Intelligent carrying vehicle cooperative carrying control method in complex environment

Abstract

The application discloses a kind of intelligent carrying car collaborative handling control methods under complex environment, comprising: multi-source signal acquisition;Vehicle-load-communication coupling error is constructed, and the system discrete model needed for predictive control is established;Enhanced observation is constructed, and the offline prediction model package is obtained by bilinear Koopman learning;Get the model after projection, network compensation state and compensation path package;Model predictive control weight, constraint boundary, predetermined performance control boundary and control mode are obtained by scheduling;Communication state perception consensus Koopman model predictive control problem is constructed, and candidate control instruction and predicted state are obtained by solving;Executor fault diagnosis, fault-tolerant projection, security certificate judgment, smoothing and limiting amplitude processing are carried out, and execution instruction sequence is obtained.The application can still generate executable control instruction under the condition of communication state change, model residual, load connection constraint and executor degradation, and improve the reliability and adaptability of control.

Description

Technical Field

[0001] This invention relates to the field of intelligent vehicle technology, and more specifically to a collaborative transport control method for intelligent vehicles in complex environments. Background Technology

[0002] Intelligent transport platooning collaborative handling systems can be used in scenarios such as park logistics, warehouse handling, engineering transportation, special operations, and autonomous driving transportation. These systems typically consist of multiple mobile vehicles and a shared load, and need to operate collaboratively under conditions of limited communication, varying path curvature, load connection constraints, and changes in actuator state.

[0003] Existing collaborative transport control systems mostly rely on preset dynamic models, centralized controllers, or ideal communication conditions. In actual operation, there is nonlinear coupling between vehicles and loads, and communication links may experience delays, packet loss, bias, or noise. Actuators may also experience efficiency degradation. Fixed-weight model predictive control, ordinary Koopman predictive models, or control methods that only schedule based on error amplitude are difficult to simultaneously consider path tracking, connection maintenance, load force constraints, and control executability under degraded conditions.

[0004] Specifically, existing technologies have at least the following shortcomings: First, when multiple vehicles jointly transport loads, it is difficult to unify vehicle status, load status, connection error, communication status, and actuator efficiency into the same predictive control framework; Second, ordinary linear predictive models are difficult to characterize the multiplicative coupling between control input, load connection error, and vehicle lateral motion, and may also cause short-time domain prediction amplification during online updates; Third, communication latency, packet loss, link quality degradation, path curvature changes, and actuator efficiency degradation are difficult to synchronously convert into weights, constraints, and control mode scheduling quantities; Fourth, candidate control commands may not meet the feasible input range of actuators, load connection constraints, or safety certificate conditions under degraded conditions.

[0005] Therefore, a collaborative transport control method is needed that can still generate executable control commands under conditions of communication state changes, model residuals, load connection constraints, and actuator degradation. Summary of the Invention

[0006] In view of this, the present invention provides a collaborative transport control method for intelligent vehicles in complex environments, which can still generate executable control commands under conditions of communication state changes, model residuals, load connection constraints and actuator degradation, thereby improving the reliability and adaptability of control.

[0007] A collaborative transport control method for intelligent vehicles in complex environments includes: Step S1: Multi-source signal acquisition to obtain the desired path, temporary desired path packet, vehicle status, load status, connection status, communication status, and actuator status; Step S2: Based on the multi-source signals output in step S1, construct the vehicle-load-communication coupling error, establish the discrete system model required for predictive control, and then output the complete dataset. Step S3: Construct augmented observations based on the complete dataset output in step S2, perform bilinear Koopman learning, and obtain the offline prediction model package; Step S4: Based on the offline prediction model package output in step S3, perform input-aware stable projection on the input-related Koopman effective operators, and perform prediction compensation for delayed or lost information to obtain the projected model, network compensation state, and compensation path package. Step S5: Based on the communication status, network compensation status, and compensation path packet output in step S4, and combined with the desired path, connection status, and actuator status, the model predictive control weights, constraint boundaries, predetermined performance control boundaries, and control modes are scheduled. Step S6: Based on the projected model and network compensation state output in step S4, and the model prediction control weights, constraint boundaries, predetermined performance control boundaries and control modes output in step S5, construct the communication state-aware consensus Koopman model prediction control problem, and solve it to obtain candidate control commands and prediction states. Step S7: Based on the candidate control commands and predicted states output in step S6, perform actuator fault diagnosis, fault-tolerant projection, safety certificate judgment, smoothing and limiting processing, and send the execution command sequence to the actuators of each actual vehicle to enable each actual vehicle to perform collaborative transport and generate an operation log.

[0008] The intelligent vehicle cooperative handling control method in complex environments provided by the present invention has the following beneficial effects: (1) This invention obtains the desired path, temporary desired path package, vehicle status, load status, connection status, communication status and actuator status through multi-source signal acquisition, and constructs vehicle-load-communication coupling error based on these multi-source signals, including load center tracking error, vehicle-load connection error, adjacent vehicle distance error and communication degradation variable compensation amount into the error model, so that vehicle status, load status, connection error, communication link status and actuator efficiency are unified into the same data object framework, and path tracking, connection maintenance, communication compensation and execution safety can be uniformly described and processed in the same control flow, solving the problem that when multiple vehicles jointly transport loads, vehicle status, load status, connection error, communication status and actuator efficiency are difficult to be uniformly incorporated into the same predictive control framework.

[0009] (2) This invention constructs an enhanced observation and establishes a bilinear Koopman prediction model. The bilinear term can characterize the multiplicative coupling relationship between the control input and the boost state. Furthermore, by performing input-sensory stable projection on the input-related Koopman effective operator, the main matrix is ​​projected and the bilinear input matrix is ​​scaled to ensure that the spectral radius of the input-related effective operator meets the corresponding conditions. This constrains the short-time domain prediction amplification effect and solves the problem that ordinary linear prediction models are difficult to characterize the multiplicative coupling between control input, load connection error and vehicle lateral motion, and may also cause short-time domain prediction amplification during online updates.

[0010] (3) This invention obtains network compensation state and compensation path packet by predicting and compensating for delay or loss information, and transforms communication state, expected path, connection state and actuator state into scheduling variables of model predictive control weights, constraint boundaries, predetermined performance control boundaries and control modes. This allows communication delay, packet loss, link quality degradation, path curvature change and actuator efficiency degradation to participate in the construction of model predictive control problem in a synchronized manner, solving the problem that communication delay, packet loss, link quality degradation, path curvature change and actuator efficiency degradation are difficult to be simultaneously transformed into weights, constraints and control mode scheduling variables.

[0011] (4) This invention predicts the control problem by constructing a communication state-aware consensus Koopman model, solves the candidate control commands, and performs actuator fault diagnosis (including exponential weighted moving average statistics and cumulative sum statistics), fault-tolerant projection, safety certificate judgment, smoothing and limiting processing. It also introduces load force proxy quantities to participate in constraints and audit items, so that the control commands under degraded conditions meet the feasible input range of the actuator, load connection constraints and safety certificate conditions. This solves the problem that the candidate control commands may not meet the feasible input range of the actuator, load connection constraints or safety certificate conditions under degraded conditions. Finally, it achieves the technical effect of generating collaborative transport control commands that can be executed for actual vehicles in complex environments, and improves the reliability and adaptability of the control. Attached Figure Description

[0012] Figure 1 A flowchart illustrating a collaborative handling control method for intelligent transport vehicles in complex environments; Figure 2 The image shows the validation results of the bilinear Koopman model and the input-sensory stable projection. Figure 3 Figure showing the verification results of hairpin bend tracking under communication / execution degradation conditions; Figure 4 The diagram shows the verification results of connection retention and load force protection. Detailed Implementation

[0013] 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 embodiments of the present invention, and should not be construed as limiting the present invention.

[0014] Please see Figure 1 The present invention provides an intelligent vehicle collaborative handling control method in complex environments, the method comprising steps S1 to S7: Step S1: Multi-source signal acquisition to obtain the desired path, temporary desired path packet, vehicle status, load status, connection status, communication status, and actuator status.

[0015] In step S1, the collaborative handling system includes multiple mobile vehicles and a common load, and the set of vehicle nodes is denoted as . Vehicle number is , The sampling period; Data sets collected or calculated at any time Defined as:

[0016] in, For global reference path information, The upper-level planning nodes are distributed to the first... The short-term temporary expected path packet of the vehicle. For the first Vehicle global status, For the first Vehicle path coordinate error status, For the first Vehicle control input, At the load center state, For the first Connection error between the vehicle and load connection points For the set of communication link states, For actuator efficiency estimation, This is a set of system parameters.

[0017] For the collection of expected path and temporary path packets, the global reference path information is represented as follows:

[0018] in, and These are the x and y coordinates of the reference path, respectively. For reference heading angle, For reference curvature, For reference velocity profile, This is the path arc length parameter.

[0019] In a preferred embodiment, the upper-level collaborative planning node generates a corresponding node based on the load center reference progress, reference curvature, and the installation position of each vehicle relative to the load center. The vehicle's temporary expected path packet:

[0020] in, This is a positional function within the path packet. For the path packet heading angle, For path envelope curvature, This refers to the local arc length parameter within the path packet. The path packet length. The temporary expected path packet can be generated from an equivalent four-wheel steering geometry model, a load rigid body velocity distribution model, a virtual structure model, or other load center trajectory decomposition model to provide a local geometric reference.

[0021] For vehicle status and control input acquisition, the first The vehicle's global state, path coordinate error state, and control input are represented as follows:

[0022]

[0023]

[0024] in, , For the first Global plane coordinates of the vehicle For the first Vehicle heading angle, and They are respectively Time of the first The longitudinal and lateral speeds of the vehicle for Time of the first yaw rate of the car Indicates transpose. For the first Vehicle path projection progress, For lateral error, For heading error, For longitudinal acceleration, This is the steering angle command.

[0025] In differential drive, omnidirectional sliding, or wheel-side independent drive platforms It can be replaced with linear velocity, angular velocity, wheel end torque, or wheel speed commands.

[0026] For load status and connection status acquisition, the load center status is represented as follows:

[0027] in, , For the global coordinates of the load center, For the load heading angle; No. The fixed position vector of the connection point between the vehicle and the load in the load coordinate system is: The position of the connection point in the global coordinate system for:

[0028] in, The load center position vector, The rotation matrix is ​​constructed from the load heading angle. , They are respectively Two components in the load coordinate system; No. Connection error between vehicle and load connection point Defined as:

[0029] in, For the first The car is The position vector at time , ; The connection structure can be equivalent to a spring-damped connection, and the load surcharge is:

[0030] in, For the first Load-bearing capacity at vehicle connection point The equivalent stiffness matrix, The equivalent damping matrix, This is the first derivative of the connection error with respect to time. This force surrogate is used for control costs, constraints, or security audits; in actual system configurations with connection force sensors, it can be replaced or corrected by measured force.

[0031] For communication link status acquisition, the communication graph is defined as follows: ,in, For upper-level planning nodes, For a moment A set of directed communication links; for links The communication status is:

[0032] in, As the information sending node, As an information receiving node, for Link quality at any given moment For communication delay, For packet loss indication, For communication bias, For communication noise, superscript For communication-related quantities; Overall network quality is defined as:

[0033] in, For overall network quality, the superscript "net" represents the network aggregation quantity. This is a link quality aggregation function, which can take the average value, weighted average value, minimum value, quantile, or its smoothed form.

[0034] For actuator status acquisition, the actuator efficiency is estimated as follows:

[0035] in, For longitudinal acceleration channel efficiency, For steering channel efficiency. In other embodiments, actuator efficiency estimates can also be set separately for braking, wheel-end torque, wheel speed, or differential drive channels.

[0036] Step S2: Based on the multi-source signals output in step S1, construct the vehicle-load-communication coupling error, establish the discrete system model required for predictive control, and then output the complete dataset.

[0037] In step S2, the multi-source data from step S1 is transformed into error variables and coupled state variables that can be used for predictive control.

[0038] Specifically, regarding vehicle kinematics and path coordinate errors, based on the output of step S1... , No. The global kinematics of the car are represented as follows:

[0039] After the vehicle is projected onto the reference path, the path coordinate error satisfies:

[0040] in, , , , , , They represent the first global X-axis coordinates of the car Global Y-axis coordinates Heading angle Path projection progress lateral error Heading error The first derivative with respect to time, and The first The longitudinal and lateral speeds of the vehicle For the first The yaw rate of the car, For reference path in The curvature at that point.

[0041] For load center error and connection error, load center tracking error is defined as:

[0042] in, For load center tracking error, , , These are the reference x-coordinate, reference y-coordinate, and reference heading angle of the load center, respectively, with the superscript ref indicating a reference value; The connection error is concatenated as follows:

[0043] in, This is a spliced ​​vector of vehicle-load connection error. To concatenate by column; The distance error between adjacent vehicles is defined as:

[0044] in, For link The corresponding distance error between adjacent vehicles, superscript For distance channels, The distance between adjacent vehicles after compensation. For reference to adjacent distances, This is the current reference path progress.

[0045] Communication degradation affects the distance between adjacent vehicles, the load center error, and the upper-layer temporary expected path packet. The communication degradation variables are incorporated into the error model as follows:

[0046]

[0047]

[0048] in, for The distance between adjacent vehicles measured at any time. Due to the delay in the information channel, This is for offsetting the distance information channel. This refers to communication noise in the distance information channel; To compensate for the lateral error of the load. for Lateral load error at time t, , and These are the time delay, offset, and communication noise of the load lateral error information channel, respectively. For the compensated temporary expected path packet, for Temporary expected path packet at time, , and These represent the latency, offset, and communication noise of the path packet information channel, respectively.

[0049] It should be noted that if the corresponding data packet is lost, it will be reconstructed using the prediction compensation method in step S4.

[0050] By integrating vehicle kinematics, load connectivity, communication degradation, and actuator states, the expression for the predictive control state-space model is obtained:

[0051]

[0052]

[0053] in, The system state at the next discrete time step. This represents the current system state. This is the current system control input. This is the system state update function formed by the combined effects of vehicle kinematics, load connectivity, and communication degradation. The system disturbance vector includes modeling residuals, communication bias, communication noise, load coupling disturbances, and actuator implementation errors.

[0054] Finally, step S2 is obtained at time [time]. The complete set of output data for: .

[0055] Step S3: Construct enhanced observations based on the complete dataset output in step S2, perform bilinear Koopman learning, and obtain the offline prediction model package.

[0056] In step S3, a short-time domain prediction model is obtained that can be embedded in model predictive control and is subject to stable projection audit.

[0057] Specifically, based on the output of step S2 Define enhanced observations :

[0058] in, To enhance the observation, The reference input or reference feature vector may include reference curvature, reference heading, path progress, local path preview points, and high curvature window markers. In one embodiment, the observation dimension is enhanced. Desirable Increase the state dimension Desirable The values ​​mentioned above are for illustrative purposes only and do not constitute necessary limitations.

[0059] Define promotion mapping:

[0060] in, for Constantly improving one's state. The parameter is Koopman lifting mapping, It is determined by parameters Determined nonlinear feature mapping, To improve the mapping parameters. It can be implemented using neural networks, radial basis functions, multinomial dictionaries, Fourier basis functions, or combinations thereof. If a neural network is used, the number of network layers, nodes, and activation functions can be set according to the actual system size.

[0061] To describe the multiplicative coupling between steering input, longitudinal acceleration input, vehicle lateral motion, connection error, and load state, a bilinear Koopman prediction model is established:

[0062] in, for Constantly improving one's state. To improve the state master matrix, To control the input matrix, For the first A bilinear input matrix, For the summation index of the input channel, To control the input dimension, For a moment The first of the overall control input vectors Each scalar input channel component This is for predicting residuals in one step.

[0063] The offline dataset is defined as:

[0064] in, This is an offline dataset, with the subscript "off" indicating offline. The offline data covers at least one of the following operating conditions: 1. Normal operation without communication degradation; 2. High communication noise or high packet loss; 3. Single actuator efficiency degradation; 4. Mixed operating conditions where communication degradation and actuator degradation coexist; 5. Operating conditions with different reference curvatures and different initial lateral errors.

[0065] The Koopman regression vector is:

[0066] in, For the Koopman regression vector, the subscripts are... For Koopman, and Representing time respectively Overall control input vector The first and the second Each scalar input channel component; concatenates the regression vectors of multiple sampling points column-wise into a regression matrix. .

[0067] Will The Koopman matrix to be identified is obtained through regularized least squares or ridge regression:

[0068] in, To improve the state matrix for the next time step in the training samples, The regression matrix is ​​composed of regression vectors; This is the regularization coefficient, used to suppress parameter amplification caused by small samples, multicollinearity, and noise. It is the identity matrix. The identification results are obtained through optimization or regression.

[0069] Simultaneously record the predicted residual boundary at one step:

[0070] in, For vector norm, For constant residual boundary parameters, For residual boundary parameters related to enhanced observation bias, For residual boundary parameters related to network quality degradation, To enhance the reference value of the observation.

[0071] Final output: offline prediction model package :

[0072] in, The offline prediction model package output from step S3. This is a projection operator for reconstructing the system state from the improved state.

[0073] Step S4: Based on the offline prediction model package output in step S3, perform input-aware stable projection on the input-related Koopman effective operators, and perform prediction compensation for delayed or lost information to obtain the projected model, network compensation state, and compensation path package.

[0074] In step S4, the online residual update is constrained within the stable projection constraint, so that the bilinear Koopman model satisfies the prediction amplification constraint within the control input envelope.

[0075] Specifically, based on the output of step S3 Given a control input At that time, the input-related valid operators for lifting the state are:

[0076] in, To control input The input-related valid operators are defined below, with the subscript eff indicating the valid operators.

[0077] Only for the main matrix Spectral radius constraints cannot cover bilinear input terms, therefore it is necessary to... Perform input-aware stable projection.

[0078] Project the principal matrix as:

[0079] in, The projected principal matrix, with superscript... The quantity after stable projection. For For a stable projection operator of projection parameters, These are the projection constraint parameters of the main matrix; Then the bilinear input matrix is ​​scaled:

[0080] in, For the projected or scaled first A bilinear input matrix, For the first The scaling factor of a bilinear input matrix. ; Constructing the input sample set The input sample set includes at least one of training input quantiles, control input boundary points, and multi-input dimension corner points, making the input sample set... The input-related valid operators within satisfy:

[0081] in, To obtain the maximum value, The spectral radius of the matrix is... for The elements in To provide the upper bound of the stability radius of the relevant effective operators, For the first input sample Components of each input channel; Alternatively, it satisfies the conservative norm constraint under the continuous input envelope:

[0082] in, It is a 2-norm. For the first The upper bound of the absolute value of each input channel, with the superscript abs indicating the absolute value boundary.

[0083] If during online operation, the residual of the sliding window is used to... , or If an update is performed, the updated model should undergo the above stable projection again to obtain the projected model after the online update:

[0084] in, For a moment The projected model, , and They are time points After projection, the principal matrix, the input matrix, and the first... A bilinear input matrix.

[0085] When the vehicle At any moment Received from vehicle When dealing with delayed state data, the delay steps are: The enhanced state is constructed by delaying the reception time of the enhanced observations, and the projected model is used to... Predicted to Using the projected model, we obtain:

[0086] in, For from The network compensation status, Based on the receiving time and status information channel delay The compensation received improved the status; When the vehicle When state data packets are lost, reconstruction is performed using the previous valid network compensation state or the projected model prediction state. The specific choice can be determined based on the length of consecutive packet loss, link quality, and computing resources. Simultaneously, temporary expected path packets are time-aligned, the previous valid path packet is preserved, or reference schedule is shrunk.

[0087] Specifically, the upper-layer temporary expected path packet If delays or data loss occur, the following methods can be used to handle the situation: 1. When a path packet is delayed but still valid, time alignment is performed based on the path progress difference; 2. When a path packet is lost for a short period of time, the previous valid path packet is used and the progress is extrapolated by combining the load center prediction. 3. When there is continuous packet loss or the link quality is below the threshold, the degradation path maintenance or reference progress contraction strategy is adopted.

[0088] The final data set output by step S4 for:

[0089] in, For link At any moment Smooth link quality, and These are link packet loss indication and communication delay, respectively.

[0090] Step S5: Based on the communication status, network compensation status, and compensation path packet output in step S4, and combined with the desired path, connection status, and actuator status, the model predictive control weights, constraint boundaries, predetermined performance control boundaries, and control modes are scheduled.

[0091] In step S5, the model predicts the control weight matrix, constraint boundary, predetermined performance control boundary and control mode, and uses the communication state, desired path, connection state and actuator state as scheduling variables.

[0092] Specifically, based on the communication status output in step S4, the link... Define smooth link quality:

[0093] in, This is the link quality smoothing coefficient. For link At any moment Smooth link quality; Neighbor consensus weight Defined as:

[0094] in, and These are the minimum and maximum consensus weights, respectively.

[0095] The higher the link quality, the stronger the cooperative constraint effect between adjacent vehicles.

[0096] When the overall network quality degrades, the upper bound of the connection error and the input constraint are defined as follows:

[0097]

[0098] in, This represents the upper bound of the connection error that varies with the overall network quality. This is the upper bound of the minimum connection error. This is the upper bound of the connection error under preset conditions (specifically, when the network quality is good). For a moment The set of input constraints, As the baseline input constraint set, To constrain the tightening amount, To constrain and tighten operations.

[0099] The predetermined performance boundary for the lateral error is defined as follows:

[0100] in, For a moment The lateral error is predetermined performance boundary. As the initial boundary, For steady-state boundary, This is the boundary attenuation coefficient; The normalization error ratio is:

[0101] in, For the first Normalized ratio of vehicle lateral error To prevent small positive numbers with a denominator of zero; when When the value approaches 1, the predetermined performance control protection is activated, increasing the weight of the lateral error or tightening the soft constraint of the lateral error.

[0102] Control Mode Defined as:

[0103] in, For control mode, For normal tracking and connection maintenance mode, For high curvature lateral priority mode, For load protection mode, This refers to a conservative fallback mode under communication or actuator degradation. The above modes correspond to weighted scheduling, constraint scheduling, and execution layer strategies in the same consensus model predictive control problem. If multiple triggering conditions occur simultaneously, the priority is as follows: .

[0104] When the reference curvature When the threshold is exceeded or the vehicle's lateral error approaches the predetermined performance control boundary, the weights of lateral error and heading error are increased. The weights in the high curvature lateral priority mode are:

[0105]

[0106] in, This is the vehicle error weight matrix in the high-curvature lateral priority mode, where the superscript hc indicates high-curvature lateral priority. For the weight scaling matrix, This is the vehicle error baseline weight matrix. It is a diagonal matrix. and These are the lateral error weighting amplification factor and the heading error weighting amplification factor, respectively. , The aforementioned weight amplification factor is used for weight scheduling in high curvature mode and does not represent actuator efficiency. Simultaneously, the steering penalty or steering smoothing penalty can be appropriately reduced to improve the lateral correction capability in high curvature road sections.

[0107] Load protection strength Defined as:

[0108] in, To limit the input to Amplitude limiting function between, To protect the exit curvature threshold, To protect the fully activated curvature threshold, This is the curvature value of the reference path at the projection point of the vehicle's current position.

[0109] In load protection mode, the cost weights for connection errors, input dispersion, and load stress proxy are increased. The load protection item in load protection mode is:

[0110]

[0111] in, For the objective function term under load protection mode, The original objective function is... , and These are the weighting coefficients for connection error, input dispersion, and load force surcharge, respectively. This is the average value of the control input for each vehicle. For the load plane force surcharge, For the number of vehicles; The final data set output by step S5 for:

[0112] in, , , These are the set of state weights, input weights, and state constraints, respectively.

[0113] Step S6: Based on the projected model and network compensation state output in step S4, and the model prediction control weights, constraint boundaries, predetermined performance control boundaries, and control modes output in step S5, construct the communication state-aware consensus Koopman model predictive control problem, and solve it to obtain candidate control commands and predicted states.

[0114] In step S6, a unified communication state-aware consensus model prediction and control problem is constructed based on the projected model, network compensation state, model prediction and control weights, constraint boundaries, predetermined performance control boundaries, and control modes.

[0115] Specifically, based on the projected model output in step S4 and the output of step S5 Define the prediction time domain length as The optimized decision vector is:

[0116] in, This represents the optimal decision vector corresponding to the overall control sequence. To predict the first element in the control sequence, For at any time The predicted first Constantly control the input. For at any time The predicted first Constantly control input; Improved state prediction satisfies:

[0117] in, To predict the state of improvement, To improve the state for the next prediction step, For at any time The predicted first Constantly control the input. This refers to the prediction step number; For time-based Information, predictions The first moment in the overall control vector at time step One scalar channel component; Predictive control state is determined by Obtain, among which, To predict the control state; The state-aware consensus Koopman model predicts the control objective function as follows:

[0118] in, The model predicts the control objective function. For the predicted load center tracking error, It is a weighted quadratic form. This is the load error weight matrix. For the first Predicting vehicle error This is the vehicle error weight matrix. For distance error consensus weights, The predicted vehicle status for the information receiving node. The predicted state of vehicle network compensation from the information sending node. The consensus error weight matrix is... The error in the predicted distance between adjacent vehicles, This is the spliced ​​vector of the predicted vehicle-load connection error. To connect the error weight matrix, For the predicted control input vector, For the input weight matrix, To predict changes in input, For the input rate of change weight matrix, For soft-constrained slack variables, This is the slack variable weight matrix; For time-based Information predicted at time The load center tracking error; This is the terminal load error weight matrix; Load protection strength It will take effect at that time.

[0119] The constraints include speed constraints, steering and longitudinal input constraints, input rate of change constraints, connection error constraints, load force surrogate constraints, and predetermined performance control soft constraints.

[0120]

[0121]

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128] in, For the first The car is Predicted longitudinal velocity at time [time] and These are the lower and upper bounds of the longitudinal velocity, respectively. For the first The car is Predicted steering angle command at any time For a moment The upper limit of the steering angle, For the first The car is Predicted longitudinal acceleration command at time of moment For a moment The upper limit of longitudinal acceleration, For the first Predicting the change in steering angle of the vehicle. This is the upper bound of the change in steering angle. For the first Predicting the change in longitudinal acceleration of the vehicle. This represents the upper bound of the change in longitudinal acceleration. For the first The car is Predicted connection error at time step To improve overall network quality The upper bound of the changing connection error For the first The car is Predicted load surcharge at time of day This is the upper bound of the load-bearing force surcharge. For a moment The lateral error is predetermined performance boundary. For the first The lateral error of the vehicle is a soft constraint relaxation variable.

[0129] After solving, the first control input in the predicted sequence is taken as the candidate control command:

[0130] in, These are candidate control commands obtained after solving the model predictive control algorithm. The superscript mpc represents the model predictive control quantity. is the first control input in the optimal prediction sequence, and the superscript opt ​​represents the optimal solution.

[0131] The final data set output by step S6 for:

[0132] in, The data set output in step S6 This is the optimal overall control sequence. , , These are the one-step predicted state, the one-step predicted connection error, and the one-step predicted load force surrogate, respectively.

[0133] Step S7: Based on the candidate control commands and predicted states output in step S6, perform actuator fault diagnosis, fault-tolerant projection, safety certificate judgment, smoothing and limiting processing, and send the execution command sequence to the actuators of each actual vehicle to enable each actual vehicle to perform collaborative transport and generate an operation log.

[0134] In step S7, the candidate control commands from step S6 are converted into execution commands that can be issued to the actual vehicle actuators, and the predicted control output of the master model is subject to bounded correction, projection, and degradation protection.

[0135] Specifically, based on the output of step S6 , for the The vehicle performs actuator fault diagnosis and command correction to obtain the actual execution input for issuing to the actual vehicle actuators. The actual execution input is represented as follows:

[0136] in, For the first The vehicle actually executes the input. For the first The vehicle issues control commands. For element-wise multiplication, The execution error is bounded. Based on the monitoring data constructed from the one-step predicted residual, the one-step predicted residual is:

[0137] in, To predict residuals in one step, For the first The car at any time The system status, The first time obtained based on the information from the previous time step The vehicle's current predicted status; The exponentially weighted moving average statistic is:

[0138] in, For the first The car is The exponentially weighted moving average statistic at time point. For the first The car is The exponentially weighted moving average statistic at time point. The residual statistical smoothing coefficient; The cumulative sum statistic is:

[0139] in, For the first The car is Accumulation and statistics over time. For the first The car is Accumulation and statistics over time. This is the statistical threshold for fault diagnosis.

[0140] when , If the actuator efficiency estimate is below a threshold, the corresponding vehicle or control channel is determined to be degraded. To reduce the probability of misjudgment due to communication noise, preheating count, continuous trigger count, and release hold count can be set.

[0141] When a decrease in actuator efficiency, degradation in communication status, or a deterioration in link quality is detected, candidate control commands are projected from the conservative set of feasible inputs:

[0142] in, The overall control commands after safe projection. Let the set of conservative feasible inputs be determined by actuator efficiency estimates and network quality. For system-level actuator efficiency estimation, For the safe projection weight matrix, This is the connection error penalty coefficient. For a given control input The projection prioritizes maintaining the main structure of the candidate control commands while reducing the control contribution of degraded vehicles or degraded channels.

[0143] Define the comprehensive error for security certificates:

[0144] in, For safety certificates, comprehensive error is used. To improve state prediction error; This embodiment uses security certificates for online decision-making and engineering auditing. The certificate function is:

[0145] in, For control mode The certificate function below, For control mode The corresponding positive definite weight matrix, Let be a vector whose elements are all 1s. It is the first The actuator efficiency estimation vector after filtering or observation. For the first The efficiency deviation weight matrix of the vehicle actuator. This is the link quality penalty coefficient; The enhanced certificate function after adding the load force proxy audit item is:

[0146] in, To enhance the certificate function, As the audit weight for load-bearing surcharge; The one-step certificate margin is:

[0147] in, For one-step certificate margin, For control mode The corresponding certificate shrinkage rate, For system disturbance weighting coefficients, To predict the residual weighting coefficients, Enhance the certificate function for the next time step; when When the candidate security instruction is deemed to meet the certificate requirements, among which... The certificate margin allowance is set; otherwise, at least one of the following commands is executed: control command projection, constraint tightening, reference progress contraction, degradation rollback mode switching, holding the previous safety control command, braking command, or preset minimum risk command.

[0148] Smooth out safety control commands that meet safety conditions:

[0149] in, For the first The vehicle issues control commands at the current moment. To control the smoothing coefficient of the instructions, , For the first The car is at the moment Safety projection control commands, For the first Control commands are issued from the vehicle at any time; Apply amplitude and rate of change constraints:

[0150]

[0151] in, and The first The car is at the moment and time The issued steering angle command This is the upper bound of the rate of change of steering angle. and The respective The car is at the moment and time The issued longitudinal acceleration command, This is the upper bound of the longitudinal acceleration rate of change; Will meet safety conditions The system is distributed to the steering, drive, or braking actuators of each actual vehicle, enabling the vehicles to perform coordinated transport and movement.

[0152] Final output data set It then issues a sequence of execution instructions to enable each actual vehicle to perform coordinated transport operations and generates an operation log, in which... The expression is:

[0153] in, The data set output and distributed in step S7 This is the set of load surrogate quantities at the connection point.

[0154] Validation Case In a verification case, a simulation platform using four intelligent transport vehicles collaboratively transporting a shared load was employed to verify the method of this invention. The verification scenarios included hairpin bend paths, communication delays, communication quality variations, and actuator degradation disturbances. For clarity, "degradation control method" refers to a control implementation that does not employ the communication state-aware scheduling, predictive compensation, and fault-tolerant safe execution of this invention; "no load protection" refers to a control implementation that does not enable load force proxy constraints and load force audit items. RMSE is the root mean square error, and RMS is the root mean square quantity.

[0155] Please see Figure 2 The maximum radius, 95th percentile, and average value of the input relevant Koopman effective operator before stable projection are approximately 1.356, 1.174, and 1.035, respectively, and after stable projection, they are approximately 0.998, 0.993, and 0.992, respectively, all compressed to near the constraint threshold. Under the hairpin bend condition, the maximum point-state error increments of the unstable projection, linear Koopman, and original linear models relative to the present invention are approximately 0.023m, 0.183m, and 0.734m, respectively. These verification results demonstrate that steps S3 and S4 of the present invention can constrain the short-time domain amplification of the input relevant effective operator and support the use of the bilinear Koopman prediction model in complex paths.

[0156] Please see Figure 3 Under communication / execution degradation conditions, the lateral RMSE of the degradation control method is approximately 0.299m, and the maximum lateral error is approximately 0.587m. After applying the method of this invention, the lateral RMSE is approximately 0.085m, and the maximum lateral error is approximately 0.180m. Furthermore, the method of this invention outperforms the control method in approximately 85.4% of path points. This verification result demonstrates that the prediction compensation in step S4, the communication state-aware scheduling in step S5, and the consensus Koopman model predictive control in step S6 of this invention can improve path tracking under high curvature paths and degradation disturbances.

[0157] Please see Figure 4In the verification of connection retention and load stress protection, the maximum connection utilization rate of the degradation control method is approximately 1.000, the maximum connection utilization rate without load protection is approximately 0.778, and the maximum connection utilization rate after enabling the load protection of the present invention is approximately 0.307. The RMS load stress surcharges of the degradation control method, the implementation without load protection, and the load protection of the present invention are approximately 2.37 kN, 2.59 kN, and 1.78 kN, respectively. These verification results demonstrate that by incorporating connection status, load stress surcharge, and actuator status into the constraint, certificate judgment, and smoothing limiting processes in steps S5 to S7 of the present invention, the occupancy of connection constraints and the load stress surcharge can be reduced.

[0158] It should be noted that the above verification cases are only used to illustrate the technical effect of the method of the present invention in one implementation scenario, and do not constitute a limitation on the number of vehicles, path type, sampling period, controller parameters or fault type.

[0159] In summary, the intelligent vehicle cooperative handling control method in complex environments according to the above embodiments has the following beneficial effects: (1) This invention obtains the desired path, temporary desired path package, vehicle status, load status, connection status, communication status and actuator status through multi-source signal acquisition, and constructs vehicle-load-communication coupling error based on these multi-source signals, including load center tracking error, vehicle-load connection error, adjacent vehicle distance error and communication degradation variable compensation amount into the error model, so that vehicle status, load status, connection error, communication link status and actuator efficiency are unified into the same data object framework, and path tracking, connection maintenance, communication compensation and execution safety can be uniformly described and processed in the same control flow, solving the problem that when multiple vehicles jointly transport loads, vehicle status, load status, connection error, communication status and actuator efficiency are difficult to be uniformly incorporated into the same predictive control framework.

[0160] (2) This invention constructs an enhanced observation and establishes a bilinear Koopman prediction model. The bilinear term can characterize the multiplicative coupling relationship between the control input and the boost state. Furthermore, by performing input-sensory stable projection on the input-related Koopman effective operator, the main matrix is ​​projected and the bilinear input matrix is ​​scaled to ensure that the spectral radius of the input-related effective operator meets the corresponding conditions. This constrains the short-time domain prediction amplification effect and solves the problem that ordinary linear prediction models are difficult to characterize the multiplicative coupling between control input, load connection error and vehicle lateral motion, and may also cause short-time domain prediction amplification during online updates.

[0161] (3) This invention obtains network compensation state and compensation path packet by predicting and compensating for delay or loss information, and transforms communication state, expected path, connection state and actuator state into scheduling variables of model predictive control weights, constraint boundaries, predetermined performance control boundaries and control modes. This allows communication delay, packet loss, link quality degradation, path curvature change and actuator efficiency degradation to participate in the construction of model predictive control problem in a synchronized manner, solving the problem that communication delay, packet loss, link quality degradation, path curvature change and actuator efficiency degradation are difficult to be simultaneously transformed into weights, constraints and control mode scheduling variables.

[0162] (4) This invention predicts the control problem by constructing a communication state-aware consensus Koopman model, solves the candidate control commands, and performs actuator fault diagnosis (including exponential weighted moving average statistics and cumulative sum statistics), fault-tolerant projection, safety certificate judgment, smoothing and limiting processing. It also introduces load force proxy quantities to participate in constraints and audit items, so that the control commands under degraded conditions meet the feasible input range of the actuator, load connection constraints and safety certificate conditions. This solves the problem that the candidate control commands may not meet the feasible input range of the actuator, load connection constraints or safety certificate conditions under degraded conditions. Finally, it achieves the technical effect of generating collaborative transport control commands that can be executed for actual vehicles in complex environments, and improves the reliability and adaptability of the control.

[0163] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for collaborative transport control of intelligent vehicles in complex environments, characterized in that, include: Step S1: Multi-source signal acquisition to obtain the desired path, temporary desired path packet, vehicle status, load status, connection status, communication status, and actuator status; Step S2: Based on the multi-source signals output in step S1, construct the vehicle-load-communication coupling error, establish the discrete system model required for predictive control, and then output the complete dataset. Step S3: Construct augmented observations based on the complete dataset output in step S2, perform bilinear Koopman learning, and obtain the offline prediction model package; Step S4: Based on the offline prediction model package output in step S3, perform input-aware stable projection on the input-related Koopman effective operators, and perform prediction compensation for delayed or lost information to obtain the projected model, network compensation state, and compensation path package. Step S5: Based on the communication status, network compensation status, and compensation path packet output in step S4, and combined with the desired path, connection status, and actuator status, the model predictive control weights, constraint boundaries, predetermined performance control boundaries, and control modes are scheduled. Step S6: Based on the projected model and network compensation state output in step S4, and the model prediction control weights, constraint boundaries, predetermined performance control boundaries and control modes output in step S5, construct the communication state-aware consensus Koopman model prediction control problem, and solve it to obtain candidate control commands and prediction states. Step S7: Based on the candidate control commands and predicted states output in step S6, perform actuator fault diagnosis, fault-tolerant projection, safety certificate judgment, smoothing and limiting processing, and send the execution command sequence to the actuators of each actual vehicle to enable each actual vehicle to perform collaborative transport and generate an operation log.

2. The intelligent vehicle collaborative handling control method in complex environments according to claim 1, characterized in that, In step S1, the collaborative handling system includes multiple mobile vehicles and a common load, and the set of vehicle nodes is denoted as . Vehicle number is , The sampling period; Data sets collected or calculated at any time Defined as: in, For global reference path information, The upper-level planning nodes are distributed to the first... The short-term temporary expected path packet of the vehicle. For the first Vehicle global status, For the first Vehicle path coordinate error status, For the first Vehicle control input, At the load center state, For the first Connection error between the vehicle and load connection points For the set of communication link states, For actuator efficiency estimation, For system parameter set; Global reference path information is represented as follows: in, and These are the x and y coordinates of the reference path, respectively. For reference heading angle, For reference curvature, For reference velocity profile, This is the path arc length parameter; The temporary expected path packet is represented as: in, This is a positional function within the path packet. For the path packet heading angle, For path envelope curvature, This refers to the local arc length parameter within the path packet. This is the length of the path packet; No. The vehicle's global state, path coordinate error state, and control input are represented as follows: in, , For the first global plane coordinates of the vehicle For the first Vehicle heading angle, and They are respectively Time of the first The longitudinal and lateral speeds of the vehicle for Time of the first The yaw rate of the car, Indicates transpose. For the first Vehicle path projection progress, For lateral error, For heading error, For longitudinal acceleration, This is a steering angle command; The load center state is represented as follows: in, , For the global coordinates of the load center, For the load heading angle; No. The fixed position vector of the connection point between the vehicle and the load in the load coordinate system is: The position of the connection point in the global coordinate system for: in, The load center position vector, The rotation matrix is ​​constructed from the load heading angle. , They are respectively Two components in the load coordinate system; No. Connection error between vehicle and load connection point Defined as: in, For the first The car is The position vector at time , ; The load-bearing capacity is: in, For the first Load-bearing capacity at vehicle connection point The equivalent stiffness matrix, The equivalent damping matrix is... This is the first derivative of the connection error with respect to time. Communication graph is defined as ,in, For upper-level planning nodes, For a moment A set of directed communication links; for links The communication status is: in, As the information sending node, As an information receiving node, for Link quality at any given moment For communication delay, For packet loss indication, For communication bias, For communication noise, superscript For communication-related quantities; Overall network quality is defined as: in, For overall network quality, the superscript "net" represents the network aggregation quantity. This is a link quality aggregation function; The actuator efficiency is estimated as follows: in, For longitudinal acceleration channel efficiency, For the efficiency of the turning channel.

3. The intelligent vehicle collaborative handling control method in complex environments according to claim 2, characterized in that, In step S2, based on the output of step S1 , No. The global kinematics of the car are represented as follows: After the vehicle is projected onto the reference path, the path coordinate error satisfies: in, , , , , , They represent the first global X-axis coordinates of the car Global Y-axis coordinates Heading angle Path projection progress lateral error Heading error The first derivative with respect to time, and The first The longitudinal and lateral speeds of the vehicle For the first The yaw rate of the car, For reference path in Curvature at that point; The load center tracking error is defined as: in, For load center tracking error, , , These are the reference x-coordinate, reference y-coordinate, and reference heading angle of the load center, respectively, with the superscript ref indicating a reference value; The connection error is concatenated as follows: in, This is a spliced ​​vector of vehicle-load connection error. To concatenate by column; The distance error between adjacent vehicles is defined as: in, For link The corresponding distance error between adjacent vehicles, superscript For distance channels, The distance between adjacent vehicles after compensation. For reference to adjacent distances, This is the current reference path progress; The communication degradation variable is incorporated into the error model as follows: in, for The distance between adjacent vehicles measured at any time. Due to the delay in the information channel, This is for offsetting the distance information channel. This refers to communication noise in the distance information channel; To compensate for the lateral error of the load. for Lateral load error at time t, , and These are the time delay, offset, and communication noise of the load lateral error information channel, respectively. For the compensated temporary expected path packet, for Temporary expected path packet at time, , and These are the latency, offset, and communication noise of the path packet information channel, respectively. By integrating vehicle kinematics, load connectivity, communication degradation, and actuator states, the expression for the predictive control state-space model is obtained: in, The system state at the next discrete time step. This represents the current system state. This is the current system control input. The system state update function is formed by the combined effects of vehicle kinematics, load connectivity, and communication degradation. This is the system disturbance vector; Finally, step S2 is obtained at time [time]. The complete set of output data for: 。 4. The intelligent vehicle collaborative handling control method in complex environments according to claim 3, characterized in that, In step S3, based on the output of step S2 Define enhanced observations : in, To enhance the observation, This can be used as a reference input or a reference feature vector. Define promotion mapping: in, for Constantly improving one's state. The parameter is Koopman lifting mapping, It is determined by parameters Determined nonlinear feature mapping, To improve the mapping parameters; Establish a bilinear Koopman prediction model: in, for Constantly improving one's state. To improve the state master matrix, To control the input matrix, For the first A bilinear input matrix, For the summation index of the input channel, To control the input dimension, For a moment The first of the overall control input vectors Each scalar input channel component To predict residuals in one step; The offline dataset is defined as: in, This is an offline dataset; the index "off" indicates offline. The Koopman regression vector is: in, For the Koopman regression vector, the subscripts are... For Koopman, and Representing time respectively Overall control input vector The first and the second Each scalar input channel component; Will The Koopman matrix to be identified is obtained through regression with regularization: in, To improve the state matrix for the next time step in the training samples, The regression matrix is ​​composed of regression vectors. The regularization coefficient is . It is the identity matrix. The identification results obtained through optimization or regression; Simultaneously record the prediction residual boundary at one step: in, For vector norm, For constant residual boundary parameters, For residual boundary parameters related to enhanced observation bias, For residual boundary parameters related to network quality degradation, To enhance the reference value of the observation; Final output: offline prediction model package : in, The offline prediction model package output from step S3. This is a projection operator for reconstructing the system state from the improved state.

5. The intelligent vehicle collaborative handling control method in complex environments according to claim 4, characterized in that, In step S4, based on the output of step S3 Given a control input At that time, the input-related valid operators for lifting the state are: in, To control input The following are the relevant valid operators for the input; Project the principal matrix as: in, This is the projected principal matrix. For For the stable projection operator of the projection parameters, These are the projection constraint parameters of the main matrix; Then the bilinear input matrix is ​​scaled: in, For the projected or scaled first A bilinear input matrix, For the first The scaling factor of a bilinear input matrix. ; Constructing the input sample set , making the input sample set The input-related valid operators within satisfy: in, To obtain the maximum value, The spectral radius of the matrix is... for The elements in To provide the upper bound of the stability radius of the relevant effective operators, For the first input sample Components of each input channel; Alternatively, it satisfies the conservative norm constraint under the continuous input envelope: in, It is a 2-norm. For the first Upper bound of the absolute values ​​of each input channel; The projected model is obtained after online updating: in, For a moment The projected model, , and They are time points After projection, the principal matrix, the input matrix, and the first... Bilinear input matrices; When the vehicle At any moment Received from vehicle When dealing with delayed state data, the projected model is used to obtain: in, For from The network compensation status, Based on the receiving time and status information channel delay The compensation received improved the status; When the vehicle When a state data packet is lost, it is reconstructed using the previous valid network compensation state or the projected model prediction state; at the same time, the temporary expected path packet is time-aligned, the previous valid path packet is preserved, or the reference schedule is shrunk. The final data set output by step S4 for: in, For link At any moment Smooth link quality, and These are link packet loss indication and communication delay, respectively.

6. The intelligent vehicle collaborative handling control method in complex environments according to claim 5, characterized in that, In step S5, based on the communication status output in step S4, the link is... Define smooth link quality: in, This is the link quality smoothing coefficient. For link At any moment Smooth link quality; Neighbor consensus weight Defined as: in, and These are the minimum and maximum values ​​of the consensus weight, respectively. When the overall network quality degrades, the upper bound of the connection error and the input constraint are defined as follows: in, This is the upper bound of the connection error that varies with the overall network quality. This is the upper bound of the minimum connection error. This is the upper bound of the connection error under the preset conditions. For a moment The set of input constraints, As the baseline input constraint set, To constrain the tightening amount, To constrain tightening operations; The predetermined performance boundary for the lateral error is defined as follows: in, For a moment The lateral error is predetermined performance boundary. As the initial boundary, For steady-state boundary, This is the boundary attenuation coefficient; The normalization error ratio is: in, For the first Normalized ratio of vehicle lateral error To prevent small positive numbers with a denominator of zero; Control Mode Defined as: in, For control mode, For normal tracking and connection maintenance mode, For high curvature lateral priority mode, For load protection mode, For conservative fallback mode under communication or actuator degradation; The weights in the high curvature lateral priority mode are: in, This is the vehicle error weight matrix in the high curvature lateral priority mode. For the weight scaling matrix, This is the vehicle error baseline weight matrix. It is a diagonal matrix. and These are the lateral error weighting amplification factor and the heading error weighting amplification factor, respectively. , ; Load protection strength Defined as: in, To limit the input to Amplitude limiting function between, To protect the exit curvature threshold, To protect the fully activated curvature threshold, The curvature value of the reference path at the projection point of the vehicle's current position; In load protection mode, the load protection items are: in, For the objective function term under load protection mode, The original objective function is... , and These are the weighting coefficients for connection error, input dispersion, and load force surcharge, respectively. This is the average value of the control input for each vehicle. For the load plane force surcharge, For the number of vehicles; The final data set output by step S5 for: in, , , These are the set of state weights, input weights, and state constraints, respectively.

7. The intelligent vehicle collaborative handling control method in complex environments according to claim 6, characterized in that, In step S6, based on the projected model output in step S4... and the output of step S5 Define the prediction time domain length as The optimized decision vector is: in, This represents the optimal decision vector corresponding to the overall control sequence. To predict the first element in the control sequence, For at any time The predicted first Constantly control the input. For at any time The predicted first Constantly control input; Improved state prediction satisfies: in, To predict the state of improvement, To improve the state for the next prediction step, For at any time The predicted first Constantly control the input. This refers to the prediction step number; For time-based Information, predictions The first moment in the overall control vector at time step One scalar channel component; Predictive control state is determined by Obtain, among which, To predict control state; The state-aware consensus Koopman model predicts the control objective function as follows: in, The model predicts the control objective function. For the predicted load center tracking error, It is a weighted quadratic form. This is the load error weight matrix. For the first Predicting vehicle error This is the vehicle error weight matrix. For distance error consensus weights, The predicted vehicle status for the information receiving node. The predicted state of vehicle network compensation from the information sending node. The consensus error weight matrix is... The error in the predicted distance between adjacent vehicles, This is the spliced ​​vector of the predicted vehicle-load connection error. To connect the error weight matrix, For the predicted control input vector, For the input weight matrix, To predict changes in input, For the input rate of change weight matrix, For soft-constrained slack variables, This is the slack variable weight matrix; For time-based Information predicted at time The load center tracking error; This is the terminal load error weight matrix; The constraints include speed constraints, steering and longitudinal input constraints, input rate of change constraints, connection error constraints, load force surrogate constraints, and predetermined performance control soft constraints. in, For the first The car is Predicted longitudinal velocity at time [time] and These are the lower and upper bounds of the longitudinal velocity, respectively. For the first The car is Predicted steering angle command at any given time. For a moment The upper limit of the steering angle, For the first The car is Predicted longitudinal acceleration command at time of moment For a moment The upper limit of longitudinal acceleration, For the first Predicting the change in steering angle of the vehicle. This is the upper bound of the change in steering angle. For the first Predicting the change in longitudinal acceleration of the vehicle. This represents the upper bound of the change in longitudinal acceleration. For the first The car is Predicted connection error at time step To improve overall network quality The upper bound of the changing connection error For the first The car is Predicted load surcharge at time of day This is the upper bound of the load-bearing force surcharge. For a moment The lateral error is predetermined performance boundary. For the first Soft constraint relaxation variable for vehicle lateral error; After solving, the first control input in the predicted sequence is taken as the candidate control command: in, Candidate control commands are obtained after solving the model predictive control. This is the first control input in the optimal prediction sequence; The final data set output by step S6 for: in, The data set output in step S6 This is the optimal overall control sequence. , , These are the one-step predicted state, the one-step predicted connection error, and the one-step predicted load force surrogate, respectively.

8. The intelligent vehicle collaborative handling control method in complex environments according to claim 7, characterized in that, In step S7, based on the output of step S6 , for the The vehicle performs actuator fault diagnosis and command correction to obtain the actual execution input for issuing to the actual vehicle actuators. The actual execution input is represented as follows: in, For the first The vehicle actually executes the input. For the first The vehicle issues control commands. For element-wise multiplication, The execution error is bounded. The one-step prediction residual is: in, To predict residuals in one step, For the first The car at any time The system status, The first time obtained based on the information from the previous time step The vehicle's current predicted status; The exponentially weighted moving average statistic is: in, For the first The car is The exponentially weighted moving average statistic at time point. For the first The car is The exponentially weighted moving average statistic at time point. The residual statistical smoothing coefficient; The cumulative sum statistic is: in, For the first The car is Accumulation and statistics over time. For the first The car is Accumulation and statistics over time. For fault diagnosis statistics thresholds; When a decrease in actuator efficiency, degradation in communication status, or a deterioration in link quality is detected, candidate control commands are projected from the conservative set of feasible inputs: in, The overall control commands after safe projection. Let the set of conservative feasible inputs be determined by actuator efficiency estimates and network quality. For system-level actuator efficiency estimation, For the safe projection weight matrix, This is the connection error penalty coefficient. For a given control input One-step prediction connection error at time; Define the comprehensive error for security certificates: in, For safety certificates, comprehensive error is used. To improve state prediction error; The certificate function is: in, For control mode The certificate function below, For control mode The corresponding positive definite weight matrix, Let be a vector whose elements are all 1s. It is the first The actuator efficiency estimation vector after filtering or observation. For the first The efficiency deviation weight matrix of the vehicle actuator. This is the link quality penalty coefficient; The enhanced certificate function after adding the load force proxy audit item is: in, To enhance the certificate function, As the audit weight for load-bearing surcharge; The one-step certificate margin is: in, For one-step certificate margin, For control mode The corresponding certificate shrinkage rate, For system disturbance weighting coefficients, To predict the residual weighting coefficients, Enhance the certificate function for the next time step; when When the candidate security instruction is deemed to meet the certificate requirements, among which... The certificate margin allowance; otherwise, execute at least one of the following: control command projection, constraint tightening, reference progress shrinking, degradation rollback mode switching, previous safety control command holding, braking command, or preset minimum risk command. Smooth out safety control commands that meet safety conditions: in, For the first The vehicle issues control commands at the current moment. To control the smoothing coefficient of the instructions, , For the first The car is at the moment Safety projection control commands, For the first Control commands are issued from the vehicle at any time; Apply amplitude and rate of change constraints: in, and The first The car is at the moment and time The issued steering angle command, This is the upper bound of the rate of change of steering angle. and The respective The car is at the moment and time The issued longitudinal acceleration command, This is the upper bound of the longitudinal acceleration rate of change; Will meet safety conditions The system is deployed to the steering, drive, or braking actuators of each actual vehicle, enabling the vehicles to perform coordinated transport and movement. Final output data set It then issues a sequence of execution instructions to enable each actual vehicle to perform coordinated transport operations and generates an operation log, in which... The expression is: in, The data set output and distributed in step S7 This is the set of load surrogate quantities at the connection point.

Images