A vehicle trajectory tracking energy consumption optimization method and system based on a linearized fuel model

Abstract

The application discloses a vehicle trajectory tracking energy consumption optimization method and system based on a linear fuel model, and relates to the field of intelligent driving.The method comprises the following steps: constructing a three-degree-of-freedom dynamic model of a target vehicle based on motion state information and a reference trajectory, and obtaining a linear discrete model after discretization; constructing a polynomial fuel consumption model, linearizing the polynomial fuel consumption model to obtain partial derivative weights, and calculating a linear fuel consumption cost; constructing a standard convex quadratic programming objective function based on the linear discrete model; and solving the standard convex quadratic programming objective function under a preset constraint condition to obtain an optimal control increment sequence, and generating an energy-saving tracking control instruction which is then sent to a controller.The application introduces a fuel model into the controller to predict high-energy-consumption transient behavior, thereby overcoming the limitation that the energy consumption target and the tracking target of a traditional MPC controller conflict with each other, and achieving the coordination of energy saving and safety while reducing the overall fuel consumption.

Description

Technical Field

[0001] This application relates to the field of intelligent driving, and in particular to a method and system for optimizing energy consumption by tracking vehicle trajectory based on a linearized fuel model. Background Technology

[0002] With the continuous development of autonomous driving systems, vehicles need to ensure safety and comfort in complex environments. Accurate perception, prediction, planning, and control are fundamental to achieving this goal. Trajectory tracking, as a safety-critical issue, requires the controller to ensure the vehicle accurately follows the planned path. Trajectory tracking must adhere to physical constraints such as vehicle dynamics. Due to its advantages in handling multivariable constraints and rolling time-domain optimization, the MPC (Model Predictive Control) controller has become the mainstream motion control solution for autonomous driving.

[0003] Traditional trajectory tracking controllers mostly focus on improving tracking accuracy and ensuring driving safety. These solutions typically ignore the impact of vehicle longitudinal behavior on energy consumption. Currently, there are several technical bottlenecks in control schemes that consider energy consumption. Traditional energy consumption models based on steady-state MAP diagrams cannot capture transient dynamic characteristics, and complex transient correction models often require data such as gear position, which are difficult to obtain in predictive planning. Traditional MPC controllers have conflicting energy consumption and tracking objectives, making it impossible to reduce overall fuel consumption while ensuring safe tracking. Summary of the Invention

[0004] The purpose of this application is to provide a vehicle trajectory tracking energy consumption optimization method and system based on a linearized fuel model. By introducing the fuel model into the controller, the transient behavior of high energy consumption can be predicted, thereby overcoming the limitation of the conflict between the energy consumption target and the tracking target of the traditional MPC controller, and achieving the coordination of energy saving and safety while reducing overall fuel consumption.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model. This method is applied to an MPC controller. The method includes: acquiring the current motion state information and reference trajectory of the target vehicle; constructing a three-degree-of-freedom dynamic model of the target vehicle based on the motion state information and reference trajectory, and discretizing the three-degree-of-freedom dynamic model to obtain a linear discrete model; the three-degree-of-freedom dynamic model is a dynamic model including longitudinal, lateral, and yaw motions; the linear discrete model is used to predict the impact of control increments on the target vehicle's state tracking error within a future finite time domain; constructing a polynomial fuel consumption model based on transient dynamic characteristics, and using a model based on the expected total traction power operating mode corresponding to the current reference trajectory. The linearization method linearizes the polynomial fuel consumption model to obtain partial derivative weights, and then uses these weights to calculate the linearized fuel consumption cost. The partial derivative weights include speed cost weights and acceleration cost weights. Based on the linear discrete model, a standard convex quadratic programming objective function is constructed using trajectory tracking error cost, control increment smoothness cost, and linearized fuel consumption cost. This objective function is a lossless convex function. Under preset constraints, the optimal control increment sequence is obtained by solving the standard convex quadratic programming objective function. Based on this sequence, an energy-saving tracking control command is generated and sent to the target vehicle's underlying drive-by-wire chassis for execution. The energy-saving tracking control command includes the optimal front wheel steering angle control command and longitudinal acceleration control command for the current moment.

[0006] In a second aspect, this application also provides a computer system, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model as described in the first aspect.

[0007] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application acquires the vehicle's motion state and reference trajectory, constructs a three-degree-of-freedom dynamic model including longitudinal, lateral, and yaw motions, and discretizes it to obtain a linear discrete model. Based on transient dynamics, a polynomial fuel consumption model is constructed. Based on the expected total traction power, a mode scheduling method is used to linearize the speed and acceleration cost weights, and the linearized fuel consumption cost is calculated. Then, based on the linear discrete model, a lossless convex standard convex quadratic programming objective function is constructed based on the trajectory tracking error cost, control increment smoothness cost, and linearized fuel consumption cost. Under constraints, the optimal control increment sequence is solved, generating front wheel steering angle and longitudinal acceleration commands that are sent to the drive-by-wire chassis for execution. This application introduces the fuel consumption model into the controller to predict high-energy-consumption transient behavior, solving the problem of non-convexity and massive computational load in MPC optimization due to the introduction of the energy consumption model. This overcomes the limitation of traditional MPC controllers where energy consumption and tracking objectives conflict, achieving a balance between energy saving and safety while reducing overall fuel consumption, meeting the real-time control requirements of the vehicle. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating a vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model, provided in an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of the planar dynamics and coordinate system projection structure of a three-degree-of-freedom monorail vehicle provided in an embodiment of this application.

[0011] Figure 3 This is a schematic diagram of the closed-loop high-fidelity co-simulation platform architecture and data flow interaction structure provided in the embodiments of this application.

[0012] Figure 4 This is a schematic diagram of the internal algorithm architecture logic of the multi-objective quadratic programming controller provided in the embodiments of this application.

[0013] Figure 5 This is an internal structure diagram of a computer system provided in an embodiment of this application. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] Example 1, such as Figure 1 As shown, this embodiment provides a vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model. This method is applied to an MPC controller and includes: S1. Obtain the current motion status information and reference trajectory of the target vehicle.

[0017] Optionally, motion state information is acquired through onboard sensors (such as inertial navigation systems, wheel speed sensors, etc.). The vehicle's motion state vector mainly includes: lateral velocity, longitudinal velocity, heading angle, yaw rate, and longitudinal and lateral position coordinates in the global coordinate system.

[0018] Optionally, the reference trajectory is issued by the planning layer of the target vehicle's autonomous driving system; the reference trajectory includes the target's reference state over a future period of time. and reference control input .

[0019] S2. Construct a three-degree-of-freedom dynamic model of the target vehicle based on motion state information and reference trajectory, and discretize the three-degree-of-freedom dynamic model to obtain a linear discrete model; the three-degree-of-freedom dynamic model is a dynamic model that includes longitudinal, lateral and yaw motions; the linear discrete model is used to predict the impact of control increments on the target vehicle state tracking error in the future finite time domain.

[0020] Furthermore, step S2 specifically includes: S21. A three-degree-of-freedom dynamic model is constructed by characterizing state variables using motion state information and control variables using a reference trajectory; the state variables... Including: lateral velocity, longitudinal velocity, heading angle, yaw rate, global x-coordinate, and global y-coordinate; the control variables This includes: front wheel steering angle and longitudinal acceleration.

[0021] S22. At the current reference trajectory working point, the three-degree-of-freedom dynamic model is linearized by first-order Taylor expansion to obtain the linearized three-degree-of-freedom dynamic model.

[0022] S23. The forward Euler method is used to discretize the linearized three-degree-of-freedom dynamic model at the sampling time to obtain the initial discretized model.

[0023] S24. Based on the initial discretized model, the control input from the previous time step is added to the state variables to obtain the augmented state vector (control increment). Augmented discrete state equations are constructed using augmented state vectors and output as linear discrete models.

[0024] In practical applications, to achieve high-precision trajectory tracking, a three-degree-of-freedom dynamic model considering tire lateral forces is used for prediction. Since the original dynamic and kinematic models are highly nonlinear and difficult to use directly for real-time optimization, in each control cycle, the reference trajectory provided by the nonlinear system at the planning layer is used. A first-order Taylor expansion is performed nearby to achieve linearization. Subsequently, the forward Euler method is used with a sampling time... The continuous-time linear model is discretized. To introduce integral action and eliminate steady-state error, while also directly constraining the control increment to ensure actuator smoothness, an augmented state vector is further constructed. Thus, a linear discrete model is established.

[0025] Among them, such as Figure 2 As shown, for the three-degree-of-freedom dynamic model, in order to reduce computational complexity while ensuring the model's prediction accuracy and facilitating real-time solution by the onboard controller, the complex vehicle suspension system is decoupled from planar motion. Specifically, the following reasonable assumptions are made for the vehicle dynamic model: the pitch and roll dynamics of the suspension are ignored; the lateral load transfer between the left and right wheels is ignored; and the left and right wheels on the same axle are equivalent to a single-track bicycle model.

[0026] like Figure 2 The force analysis of the entire vehicle and tires shown is based on the Newton-Euler equations, establishing a three-free nonlinear dynamic differential equation that includes the longitudinal, lateral, and yaw motions of the vehicle: .

[0027] in, For the overall vehicle quality, Let yaw moment of inertia be the vehicle's rotation about the Z-axis. and These are the longitudinal and lateral velocities at the center of mass, respectively. and These are the vehicle's heading angle and yaw rate, respectively. and These are the distances from the center of mass to the front and rear axles, respectively. This refers to the front wheel steering angle; and These are the longitudinal forces of the front and rear tires, respectively. and These are the lateral forces of the front and rear tires, respectively. This is the longitudinal resultant resistance force acting on the vehicle.

[0028] In practical applications, in order to reflect normal driving conditions (when the tire slip angle is small, i.e.) The coupling relationship between tire slip angle and lateral force is modeled in this embodiment using a linearization method to model the interaction between the tire and the road surface. The lateral forces of the front and rear tires can be approximated as: .

[0029] in, and These are the combined lateral stiffness of the front and rear wheels, respectively. Tire slip angle. and It can be derived from the geometric relationship between the vehicle's velocity vector and steering angle: .

[0030] Furthermore, to achieve accurate tracking of the planned trajectory in the global inertial coordinate system, global kinematic equations for coordinate transformation must be introduced. This involves combining the velocity in the vehicle body coordinate system with the rate of change of longitudinal position in the global coordinate system. and lateral position change rate Represented as: .

[0031] By substituting the linearized tire force formula and the global kinematic equations into the vehicle dynamics equations, a complete control-guided nonlinear state-space model can be constructed. This model can capture the coupled effects of tire slippage on the vehicle's lateral and yaw motions, meeting the requirements of high-precision trajectory tracking tasks where tire coupling characteristics cannot be ignored.

[0032] S3. Construct a polynomial fuel consumption model based on transient dynamic characteristics. Based on the expected total traction power working mode corresponding to the current reference trajectory, linearize the polynomial fuel consumption model using the mode scheduling method to obtain partial derivative weights. Calculate the linearized fuel consumption cost using the partial derivative weights. The partial derivative weights include: velocity cost weights and acceleration cost weights.

[0033] Furthermore, the expression for the polynomial fuel consumption model is as follows: .

[0034] .

[0035] in, Transient fuel consumption rate; Longitudinal velocity; It is longitudinal acceleration; This represents the expected total traction power. Idle fuel consumption rate; , , , , , and All are polynomial fitting coefficients; For vehicle quality; It is the acceleration due to gravity; This is the rolling resistance coefficient; Road slope; This refers to the drag coefficient; For windward area; air density; This is the rotational mass conversion factor.

[0036] Furthermore, based on the expected total traction power operating mode corresponding to the current reference trajectory, the partial derivative weights are obtained after linearizing the polynomial fuel consumption model using the mode scheduling method, specifically including: 1) When the expected total traction power corresponding to the current reference trajectory is greater than 0, the target vehicle's operating mode is determined to be in the traction phase. The polynomial characteristic of the polynomial fuel consumption model is activated, and the first-order partial derivative is calculated at the operating point of the current reference trajectory to obtain the speed cost weight and acceleration cost weight, which are used as the partial derivative weight. At this time, the speed cost weight is... The acceleration cost weight is: ; Transient fuel consumption rate; Longitudinal velocity; This is the longitudinal acceleration.

[0037] 2) When the expected total traction power corresponding to the current reference trajectory is less than or equal to When the target vehicle is in coasting or braking mode, the fuel consumption rate is used as the idle constant, and the partial derivative weight is set to zero; at this time, the speed cost weight is... The acceleration cost weight is: .

[0038] Furthermore, the expression for the linearized fuel consumption cost is as follows: .

[0039] In the formula, The cost of linearizing fuel consumption; This is a weighting coefficient for fuel economy. For prediction in the time domain; To control the time domain; for The speed cost weight at any given moment; for The acceleration cost weight at each moment; for The longitudinal velocity at any given moment; for Longitudinal acceleration at any given moment.

[0040] In practical applications, the fuel consumption rate of the target vehicle varies significantly under traction, coasting, or braking conditions. Conventional fuel consumption models that involve state transitions introduce discrete transition conditions into the optimization problem, transforming it into a MIQP (Mixed Integer Quadratic Programming) problem, which fails to meet the real-time requirements of the vehicle controller.

[0041] In this embodiment, the reference trajectory feedforward mode scheduling method is mainly applied. First, based on the reference velocity and reference acceleration issued by the planning layer, and combined with the vehicle's longitudinal dynamics model, the expected total traction power required to maintain the reference trajectory is calculated. The formula for calculating the expected total traction power takes into account the instantaneous power required for the vehicle to overcome acceleration resistance, air resistance, rolling resistance, and gradient resistance, and can reflect the expected load pattern of the vehicle on the future reference trajectory.

[0042] When the reference trajectory indicates that the vehicle is in the traction phase (i.e.) At the linearization point, activate the polynomial fuel consumption model and at the linearization point... The partial derivatives of fuel consumption rate with respect to vehicle speed and acceleration are calculated to obtain the sensitivity coefficient for fuel consumption. and Conversely, when the reference trajectory indicates that the vehicle is in a coasting or braking phase (i.e., When the engine is idling, the fuel consumption rate is constant. At this time, the sensitivity coefficient is forcibly set to [value missing]. and .

[0043] In the subsequent construction of the QP objective function, the Hessian matrix, which determines the convexity of the problem, is entirely determined by the trajectory tracking weight matrix. and control smoothness weight matrix It is constitutive and positive definite; at the same time, the cost of controlling smoothness is... Control increment A quadratic penalty is applied. Therefore, even if the feedforward fuel gradient term changes abruptly during mode switching, the underlying actual control command can still remain smooth and continuous. This mode scheduling method not only perfectly transforms the complex transient fuel consumption dynamics into a time-varying linear function, completely avoiding the MIQP problem, but also theoretically guarantees the high stability of the control system.

[0044] Since the quadratic form matrix is ​​composed of a positive semi-definite error and control weight matrix, the objective function is a lossless convex function, ensuring the existence of a global optimal solution.

[0045] S4. Based on the linear discrete model, a standard convex quadratic programming objective function is constructed using the trajectory tracking error cost, the control increment smoothness cost, and the linearized fuel consumption cost; the standard convex quadratic programming objective function is a lossless convex function.

[0046] Furthermore, the expression for the standard convex quadratic programming objective function is as follows: .

[0047] .

[0048] In the formula, Let the objective function be a standard convex quadratic programming problem. To control the incremental sequence; The Hessian matrix; The gradient vector contains the gradient descent direction of the linearized fuel consumption cost; The objective function is a quadratic programming problem. This is to compensate for trajectory tracking errors; To control the cost of incremental smoothness; This comes at the cost of linearizing fuel consumption.

[0049] Furthermore, the trajectory tracking error cost is used to penalize the trajectory tracking error in the prediction time domain by using quadratic matrix weights; the control increment smoothness cost is used to suppress high-frequency jitter of the actuator in the control time domain by using quadratic weights to penalize the control increment.

[0050] S5. Under preset constraints, solve the standard convex quadratic programming objective function to obtain the optimal control increment sequence, generate energy-saving tracking control commands based on the optimal control increment sequence, and send the tracking control commands to the underlying drive-by-wire chassis of the target vehicle for execution; the energy-saving tracking control commands include: the optimal front wheel steering angle control command and the longitudinal acceleration control command at the current moment.

[0051] Furthermore, step S5 specifically includes: S51. Under preset constraints, a real-time convex optimization solver is used to solve the standard convex quadratic programming objective function to obtain the optimal control increment sequence.

[0052] Furthermore, the constraints specifically include: control input amplitude constraints and control increment constraints; the control input amplitude constraints are used to limit the maximum amplitude of the front wheel steering angle and the maximum amplitude of the longitudinal acceleration of the target vehicle; the control increment constraints are used to limit the rate of change of the front wheel steering angle and the longitudinal acceleration of the target vehicle.

[0053] Optionally, the expression for controlling the input amplitude constraint is as follows: .

[0054] The expression for the control increment constraint is as follows: .

[0055] Optionally, the real-time convex optimization solver includes at least the following: an effective set algorithm.

[0056] In practical applications, preset constraints are used to ensure that the calculated control commands are physically executable and safe; the model predictive controller must adhere to system constraints. Preset constraints include actuator amplitude saturation limits (i.e., upper and lower bounds of the front wheel steering angle and longitudinal acceleration) and actuator rate limits (i.e., limits on steering angular velocity and jerk). These preset constraints are then transformed into constraints concerning the control increment. After applying linear inequality constraints, the constrained standard QP problem is solved using real-time convex optimization solvers such as the effective set method.

[0057] S52. Take the first element of the optimal control increment sequence and add it to the actual control quantity of the previous moment to generate the optimal front wheel steering angle control command and longitudinal acceleration control command for the current moment.

[0058] S53. The optimal front wheel steering angle control command and longitudinal acceleration control command at the current moment are sent as tracking control commands to the underlying drive-by-wire chassis of the target vehicle for execution.

[0059] In practical applications, the vehicle trajectory tracking energy consumption optimization method based on the linearized fuel model in this embodiment is applied to the program or algorithm module of the autonomous driving domain controller or the on-board computing platform.

[0060] As an optional implementation method, such as Figure 3 As shown, this application also provides a closed-loop high-fidelity co-simulation platform architecture and data flow interaction structure. Figure 3As shown, the architecture mainly consists of a reference trajectory planning module, a model prediction control module, and a controlled vehicle model actual driving state module. The modules form a complete closed-loop control link through real-time data interaction.

[0061] 1) Reference Trajectory Planning Module: As the upper-level decision-making unit of the control system, the reference trajectory planning module is responsible for generating the global path and speed distribution of the vehicle's expected travel. The reference trajectory planning module outputs reference trajectory data to the model predictive control module in real time, serving as the benchmark target for trajectory tracking by the lower-level model predictive controller.

[0062] 2) Multi-objective optimizer module: Based on the derivation results of the internal prediction model, the multi-objective optimizer is responsible for solving the control actions. The optimizer internally constructs a standard convex quadratic programming objective function to comprehensively penalize and trade off trajectory tracking accuracy, actuator control smoothness, and transient fuel consumption costs.

[0063] 3) Internal Prediction Model Module; The internal prediction model module, together with the multi-objective optimizer, constitutes the main part of the model predictive control module. The internal prediction model mainly includes a vehicle dynamics prediction model and a polynomial fuel consumption model. This module receives reference trajectory data and vehicle state feedback, and uses a linear augmented prediction model to continuously extrapolate the evolution of the vehicle's future trajectory and transient energy consumption characteristics within a set prediction time domain.

[0064] 4) Controlled vehicle model actual driving state module; The controlled vehicle model, as the controlled object, receives control commands from the optimizer. In actual physical deployment scenarios, the controlled object is a real autonomous vehicle equipped with a drive-by-wire chassis system. Its underlying actuators perform mechanical actions according to the received commands and feed back the vehicle's actual physical driving state to the control module through onboard sensors (such as inertial navigation, wheel speedometers, etc.).

[0065] In an optional simulation embodiment for verifying the feasibility of the algorithm of the present invention, the controlled object can be represented by a high-fidelity vehicle dynamics model. For example, but not limited to, a digital twin model can be constructed using professional vehicle dynamics simulation software (such as CarSim, Adams, etc.), and dynamic parameters (such as vehicle mass, wheelbase, lateral stiffness, drag coefficient, etc.) corresponding to those of a conventional passenger car can be configured in it. The controlled vehicle model performs high-precision dynamic response calculations under a set complex test scenario (such as WLTC conditions), and feeds back the actual driving state of the vehicle to the model predictive control module at the end of each step. The model predictive control module can be compiled and deployed in an actual vehicle domain controller (such as an embedded MCU or computing platform based on ARM or x86 architecture), or it can be deployed in a simulation environment (such as Simulink, etc.) during the research and development testing phase. When deployed in a simulation environment, it can interact in real time with the simulation software that acts as the controlled vehicle model through standard communication interfaces (such as, but not limited to, the S-Function interface or the FMI / FMU standard interface), efficiently managing the flow of control commands and status feedback data in each control cycle, thereby building a reliable closed-loop verification environment.

[0066] In practical applications, for steps S4-S5, Figure 4 The internal algorithm architecture logic structure of the multi-objective quadratic programming controller is shown, such as... Figure 4 As shown, this architecture details the construction and solution process of the objective function in the aforementioned embodiments, and is the core of the algorithm that balances trajectory tracking, control smoothness, and energy consumption optimization. This architecture runs in real-time within each control cycle and specifically includes the following modules and detailed calculation steps: Step 1: Input the augmented state and prediction matrix.

[0067] To implement the algorithm in the digital controller and ensure smooth operation of the actuator, the continuous-time linear vehicle dynamics model needs to be transformed using the forward Euler method according to the sampling time. Discretize the data.

[0068] Unlike directly optimizing the absolute control input, this algorithm instead optimizes the control increment. This introduces an integral action and punishes high-frequency jitter in the actuator.

[0069] Subsequently, a discrete-time augmented state vector is constructed, which includes the state tracking error and the control input from the previous time step. .

[0070] Based on this, according to the set prediction time domain and control time domain By iterating through the system matrix, the future system prediction output matrix is ​​derived. and These matrices establish future prediction sequences. With future control increment sequence The explicit linear mapping relationship between them satisfies the following formula: .

[0071] Step 2: Control and track weight input.

[0072] The input is a positive semidefinite weight matrix used to adjust the system's multi-objective performance. Specifically, it includes a tracking weight matrix used to penalize system state deviations such as lateral error, heading error, and velocity error. And the control increment weight matrix used to penalize the increase in front wheel steering angle and longitudinal acceleration. .

[0073] Through dynamic adjustment and The values ​​of the diagonal elements allow for a flexible trade-off between tracking accuracy and ride comfort.

[0074] Step 3: Linearize the fuel gradient input.

[0075] To avoid compromising the convexity of the optimization problem by introducing discrete gear decisions or mode switching, a mode scheduling method based on reference trajectories is adopted to calculate the fuel gradient online.

[0076] When the total traction power corresponding to the reference trajectory At that time, the polynomial fuel characteristics were activated, and the transient fuel consumption rate was calculated separately. Regarding longitudinal velocity and longitudinal acceleration The partial derivatives are used to obtain the fuel sensitivity coefficient. and : .

[0077] .

[0078] when At this time, the engine is running at idle speed, and its fuel consumption is constant. The settings at this time are: .

[0079] By setting the gradient to zero, the controller avoids incorrectly penalizing necessary safe braking actions of the vehicle, thereby achieving energy saving while ensuring safety.

[0080] Step 4: Trajectory Tracking Calculation Unit.

[0081] The trajectory tracking calculation unit calculates the trajectory tracking cost in the prediction time domain based on the input prediction matrix and tracking weights. The controller's requirement is to eliminate tracking errors and ensure safe vehicle operation. Specifically, a state tracking error vector is defined. This error vector includes the vehicle's lateral error, heading error, and longitudinal velocity error, and satisfies the following formula: .

[0082] in, Indicates based on the current Moment information for the future The predicted value at any given time; This is a positive semi-definite weight matrix set to account for tracking error. The matrix is ​​adjusted... The size of the elements on the diagonal allows for flexible configuration of the penalty priority for various errors. For example, in high-speed cruising scenarios, the weight of lateral errors can be increased to ensure that the vehicle stays near the center line of the lane, thereby meeting the safety constraints of autonomous driving.

[0083] Step 5: Control smoothness calculation unit.

[0084] The control smoothness calculation unit is responsible for calculating the control time domain. Internal control smoothing cost Besides accurate trajectory tracking, the smoothness of control commands directly affects vehicle ride comfort and the lifespan of the underlying physical actuators. Therefore, defining a control increment vector... Used to characterize the increase in front wheel steering angle and longitudinal acceleration increment .

[0085] By applying a quadratic penalty to the control increment, the actuator's drastic movements are limited, satisfying the following formula: .

[0086] in, This is a semidefinite weighting matrix for control force. If acceleration or steering angle changes too rapidly, it not only degrades the passenger experience but also easily triggers additional transient fuel consumption. This is achieved through the matrix in this unit. Through adjustment, the controller can actively smooth the control signal and reserve a reasonable response time for the actuator, thus suppressing the high-frequency actuator jitter phenomenon from the source.

[0087] Step 6: Fuel Consumption Calculation Unit.

[0088] This unit is key to achieving energy consumption optimization. This is achieved by introducing a fuel weighting factor. By using time-varying partial derivatives as linear weights, an approximate transient fuel consumption cost function is constructed. : .

[0089] This transformation integrates transient fuel dynamics characteristics while preserving the convexity of the objective function.

[0090] Step 7: Constructing the objective function unit.

[0091] Given the stringent millisecond-level real-time computation requirements of automotive domain controllers (such as embedded MCUs), the comprehensive nonlinear objective function, encompassing trajectory tracking, control smoothing, and fuel consumption, must be transformed into a standard and convex quadratic programming form. The objective function construction unit performs algebraic transformations and fuses the three independent cost calculation modules to extract the matrix parameters required by the standard QP solver. This process includes two sub-steps: Standard quadratic form transformation and construction of positive definite Hessian matrices: The quadratic term is mainly generated by substituting the system state prediction equations of the trajectory tracking unit and the control smoothing unit. This is achieved by predicting the output equation... Substituting the cost components, the Hessian matrix of the objective function is constructed, satisfying the following formula: .

[0092] in, and Weight matrices and The block diagonal extended matrix in the prediction and control time domains. Since the tracking error weight matrix has been ensured during the algorithm design phase. and control increment weight matrix Given a positive semi-definite or positive definite matrix, the Hessian matrix is ​​constructed according to the matrix operation rules. Positive definiteness. This mathematical transformation process fundamentally guarantees that the objective function is a convex function, thereby ensuring that the optimization algorithm can find a unique global optimum within the feasible region, avoiding vehicle control divergence or high-frequency actuator jitter caused by local optima.

[0093] Multi-objective gradient vector fusion and dimensionality reduction: The embodiments in this specification generate a gradient vector by fusing state feedback with a fuel consumption sensitivity coefficient. It satisfies the following formula: .

[0094] In mathematical and engineering implementation, it is essential to ensure the alignment of the two terms on the right-hand side of the equation in terms of matrix dimensions and physical dimensions. This involves defining the future control increment sequence. The dimension is (in, The number of system control inputs, where (i.e., the increment of the front wheel steering angle and the increment of the longitudinal acceleration). Therefore, the first-order term obtained by expanding the trajectory tracking error prediction equation... The dimension is It is used to guide the vehicle to eliminate the current state deviation in the direction of fastest descent.

[0095] In the formula This is the fuel-sensitive gradient vector. Specifically, It is the transient fuel consumption cost function After the prediction model system matrix mapping, regarding the future control increment sequence The first-order partial derivative vector of is also mapped to the matrix dimension of . .

[0096] Step 8: Output and solve the objective function of the standard convex quadratic programming.

[0097] After constructing the standard cost function, the operational boundaries of the physical actuators also need to be introduced. Define the actuator amplitude constraint boundaries. and and actuator rate constraint boundary and Transform absolute control constraints into constraints on control increments. Linear inequality constraints: .

[0098] .

[0099] Finally, using efficient solvers such as the effective set algorithm, the solution is obtained in real time under the above linear constraints. Standard quadratic programming problem with independent variables The optimal control increment sequence obtained by solving It not only eliminates trajectory tracking errors, but also follows the energy-saving characteristics of the polynomial fuel model, achieving economical and smooth driving of the vehicle.

[0100] The technical effects of this application are as follows: 1) By using pattern scheduling and online partial derivative calculation, the nonlinear transient fuel model is linearized, maintaining the convex quadratic programming form of the objective function, avoiding mixed integer programming, and ensuring high-frequency real-time solution capability in the vehicle environment.

[0101] 2) Integrating the energy consumption model into a single-layer controller avoids conflicts between economy and tracking objectives, predicts and punishes high-energy-consuming transient behaviors, suppresses high-frequency jitter of actuators, and reduces overall fuel consumption.

[0102] 3) Even after introducing energy consumption optimization targets, the vehicle maintains stability, and the yaw rate response is highly consistent in extreme obstacle avoidance scenarios. The peak lateral tracking error is limited within the safety boundary, achieving the best balance between energy saving and safety performance.

[0103] This application introduces a fuel consumption model into the controller to predict high-energy-consumption transient behavior, solving the problem of non-convexity and large computational load in MPC optimization due to the introduction of the energy consumption model. This overcomes the limitation of traditional MPC controllers where energy consumption targets and tracking targets conflict, achieving energy saving and safety coordination while reducing overall fuel consumption and meeting the real-time control requirements of vehicles.

[0104] Example 2: This example provides a computer system, which can be a server or a terminal, and its internal structure diagram can be as follows. Figure 5 As shown, the computer system includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media to run. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the methods described above.

[0105] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer system to which the present application is applied. A specific computer system may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0106] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0107] The processors involved in the various embodiments provided in this application may be general-purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited thereto.

[0108] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0109] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model, wherein the vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model is applied to an MPC controller, characterized in that, The method includes: Obtain the current motion status information and reference trajectory of the target vehicle; A three-degree-of-freedom dynamic model of the target vehicle is constructed based on motion state information and reference trajectory, and the three-degree-of-freedom dynamic model is discretized to obtain a linear discrete model; the three-degree-of-freedom dynamic model is a dynamic model that includes longitudinal, lateral and yaw motions; the linear discrete model is used to predict the impact of control increments on the target vehicle state tracking error in the future finite time domain; A polynomial fuel consumption model is constructed based on transient dynamic characteristics. Based on the expected total traction power working mode corresponding to the current reference trajectory, the polynomial fuel consumption model is linearized using the mode scheduling method to obtain partial derivative weights. The linearized fuel consumption cost is then calculated using the partial derivative weights. The partial derivative weights include: velocity cost weights and acceleration cost weights. Based on the linear discrete model, a standard convex quadratic programming objective function is constructed using the trajectory tracking error cost, the control increment smoothness cost, and the linearized fuel consumption cost; the standard convex quadratic programming objective function is a lossless convex function. Under preset constraints, the optimal control increment sequence is obtained by solving the standard convex quadratic programming objective function. Based on the optimal control increment sequence, an energy-saving tracking control command is generated and sent to the underlying drive-by-wire chassis of the target vehicle for execution. The energy-saving tracking control command includes: the optimal front wheel steering angle control command and the longitudinal acceleration control command at the current moment.

2. The vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to claim 1, characterized in that, A three-degree-of-freedom dynamic model of the target vehicle is constructed based on motion state information and a reference trajectory. This three-degree-of-freedom dynamic model is then discretized to obtain a linear discrete model, specifically including: A three-degree-of-freedom dynamic model is constructed by using motion state information to characterize state variables and using a reference trajectory to characterize control variables. The state variables include: lateral velocity, longitudinal velocity, heading angle, yaw rate, global abscissa, and global ordinate. The control variables include: front wheel steering angle and longitudinal acceleration. At the current reference trajectory working point, the three-degree-of-freedom dynamic model is linearized by first-order Taylor expansion to obtain the linearized three-degree-of-freedom dynamic model. The forward Euler method is used to discretize the linearized three-degree-of-freedom dynamic model at the sampling time to obtain the initial discretized model; Based on the initial discretized model, the control input from the previous time step is added to the state variables to obtain the augmented state vector. The augmented state vector is then used to construct the augmented discrete state equation, which is then used as the output of the linear discrete model.

3. The vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to claim 1, characterized in that, The expression for the polynomial fuel consumption model is as follows: ； ； in, Transient fuel consumption rate; Longitudinal velocity; It is longitudinal acceleration; This represents the expected total traction power. Idle fuel consumption rate; , , , , , and All are polynomial fitting coefficients; For vehicle quality; It is the acceleration due to gravity; This is the rolling resistance coefficient; Road slope; This refers to the drag coefficient; For windward area; air density; This is the rotational mass conversion factor.

4. The vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to claim 1, characterized in that, Based on the expected total traction power operating mode corresponding to the current reference trajectory, the partial derivative weights are obtained after linearizing the polynomial fuel consumption model using the mode scheduling method, specifically including: When the expected total traction power corresponding to the current reference trajectory is greater than 0, the target vehicle's operating mode is determined to be in the traction phase. The polynomial characteristic of the polynomial fuel consumption model is activated, and the first-order partial derivative is calculated at the operating point of the current reference trajectory to obtain the speed cost weight and acceleration cost weight, which are used as the partial derivative weight. At this time, the speed cost weight is... The acceleration cost weight is: ; Transient fuel consumption rate; Longitudinal velocity; It is longitudinal acceleration; When the expected total traction power corresponding to the current reference trajectory is less than or equal to When the target vehicle is in coasting or braking mode, the fuel consumption rate is used as the idle constant, and the partial derivative weight is set to zero; at this time, the speed cost weight is... The acceleration cost weight is: .

5. The vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to claim 1, characterized in that, The expression for the linearized fuel consumption cost is as follows: ； In the formula, The cost of linearizing fuel consumption; This is a weighting coefficient for fuel economy. For prediction in the time domain; To control the time domain; for The speed cost weight at any given moment; for The acceleration cost weight at each moment; for The longitudinal velocity at any given moment; for Longitudinal acceleration at any given moment.

6. The vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to claim 1, characterized in that, The expression for the objective function of the standard convex quadratic programming is as follows: ； ； In the formula, Let the objective function be a standard convex quadratic programming problem. To control the incremental sequence; The Hessian matrix; The gradient vector contains the gradient descent direction of the linearized fuel consumption cost; The objective function is a quadratic programming problem. This is to compensate for trajectory tracking errors; To control the cost of incremental smoothness; This comes at the cost of linearizing fuel consumption.

7. The vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to claim 1, characterized in that, Under preset constraints, the optimal control increment sequence is obtained by solving the standard convex quadratic programming objective function. Based on the optimal control increment sequence, energy-saving tracking control commands are generated and sent to the underlying drive-by-wire chassis of the target vehicle for execution. Specifically, this includes: Under preset constraints, a real-time convex optimization solver is used to solve the standard convex quadratic programming objective function to obtain the optimal control increment sequence; The first element of the optimal control increment sequence is added to the actual control quantity at the previous moment to generate the optimal front wheel steering angle control command and longitudinal acceleration control command at the current moment. The optimal front wheel steering angle control command and longitudinal acceleration control command at the current moment are sent as tracking control commands to the underlying drive-by-wire chassis of the target vehicle for execution.

8. The vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to claim 1, characterized in that, The constraints specifically include: control input amplitude constraints and control increment constraints; the control input amplitude constraints are used to limit the maximum amplitude of the front wheel steering angle and the maximum amplitude of the longitudinal acceleration of the target vehicle; the control increment constraints are used to limit the rate of change of the front wheel steering angle and the longitudinal acceleration of the target vehicle.

9. The vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to claim 1, characterized in that, The trajectory tracking error cost is used to penalize the trajectory tracking error in the prediction time domain by using quadratic matrix weights; the control increment smoothness cost is used to suppress high-frequency jitter of the actuator in the control time domain by using quadratic weights to penalize the control increment.

10. A computer system, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the vehicle trajectory tracking energy consumption optimization method based on a linearized fuel model according to any one of claims 1-9.

Images