High-speed train group distributed data-driven control method under fault saturation

Abstract

This invention discloses a distributed data-driven control method for high-speed train groups under fault saturation. Specifically, it involves: establishing a dynamic model of the high-speed train group using Newton's laws, followed by discretization and normalization; using pseudo-partial derivatives and the mean value theorem to linearize the nonlinear model; establishing criterion functions for the system data model parameters and control signals, and preliminarily obtaining a distributed data-driven control scheme for the train group based on optimization conditions; establishing a model for intermittent actuator faults and nonlinear input saturation during high-speed train operation, introducing multi-agent theory and parameter reset algorithms to obtain a data-driven distributed anti-saturation robust control method for high-speed train groups; testing and verification on a numerical simulation platform, and then porting it to a high-speed train automatic operation system. This invention eliminates the dependence on train models, has a simple structure, is easy to implement, and can improve the utilization rate of inter-station lines and the transportation efficiency of high-speed trains.

Description

Technical Field

[0001] This invention belongs to the field of distributed control of high-speed train groups, and particularly relates to a distributed data-driven control method for high-speed train groups under fault saturation. Background Technology

[0002] High-speed trains are the core equipment of high-speed railways and a crucial infrastructure and public transportation tool in my country. Vigorously developing high-speed train system technology is a major strategic need for my country's economy, society, and even national security, and a key support for action plans and roadmaps such as "Made in China 2025," the "Medium- and Long-Term Railway Network Plan (2016-2025)," the "Outline for Building a Transportation Power," and the "14th Five-Year Plan for Railway Science and Technology Innovation." Based on high technologies such as big data, mobile internet, and cloud computing, domestic and international rail transit equipment is undergoing a rapid development phase towards intelligence. Currently, intelligent driving has become an inevitable trend in the rail transit industry abroad. Major international train design and manufacturing companies (Siemens, Alstom, GE, etc.) are increasing their efforts in researching assisted driving, automatic driving, and driverless driving. Therefore, conducting research on key technologies for intelligent driving for the next generation of high-speed trains is an inevitable trend in the development of my country's rail transit.

[0003] On the one hand, to further enhance the passenger carrying capacity of passenger railways, the EU has clearly identified multi-train virtual cooperative control based on train-to-train communication as a key technology for enhancing the resilience and interoperability of passenger railways. Under the high-speed rail cooperative control framework based on direct train-to-train communication, trains can directly communicate with trains within their communication range that are "topologically adjacent." Multi-train cooperative control technology can coordinate the operating status of multiple trains, consider the safe distance between adjacent trains, and treat the dynamic operation of multiple trains as a whole, uniformly formulating control strategies, thereby achieving rapid execution of dispatching commands and robustness against external interference for the railway network. Train cooperative control based on the train-to-train communication information interaction structure, as the core technology of the next-generation train control system, is an effective means to improve the passenger carrying capacity of the high-speed rail system and to safely, quickly, and efficiently adjust trains in case of emergencies. Some technologies have been successfully applied in urban rail transit, but are still lacking in high-speed rail systems.

[0004] On the other hand, most existing high-speed train control methods rely on train dynamics models or partial model information to design control strategies. However, in reality, the internal mechanisms of train operation are complex, and the external environment is highly variable and uncertain, making it difficult to establish accurate dynamic models. In recent years, with the rapid development of science and technology, especially information science and technology, transportation systems have undergone significant changes, and the requirements for the quality of high-speed train operation are becoming increasingly stringent, making the application of control theories and methods based on accurate mathematical models of the controlled object increasingly difficult in practice. Furthermore, high-speed trains constantly generate a large amount of process data during operation, which implicitly contains information such as changes in system state. How to effectively utilize this data and the knowledge it contains to achieve optimized control of the system and production process under the condition that it is difficult to establish a relatively accurate model of the controlled system has become an urgent problem to be solved by the control theory community. Therefore, researching and developing data-driven control theories and methods is an inevitable choice for the development and application of control theory in the new era, and has significant theoretical and practical implications.

[0005] Model-free adaptive control is a typical data-driven control method that uses only the input and output data of the controlled system for controller design and analysis. It can realize parameter adaptive control and structural adaptive control of unknown nonlinear controlled systems, freeing the controller design from dependence on the mathematical model of the controlled system and the theoretical problems of bionic twins driven by various models. It provides a brand-new control theory and method for control theory research and practical application. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a distributed data-driven control method for high-speed train groups under fault saturation.

[0007] The present invention provides a distributed data-driven control method for high-speed train groups under fault saturation, comprising the following steps:

[0008] Step 1: Establish a normalized discrete model of the high-speed train group's dynamics. Consider a group of high-speed trains running on the same railway line. Based on Newton's theorem, establish the dynamic model of the j-th train:

[0009] v(k+1,j)=v(k,j)+T(u(k,j)-F(k,j))

[0010] s(k+1,j)=s(k,j)+Tv(k,j)+0.5T 2 (u(k,j)-F(k,j))

[0011] Where k is the sampling time, j is the train number, and T is the sampling period; the variables v(k,j), s(k,j), u(k,j) and F(k,j) represent the speed, position, control input and total resistance of train j, respectively.

[0012] Step 2: Establish an equivalent linearization model suitable for controller design.

[0013] The train group dynamics model satisfies the following: 1. Except for finite sampling time points, the partial derivative of the function f(·) with respect to u(k,j) is continuous; 2. Except for finite sampling time points, the function f(·) satisfies the generalized Lipschitz condition, that is, for Δu(k,j)≠0, for any k, |Δs(k+1,j)|≤b|Δu(k,j)|, where Δu(k,j)=u(k,j)-u(k-1,j) and Δs(k+1,j)=s(k+1,j)-s(k,j), and b>0 represents a constant.

[0014] Constructing nonlinear terms:

[0015] f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k-1,j),u(k-1,j))

[0016] And let:

[0017]

[0018] Then, by reviewing the definition of Δs(k,j), Assumption 1, and Cauchy's Mean Value Theorem, we obtain the following relationship:

[0019]

[0020] in Let f(·) represent the partial derivative of f(·) with respect to u(k,j) at a point between f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k-1,j),u(k-1,j)) and f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k,j),u(k-1,j)).

[0021] For each sampling time k, consider the following data equation κ(k,j) with variables:

[0022]

[0023] Since Δu(k,j)≠0, there exists a unique solution κ for the above equation. * (k,j); Definition Then, the nonlinear model is written as an equivalent linearized model Δs(k+1,j)=φ(k,j)Δu(k,j); the boundedness of φ(k,j) is proved according to hypothesis 2; therefore, the nonlinear model of the high-speed train group with unknown dynamics is transformed into the above equivalent linear data model by means of φ(k,j).

[0024] Step 3: Design a data-driven control scheme for high-speed train groups.

[0025] Consider the following criterion function with respect to the control input u(k,j):

[0026] J(u(k,j))=|s d (k+1)-s(k+1,j)| 2 +λ|u(k,j)-u(k-1,j)| 2

[0027] Where λ is a positive weighting factor that restricts the variation of u(k,j).

[0028] Substituting the obtained linear data model into the criterion function and minimizing it yields the control algorithm:

[0029]

[0030] Design the following criterion function to estimate the value of φ(k,j):

[0031]

[0032] Where μ > 0 is the weighting factor.

[0033] Minimizing the above criterion function yields:

[0034]

[0035] in, Let φ(k,j) be the estimated value, and η∈(0,1] be the step size factor.

[0036] Step 4: Establish a model of intermittent actuator failures during high-speed train operation:

[0037] s fault (k,j)=α(k,j)s(k,j)

[0038] Among them, s fault (k,j) represents the actual usable train position information under actuator failure. α(k,j)={0,1} is a random Bernoulli variable representing the health status of the actuator. If α(k,j)=1, it means that the actuator is in good health at the current moment. If α(k,j)=0, it means that the data at the current moment is lost due to the failure. s(k,j) is the actual position signal output by the high-speed train.

[0039] The traction power of high-speed trains is provided by the traction net above them, and this energy exhibits saturation characteristics due to the limitation of net voltage; specifically, the saturation model for the control input is constructed as follows:

[0040]

[0041] Among them, u sat (k,j) represents the actual control input signal that the traction network can provide, and u(k,j) represents the desired control input signal calculated by the controller. min u is the minimum energy that the traction network can provide. max This is the maximum energy that the traction net can provide.

[0042] A distributed data-driven control method for high-speed train groups is established based on multi-agent theory. Graph theory is used to model the communication topology of J train groups, where each train is considered an independent agent and can only communicate with adjacent trains. G = (V, E, A) is defined as a weighted directed graph of order J, where V = {1, 2, ..., J} represents J vertices, E ∈ V × V is the set of edges, and A = [a...]. ij ] J×J Let represent the adjacency matrix, where all elements are non-negative; if information from train i can be received by train j, then is a. ij =1, otherwise a ij =0; the neighbors of train i are represented as N. i =(j∈V:((j,i)∈E)); Furthermore, the desired trajectory is generated by the virtual leader train, which is numbered 0; therefore, the extended directed graph is The order is J+1, where and These are the corresponding edge matrix and adjacency matrix, respectively. The distributed error is defined as follows:

[0043]

[0044] J j Let d represent the neighborhood set of train j. j In {0,1}, d represents the communication relationship between the leading train 0 and the following train j; if train j can access the required trajectory from train 0, then d j =1, otherwise d j =0; S represents the minimum safe distance between two adjacent trains, s d (k)-jS represents the desired trajectory of train j.

[0045] Considering intermittent actuator failures and nonlinear saturation of control inputs, the following distributed data-driven robust control scheme for high-speed train groups is designed:

[0046]

[0047]

[0048]

[0049] Where, Δu sat (k-1,j)=u sat (k-1,j)-u sat (k-2,j), Δs fault (k,j)=s fault (k,j)-s fault (k-1,j); the second equation in the formula is about the parameter The reset algorithm, where ε is a small positive integer.

[0050] Step 5: Based on the control scheme, the control algorithm is first verified on a numerical simulation platform, and a complete set of controller parameters with optimal train operation performance is obtained. Furthermore, this scheme is then ported to a high-speed train automatic operation system.

[0051] Furthermore, in step 1, according to the Davis equation, F(k,j) is further described as: F(k,j) = F b (k,j)+F a (k,j), where F b (k,j) represents the basic operating resistance:

[0052] F b (k,j)=c0(k,j)+c v (k,j)v(k,j)+c t (k,j)v(k,j) 2

[0053] Where c0(k,j), c v (k,j) and c t (k,j) represents the unknown time-varying drag coefficient.

[0054] F a (k,j) represents the additional resistance, which satisfies:

[0055] F a (k,j)=M(j)g sin(θ(s(k,j)))+f c (s(k,j))+f t (s(k,j))

[0056] Where M(j), g, and θ(s(k,j)) represent the train mass, gravitational acceleration, and track gradient, respectively; the unknown nonlinear function f c (·) and f t (·) is related to curves and tunnels.

[0057] The train model and operating resistance are only related to the historical speed and position data of the train group, and involve complex nonlinear and unknown dynamics; therefore, the above dynamic model can be rewritten in the following compact form:

[0058] s(k+1,j)=f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k,j),u(k-1,j))

[0059] Where f(·) is an unknown nonlinear function.

[0060] Furthermore, the selection of controller parameters in step 5 needs to meet the following conditions: μ,λ>0, 0<η≤1, 0<ρ<0.5, and ε is set to 10. -5 Or 10 -6 .

[0061] The beneficial technical effects of this invention are as follows:

[0062] I. This invention proposes a data-driven modeling method for high-speed train groups, resulting in a dynamic linearized data model of the train group. This modeling method and model utilize only the readily available position signals and control input information of the high-speed train group, eliminating the dependence on prior information of the system model. It overcomes the difficulty in establishing accurate mathematical models of train dynamics under actual high-speed train operating conditions and can simultaneously provide new ideas for modeling other typical nonlinear systems.

[0063] Second, based on the constructed equivalent linearized data model, this invention designs a distributed data-driven control scheme suitable for high-speed train groups, consisting of a parameter estimation algorithm, a parameter reset algorithm, and a control input algorithm. This scheme does not require knowledge of the precise nonlinear model of the high-speed train, and its design method is simple and easy to apply in practical engineering.

[0064] Third, the distributed control scheme for high-speed train groups proposed in this invention relies on a reliable direct communication network between trains. By utilizing multi-train distributed tracking error signals within a directed communication topology to replace the single-train tracking error signals used in traditional algorithms, the relative distance between high-speed trains can be reduced, increasing the operating density of train groups within the section and further improving the line's transportation efficiency and carrying capacity. Attached Figure Description

[0065] Figure 1 This is a diagram illustrating the safe distance between adjacent vehicles.

[0066] Figure 2 This is a diagram of the fixed communication topology between high-speed train groups.

[0067] Figure 3 This is a trajectory diagram showing the expected speed and position of a train group.

[0068] Figure 4 The speed and position response curves are shown under a distributed data-driven control scheme.

[0069] Figure 5 The speed and position tracking error curves for the train group.

[0070] Figure 6 This section compares the performance of several typical control methods. Detailed Implementation

[0071] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0072] To ensure the safety of high-speed trains traveling at high speeds, sufficient safety distances should be maintained between high-speed train groups, such as... Figure 1 As shown in Figure S. During multi-vehicle cooperative control, the minimum distance between adjacent vehicles must be greater than this safe distance. In vehicle-to-vehicle cooperative control mode, high-speed trains have wireless transmission channels that enable communication between them, such as... Figure 2 As shown, train 0 serves as a virtual train, providing the desired trajectory signal required by subsequent trains. This invention considers reliable bidirectional communication only between adjacent trains in the simulation, resulting in a strongly connected communication topology.

[0073] The present invention provides a distributed data-driven control method for high-speed train groups under fault saturation, comprising the following steps:

[0074] Step 1: Establish a normalized discrete model of the high-speed train group's dynamics. Consider a group of high-speed trains running on the same railway line. Based on Newton's theorem, establish the dynamic model of the j-th train:

[0075] v(k+1,j)=v(k,j)+T(u(k,j)-F(k,j))

[0076] s(k+1,j)=s(k,j)+Tv(k,j)+0.5T 2 (u(k,j)-F(k,j))

[0077] Where k is the sampling time, j is the train number, and T is the sampling period; the variables v(k,j), s(k,j), u(k,j) and F(k,j) represent the speed, position, control input and total resistance of train j, respectively.

[0078] According to the Davis equation, F(k,j) can be further described as: F(k,j) = F b (k,j)+F a (k,j), where F b (k,j) represents the basic operating resistance:

[0079] F b (k,j)=c0(k,j)+c v (k,j)v(k,j)+c t (k,j)v(k,j) 2

[0080] Where c0(k,j), c v (k,j) and c t (k,j) represents the unknown time-varying drag coefficient.

[0081] F a (k,j) represents the additional resistance, which satisfies:

[0082] F a (k,j)=M(j)gsin(θ(s(k,j)))+f c (s(k,j))+f t (s(k,j))

[0083] Where M(j), g, and θ(s(k,j)) represent the train mass, gravitational acceleration, and track gradient, respectively; the unknown nonlinear function f c (·) and f t (·) is related to curves and tunnels.

[0084] The train model and operating resistance are only related to the historical speed and position data of the train group, and involve complex nonlinear and unknown dynamics; therefore, the above dynamic model can be rewritten in the following compact form:

[0085] s(k+1,j)=f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k,j),u(k-1,j))

[0086] Where f(·) is an unknown nonlinear function.

[0087] Step 2: Establish an equivalent linearization model suitable for controller design.

[0088] The train group dynamics model satisfies the following: 1. Except for finite sampling time points, the partial derivative of the function f(·) with respect to u(k,j) is continuous; 2. Except for finite sampling time points, the function f(·) satisfies the generalized Lipschitz condition, that is, for Δu(k,j)≠0, for any k, |Δs(k+1,j)|≤bΔu(k,j)|, where Δu(k,j)=u(k,j)-u(k-1,j) and Δs(k+1,j)=s(k+1,j)-s(k,j), and b>0 represents a constant.

[0089] Constructing nonlinear terms:

[0090] f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k-1,j),u(k-1,j))

[0091] And let:

[0092]

[0093] Then, by reviewing the definition of Δs(k,j), Assumption 1, and Cauchy's Mean Value Theorem, we obtain the following relationship:

[0094]

[0095] in Let f(·) represent the partial derivative of f(·) with respect to u(k,j) at a point between f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k-1,j),u(k-1,j)) and f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k,j),u(k-1,j)).

[0096] For each sampling time k, consider the following data equation κ(k,j) with variables:

[0097]

[0098] Since Δu(k,j)≠0, there exists a unique solution κ for the above equation. * (k,j); Definition Then, the nonlinear model is written as an equivalent linearized model Δs(k+1,j)=φ(k,j)Δu(k,j); the boundedness of φ(k,j) is proved according to hypothesis 2; therefore, the nonlinear model of the high-speed train group with unknown dynamics is transformed into the above equivalent linear data model by means of φ(k,j).

[0099] Step 3: Design a data-driven control scheme for high-speed train groups.

[0100] Consider the following criterion function with respect to the control input u(k,j):

[0101] J(u(k,j))=|s d (k+1)-s(k+1,j)| 2 +λ|u(k,j)-u(k-1,j)| 2

[0102] Where λ is a positive weighting factor that restricts the variation of u(k,j).

[0103] Substituting the obtained linear data model into the criterion function and minimizing it yields the control algorithm:

[0104]

[0105] Design the following criterion function to estimate the value of φ(k,j):

[0106]

[0107] Where μ>0 is the weighting factor.

[0108] Minimizing the above criterion function yields:

[0109]

[0110] in, Let φ(k,j) be the estimated value, and η∈(0,1] be the step size factor.

[0111] Step 4: Establish a model of intermittent actuator failures during high-speed train operation:

[0112] s fault (k,j)=α(k,j)s(k,j)

[0113] Among them, s fault (k,j) represents the actual usable train position information under actuator failure. α(k,j)={0,1} is a random Bernoulli variable representing the health status of the actuator. If α(k,j)=1, it means that the actuator is in good health at the current moment. If α(k,j)=0, it means that the data at the current moment is lost due to the failure. s(k,j) is the actual position signal output by the high-speed train.

[0114] The traction power of high-speed trains is provided by the traction net above them, and this energy exhibits saturation characteristics due to the limitation of net voltage; specifically, the saturation model for the control input is constructed as follows:

[0115]

[0116] Among them, u sat (k,j) represents the actual control input signal that the traction network can provide, and u(k,j) represents the desired control input signal calculated by the controller. min u is the minimum energy that the traction network can provide. max This is the maximum energy that the traction net can provide.

[0117] A distributed data-driven control method for high-speed train groups is established based on multi-agent theory. Graph theory is used to model the communication topology of J train groups, where each train is considered an independent agent and can only communicate with adjacent trains. G = (V, E, A) is defined as a weighted directed graph of order J, where V = {1, 2, ..., J} represents J vertices, E ∩ V × V is the set of edges, and A = [a...]. ij ] J×J Let represent the adjacency matrix, where all elements are non-negative; if information from train i can be received by train j, then is a. ij =1, otherwise a ij =0; the neighbors of train i are represented as N. i =(j∈V:((j,i)∈E)); Furthermore, the desired trajectory is generated by the virtual leader train, which is numbered 0; therefore, the extended directed graph is The order is J+1, where and These are the corresponding edge matrix and adjacency matrix, respectively. The distributed error is defined as follows:

[0118]

[0119] J j Let d represent the neighborhood set of train j. j In {0,1}, d represents the communication relationship between the leading train 0 and the following train j; if train j can access the required trajectory from train 0, then d j =1, otherwise d j =0; S represents the minimum safe distance between two adjacent trains, s d (k)-jS represents the desired trajectory of train j.

[0120] Considering intermittent actuator failures and nonlinear saturation of control inputs, the following distributed data-driven robust control scheme for high-speed train groups is designed:

[0121]

[0122]

[0123]

[0124] Where, Δu sat (k-1,j)=u sat (k-1,j)-u sat (k-2,j), Δs fault (k,j)=s fault (k,j)-s fault (k-1,j); the second equation in the formula is about the parameter The reset algorithm uses ε as a small positive constant. Generally, the controller parameters should be selected to meet the following conditions: μ,λ > 0, 0 < η ≤ 1, 0 < ρ < 0.5, and ε is usually set to 10. -5 Or 10 -6 .

[0125] Step 5: Based on the control scheme, the control algorithm is first verified on a numerical simulation platform, and a complete set of controller parameters with optimal train operation performance is obtained. Furthermore, this scheme is then ported to a high-speed train automatic operation system.

[0126] The speed and position trajectory required for high-speed train groups, such as Figure 3 As shown, the maximum acceleration is 0.5 m / s². 2 The total running time was 2000 seconds, and the total distance was 172.15 kilometers. During the operation, the train group experienced acceleration, cruising, and deceleration to 0, which fully tested the performance of the control scheme.

[0127] Figure 4 The speed and position response curves of the 1st, 3rd, and 5th high-speed trains under the distributed data-driven control method are shown. The results demonstrate that, under the distributed data-driven control method of this invention, the speed and position curves of the high-speed train group can quickly track a given desired trajectory curve.

[0128] Figure 5 The speed and position tracking error curves of the train group are shown. It can be seen that the final speed and position tracking errors are 0.02 m / s and 1.5 × 10⁻⁶ m / s, respectively. -3 The train tracking performance, which fluctuates around m, can meet the operational requirements of high-speed trains.

[0129] To facilitate the performance verification of the control scheme proposed in this invention, it was compared with several typical control methods, including PID control and sliding mode control. The results show that all schemes can converge the tracking error to near zero in a relatively short time. However, closer observation reveals that the distributed data-driven control method involved in this invention minimizes the oscillation range of the tracking error, i.e., it exhibits the best tracking performance.

Claims

1. A distributed data-driven control method for high-speed train group under fault saturation, characterized in that, Comprising the following steps: Step 1: Establish a high-speed train group dynamics normalized discrete model; considering a group of high-speed trains running on the same railway line, a dynamic model of the jth train is established according to Newton's theorem: v(k+1,j) = v(k,j) + T(u(k,j) - F(k,j)) s(k + 1, j) = s(k, j) + Tv(k, j) + 0.5T 2 (u(k, j) - F(k, j)) Wherein, k is the sampling time, j is the train number, T is the sampling period; variables v(k,j), s(k,j), u(k,j) and F(k,j) represent the speed, position, control input and total resistance of train j, respectively; Step 2: Establish an equivalent linearization model suitable for controller design; The train group dynamics model satisfies:

1. Except for a limited number of sampling time points, the partial derivative of the function f(·) with respect to u(k,j) is continuous; 2. Except for a limited number of sampling time points, the function f(·) satisfies the generalized Lipschitz condition, that is, for Δu(k,j)≠0, for any k, |Δs(k+1,j)|≤b|Δu(k,j)|, where Δu(k,j)=u(k,j)-u(k-1,j), Δs(k+1,j)=s(k+1,j)-s(k,j), and b>0 represents a constant; Construct the nonlinear term: f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k-1,j),u(k-1,j)) And let: Then, by reviewing the definition of Δs(k,j), assumption 1 and the Cauchy mean value theorem, the following relationship is obtained: wherein denotes the value of the partial derivative of f(·) with respect to u(k,j) at some point between f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k-1,j),u(k-1,j)) and f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k,j),u(k-1,j)); For each sampling time k, consider the following data equation with variables κ(k,j): Since Δu(k,j)≠0, there is a unique solution κ for the above equation * (k,j) ; define After that, the nonlinear model is written as an equivalent linearization model Δs(k+1,j) = φ(k,j)Δu(k,j); the boundedness of φ(k,j) is proved according to assumption 2; therefore, the high-speed train group nonlinear model with unknown dynamics is converted into the above equivalent linear data model with the help of φ(k,j). Step 3: Design a high-speed train group data-driven control scheme; Consider the following criterion function for the control input u(k,j): J(u(k, j)) = |s d (k+1)- s(k+1, j)| 2 + λ |u(k, j) - u(k-1, j)| 2 Where λ is a positive weighting factor that limits the change of u(k,j); The resulting linear data model is substituted into the criterion function and minimized to obtain the control algorithm: Design the following criterion function to estimate the value of φ(k,j): Where μ>0 is a weight factor; Minimizing the above criterion function gives: wherein φ(k, j) is an estimate of φ(k, j), and η e (0, 1] is a step size factor. Step 4: Establish an actuator fault model that occurs intermittently during high-speed train operation: s fault (k, j) = a(k, j)s(k, j) where s faul t(k, j) represents the actual available train position information under actuator failure, a(k, j) = {0, 1} is a random Bernoulli variable representing the health status of the actuator; if a(k, j) = 1, it represents that the actuator is in good health at the current time, and if a(k, j) = 0, it represents that the data at the current time is lost due to failure; s(k, j) is the actual output position signal of the high-speed train. The traction power of a high-speed train is provided by the traction network above it, and this energy exhibits saturation characteristics due to network pressure limitations; Specifically, the saturation model for the control input is constructed as follows: wherein u sat (k, j) is the control input signal that the traction network is actually able to provide, u(k, j) is the desired control input signal calculated by the controller, u min is the minimum energy that the traction network is able to provide, u max is the maximum energy that the traction network is able to provide; A distributed data-driven control method for high-speed train platoon is established based on multi-agent theory. The communication topology of J train platoon is modeled by graph theory, where each train is regarded as an independent agent and can only communicate with adjacent trains. G=(V, E, A) is defined as a weighted directed graph with order J, where V={1, 2, …J} represents J vertices, E∈V×V is the edge set, and A=[a ij ] J×J is the adjacency matrix, where all elements are non-negative; a ij =1 if the information from train i can be received by train j, otherwise a ij =0; the neighbors of train i are denoted as N i =(j∈V:((j,i)∈E)); in addition, the desired trajectory is generated by a virtual leader train, whose number is 0; therefore, the extended directed graph is with order J+1, where and are the corresponding edge matrix and adjacency matrix, respectively; the distributed error is defined as follows: where J j denotes the neighborhood set of train j, d j ∈{0,1} denotes the communication relation between the lead train 0 and the following train j; if train j can access the required track from train 0, then d j = 1, otherwise d i = 0; S denotes the minimum safety distance between two adjacent trains, s d (k) - jS denotes the desired trajectory of train j; Considering the intermittent actuator fault and the nonlinear saturation of the control input, a distributed data-driven anti-saturation robust control scheme suitable for high-speed train groups is designed as follows: where Δu sat (k-1, j) = u sat (k-1, j) - u sat (k-2, j), Δs fault (k, j) = s fault (k, j) - s fault (k-1, j); where the second equation is a reset algorithm for the parameter ε is a small positive number. Step 5: On the basis of the control scheme, the control algorithm is first verified on a numerical simulation platform, and a complete set of controller parameters that optimize train operation performance is adjusted; Further, the scheme is transplanted to the high-speed train automatic operation system.

2. The distributed data-driven control method for high-speed train group under fault saturation according to claim 1, characterized in that, In step 1, F(k,j) is further described according to Davis equation as: F(k,j) = F b (k,j) + F a (k,j), where F b (k,j) represents the basic running resistance: F b (k, j) = c0(k, j) + c v (k, j) v(k, j) + c t (k, j) v(k, j) 2 where c0(k, j), c v (k, j) and c t (k, j) represent unknown time-varying drag coefficients; F a (k, j) represents an additional resistance, which satisfies: F a (k, j) = M(j) g sin(θ(s(k, j))) + f c (s(k, j)) + f t (s(k, j)) where M(j), g and θ(s(k, j)) are the train mass, the gravity acceleration and the slope of the track grade, respectively; f c (·) and f t (·) are related to curves and tunnels; The train model and running resistance are only related to the historical speed and position data of the train group, and contain complex nonlinearities and unknown dynamics; Therefore, the above dynamic model can be rewritten in the following compact form: s(k+1,j) = f(s(k,j),s(k-1,j),v(k,j),v(k-1,j),u(k,j),u(k-1,j)) Where f() is an unknown nonlinear function.

3. The distributed data-driven control method for high-speed train group under fault saturation according to claim 1, characterized in that, The selection of the controller parameters in step 5 needs to satisfy the following conditions: μ, λ > 0, 0 < η ≤ 1, 0 < ρ < 0.5, ε is set to 10 -5 or 10 -6 .

Images