligent connected vehicle control system and method, which improves the accuracy and stability of vehicle control.

[0005] Note that the description of these objectives does not preclude the existence of other objectives. One aspect of the invention does not require achieving all of the above objectives. Objectives other than those described above can be extracted from the description, drawings, and claims.

[0006] This invention improves the control accuracy of vehicles by integrating lateral and longitudinal controllers in the cloud. The vehicle-side design incorporates a multi-agent coordinated controller to coordinate the control of yaw torque, thereby improving the stability of the vehicle under adverse driving conditions. It is also equipped with lateral and longitudinal controllers based on a linear vehicle dynamics model, which enables control redundancy when cloud control quantities cannot be accurately transmitted to the vehicle.

[0007] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0008] A cloud-based intelligent connected vehicle control system includes an edge-cloud layer, a vehicle-side control layer, and a vehicle-side execution layer;

[0009] The edge cloud layer is used to collect information from all roadside sensors and the status information of all vehicles within the control area. Based on the collected information, it makes a comprehensive decision to plan an optimal path and calls the lateral controller and longitudinal controller respectively to perform path tracking, perform lateral control quantity fusion and longitudinal control quantity fusion respectively, and transmit them to the vehicle control layer.

[0010] The vehicle-side control layer is used to receive control commands transmitted from the edge cloud layer and perform lower-level tracking control. At the same time, it combines vehicle status information to make real-time vehicle stability judgments. When vehicle instability occurs, a multi-agent coordination controller is used to distribute additional yaw moment to the active steering module, differential braking module, and differential drive module. When the control information from the edge cloud layer is not transmitted to the vehicle end in time, the vehicle end calls the vehicle-side lateral L-MPC controller and the vehicle-side longitudinal dual PID controller for tracking control.

[0011] The vehicle-side execution layer is used to receive control information transmitted from the vehicle-side control layer and control each actuator to execute the control quantity.

[0012] In the above scheme, the edge cloud layer includes a perception filtering fusion module, a state receiving and estimation module, a decision planning module, an edge horizontal N-MPC controller, an edge horizontal LSTM controller, a horizontal control quantity fusion module, an edge vertical nonlinear controller, an edge vertical LSTM controller, and a vertical control quantity fusion module;

[0013] The perception filtering and fusion module is used to collect information from all roadside sensors within the control area, then filter out noise from the collected information, and fuse multi-source information. The information includes the location, shape, speed information and traffic signals of all traffic participants in the area who cannot actively communicate with the cloud control center.

[0014] The state receiving and estimation module is used to collect state information of all vehicles in the area, including vehicle-level information, system-level information, and actuator-level information. The vehicle-level information includes vehicle speed, vehicle acceleration, yaw angle, pitch angle, roll angle, and center of gravity sideslip angle. The system-level information includes steering system state information, braking system state information, and drive system state information. The actuator-level information includes steering motor state information, brake motor state information, and drive motor state information.

[0015] The decision planning module is used to combine all roadside sensor information and all received vehicle status information within the control area, and to make comprehensive decision planning based on vehicle economy, traffic efficiency and vehicle safety as performance indicators, and output a planned path with speed and direction characteristics.

[0016] The edge lateral N-MPC controller is based on a nonlinear vehicle lateral dynamics model and uses a nonlinear MPC method for calculation to solve for the front wheel steering angle control quantity.

[0017] The edge lateral LSTM controller is a data-driven lateral controller trained with a long short-term memory network. It takes vehicle-level information, steering system-level information, and steering motor actuator-level information during vehicle operation as inputs to the LSTM network and front wheel steering angle control as outputs to the LSTM network.

[0018] The lateral control quantity fusion module is used to fuse the control quantity of the edge lateral N-MPC controller and the control quantity of the edge lateral LSTM controller. Based on fuzzy control, the control quantity weight coefficients of the two controllers are solved. Then, the control quantity of the edge lateral N-MPC controller and the control quantity of the edge lateral LSTM controller are calculated respectively through the weight coefficients. Finally, the two are fused to obtain the final lateral control quantity output.

[0019] The edge longitudinal nonlinear controller is based on a nonlinear vehicle longitudinal dynamics model, which includes the nonlinear longitudinal slip and sideslip characteristics of the tire, and solves for the longitudinal control quantity.

[0020] The edge longitudinal LSTM controller is a data-driven longitudinal controller trained with a long short-term memory network. It takes vehicle-level information, braking system and drive system-level information, and brake motor and drive motor actuator-level information during vehicle operation as inputs to the LSTM network, and takes vehicle speed and acceleration control quantities as outputs of the LSTM network.

[0021] The longitudinal control quantity fusion module fuses the control quantity of the edge longitudinal nonlinear controller and the control quantity of the edge longitudinal LSTM controller. Based on fuzzy control, the control quantity weight coefficients of the two controllers are solved. Then, the control quantity of the edge longitudinal nonlinear controller and the control quantity of the edge longitudinal LSTM controller are calculated separately using the weight coefficients. Finally, the two are fused to obtain the final longitudinal control quantity output.

[0022] Furthermore, in the fused lateral control quantities, the control quantity of the edge lateral N-MPC controller is represented as u by the lateral control quantity fusion module. y1 (t), the weight is represented by k y The control quantity of the edge lateral LSTM controller is represented as u y2 (t), the weight is represented as 1-k y Let t be time. Based on fuzzy control, the control input weight coefficients of the two controllers are obtained:

[0023] u y (t)=k y uy1 (t)+(1-k y )u y2 (t).

[0024] Furthermore, in the fused longitudinal control quantities, the control quantity of the edge longitudinal nonlinear controller is represented as u by the longitudinal control quantity fusion module. x1 (t), the weight is represented by k x The control quantity of the edge longitudinal LSTM controller is represented as u x2 (t), the weight is represented as 1-k x Let t be time. Based on fuzzy control, the control input weight coefficients of the two controllers are obtained:

[0025] u x (t)=k x u x1 (t)+(1-k x )u x2 (t).

[0026] In the above scheme, the vehicle-side control layer includes a stability judgment module, a multi-agent coordination controller, an active steering module, a differential braking module, a differential drive module, a vehicle-side lateral L-MPC controller, and a vehicle-side longitudinal dual PID controller.

[0027] The stability judgment module makes stability judgments based on the real-time status information of the vehicle. When the vehicle is in a stable state, it directly transmits the control information received from the edge cloud layer to the vehicle execution layer. When the vehicle is in an unstable state, it calls the multi-agent coordination controller to distribute additional yaw torque.

[0028] The multi-agent coordination controller is used to first calculate the total additional yaw moment required to maintain vehicle stability, and then distribute the total additional yaw moment to the active steering module, differential braking module and differential drive module according to the vehicle operating conditions.

[0029] The active steering module is used to receive the allocated additional yaw moment and then add an additional front wheel angle on top of the front wheel angle that maintains normal steering to achieve the additional yaw moment.

[0030] The differential braking module is used to receive the allocated additional yaw moment and then apply different braking torques to the left and right wheels to achieve the additional yaw moment.

[0031] The differential drive module is used to receive the allocated additional yaw torque and then apply different driving torques to the left and right wheels to achieve the additional yaw torque.

[0032] The vehicle-side lateral L-MPC controller is based on a linear vehicle lateral dynamics model. When the control system is working normally, the control quantity calculated by the vehicle-side lateral L-MPC controller is not executed. When the control quantity information of the edge cloud layer is not transmitted to the vehicle in time, the vehicle executes the control quantity calculated by the vehicle-side lateral L-MPC controller.

[0033] The vehicle-side longitudinal dual PID controller is based on a linear vehicle longitudinal dynamics model. When the control system is working normally, the control quantity calculated by the vehicle-side longitudinal dual PID controller is not executed. When the control quantity information of the edge cloud layer is not transmitted to the vehicle in time, the vehicle executes the control quantity calculated by the vehicle-side longitudinal dual PID controller.

[0034] In the above scheme, the vehicle-side execution layer includes a drive motor subsystem, a steering motor subsystem, a brake motor subsystem, and a sensor subsystem;

[0035] The drive motor subsystem is used to execute vehicle longitudinal control information to achieve normal driving speed control and differential drive control of the vehicle.

[0036] The steering motor subsystem is used to execute vehicle lateral control information to achieve normal vehicle steering control and active vehicle steering control.

[0037] The brake motor subsystem executes vehicle longitudinal control information to achieve normal vehicle speed control and differential braking control.

[0038] The sensor subsystem is used to collect vehicle driving information and actuator status information, send them to the vehicle control layer, and simultaneously send them to the edge cloud layer.

[0039] A control method for a cloud-based intelligent connected vehicle control system includes the following steps:

[0040] The edge cloud layer collects information from all roadside sensors and the status information of all vehicles within the control area. Based on the collected information, it makes a comprehensive decision to plan an optimal path and calls the lateral controller and longitudinal controller to perform path tracking. It also performs lateral control quantity fusion and longitudinal control quantity fusion and transmits the results to the vehicle control layer.

[0041] The vehicle-side control layer receives control commands transmitted from the edge cloud layer and performs lower-level tracking control. At the same time, it combines vehicle status information to make real-time vehicle stability judgments. When vehicle instability occurs, a multi-agent coordination controller is used to distribute additional yaw moment to the active steering module, differential braking module, and differential drive module. When the control information from the edge cloud layer is not transmitted to the vehicle end in time, the vehicle end calls the vehicle-side lateral L-MPC controller and the vehicle-side longitudinal dual PID controller for tracking control.

[0042] The vehicle-side execution layer receives control information transmitted from the vehicle-side control layer and controls each actuator to execute the control quantity.

[0043] The above solution specifically includes the following steps:

[0044] Step S1: All vehicles within the edge cloud layer control area send their own vehicle status information to the cloud. The edge cloud layer simultaneously collects all roadside sensor data and traffic information within the control area. The edge cloud layer then filters and fuses all the data.

[0045] Step S2: Based on the collected information, the edge cloud layer makes a comprehensive decision-making and planning process using vehicle economy, traffic efficiency, and vehicle safety as performance indicators, and outputs a planned path with speed and direction characteristics.

[0046] Step S3: Based on the vehicle status information collected by the edge cloud layer, train an edge lateral LSTM controller and an edge longitudinal LSTM controller based on the LSTM network. The input of the edge lateral LSTM controller is vehicle-level information, steering system-level information and steering motor actuator-level information during vehicle movement, and the output is the front wheel steering angle control quantity. The input of the edge longitudinal LSTM controller is vehicle-level information, braking system and drive system-level information and brake motor and drive motor actuator-level information during vehicle movement, and the output is vehicle speed and acceleration control quantity.

[0047] Step S4: Design an edge lateral N-MPC controller based on a nonlinear vehicle lateral dynamics model, and use the nonlinear MPC method to solve and calculate the front wheel steering angle control quantity. Based on a nonlinear vehicle longitudinal dynamics model, design an edge longitudinal nonlinear controller that includes the tire's nonlinear longitudinal slip and sideslip characteristics.

[0048] Step S5: Based on fuzzy control theory, fuse the control input of the edge lateral LSTM controller and the control input of the edge lateral N-MPC controller, fuse the control input of the edge longitudinal LSTM controller and the control input of the edge longitudinal nonlinear controller, and send them to the vehicle-end control layer.

[0049] Step S6: The stability judgment module of the vehicle control layer judges the stability of the vehicle based on the received information. When the vehicle is judged to be in a normal driving state, the lateral fusion control quantity and the longitudinal fusion control quantity are sent to the vehicle execution layer to complete the action execution. When the vehicle is judged to be in an unstable state, the multi-agent coordination controller is called to perform stability control.

[0050] Step S7: When the vehicle becomes unstable, the multi-agent coordination controller first calculates the total additional yaw moment required to maintain vehicle stability. Then, according to the vehicle's operating conditions, it distributes the total additional yaw moment to the active steering module, differential braking module, and differential drive module to collaboratively complete the application of the total additional yaw moment.

[0051] Step S8: The active steering module calculates the required additional steering angle, the differential braking module calculates the additional braking torque of the left and right wheels, and the differential drive module calculates the additional driving torque of the left and right wheels, and sends them to the vehicle-side execution layer for execution by the steering motor, brake motor and drive motor respectively.

[0052] Furthermore, the stability control of the multi-agent coordination controller in step S6 specifically includes the following steps:

[0053] Step S1) The horizontal fusion control quantity, vertical fusion control quantity, cloud perception fusion information and vehicle status information of the edge cloud layer are transmitted to the vehicle layer.

[0054] Step S2) The vehicle-side layer stability judgment module judges the vehicle's stability based on the received full-dimensional information and divides the vehicle's driving area into: stable area, transition area, unstable area and severe unstable area. When the vehicle is judged to be in the stable area, the vehicle can drive normally without stability control. The lateral fusion control quantity and the longitudinal fusion control quantity are sent to the execution layer to complete the action execution.

[0055] Step S3) When the vehicle is determined to be in an unstable state, first determine which region the vehicle is in. When it is in the transition region, it belongs to the low yaw moment control range, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module. When it is in the instability region, it belongs to the medium yaw moment control range, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module and the differential braking module in collaboration. When it is in the severe instability region, it belongs to the high yaw moment control range, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module, the differential braking module, and the differential drive module in collaboration.

[0056] Step S4) After determining the unstable region where the vehicle is located, the multi-agent coordination controller first calculates the total additional yaw moment required to maintain vehicle stability, and then distributes the total additional yaw moment to the active steering module, differential braking module and differential drive module through multi-agent coordination according to the unstable region where the vehicle is located.

[0057] Step S5) The active steering module calculates the required additional steering angle, the differential braking module calculates the additional braking torque of the left wheel and the additional braking torque of the right wheel, and the differential drive module calculates the additional driving torque of the left wheel and the additional driving torque of the right wheel.

[0058] Step S6) The calculated additional steering angle, additional braking torque and additional driving torque are sent to the execution layer to complete the action execution.

[0059] Furthermore, the total additional yaw moment is distributed in a coordinated manner in step S4) as follows:

[0060] ΔM Z =ΔM Z1 +ΔM Z2 +ΔM Z3

[0061] ΔM Z1 =k1ΔM Z ,ΔM Z2 =k2ΔM Z ,ΔM Z3 =k3ΔM Z

[0062] k1+k2+k3=1

[0063] In the formula, ΔM Z ΔM represents the total additional yaw moment required to maintain vehicle stability. Z1 This indicates the additional yaw moment required from the active steering module, ΔM. Z2 This indicates the additional yaw moment required from the differential braking module, ΔM. Z3 The differential drive module is required to provide additional yaw torque. k1 represents the allocation coefficient of the active steering module, k2 represents the allocation coefficient of the differential braking module, and k3 represents the allocation coefficient of the differential drive module. The coordination allocation coefficients of the three modules are calculated by the multi-agent coordination controller.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] 1. This invention fully utilizes the advantages of high computing power and full-dimensional information perception in the edge cloud to construct a cloud-based intelligent connected vehicle control system. By designing a lateral N-MPC controller based on a nonlinear vehicle dynamics model and a lateral LSTM controller based on a long short-term memory network, and by fusing lateral control quantities based on fuzzy control, the lateral control accuracy of the vehicle is improved. At the same time, by designing a longitudinal controller based on a nonlinear vehicle dynamics model and a longitudinal LSTM controller based on a long short-term memory network, and by fusing longitudinal control quantities based on fuzzy control, the longitudinal control accuracy of the vehicle is improved.

[0066] 2. The control method of this invention is based on the multi-agent theory. By judging the stable state of the vehicle and intelligently allocating the total additional yaw moment required to maintain the stability of the vehicle, the vehicle can achieve stable and safe driving. At the same time, a lateral and longitudinal controller based on a linear vehicle model is designed at the vehicle end to maintain vehicle control safety when the control information of the edge cloud layer is not transmitted to the vehicle end in time. This is of great significance for improving the driving safety of intelligent connected vehicles.

[0067] 3. In this invention's cloud-based control architecture, cloud computing offers the advantage of high computing power, enabling the use of more precise and high-dimensional nonlinear vehicle dynamics models for control calculations. Furthermore, it allows for the creation of data-driven control models for vehicles using vast amounts of cloud data. The fusion of mechanism-based and data-based control models effectively improves vehicle control accuracy. In this cloud-based architecture, a cloud controller serves as the vehicle's main controller, while a low-power onboard linear dynamics controller is designed at the vehicle end. Redundant control is implemented when the control input from the cloud controller is not accurately transmitted to the vehicle, further enhancing vehicle driving safety.

[0068] Note that the description of these effects does not preclude the existence of other effects. One aspect of the invention does not necessarily have all the aforementioned effects. Effects other than those described above can be readily observed and extracted from the description, drawings, claims, etc. Attached Figure Description

[0069] Figure 1 This is a diagram of a cloud-based intelligent connected vehicle control architecture according to one embodiment of the present invention.

[0070] Figure 2 This is a schematic diagram of a horizontal LSTM controller according to an embodiment of the present invention;

[0071] Figure 3 The chart shows a performance comparison of three lateral control methods. Figure 3 (a) is a comparison chart of lateral displacement errors. Figure 3 (b) is a comparison chart of yaw angular velocities;

[0072] Figure 4 This is a schematic diagram of the longitudinal LSTM controller of the present invention;

[0073] Figure 5 The chart shows a performance comparison of three longitudinal control methods. Figure 5 (a) is a comparison diagram of longitudinal displacement. Figure 5 (b) is a comparison diagram of longitudinal velocities;

[0074] Figure 6 This is a schematic diagram of the multi-agent cooperative control principle of the present invention;

[0075] Figure 7This is a flowchart of the control method of the present invention;

[0076] Figure 8 This is a diagram showing the vehicle driving area division of the present invention;

[0077] Figure 9 This is a result diagram of an application example of the present invention, wherein... Figure 9 (a) is a comparison chart of lateral errors. Figure 9 (b) is a comparison diagram of the centroid sideslip angle. Detailed Implementation

[0078] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0079] Reference Figure 1 As shown, a cloud-based intelligent connected vehicle control architecture includes: an edge cloud layer, a vehicle control layer, and a vehicle execution layer. The edge cloud layer is used to collect information from all roadside sensors and the status information of all vehicles within the control area. Based on the collected information, it comprehensively plans an optimal path and calls the lateral controller and longitudinal controller for path tracking, performing lateral control quantity fusion and longitudinal control quantity fusion respectively, and then transmits them to the vehicle control layer. The vehicle control layer receives control commands from the edge cloud layer and performs lower-level tracking control. At the same time, it combines vehicle status information to perform real-time vehicle stability judgment. When vehicle instability occurs, a multi-agent coordination controller is used to distribute additional yaw torque to the active steering module, differential braking module, and differential drive module. When the control information from the edge cloud layer is not transmitted to the vehicle end in time, the vehicle end calls the vehicle-side lateral L-MPC controller and the vehicle-side longitudinal dual PID controller for tracking control.

[0080] The vehicle-side execution layer is used to receive control information transmitted from the vehicle-side control layer and control each actuator to execute the control quantity.

[0081] The edge cloud layer includes a perception filtering fusion module, a state receiving and estimation module, a decision planning module, an edge horizontal N-MPC controller, an edge horizontal LSTM controller, a horizontal control quantity fusion module, an edge vertical nonlinear controller, an edge vertical LSTM controller, and a vertical control quantity fusion module. Based on the collected multi-dimensional information, the edge cloud layer comprehensively plans an optimal path and calls the horizontal and vertical controllers respectively for path tracking, and performs horizontal and vertical control quantity fusion to improve control accuracy.

[0082] The perception filtering and fusion module is used to collect information from all roadside sensors within the control area, then remove noise from the collected information through filtering technology, and fuse multi-source information. The information includes the location, shape, speed, and other information of all traffic participants in the area who cannot actively communicate with the cloud control center, as well as traffic signals.

[0083] The state receiving and estimation module is used to collect state information of all vehicles within the area, including vehicle-level information, system-level information, and actuator-level information. Vehicle-level information includes: vehicle speed, vehicle acceleration, yaw angle, pitch angle, roll angle, and sideslip angle; system-level information includes: steering system state information, braking system state information, and drive system state information; actuator-level information includes: steering motor state information, brake motor state information, and drive motor state information.

[0084] The decision planning module combines edge cloud perception information and all received vehicle status information to make comprehensive decision planning based on vehicle economy, traffic efficiency, and vehicle safety as performance indicators, and outputs a planned path with characteristics such as speed and direction to the control module.

[0085] The edge lateral N-MPC controller is based on a nonlinear vehicle lateral dynamics model and uses a nonlinear MPC method for solution calculation. It fully leverages the high computing power of the cloud to quickly solve the problem, thus obtaining the front wheel steering angle control quantity based on the lateral mechanism model. The nonlinear vehicle lateral dynamics model considering tire characteristics is as follows:

[0086]

[0087]

[0088] Where m is the mass of the car. It is the lateral acceleration in the vehicle body coordinate system, F yf It is the lateral force of the front tire, δ f For the front wheel steering angle, F xf It is the longitudinal force of the front tire, F yr It is the lateral force of the rear tire, V x It is the longitudinal velocity in the vehicle body coordinate system. It is the yaw rate, I z It is the moment of inertia of the vehicle about the z-axis. It is the first derivative of the yaw rate, l f It is the distance from the vehicle's center of gravity to the front axle, l r It is the distance from the vehicle's center of gravity to the rear axle.

[0089] The design principle of the edge lateral LSTM controller is referenced. Figure 2As shown, by collecting historical information on vehicle lateral control, a lateral controller is trained using a Long Short-Term Memory (LSTM) network in a data-driven manner. The formulas for the forget gate, input gate, cell state update, and output gate in the LSTM network are expressed as follows:

[0090] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0091]

[0092]

[0093] o t =σ(W o [h t-1 ,x t ]+b o ),h t =o t *tanh(C t )

[0094] In the formula, f t Let i represent the activation vector of the forget gate in the network. t W represents the activation vector of the output gate in the network. f W i W C and W o Here are the weight matrices for the forget gate, input gate, cell state update, and output gate, respectively. f b i b C and b o Here, represents the bias vectors corresponding to the forget gate, input gate, cell state update, and output gate, respectively; σ represents the non-linear activation function used to compress the input to between 0 and 1; h... t-1 h represents the hidden state at the previous time step. t Indicates the hidden state at the current time step, x t This represents the current time step input. Indicates the candidate cell state, C t C represents the current state of the cell. t-1 represents the cell state at the previous time step, tanh represents another activation function with an output range between -1 and 1, and t represents time.

[0095] Vehicle-level information, steering system-level information, and steering motor actuator-level information from historical vehicle driving data are used as inputs to the LSTM network, and the front wheel steering angle control value is used as the output of the LSTM network. An edge lateral LSTM controller network is trained offline, and real-time data under different operating conditions are used to optimize the network online during subsequent use of the network.

[0096] The lateral control quantity fusion module fuses the control quantities of the edge lateral N-MPC controller based on the lateral mechanism model and the control quantities of the edge lateral LSTM controller based on data-driven mechanisms. In the fused lateral control quantities, the control quantity of the edge lateral N-MPC controller is represented as u. y1 (t), the weight is represented by k y The control quantity of the edge lateral LSTM controller is represented as u y2 (t), the weight is represented as 1-k y Let t be time. Based on fuzzy control, the control input weight coefficients of the two controllers are obtained:

[0097] u y (t)=k y u y1 (t)+(1-k y )u y2 (t)

[0098] Figure 3 The figure shows the actual control effect after lateral control quantity fusion of the present invention. Simulations of dual lane change trajectory tracking were conducted at a vehicle speed of 60 km / h and a road adhesion coefficient of 0.75. Among the lateral displacement error and yaw rate indices characterizing vehicle lateral control performance, the control error of lateral control quantity fusion was improved by 52.3% compared to edge lateral N-MPC control and by 15.6% compared to edge lateral LSTM control. Figure 3 As shown in (a), the yaw rate index is more stable compared to edge lateral N-MPC control and edge lateral LSTM control, as... Figure 3 As shown in (b).

[0099] The edge longitudinal nonlinear controller is based on a nonlinear vehicle longitudinal dynamics model, which takes into account the nonlinear characteristics of the tires. It leverages the high computing power of the cloud to solve for high-precision longitudinal control quantities.

[0100]

[0101] in, It is the longitudinal acceleration in the vehicle body coordinate system, F xr It is the longitudinal force of the rear wheel, V yThis is the lateral velocity in the vehicle's coordinate system. The longitudinal and lateral forces of the front and rear wheels are calculated using the magic formula tire model.

[0102] Y=D·sin(C·arctan(B·xE(B·x-arctan(B·x))))+S v

[0103] x = X + S h

[0104] In the formula, Y can represent the lateral force and longitudinal force of the front and rear wheels, respectively; X can represent the slip angle and longitudinal slip ratio of the front and rear wheels, respectively; D is the peak factor; B is the stiffness factor; C is the curve shape factor; E is the curve curvature factor; and S... h For the horizontal drift of the curve, S v This refers to the vertical drift of the curve.

[0105] The design principle of the edge longitudinal LSTM controller is referenced. Figure 4 As shown, by collecting historical information on vehicle longitudinal control, a longitudinal controller is trained using a Long Short-Term Memory (LSTM) network based on a data-driven approach. The vehicle-level information, braking system and drive system-level information, and brake motor and drive motor actuator-level information during vehicle operation are used as inputs to the LSTM network, and the vehicle speed and acceleration control quantities are used as outputs to the LSTM network. An edge longitudinal LSTM controller network is trained offline, and in subsequent use of the network, real-time data under different operating conditions are used to optimize the network online.

[0106] The longitudinal control quantity fusion module fuses the control quantity of the edge longitudinal nonlinear controller based on the longitudinal mechanism model and the control quantity of the edge longitudinal LSTM controller based on data driving. The control quantity weight coefficients of the two controllers are solved based on fuzzy control. Then, the control quantity of the edge longitudinal nonlinear controller and the control quantity of the edge longitudinal LSTM controller are calculated separately using the weight coefficients. Finally, the two are fused to obtain the final longitudinal control quantity output.

[0107] The longitudinal control quantity fusion module fuses the control quantity of the edge longitudinal nonlinear controller based on the longitudinal mechanism model and the control quantity of the data-driven edge longitudinal LSTM controller. In the fused longitudinal control quantity, the control quantity of the edge longitudinal nonlinear controller is represented as u. x1 (t), the weight is represented by k x The control quantity of the edge longitudinal LSTM controller is represented as u x2 (t), the weight is represented as 1-k x The control weight coefficients of the two controllers are obtained based on fuzzy control.

[0108] u x (t)=k x u x1 (t)+(1-k x )u x2 (t)

[0109] Figure 5 The diagram shows the actual control effect after longitudinal control quantity fusion of the present invention. Vehicle speed tracking simulation was conducted on a straight road under a road adhesion coefficient of 0.75. In the speed error index, which characterizes the vehicle's longitudinal control performance, the control error of longitudinal control quantity fusion was improved by 3.7% compared to edge longitudinal nonlinear control and by 5.2% compared to edge longitudinal LSTM control. In the longitudinal displacement error index, which characterizes the vehicle's longitudinal control performance, the control error of longitudinal control quantity fusion was improved by 1.7% compared to edge longitudinal nonlinear control and by 2.1% compared to edge longitudinal LSTM control. Figure 5 (a) and Figure 5 As shown in (b).

[0110] The vehicle-side control layer includes: a stability assessment module, a multi-agent coordination controller, an active steering module, a differential braking module, a differential drive module, a vehicle-side lateral L-MPC controller, and a vehicle-side longitudinal dual PID controller. The vehicle-side control layer receives control commands from the edge cloud layer and performs lower-level tracking control. Simultaneously, it combines vehicle status information to perform real-time vehicle stability assessment. When vehicle instability occurs, the multi-agent coordination controller distributes additional yaw torque to the active steering module, differential braking module, and differential drive module. Furthermore, when control information from the edge cloud layer is not transmitted to the vehicle-side in a timely manner, the vehicle-side invokes the vehicle-side lateral L-MPC controller and the vehicle-side longitudinal dual PID controller for tracking control, achieving vehicle control redundancy.

[0111] The stability judgment module makes stability judgments based on the real-time status information of the vehicle. When the vehicle is in a stable state, it directly transmits the control information received from the edge cloud layer to the vehicle execution layer. When the vehicle is in an unstable state, it calls the multi-agent coordination controller to distribute additional yaw torque.

[0112] The principle of multi-agent coordinated control is as follows: Figure 6 As shown, the vehicle status is first determined based on all the perceived information, then the total additional yaw moment required to maintain vehicle stability is calculated, and then the total additional yaw moment is distributed to the active steering module, differential braking module and differential drive module according to the vehicle operating conditions.

[0113] After receiving the allocated additional yaw moment, the active steering module adds an extra front wheel angle on top of the front wheel angle that maintains normal steering to achieve the additional yaw moment.

[0114] After receiving the allocated additional yaw moment, the differential braking module applies different braking torques to the left and right wheels to achieve the additional yaw moment.

[0115] After receiving the allocated additional yaw torque, the differential drive module applies different driving torques to the left and right wheels to achieve the additional yaw torque.

[0116] The vehicle-side lateral L-MPC controller is based on a linear vehicle lateral dynamics model. While its accuracy is lower than that of a nonlinear vehicle lateral dynamics model, it features rapid solution and is suitable for vehicle-mounted computing platforms with low computing power. During normal operation of the control system, the control quantities calculated by the vehicle-side lateral L-MPC controller are not executed. However, when control quantity information from the edge cloud layer is not transmitted to the vehicle in a timely manner, the vehicle executes the control quantities calculated by the vehicle-side lateral L-MPC controller to ensure lateral driving safety.

[0117] The vehicle-side longitudinal dual PID controller is based on a linear vehicle longitudinal dynamics model. While its accuracy is lower than that of a nonlinear vehicle longitudinal dynamics model that considers tire characteristics, it features rapid solution and is suitable for vehicle-mounted computing platforms with low computing power. During normal operation of the control system, the control inputs calculated by the vehicle-side longitudinal dual PID controller are not executed. However, when control input information from the edge cloud layer is not transmitted to the vehicle in a timely manner, the vehicle executes the control inputs calculated by the vehicle-side longitudinal dual PID controller to ensure longitudinal driving safety.

[0118] The vehicle-side execution layer includes a drive motor subsystem, a steering motor subsystem, a brake motor subsystem, and a sensor subsystem. The vehicle-side execution layer receives control information transmitted from the vehicle-side control layer and calls on each actuator to perform the control quantity.

[0119] The drive motor subsystem is used to execute longitudinal control information of the vehicle, thereby realizing normal driving speed control and differential drive control of the vehicle.

[0120] The steering motor subsystem is used to execute vehicle lateral control information to achieve normal vehicle steering control and active vehicle steering control.

[0121] The brake motor subsystem executes longitudinal control information for the vehicle, enabling normal driving speed control and differential braking control.

[0122] The sensor subsystem is used to collect vehicle driving information and actuator status information, send them to the vehicle control layer, and simultaneously send them to the edge cloud layer.

[0123] A control method for a cloud-based intelligent connected vehicle control system includes the following steps:

[0124] The edge cloud layer collects information from all roadside sensors and the status information of all vehicles within the control area. Based on the collected information, it makes a comprehensive decision to plan an optimal path and calls the lateral controller and longitudinal controller to perform path tracking. It also performs lateral control quantity fusion and longitudinal control quantity fusion and transmits the results to the vehicle control layer.

[0125] The vehicle-side control layer receives control commands transmitted from the edge cloud layer and performs lower-level tracking control. At the same time, it combines vehicle status information to make real-time vehicle stability judgments. When vehicle instability occurs, a multi-agent coordination controller is used to distribute additional yaw moment to the active steering module, differential braking module, and differential drive module. When the control information from the edge cloud layer is not transmitted to the vehicle end in time, the vehicle end calls the vehicle-side lateral L-MPC controller and the vehicle-side longitudinal dual PID controller for tracking control.

[0126] The vehicle-side execution layer receives control information transmitted from the vehicle-side control layer and controls each actuator to execute the control quantity.

[0127] Figure 7 The diagram illustrates the specific workflow of the control method for the cloud-based intelligent connected vehicle control system of the present invention, which includes the following steps:

[0128] Step S1: All vehicles within the edge cloud control area send their own vehicle status information to the cloud. At the same time, the edge cloud collects all roadside sensor data and traffic information within the control area, and performs filtering and fusion processing on all the data.

[0129] Step S2: Based on the collected full-dimensional information, the edge cloud performs comprehensive decision-making and planning based on existing technologies, using vehicle economy, traffic efficiency, and vehicle safety as performance indicators, and outputs a planned path with characteristics such as speed and direction.

[0130] Step S3: Based on the vehicle status information collected from the edge cloud, train an edge lateral LSTM controller and an edge longitudinal LSTM controller using an LSTM network. The inputs to the edge lateral LSTM controller are vehicle-level information, steering system-level information, and steering motor actuator-level information during vehicle operation; the output is the front wheel steering angle control value. The inputs to the edge longitudinal LSTM controller are vehicle-level information, braking system and drive system-level information, and brake motor and drive motor actuator-level information during vehicle operation; the outputs are vehicle speed and acceleration control values. The edge lateral and longitudinal LSTM controller networks are trained offline, and then optimized online using real-time data under different operating conditions during subsequent network usage.

[0131] Step S4: Design an edge lateral N-MPC controller based on the nonlinear vehicle lateral dynamics model and nonlinear MPC control theory. Design an edge longitudinal nonlinear controller based on the nonlinear vehicle longitudinal dynamics model, which considers the nonlinear longitudinal slip / lateral slip characteristics of the tire.

[0132] Step S5: Based on fuzzy control theory, fuse the control input of the edge lateral LSTM controller and the control input of the edge lateral N-MPC controller, fuse the control input of the edge longitudinal LSTM controller and the control input of the edge longitudinal nonlinear controller, and send them to the vehicle end.

[0133] Step S6: The vehicle-side stability judgment module judges the vehicle's stability based on the received full-dimensional information. When the vehicle is judged to be in a normal driving state, the lateral fusion control quantity and longitudinal fusion control quantity are sent to the execution layer to complete the action execution. When the vehicle is judged to be in an unstable state, the multi-agent coordination controller is called to perform stability control.

[0134] Step S7: When the vehicle becomes unstable, the multi-agent coordination controller first calculates the total additional yaw moment required to maintain vehicle stability. Then, based on the vehicle's operating conditions, it distributes the total additional yaw moment to the active steering module, differential braking module, and differential drive module to collaboratively complete the application of the total additional yaw moment.

[0135] Step S8: The active steering module calculates the required additional steering angle, the differential braking module calculates the additional braking torque of the left and right wheels, and the differential drive module calculates the additional driving torque of the left and right wheels, and sends them to the execution layer for execution by the steering motor, brake motor and drive motor respectively.

[0136] The stability control steps of the multi-agent coordination controller are as follows:

[0137] Step S1) Transmit the horizontal fusion control quantity, vertical fusion control quantity, cloud perception fusion information and vehicle status information from the edge cloud to the vehicle.

[0138] Step S2) The vehicle-side stability judgment module performs a stability judgment on the vehicle based on the received full-dimensional information, and divides the vehicle's driving area as follows: Figure 8 The division shown Figure 8The horizontal axis represents the sideslip angle, and the vertical axis represents the yaw rate. The vehicle's driving area is divided into four regions: stable region, transition region, unstable region, and severe instability region. The coordinates of the stable region are (-0.2, -0.2), (-0.18, 0.2), (0.18, -0.2), (0.2, 0.2); the coordinates of the transition region are (-0.4, 0.4), (-0.35, 0.4), (0.35, -0.4), (0.4, 0.4); the coordinates of the unstable region are (-0.6, -0.6), (-0.5, 0.6), (0.5, -0.6), (0.6, 0.6); and the coordinates of the severe instability region are... Figure 7 The remaining area. When it is determined that the vehicle is in a stable area, the vehicle can drive normally without stability control. The lateral fusion control quantity and the longitudinal fusion control quantity are sent to the execution layer to complete the action execution.

[0139] Step S3) When the vehicle is determined to be in an unstable state, first determine which region the vehicle is in. When it is in the transition region, it belongs to the low yaw moment control range (0, 1500) Nm, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module. When it is in the instability region, it belongs to the medium yaw moment control range (1500, 5000) Nm, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module and the differential braking module in collaboration. When it is in the severe instability region, it belongs to the high yaw moment control range (greater than 5000) Nm, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module, the differential braking module, and the differential drive module in collaboration.

[0140] Step S4) After determining the unstable region where the vehicle is located, the multi-agent coordination controller first calculates the total additional yaw moment required to maintain vehicle stability. Then, based on the unstable region where the vehicle is located, it distributes the total additional yaw moment to the active steering module, differential braking module, and differential drive module through multi-agent coordination. The total additional yaw moment coordination distribution method is as follows:

[0141] ΔM Z =ΔM Z1 +ΔM Z2 +ΔM Z3

[0142] ΔM Z1 =k1ΔM Z ,ΔM Z2 =k2ΔM Z ,ΔM Z3 =k3ΔM Z

[0143] k1+k2+k3=1

[0144] In the formula, ΔMZ ΔM represents the total additional yaw moment required to maintain vehicle stability. Z1 This indicates the additional yaw moment required from the active steering module, ΔM. Z2 This indicates the additional yaw moment required from the differential braking module, ΔM. Z3 The differential drive module is required to provide additional yaw torque. k1 represents the allocation coefficient of the active steering module, k2 represents the allocation coefficient of the differential braking module, and k3 represents the allocation coefficient of the differential drive module. The coordination allocation coefficients of the three modules are calculated by the multi-agent coordination controller.

[0145] Step S5) The active steering module calculates the required additional steering angle, the differential braking module calculates the additional braking torque of the left and right wheels, and the differential drive module calculates the additional driving torque of the left and right wheels.

[0146] Step S6) The calculated additional steering angle, additional braking torque and additional driving torque are sent to the execution layer to complete the action execution.

[0147] To verify the actual control effect of the present invention, the vehicle state was set to [condition missing] in the simulation environment. Figure 8 The area of ​​severe instability shown has a vehicle center of gravity sideslip angle of 0.6 radians and a yaw rate of 0.5 radians / second. Figure 9 This paper demonstrates the actual control performance of a cloud-based intelligent connected vehicle control architecture and multi-agent cooperative stability control method based on this invention under severe instability conditions. In the centroid sideslip angle index, which characterizes vehicle stability, the control error of the proposed multi-agent cooperative stability control method is improved by 43.2% and 14.7% compared to the active steering cooperative differential drive method and the active steering cooperative differential braking method, respectively. In the lateral error index, which characterizes path tracking performance, the proposed multi-agent cooperative stability control method maintains the same stability as the active steering cooperative differential braking method and outperforms the active steering cooperative differential drive method. Figure 9 (a) and Figure 9 As shown in (b).

[0148] This invention relates to a cloud-based intelligent connected vehicle control system. The edge-cloud layer comprises a perception filtering and fusion module, a state receiving and estimation module, a decision-making and planning module, a lateral N-MPC controller based on a nonlinear vehicle dynamics model, a lateral LSTM controller based on a long short-time memory network, a longitudinal controller based on a nonlinear vehicle dynamics model, and a longitudinal LSTM controller based on a long short-time memory network. Vehicle-road-cloud collaborative perception and communication technologies acquire vehicle driving environment and traffic information. Combined with the state information uploaded by the vehicle to the cloud, a suitable path is planned for the vehicle. The lateral N-MPC controller and the lateral LSTM controller track the planned path to obtain two lateral control variables. Based on fuzzy control, the weight coefficients of the two lateral control variables are determined, and a fused lateral control variable is calculated. Similarly, the longitudinal nonlinear controller and the longitudinal LSTM controller track the speed and acceleration information of the planned path to obtain two longitudinal control variables. Based on fuzzy control, the weight coefficients of the two longitudinal control variables are determined, and a fused longitudinal control variable is calculated. The vehicle-side control layer includes a lateral L-MPC controller based on a linear vehicle dynamics model, a longitudinal dual PID controller based on the same model, a stability assessment module, and a multi-agent coordination controller based on multi-agent theory to coordinate additional yaw moment. When the autonomous intelligent connected vehicle is driving normally, the vehicle receives fused lateral and longitudinal control signals from the cloud and sends them to each actuator for execution. When the stability assessment module detects that the vehicle is in an unstable state, it calls the multi-agent coordination controller to coordinate and distribute additional yaw moment, which is then sent to each actuator for execution. When packet loss or high latency occurs in the vehicle-cloud communication network, the vehicle-side lateral MPC controller and longitudinal dual PID controller are called to calculate the lateral and longitudinal control signals respectively, maintaining stable and safe vehicle operation.

[0149] The cloud-based intelligent connected vehicle control system of the present invention improves the control accuracy of the vehicle by integrating the mechanism model and the data-driven model. The lateral MPC controller, longitudinal dual PID controller and additional yaw moment multi-agent coordination controller deployed at the vehicle end ensure the driving safety of the vehicle.

[0150] This invention designs a lateral N-MPC controller based on a nonlinear vehicle dynamics model and a lateral LSTM controller based on a long short-time memory network. Fuzzy control is used to determine the weight coefficients of the two lateral control variables for lateral control quantity fusion. Simultaneously, a longitudinal controller based on a nonlinear vehicle dynamics model and a longitudinal LSTM controller based on a long short-time memory network are designed, and fuzzy control is used to determine the weight coefficients of the two lateral control variables for longitudinal control quantity fusion. The cloud-based lateral and longitudinal fusion control effectively improves the vehicle's control accuracy. At the vehicle end, a stability coordination controller based on multi-agent theory is designed to compensate for yaw moments under various adverse operating conditions, maintaining vehicle stability. A lateral MPC controller based on a linear vehicle dynamics model and a longitudinal dual PID controller are also designed. When the control quantity from the cloud controller is not accurately transmitted to the vehicle end, the on-board controller is used for vehicle control, ensuring vehicle control safety.

[0151] It should be understood that although this specification is described according to various embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0152] The detailed descriptions listed above are merely specific illustrations of feasible embodiments of the present invention and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. A cloud-based intelligent connected vehicle control system, characterized in that, This includes the edge cloud layer, the vehicle control layer, and the vehicle execution layer; The edge cloud layer is used to collect information from all roadside sensors and the status information of all vehicles within the control area. Based on the collected information, it makes a comprehensive decision to plan an optimal path and calls the lateral controller and longitudinal controller respectively to perform path tracking, perform lateral control quantity fusion and longitudinal control quantity fusion respectively, and transmit them to the vehicle control layer. The vehicle-side control layer is used to receive control commands transmitted from the edge cloud layer and perform lower-level tracking control. At the same time, it combines vehicle status information to make real-time vehicle stability judgments. When vehicle instability occurs, a multi-agent coordination controller is used to distribute additional yaw moment to the active steering module, differential braking module, and differential drive module. When the control information from the edge cloud layer is not transmitted to the vehicle end in time, the vehicle end calls the vehicle-side lateral L-MPC controller and the vehicle-side longitudinal dual PID controller for tracking control. The vehicle-side execution layer is used to receive control information transmitted from the vehicle-side control layer and control each actuator to execute the control quantity.

2. The cloud-based intelligent connected vehicle control system according to claim 1, characterized in that, The edge cloud layer includes a perception filtering fusion module, a state receiving and estimation module, a decision planning module, an edge horizontal N-MPC controller, an edge horizontal LSTM controller, a horizontal control quantity fusion module, an edge vertical nonlinear controller, an edge vertical LSTM controller, and a vertical control quantity fusion module; The perception filtering and fusion module is used to collect information from all roadside sensors within the control area, then filter out noise from the collected information, and fuse multi-source information. The information includes the location, shape, speed information and traffic signals of all traffic participants in the area who cannot actively communicate with the cloud control center. The state receiving and estimation module is used to collect state information of all vehicles in the area, including vehicle-level information, system-level information, and actuator-level information. The vehicle-level information includes vehicle speed, vehicle acceleration, yaw angle, pitch angle, roll angle, and center of gravity sideslip angle. The system-level information includes steering system state information, braking system state information, and drive system state information. The actuator-level information includes steering motor state information, brake motor state information, and drive motor state information. The decision planning module is used to combine all roadside sensor information and all received vehicle status information within the control area, and to make comprehensive decision planning based on vehicle economy, traffic efficiency and vehicle safety as performance indicators, and output a planned path with speed and direction characteristics. The edge lateral N-MPC controller is based on a nonlinear vehicle lateral dynamics model and uses a nonlinear MPC method for calculation to solve for the front wheel steering angle control quantity. The edge lateral LSTM controller is a data-driven lateral controller trained with a long short-term memory network. It takes vehicle-level information, steering system-level information, and steering motor actuator-level information during vehicle operation as inputs to the LSTM network and front wheel steering angle control as outputs to the LSTM network. The lateral control quantity fusion module is used to fuse the control quantity of the edge lateral N-MPC controller and the control quantity of the edge lateral LSTM controller. Based on fuzzy control, the control quantity weight coefficients of the two controllers are solved. Then, the control quantity of the edge lateral N-MPC controller and the control quantity of the edge lateral LSTM controller are calculated respectively through the weight coefficients. Finally, the two are fused to obtain the final lateral control quantity output. The edge longitudinal nonlinear controller is based on a nonlinear vehicle longitudinal dynamics model, which includes the nonlinear longitudinal slip and sideslip characteristics of the tire, and solves for the longitudinal control quantity. The edge longitudinal LSTM controller is a data-driven longitudinal controller trained with a long short-term memory network. It takes vehicle-level information, braking system and drive system-level information, and brake motor and drive motor actuator-level information during vehicle operation as inputs to the LSTM network, and takes vehicle speed and acceleration control quantities as outputs of the LSTM network. The longitudinal control quantity fusion module fuses the control quantity of the edge longitudinal nonlinear controller and the control quantity of the edge longitudinal LSTM controller. Based on fuzzy control, the control quantity weight coefficients of the two controllers are solved. Then, the control quantity of the edge longitudinal nonlinear controller and the control quantity of the edge longitudinal LSTM controller are calculated separately using the weight coefficients. Finally, the two are fused to obtain the final longitudinal control quantity output.

3. The cloud-based intelligent connected vehicle control system according to claim 2, characterized in that, The control quantity of the edge lateral N-MPC controller in the fused lateral control quantity is represented as u by the lateral control quantity fusion module. y1 (t), the weight is represented by k y The control quantity of the edge lateral LSTM controller is represented as u y2 (t), the weight is represented as 1-k y Let t be time. Based on fuzzy control, the control input weight coefficients of the two controllers are obtained: u y (t)=k y u y1 (t)+(1-k y )u y2 (t)。 4. The cloud-based intelligent connected vehicle control system according to claim 2, characterized in that, The longitudinal control quantity fusion module represents the control quantity of the edge longitudinal nonlinear controller in the fused longitudinal control quantity as u. x1 (t), the weight is represented by k x The control quantity of the edge longitudinal LSTM controller is represented as u x2 (t), the weight is represented as 1-k x Let t be time. Based on fuzzy control, the control input weight coefficients of the two controllers are obtained: u x (t)=k x u x1 (t)+(1-k x )u x2 (t)。 5. The cloud-based intelligent connected vehicle control system according to claim 1, characterized in that, The vehicle-side control layer includes a stability judgment module, a multi-agent coordination controller, an active steering module, a differential braking module, a differential drive module, a vehicle-side lateral L-MPC controller, and a vehicle-side longitudinal dual PID controller. The stability judgment module makes stability judgments based on the real-time status information of the vehicle. When the vehicle is in a stable state, it directly transmits the control information received from the edge cloud layer to the vehicle execution layer. When the vehicle is in an unstable state, it calls the multi-agent coordination controller to distribute additional yaw torque. The multi-agent coordination controller is used to first calculate the total additional yaw moment required to maintain vehicle stability, and then distribute the total additional yaw moment to the active steering module, differential braking module and differential drive module according to the vehicle operating conditions. The active steering module is used to receive the allocated additional yaw moment and then add an additional front wheel angle on top of the front wheel angle that maintains normal steering to achieve the additional yaw moment. The differential braking module is used to receive the allocated additional yaw moment and then apply different braking torques to the left and right wheels to achieve the additional yaw moment. The differential drive module is used to receive the allocated additional yaw torque and then apply different driving torques to the left and right wheels to achieve the additional yaw torque. The vehicle-side lateral L-MPC controller is based on a linear vehicle lateral dynamics model. When the control system is working normally, the control quantity calculated by the vehicle-side lateral L-MPC controller is not executed. When the control quantity information of the edge cloud layer is not transmitted to the vehicle in time, the vehicle executes the control quantity calculated by the vehicle-side lateral L-MPC controller. The vehicle-side longitudinal dual PID controller is based on a linear vehicle longitudinal dynamics model. When the control system is working normally, the control quantity calculated by the vehicle-side longitudinal dual PID controller is not executed. When the control quantity information of the edge cloud layer is not transmitted to the vehicle in time, the vehicle executes the control quantity calculated by the vehicle-side longitudinal dual PID controller.

6. The cloud-based intelligent connected vehicle control system according to claim 1, characterized in that, The vehicle-side execution layer includes a drive motor subsystem, a steering motor subsystem, a brake motor subsystem, and a sensor subsystem; The drive motor subsystem is used to execute vehicle longitudinal control information to achieve normal driving speed control and differential drive control of the vehicle. The steering motor subsystem is used to execute vehicle lateral control information to achieve normal vehicle steering control and active vehicle steering control. The brake motor subsystem executes vehicle longitudinal control information to achieve normal driving speed control and differential braking control of the vehicle. The sensor subsystem is used to collect vehicle driving information and actuator status information, send them to the vehicle control layer, and simultaneously send them to the edge cloud layer.

7. A control method for a cloud-based intelligent connected vehicle control system according to any one of claims 1 to 6, characterized in that, Includes the following steps: The edge cloud layer collects information from all roadside sensors and the status information of all vehicles within the control area. Based on the collected information, it makes a comprehensive decision to plan an optimal path and calls the lateral controller and longitudinal controller to perform path tracking. It also performs lateral control quantity fusion and longitudinal control quantity fusion and transmits the results to the vehicle control layer. The vehicle-side control layer receives control commands transmitted from the edge cloud layer and performs lower-level tracking control. At the same time, it combines vehicle status information to make real-time vehicle stability judgments. When vehicle instability occurs, a multi-agent coordination controller is used to distribute additional yaw moment to the active steering module, differential braking module, and differential drive module. When the control information from the edge cloud layer is not transmitted to the vehicle end in time, the vehicle end calls the vehicle-side lateral L-MPC controller and the vehicle-side longitudinal dual PID controller for tracking control. The vehicle-side execution layer receives control information transmitted from the vehicle-side control layer and controls each actuator to execute the control quantity.

8. The control method for a cloud-based intelligent connected vehicle control system according to claim 7, characterized in that, Specifically, the following steps are included: Step S1: All vehicles within the edge cloud layer control area send their own vehicle status information to the cloud. The edge cloud layer simultaneously collects all roadside sensor data and traffic information within the control area. The edge cloud layer then filters and fuses all the data. Step S2: Based on the collected information, the edge cloud layer makes a comprehensive decision-making and planning process using vehicle economy, traffic efficiency, and vehicle safety as performance indicators, and outputs a planned path with speed and direction characteristics. Step S3: Based on the vehicle status information collected by the edge cloud layer, train an edge lateral LSTM controller and an edge longitudinal LSTM controller based on the LSTM network. The input of the edge lateral LSTM controller is vehicle-level information, steering system-level information and steering motor actuator-level information during vehicle movement, and the output is the front wheel steering angle control quantity. The input of the edge longitudinal LSTM controller is vehicle-level information, braking system and drive system-level information and brake motor and drive motor actuator-level information during vehicle movement, and the output is vehicle speed and acceleration control quantity. Step S4: Design an edge lateral N-MPC controller based on a nonlinear vehicle lateral dynamics model, and use the nonlinear MPC method to solve and calculate the front wheel steering angle control quantity. Based on a nonlinear vehicle longitudinal dynamics model, design an edge longitudinal nonlinear controller that includes the tire's nonlinear longitudinal slip and sideslip characteristics. Step S5: Based on fuzzy control theory, fuse the control input of the edge lateral LSTM controller and the control input of the edge lateral N-MPC controller, fuse the control input of the edge longitudinal LSTM controller and the control input of the edge longitudinal nonlinear controller, and send them to the vehicle-end control layer. Step S6: The stability judgment module of the vehicle control layer judges the stability of the vehicle based on the received information. When the vehicle is judged to be in a normal driving state, the lateral fusion control quantity and the longitudinal fusion control quantity are sent to the vehicle execution layer to complete the action execution. When the vehicle is judged to be in an unstable state, the multi-agent coordination controller is called to perform stability control. Step S7: When the vehicle becomes unstable, the multi-agent coordination controller first calculates the total additional yaw moment required to maintain vehicle stability. Then, according to the vehicle's operating conditions, it distributes the total additional yaw moment to the active steering module, differential braking module, and differential drive module to collaboratively complete the application of the total additional yaw moment. Step S8: The active steering module calculates the required additional steering angle, the differential braking module calculates the additional braking torque of the left and right wheels, and the differential drive module calculates the additional driving torque of the left and right wheels, and sends them to the vehicle-side execution layer for execution by the steering motor, brake motor and drive motor respectively.

9. The control method for a cloud-based intelligent connected vehicle control system according to claim 8, characterized in that, The stability control of the multi-agent coordination controller in step S6 specifically includes the following steps: Step S1) The horizontal fusion control quantity, vertical fusion control quantity, cloud perception fusion information and vehicle status information of the edge cloud layer are transmitted to the vehicle layer. Step S2) The vehicle-side layer stability judgment module judges the vehicle's stability based on the received full-dimensional information and divides the vehicle's driving area into: stable area, transition area, unstable area and severe unstable area. When the vehicle is judged to be in the stable area, the vehicle can drive normally without stability control. The lateral fusion control quantity and the longitudinal fusion control quantity are sent to the execution layer to complete the action execution. Step S3) When the vehicle is determined to be in an unstable state, first determine which region the vehicle is in. When it is in the transition region, it belongs to the low yaw moment control range, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module. When it is in the instability region, it belongs to the medium yaw moment control range, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module and the differential braking module in collaboration. When it is in the severe instability region, it belongs to the high yaw moment control range, and the total additional yaw moment required to maintain vehicle stability is applied by the active steering module, the differential braking module, and the differential drive module in collaboration. Step S4) After determining the unstable region where the vehicle is located, the multi-agent coordination controller first calculates the total additional yaw moment required to maintain vehicle stability, and then distributes the total additional yaw moment to the active steering module, differential braking module and differential drive module through multi-agent coordination according to the unstable region where the vehicle is located. Step S5) The active steering module calculates the required additional steering angle, the differential braking module calculates the additional braking torque of the left wheel and the additional braking torque of the right wheel, and the differential drive module calculates the additional driving torque of the left wheel and the additional driving torque of the right wheel. Step S6) The calculated additional steering angle, additional braking torque and additional driving torque are sent to the execution layer to complete the action execution.

10. The control method for a cloud-based intelligent connected vehicle control system according to claim 9, characterized in that, The total additional yaw moment is distributed in the following manner in step S4): ΔM Z =ΔM Z1 +ΔM Z2 +ΔM Z3 ΔM Z1 =k1ΔM Z ,ΔM Z2 =k2ΔM Z ,ΔM Z3 =k3ΔM Z k1+k2+k3=1 In the formula, ΔM Z ΔM represents the total additional yaw moment required to maintain vehicle stability. Z1 This indicates the additional yaw moment required from the active steering module, ΔM. Z2 This indicates the additional yaw moment required from the differential braking module, ΔM. Z3 The differential drive module is required to provide additional yaw torque. k1 represents the allocation coefficient of the active steering module, k2 represents the allocation coefficient of the differential braking module, and k3 represents the allocation coefficient of the differential drive module. The coordination allocation coefficients of the three modules are calculated by the multi-agent coordination controller.

Images