High-speed train intelligent model prediction control method and system

Abstract

The invention provides a high-speed train intelligent model prediction control method and system, and belongs to the technical field of train operation control, and the method comprises the steps: carrying out the prediction of the operation state prediction data of a future high-speed train through employing a pre-trained long-short term memory neural network based on historical operation state data; based on the running state prediction data of the future high-speed train, in combination with the current real-time running state data, performing state estimation by using a fixed lag Kalman smoother to obtain an optimized running state estimation result; based on the optimal control input sequence, the traction force and braking force of the train are controlled and adjusted in real time, and it is ensured that the train runs according to the expected trajectory. According to the method, the problems of nonlinear dynamic and non-Gaussian noise interference are solved, and the state estimation precision is improved; by introducing the multi-particle model, the position and speed dynamic state of each particle of the train is accurately described, the physical consistency of the control strategy is improved, the physical authenticity of the model is enhanced, and the high-precision tracking of the position and speed of the train is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of train operation control, and particularly relates to a high-speed train intelligent model predictive control method and system based on a long short-term memory neural network fixed-lag Kalman smoother and multi-particle modeling. Background Art

[0002] The operation control of high-speed trains is a core issue in rail transit systems, and its performance directly affects the safety, comfort, and energy efficiency of trains. However, due to the complex train operation environment, factors such as track irregularities, aerodynamic effects, and wheel-rail interactions make the train's dynamic system exhibit significant nonlinear characteristics. At the same time, due to sensor measurement errors, external disturbances, etc., the train's state observation data is usually disturbed by non-Gaussian noise, posing challenges to accurate state estimation and control. To address these issues, researchers have proposed various state estimation and control methods. Traditional Kalman filters (KF) and their extended forms, such as extended Kalman filters (EKF), unscented Kalman filters (UKF), and particle filters (PF), etc., have been widely used for train state estimation. These methods estimate the train state based on a dynamic system model and provide feedback for the train control system. However, they have limitations in dealing with nonlinear and non-Gaussian noise problems. For example, EKF uses first-order Taylor expansion to linearize the system, which is prone to introducing linearization errors, while UKF, although improving the nonlinear adaptability, still has estimation errors in non-Gaussian noise environments. The PF method estimates the non-Gaussian distribution through sampling, but has a large computational amount and insufficient real-time performance, making it difficult to meet the requirements of high-speed train control.

[0003] In terms of control methods, model predictive control (MPC) has been widely applied in the field of rail transit to optimize the train operation trajectory and improve energy efficiency. MPC optimizes the control input within a future period of time to make the train operation state as close as possible to the desired trajectory. However, the control accuracy of MPC depends on accurate state estimation. Traditional MPC methods usually rely on single-particle modeling and assume that the system noise is Gaussian distributed, making it difficult to handle the nonlinear dynamic characteristics and complex noise environment of high-speed trains. In recent years, researchers have tried to combine deep learning techniques to improve train state estimation and control. For example, the long short-term memory network (LSTM) has been used for train operation state prediction due to its excellent time series modeling ability. LSTM can learn the dynamic characteristics of trains from historical data and predict future states, enhancing the system's adaptability to nonlinear and non-Gaussian noise. However, the pure LSTM method lacks physical constraints and it is difficult to ensure the physical rationality of the prediction results. Therefore, how to combine data-driven methods with model-driven methods to give full play to their respective advantages has become an important research direction.

[0004] Currently, existing methods include MPC methods based on EKF / UKF, PF-based MPC methods, and state estimation methods based on LSTM. The EKF / UKF-MPC method uses EKF or UKF to estimate the train state and inputs the estimated value into MPC for optimal control. However, the linearization error of EKF is relatively large, and although UKF improves the nonlinear processing ability, there is still an estimation bias in the non-Gaussian noise environment, affecting the MPC control accuracy. The PF-MPC method estimates the train state through PF and inputs the result into MPC for optimal control. Although PF can handle non-Gaussian noise, its computational complexity is relatively high, and a large number of particles are required to ensure the estimation accuracy, resulting in a heavy computational burden and difficulty in meeting the real-time control requirements of high-speed trains. In addition, PF has the problem of particle degeneracy, affecting the estimation accuracy. In recent years, researchers have used LSTM to model and predict the state of high-speed trains. LSTM can capture the nonlinear dynamic characteristics of the system and predict future states through historical data, enhancing the system's adaptability to complex environments. However, although LSTM can model nonlinear and non-Gaussian noise, it is difficult to ensure the physical rationality of the estimation by using deep learning methods alone. In addition, the prediction performance of LSTM depends on a large amount of training data, and its generalization ability is limited.

[0005] In summary, the deficiencies of existing traditional methods include: Insufficient processing of nonlinear and non-Gaussian noise: Traditional Kalman filters (such as EKF, UKF) rely on linear approximation and Gaussian noise assumptions and cannot effectively handle the nonlinear dynamics and non-Gaussian noise of high-speed train systems, resulting in a decrease in state estimation and control accuracy. High computational complexity: Particle filters (PF) require a large number of particle calculations, have the problem of particle degeneracy, and have poor real-time performance; high-order nonlinear filters (such as CKF) have high computational complexity and are difficult to meet the real-time control requirements. Accuracy loss due to model simplification: Existing methods often linearize the high-speed train dynamics model, ignoring high-order nonlinear characteristics (such as air resistance, coupling spring nonlinearity), and the deviation between the model and the actual dynamics is relatively large. Lack of support for multi-particle models: Traditional control methods do not fully combine multi-particle models and cannot accurately describe the dynamic coupling relationship between train cars, and the generalization ability of control strategies is weak. Summary of the Invention

[0006] The purpose of the present invention is to provide a high-speed train intelligent model predictive control method and system based on long short-term memory neural network fixed-lag Kalman smoother and multi-particle modeling to solve at least one of the technical problems existing in the above background technology.

[0007] To achieve the above purpose, the present invention adopts the following technical solutions:

[0008] In the first aspect, the present invention provides a high-speed train intelligent model predictive control method, including:

[0009] Obtain the real-time operation status data and historical operation status data of the high-speed train;

[0010] Based on the historical operation status data, use the pre-trained long short-term memory neural network to predict the operation status prediction data of the future high-speed train;

[0011] Based on the operation status prediction data of the future high-speed train, combined with the current real-time operation status data, use the fixed-lag Kalman smoother for state estimation to obtain the optimized operation status estimation result; among them, use the current real-time operation status data and the operation status prediction data of the future high-speed train for forward filtering to generate a preliminary state estimation; use the future observation data for backward smoothing to optimize the preliminary state estimation at the current moment;

[0012] Based on the optimized preliminary state estimation at the current moment, predict the state change of the train in the future for a period of time; generate the optimal control input sequence by optimizing the objective function; among them, the objective function considers the state error and the constraints of the control input;

[0013] Based on the optimal control input sequence, adjust the traction force and braking force of the train in real time to ensure that the train runs according to the desired trajectory.

[0014] As a further limitation of the first aspect of the present invention, in the fixed-delay Kalman smoother, a two-way filtering structure is established. In the forward filtering, the time information of the high-speed train is collected. In the backward smoothing, the operation status prediction data of the future high-speed train is used for early observation to achieve the optimal estimation of the current state.

[0015] As a further limitation of the first aspect of the present invention, model predictive control MPC needs to predict the future system state. In the next p control cycles, the predicted state of the HST system is expressed as:

[0016] X k+p =[X(k + 1|k), X(k + 2|k), …, X(k + p|k)] T ;

[0017] p is the prediction horizon;

[0018] When predicting the future state of the dynamic system, it is also necessary to know the control input U within the prediction horizon k+p ;

[0019] U k+p =[U(k|k), U(k + 1|k), …, U(k + p - 1|k)] T ;

[0020] Based on X k+p and U k+p, determine the optimization problem expression for controlling the HST system.

[0021] As a further limitation of the first aspect of the present invention, determining the optimization problem expression for controlling the HST system includes: First, sequentially predict the states of the HST in the next p control cycles; then, design the objective function of the HST control to determine the optimal control input U k+p , such that the predicted HST state vector is closer to the desired state within the prediction horizon p This is an open-loop optimal control problem. For this purpose, define the objective function by calculating the cumulative error between the predicted state vector and the desired state, and add control constraints to reduce the control amplitude.

[0022] As a further limitation of the first aspect of the present invention, the model constraints include: the range of each speed v i is limited to 0 ≤ v i ≤ v lim ; the displacement difference between adjacent positions x i and x i-1 is limited to -(τ - δ) ≤ x i - x i-1 ≤ -(τ + δ), ensuring sufficient safety distance between mass points; the control input u i of each point mass is subject to . v lim is the speed limit of each point mass, δ is the hard safety constraint between HST point masses, is the maximum braking force, is the maximum traction force; these constraints define the feasible set of control variables, ensuring that the control input u i always follows the physical limitations of the system during the optimization process, thereby ensuring the stability and safety of the HST system.

[0023] As a further limitation of the first aspect of the present invention, the objective function and constraints are combined to transform the MPC problem into a standard quadratic programming problem. In each control cycle, the objective function and constraints are redefined based on the current system state. By solving the QP problem, the optimal control input sequence U k+p within the prediction horizon is obtained, and the first control input is extracted and applied to the current moment k.

[0024] Second aspect, the present invention provides an intelligent model predictive control system for high-speed trains, including:

[0025] An acquisition module for acquiring the real-time operation state data and historical operation state data of the high-speed train;

[0026] A prediction module, configured to predict future operation state prediction data of a high-speed train based on historical operation state data by using a pre-trained long short-term memory neural network;

[0027] A state estimation module, configured to perform state estimation by using a fixed-lag Kalman smoother based on the future operation state prediction data of the high-speed train and in combination with current real-time operation state data to obtain an optimized operation state estimation result; wherein, forward filtering is performed by using the current real-time operation state data and the future operation state prediction data of the high-speed train to generate a preliminary state estimation; and reverse smoothing is performed by using future observation data to optimize the preliminary state estimation at the current moment;

[0028] An optimization module, configured to predict the state change of the train within a future period of time based on the optimized preliminary state estimation at the current moment; and generate an optimal control input sequence by optimizing an objective function; wherein, the objective function takes into account state errors and constraints of control inputs;

[0029] A control module, configured to adjust the traction force and braking force of the train in real time based on the optimal control input sequence to ensure that the train runs along a desired trajectory.

[0030] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions, and when the computer instructions are executed by a processor, the intelligent model predictive control method for a high-speed train based on a fixed-lag Kalman smoother and multi-particle modeling as described in the first aspect is implemented.

[0031] In a fourth aspect, the present invention provides a computer device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the intelligent model predictive control method for a high-speed train based on a fixed-lag Kalman smoother and multi-particle modeling as described in the first aspect.

[0032] In a fifth aspect, the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes instructions for implementing the intelligent model predictive control method for a high-speed train based on a fixed-lag Kalman smoother and multi-particle modeling as described in the first aspect.

[0033] Advantages of the present invention: The intelligent model predictive control method for high-speed trains based on the long short-term memory neural network fixed-lag Kalman smoother and multi-particle modeling demonstrates significant advantages. By integrating the LSTM network with the fixed-lag Kalman smoother (FLKS), it solves the problems of nonlinear dynamics and non-Gaussian noise interference, providing high-precision state estimation. By introducing the multi-particle model, it accurately depicts the position and velocity dynamics of each particle of the train, improves the physical consistency of the control strategy, and enhances the physical authenticity of the model. Combined with the improved model predictive control (MPC), based on the optimized state estimation, it achieves high-precision tracking of the train's position and velocity.

[0034] The advantages of the additional aspects of the present invention will be more clearly presented in the following description section, or understood through the practice of the present invention. Brief Description of the Drawings

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0036] Figure 1 It is a flowchart of the intelligent model predictive control method for high-speed trains according to the embodiments of the present invention.

[0037] Figure 2 It is a flowchart of the intelligent model predictive control method for high-speed trains based on the long short-term memory neural network fixed-lag Kalman smoother and multi-particle modeling according to the embodiments of the present invention. Detailed Embodiments

[0038] The following details the embodiments of the present invention. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

[0039] Those skilled in the art of the present technology can understand that unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the art in the field to which the present invention belongs.

[0040] It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless defined as here.

[0041] Those skilled in the art can understand that, unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the description of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or their groups.

[0042] In the description of this specification, the descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. Without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0043] To facilitate the understanding of the present invention, the following takes specific embodiments in conjunction with the drawings to further explain the present invention, and the specific embodiments do not constitute a limitation to the embodiments of the present invention.

[0044] Those skilled in the art should understand that the drawings are only schematic diagrams of the embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0045] During the operation of a high-speed train, the dynamic relationship between the carriages of the train is complex and is affected by various factors such as traction force, braking force, air resistance, and track irregularity. Traditional train dynamics modeling methods usually adopt a lumped mass model or a simplified rigid car body model, which are difficult to accurately describe the dynamic characteristics of each part of the train, especially the relative displacement and speed changes during the train operation. The present invention adopts a multi-point mass model (MPMM), models the train as a system of multiple mass points, and considers factors such as traction / braking force, spring-damping coupling force, and additional resistance, so as to more accurately describe the position, speed, and force conditions of each part of the train. This model can more precisely depict the dynamic behavior of the train and provide higher modeling accuracy for subsequent state estimation and control.

[0046] During the operation of high-speed trains, there are significant non-linear and non-Gaussian noises, making it difficult for traditional filtering methods to accurately estimate the train state. The present invention proposes a Fixed-Lag Kalman Smoother (FLKS) based on Long Short-Term Memory (LSTM) neural network, which uses LSTM to learn non-linear dynamic characteristics from historical data and predict future observation values to provide more accurate input data for Kalman smoothing estimation. It can effectively suppress the influence of non-Gaussian noise and improve the robustness of train state estimation, providing a more reliable basis for high-precision control.

[0047] Traditional Model Predictive Control (MPC) methods are difficult to fully utilize multi-particle dynamics information and have insufficient adaptability to non-linear and non-Gaussian noises, affecting the control accuracy of high-speed trains. The present invention proposes a high-precision MPC method that integrates a multi-particle mass model with LSTM-FLKS state estimation. By combining the high-precision state information provided by LSTM-FLKS, the MPC control strategy is optimized to enable more refined adjustment of the train operation state. In addition, this method takes into account the speed constraint, adjacent mass point spacing limit, and traction / braking force constraint during train operation to improve the safety, stability, and energy efficiency of train operation.

[0048] Embodiment 1

[0049] In this Embodiment 1, first, an intelligent model predictive control system for high-speed trains is provided, including: an acquisition module for acquiring real-time operation state data and historical operation state data of the high-speed train; a prediction module for predicting future operation state prediction data of the high-speed train based on the historical operation state data using a pre-trained long short-term memory neural network; a state estimation module for performing state estimation using a fixed-lag Kalman smoother based on the future operation state prediction data of the high-speed train in combination with the current real-time operation state data to obtain an optimized operation state estimation result; wherein, forward filtering is performed using the current real-time operation state data and the future operation state prediction data of the high-speed train to generate a preliminary state estimation; backward smoothing is performed using future observation data to optimize the preliminary state estimation at the current moment; an optimization module for predicting the state change of the train within a future period based on the optimized preliminary state estimation at the current moment; generating an optimal control input sequence by optimizing the objective function; wherein the objective function takes into account state errors and control input constraints; a control module for adjusting the traction force and braking force of the train in real time based on the optimal control input sequence to ensure that the train runs along the desired trajectory.

[0050] As Figure 1As shown in the figure, in this embodiment, based on the above system, a high-speed train intelligent model predictive control method is implemented, including: obtaining the real-time operation state data and historical operation state data of the high-speed train; based on the historical operation state data, using a pre-trained long short-term memory neural network to predict the future operation state prediction data of the high-speed train; based on the future operation state prediction data of the high-speed train, combining with the current real-time operation state data, using a fixed-lag Kalman smoother for state estimation to obtain an optimized operation state estimation result; wherein, using the current real-time operation state data and the future operation state prediction data of the high-speed train to perform forward filtering to generate a preliminary state estimation; using future observation data to perform backward smoothing to optimize the preliminary state estimation at the current moment; based on the optimized preliminary state estimation at the current moment, predicting the state change of the train in the next period of time; generating an optimal control input sequence by optimizing the objective function; wherein, the objective function considers the state error and the constraints of the control input; based on the optimal control input sequence, adjusting the traction and braking force of the train in real time to ensure that the train runs along the desired trajectory.

[0051] In this embodiment, it aims to solve the problem of high-precision control of high-speed trains in a non-linear dynamic environment. Existing methods, such as extended Kalman filter (EKF), unscented Kalman filter (UKF), and particle filter (PF), etc., have limitations in dealing with non-Gaussian noise and complex non-linear dynamic characteristics of the high-speed train system, resulting in a decrease in state estimation accuracy and affecting the stability and safety of train operation. In addition, traditional predictive control methods rely strongly on state estimation and fail to fully utilize data-driven methods to optimize the accuracy of state estimation. Therefore, as Figure 2 shown, this embodiment 2 proposes a high-precision model predictive control (MPC) method for high-speed trains based on a fixed-lag Kalman smoother (FLKS) combined with a multi-particle mass model (MPMM) using long short-term memory (LSTM) neural networks. Through the high-precision MPC method of this embodiment, the operation stability, energy efficiency, and safety of high-speed trains can be significantly improved, enabling them to maintain an optimal operation state in a complex track environment.

[0052] The HST system consists of a multi-point mass model, and its state variables include the position and velocity of each point. Assuming the train consists of N points, the state vector of the train system can be expressed as:

[0053] X = [x1 x2 …x N v1 v2 …v N T (1)

[0054] where, x i represents the position of the i-th particle, v i ​represents the velocity of the \(i\)-th particle. During operation, the HST is subject to various forces, including spring force, drag force, and traction force, etc. The coupler system connecting adjacent carriages can be regarded as a spring with damping effect. The spring force can be determined according to the generalized Hooke's law, and the influence of the damping force and traction force also needs to be considered. The motion equation of the multi-particle high-speed train can be expressed as:

[0055]

[0056] where \(m\) i is the mass of the \(i\)-th particle, \(F\) i,t / b is the traction / braking force acting on the \(i\)-th particle, \(F\) i,s is the spring force of the \(i\)-th coupler system, \(F\) i,a is the additional resistance suffered by the \(i\)-th particle.

[0057] According to formula (2), a more specific running differential equation of the high-speed train (HST) can be derived:

[0058]

[0059] where \(\kappa\) i is the stiffness coefficient of the \(i\)-th coupler, \(u\) i is the control input (traction or braking force) of the \(i\)-th particle, with the unit of kN, \(\tau\) is the free tension of the coupler system, and \(c_0\), \(c_1\) and \(c_2\) are the empirical coefficients of the additional resistance.

[0060] The Runge - Kutta method is a set of numerical integration techniques for solving the initial value problems of ordinary differential equations. These methods improve the accuracy of the numerical solution through multiple estimation steps. The widely used fourth-order Runge - Kutta method (RK4) is adopted to deal with equation (3), and the specific process is as follows:

[0061]

[0062] By applying the RK4 method, the continuous-time running equation of the high-speed train (HST) system is discretized, so as to simulate its dynamic behavior changing with time. To meet the basic problem modeling requirements of model predictive control (MPC) and Kalman smoother, a state space model needs to be constructed for the HST system.

[0063] \(x\) i (k + 1)=\(\varPhi\) i (k + 1)x i (k)+\(\varTheta\) i u i (k)+w ix (k)(5)

[0064] \(v\) i (k + 1)=\(\varXi\)i (k + 1)v i (k) + B i u i (k) + w iv (k)(6)

[0065] Where: w ix (k) and w iv (k) is noise with independent distribution characteristics, which can be Gaussian noise or non-Gaussian noise.

[0066]

[0067] Where: X(k) = [x(k), v(k)] T , U(k) = [0 N×1 , u(k)] T , W(k) = [w x (k), w v (k)] T , 0 N×1 is an N×1 zero vector, Y(k + 1) = [y x (k + 1), y v (k + 1)] T , V(k + 1) is a noise sequence with a specific distribution similar to W(k).

[0068] The specific parameter vector is given by formula (8).

[0069]

[0070] In formula (7), the state transition matrix A(k + 1), the control parameter matrix Γ, and the observation matrix H(k + 1) are given by formulas (9), (10), and (11) respectively.

[0071]

[0072] Where: Φ N×N (k + 1) = Diag(Φ1, Φ2, …, Φ N ), Ξ N×N (k + 1) = Diag(Ξ1, Ξ2, …, Ξ N ). B N×N = Diag(B1, B2, …, B N ), H x (k + 1) and H v (k + 1) are the displacement and velocity observation matrices of multiple points of mass in the high-speed train (HST) system, respectively, and need to be obtained according to the displacement sensors and velocity sensors of the HST.

[0073] In this embodiment, based on the model of the above HST system, to achieve the above invention object, the following technical solutions are adopted for the high-precision model predictive control (MPC) method of a high-speed train that combines a fixed-lag Kalman smoother (FLKS) based on a long short-term memory (LSTM) neural network and a multi-particle mass model (MPMM):

[0074] Steps for predicting future train state data based on LSTM:

[0075] In view of the long-term dependence and non-linear characteristics of train dynamics, an LSTM network is used in this embodiment to predict future train state data. LSTM is a special type of recurrent neural network (RNN) that can learn long-term dependencies. The LSTM network controls the information flow by using a gating mechanism, making it very effective in processing sequential data. Using LSTM to predict the observations of a high-speed train system aims to capture the non-linear state changes and non-Gaussian noise experienced by the high-speed train system based on the expected operation trajectory. At the same time, during the forward filtering process, the Kalman smoother requires additional future observation data to optimize the real-time estimation of the high-speed train operation state, so as to provide more accurate state parameters for model predictive control (MPC).

[0076] Input gate:

[0077]

[0078] Forget gate:

[0079]

[0080] Output gate:

[0081]

[0082] Among them, I(k), F(k), and O(k) represent the activation values of the input gate, forget gate, and output gate respectively. And C(k) are the candidate memory cell state and the current memory cell state respectively. σ(*) = (1 + e (*) ) -1 ∈(0,1) is the sigmoid activation function, is the hyperbolic tangent activation function, ⊙ represents element-wise multiplication, ω * and b * correspond to the weight matrix and the bias term respectively. y(k) and represent the input value and the hidden state respectively.

[0083] Fixed Lag Kalman Smoother (FLKS): To obtain the estimates from the smoother, a two-way filtering structure needs to be established. During the normal operation of the forward filter, the time information of the High-Speed Train (HST) is collected. Meanwhile, during the backward derivation process, the LSTM-based prediction is used for the advance observation, thus achieving the optimal estimation of the current state.

[0084]

[0085] Where:

[0086]

[0087] Based on Equation (18), the one-step prediction estimate of can be calculated.

[0088]

[0089] The prediction estimation error is

[0090]

[0091] The prediction estimation error has a covariance matrix of P(k + 1|k):

[0092]

[0093] Similarly, the one-step prediction estimate of the observed value can be calculated:

[0094]

[0095] The prediction estimation error of the observed value is

[0096]

[0097] Therefore, the state estimate obtained by the Kalman Filter (KF) at time k + 1 is

[0098]

[0099] where K(k + 1) is the Kalman optimal gain matrix to be determined, which can be solved by the orthogonality principle as shown in Equation (25).

[0100]

[0101]

[0102] To achieve the iterative update of the Kalman filter (KF), it is necessary to update the state estimate at time k+1 of the estimated error covariance matrix.

[0103]

[0104] Similar to the principle of the Kalman filter (KF), the Kalman smoother uses a series of future observations to optimize the estimate at the current time to make it optimal, and its smoothed estimate can still be obtained through innovation analysis and the orthogonality principle.

[0105]

[0106] In the given observation sequence {Y(1),…,Y(k),…,Y(k+n)}, is the optimal smoothed estimate, and its expression form is as follows:

[0107]

[0108] Based on the above formula, the Kalman smoother is designed and completed.

[0109]

[0110] Among them, Ψ(j) is the Kalman smoothing gain matrix, which can be solved by the orthogonality principle, as shown in Equation (29).

[0111]

[0112] can be used to determine Ψ(j).

[0113] Ψ(j) = P(k|j)H T (j)[H(j)P(j|j - 1)H T (j) + R(j)] -1 (32)

[0114] Among them, P(k|j) = P(k|j - 1)A T (j).

[0115] Model Predictive Control (MPC): MPC is a model-based control method. It predicts the future behavior of the HST system according to its dynamic model and generates control actions by optimizing the performance index to achieve the desired operating state. The advantage of this method is that it considers the dynamic characteristics of the HST system and possible future changes, thus coping with the uncertainty and changes of the HST system to a certain extent. The unique feature of the MPC method is that it needs to predict the future system state. In the next p control cycles, the predicted state of the HST system is expressed as:

[0116] X k+p =[X(k + 1|k), X(k + 2|k), …, X(k + p|k)] T (33)

[0117] p is called the prediction horizon. When predicting the future state of a dynamic system, the control input U within the prediction horizon is also required to be known k+p .

[0118] U k+p =[U(k|k), U(k + 1|k), …, U(k + p - 1|k)] T (34)

[0119] Based on X k+p and U k+p , an expression for the optimization problem of controlling the HST system can be formulated. First, the states of the HST in the next p control cycles are predicted sequentially:

[0120]

[0121] Its matrix form is:

[0122]

[0123] Where:

[0124]

[0125] The discretization and linearization of the HST multi - point mass model, as well as the derivation of the aforementioned prediction model, have been completed. The next step is to design the objective function for HST control. To find the optimal control input U k+p such that the predicted HST state vector is closer to the desired state within the prediction horizon p This is an open - loop optimal control problem. For this purpose, an objective function is defined by calculating the cumulative error between the predicted state vector and the desired state and adding control constraints to reduce the control amplitude.

[0126]

[0127] Where C = Diag(C1, C2, …, C p ) is the state error weight matrix, M = Diag(M1, M2, …, M p-1 ) is the control input weight matrix, while

[0128] Based on the above equation, the final optimization objective function can be obtained:

[0129]

[0130] In formula (40), the constraint conditions clearly define the feasible range of the control variables, effectively forming a feasible set. Specifically, the constraints include: the range of each velocity v i is limited to 0 ≤ v i ≤ v lim ; the displacement difference between adjacent positions x i and x i-1 is limited to -(τ - δ) ≤ x i - x i-1 ≤ -(τ + δ), ensuring sufficient safety distance between the mass points; the control input u i of each point mass is subject to . v lim is the velocity limit of each point mass, δ is the hard safety constraint between HST point masses, is the maximum braking force, is the maximum traction force. These constraints define the feasible set of the control variables, ensuring that the control input u i always follows the physical limitations of the system during the optimization process, thus guaranteeing the stability and safety of the HST system. Formula (40) provides the basic constraints of the HST system, but in order to achieve the control of the HST multi-point mass model, additional constraints are required, and these constraints must conform to the mechanism characteristics of HST distributed traction and braking.

[0131] Combining the objective function and the constraints transforms the MPC problem into a standard quadratic programming (QP) problem. In each control cycle, the objective function and the constraints are redefined based on the current system state. By solving the QP problem, the optimal control input sequence U k+p within the prediction horizon is obtained, and the first control input is extracted and applied to the current time k.

[0132] Finally, it is necessary to prove the stability of the controller design. Assume that the initial state X(0) is feasible, and there exists an optimal control sequence and the corresponding state trajectory By inputting into the HST system, is obtained. At X(1), the control sequence is also feasible (when X p = 0, in order to reach X p+1 = 0, the control quantity must be 0). The goal of the stability proof is to prove that the objective function is a Lyapunov function, that is:

[0133]

[0134] The proof process is as follows:

[0135]

[0136] where: q(0, 0) = 0, and it follows that:

[0137]

[0138] In summary, the fixed-lag Kalman smoother based on the long short-term memory neural network and the intelligent high-speed train model predictive control method with multi-particle modeling provided in this Embodiment 2 address the issues of fine multi-particle modeling of high-speed trains, the non-linearity of high-speed train operation and the interference by non-Gaussian noise, and the design of the intelligent model predictive control method for high-speed train multi-particles. Traditional train modeling methods (such as the lumped mass model or the rigid car body model) cannot accurately depict the dynamic relationships between carriages. This embodiment adopts the multi-particle mass model (Multi-Point Mass Model, MPMM), regarding the train as a system of multiple interacting mass points. This model takes into account factors such as traction / braking force, coupler spring-damping coupling force, and air resistance, and can more accurately describe the running state of the train. The operating environment of high-speed trains is complex, affected by non-linear dynamic characteristics and non-Gaussian noise, and traditional filtering methods (EKF, UKF, PF) have limitations. The present invention integrates the LSTM network with the fixed-lag Kalman smoother (FLKS), uses LSTM to learn the dynamic characteristics of the train, predicts future observation values, and enhances the estimation ability of the Kalman smoother. This method can effectively suppress non-Gaussian noise, improve the state estimation accuracy, and provide high-precision input data for subsequent control optimization. Traditional MPC methods are limited by single-particle modeling and Gaussian noise assumptions and are difficult to adapt to the complex environment of high-speed trains. The present invention proposes a high-precision MPC method that integrates multi-particle modeling + LSTM-FLKS estimation: using the high-precision state estimation provided by LSTM-FLKS to enhance the trajectory tracking ability of MPC. During the MPC optimization process, speed constraints, adjacent mass point spacing limits, and traction / braking force constraints are considered to ensure safe, stable, and optimal energy efficiency operation.

[0139] In summary, this embodiment uses LSTM-FLKS combined with multi-particle modeling, which can learn non-linear characteristics from data while maintaining physical constraints, significantly improving the accuracy and stability of state estimation. LSTM predicts future observed values, providing additional information for FLKS, and combined with filtering and smoothing techniques, enhancing the anti-noise ability. This method can still maintain high-precision estimation in a complex non-Gaussian noise environment, and experimental verification shows that the accuracy is improved by at least 27.58%. Using a fixed-lag Kalman smoother (FLKS) to replace the traditional Kalman filter reduces the computational complexity. Combining with a lightweight MPC optimization strategy improves the computational efficiency while ensuring accuracy, meeting the real-time control requirements of high-speed trains. Through high-precision state estimation of LSTM-FLKS and combined with a multi-particle MPC optimization strategy, precise trajectory tracking is achieved. Optimize the traction / braking force distribution, and improve the energy efficiency of the train on the premise of ensuring stable operation. Through the innovative integration of multi-particle mass modeling, LSTM-FLKS state estimation, and high-precision MPC control, the limitations of existing technologies are broken through, and there are obvious advantages in non-linear dynamic modeling, anti-non-Gaussian noise ability, computational efficiency, and trajectory tracking accuracy, which is particularly suitable for the intelligent control optimization of high-speed trains.

[0140] Embodiment 2

[0141] This Embodiment 2 provides a non-transitory computer-readable storage medium for storing computer instructions, which when executed by a processor, implement the above-mentioned intelligent model predictive control method for high-speed trains based on a long short-term memory neural network and multi-particle modeling. The method includes:

[0142] Obtain the real-time operation state data and historical operation state data of the high-speed train;

[0143] Based on the historical operation state data, use the pre-trained long short-term memory neural network to predict the future operation state prediction data of the high-speed train;

[0144] Based on the future operation state prediction data of the high-speed train and combined with the current real-time operation state data, use a fixed-lag Kalman smoother for state estimation to obtain an optimized operation state estimation result; wherein, use the current real-time operation state data and the future operation state prediction data of the high-speed train for forward filtering to generate a preliminary state estimation; use the future observed data for backward smoothing to optimize the preliminary state estimation at the current moment;

[0145] Based on the optimized preliminary state estimation at the current moment, predict the state change of the train in the future for a period of time; generate an optimal control input sequence by optimizing the objective function; wherein, the objective function considers the state error and the constraints of the control input;

[0146] Based on the optimal control input sequence, the traction force and braking force of the train are controlled and adjusted in real time to ensure that the train runs according to the desired trajectory.

[0147] Embodiment 3

[0148] Embodiment 3 provides a computer device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the above-mentioned high-speed train intelligent model predictive control method based on the long short-term memory neural network and multi-particle modeling, and this method includes:

[0149] Obtain the real-time operation state data and historical operation state data of the high-speed train;

[0150] Based on the historical operation state data, use the pre-trained long short-term memory neural network to predict the future operation state prediction data of the high-speed train;

[0151] Based on the future operation state prediction data of the high-speed train, combined with the current real-time operation state data, use a fixed-lag Kalman smoother for state estimation to obtain an optimized operation state estimation result; among them, use the current real-time operation state data and the future operation state prediction data of the high-speed train to perform forward filtering to generate a preliminary state estimation; use the future observation data to perform backward smoothing to optimize the preliminary state estimation at the current moment;

[0152] Based on the optimized preliminary state estimation at the current moment, predict the state change of the train in the next period of time; generate an optimal control input sequence by optimizing the objective function; among them, the objective function considers the state error and the constraints of the control input;

[0153] Based on the optimal control input sequence, the traction force and braking force of the train are controlled and adjusted in real time to ensure that the train runs according to the desired trajectory.

[0154] Embodiment 4

[0155] Embodiment 4 provides an electronic device, including: a processor, a memory, and a computer program; among them, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes the instruction to implement the above-mentioned high-speed train intelligent model predictive control method based on the long short-term memory neural network and multi-particle modeling, and this method includes:

[0156] Obtain the real-time operation state data and historical operation state data of the high-speed train;

[0157] Based on historical operation status data, use a pre-trained long short-term memory neural network to predict the operation status prediction data of future high-speed trains;

[0158] Based on the operation status prediction data of future high-speed trains, combined with the current real-time operation status data, use a fixed-lag Kalman smoother for state estimation to obtain an optimized operation status estimation result; among them, use the current real-time operation status data and the operation status prediction data of future high-speed trains for forward filtering to generate a preliminary state estimation; use future observation data for backward smoothing to optimize the preliminary state estimation at the current moment;

[0159] Based on the optimized preliminary state estimation at the current moment, predict the state change of the train in the next period of time; generate an optimal control input sequence by optimizing the objective function; where the objective function takes into account the state error and the constraints of the control input;

[0160] Based on the optimal control input sequence, adjust the traction and braking force of the train in real time to ensure that the train runs according to the desired trajectory.

[0161] In summary, the MPC method of a Fixed-lag Kalman Smoother (FLKS) based on LSTM combined with a multi-point mass model described in the embodiments of the present invention. This method combines the prediction ability of LSTM and the real-time optimization characteristics of FLKS to improve the state estimation accuracy, especially the robustness in a non-Gaussian noise environment. LSTM can predict future observations, providing richer time series information for FLKS, enabling the smoothing estimation to more accurately suppress noise and improve the train state estimation accuracy. At the same time, the present invention introduces a multi-point mass model (MPMM) to finely model the high-speed train, more accurately describing the dynamic characteristics of each part of the train. By modeling the interactions between the mass points of the train, the optimization ability of MPC for the train operation state is improved, reducing the influence of modeling errors on the control effect. In addition, the improved MPC optimization strategy uses the high-precision state estimation provided by LSTM-FLKS to improve the control accuracy and response speed of MPC in complex environments. This method also adopts a constraint optimization method, combined with the speed constraint of train operation, the spacing limit between adjacent mass points, and the traction / braking force constraint, to improve the safety, stability, and energy efficiency of train operation. Experimental results show that the method of the present invention in the high-speed train operation environment, compared with the traditional MPC methods based on EKF, UKF, and PF, the control accuracy is increased by at least 27.58%, and it can still maintain high robustness in a complex non-Gaussian noise environment. In addition, this method shows excellent control performance in the experimental verification of the high-speed railway from Zhengzhou East to Xi'an North, demonstrating its engineering application value in the actual rail transit system. In summary, the present invention provides a new solution for high-precision MPC of high-speed trains by combining LSTM, FLKS, and a multi-point mass model, improving the safety, stability, and energy efficiency of train operation, and providing a new technical direction for the development of intelligent rail transit control systems.

[0162] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0163] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in the flow Figure 1 one or more flows and / or blocks Figure 1 means for implementing the functions specified in one or more blocks or multiple blocks.

[0164] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in the flow Figure 1 one or more flows and / or blocks Figure 1 means for implementing the functions specified in one or more blocks or multiple blocks.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device to perform a series of operational steps on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the flow Figure 1 one or more flows and / or blocks Figure 1 means for implementing the functions specified in one or more blocks or multiple blocks.

[0166] Although the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation to the protection scope of the present invention. Those skilled in the art should understand that based on the technical solutions disclosed in the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts should be covered within the protection scope of the present invention.

Claims

1. An intelligent model predictive control method for high-speed trains, characterized in that, Including: Obtain the real-time operation status data and historical operation status data of the high-speed train; Based on the historical operation status data, use the pre-trained long short-term memory neural network to predict the operation status prediction data of the future high-speed train; Based on the operation status prediction data of the future high-speed train, combined with the current real-time operation status data, use the fixed-lag Kalman smoother for state estimation to obtain the optimized operation status estimation result; among them, use the current real-time operation status data and the operation status prediction data of the future high-speed train for forward filtering to generate a preliminary state estimation; use the future observation data for backward smoothing to optimize the preliminary state estimation at the current moment; Based on the optimized preliminary state estimation at the current moment, predict the state change of the train in a future period of time; generate the optimal control input sequence by optimizing the objective function; among them, the objective function considers the state error and the constraints of the control input; Based on the optimal control input sequence, adjust the traction force and braking force of the train in real time to ensure that the train runs according to the expected trajectory.

2. The intelligent model predictive control method for high-speed trains according to claim 1, wherein In the fixed-delay Kalman smoother, a two-way filtering structure is established. In forward filtering, the time information of the high-speed train is collected. In backward smoothing, the operation status prediction data of the future high-speed train is used for early observation to achieve the optimal estimation of the current state.

3. The intelligent model predictive control method for high-speed trains according to claim 2, wherein Model predictive control (MPC) needs to predict the future system state. In the next p control cycles, the predicted state of the HST system is expressed as: X k+p = [X(k + 1|k), X(k + 2|k), …, X(k + p|k)] T ; p is the prediction horizon; When predicting the future state of a dynamic system, it is also necessary to know the control input U within the prediction horizon k+p ; U k+p = [U(k|k), U(k + 1|k), …, U(k + p - 1|k)] T ; Based on X k+p and U k+p , determine the optimization problem expression for controlling the HST system.

4. The intelligent model predictive control method for high-speed trains according to claim 3, wherein Determine the optimization problem expression for controlling the HST system, including: First, sequentially predict the state of the HST in the next p control cycles; Then, design the objective function of HST control to determine the optimal control input U k+p , such that the predicted HST state vector is closer to the desired state within the prediction horizon p This is an open-loop optimal control problem. To this end, define the objective function by calculating the cumulative error between the predicted state vector and the desired state, and adding control constraints to reduce the control amplitude.

5. The intelligent model predictive control method for high-speed trains according to claim 4, wherein The model constraints include: for each velocity v i the range is limited to 0 ≤ v i ≤ v lim ; the displacement difference between adjacent positions x i and x i-1 is limited to -(τ - δ) ≤ x i - x i-1 ≤ -(τ + δ), ensuring sufficient safety distance between mass points; the control input u i for each point mass is subject to . v lim is the velocity limit of each point mass, δ is the hard safety constraint between HST point masses, is the maximum braking force, is the maximum traction force; these constraints define the feasible set of control variables, ensuring that the control input u i always follows the physical limitations of the system, thus guaranteeing the stability and safety of the HST system.

6. The intelligent model predictive control method for high-speed trains according to claim 5, characterized in that, The objective function and constraints are combined to transform the MPC problem into a standard quadratic programming problem. In each control period, the objective function and constraints are redefined based on the current system state. By solving the QP problem, an optimal control input sequence U within the prediction horizon is obtained k+p , and the first control input is extracted and applied to the current time k.

7. An intelligent model predictive control system for high-speed trains, characterized in that, Including: An acquisition module for obtaining the real-time operation status data and historical operation status data of the high-speed train; A prediction module for predicting the operation status prediction data of the future high-speed train based on the historical operation status data using the pre-trained long short-term memory neural network; A state estimation module for performing state estimation using the fixed-lag Kalman smoother based on the operation status prediction data of the future high-speed train and combined with the current real-time operation status data to obtain the optimized operation status estimation result; among them, use the current real-time operation status data and the operation status prediction data of the future high-speed train for forward filtering to generate a preliminary state estimation; use the future observation data for backward smoothing to optimize the preliminary state estimation at the current moment; An optimization module for predicting the state change of the train in a future period of time based on the optimized preliminary state estimation at the current moment; generating the optimal control input sequence by optimizing the objective function; among them, the objective function considers the state error and the constraints of the control input; A control module for adjusting the traction force and braking force of the train in real time based on the optimal control input sequence to ensure that the train runs according to the expected trajectory.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the intelligent model predictive control method for high-speed trains described in any one of claims 1-6 is implemented.

9. A computer device, characterized in that, It includes a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the high-speed train intelligent model predictive control method according to any one of claims 1-6.

10. An electronic device, characterized in that, It includes: A processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes the instructions for implementing the high-speed train intelligent model predictive control method according to any one of claims 1-6.

Images