Train parameter uncertain anti-saturation adaptive neural network control method and system

Abstract

The application provides a train parameter uncertain anti-saturation adaptive neural network control method and system, and belongs to the technical field of train control. The technical scheme is: a train longitudinal motion dynamics model is established, mass uncertainty, resistance parameter uncertainty, additional resistance and unmodeled dynamics are taken as compound unknown dynamics; an actuator saturation constraint model and a saturation deviation auxiliary system are constructed to form a comprehensive error; an RBF neural network is used to online approximate the compound unknown dynamics, and an adaptive law with a leakage term correction and an adaptive anti-saturation control law are used to output actual control input. The beneficial effect is that the train displacement and speed tracking precision can be improved under the conditions of actuator saturation and parameter uncertainty, parameter drift can be inhibited, and the stability of closed-loop control can be enhanced.

Description

Technical Field

[0001] This invention relates to the field of train control technology, and more particularly to a train parameter uncertainty anti-saturation adaptive neural network control method and system. Background Technology

[0002] As rail transit systems develop towards higher speeds, greater intelligence, and automation, the requirements for position tracking accuracy, speed tracking accuracy, operational stability, and safety and reliability of train operation control systems are constantly increasing. Automatic train control (ATC) technology, by adjusting train traction and braking forces in real time to ensure safe and efficient train operation along the desired trajectory, is one of the key technologies in modern rail transit systems. Among these, train position and speed tracking control, as a crucial component of automatic train operation and cooperative operation control, directly impacts train operating efficiency, passenger comfort, and driving safety.

[0003] In actual operation, the longitudinal motion dynamics system of a train is characterized by nonlinearity, parameter uncertainty, and complex external disturbances. For example, the train's running mass changes with the number of passengers and load; the basic running resistance is affected by the perturbation of the Davis equation coefficients; additional resistance varies with track gradient, curve radius, tunnel environment, and aerodynamic factors; and unmodeled dynamics can cause the actual train operating state to deviate from the ideal model. These factors make it difficult for control methods based on precise model parameters to maintain high tracking accuracy and stability under complex operating conditions.

[0004] Furthermore, both the train traction and braking actuators have limitations on their actual output capacity. When the desired control input calculated by the controller exceeds the upper limit of traction force or falls below the lower limit of braking force, it is subject to actuator saturation constraints, resulting in a saturation deviation between the desired and actual control inputs. If this saturation deviation is not effectively compensated, it can easily lead to increased train displacement and speed tracking errors, thereby affecting the stability and tracking performance of the closed-loop control system.

[0005] Existing train operation control methods mainly include proportional-integral control, adaptive control, sliding mode control, model predictive control, and neural network control. These methods can improve train control performance to some extent, but some rely on relatively accurate system model parameters. Control performance may degrade or even affect system stability when there are uncertainties in running quality, perturbations in running resistance parameters, changes in additional resistance, unmodeled dynamics, and actuator saturation constraints. Neural networks have strong nonlinear approximation capabilities and can be used to estimate unknown dynamics in the longitudinal motion dynamics of trains, but problems still need to be solved such as actuator saturation impact compensation, parameter estimation drift suppression, and closed-loop stability constraints. Summary of the Invention

[0006] The purpose of this invention is to provide a train parameter uncertainty anti-saturation adaptive neural network control method and system that can perform online approximation of complex unknown dynamics and dynamic compensation for saturation deviation under actuator saturation constraints and uncertain train operating parameters, so as to improve the train displacement tracking accuracy, speed tracking accuracy and closed-loop control system stability.

[0007] This invention is achieved through the following measures: A method for train parameter uncertainty-resistant adaptive neural network control, characterized by the following steps: S1, Obtain the first Based on the real-time operating status, expected operating curve, and train operating parameters of the train, a longitudinal motion dynamics model of the train is established. The uncertainties in operating quality, the parameter uncertainties in basic operating resistance, additional resistance, and unmodeled dynamics are identified as composite unknown dynamics. The longitudinal motion dynamics model and composite unknown dynamics of the train are then output. S2. Establish an actuator saturation constraint model based on the lower limit of braking force and the upper limit of traction force. The actuator saturation constraint model is used to limit the expected control input generated by the controller, determine the actual control input and saturation deviation, and output saturation auxiliary variables through the saturation deviation auxiliary system. S3. Using the real-time operating status and the desired operating curve as input, determine the displacement tracking error and the velocity tracking error, and construct the displacement tracking error, velocity tracking error and saturated auxiliary variable into a comprehensive error; S4. Using the comprehensive error and real-time operating status as input, the RBF neural network is used to approximate the composite unknown dynamics online, and the estimated value of the composite unknown dynamics, the weight parameters of the RBF neural network and their estimation error are output. S5. Using the comprehensive error, composite unknown dynamic estimate, RBF neural network weight parameter estimate and its estimated parameters as input, construct an adaptive law with leakage term correction, and construct an adaptive anti-saturation control law based on the adaptive law, comprehensive error, saturation auxiliary variable and composite unknown dynamic estimate, and output the actual control input that satisfies the lower limit of braking force and the upper limit of traction force constraints. S6. Apply the actual control input to the train longitudinal motion dynamics model to form a closed-loop control system, and constrain the stability of the closed-loop control system based on the Lyapunov function, so that the actual displacement and actual speed of the train follow the desired running curve in a bounded manner, and the displacement tracking error and speed tracking error converge to a preset neighborhood.

[0008] The invention also has the following specific features: Step S1 includes: obtaining the first The train's running time, real-time displacement, real-time speed, desired displacement curve, desired speed curve, known mass, running mass uncertainty, control input variables, basic running resistance, additional resistance, and unmodeled dynamics are used to establish a longitudinal motion dynamics model based on Newton's second law. The running mass uncertainty, the parameter uncertainty of the basic running resistance, the additional resistance, and the unmodeled dynamics are collectively considered as the composite unknown dynamics. After equivalent processing using the train longitudinal motion dynamics model, the running mass uncertainty is combined with the parameter uncertainty of the basic running resistance, the additional resistance, and the unmodeled dynamics into the composite unknown dynamics.

[0009] Step S2 includes: denoting the desired control input calculated by the controller as the first... The desired control input of the train, after being processed by the actuator saturation constraint model, is denoted as the control input acting on the train's longitudinal motion dynamics model. The actual control input of the train; when the desired control input is less than the lower limit of braking force, the actual control input is limited to the lower limit of braking force; when the desired control input is between the lower limit of braking force and the upper limit of traction force, the actual control input is equal to the desired control input; when the desired control input is greater than the upper limit of traction force, the actual control input is limited to the upper limit of traction force; and the difference between the actual control input and the desired control input is determined as the saturation deviation.

[0010] Step S2 further includes: constructing a saturation deviation auxiliary system, wherein the saturation deviation auxiliary system takes the saturation deviation as input and generates a saturation auxiliary variable through a first-order dynamic compensation method, the saturation auxiliary variable being used to characterize the influence of the actuator saturation deviation after dynamic compensation; wherein, when the desired control input does not exceed the actuator's allowable output range defined by the lower limit of braking force and the upper limit of traction force, the saturation deviation is zero, and the saturation auxiliary variable gradually decays with the saturation deviation auxiliary system; when the desired control input exceeds the actuator's allowable output range, the saturation deviation is input to the saturation deviation auxiliary system, causing the saturation auxiliary variable to be dynamically updated according to the change of the saturation deviation.

[0011] Step S3 includes: according to the first The displacement tracking error is determined by comparing the real-time displacement with the expected displacement curve of the train. Based on the first... The speed tracking error is determined by comparing the real-time speed of the train with the expected speed curve, and the displacement tracking error, speed tracking error, and saturation auxiliary variable are combined to construct a comprehensive error. The comprehensive error is used to simultaneously characterize the tracking deviation of the actual train trajectory relative to the expected running curve, as well as the impact of the actuator saturation deviation after reconstruction by the saturation deviation auxiliary system.

[0012] Step S4 includes: representing the composite unknown dynamic as the sum of the ideal approximation term and the approximation error of the composite unknown dynamic by the RBF neural network, and using the RBF neural network to perform online estimation of the composite unknown dynamic; wherein, the input of the RBF neural network is determined by the first... The system consists of the real-time operating status of the train and the comprehensive error correlation quantity, which includes displacement tracking error, speed tracking error and saturation auxiliary variable. The basis function vector and weight vector of the RBF neural network are both n-dimensional vectors, where n represents the number of nodes in the RBF neural network. The approximation error satisfies the bounded condition, so that the saturation auxiliary variable is dynamically updated according to the change of the saturation deviation.

[0013] Step S5 includes: constructing an adaptive law for the estimated weight parameters and the estimated parameters related to the upper bound of the approximation error of the RBF neural network based on the comprehensive error, and introducing a leakage term correction coefficient into the adaptive law to form an adaptive law with leakage term correction; wherein, the leakage term correction coefficient is used to suppress the parameter estimation drift of the estimated weight parameters and the estimated parameters related to the upper bound of the approximation error of the RBF neural network during long-term operation, so that the parameter estimates remain bounded.

[0014] Step S5 further includes: constructing an adaptive anti-saturation control law and generating desired control input based on the adaptive law with leakage term correction, comprehensive error, saturation auxiliary variable, composite unknown dynamic estimate, approximation error upper bound related parameter estimate and speed tracking error; wherein, vehicle 1 generates desired control input with preset desired displacement curve and desired speed curve as tracking target, and vehicles 2 to 4 generate desired control input with the desired displacement curve and desired speed curve determined by the real-time displacement of the preceding vehicle, the real-time speed of the preceding vehicle and the preset relative position relationship as tracking target, and the desired control input is processed by the actuator saturation constraint model to obtain the actual control input.

[0015] Step S6 includes: applying the actual control input to the train longitudinal motion dynamics model to form a closed-loop control system comprising the train longitudinal motion dynamics model, actuator saturation constraint model, saturation deviation auxiliary system, RBF neural network, adaptive law with leakage term correction, and adaptive anti-saturation control law; and selecting a Lyapunov function including the comprehensive error term, the estimation error term of the upper bound correlation estimate of the approximation error, and the estimation error term of the RBF neural network weight parameters to perform stability analysis on the closed-loop control system to determine the stability conditions satisfied by the control parameters and adaptive parameters. The displacement tracking error and speed tracking error of the train asymptotically converge to zero or remain within a bounded range.

[0016] A train parameter uncertainty anti-saturation adaptive neural network control system, characterized in that it includes: The dynamic modeling module is used to obtain the first... Based on the real-time operating status, expected operating curve, and train operating parameters of the train, a longitudinal motion dynamic model of the train is established, and the uncertainties in operating quality, parameters of basic operating resistance, additional resistance, and unmodeled dynamics are identified as composite unknown dynamics. The saturation processing module is used to establish an actuator saturation constraint model based on the lower limit of braking force and the upper limit of traction force, taking the train longitudinal motion dynamics model and the desired control input as inputs, to determine the actual control input and saturation deviation, and to output saturation auxiliary variables through the saturation deviation auxiliary system. The comprehensive error construction module is used to determine the displacement tracking error and velocity tracking error by taking the real-time operating status, the expected operating curve and the saturated auxiliary variable as input, and to construct the displacement tracking error, velocity tracking error and saturated auxiliary variable as a comprehensive error; The composite unknown dynamic estimation module is used to approximate the composite unknown dynamic online using the comprehensive error and real-time operating status as inputs, and output the composite unknown dynamic estimate and the RBF neural network weight parameter estimate. An adaptive anti-saturation control module is used to construct an adaptive law with leakage term correction by taking the comprehensive error, composite unknown dynamic estimate, and RBF neural network weight parameter estimate as inputs, and construct an adaptive anti-saturation control law based on the adaptive law, comprehensive error, saturation auxiliary variable, and composite unknown dynamic estimate, and output the actual control input that satisfies the lower limit of braking force and the upper limit of traction force constraints. The stability constraint module is used to apply the actual control input to the train longitudinal motion dynamics model to form a closed-loop control system, and to determine the stability of the closed-loop control system based on the Lyapunov function, so that the train displacement tracking error and speed tracking error remain bounded and converge to a preset neighborhood.

[0017] The beneficial effects of this invention are as follows: 1. It can perform online estimation of complex unknown dynamics under conditions of uncertain operating quality, uncertain basic operating resistance parameters, changes in additional resistance, and the existence of unmodeled dynamics, thereby improving the adaptability of the control method to the uncertainty of train parameters; 2. It can dynamically compensate for the saturation deviation between the desired control input and the actual control input through the saturation deviation auxiliary system, so that the saturation effect of the actuator can participate in the comprehensive error construction and adaptive anti-saturation control process, thereby improving the tracking control performance of the actuator under saturation conditions. 3. It can use the comprehensive error constructed from displacement tracking error, speed tracking error and saturated auxiliary variables to make the actual displacement and actual speed of the train stably track the desired running curve, thereby improving the position tracking accuracy and speed tracking accuracy; 4. It can suppress the long-term drift of the estimated values ​​of the weight parameters and the upper bound of the approximation error related parameters of the RBF neural network through the adaptive law with leakage term correction, so that the parameter estimates remain bounded and improve the stability and reliability of the closed-loop control system. Attached Figure Description

[0018] Figure 1 This is a flowchart of an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the expected displacement and actual displacement curves in Embodiment 2 of the present invention.

[0020] Figure 3 This is a schematic diagram of the expected speed versus actual speed curves in Embodiment 2 of the present invention.

[0021] Figure 4 This is a schematic diagram of the control input in Embodiment 2 of the present invention. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0023] To clearly illustrate the technical features of this solution, the following detailed implementation method will be used to explain the solution.

[0024] To achieve stable tracking of desired displacement and speed curves during train operation and effectively address uncertainties in the train's longitudinal motion dynamics model, including uncertainties in running quality, parameters of basic running resistance, additional resistance, unmodeled dynamics, and actuator saturation constraints, this invention provides a train parameter uncertainty anti-saturation adaptive neural network control method and system. This method establishes a train longitudinal motion dynamics model, identifies uncertainties in running quality, parameters of basic running resistance, additional resistance, and unmodeled dynamics as composite unknown dynamics, and utilizes an RBF neural network to approximate these composite unknown dynamics online, thereby improving the control method's adaptability to train parameter uncertainties and complex operating conditions.

[0025] To achieve the above objectives, this invention focuses on the closed-loop interaction between actuator saturation deviation, saturation deviation auxiliary system, saturation auxiliary variable, comprehensive error, composite unknown dynamic estimation, adaptive law with leakage term correction, and adaptive anti-saturation control law. Specifically, an actuator saturation constraint model is established based on the lower limit of braking force and the upper limit of traction force to determine the actual control input and saturation deviation, and a saturation auxiliary variable is generated through the saturation deviation auxiliary system. The displacement tracking error, speed tracking error, and saturation auxiliary variable are combined to construct a comprehensive error. Then, based on the comprehensive error, the composite unknown dynamic estimation value, and the RBF neural network weight parameter estimation value, an adaptive law with leakage term correction and an adaptive anti-saturation control law are constructed to output the actual control input that satisfies the lower limit of braking force and the upper limit of traction force constraints, enabling the actual displacement and actual speed of the train to track the desired running curve. The details are as follows: Example 1 See Figure 1 A method for train parameter uncertainty anti-saturation adaptive neural network control includes the following steps: S1, Obtain the first Based on the real-time operating status, expected operating curve, and train operating parameters of the train, a longitudinal motion dynamics model of the train is established. The uncertainties in operating quality, the parameter uncertainties in basic operating resistance, additional resistance, and unmodeled dynamics are identified as composite unknown dynamics, and the longitudinal motion dynamics model and composite unknown dynamics of the train are output.

[0026] Specifically, this embodiment provides a train parameter uncertainty anti-saturation adaptive neural network control method for train position and speed tracking control. For the first... For each train, its running time, real-time displacement, real-time speed, real-time acceleration, desired displacement curve, desired speed curve, known mass, running mass uncertainty, control input variables, basic running resistance, additional resistance, and unmodeled dynamics are obtained; among them, =1,2,3,4.

[0027] In some implementations, vehicle 1 acts as the lead vehicle, and vehicles 2 to 4 act as follower vehicles. Vehicle 1 tracks a preset desired displacement curve and a desired speed curve. Vehicles 2 to 4 adjust their traction or braking force according to the operating status of the lead vehicle to maintain the desired relative position and speed consistency among the vehicles during operation. For vehicles 2 to 4, their desired displacement and speed curves are determined by the real-time displacement and speed of the lead vehicle, as well as the preset relative position relationship. The preset relative position relationship is used to define the target relative position of adjacent vehicles during operation.

[0028] Based on Newton's second law, the first The forces acting on the train during longitudinal motion are analyzed, and the following dynamic equations for the train's longitudinal motion are established:

[0029] in, T represents the train's travel time; and They represent the first Real-time displacement, real-time velocity, and real-time acceleration of a train at time t; Indicates the first Knowable quality during train operation; Indicates the first Uncertainties in the quality of train operation; Indicates the first In the longitudinal motion dynamics model of a train, the control input variables are determined as the actual control inputs in step S2 after the desired control inputs are processed by the actuator saturation constraint model. Indicates the first The basic operating resistance of a train; Indicates the first Additional resistance of the train; Indicates the first Unmodeled dynamics in the train dynamics model.

[0030] In actual operation, train resistance includes basic operating resistance and additional resistance. Basic operating resistance is related to train speed and is expressed as:

[0031] in, and For the first Davis equation coefficients for the trains; , and This represents the uncertainties in the parameters of the basic operating resistance. Additional resistance includes curve resistance, slope resistance, tunnel resistance, and other resistances caused by unmodeled factors when the train is running on a fixed track.

[0032] Furthermore, to clarify the estimation target of the subsequent RBF neural network, the quality uncertainty term will be run. Uncertainty in the parameters of basic operating resistance , and Additional resistance and unmodeled dynamics These together constitute a complex unknown dynamic during train operation. The complex unknown dynamic is denoted as... The operational quality uncertainty term After equivalent rearrangement of the train's longitudinal motion dynamics equations, they are incorporated into the aforementioned complex unknown dynamics. In this context, the composite unknown dynamics are made It can uniformly characterize the combined impact of operational quality uncertainties, parameter uncertainties of basic operational resistance, additional resistance, and unmodeled dynamics on the train acceleration channel; the composite unknown dynamics In subsequent online approximation of the RBF neural network, construction of the adaptive law, and design of the adaptive anti-saturation control law, it is uniformly regarded as the unknown dynamic estimation object; the composite unknown dynamic is used to characterize the unknown parts in the train longitudinal motion dynamics model that are difficult to model accurately or change with the operating conditions, and serves as the object of subsequent online approximation of the RBF neural network.

[0033] Through the above processing, step S1 outputs a train longitudinal motion dynamics model that includes control input variables, basic running resistance, additional resistance, running quality uncertainty, and unmodeled dynamics, and outputs composite unknown dynamics; the train longitudinal motion dynamics model is used to subsequently establish an actuator saturation constraint model, and the composite unknown dynamics are used for subsequent RBF neural network estimation.

[0034] S2. Establish an actuator saturation constraint model based on the lower limit of braking force and the upper limit of traction force. The actuator saturation constraint model is used to limit the expected control input generated by the controller, determine the actual control input and saturation deviation, and output saturation auxiliary variables through the saturation deviation auxiliary system.

[0035] Specifically, due to the limited output capabilities of the train traction and braking actuators, the desired control input calculated by the controller cannot be directly used as the actual control input in the train's longitudinal motion dynamics model, but is subject to the limitations of the actuator's allowable output range. For the first... For each train, the desired control input calculated by the controller is denoted as... The actual control input applied to the train's longitudinal motion dynamics model after actuator saturation constraints is denoted as... The actual control input For the first The actual output of the train's traction actuator or braking actuator is applied to the train's longitudinal motion dynamics model.

[0036] In some implementations, the actuator saturation constraint model is represented as:

[0037] in, Indicates the lower limit of braking force. Indicates the upper limit of traction force; Indicates the first The desired control input for each train; Indicates the first The actual control input of the train. Specifically, when the desired control input... Less than the lower limit of braking force At that time, actual control input Restricted to When the desired control input Located at the lower limit of braking force and traction limit In between, actual control input Equal to expected control input When the desired control input Greater than the upper limit of traction force At that time, actual control input Restricted to .

[0038] Furthermore, when the desired control input When the output exceeds the actuator's permissible range, the desired control input is... With actual control input A saturation deviation occurs between the actual control input and the desired control input. In some implementations, the saturation deviation is defined as the difference between the actual control input and the desired control input, i.e.:

[0039] The saturation deviation Used to characterize the saturation deviation of the actuator saturation constraint on the controller calculation results.

[0040] This saturation deviation, as an additional input, affects the acceleration channel in the train's longitudinal motion dynamics model. If it is not compensated, it will reduce the tracking performance of the train's actual displacement and speed on the desired running curve.

[0041] To compensate for the impact of actuator saturation on train operation control, a saturation deviation auxiliary system is constructed. This system takes the saturation deviation as input and generates a saturation auxiliary variable through first-order dynamic compensation. This auxiliary variable characterizes the impact of the actuator saturation deviation after dynamic compensation. Its form is as follows:

[0042] in, Indicates the first Saturation auxiliary variables for each train; This represents the rate of change of the saturated auxiliary variable; Indicates the first Saturation deviation of trains; Indicates the first Knowable quality during train operation; For the first The design parameters corresponding to each train.

[0043] In some implementations, saturation deviation Can be controlled by actual input With desired control input Confirmed, and When the desired control input does not exceed the actuator's allowable output range, the actual control input is equal to the desired control input, the saturation deviation is zero, and the saturation auxiliary variable gradually decays with the saturation deviation auxiliary system. When the desired control input exceeds the actuator's allowable output range, the actual control input is limited to the actuator's allowable output range, and the saturation deviation is input to the saturation deviation auxiliary system, causing the saturation auxiliary variable to be updated with the change in saturation deviation.

[0044] Through the above processing, step S2 outputs the actual control input, saturation deviation, and saturation auxiliary variable. The actual control input is applied to the train's longitudinal motion dynamics model, and the saturation auxiliary variable is used to construct a comprehensive error together with the displacement tracking error and speed tracking error, allowing the actuator's saturation effect to participate in the subsequent adaptive anti-saturation control process.

[0045] S3. Using the real-time operating status and the desired operating curve as input, determine the displacement tracking error and the velocity tracking error, and construct the displacement tracking error, velocity tracking error and saturation auxiliary variable into a comprehensive error.

[0046] Specifically, the real-time operating status includes the first The real-time displacement and real-time speed of the trains are measured. The desired operating curve includes a desired displacement curve and a desired speed curve. The saturation auxiliary variable is the dynamic compensation amount output by the saturation deviation auxiliary system in step S2. For vehicle 1, the desired displacement curve and desired speed curve are preset desired operating curves; for vehicles 2 to 4, the desired displacement curve and desired speed curve are determined by the real-time displacement and real-time speed of the preceding vehicle and a preset relative position relationship. According to the... The real-time displacement and expected displacement curves of the train are determined. The displacement tracking error of the train, according to the first The real-time speed vs. desired speed curve of the train is determined. Speed ​​tracking error of the train.

[0047] In some implementations, the first The displacement tracking error of the train is denoted as The speed tracking error is denoted as Its error dynamically expressed as:

[0048] in, Indicates the first Displacement tracking error of the train Indicates the first Speed ​​tracking error of trains; Indicates the first Knowable quality during train operation; Indicates the first The actual control input of the train; and For the first Davis equation coefficients for the trains; Indicates the first The real-time speed of the train; This represents the expected displacement of the train. Indicates the desired speed of the train operation. This represents the desired acceleration of the train; in the dynamic analysis of speed tracking error, it uses... As a reference value for the acceleration channel.

[0049] Furthermore, the displacement tracking error, velocity tracking error, and saturation auxiliary variable are collectively constructed into a comprehensive error. This comprehensive error reflects both the tracking deviation of the actual train trajectory relative to the desired trajectory and the impact of actuator saturation deviation reconstructed by the saturation deviation auxiliary system. It is the core driving variable for subsequent adaptive and control law design. The comprehensive error is expressed as:

[0050] in, Indicates the first The overall error of each train; These are the filter coefficients; It is a saturation auxiliary variable.

[0051] In some implementations, the comprehensive error in Used to characterize position tracking deviation Used to characterize velocity tracking deviation This is used to characterize the impact of actuator saturation deviation after dynamic compensation; by introducing the three together into the comprehensive error, the actuator saturation effect no longer exists only as the amplitude limiting result after actuator saturation constraint, but participates in the subsequent adaptive anti-saturation control process as part of the error feedback.

[0052] Specifically, when the actuator does not saturate or the saturation deviation is small, the influence of the saturation auxiliary variable on the comprehensive error decreases, and the comprehensive error mainly reflects the displacement tracking error and the velocity tracking error. When the actuator saturates and generates a saturation deviation, the saturation auxiliary variable enters the comprehensive error, enabling the subsequent control law to adjust the actual control input together with the trajectory tracking deviation and the saturation compensation amount.

[0053] Through the above processing, step S3 outputs the comprehensive error. The comprehensive error serves as the input for subsequent online approximation by the RBF neural network, the construction of the adaptive law with leakage term correction, and the design of the adaptive anti-saturation control law.

[0054] S4. Using the comprehensive error and real-time operating status as input, the RBF neural network is used to approximate the composite unknown dynamics online, and the estimated value of the composite unknown dynamics, the estimated value of the RBF neural network weight parameters and their estimation error are output.

[0055] Specifically, due to the first The longitudinal motion dynamics model of a train contains uncertainties in running quality, parameters of basic running resistance, additional resistance, and unmodeled dynamics. These uncertainties are difficult to accurately represent using a fixed model. Therefore, an RBF neural network is used to approximate the complex unknown dynamics during train operation online.

[0056] In some implementations, the composite unknown dynamic is denoted as The approximate form of the RBF neural network is:

[0057] in, This is the ideal approximation term of the RBF neural network for complex unknown dynamics. To approximate the error, and satisfy... ; In order to make Minimize the ideal weight vector.

[0058] In some implementations, the ideal weight vector is defined as:

[0059] in, The basis function vectors of a neural network; This represents the weight vector of the RBF neural network; n represents the number of nodes in the neural network. Indicates the first The longitudinal motion dynamics of a train consists of a complex unknown dynamic formed by the uncertainty of running mass, the parameter uncertainty of basic running resistance, additional resistance, and unmodeled dynamics. Indicates the overall error The range of values ​​for .

[0060] In some implementations, the input to the RBF neural network is determined by the first... The system consists of the real-time operating status of a train and comprehensive error correlation quantities. The real-time operating status includes real-time displacement and real-time speed, and the comprehensive error correlation quantities include displacement tracking error, speed tracking error, and saturation auxiliary variables. Through the above input quantities, the RBF neural network can perform online estimation of complex unknown dynamics based on the train operating status, trajectory tracking deviation, and actuator saturation effects.

[0061] Furthermore, based on the RBF neural network approximation results, the train longitudinal motion dynamics equations considering parameter uncertainties, mass fluctuations, and unmodeled dynamics are reorganized into the following form:

[0062] in, This represents the complex unknown dynamics approximated by the RBF neural network; Indicates the first Knowable quality during train operation; Indicates the first The actual control input of the train; and For the first Davis equation coefficients for the trains; and They represent the first Real-time displacement and real-time speed of the train.

[0063] For the For a train, whose tracking targets are the corresponding desired displacement curve and desired velocity curve, the tracking error model is defined as follows:

[0064] in, Indicates displacement tracking error. Indicates speed tracking error. Represents the desired displacement curve. Represents the desired speed curve. This represents the reference acceleration corresponding to the desired running curve.

[0065] Specifically, the composite unknown dynamic estimate output by the RBF neural network is used to replace the unknown parts that are difficult to model accurately in the longitudinal motion dynamics model of the train, so that the subsequent adaptive anti-saturation control law can simultaneously consider the train trajectory tracking deviation, actuator saturation effect and parameter uncertainty effect.

[0066] Furthermore, the estimated weight parameters of the RBF neural network are updated in subsequent steps using an adaptive law. The estimation error is used to characterize the deviation between the composite unknown dynamics and the estimation results of the RBF neural network, and serves as a constraint object in subsequent stability analysis.

[0067] Through the above processing, step S4 outputs a composite unknown dynamic estimate and RBF neural network weight parameter estimates. The composite unknown dynamic estimate is used to subsequently construct an adaptive anti-saturation control law, and the RBF neural network weight parameter estimates and their estimation errors are used to subsequently construct an adaptive law with leakage term correction.

[0068] S5. Using the comprehensive error, composite unknown dynamic estimate, RBF neural network weight parameter estimate and its estimated parameters as input, construct an adaptive law with leakage term correction, and construct an adaptive anti-saturation control law based on the adaptive law, comprehensive error, saturation auxiliary variable and composite unknown dynamic estimate, and output the actual control input that satisfies the lower limit of braking force and the upper limit of traction force constraints.

[0069] Specifically, after obtaining the online approximation result of the RBF neural network for the complex unknown dynamics in step S4, for the first... For the parameter uncertainties of a train, we define the estimation errors of the weight parameters of the RBF neural network and the estimation errors of the upper bound of the approximation error, and construct the parameter adaptive law based on the comprehensive error.

[0070] In some implementations, the parameter estimation error is defined as:

[0071] in, This indicates the estimation error of the weight parameters in the RBF neural network; Represents the ideal weight vector; This represents the estimated weight parameters of the RBF neural network; This represents the estimation error of the correlation estimator related to the upper bound of the approximation error; This represents the parameters related to the upper bound of the approximation error; This represents the estimated value of the relevant parameters related to the upper bound of the approximation error.

[0072] In some implementations... Parameters used to represent the upper bound of the approximation error The adaptive estimator serves to compensate for the approximation error of the RBF neural network. The residual effects of dynamic estimation of complex unknowns; In conjunction with the composite unknown dynamic estimate in step S4, it is used to improve the control law's adaptability to parameter uncertainties and unmodeled dynamics.

[0073] Furthermore, the following parameter adaptive law is constructed:

[0074] in, This represents the update rate of the estimated weight parameters of the RBF neural network; This represents the update rate of the parameter estimates related to the upper bound of the approximation error; Indicates the overall error; It is a positive definite matrix; Represents the basis function vector of an RBF neural network; It is a positive constant; A sign function representing the overall error.

[0075] Adaptive laws may experience parameter estimate drift during long-term operation, thus affecting control performance. To suppress parameter estimate drift and ensure the boundedness of the estimated parameters, a leakage term correction mechanism is introduced to modify the parameter adaptive law. The modified adaptive law is as follows:

[0076] in, This represents the correction coefficient for the leakage term in the estimated weight parameters of the RBF neural network; The leakage term correction coefficient represents the parameter estimate related to the upper bound of the approximation error; the leakage term correction coefficient is used to limit the unbounded growth of the parameter estimate during the long-term adaptive update process.

[0077] Specifically, the adaptive law with leakage term correction is driven by the comprehensive error. When the comprehensive error is large, the adaptive law adjusts the weight parameter estimates based on the comprehensive error and the RBF neural network basis function vector, and adjusts the approximation error upper bound related parameter estimates based on the sign and magnitude of the comprehensive error. When the train runs for a long time and the parameter estimates show a drift trend, the leakage term correction coefficient has a suppressive effect on the corresponding estimates, keeping the parameter estimates bounded.

[0078] Furthermore, based on the adaptive law with leakage term correction, the comprehensive error, the saturation auxiliary variable, and the composite unknown dynamic estimate, an adaptive anti-saturation control law is constructed. For vehicle 1, its desired control input is expressed as:

[0079] For vehicles 2 to 4, the desired control input is expressed as follows:

[0080] in, This represents the desired control input for vehicle 1; Indicates the first The desired control input for each train; This represents the reference acceleration corresponding to the desired running curve; The first derivative represents the real-time speed of the vehicle in front, i.e., the real-time acceleration of the vehicle in front. and For the first Davis equation coefficients for the trains; Indicates the first The real-time speed of the train; This represents the composite unknown dynamic estimate of the output of the RBF neural network; This represents the estimated value of the relevant parameters related to the upper bound of the approximation error; Indicates a saturated auxiliary variable; Indicates the overall error; Indicates speed tracking error; and All are normal numbers.

[0081] In the above adaptive anti-saturation control law Used to compensate for complex unknown dynamics; Used to compensate for the approximation error of RBF neural networks; Used to introduce the dynamic compensation amount formed by the saturation deviation auxiliary system of the actuator saturation deviation; Used for feedback adjustment based on comprehensive error; Used for error feedback adjustment based on speed tracking error.

[0082] In some implementations, vehicle 1 generates desired control inputs based on preset desired displacement and desired speed curves as tracking targets; vehicles 2 to 4 generate desired control inputs based on the real-time displacement and real-time speed of the preceding vehicle, as well as the desired displacement and desired speed curves determined by a preset relative positional relationship. This enables vehicle 1 to track the preset desired operating curves, and allows vehicles 2 to 4 to adjust traction or braking force according to the operating state of the preceding vehicle, thereby maintaining the desired relative positional relationship and speed consistency.

[0083] Furthermore, the desired control input After the actuator saturation constraint model processing in step S2, the actual control input is obtained. The actual control input It satisfies the lower limit of braking force and the upper limit of traction force, and applies them to the longitudinal motion dynamics model of the train.

[0084] Specifically, when the desired control input is between the lower limit of braking force and the upper limit of traction force, the actual control input is equal to the desired control input; when the desired control input exceeds the upper limit of traction force or is lower than the lower limit of braking force, the actual control input is limited to the range allowed by the actuator. At the same time, the saturation auxiliary variable generated by the saturation deviation auxiliary system continues to participate in the comprehensive error and adaptive anti-saturation control law, so that the saturation effect of the actuator enters the closed-loop adjustment process.

[0085] Through the above processing, step S5 outputs the actual control input that satisfies the constraints of the lower limit of braking force and the upper limit of traction force. This actual control input is used to act on the train's longitudinal motion dynamics model, forming the subsequent closed-loop control system.

[0086] S6. Apply the actual control input to the train longitudinal motion dynamics model to form a closed-loop control system, and constrain the stability of the closed-loop control system based on the Lyapunov function, so that the actual displacement and actual speed of the train follow the desired running curve in a bounded manner, and the displacement tracking error and speed tracking error converge to a preset neighborhood.

[0087] Specifically, the actual control input output in step S5 The longitudinal motion dynamics model of the train established in step S1 is applied to make the first Real-time displacement of trains and real-time speed In the actual control input The system is updated under the influence of various factors, thereby forming a closed-loop control system that includes a train longitudinal motion dynamics model, an actuator saturation constraint model, a saturation deviation auxiliary system, an RBF neural network, an adaptive law with leakage term correction, and an adaptive anti-saturation control law.

[0088] In some implementations, the closed-loop control system does not simply adjust the train's motion state unidirectionally based on the actual control input. Instead, it feeds back the train's real-time operating state to the comprehensive error construction process, feeds back the comprehensive error to the RBF neural network and the adaptive law, and feeds back the adaptive law update result to the adaptive anti-saturation control law, thereby forming a closed-loop interaction relationship between trajectory tracking, unknown dynamic estimation, saturation compensation, and actual control input.

[0089] Furthermore, to verify the stability of the closed-loop control system under actuator saturation constraints and parameter uncertainties, the following Lyapunov function is selected:

[0090] in, Indicates the first The Lyapunov function corresponding to each train; Indicates the first Knowable quality during train operation; Indicates the overall error; This represents the estimation error of the correlation estimator related to the upper bound of the approximation error; It is a positive constant; This indicates the estimation error of the weight parameters in the RBF neural network; It is a positive definite matrix.

[0091] Specifically, the Lyapunov function includes a comprehensive error term, an estimation error term for the upper bound of the approximation error, and an estimation error term for the weight parameters of the RBF neural network. The comprehensive error term characterizes the error energy resulting from the combined effects of train displacement tracking error, speed tracking error, and saturated auxiliary variables; the estimation error term for the upper bound of the approximation error characterizes the residual uncertainty of the RBF neural network when performing online approximation of the complex unknown dynamics; and the estimation error term for the weight parameters of the RBF neural network characterizes the deviation between the estimated weight parameters of the RBF neural network and the ideal weight vector.

[0092] Combining the adaptive law with leakage term correction and the adaptive anti-saturation control law in step S5, the derivative of the Lyapunov function is differentiated and rearranged to ensure that the derivative of the Lyapunov function satisfies the negative semi-definite or final bounded stability inequality, thereby determining that the closed-loop control system satisfies the stability condition. Based on Lyapunov stability theory, the stability condition of the closed-loop control system is confirmed.

[0093] In some implementations, an adaptive law with leakage term correction is used to constrain the long-term drift of the weight parameter estimates and the approximation error upper bound related parameter estimates of the RBF neural network; an adaptive anti-saturation control law is used to introduce the comprehensive error, saturation auxiliary variables, and composite unknown dynamic estimates into the desired control input generation process; the two work together to constrain the comprehensive error term and parameter estimation error term in the Lyapunov function, thereby improving the stability of the closed-loop control system under actuator saturation conditions.

[0094] Furthermore, if the closed-loop control system satisfies the stability condition, the first... The displacement tracking error and speed tracking error of the first train asymptotically converge to zero or remain within a bounded range, making the first train... The actual displacement of each train approaches the desired displacement curve, and the actual speed approaches the desired speed curve. For vehicle 1, the desired operating curve is a preset desired displacement curve and desired speed curve; for vehicles 2 to 4, the desired operating curve is related to the operating state of the preceding vehicle, so that the following vehicles can maintain the desired relative position and speed consistency according to the operating state of the preceding vehicle.

[0095] Specifically, when the train's running quality changes, the basic running resistance parameters are perturbed, the additional resistance of the line changes, or there are unmodeled dynamics, the RBF neural network performs online approximation of the complex unknown dynamics, the adaptive law with leakage term correction performs bounded updates to the estimated parameters, the adaptive anti-saturation control law generates control input based on the comprehensive error, and obtains the actual control input through the actuator saturation constraint model. This enables the train to achieve stable tracking of the desired running curve even under parameter uncertainty and actuator saturation conditions.

[0096] Through the above processing, step S6 outputs the tracking control results of the train's actual displacement and actual speed on the desired running curve, realizing the train's closed-loop stable control under actuator saturation constraints and parameter uncertainty conditions.

[0097] Example 2 See Figure 2-4 To verify the feasibility and effectiveness of the train parameter uncertainty anti-saturation adaptive neural network control method provided in Example 1, this example uses MATLAB for simulation experiments.

[0098] This embodiment employs a single-mass model of the train and considers the impact of actuator saturation constraints and parameter uncertainties on position and speed tracking during train operation. Based on the longitudinal motion force analysis of the train, a longitudinal motion dynamic model of the train is established. A saturation deviation auxiliary system is introduced to dynamically compensate for actuator saturation deviations. An RBF neural network is used to approximate the complex unknown dynamics online. An adaptive law with leakage term correction and an adaptive anti-saturation control law are used to constrain the comprehensive error, thereby improving the train position tracking accuracy, speed tracking accuracy, and the stability of the closed-loop control system.

[0099] In the simulation experiment, the train running time was 2000s, the total train mass was 500t, the upper limit of the traction force under actuator saturation constraint was 60kN, and the lower limit of the braking force was -12kN. Based on the above simulation conditions, the train parameter uncertainty anti-saturation adaptive neural network control method of the present invention was simulated using MATLAB, and the train displacement response curve, speed response curve, and control input curve were obtained.

[0100] Simulation results show that, under actuator saturation constraints and parameter uncertainties, this invention enables the actual train displacement to track the desired displacement curve and the actual train speed to track the desired speed curve. Simultaneously, the overall error remains within a bounded range, and the actual control input meets the constraints of the upper limit of traction force and the lower limit of braking force. Therefore, this invention can achieve stable train position and speed tracking control under actuator saturation constraints and parameter uncertainties.

[0101] Among them, the train displacement response curve is as follows Figure 2 As shown, the train speed response curve is as follows: Figure 3 As shown, the control input curve is as follows Figure 4 As shown.

[0102] Example 3 This solution also provides a train parameter uncertainty anti-saturation adaptive neural network control system, including: The dynamic modeling module is used to obtain the first... Based on the real-time operating status, expected operating curve, and train operating parameters of the train, a longitudinal motion dynamic model of the train is established, and the uncertainties in operating quality, parameters of basic operating resistance, additional resistance, and unmodeled dynamics are identified as composite unknown dynamics. The saturation processing module is used to establish an actuator saturation constraint model based on the lower limit of braking force and the upper limit of traction force, taking the train longitudinal motion dynamics model and the desired control input as inputs, to determine the actual control input and saturation deviation, and to output saturation auxiliary variables through the saturation deviation auxiliary system. The comprehensive error construction module is used to determine the displacement tracking error and velocity tracking error by taking the real-time operating status, the expected operating curve and the saturated auxiliary variable as input, and to construct the displacement tracking error, velocity tracking error and saturated auxiliary variable as a comprehensive error; The composite unknown dynamic estimation module is used to approximate the composite unknown dynamic online using the comprehensive error and real-time operating status as inputs, and output the composite unknown dynamic estimate and the RBF neural network weight parameter estimate. An adaptive anti-saturation control module is used to construct an adaptive law with leakage term correction by taking the comprehensive error, composite unknown dynamic estimate, and RBF neural network weight parameter estimate as inputs, and construct an adaptive anti-saturation control law based on the adaptive law, comprehensive error, saturation auxiliary variable, and composite unknown dynamic estimate, and output the actual control input that satisfies the lower limit of braking force and the upper limit of traction force constraints. The stability constraint module is used to apply the actual control input to the train longitudinal motion dynamics model to form a closed-loop control system, and to determine the stability of the closed-loop control system based on the Lyapunov function, so that the train displacement tracking error and speed tracking error remain bounded and converge to a preset neighborhood.

[0103] Example 4 This embodiment also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the train parameter uncertainty anti-saturation adaptive neural network control method described in Embodiment 1.

[0104] This embodiment also provides an electronic device, which includes an input unit, a memory, a processor, and an output unit; wherein, the input unit is used to acquire the real-time operating status, desired operating curve, and train operating parameters of the i-th train; the memory is used to store a computer program; the processor is used to execute the computer program stored in the memory to implement the steps of the train parameter uncertainty anti-saturation adaptive neural network control method described in Embodiment 1; the output unit is used to output the actual control input that satisfies the lower limit of braking force and the upper limit of traction force constraints.

[0105] Furthermore, when the processor executes the computer program, it is at least used to establish a longitudinal motion dynamics model of the train, determine the complex unknown dynamics, establish an actuator saturation constraint model, generate saturation auxiliary variables, construct a comprehensive error, use an RBF neural network to approximate the complex unknown dynamics online, construct an adaptive law with leakage term correction and an adaptive anti-saturation control law, and constrain the stability of the closed-loop control system based on Lyapunov functions.

[0106] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for train parameter uncertainty-resistant adaptive neural network control, characterized in that, Includes the following steps: S1, Obtain the first Based on the real-time operating status, expected operating curve, and train operating parameters of the train, a longitudinal motion dynamics model of the train is established. The uncertainties in operating quality, the parameter uncertainties in basic operating resistance, additional resistance, and unmodeled dynamics are identified as composite unknown dynamics. The longitudinal motion dynamics model and composite unknown dynamics of the train are then output. S2. Establish an actuator saturation constraint model based on the lower limit of braking force and the upper limit of traction force. The actuator saturation constraint model is used to limit the expected control input generated by the controller, determine the actual control input and saturation deviation, and output saturation auxiliary variables through the saturation deviation auxiliary system. S3. Using the real-time operating status and the desired operating curve as input, determine the displacement tracking error and the velocity tracking error, and construct the displacement tracking error, velocity tracking error and saturated auxiliary variable into a comprehensive error; S4. Using the comprehensive error and real-time operating status as input, the RBF neural network is used to approximate the composite unknown dynamics online to obtain the estimated value of the composite unknown dynamics, and the weight estimation parameters of the RBF neural network are updated online according to the comprehensive error. S5. Using the comprehensive error, composite unknown dynamic estimate, RBF neural network weight parameter estimate and its estimated parameters as input, construct an adaptive law with leakage term correction, and construct an adaptive anti-saturation control law based on the adaptive law, comprehensive error, saturation auxiliary variable and composite unknown dynamic estimate, and output the actual control input that satisfies the lower limit of braking force and the upper limit of traction force constraints. S6. Apply the actual control input to the train longitudinal motion dynamics model to form a closed-loop control system, and constrain the stability of the closed-loop control system based on the Lyapunov function, so that the actual displacement and actual speed of the train follow the desired running curve in a bounded manner, and the displacement tracking error and speed tracking error converge to a preset neighborhood.

2. The train parameter uncertainty anti-saturation adaptive neural network control method according to claim 1, characterized in that, Step S1 includes: obtaining the first The train's running time, real-time displacement, real-time speed, desired displacement curve, desired speed curve, known mass, running mass uncertainty, control input variables, basic running resistance, additional resistance, and unmodeled dynamics are used to establish a longitudinal motion dynamics model based on Newton's second law. The running mass uncertainty, the parameter uncertainty of the basic running resistance, the additional resistance, and the unmodeled dynamics are collectively considered as the composite unknown dynamics. After equivalent processing using the train longitudinal motion dynamics model, the running mass uncertainty is combined with the parameter uncertainty of the basic running resistance, the additional resistance, and the unmodeled dynamics into the composite unknown dynamics.

3. The train parameter uncertainty anti-saturation adaptive neural network control method according to claim 1, characterized in that, Step S2 includes: denoting the desired control input calculated by the controller as the first... The desired control input of the train, after being processed by the actuator saturation constraint model, is denoted as the control input acting on the train's longitudinal motion dynamics model. The actual control input of the train; when the desired control input is less than the lower limit of braking force, the actual control input is limited to the lower limit of braking force; when the desired control input is between the lower limit of braking force and the upper limit of traction force, the actual control input is equal to the desired control input; when the desired control input is greater than the upper limit of traction force, the actual control input is limited to the upper limit of traction force; and the difference between the actual control input and the desired control input is determined as the saturation deviation.

4. The train parameter uncertainty anti-saturation adaptive neural network control method according to claim 3, characterized in that, Step S2 further includes: constructing a saturation deviation auxiliary system, wherein the saturation deviation auxiliary system takes the saturation deviation as input and generates a saturation auxiliary variable through a first-order dynamic compensation method, the saturation auxiliary variable being used to characterize the influence of the actuator saturation deviation after dynamic compensation; wherein, when the desired control input does not exceed the actuator's allowable output range defined by the lower limit of braking force and the upper limit of traction force, the saturation deviation is zero, and the saturation auxiliary variable gradually decays with the saturation deviation auxiliary system; when the desired control input exceeds the actuator's allowable output range, the saturation deviation is input to the saturation deviation auxiliary system, causing the saturation auxiliary variable to be dynamically updated according to the change of the saturation deviation.

5. The train parameter uncertainty anti-saturation adaptive neural network control method according to claim 4, characterized in that, Step S3 includes: according to the first The displacement tracking error is determined by comparing the real-time displacement with the expected displacement curve of the train. Based on the first... The speed tracking error is determined by comparing the real-time speed of the train with the expected speed curve, and the displacement tracking error, speed tracking error, and saturation auxiliary variable are combined to construct a comprehensive error. The comprehensive error is used to simultaneously characterize the tracking deviation of the actual train trajectory relative to the expected running curve, as well as the impact of the actuator saturation deviation after reconstruction by the saturation deviation auxiliary system.

6. The train parameter uncertainty anti-saturation adaptive neural network control method according to claim 5, characterized in that, Step S4 includes: representing the composite unknown dynamic as the sum of the ideal approximation term and the approximation error of the composite unknown dynamic by the RBF neural network, and using the RBF neural network to perform online estimation of the composite unknown dynamic; wherein, the input of the RBF neural network is determined by the first... The system consists of the real-time operating status of the train and the comprehensive error correlation quantity, which includes displacement tracking error, speed tracking error and saturation auxiliary variable. The basis function vector and weight vector of the RBF neural network are both n-dimensional vectors, where n represents the number of nodes in the RBF neural network. The approximation error satisfies the bounded condition, and a composite unknown dynamic estimate is obtained based on the weight estimation vector and the basis function vector.

7. The train parameter uncertainty anti-saturation adaptive neural network control method according to claim 6, characterized in that, Step S5 includes: constructing an adaptive law for the estimated weight parameters and the estimated parameters related to the upper bound of the approximation error of the RBF neural network based on the comprehensive error, and introducing a leakage term correction coefficient into the adaptive law to form an adaptive law with leakage term correction; wherein, the leakage term correction coefficient is used to suppress the parameter estimation drift of the estimated weight parameters and the estimated parameters related to the upper bound of the approximation error of the RBF neural network during long-term operation, so that the parameter estimates remain bounded.

8. The train parameter uncertainty anti-saturation adaptive neural network control method according to claim 7, characterized in that, Step S5 further includes: constructing an adaptive anti-saturation control law and generating desired control input based on the adaptive law with leakage term correction, comprehensive error, saturation auxiliary variable, composite unknown dynamic estimate, approximation error upper bound related parameter estimate and speed tracking error; wherein, vehicle 1 generates desired control input with preset desired displacement curve and desired speed curve as tracking target, and vehicles 2 to 4 generate desired control input with the desired displacement curve and desired speed curve determined by the real-time displacement of the preceding vehicle, the real-time speed of the preceding vehicle and the preset relative position relationship as tracking target, and the desired control input is processed by the actuator saturation constraint model to obtain the actual control input.

9. The train parameter uncertainty anti-saturation adaptive neural network control method according to claim 8, characterized in that, Step S6 includes: applying the actual control input to the train longitudinal motion dynamics model to form a closed-loop control system comprising the train longitudinal motion dynamics model, actuator saturation constraint model, saturation deviation auxiliary system, RBF neural network, adaptive law with leakage term correction, and adaptive anti-saturation control law; and selecting a Lyapunov function including the comprehensive error term, the estimation error term of the upper bound correlation estimate of the approximation error, and the estimation error term of the weight parameter of the RBF neural network to perform stability analysis on the closed-loop control system to determine the stability conditions satisfied by the control parameters and adaptive parameters, such that the first... The displacement tracking error and speed tracking error of the train asymptotically converge to zero or remain within a bounded range.

10. A train parameter uncertainty anti-saturation adaptive neural network control system, characterized in that, include: The dynamic modeling module is used to obtain the first... Based on the real-time operating status, expected operating curve, and train operating parameters of the train, a longitudinal motion dynamic model of the train is established, and the uncertainties in operating quality, parameters of basic operating resistance, additional resistance, and unmodeled dynamics are identified as composite unknown dynamics. The saturation processing module is used to establish an actuator saturation constraint model based on the lower limit of braking force and the upper limit of traction force, taking the train longitudinal motion dynamics model and the desired control input as inputs, to determine the actual control input and saturation deviation, and to output saturation auxiliary variables through the saturation deviation auxiliary system. The comprehensive error construction module is used to determine the displacement tracking error and velocity tracking error by taking the real-time operating status, the expected operating curve and the saturated auxiliary variable as input, and to construct the displacement tracking error, velocity tracking error and saturated auxiliary variable as a comprehensive error; The composite unknown dynamic estimation module is used to approximate the composite unknown dynamic online using the comprehensive error and real-time operating status as inputs, and output the composite unknown dynamic estimate and the RBF neural network weight parameter estimate. An adaptive anti-saturation control module is used to construct an adaptive law with leakage term correction by taking the comprehensive error, composite unknown dynamic estimate, and RBF neural network weight parameter estimate as inputs, and construct an adaptive anti-saturation control law based on the adaptive law, comprehensive error, saturation auxiliary variable, and composite unknown dynamic estimate, and output the actual control input that satisfies the lower limit of braking force and the upper limit of traction force constraints. The stability constraint module is used to apply the actual control input to the train longitudinal motion dynamics model to form a closed-loop control system, and to determine the stability of the closed-loop control system based on the Lyapunov function, so that the train displacement tracking error and speed tracking error remain bounded and converge to a preset neighborhood.

Images