Virtual marshalling control method and system based on data-driven predictive control

Abstract

The embodiment of the present specification provides a virtual marshalling control method and system based on data-driven predictive control, wherein the method comprises the following steps: predicting the speed values of the front vehicle in the next m periods through a pre-trained short-term dynamic model by using the speed, control quantity and slope information of the front vehicle in the previous n periods, wherein m and n are natural numbers; obtaining a predicted speed sequence according to the speed values in the next m periods and sending the predicted speed sequence to the rear vehicle; obtaining the predicted speed sequence by the rear vehicle, determining the marshalling distance in the next m periods according to the predicted speed sequence, calculating the target speed of the rear vehicle in the next m periods according to the marshalling distance, solving the control quantity required to achieve the target speed of the rear vehicle according to the target speed of the rear vehicle, and controlling the driving of the rear vehicle according to the control quantity. The embodiment of the present specification can more accurately predict the trajectory of the front vehicle, thereby optimizing the tracking performance of the rear vehicle.

Description

TECHNICAL FIELD

[0001] The present document relates to the technical field of model predictive control, and in particular to a virtual marshalling control method and system based on data-driven predictive control. BACKGROUND

[0002] With the rapid development of urban rail transit and the continuous improvement of transportation demand, higher requirements are put forward for the operation capacity of train control systems. The operation mode based on moving block becomes the core factor affecting train operation interval. In order to further improve the operation efficiency, it is necessary to break through the limitation of block section. One of the development directions can be considered from the perspective of train operation marshalling mode. The application of advanced train marshalling mode can improve the transportation efficiency of rail transit, shorten the departure time interval between trains, and improve the transportation capacity of trains, so as to meet the needs of traffic volume demand and national economic development.

[0003] Therefore, the concept of virtual marshalling is proposed. As shown in Figure 1 , the virtual marshalling technology realizes the cooperative operation mode of multiple trains at the same speed and minimum interval through wireless communication between trains. Through this way, the traditional physical coupling is changed into wireless communication coupling. The virtual marshalling breaks through the limitation of the original block section, further shortens the safety protection distance of the train, and the train adjustment is more flexible, which further improves the operation efficiency.

[0004] The concept of virtual marshalling also poses new challenges to the train control method. When the virtual train runs, the running track of the following train needs to be dynamically generated and adjusted in real time. Traditional control methods such as PID are difficult to achieve good control effect, so many control methods for virtual marshalling have been proposed. One of the commonly used methods is the model predictive control (MPC) method.

[0005] MPC, also known as moving horizon control (MHC), is an advanced process control method used to control processes under a series of constraints. Since the 1980s, it has been widely used in process industries such as chemical plants and oil refineries. MPC has the advantage of being able to handle strong constraints of processing systems, making it widely used in various industrial controls. In the prior art, as shown in Figure 2 , y k is the predicted trajectory, u k is the optimal control sequence, and r(t) is the desired trajectory. MPC can be divided into three steps:

[0006] (1) Model prediction: measure the current state of the system, take the measured state information as the initial state of the system, and predict the state of the system in a limited time domain through the established system model.

[0007] (2) Optimization solution: design a target function, obtain the state according to the prediction, and solve the optimization problem based on the establishment of the target function, and obtain the control sequence of the current state of the system through the optimization solution.

[0008] (3) Execution: apply the first element of the control sequence of the optimization solution to the control object, and repeat step (1) in the next sampling period.

[0009] In the MPC method currently proposed for virtual marshalling, more attention is usually paid to the algorithm of the optimization solution part, and the model part is basically a simple kinematics model or a simplified train dynamics model, resulting in certain deficiencies in prediction accuracy, and unable to adjust to changes in train performance during operation, thereby affecting the final control effect. SUMMARY

[0010] The purpose of the present application is to provide a virtual marshalling control method and system based on data-driven predictive control, which aims to solve the above problems in the prior art.

[0011] The present application provides a virtual marshalling control method based on data-driven predictive control, which is used for double-car virtual marshalling composed of a front car and a rear car, comprising:

[0012] The speed value of the rear m periods of the front car is predicted through the pre-trained short-term dynamics model by using the speed, control amount and slope information of the front n periods of the front car, the predicted speed sequence is obtained according to the speed value of the rear m periods, and the predicted speed sequence is sent to the rear car, wherein m and n are natural numbers.

[0013] The predicted speed sequence is obtained by the rear car, the marshalling distance of the rear m periods is determined according to the predicted speed sequence, the target speed of the rear m periods of the rear car is calculated according to the marshalling distance, the control amount required to achieve the target speed of the rear car is solved according to the target speed of the rear car, and the rear car is controlled according to the control amount.

[0014] The present application provides a virtual marshalling control system based on data-driven predictive control, which is used for double-car virtual marshalling composed of a front car and a rear car, comprising:

[0015] The prediction module is located in the front car, and is used for predicting the speed value of the rear m periods of the front car through the pre-trained short-term dynamics model by using the speed, control amount and slope information of the front n periods of the front car, obtaining the predicted speed sequence according to the speed value of the rear m periods, and sending the predicted speed sequence to the rear car, wherein m and n are natural numbers.

[0016] The solving module is located in the rear vehicle, and is configured to acquire the predicted speed sequence by the rear vehicle, determine a marshalling distance of m periods in the rear according to the predicted speed sequence, calculate a target speed of the rear vehicle of m periods in the rear according to the marshalling distance, solve a control quantity required for realizing the target speed of the rear vehicle according to the target speed of the rear vehicle, and control the rear vehicle to run according to the control quantity.

[0017] By using the embodiment of the present application, the trajectory of the front vehicle can be more accurately predicted, so as to optimize the tracking performance of the rear vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the one or more embodiments of the present application or the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without any creative labor.

[0019] Figure 1 is a schematic diagram of virtual marshalling in the prior art;

[0020] Figure 2 is a flowchart of MPC processing in the prior art;

[0021] Figure 3 is a flowchart of the virtual marshalling control method based on data-driven predictive control in the embodiment of the present application;

[0022] Figure 4 is a flowchart of detailed processing of the virtual marshalling control method based on data-driven predictive control in the embodiment of the present application;

[0023] Figure 5 is a schematic diagram of solving the control quantity by using the search algorithm in the embodiment of the present application;

[0024] Figure 6 is a schematic diagram of the virtual marshalling control system based on data-driven predictive control in the embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to solve the problems in the prior art, the short-term dynamic model of the virtual marshalling front vehicle is established by a data-driven manner in the embodiment of the present application. The short-term speed curve is predicted by the front vehicle during marshalling and sent to the rear vehicle of the marshalling. The control level is calculated by the rear vehicle of the marshalling according to the predicted speed curve of the front vehicle, so as to realize the tracking of the speed and the distance. Compared with the model predictive control method in the prior art, the technical solution of the embodiment of the present application is based on the machine learning manner, and the model is established by the data-driven manner, so as to more accurately predict the trajectory of the front vehicle, thereby optimizing the tracking performance of the rear vehicle.

[0026] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0027] Method Implementation Examples

[0028] According to embodiments of the present invention, a virtual train formation control method based on data-driven predictive control is provided for virtual train formation of two vehicles consisting of a lead vehicle and a follower vehicle. Figure 3 This is a flowchart of a virtual grouping control method based on data-driven predictive control according to an embodiment of the present invention, such as... Figure 3 As shown, the virtual grouping control method based on data-driven predictive control according to an embodiment of the present invention specifically includes:

[0029] Step S301: Based on the speed, control quantity, and gradient information of the preceding vehicle in the first n cycles, predict the speed value of the preceding vehicle in the next m cycles using a pre-trained short-time dynamics model. Obtain the predicted speed sequence based on the speed values ​​of the next m cycles and send it to the following vehicle, where m and n are natural numbers.

[0030] In this embodiment of the invention, all cycles of the preceding train within a certain period are grouped into n cycles. The speed, control parameters, and gradient information of each group's n cycles are used as input variables for a short-time dynamics model. The short-time dynamics model uses these input variables to predict the speed values ​​of the subsequent m cycles. Each group's input variables and speed values ​​are treated as a sample, and multiple samples are obtained within the given period. The input variables of these multiple samples are used as the input vector x of the short-time dynamics model, and the speed values ​​of these multiple samples are used as the ground truth y of the short-time dynamics model. A supervised learning approach is used to obtain the mapping relationship from x to y, resulting in a trained short-time dynamics model. In other words, this embodiment of the invention constructs a self-learning short-term train dynamics model using machine learning methods, and uses this model for model predictive control of virtual train formations.

[0031] In this embodiment of the invention, the controller of the preceding vehicle tracks the target speed of the current vehicle and calculates the control quantity based on the target speed of the current vehicle, drives the train based on the control quantity, and records the speed of the train in this cycle.

[0032] Step S302, the rear vehicle acquires the predicted speed sequence, determines the marshalling distance of the rear m periods according to the predicted speed sequence, calculates the rear vehicle target speed of the rear m periods according to the marshalling distance, solves the control amount required to realize the rear vehicle target speed according to the rear vehicle target speed, and controls the rear vehicle driving according to the control amount.

[0033] Wherein, the marshalling distance of the rear m periods according to the predicted speed sequence specifically includes: in the case of non-virtual marshalling driving, the minimum train driving distance is usually determined by the emergency braking "soft wall" protection model according to the front and rear vehicle speed, that is, the minimum value of the displacement difference between the front and rear vehicles during the emergency braking process under the speed of each vehicle, if the value is less than 0, the absolute value of the value is the minimum driving distance, if the value is greater than or equal to 0, the minimum driving distance is 0. In virtual marshalling operation, due to the characteristics that the front and rear vehicle speeds tend to be consistent, it can be determined that the difference between the rear vehicle speed and the front vehicle speed is less than a certain deviation, so that the front vehicle speed plus the deviation can be taken as the rear vehicle speed to be brought into the model, so as to calculate the marshalling distance under a certain front vehicle speed.

[0034] Wherein, the rear vehicle target speed of the rear m periods according to the marshalling distance specifically includes:

[0035] The rear vehicle target speed function of the rear m periods is determined according to formula 1:

[0036] Formula 1;

[0037] Wherein, is the i-th period speed of the front vehicle, is the i-th period speed of the rear vehicle, represents the distance term weight, is the distance between the i-th period head of the rear vehicle and the tail of the front vehicle, is the i-th period target marshalling distance, As shown in formula 2:

[0038] Formula 2;

[0039] Wherein, is the control period, is the initial distance;

[0040] In one control period, the initial distance of the front and rear vehicles and the front vehicle predicted speed of the rear m periods ~ Given that, constant, the rear vehicle speed of the rear m periods ~ is the solution target, and the constraints are formula 3 and formula 4:

[0041] Formula 3;

[0042] Formula 4;

[0043] wherein, is the minimum acceleration that the train can generate, is the maximum acceleration that the train can generate, is the speed limit considering the line, vehicle and signal factors;

[0044] The target speed of the rear vehicle in the next m periods is obtained by solving Formula 1, Formula 3 and Formula 4.

[0045] Solving the control amount required to achieve the target speed of the rear vehicle specifically includes:

[0046] Based on the short-term dynamic model of the rear vehicle, a mapping relationship between the control amount and the short-term speed is obtained, and a target function that minimizes the error between the short-term speed and the target speed with the control amount u as the variable is derived, as shown in Formula 5:

[0047] Formula 5;

[0048] wherein, is the speed sequence of the rear vehicle in the next m periods predicted by the short-term model under the control amount u;

[0049] The control amount required by the target speed of the rear vehicle based on Formula 5 is solved, and the basic flow is as follows Figure 5 , specifically including the following processing:

[0050] Step 1, input the target speed sequence;

[0051] Step 2, determine the initial value of the control amount u, and initialize the minimum error to infinity;

[0052] Step 3, predict the speed sequence of the control amount u through the short-term dynamic model;

[0053] Step 4, calculate the error between the speed sequence and the target speed sequence;

[0054] Step 5, judge whether the error is less than the current optimal value, if yes, take the current error as the minimum error, and record the corresponding control amount u, if no, judge whether the search condition is met, if yes, select the next search control amount u according to the search rule, return to Step 3, if no, output the solving result control amount u.

[0055] The above technical solutions of the embodiments of the present application will be described in detail below in combination with the drawings.

[0056] The embodiment of the present application is directed to the current implementation possibility of double car virtual marshalling design, wherein the car before marshalling only needs to run according to the mode specified in the operation plan to ensure the accuracy of the plan, and the distance control of the virtual marshalling is guaranteed by the car after marshalling. As shown in Figure 4 The overall flow of the algorithm still adopts the three steps of prediction-solution-execution of MPC, wherein the prediction part is to predict the motion state (short-time speed sequence) of the car before marshalling, which is performed by the car before marshalling; the solution part calculates the control amount to be performed by the car after marshalling, which is performed by the car after marshalling. This design method has two considerations, one of which is that a large amount of data is needed to predict the train motion state, and if these data are sent to the car after marshalling for processing, the burden of train communication will be increased, and the other is that the car after marshalling needs to perform the solution step, and the calculation amount is large, so the operation work of the car after marshalling should be reduced as much as possible.

[0057] I. Establishment of short-time dynamics model:

[0058] Establishing an accurate speed prediction model is an important prerequisite for effective control. There are usually two ways to predict the speed of a train. One is to calculate a train speed-time curve in the interval according to the historical running information of the train. This method is better for trains running in a highly consistent manner in the interval, and the calculation amount is small, but in the case of changes in train running mode, such as setting temporary speed limit, adjusting train time, etc., the prediction effect cannot be guaranteed. The second is to input a model simulating train dynamics by combining environmental information, train state, control amount sequence, etc., to deduce the train running state in the short future. This method is not affected by changes in train running mode in the interval, and is suitable for various situations, but the prediction time range is short (basically depends on the control delay of the train), and the establishment of the model has certain difficulty, and the calculation amount is also larger. Considering the flexibility of the marshalling running environment and the demand for the length of the speed sequence by the subsequent control, the second prediction method is used in real time.

[0059] ① Variable selection:

[0060] The physical law should be as close to reality as possible, and the factors that will affect the train running speed in the physical range under actual conditions should be considered to avoid the influence of artificial defined factors such as control strategy, line speed limit, driving habit, etc. According to the above consideration, the train speed, control amount (divided into traction control amount and braking control amount) and slope in a certain time window (n period) are selected as the input variables of the model.

[0061] ② Data collection:

[0062] Whether it is automatic driving or manual driving, as long as the running data of the train is recorded for a period of time, including speed, slope and control amount information (if the difference in dynamics of the same type of train is within an acceptable range, the vehicle does not need to be limited, as long as the same type of vehicle can be used), it can be collected as a data set for the model to learn. Generally speaking, for an automatically driven train, the ATO log can automatically record the above data, and no additional collection is required. If you want to collect data from a non-automatic train, you need to install additional collection equipment.

[0063] ③ Model establishment:

[0064] The short-term dynamic model can select a neural network model, a time series model, a decision tree series model, etc. The embodiments of the present application do not limit the use of a specific machine learning model. If the model needs to predict the speed of the m period after the n period variable, group all periods in a period of time of the preceding vehicle into groups with n periods, and take the speed, control amount and slope information of n periods in each group as the input variables of the short-term dynamic model. The speed value of the m period after the short-term dynamic model is predicted by the input variable, and the input variable and speed value of each group are taken as a sample. A plurality of samples in the period of time are obtained, the input variables of the plurality of samples are taken as the input vector x of the short-term dynamic model, and the speed values of the plurality of samples are taken as the true value y of the short-term dynamic model. The mapping relationship from x to y is obtained in a supervised learning manner, and the trained short-term dynamic model is obtained.

[0065] In addition, although only the preceding vehicle and the model needs to be used for speed prediction during virtual marshalling control, a speed prediction model also needs to be established for the following vehicle after virtual marshalling. One is because the relationship between the preceding and following vehicles will change after turning back, and the other is because the model of the following vehicle is also needed in the subsequent control amount solving.

[0066] II. Calculate the target speed of the following vehicle:

[0067] The tracking goal of virtual marshalling is to make the two trains perform as much as possible like a train, that is, at each time: ① the distance between the tail of the preceding vehicle and the head of the following vehicle approaches 0; ② the speeds of the two vehicles approach the same. Considering that the two goals cannot be simultaneously optimal in some scenarios (for example, the interval between the preceding and following vehicles is widened due to interference, at which time the following vehicle needs to exceed the speed of the preceding vehicle to compensate for the widened interval), the two goals can be combined into a single optimization goal by assigning weights, considering the m period case, that is:

[0068] Formula 1;

[0069] wherein, is the speed of the i period of the preceding vehicle, is the speed of the i period of the following vehicle, representative distance term weight, is the distance between the head of the ith vehicle and the tail of the (i-1)th vehicle, is the target consist distance of the ith vehicle, as shown in Equation 2:

[0070] Equation 2;

[0071] is the control period.

[0072] In one control period, the initial distance between the front and rear vehicles and the predicted speed of the front vehicle after m periods ~ is known, i.e., a constant, the speed of the rear vehicle after m periods ~ is the target, the constraint of which is:

[0073]

[0074]

[0075] where is the initial speed of the rear vehicle, is the minimum acceleration that the train can produce, is the maximum acceleration that the train can produce, is the speed limit considering factors such as the line, vehicle, signal, etc.

[0076] It is not difficult to see that the above optimization problem is a constrained multivariate quadratic programming problem, and there are mature solution algorithms, which will not be described here.

[0077] III. Control quantity solution

[0078] The ultimate goal of control is to calculate the control quantity u that the rear vehicle needs to issue after the consist. Since the target speed of the rear vehicle has been obtained, a variety of control algorithms can be used to solve it here. In order to reduce the additional controller design and tuning work, the short-term dynamics model of the rear vehicle can be used here to obtain the mapping relationship between the control quantity and the short-term speed, so as to obtain the optimization problem of minimizing the error between the short-term speed and the target speed with the control quantity u as the variable, as follows:

[0079]

[0080] where is the speed sequence of the rear vehicle after m periods predicted by the short-term model under the application of the control quantity u.

[0081] Since the short-term dynamics model is usually a nonlinear model, the optimization problem cannot be solved by quadratic programming or the like, and a search algorithm such as simulated annealing, artificial fish swarm or the like can be used for solving.

[0082] As shown in Figure 5 The specific processing includes the following: first, input the target speed sequence, select the initial value of the control quantity u, and initialize the minimum error to infinity; predict the speed sequence of the u control quantity through the dynamics model, calculate the error of the speed sequence and the target speed; judge whether the error is less than the current optimal value, if yes, record the current error as the minimum error, and record the corresponding control quantity, otherwise, judge whether the search condition is met, if yes, select the next search u value according to the search rule, and return to the step of predicting the speed sequence of the u control quantity through the dynamics model, if not, output the solving result, that is, the recorded corresponding control quantity.

[0083] In summary, the embodiment of the present application establishes a short-term dynamics model of a virtual marshalling front vehicle in a data-driven manner, the front vehicle predicts a short-term speed curve during marshalling and sends it to a marshalling rear vehicle, the rear vehicle calculates a control level according to the predicted speed curve of the front vehicle, and realizes tracking of the speed and the distance. Compared with the existing model predictive control method, the technical scheme of the embodiment of the present application is based on machine learning, the model is established through data driving, the trajectory of the front vehicle is more accurately predicted, and the tracking performance of the rear vehicle is optimized.

[0084] System embodiment

[0085] According to the embodiment of the present application, a virtual marshalling control system based on data-driven predictive control is provided for a double-vehicle virtual marshalling composed of a front vehicle and a rear vehicle, Figure 6 is a schematic diagram of the virtual marshalling control system based on data-driven predictive control of the embodiment of the present application, as Figure 6 The virtual marshalling control system based on data-driven predictive control according to the embodiment of the present application specifically includes:

[0086] The prediction module 60 is located in the front vehicle, and is used for predicting the speed values of the last m periods of the front vehicle through the first n period speed, the control quantity and the slope information of the front vehicle, and through a pre-trained short-term dynamics model, acquiring a predicted speed sequence from the speed values of the last m periods and sending it to the rear vehicle, wherein m and n are natural numbers;

[0087] The solving module 62 is located in the rear vehicle, and is used for acquiring the predicted speed sequence by the rear vehicle, determining the marshalling distance of the last m periods according to the predicted speed sequence, calculating the target speed of the rear vehicle for the last m periods according to the marshalling distance, solving the control quantity required for realizing the target speed of the rear vehicle according to the target speed of the rear vehicle, and controlling the vehicle of the rear vehicle according to the control quantity. The solving module 62 is specifically used for:

[0088] The rear vehicle target speed function of the last m periods is determined according to formula 1:

[0089] Formula 1;

[0090] Wherein, is the speed of the front vehicle in the i th period, is the speed of the rear vehicle in the i th period, represents the distance term weight, is the distance between the front of the rear vehicle and the tail of the front vehicle in the i th period, is the target formation distance in the i th period, As shown in formula 2:

[0091] Formula 2;

[0092] Wherein, is the control period, is the initial distance;

[0093] In a control period, the initial distance between the front and rear vehicles And the predicted speed of the front vehicle in the last m periods ~ Given that, i.e. constant, the speed of the rear vehicle in the last m periods ~ To solve the target, the constraints are formula 3 and formula 4:

[0094] Formula 3;

[0095] Formula 4;

[0096] Wherein, is the initial speed of the rear vehicle, is the minimum acceleration that the train can generate, is the maximum acceleration that the train can generate, is the speed limit considering the factors of line, vehicle and signal;

[0097] The target speed of the rear vehicle in the last m periods is obtained by solving formula 1, formula 3 and formula 4.

[0098] Wherein, the control quantity required to achieve the rear vehicle target speed specifically includes:

[0099] Based on the short-term dynamics model of the rear vehicle, the mapping relationship between the control quantity and the short-term speed is obtained, and the objective function of minimizing the error between the short-term speed and the target speed with the control quantity u as the variable is obtained, as shown in formula 5:

[0100] Formula 5;

[0101] wherein, is to predict the speed sequence of the rear vehicle in the next m periods under the control amount u by the short-time model;

[0102] Solve the control amount required for the target speed of the rear vehicle based on formula 5. Specifically, the following processing is included:

[0103] Step 1, input the target speed sequence;

[0104] Step 2, determine the initial value of the control amount u, and initialize the minimum error to infinity;

[0105] Step 3, predict the speed sequence of the control amount u by the short-time dynamics model;

[0106] Step 4, calculate the error of the speed sequence and the target speed sequence;

[0107] Step 5, judge whether the error is less than the current optimal value, if yes, take the current error as the minimum error, and record the corresponding control amount u, if no, judge whether the search condition is met, if yes, select the next search control amount u according to the search rule, and return to step 3, if no, output the solving result control amount u.

[0108] The above system can further include:

[0109] A model training module is located in the front vehicle and the rear vehicle, which is used to group all periods of the front vehicle in a period of time into groups with n periods, take the speed, control amount and slope information of n periods in each group as the input variables of the short-time dynamics model, predict the speed value of the next m periods by the input variables, take the input variables and speed value of each group as a sample, obtain multiple samples in the period of time, take the input variables of the multiple samples as the input vector x of the short-time dynamics model, take the speed value of the multiple samples as the true value y of the short-time dynamics model, obtain the mapping relationship from x to y in a supervised learning manner, and obtain the trained short-time dynamics model.

[0110] The embodiment of the application is a system embodiment corresponding to the above-mentioned method embodiment, and the specific operation of each module can be understood with reference to the description of the method embodiment, which will not be repeated here.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A virtual grouping control method based on data-driven predictive control, characterized in that, The method for virtual formation of two vehicles consisting of a leading vehicle and a trailing vehicle specifically includes: Using the speed, control quantity, and gradient information of the preceding vehicle in the first n cycles, the speed value of the preceding vehicle in the next m cycles is predicted by a pre-trained short-time dynamics model. The predicted speed sequence is obtained based on the speed values ​​of the next m cycles and sent to the following vehicle, where m and n are natural numbers. The predicted speed sequence is obtained by the following vehicle. The grouping distance for the next m cycles is determined based on the predicted speed sequence. The target speed of the following vehicle for the next m cycles is calculated based on the grouping distance. The control quantity required to achieve the target speed of the following vehicle is solved based on the target speed of the following vehicle. The following vehicle is controlled to move according to the control quantity.

2. The method according to claim 1, characterized in that, The method further includes: The preceding vehicle's cycles over a given period are grouped into n-cycle groups. The speed, control parameters, and gradient information of each group's n cycles are used as input variables for a short-time dynamics model. The short-time dynamics model uses these input variables to predict the speed values ​​of the subsequent m cycles. Each group's input variables and speed values ​​are treated as a sample, and multiple samples are obtained over the given period. The input variables of these multiple samples are used as the input vector x of the short-time dynamics model, and the speed values ​​of these multiple samples are used as the ground truth y of the short-time dynamics model. The mapping relationship from x to y is obtained through supervised learning, resulting in a trained short-time dynamics model.

3. The method according to claim 1, characterized in that, The calculation of the target speed of the following vehicle over m cycles based on the grouping distance specifically includes: The target speed function of the following vehicle for the last m cycles is determined according to Formula 1: Formula 1: in, Let the speed of the preceding vehicle be the speed during the i-th cycle. Let the speed of the following vehicle in the i-th cycle be denoted as . Represents the weight of the distance term. Let be the distance between the front of the following vehicle and the rear of the preceding vehicle in the i-th cycle. Let be the target grouping distance in the i-th cycle. As shown in Formula 2: Formula 2: in, To control the cycle, This is the initial distance; The initial distance between the front and rear vehicles during one control cycle. Predicted speed of the vehicle m cycles ahead ~ Given, i.e., constant, the speed of the following vehicle over m cycles. ~ To solve the objective, the constraints are Equations 3 and 4: Formula 3: Formula 4: in, The minimum acceleration that the train can produce. The maximum acceleration that the train can produce. Speed ​​limits are implemented to take into account factors such as road conditions, vehicles, and signaling. The target speed of the following vehicle in the last m cycles is obtained by solving formulas 1, 3, and 4.

4. The method according to claim 3, characterized in that, The specific control quantity required to achieve the target speed of the following vehicle based on the target speed of the following vehicle includes: Based on the short-time dynamics model of the following vehicle, the mapping relationship between the control quantity and the short-time speed is obtained, and the objective function that minimizes the error between the short-time speed and the target speed with the control quantity u as the variable is derived, as shown in Formula 5: Formula 5: in, To predict the velocity sequence of the following vehicle over the last m cycles under the applied control quantity u using a short-time model; The required control quantity for the target speed of the rear vehicle is calculated based on Formula 5.

5. The method according to claim 4, characterized in that, The control quantity required to calculate the target speed of the rear vehicle based on Formula 5 specifically includes: Step 1, input the target velocity sequence; Step 2: Determine the initial value of the control variable u, and initialize the minimum error to infinity; Step 3: Predict the velocity sequence of the control variable u using a short-time dynamics model; Step 4: Calculate the error between the velocity sequence and the target velocity sequence; Step 5: Determine whether the error is less than the current optimal value. If the determination is yes, then take the current error as the minimum error and record the corresponding control quantity u. If the determination is no, then determine whether the conditions for continuing the search are met. If they are met, then select the next control quantity u to search according to the search rules and return to step 3. If the conditions are not met, output the solution result control quantity u.

6. A virtual train control system based on data-driven predictive control, characterized in that, The system is used for virtual formation of two vehicles consisting of a leading vehicle and a trailing vehicle, and specifically includes: The prediction module, located in the leading vehicle, is used to predict the speed value of the leading vehicle in the next m cycles based on the speed, control quantity, and gradient information of the leading vehicle in the first n cycles, using a pre-trained short-time dynamics model. Based on the speed values ​​of the next m cycles, the predicted speed sequence is obtained and sent to the following vehicle, where m and n are natural numbers. The solution module, located in the rear vehicle, is used to obtain the predicted speed sequence through the rear vehicle, determine the grouping distance for the next m cycles based on the predicted speed sequence, calculate the target speed of the rear vehicle for the next m cycles based on the grouping distance, solve for the control quantity required to achieve the target speed of the rear vehicle based on the target speed of the rear vehicle, and control the movement of the rear vehicle according to the control quantity.

7. The system according to claim 6, characterized in that, The system further includes: The model training module, located on the front and rear vehicles, is used to group all cycles of the front vehicle within a certain period into groups of n cycles. The speed, control parameters, and gradient information of each group of n cycles are used as input variables for the short-time dynamics model. The short-time dynamics model uses these input variables to predict the speed values ​​of the subsequent m cycles. Each group's input variables and speed values ​​are treated as a sample, and multiple samples are obtained within the specified time period. The input variables of these multiple samples are used as the input vector x of the short-time dynamics model, and the speed values ​​of these multiple samples are used as the ground truth y of the short-time dynamics model. The mapping relationship from x to y is obtained through supervised learning, resulting in the trained short-time dynamics model.

8. The system according to claim 6, characterized in that, The solution module is specifically used for: The target speed function of the following vehicle for the last m cycles is determined according to Formula 1: Formula 1: in, Let the speed of the preceding vehicle be the speed during the i-th cycle. Let the speed of the following vehicle in the i-th cycle be denoted as . Represents the weight of the distance term. Let be the distance between the front of the following vehicle and the rear of the preceding vehicle in the i-th cycle. Let be the target grouping distance in the i-th cycle. As shown in Formula 2: Formula 2: in, To control the cycle, This is the initial distance; The initial distance between the front and rear vehicles during one control cycle. Predicted speed of the vehicle m cycles ahead ~ Given, i.e., constant, the speed of the following vehicle over m cycles. ~ To solve the objective, the constraints are Equations 3 and 4: Formula 3: Formula 4: in, The minimum acceleration that the train can produce. The maximum acceleration that the train can produce. Speed ​​limits are implemented to take into account factors such as road conditions, vehicles, and signaling. The target speed of the following vehicle in the last m cycles is obtained by solving formulas 1, 3, and 4.

9. The system according to claim 8, characterized in that, The solution module is specifically used for: Based on the short-time dynamics model of the following vehicle, the mapping relationship between the control quantity and the short-time speed is obtained, and the objective function that minimizes the error between the short-time speed and the target speed with the control quantity u as the variable is derived, as shown in Formula 5: Formula 5: in, To predict the velocity sequence of the following vehicle over the last m cycles under the applied control quantity u using a short-time model; The required control quantity for the target speed of the rear vehicle is calculated based on Formula 5.

10. The system according to claim 9, characterized in that, The solution module is specifically used for: Step 1, input the target velocity sequence; Step 2: Determine the initial value of the control variable u, and initialize the minimum error to infinity; Step 3: Predict the velocity sequence of the control variable u using a short-time dynamics model; Step 4: Calculate the error between the velocity sequence and the target velocity sequence; Step 5: Determine whether the error is less than the current optimal value. If the determination is yes, then take the current error as the minimum error and record the corresponding control quantity u. If the determination is no, then determine whether the conditions for continuing the search are met. If they are met, then select the next control quantity u to search according to the search rules and return to step 3. If the conditions are not met, output the solution result control quantity u.

Images