A method, device, equipment and medium for predictive control of automatic driving of high-speed train

Abstract

The present application discloses a method, device, equipment and medium for predictive control of automatic driving of high-speed trains, which relates to the field of automatic driving of trains. The method comprises: based on a distributed update strategy under the collaboration of end-edge-cloud, the edge-side and cloud-side digital twin models are updated in real time to obtain an edge return model and a cloud return model; the end-side digital twin model is equivalent to the edge return model updated last time; the prediction errors of the end-side, edge-side and cloud-side digital twin models are evaluated, and when the end-side digital twin model is not optimal, the edge-side digital twin model and the cloud return model are used to correct the end-side digital twin model by means of a self-correction mechanism, and the optimal control instruction sequence of the predicted time domain of the target high-speed train is determined according to the predicted train speed at the next moment, so as to realize uninterrupted control of high-speed trains without delay. The present application improves the tracking performance and ride comfort of high-speed trains.

Description

Technical Field

[0001] The present application relates to the field of automatic train driving, and in particular to a method, device, equipment and medium for predictive control of automatic high-speed train driving. Background Art

[0002] High-speed trains (HST), as a fast and convenient means of transportation, have become the future trend of global railway transportation. Automatic Train Operation (ATO) plays a vital role in controlling, monitoring, optimizing operations, improving safety, and ensuring accuracy. With the increase in HST operating speed and the complexity of the operating environment, the maneuverability of ATO faces new challenges. Therefore, there is an urgent need to develop advanced and reliable modeling and control technologies to support autonomous driving operations.

[0003] Currently, a model predictive control (MPC) algorithm based on an autoregressive integrated moving average model (ARIMA) can be used. However, the ARIMA-based MPC algorithm or other data-driven control algorithms are difficult to meet the requirements for tracking performance and ride comfort of HST. Summary of the invention

[0004] The purpose of this application is to provide a high-speed train automatic driving predictive control method, device, equipment and medium, which can improve the tracking performance and ride comfort of high-speed trains.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a high-speed train automatic driving predictive control method, comprising the following steps:

[0007] Based on the distributed update strategy under the collaboration of edge, edge and cloud, the model parameters of the edge-side digital twin model and the cloud-side digital twin model are updated in real time to obtain the edge backhaul model and the cloud backhaul model.

[0008] Obtaining train operation information of a target high-speed train during a set period; the set period includes multiple historical moments; the train operation information includes train operation speed, track conditions and control instructions;

[0009] For each historical moment within the set period, the train operation information of the target high-speed train at the historical moment is input into the end-side digital twin model to obtain the predicted train speed at the next moment of the historical moment; the end-side digital twin model is the edge feedback model updated last time;

[0010] Calculate the end-side error evaluation value of the end-side digital twin model based on the predicted train speed at the next moment and the actual train running speed of all historical moments within the set period;

[0011] According to the end-side error evaluation value of the end-side digital twin model, determine whether the end-side digital twin model needs to be corrected. If so, use the self-correction mechanism to correct the end-side digital twin model using the edge-side digital twin model and the cloud-based backhaul model to obtain a corrected end-side digital twin model.

[0012] The corrected end-side digital twin model is used to cyclically predict the train running speed at multiple prediction moments, and the optimal control instruction sequence of the target high-speed train in the prediction time domain is determined based on the train running speed at the prediction moment; the optimal control instruction sequence includes the optimal control input corresponding to several prediction moments in the prediction time domain; when the execution of the control instruction at the previous moment is completed, the target high-speed train is controlled to execute the optimal control instruction corresponding to the next moment, so as to realize real-time control of the high-speed train.

[0013] In a second aspect, the present application provides a high-speed train automatic driving prediction control device, comprising the following modules:

[0014] The end-edge-cloud collaborative distributed update module is used to: based on the distributed update strategy under the end-edge-cloud collaboration, update the model parameters of the edge-side digital twin model and the cloud-side digital twin model in real time to obtain the edge return model and the cloud return model;

[0015] The train operation information acquisition module is used to: acquire the train operation information of the target high-speed train during a set period; the set period includes multiple historical moments; the train operation information includes the train operation speed, track conditions and control instructions;

[0016] The train speed prediction module is used to: for each historical moment within a set period, input the train operation information of the target high-speed train at the historical moment into the end-side digital twin model to obtain the predicted train speed at the next moment of the historical moment; the end-side digital twin model is the edge feedback model updated last time;

[0017] The terminal-side error evaluation value calculation module is used to calculate the terminal-side error evaluation value of the terminal-side digital twin model according to the predicted train speed at the next moment of all historical moments within a set period and the actual train running speed;

[0018] The device-side digital twin model self-correction module is used to: determine whether the device-side digital twin model needs to be corrected according to the device-side error evaluation value of the device-side digital twin model; if so, use the self-correction mechanism to correct the device-side digital twin model using the edge-side digital twin model and the cloud-based backhaul model to obtain a corrected device-side digital twin model;

[0019] The optimal control instruction sequence determination module is used to: use the corrected end-side digital twin model to cyclically predict the train running speed at multiple prediction moments, and determine the optimal control instruction sequence of the target high-speed train in the prediction time domain based on the train running speed at the prediction moment; the optimal control instruction sequence includes the optimal control input corresponding to several prediction moments in the prediction time domain; when the control instruction at the previous moment is executed, the target high-speed train is controlled to execute the optimal control instruction corresponding to the next moment, so as to realize real-time control of the high-speed train.

[0020] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned high-speed train automatic driving predictive control method.

[0021] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned high-speed train automatic driving predictive control method.

[0022] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0023] The present application provides a method, device, equipment and medium for predictive control of automatic driving of high-speed trains. The predicted train speed at the next moment is obtained through the end-side digital twin model, and the end-side digital twin model is used as the prediction model of the MPC controller. Thanks to the high accuracy and robustness of the end-side digital twin model during the online period, the high-speed train under this algorithm has higher tracking accuracy. On this basis, combined with the MPC multi-objective loss function, the control instructions generated by the proposed algorithm are smoother, thereby improving ride comfort. In addition, based on the multi-step cycle prediction capability of the digital twin model, the MPC control decision is processed one step ahead, the MPC control delay problem on the on-board computer is solved, and the uninterrupted control of the high-speed train without delay in the control instructions is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 This is an application environment diagram of a high-speed train automatic driving predictive control method in one embodiment of the present application;

[0026] Figure 2A schematic diagram of a multi-dimensional attention GRU network framework provided in an embodiment of the present application;

[0027] Figure 3 A flowchart of a high-speed train automatic driving predictive control method provided in one embodiment of the present application;

[0028] Figure 4 A schematic diagram of a high-speed train cooperative operation architecture provided in an embodiment of the present application;

[0029] Figure 5 A schematic diagram of a one-step-ahead control timing diagram provided in an embodiment of the present application;

[0030] Figure 6 A schematic diagram showing comparison results of prediction model identification errors of four control schemes provided in one embodiment of the present application;

[0031] Figure 7 A schematic diagram of the speed tracking effect of a high-speed train under four control schemes provided in an embodiment of the present application;

[0032] Figure 8 for Figure 7 A magnified view of part A;

[0033] Fig. 9 A schematic diagram of tracking error distribution during high-speed train operation under four control schemes provided in an embodiment of the present application;

[0034] Fig.10 A schematic diagram of control force curves calculated from four control schemes provided in an embodiment of the present application;

[0035] Fig.11 Schematic diagram of acceleration curves of a high-speed train under four control schemes provided in an embodiment of the present application;

[0036] Fig.12 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0038] Although traditional predictive control systems can achieve automatic tracking of high-speed trains, their complexity makes online updates difficult, and all simulation experiments are implemented on personal computers without considering the limited hardware resources of on-board computers. There is a control delay problem caused by the MPC nested optimization problem on low-performance computers.

[0039] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0040] The high-speed train automatic driving prediction control method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the train operation information of the historical moment to the server 104. After the server 104 receives the train operation information of the historical moment, the server 104 updates the edge side and cloud side digital twin models in real time based on the distributed update strategy under the end-edge-cloud collaboration to obtain the edge return model and the cloud return model; the end-side digital twin model is equivalent to the edge return model updated last time; the prediction errors of the end-side, edge-side, and cloud-side digital twin models are evaluated. When the end-side digital twin model is not optimal, the self-correction mechanism is used to correct the end-side digital twin model using the edge-side digital twin model and the cloud return model, and the corrected end-side digital twin model is used to cyclically predict the train running speed at multiple prediction moments, and the optimal control instruction sequence of the target high-speed train prediction time domain is determined based on the train running speed at the prediction moment. The server 104 can feed back the obtained optimal control instruction sequence to the terminal 102. In addition, in some embodiments, the high-speed train automatic driving predictive control method can also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 can directly perform module predictive control on the train operation information at the historical moment, or the server 104 can obtain the train operation information at the historical moment from the data storage system and perform module predictive control on the train operation information at the historical moment.

[0041] The terminal 102 is a digital computer of the train control system. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers, or a cloud server.

[0042] In an exemplary embodiment, Figure 3As shown, a high-speed train automatic driving prediction control method is provided, which is executed by a computer device, and can be specifically executed by a computer device such as a terminal or a server alone, or can be executed by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used for explanation, and the steps include the following steps 201 to 206.

[0043] Step 201: Based on the distributed update strategy under the collaboration of end, edge and cloud, the model parameters of the edge-side digital twin model and the cloud-side digital twin model are updated in real time to obtain the edge backhaul model and the cloud backhaul model.

[0044] Step 202: Acquire the train operation information of the target high-speed train during a set period; the set period includes multiple historical moments; the train operation information includes the train operation speed, track conditions and control instructions.

[0045] Step 203: For each historical moment within the set time period, the train operation information of the target high-speed train at the historical moment is input into the end-side digital twin model to obtain the predicted train speed at the next moment of the historical moment; the end-side digital twin model is the edge feedback model updated last time.

[0046] Step 204: Calculate the end-side error evaluation value of the end-side digital twin model based on the predicted train speed at the next moment and the actual train running speed of all historical moments within the set time period.

[0047] Step 205: According to the device-side error evaluation value of the device-side digital twin model, determine whether the device-side digital twin model is to be corrected. If so, use the self-correction mechanism to correct the device-side digital twin model using the edge-side digital twin model and the cloud-based return model to obtain a corrected device-side digital twin model. The corrected device-side digital twin model is the model updated this time.

[0048] Step 206: Use the corrected end-side digital twin model to cyclically predict the train running speed at multiple prediction moments, and determine the optimal control instruction sequence of the target high-speed train in the prediction time domain based on the train running speed at the prediction moment; the optimal control instruction sequence includes the optimal control input corresponding to several prediction moments in the prediction time domain; when the control instruction at the previous moment is executed, the target high-speed train is controlled to execute the optimal control instruction corresponding to the next moment, so as to realize real-time control of the high-speed train.

[0049] The predictive control algorithm is implemented as a whole in adjacent sampling stages. When the control instruction at the previous moment is applied, it will be immediately replaced by the calculated optimal control instruction at the next moment to achieve real-time control of the high-speed train.

[0050] Implementing the above steps 201 to 206, this application combines the self-attention mechanism with the GRU neural network unit based on the dynamic characteristics of the high-speed train and the I / O data generated during operation, and then establishes an end-edge-cloud collaborative high-speed train digital twin model. While ensuring high-precision identification, the neural network framework is more concise. The industrial Internet end-edge-cloud collaboration (EECC) technology is also enabled in the high-speed train MPC control algorithm based on the digital twin model, and the multi-end collaborative model update strategy is used to solve the model drift problem caused by the strong time-varying system.

[0051] The force analysis of the high-speed train model in the running direction can reflect the longitudinal dynamic relationship of the high-speed train body. Represents the current moment, It is the train traction / braking force, and are the basic resistance and additional resistance of the train respectively. is the train running speed.

[0052] According to Newton's laws of kinematics, the high-speed train operation model under complex working conditions can be expressed as shown in formula (1).

[0053] (1).

[0054] in, for The train speed at the time, is the total running quality of the high-speed train, is a constant mass; It is a random disturbance equivalent to 0.5 tons, used to simulate the flow of people; It is a random disturbance equivalent to 200N, used to simulate external environmental disturbance; is the sampling time. The basic resistance and additional resistance can be expressed as shown in equations (2) and (3).

[0055] (2).

[0056] (3).

[0057] in, is the basic resistance coefficient, , , are random parameter perturbations corresponding to the three coefficients, which are used to simulate the mechanical coupling of the train and the influence of gusts on the train; , , They are unit ramp additional resistance, unit curve additional resistance, and unit tunnel additional resistance, and the numerical unit is N / KN. The calculation formulas of the three forces are shown in the following formula (4).

[0058] (4).

[0059] in, is the track slope at the current train location, which is approximately equal to the additional resistance of the ramp. Generally a constant of 450-800, is the radius of curvature of the track, is the tunnel length.

[0060] The train running speed at the historical moment is sampled from the high-speed train dynamics model. Since the basic resistance coefficient and additional resistance parameters are affected by many factors, and the train mass changes with the number of passengers, the high-speed train model presents strong time-varying and nonlinear characteristics. Considering that a large amount of data will be generated during the operation of the train, these data can be used to re-describe the actual dynamic relationship of the high-speed train, forming the nonlinear mapping relationship shown below, that is, the high-speed train dynamics model is shown in the following formula (5).

[0061] (5).

[0062] in, is the nonlinear transfer function of the train, for The train speed at the time, for The train speed at the time, for The control input at each moment is the traction or braking force of the train; for The control input at the moment, Indicates real-time track conditions, including slope, curvature radius and tunnel length; and are two unknown parameters, representing the order of the input variables of the dynamic system.

[0063] Based on a deep understanding of the end-edge-cloud collaboration mechanism, this application proposes the following Figure 4 The high-speed train collaborative operation architecture shown in the figure. Among them, the cloud-side artificial intelligence platform executes the cloud-side train digital twin model, the edge-side edge computer executes data processing and the edge-side train digital twin model, and the terminal-side train control system executes data collection and autonomous driving algorithms.

[0064] Specifically, the end side executes the MPC control algorithm based on the end side's optimal train digital twin model, and collects train operation information at historical moments in real time and sends it to the edge side and cloud side step by step to provide basic data for model training. Train operation information includes train traction or braking force, speed, track conditions and other data. The edge side preprocesses the train operation information and uses a window size of The second sample set of the edge side train digital twin model is partially updated; the cloud side uses the artificial intelligence platform to use a larger data window to The third sample set is used to globally update the cloud-side digital twin model.

[0065] In another exemplary embodiment of the present application, the above step 201 includes the following steps 301 to 302.

[0066] Step 301: Use a first sample set to train an initial digital twin model to obtain an offline digital twin model; the first sample set includes train operation information at several historical moments within a sample historical time period.

[0067] Step 302: Use the second sample set to update the parameters of the self-attention mechanism layer in the edge-side digital twin model online to obtain an edge feedback model; the second sample set includes The train operation information at each historical moment; the initial model parameters of the edge-side digital twin model are the model parameters of the offline digital twin model.

[0068] Step 303: Use the third sample set to perform online global correction on the cloud-side digital twin model to obtain a cloud-side return model; the third sample set includes The train operation information at each historical moment; the initial model parameters of the cloud-side digital twin model are the model parameters of the offline digital twin model; .

[0069] According to formula (5), the running state of the train at the next moment is affected by both internal and external factors. Internal factors include the control force and speed at the historical moment, and external factors include factors such as running road conditions and weather. Therefore, the sample set of the constructed digital twin model can be defined as shown in the following formula (6).

[0070] (6).

[0071] Under high-frequency sampling, the speed of the train at the next moment is highly correlated with the historical speed, which is due to the continuity and inertia of the train's movement. Therefore, when designing the neural network framework, the historical speed feature should be placed at the end of the network to capture the dynamic characteristics and trends of speed changes more quickly.

[0072] In addition, the changing trend and amount of the train state are mainly affected by external forces and train driving, which have different action characteristics. For example, factors such as slope, curve curvature and tunnel affect the train relatively quickly, while the control force will act on the train with a certain delay due to the characteristics of the actuator. Therefore, this embodiment adopts two sets of parallel network structures: one set processes the control force data in the time dimension, and the other set processes the track data in the feature dimension, so as to better identify the basic influence of each driving factor on the train state.

[0073] Finally, the data-driven modeling problem of high-speed trains can be defined as a time series problem with multivariable inputs. In order to improve the performance of the model, the self-attention mechanism is introduced into the GRU network, that is, the end-side digital twin model is an improved GRU network; the improved GRU network is obtained by introducing a self-attention mechanism layer into the GRU network.

[0074] like Figure 2 As shown, in order to better capture the local relationship between the input sequence and the train state, this application uses the hidden states of multiple GRU neurons to progressively calculate the attention weight. Here, the weight calculation of the first feature dimension is taken as an example, and the weight matrix The calculation process of can be expressed as shown in formula (7).

[0075] (7).

[0076] Among them, the control sequence ; GRU hidden state ; Indicates that the data is merged in the last dimension; and are the weight parameters of the first linear layer and the second linear layer respectively; and are the bias parameters for the first linear layer and the second linear layer respectively; is a key model hyperparameter, representing the hidden state scale; Used to enhance the nonlinear mapping of hidden layers, It is used to normalize the scores. The specific calculation method is shown in the following formula (8).

[0077] (8).

[0078] in, For each element in the matrix, is the first elements.

[0079] Due to the addition of the self-attention mechanism, a sliding time window is uniformly adopted For characteristic data order and Equivalent substitutions are made to reduce the number of parameters that need to be adjusted.

[0080] Offline training of the high-speed train digital twin model is a prerequisite for embedding it into the device-edge-cloud collaborative architecture. This operation is necessary to further determine the key model hyperparameters. and At the same time, a more accurate pre-model can be obtained, which can accelerate the convergence speed of the deployed model under new working conditions. The training goal of the neural network is to optimize the network weights and biases to minimize the predicted value. and the true label value To this end, we define the loss function As shown in the following formula (9).

[0081] (9).

[0082] in, is the number of training samples, according to Figure 2 The multi-dimensional attention GRU network framework shown in Figure 1 has a network output that can be expressed as formula (10).

[0083] (10).

[0084] in, is the output layer weight, For its bias, . , They are the weighted control force data and line data, is the hidden state of the last GRU unit, and their calculation formula is (11).

[0085] (11).

[0086] in, is the Hadamard product (element-wise product) operator, and are the weight matrices obtained by the attention mechanism, Update gate for GRU, is the hidden state at the previous moment, is the candidate hidden state, and the calculation formula is shown in the following formula (12).

[0087] (12).

[0088] in, and is the weight parameter, is the bias term, is the GRU unit input, Reset gates for GRU, and The calculation formula is (13).

[0089] (13).

[0090] in, and is the weight parameter, is the bias parameter. for Activation function, is a hyperbolic sine function, and the calculation formulas are shown in the following formula (14).

[0091] (14).

[0092] In the formula, are the elements in the corresponding matrix.

[0093] Finally, the gradient descent method is used to update the parameters of the internal weights and biases of the improved GRU network, where and The training algorithm is shown in equation (15).

[0094] (15).

[0095] in, is the offline learning rate and decays at a fixed rate. is the number of training iterations, and the remaining parameters are trained in the same way.

[0096] This application uses the root mean square error (RMSE) and mean absolute error (MAE) to evaluate the effectiveness of the training model and to select the best training hyperparameters and model hyperparameters, as shown in the following formula (16).

[0097] (16).

[0098] in, is the number of samples in the validation set.

[0099] The actual high-speed train operating environment is complex and changeable, so it is necessary to use offline model parameters to initialize the edge-side online model parameters and update them in real time. In the edge-cloud collaborative framework, considering that the edge side mainly deals with the model drift problem caused by small nonlinear fluctuations that occur in a short period of time, a window size of The second sample set is used to update the attention layer parameters in the edge digital twin model online. The update algorithm is shown in Equation (17).

[0100] (17).

[0101] in, , Represent all weight parameters and bias parameters of the attention layer respectively; is the online learning rate; is the iteration round; online loss The calculation formula is shown in formula (18).

[0102] (18).

[0103] In addition, in order to achieve rapid convergence of the edge-side digital twin model under various working conditions, an online learning rate switching strategy is adopted to dynamically adjust the online learning rate by monitoring the MAE loss, and finally save the model parameters that achieve the minimum loss. The online learning rate switching strategy can be expressed as shown in the following formula (19).

[0104] (19).

[0105] Among them, the initial learning rate search space is set before the train runs , is the initial model loss.

[0106] Since the training of the edge-side digital twin model is based on small batch data and insufficient update of model parameters, when there are large dynamic changes in train operation, the edge-side digital twin model may not be sufficient to achieve sufficient prediction accuracy within the allowed time. In order to ensure the high-precision prediction of the final edge-side digital twin model, a cloud-side digital twin model is established on the cloud side. The offline digital twin model parameters are used as the initial parameters of the cloud-side digital twin model, and a larger window size is used. N The third data set is used to perform online global correction on the cloud-side digital twin model. The parameter update method and learning rate switching strategy of the cloud-side digital twin model are the same as those of the edge-side digital twin model, which will not be repeated here.

[0107] In the above step 205, the self-correction mechanism is used to calibrate the device-side digital twin model using the edge-side digital twin model and the cloud-based backhaul model, and obtaining the corrected device-side digital twin model may include the following steps 401 to 404.

[0108] Step 401: Use the fourth sample set to evaluate the cloud backhaul model and the edge digital twin model respectively to obtain a first error evaluation value and a second error evaluation value; the fourth sample set includes Train operation information at historical moments.

[0109] Step 402: When the first error evaluation value is less than the second error evaluation value, the model parameters of the edge-side digital twin model are corrected using the model parameters of the cloud-transmitted model to obtain a corrected edge-side digital twin model.

[0110] Step 403: Use the fourth sample set to evaluate the corrected edge-side digital twin model and edge feedback model, respectively, to obtain a third error evaluation value and a fourth error evaluation value.

[0111] Step 404: When the third error evaluation value is less than the fourth error evaluation value, the device-side digital twin model is corrected using the model parameters of the edge-side digital twin model to obtain a corrected device-side digital twin model.

[0112] The window size is The fourth sample set and formula (20) are used to evaluate the three-terminal model, and the calculation formula of the error evaluation value is shown in the following formula (20).

[0113] (20).

[0114] in, Indicates The error value of the digital twin model; , representing the end side, edge side, and cloud side respectively.

[0115] The edge side device receives the cloud side return model in real time and stores the edge side return model. This application uses a self-correction mechanism to monitor the accuracy of the digital twin model of each end train in real time. When the end side train digital twin model is not optimal, the edge side train digital twin model is used to correct it. First, the performance of the cloud side return model and the edge side digital twin model is evaluated based on the above formula (20). , the parameters of the edge digital twin model are corrected using the cloud-based backhaul model parameters. The performance of the edge digital twin model and the edge backhaul model are then evaluated. , the edge-side digital twin model parameters are used to correct the end-side digital twin model. Through this self-correction mechanism, the best model obtained through distributed training can be continuously provided to the high-speed train controller to maximize the potential of the MPC control algorithm.

[0116] In order to solve the problem that conventional MPC optimization solutions consume relatively more hardware computing resources, which may lead to a long solution time on onboard equipment with limited hardware performance, thus causing high-speed train response delays and inability to achieve effective control, this application uses the multi-step cycle prediction capability of the digital twin model to move the optimization solution process forward to the period when the train actuator is in action, rather than leaving a separate solution time for the controller after obtaining the train operation status. The following is a detailed introduction to the design of a deployable MPC controller.

[0117] For the cyclic prediction in the prediction time domain, the terminal digital twin model is used as the prediction model of the MPC controller. Therefore, for the nonlinear mapping system of the high-speed train represented by equation (5), it can be expressed as shown in the following equation (21) in the prediction time domain.

[0118] (twenty one).

[0119] in, To predict the train running speed at the predicted time in the time domain; , , are the neural network input features corresponding to each prediction moment in the prediction time domain, namely , , They are the train running speed, control input and real-time track conditions corresponding to each prediction time in the prediction time domain; Represents the input and output mapping of the end-side digital twin model, Represents all weight parameters of the end-side digital twin model.

[0120] In step 206, the optimal control instruction sequence of the target high-speed train in the prediction time domain is determined based on the train running speed at the prediction moment, specifically including: using an MPC controller, during the execution of the control instruction at the previous moment, describing the train prediction control as a multi-constraint tracking optimization problem, solving the multi-constraint tracking optimization problem, and determining the optimal control instruction at the next moment according to the train running speed at the previous moment, and all predicted optimal control instructions at the next moment constitute the optimal control instruction sequence of the target high-speed train in the prediction time domain; the objective function of the multi-constraint tracking optimization problem is determined by the tracking error and energy consumption loss in the prediction time domain.

[0121] The goal of the MPC controller is to minimize the final tracking error and energy consumption by optimizing the control inputs of each step in the control time domain. In the actual train operation system, certain constraints must be met to ensure safety. For example, the control force of the train cannot exceed the upper limit of the actuator, and the train speed must not exceed the maximum speed allowed by the line. Therefore, this optimal control problem with constraints, that is, the multi-constraint tracking optimization problem, can be expressed as shown in the following equation (22).

[0122] (twenty two).

[0123] in, represents the objective function, To optimize the control input in the control time domain, is the reference trajectory in the predicted time domain, is the predicted output in the prediction time domain, is the tracking error weight matrix, is the energy consumption weight matrix, is the control input increment in the control time domain, represents transpose, is the predicted value at each prediction time in the prediction time domain; To control the input increment, is the maximum value of the control input increment; are the minimum and maximum values ​​of the optimized control input respectively; To predict the time The predicted output of The predicted time The minimum and maximum values ​​of the predicted output.

[0124] in, , , , They are respectively expressed as shown in the following formula (23).

[0125] (twenty three).

[0126] It can be seen from formula (22) that MPC actually optimizes and solves the tracking problem in the prediction time domain, and implements the same steps at each discrete moment, which requires more solution time than other control methods. Compared with high-performance simulation computers, the computing performance of high-speed train onboard computers is usually lower, so ordinary MPC cannot meet the high real-time requirements of the control algorithm for the train control system. To address similar problems, this application proposes a delay-free MPC controller based on a terminal-side digital twin model for high-speed trains. Its optimized control timing diagram is shown in the figure below. Figure 5 Assume here that .

[0127] Figure 5 middle, is a known dataset of train speed, road conditions and historical control force characteristics. The predicted value in the prediction time domain is Feature datasets with the same element type, is the input feature dataset of the neural network, To predict the time domain Feature datasets with the same element type, To solve the vehicle control command, is the target trajectory to be tracked, is the control instruction holding time and sampling period, where ; ; .

[0128] against Figure 5 In the MPC optimization process described in , each sampling moment predicts the future state within a period of time based on the current state. For example, by collecting Train operation information and upcoming control instructions , using the end digital twin model to predict The train speed at the time is obtained after being processed by the characteristic calculation unit of the MPC controller Train operation information, combined with the Continuous cycle prediction of train control instructions The train speed at the predicted time in the domain is optimized to obtain the optimal control instruction by solving the loss function in the predicted time domain. , and in the control instruction Transmitted to the executor immediately after execution The first control instruction is sent to the train, thereby achieving uninterrupted control of the high-speed train without delay in the control instructions.

[0129] Considering that the MPC prediction model is based on deep learning and cannot be expressed explicitly using mathematical formulas, in order to reduce the number of iterations and accelerate the controller solution process, this application uses the gradient descent method to solve the multi-constraint tracking optimization problem shown in formula (22).

[0130] (twenty four).

[0131] (25).

[0132] in, is the number of iterations, is the learning rate. Combined with formula (22), the objective function The derivative of the control sequence can be expressed as shown below.

[0133] (26).

[0134] Based on the above, formula (25) can be expressed as shown below.

[0135] (27).

[0136] In the following, by introducing the composite disturbance shown in Table 1 (including CRH380A train simulation parameters), enhancing the coupled nonlinearity of the train, simulating the actual operating environment of the train, and comparing four control algorithms: the EECC-MPC control algorithm (i.e., the high-speed train automatic driving predictive control method proposed in this application), the MPC control algorithm based on end-edge collaboration (EEC-MPC), the MPC control algorithm based on end-cloud collaboration (ECC-MPC), and the online local linear MPC control algorithm (RTL-MPC), in order to verify the effectiveness of end-edge-cloud collaborative control and its feasibility in online control of high-speed trains.

[0137] Table 1 CRH380A train simulation parameters

[0138]

[0139] Considering the functional differences between edge-side and cloud-side models, we conducted experimental research on the running data with a time series length of 200 for the edge-side model and 2000 for the cloud-side model. Half of the data was used for evaluation, and the other half was used as a selection. M and experimental data of N. In addition, to simulate the data transmission delay in reality and reserve sufficient training space for model training, we set the train to upload the latest data to the edge side every 2 seconds and upload the latest data from the edge side to the cloud side every 10 seconds. Figure 6-Figure 11 The model performance, velocity tracking, control force and acceleration results of the four online control algorithms are shown in Table 2.

[0140] Table 2 Control algorithm performance comparison

[0141]

[0142] Figure 6The prediction model identification errors of four control schemes during actual operation are shown. The parameter identification of the online identified local linearization model (RTL-MPC) is greatly affected when facing stronger nonlinear situations or rapid changes in train dynamic characteristics. This impact will be further amplified in the cyclic prediction in the prediction time domain, increasing the uncertainty and risk of operation. In contrast, the digital twin model based on multi-terminal collaboration shows better robustness, especially the EECC collaboration method, which can identify the train operation dynamics more timely and accurately.

[0143] Figure 7 The speed tracking effect of high-speed train under four control schemes is shown. Overall, the four schemes can effectively track the target trajectory. Figure 8 It can be seen that the EECC-MPC control algorithm combines the advantages of edge-end collaboration and cloud-end collaboration, and can show better tracking effects even when the target trajectory changes significantly. This shows that EECC-MPC has significant advantages in dealing with complex dynamic changes.

[0144] Fig. 9 The tracking error distribution of high-speed trains under four control schemes is shown. The results show that the tracking effect of the RTL-MPC control algorithm varies greatly at different times and is generally poor. In contrast, the MPC control algorithm based on multi-terminal collaboration shows better tracking stability, especially the EECC-MPC control algorithm, which combines the advantages of the other two collaborative control algorithms and shows higher tracking accuracy.

[0145] Fig.10 The control force curves calculated by the four control schemes are shown. It can be clearly seen that the control force fluctuation of the RTL-MPC control algorithm is large, which may cause large irreversible damage to the actuator and reduce the service life of the power system. In contrast, the control force generated by the collaborative control algorithm is relatively smooth, among which the EECC-MPC control algorithm performs best. The smooth control force curve shows that EECC-MPC can better balance the control accuracy and system stability.

[0146] Fig.11 The acceleration curves of high-speed trains under four control schemes are shown. The results show that the RTL-MPC control algorithm will cause a large fluctuation in the train running state, while the EECC-MPC control algorithm makes the train run more smoothly and improves the ride comfort.

[0147] The above four control schemes are quantitatively evaluated using IASE and IAAV quantitative indicators. The evaluation results are shown in Table 3.

[0148] Table 3 IASE and IAAE under four control schemes

[0149]

[0150] From the above content, we can see the effectiveness and identification robustness of the digital twin model, as well as the efficiency of end-edge-cloud collaborative training. Considering the overall quantitative indicators of complex operation scenarios and the dynamic operation characteristics of trains, the high-speed train automatic driving predictive control method (EECC-MPC control algorithm) proposed in this application has obvious advantages and can effectively deal with high-speed train control problems under complex dynamic operation conditions.

[0151] The present application also provides an application scenario, which applies the above-mentioned high-speed train automatic driving prediction control method. Specifically: the high-speed train automatic driving prediction control method provided in this embodiment can be applied in the high-speed train automatic driving scenario. The high-speed train automatic driving scenario includes an information acquisition link, a model prediction control link and an automatic driving link; the train operation information at the historical moment enters the model prediction control link from the information acquisition link, obtains the optimal control instruction sequence of the corresponding prediction time domain, and enters the downstream automatic driving link. The high-speed train automatic driving prediction control method provided in this embodiment belongs to the model prediction control link. Specifically, in the model prediction control link process for high-speed trains, the edge-side and cloud-side digital twin models can be updated in real time based on the distributed update strategy under the collaboration of the end-edge-cloud to obtain the edge backhaul model and the cloud backhaul model; the end-side digital twin model is equivalent to the edge backhaul model updated last time; the prediction errors of the end-side, edge-side, and cloud-side digital twin models are evaluated, and when the end-side digital twin model is not optimal, the self-correction mechanism is used to correct the end-side digital twin model using the edge-side digital twin model and the cloud backhaul model, and the corrected end-side digital twin model is used to cyclically predict the train running speed at multiple prediction times, and the optimal control instruction sequence of the target high-speed train in the prediction time domain is determined based on the train running speed at the prediction time.

[0152] Based on the same inventive concept, the embodiment of the present application also provides a high-speed train automatic driving prediction control device for implementing the high-speed train automatic driving prediction control method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more high-speed train automatic driving prediction control device embodiments provided below can refer to the limitations of the high-speed train automatic driving prediction control method above, and will not be repeated here.

[0153] In an exemplary embodiment, the present application also provides a high-speed train automatic driving prediction control device including the following modules:

[0154] The end-edge-cloud collaborative distributed update module is used to: based on the distributed update strategy under the end-edge-cloud collaboration, update the model parameters of the edge-side digital twin model and the cloud-side digital twin model in real time to obtain the edge return model and the cloud return model;

[0155] The train operation information acquisition module is used to: acquire the train operation information of the target high-speed train during a set period; the set period includes multiple historical moments; the train operation information includes the train operation speed, track conditions and control instructions;

[0156] The train speed prediction module is used to: for each historical moment within a set period, input the train operation information of the target high-speed train at the historical moment into the end-side digital twin model to obtain the predicted train speed at the next moment of the historical moment; the end-side digital twin model is the edge feedback model updated last time;

[0157] The terminal-side error evaluation value calculation module is used to calculate the terminal-side error evaluation value of the terminal-side digital twin model according to the predicted train speed at the next moment of all historical moments within a set period and the actual train running speed;

[0158] The device-side digital twin model self-correction module is used to: determine whether the device-side digital twin model needs to be corrected according to the device-side error evaluation value of the device-side digital twin model; if so, use the self-correction mechanism to correct the device-side digital twin model using the edge-side digital twin model and the cloud-based backhaul model to obtain a corrected device-side digital twin model;

[0159] The optimal control instruction sequence determination module is used to: use the corrected end-side digital twin model to cyclically predict the train running speed at multiple prediction moments, and determine the optimal control instruction sequence of the target high-speed train in the prediction time domain based on the train running speed at the prediction moment; the optimal control instruction sequence includes the optimal control input corresponding to several prediction moments in the prediction time domain; when the control instruction at the previous moment is executed, the target high-speed train is controlled to execute the optimal control instruction corresponding to the next moment, so as to realize real-time control of the high-speed train.

[0160] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig.12As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store automatic driving prediction control data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a high-speed train automatic driving prediction control method is implemented.

[0161] Those skilled in the art will understand that Fig.12 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0162] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0163] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0164] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0165] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0166] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0167] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0168] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A high-speed train automatic driving predictive control method, characterized in that: The high-speed train automatic driving predictive control method comprises: Based on the distributed update strategy under the collaboration of edge, edge and cloud, the model parameters of the edge-side digital twin model and the cloud-side digital twin model are updated in real time to obtain the edge backhaul model and the cloud backhaul model. Obtaining train operation information of a target high-speed train during a set period; the set period includes multiple historical moments; the train operation information includes train operation speed, track conditions and control instructions; For each historical moment within the set time period, the train operation information of the target high-speed train at the historical moment is input into the end-side digital twin model to obtain the predicted train speed at the next moment of the historical moment; the end-side digital twin model is the edge feedback model updated last time; the end-side digital twin model is an improved GRU network; the improved GRU network is a network obtained after introducing a self-attention mechanism layer in the GRU network; the hidden states of multiple GRU neurons are used to progressively calculate the attention weights, and the weight matrix The calculation process is: ;in, To control the sequence; is the hidden state of GRU; Indicates that the data is merged in the last dimension; and are the weight parameters of the first linear layer and the second linear layer respectively; and are the bias parameters for the first linear layer and the second linear layer respectively; is a key model hyperparameter, representing the hidden state scale; Used to enhance the nonlinear mapping of hidden layers, Used to normalize the scores; Calculate the end-side error evaluation value of the end-side digital twin model based on the predicted train speed at the next moment and the actual train running speed of all historical moments within the set period; According to the end-side error evaluation value of the end-side digital twin model, determine whether the end-side digital twin model needs to be corrected. If so, use the self-correction mechanism to correct the end-side digital twin model using the edge-side digital twin model and the cloud-based backhaul model to obtain a corrected end-side digital twin model. The corrected end-side digital twin model is used to cyclically predict the train running speed at multiple prediction moments, and the optimal control instruction sequence of the target high-speed train in the prediction time domain is determined based on the train running speed at the prediction moment; the optimal control instruction sequence includes the optimal control input corresponding to several prediction moments in the prediction time domain; when the control instruction of the previous moment is executed, the target high-speed train is controlled to execute the optimal control instruction corresponding to the next moment, so as to realize real-time control of the high-speed train; The optimal control instruction sequence of the target high-speed train in the prediction time domain is determined based on the train running speed at the prediction moment, specifically including: using an MPC controller, during the execution of the control instruction at the previous moment, describing the train prediction control as a multi-constraint tracking optimization problem, solving the multi-constraint tracking optimization problem, determining the optimal control instruction at the next moment according to the train running speed at the previous moment, and using the multi-step cyclic prediction capability of the digital twin model to move the optimization solution process forward to the period when the train actuator is in action, and all predicted optimal control instructions at the next moment constitute the optimal control instruction sequence of the target high-speed train in the prediction time domain; the objective function of the multi-constraint tracking optimization problem is determined by the tracking error and energy consumption loss in the prediction time domain.

2. The high-speed train automatic driving prediction control method according to claim 1, characterized in that: Based on the distributed update strategy under the collaboration of edge, edge and cloud, the model parameters of the edge-side digital twin model and the cloud-side digital twin model are updated in real time to obtain the edge return model and the cloud return model, including: The initial digital twin model is trained using a first sample set to obtain an offline digital twin model; the first sample set includes train operation information at several historical moments within a sample historical time period; The second sample set is used to update the parameters of the self-attention mechanism layer in the edge-side digital twin model online to obtain the edge feedback model; the second sample set includes The train operation information at each historical moment; the initial model parameters of the edge-side digital twin model are the model parameters of the offline digital twin model; The third sample set is used to perform online global correction on the cloud-side digital twin model to obtain a cloud-side return model; the third sample set includes The train operation information at each historical moment; the initial model parameters of the cloud-side digital twin model are the model parameters of the offline digital twin model; .

3. The high-speed train automatic driving prediction control method according to claim 1, characterized in that: The self-correction mechanism is used to correct the device-side digital twin model using the edge-side digital twin model and the cloud-based backhaul model to obtain the corrected device-side digital twin model, which specifically includes: The fourth sample set is used to evaluate the cloud backhaul model and the edge digital twin model respectively to obtain the first error evaluation value and the second error evaluation value; the fourth sample set includes Train operation information at historical moments; When the first error evaluation value is less than the second error evaluation value, the model parameters of the edge-side digital twin model are corrected using the model parameters of the cloud-transmitted model to obtain a corrected edge-side digital twin model; Using the fourth sample set, respectively evaluating the corrected edge-side digital twin model and the edge feedback model to obtain a third error evaluation value and a fourth error evaluation value; When the third error evaluation value is less than the fourth error evaluation value, the device-side digital twin model is corrected using the model parameters of the edge-side digital twin model to obtain a corrected device-side digital twin model.

4. The high-speed train automatic driving prediction control method according to claim 1, characterized in that: The multi-constraint tracking optimization problem is expressed as follows: ； in, represents the objective function, To optimize the control input in the control time domain, is the reference trajectory in the predicted time domain, is the predicted output in the prediction time domain, is the tracking error weight matrix, is the energy consumption weight matrix, is the control input increment in the control time domain, represents transpose, is the predicted value at each prediction time in the prediction time domain; , , They are the train running speed, control input and real-time track conditions corresponding to each prediction time in the prediction time domain; Represents the input and output mapping of the end-side digital twin model, Represents all weight parameters of the end-side digital twin model, To control the input increment, is the maximum value of the control input increment; are the minimum and maximum values ​​of the optimized control input respectively; To predict the time The predicted output of The predicted time The minimum and maximum values ​​of the predicted output.

5. The high-speed train automatic driving prediction control method according to claim 1, characterized in that: The train running speed at the historical moment is sampled from the high-speed train dynamics model, which is as follows: ； in, for The train speed at the time, is the nonlinear transfer function of the train, for The train speed at the time, for The train speed at the time, for The control input at the moment, for The control input at the moment, Indicates real-time track conditions; real-time track conditions include slope, curvature radius and tunnel length; and are the orders of the two input variables representing the dynamic system.

6. A high-speed train automatic driving prediction control device, characterized in that: The high-speed train automatic driving prediction control device comprises: The end-edge-cloud collaborative distributed update module is used to: based on the distributed update strategy under the end-edge-cloud collaboration, update the model parameters of the edge-side digital twin model and the cloud-side digital twin model in real time to obtain the edge return model and the cloud return model; The train operation information acquisition module is used to: acquire the train operation information of the target high-speed train during a set period; the set period includes multiple historical moments; the train operation information includes the train operation speed, track conditions and control instructions; The train speed prediction module is used to: for each historical moment within a set period, input the train operation information of the target high-speed train at the historical moment into the end-side digital twin model to obtain the predicted train speed at the next moment of the historical moment; the end-side digital twin model is the edge feedback model updated last time; the end-side digital twin model is an improved GRU network; the improved GRU network is a network obtained after introducing a self-attention mechanism layer in the GRU network; the hidden states of multiple GRU neurons are used to progressively calculate the attention weights, and the weight matrix The calculation process is: ;in, To control the sequence; is the hidden state of GRU; Indicates that the data is merged in the last dimension; and are the weight parameters of the first linear layer and the second linear layer respectively; and are the bias parameters for the first linear layer and the second linear layer respectively; is a key model hyperparameter, representing the hidden state scale; Used to enhance the nonlinear mapping of hidden layers, Used to normalize the scores; The terminal-side error evaluation value calculation module is used to calculate the terminal-side error evaluation value of the terminal-side digital twin model according to the predicted train speed at the next moment of all historical moments within a set period and the actual train running speed; The device-side digital twin model self-correction module is used to: determine whether the device-side digital twin model needs to be corrected according to the device-side error evaluation value of the device-side digital twin model; if so, use the self-correction mechanism to correct the device-side digital twin model using the edge-side digital twin model and the cloud-based backhaul model to obtain a corrected device-side digital twin model; The optimal control instruction sequence determination module is used to: use the corrected end-side digital twin model to cyclically predict the train running speed at multiple prediction moments, and determine the optimal control instruction sequence of the target high-speed train in the prediction time domain based on the train running speed at the prediction moment; the optimal control instruction sequence includes the optimal control input corresponding to several prediction moments in the prediction time domain; when the control instruction at the previous moment is executed, the target high-speed train is controlled to execute the optimal control instruction corresponding to the next moment, so as to realize real-time control of the high-speed train; The optimal control instruction sequence of the target high-speed train in the prediction time domain is determined based on the train running speed at the prediction moment, specifically including: using an MPC controller, during the execution of the control instruction at the previous moment, describing the train prediction control as a multi-constraint tracking optimization problem, solving the multi-constraint tracking optimization problem, determining the optimal control instruction at the next moment according to the train running speed at the previous moment, and using the multi-step cyclic prediction capability of the digital twin model to move the optimization solution process forward to the period when the train actuator is in action, and all predicted optimal control instructions at the next moment constitute the optimal control instruction sequence of the target high-speed train in the prediction time domain; the objective function of the multi-constraint tracking optimization problem is determined by the tracking error and energy consumption loss in the prediction time domain.

7. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the high-speed train automatic driving predictive control method described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the high-speed train automatic driving predictive control method described in any one of claims 1 to 5 is implemented.

Images