A scenario-adaptive ATO control system for rail transit

Abstract

This invention proposes a scenario-adaptive ATO (Automatic Train Operation) control system for rail transit, comprising an operation information acquisition module 201, a scenario-adaptive control module 202, and a train traction / braking execution module 203. The scenario-adaptive control module 202 receives information input from the operation information acquisition module 201, and after learning, prediction, and optimization, generates control commands which are then input to the train traction / braking execution module 203. In this system, the parameter tuning module performs self-tuning of the five controller parameters of the sliding mode PID controller, achieving higher efficiency and accuracy compared to manual parameter tuning, and saving more time spent tuning controller parameters during train trial runs. The system prediction module effectively suppresses excessive overshoot that easily occurs when the train switches operating conditions, avoiding emergency braking triggered by overshoot exceeding the maximum overshoot threshold during operation, while also improving the accuracy of the train ATO controller in tracking the target curve.

Description

Technical Field

[0001] This application relates to the field of rail transit technology, and more specifically, to a rail transit ATO control system applicable to multiple scenarios. Background Technology

[0002] Urban rail transit, with its advantages of high capacity, high efficiency, and low pollution, has rapidly become the primary choice for many cities to solve traffic problems. During train operation, the speed of the wheelsets is obtained through speed sensors and transmitted to the train's ATP / ATO module. The actual speed of the train is compared with the target speed, and the control algorithm controls the train's traction and braking modules to ensure that the train runs at the predetermined target speed. Among them, the ATP module is responsible for the train's safety protection, while the ATO module performs the traction and braking process of the train under normal operating conditions.

[0003] ATO stands for Automatic Train Operation. Figure 1 A schematic diagram of an existing ATO (Automatic Train Operation) control system for rail transit is shown, including an operation information acquisition module, a PID control module / sliding mode PID control module, and a train traction / braking execution module. The operation information acquisition module is used to acquire real-time train status information, target curve information, and track condition information during train operation. The PID control module / sliding mode PID control module is used to calculate the appropriate command acceleration for the train based on the real-time train status information and transmit it to the train traction / braking execution module. The train traction / braking execution module is used to control the train to run according to the acceleration command value. Traditional ATO control algorithms, after acquiring train operation information, only use the sliding mode PID / PID control module and the acceleration command control module to control the train traction / braking process.

[0004] Existing ATO control algorithms mainly include PID control algorithms and sliding mode PID control algorithms.

[0005] The actual operation of trains usually has characteristics such as nonlinearity, time-varying uncertainty, and strong interference. It is difficult to achieve the ideal control effect by using conventional PID controllers. In actual operation, PID controllers need to adjust relevant parameters to obtain better control effects, but parameter tuning methods are complicated. Therefore, conventional PID controllers often have poor performance.

[0006] Sliding mode PID controllers are an improvement on traditional PID controllers. They can purposefully and continuously change according to the system's current state during dynamic processes, forcing the system to move along a predetermined "sliding mode" trajectory. They can precisely control the train to run along the target speed curve by comprehensively considering changes in speed and acceleration differences. The main drawbacks of sliding mode PID controllers are: they have two more parameters than PID controllers, making parameter tuning more complex and cumbersome; and they are prone to overshoot during operating condition transitions.

[0007] The ATO controller sets limits on the value by which the actual train speed exceeds the target speed curve. When the train speed is between 0 and 13 km / h, the maximum overshoot is approximately 1.2 km / h. When the train speed is above 13 km / h, the maximum overshoot is 2–2.5 km / h. The specific value of the maximum overshoot depends on the specifications of different manufacturers. If the overshoot exceeds the maximum overshoot threshold during operation, the train will automatically implement emergency braking, affecting operational efficiency. Considering the characteristics of operating condition switching, both PID controllers and sliding mode PID controllers will generate a certain amount of overshoot during operating condition switching, and in some scenarios, this may exceed the maximum overshoot limit, thus triggering emergency braking and affecting operational efficiency.

[0008] Therefore, a new type of ATO control system that can adapt to multiple scenarios needs to be designed based on the existing train control system in order to suppress the overshoot generated by the train when switching operating conditions. Summary of the Invention

[0009] The purpose of this application is to overcome the shortcomings of existing ATO controllers, which are prone to overshooting during operating condition switching, and to disclose a scenario-adaptive rail transit ATO control system that can meet the control requirements of train ATO controllers. In application, the scenarios include static target curve tracking during rail transit ATO operation between stations and tracking of dynamic stopping target curves generated due to temporary braking requirements during rail transit ATO operation between stations.

[0010] The objective of this application can be achieved through the following methods:

[0011] An adaptive rail transit ATO control system includes an operation information acquisition module 201, an adaptive control module 202, and a train traction / braking execution module 203. The adaptive control module 202 receives information input from the operation information acquisition module 201, and forms control commands through learning, prediction, and optimization, which are then input to the train traction / braking execution module 203.

[0012] The scene adaptive control module 202 includes a parameter tuning module 2021 and a train control module 2022. The parameter tuning module 2021 includes a machine learning module 20211, a model optimization and evaluation module 20212, and a train state update module 20213. The train control module 2022 includes a working condition judgment module 20221, a prediction module 20222, a sliding mode PID control module 20223, a command control module 20224, and an acceleration command generation module 20225, wherein:

[0013] The machine learning module 20211 is used to input the information of the operation information acquisition module 201 during the train trial operation, and continuously adjust the controller parameters of the sliding mode PID control module before the actual operation control of the train to determine the optimal controller parameter values ​​for the current line.

[0014] The model optimization and evaluation module 20212 is used to calculate and analyze different index requirements of the sliding mode PID control module 20223, determine the number of iterations of the machine learning module 20211, and finally output the results to the train control module 2022.

[0015] The train status update module 20213 is used to construct a simulation scenario based on relevant operating information, continuously provide input parameters to the machine learning module 20211, and simultaneously achieve the goal of parameter optimization based on the complete operating range of the train.

[0016] The operating condition judgment module 20221 receives information input from the operation information acquisition module 201 and is used to calculate the real-time operating condition of the train and provide it to the prediction module 20222.

[0017] The prediction module 20222 is used to determine whether the train's operating condition is about to change and to calculate a suitable command acceleration value that can suppress overshoot of the train speed before the operating condition changes. It is provided to the sliding mode PID control module 20223 and the command control module 20224.

[0018] The sliding mode PID control module 20223 provides the output command acceleration value to the acceleration command generation module 20225;

[0019] The command control module 20224 is used to transmit the appropriate command acceleration value generated by the prediction module 20222 when the train operating condition changes, and provide it to the acceleration command generation module 20225.

[0020] The acceleration command generation module 20225 is used to transmit real-time command acceleration values ​​to the train traction / braking execution module 203.

[0021] Specifically, within the same control cycle, only one module among the command control module 20224 and the sliding mode PID control module 20223 will send command acceleration. The sending of command acceleration is determined by the prediction module 20222. When the prediction module 20222 predicts that the train's operating condition is about to change, and continuing to use the command acceleration sent by the sliding mode PID control module 20223 will cause the train to overshoot during the operating condition change, it will send the target curve acceleration value after the operating condition change to the command control module 20224 in advance to suppress the train's speed overshoot during the upcoming operating condition change phase. At the same time, it will stop the sliding mode PID control module from outputting command acceleration in this control cycle.

[0022] Specifically, the acceleration command output by the acceleration command generation module 20225 undergoes an acceleration transmission and response process before being transmitted to the train traction / braking execution module 203.

[0023] Specifically, the sliding mode PID control module includes a sliding mode controller and a PID controller, namely a sliding mode control model and a PID model;

[0024] The neural network control algorithm mainly includes sliding mode control, PID control, intelligent learning algorithms, and simulation execution, running in the sliding mode PID control module and the machine learning module:

[0025] The system operation control model requires initial values ​​for velocity gain parameter kv, acceleration gain parameter ka, proportional element parameter kp, integral element parameter ki, and derivative element parameter kd to initialize the sliding mode controller and PID controller.

[0026] The main parameters included in the sliding mode control section are kv and ka. Once these two parameters are determined, the controller of the sliding mode control model is determined.

[0027] II. The main parameters included in the PID control section are kp, ki, and kd. Once these three parameters are determined, the controller effect of the PID model is determined.

[0028] III. Intelligent Learning Algorithm Section: By using the difference between the simulated acceleration of the train in the current simulation results and the target acceleration given in the ATO target speed curve, optimization learning is performed to iterate the values ​​of kp, ki, and kd in the PID controller;

[0029] Fourth, the system operation control model needs to read the existing ATO target speed curve data, track environmental condition data, etc. in advance to support the simulation operation part; the relevant parts will update the train's operation status based on the kinematic and dynamic calculations performed on the model corresponding to the current train control parameters and the track and environmental conditions; when the sliding mode controller and PID controller are determined, the train's operation control model at this moment is determined, and the simulation calculation and control of the train are realized through the simulation operation part.

[0030] The specific process is as follows:

[0031] Initially, the sliding mode controller outputs a control acceleration 'a'. s1 , and the actual acceleration a of the controlled train at the previous moment cu The input is fed into the PID controller, and a new control acceleration 'a' is obtained through PID control. s2 The two control accelerations are weighted to obtain the actual output command acceleration 'a'. s ;

[0032] Through the linear superposition of track conditions and environmental factors, and the nonlinear control of the controlled train, the actual running acceleration a of the train is obtained. cu And substitute it into the closed-loop control of the next stage;

[0033] In addition, a cu This will also serve as input to the intelligent control algorithm, determining whether the control parameters of the PID controller need optimization and iteration to ensure the train maintains good operating performance; if the target acceleration a in the ATO target speed curve data... target Compared with actual running acceleration a cu If the deviation does not exceed the set threshold, the calculation will proceed directly to the next time window; otherwise, before the next control, the PID controller needs to be optimized and learned to generate a completely new set of kp, ki, and kd for subsequent calculation and control.

[0034] Furthermore, the network structure parameters and iteration rules of the neural network are set as follows:

[0035] We choose to construct a 4-5-3 BP neural network, with the input layer taking the three indices k of the PID controller. p k i k d In addition to constant terms, the output of the output layer is the index value of the new PID controller; the activation functions of the hidden layer and the output layer are respectively Equation (1) and Equation (2), and Equation (3) is the calculation method of performance index parameters, where:

[0036] u(k) represents the actual output acceleration of the train at time k;

[0037] a cu This indicates the actual acceleration of the train.

[0038] a target This indicates the target acceleration that the train should achieve according to the given ATO curve;

[0039] E(k) represents the selected performance index parameter value at time k;

[0040]

[0041]

[0042]

[0043] The update methods for the output layer connection weights are Equations (4) and (5), respectively, and the update methods for the hidden layer connection weights are Equations (6) and (7), respectively.

[0044] η represents the model learning rate;

[0045] α represents the coefficient of inertia;

[0046] The output of the i-th neuron in the hidden layer at time k;

[0047] At time k, the output of the l-th neuron in the output layer;

[0048] At time k, the input of the l-th neuron in the output layer;

[0049] At time k, the output of the j-th neuron in the input layer;

[0050] Δω ij 2 (k+1) represents the change in weights of the input layer and hidden layer at time k+1;

[0051] Δω li 3 (k+1) represents the change in weights of the hidden layer and the output layer at time k+1;

[0052]

[0053]

[0054]

[0055]

[0056] Compared with the prior art, this application has the following advantages:

[0057] The parameter tuning module in this application system performs self-tuning of the five controller parameters of the sliding mode PID controller. Compared with manual parameter tuning, it has higher efficiency and accuracy and saves more time for controller parameter tuning during train test runs.

[0058] The system prediction module in the application can effectively suppress the excessive overshoot that the train is prone to when the operating conditions change, avoid the emergency braking triggered by the overshoot exceeding the maximum overshoot threshold during the operation of the train, and improve the accuracy of the train ATO controller in tracking the target curve. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of an existing rail transit ATO control system.

[0060] Figure 2 This is a schematic diagram of an adaptive rail transit ATO control system for the scenario described in this application.

[0061] Figure 3 This is the BP neural network control algorithm in the parameter tuning module of this application.

[0062] Figure 4 This is a flowchart of the neural network iteration in this application.

[0063] Figure 5 This is a flowchart of the adaptive rail transit ATO control method for the scenario described in this application.

[0064] Figure 6 This is a flowchart illustrating the operation of the prediction module when the train is under traction conditions.

[0065] Figure 7 This is a flowchart illustrating the operation of the prediction module for the train under braking conditions in this application.

[0066] Figure 8 This is a schematic diagram illustrating the application scenario and control results of the example. Detailed Implementation

[0067] The technical solutions provided in this application will be further described below with reference to specific embodiments and accompanying drawings. The advantages and features of this application will become clearer from the following description.

[0068] It should be noted that the embodiments of this application are preferred for implementation and are not intended to limit the application in any way. The technical features or combinations of technical features described in the embodiments of this application should not be considered isolated; they can be combined with each other to achieve better technical effects. The scope of the preferred embodiments of this application may also include other implementations, and this should be understood by those skilled in the art to which the embodiments of this application pertain.

[0069] Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and apparatus should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of exemplary embodiments may have different values.

[0070] The accompanying drawings in this application are all in a very simplified form and use non-precise proportions, intended only to facilitate and clarify the illustration of the embodiments of this application, and are not intended to limit the implementation of this application. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size, without affecting the effects and purposes achieved by this application, should fall within the scope of the technical content disclosed in this application. Furthermore, the same reference numerals appearing in the various drawings of this application represent the same features or components, and can be applied to different embodiments.

[0071] Figure 2 This is a schematic diagram of a rail transit ATO control system applicable to multiple scenarios according to an embodiment of this application. The rail transit ATO control system includes an operation information acquisition module 201, a scenario adaptive control module 202, and a train traction / braking execution module 203. The operation information acquisition module 201 and the train traction / braking execution module 203 are based on... Figure 1 The same module as the existing ATO control system in China; the scenario adaptive control module 202 receives information input from the operation information acquisition module 201, and after learning, prediction and optimization, forms control commands which are input to the train traction / braking execution module 203.

[0072] The scene adaptive control module 202 includes a parameter tuning module 2021 and a train control module 2022. The parameter tuning module 2021 includes a machine learning module 20211, a model optimization and evaluation module 20212, and a train state update module 20213. The train control module 2022 includes a working condition judgment module 20221, a prediction module 20222, a sliding mode PID control module 20223, a command control module 20224, and an acceleration command generation module 20225, wherein:

[0073] The machine learning module 20211 is used to input the information of the operation information acquisition module 201 during the train trial operation, and continuously adjust the controller parameters of the sliding mode PID control module before the actual operation control of the train to determine the optimal controller parameter values ​​for the current line.

[0074] The model optimization and evaluation module 20212 is used to calculate and analyze different index requirements of the sliding mode PID control module 20223, determine the number of iterations of the machine learning module 20211, and finally output the results to the train control module 2022.

[0075] The train status update module 20213 is used to construct a simulation scenario based on relevant operating information, continuously provide input parameters to the machine learning module 20211, and simultaneously achieve the goal of parameter optimization based on the complete operating range of the train.

[0076] The operating condition judgment module 20221 receives information input from the operation information acquisition module 201 and is used to calculate the real-time operating condition of the train and provide it to the prediction module 20222.

[0077] The prediction module 20222 is used to determine whether the train's operating condition is about to change and to calculate a suitable command acceleration value that can suppress overshoot of the train speed before the operating condition changes. It is provided to the sliding mode PID control module 20223 and the command control module 20224.

[0078] The sliding mode PID control module 20223 is adopted. Figure 1 It has the same function as the PID control module / sliding mode PID control module in the existing ATO control system, and provides the output command acceleration value to the acceleration command generation module 20225.

[0079] The command control module 20224 is used to transmit the appropriate command acceleration value generated by the prediction module 20222 when the train operating condition changes, and provide it to the acceleration command generation module 20225.

[0080] The acceleration command generation module 20225 is used to transmit real-time command acceleration values ​​to the train traction / braking execution module 203.

[0081] According to the embodiment, during the same control cycle, only one module among the command control module 20224 and the sliding mode PID control module 20223 sends command acceleration, and the sending of command acceleration is determined by the prediction module 20222. When the prediction module 20222 predicts that the train's operating condition is about to change, and continuing to use the command acceleration sent by the sliding mode PID control module 20223 would cause the train to overshoot during the operating condition change, it will send the target curve acceleration value after the operating condition change to the command control module 20224 in advance to suppress the train's speed overshoot during the upcoming operating condition change phase. At the same time, it will stop the sliding mode PID control module from outputting command acceleration during this control cycle.

[0082] According to the embodiment, the command acceleration output by the acceleration command generation module 20225 undergoes an acceleration transmission and response process before being transmitted to the train traction / braking execution module 203.

[0083] Specifically, if at least one of the operating condition judgment module 20221, prediction module 20222, or command control module 20224 fails, the train will continue to run on the line using only the command acceleration generated by the sliding mode PID control module 20223.

[0084] The sliding mode PID control module includes a sliding mode controller and a PID controller, namely a sliding mode control model and a PID model.

[0085] Figure 3 The BP neural network control algorithm mainly includes sliding mode control, PID control, intelligent learning algorithm, and simulation operation, running in the sliding mode PID control module and the machine learning module:

[0086] First, the system operation control model needs to be given initial values ​​for velocity gain parameter kv, acceleration gain parameter ka, proportional element parameter kp, integral element parameter ki, and derivative element parameter kd to initialize the sliding mode controller and PID controller.

[0087] The main parameters included in the sliding mode control section are kv and ka. Once these two parameters are determined, the controller of the sliding mode control model is determined.

[0088] II. The main parameters included in the PID control section are kp, ki, and kd. Once these three parameters are determined, the controller effect of the PID model is determined.

[0089] III. Intelligent Learning Algorithm Section:

[0090] Taking the BP neural network learning algorithm as an example, the difference between the simulated acceleration of the train in the current simulation results and the target acceleration given in the ATO target speed curve is used to perform optimization learning and iterate the values ​​of kp, ki, and kd in the PID controller.

[0091] Fourth, the system operation control model needs to read existing ATO target speed curve data and track environmental condition data in advance to support the simulation operation section. The relevant parts will update the train's operating state based on the kinematic and dynamic calculations performed on the model corresponding to the current train control parameters and the track and environmental conditions. Once the sliding mode controller and PID controller are determined, the train's current operation control model is determined, and the simulation operation section realizes the simulation calculation and control of the train.

[0092] The specific process is as follows:

[0093] Initially, the sliding mode controller outputs a control acceleration 'a'. s1 , and the actual acceleration a of the controlled train at the previous moment cu The input is fed into the PID controller, and a new control acceleration 'a' is obtained through PID control. s2 The two control accelerations are weighted together to obtain the actual output command acceleration 'a'. s .

[0094] Through the linear superposition of track conditions and environmental factors, and the nonlinear control of the controlled train, the actual running acceleration a of the train is obtained. cu This information is then incorporated into the closed-loop control of the next stage.

[0095] In addition, a cu This will also serve as input to the intelligent control algorithm (using the BP neural network learning algorithm as an example here) to determine whether the control parameters of the PID controller need to be optimized and iterated to ensure that the train can maintain good operating performance. If the target acceleration a in the ATO target speed curve data... target Compared with actual running acceleration a cu If the deviation does not exceed the set threshold, the calculation will proceed directly to the next time window; otherwise, before the next control, the PID controller needs to be optimized and learned to generate a completely new set of kp, ki, and kd for subsequent calculation and control.

[0096] based on Figure 3 The BP neural network control algorithm, the intelligent learning algorithm (running in the machine learning module) is as follows: Figure 4 As shown:

[0097] First, the structure of the neural network needs to be determined, namely the number of its layers and the specific number of neurons in each layer. Next, the optimization objective, i.e., the performance metrics that researchers are interested in, needs to be set. This objective can be set differently depending on different needs. Then, various parameters of the neural network need to be adjusted, mainly the optimization step size, maximum number of iterations, target error, and parameter learning rate. This step often involves empirical methods for trial and error. Finally, the iteration method for the parameters, i.e., the update rule for each weight in the neural network model, needs to be determined. In practical applications, there are multiple update methods, and different methods need to be selected.

[0098] After completing the architecture of the neural network model, the iterative training process begins. First, the model is iteratively evaluated. If the iteration conditions are not met, the process moves to the next calculation. When the iteration conditions are met, the model needs to acquire the required input value, which in this study corresponds to K. p K i K d And unit 1. After model training, the updated parameters are transferred into the basic control model to obtain the control index corresponding to the current parameters. The index is compared with the set optimization objective and the maximum number of iterations. If neither of the requirements is met, the iteration process is restarted. When one of the requirements is met, the optimized parameters are output, the optimization process ends, and the next calculation stage begins.

[0099] In practical applications, network structure parameters and iterative process functions different from those in the example can be selected to achieve similar effects. The following are examples, not limitations. The example BP neural network provides specific network structure parameter settings and iteration rules as follows:

[0100] We choose to construct a 4-5-3 BP neural network, with the input layer taking the three indices k of the PID controller. p k i k d In addition to the constant term, the output of the output layer is the index value of the new PID controller. The activation functions of the hidden layer and the output layer are Equation (1) and Equation (2) respectively, and Equation (3) is the calculation method of the performance index parameters, where:

[0101] u(k) represents the actual output acceleration of the train at time k;

[0102] a cu This indicates the actual acceleration of the train.

[0103] a target This indicates the target acceleration that the train should achieve according to the given ATO curve;

[0104] E(k) represents the selected performance index parameter value at time k.

[0105]

[0106]

[0107]

[0108] The update methods for the output layer connection weights are Equations (4) and (5), respectively, and the update methods for the hidden layer connection weights are Equations (6) and (7), respectively.

[0109] η represents the model learning rate;

[0110] α represents the coefficient of inertia;

[0111] The output of the i-th neuron in the hidden layer at time k;

[0112] At time k, the output of the l-th neuron in the output layer;

[0113] At time k, the input of the l-th neuron in the output layer;

[0114] At time k, the output of the j-th neuron in the input layer;

[0115] Δω ij 2 (k+1) represents the change in weights of the input layer and hidden layer at time k+1;

[0116] Δω li 3 (k+1) represents the change in weights of the hidden layer and the output layer at time k+1.

[0117]

[0118]

[0119]

[0120]

[0121] The aforementioned control system can be used to implement and execute ATO control processes to adapt to various scenarios, such as... Figure 5 As shown:

[0122] In step S301, the parameter tuning module 2021 sets the initial controller parameters for the sliding mode PID control module 20223, and continuously adjusts the controller parameters of the sliding mode PID control module 20223 during the train trial run to determine the optimal controller parameter values ​​for the current line; then proceed to step S302.

[0123] In step S302, the sliding mode PID control module 20223 calculates the command acceleration in real time within the current control cycle based on the train's current status information and target curve information in the operation information acquisition module 201, and transmits it to the acceleration command generation module 20225; then proceeds to step S303.

[0124] In S303, the operating condition judgment module 20221 calculates the current operating condition of the train by using the basic operating parameters of the train and the current status information of the train in the operating information acquisition module 201, and transmits the operating condition information to the prediction module 20222; then proceeds to step S304.

[0125] In S304, the prediction module 20222 generates a prediction point based on the current operating conditions of the train, and determines the generation of a command acceleration with overshoot suppression function based on the target value of the prediction point on the target curve and transmits it to the command control module 20224 or directly uses the general command acceleration generated by the sliding mode PID control module 20223; proceed to step S305.

[0126] In step S305, the acceleration command generation module 20225 receives the acceleration command sent from the command control module 20224 or the sliding mode PID control module 20223 and transmits it to the train traction / braking execution module 203. The train traction / braking execution module 203 controls the acceleration value of the train to the command acceleration value in real time; then proceed to step S302.

[0127] According to the embodiment, the prediction module 20222 performs different prediction point generation and judgment methods when the train is in traction and braking conditions to realize the overshoot suppression function when the train's operating conditions switch.

[0128] Among them, the prediction module 20222, when the train is in traction mode, determines whether the current control cycle is the optimal control cycle for switching the train from traction mode to braking mode by judging whether the prediction point overshoots on the target curve; for example, Figure 6 As shown.

[0129] Specifically, the prediction module 20222, when the train is in braking condition, determines whether the current control cycle restricts the train from switching from braking to traction mode by judging whether the predicted point's position on the target curve is in a braking state; for example, Figure 7 As shown.

[0130] Specifically, Figure 6The flowchart of the prediction module 20222 under train traction conditions is shown.

[0131] In S601, the prediction module 20222 obtains the train's current status information from the operation information acquisition module 201. According to the embodiment, the train's current status information includes the train's real-time speed, acceleration, position, and gradient information.

[0132] In S602, based on the train's current speed, acceleration, position, and gradient information obtained in S601, the prediction module 20222 calculates the coasting acceleration under the current train speed condition. According to the embodiment, the acceleration under the train's coasting condition is calculated by superimposing the train's basic resistance affected by the real-time speed of the hand-operated train and the gradient curve resistance.

[0133] In S603, the prediction module 20222 calculates the initial speed and position of the train after it decreases from its current acceleration to the acceleration under the coasting condition.

[0134] In S604, the prediction module 20222 calculates the train speed and position after a coasting time delay, which are denoted as the predicted point speed and the predicted point position, respectively.

[0135] In step S605, the prediction module 20222 compares the predicted train speed at the predicted point with the target speed of the target curve at the predicted point. If the predicted speed is greater than the target speed, proceed to step S606; otherwise, proceed to step S607.

[0136] In S606, the curve target acceleration of the predicted point position is sent to the command control module 20224 as the command acceleration for overshoot suppression.

[0137] In S607, the prediction module 20222 uses the sliding mode PID control module to generate command acceleration.

[0138] According to the embodiment, the prediction module performs calculations and outputs data in each control cycle when the train is in traction condition.

[0139] Specifically, Figure 7 The flowchart shows the operation of the prediction module when the train is braking.

[0140] In S701, the prediction module 20222 obtains the train's current status information from the operation information acquisition module 201. According to the embodiment, the train's current status information includes the train's real-time speed, acceleration, position, and gradient information.

[0141] In S702, based on the train's current speed, acceleration, position, and gradient information obtained in S701, the prediction module 20222 calculates the coasting acceleration under the current train speed condition. According to the embodiment, the acceleration under the train coasting condition is calculated by superimposing the train's basic resistance affected by the real-time speed of the manual train and the gradient curve resistance.

[0142] In S703, the prediction module 20222 calculates the initial speed and position of the train after it accelerates from the current acceleration to the acceleration under the coasting condition.

[0143] In S704, the prediction module 20222 calculates the train speed and position after two coasting time delays, which are denoted as the prediction point speed and prediction point position, respectively.

[0144] In S705, the prediction module 20222 determines whether the predicted point is in the parking braking stage on the target curve. If so, proceed to step S706; otherwise, proceed to step S707.

[0145] In S706, the curve target acceleration of the predicted point position is sent to the command control module 20224 as the command acceleration to suppress the switching of operating conditions.

[0146] In S707, the prediction module 20222 uses the sliding mode PID control module to generate command acceleration.

[0147] According to the embodiment, the prediction module for train braking conditions performs calculations and outputs results in each control cycle.

[0148] Figure 8 The illustration shows an application scenario and control result diagram of an ATO control method for rail transit applicable to multiple scenarios provided in this application embodiment.

[0149] The multiple scenarios include static target speed curve tracking scenario 8011 between stations, dynamic target speed curve tracking scenario 8021 with temporary braking during train traction, and dynamic target speed curve tracking scenario 8031 ​​with temporary braking during train cruising.

[0150] In the overshoot simulation results 8012, 8022 and 8032 for the three scenarios, the overshoot caused by the train speed did not exceed the maximum overshoot threshold of 2km / h.

[0151] The above description is merely a description of preferred embodiments of this application and is not intended to limit the scope of this application in any way. Any changes or modifications made by those skilled in the art based on the above-disclosed technical content should be considered as equivalent and valid embodiments and fall within the scope of protection of the technical solution of this application.

Claims

1. A scenario-adaptive ATO control system for rail transit, characterized in that, It comprises a running information acquisition module (201), a scene adaptive control module (202) and a train traction / braking execution module (203), the scene adaptive control module (202) receives information input of the running information acquisition module (201), forms control instructions after learning, prediction and optimization, and inputs the control instructions to the train traction / braking execution module (203); The scene adaptive control module (202) comprises a parameter setting module (2021) and a train control module (2022), the parameter setting module (2021) comprises a machine learning module (20211), a model optimization evaluation module (20212) and a train state updating module (20213), the train control module (2022) comprises a working condition judgment module (20221), a prediction module (20222), a sliding mode PID control module (20223), an instruction control module (20224) and an acceleration instruction generation module (20225), wherein: The machine learning module (20211) is used for inputting information of the running information acquisition module (201) in the train trial operation process, and continuously adjusting controller parameters of the sliding mode PID control module to determine optimal controller parameter values for the current line before actual train operation control; The model optimization evaluation module (20212) is used for calculating and analyzing different index requirements of the sliding mode PID control module (20223), judging the iteration number of the machine learning module (20211), and finally outputting the result to the train control module (2022); The train state updating module (20213) is used for constructing a simulation scene according to relevant running information, continuously providing input parameters for the machine learning module (20211), and achieving the purpose of parameter optimization based on the complete running interval of the train; The working condition judgment module (20221) receives information input of the running information acquisition module (201), and is used for realizing the function of calculating the real-time running working condition of the train and providing the prediction module (20222); The prediction module (20222) is used for realizing the functions of judging whether the working condition of the train is about to switch and calculating a suitable instruction acceleration value for suppressing overshoot of the train speed before the working condition of the train switches, and providing the sliding mode PID control module (20223) and the instruction control module (20224); The sliding mode PID control module (20223) provides the output instruction acceleration value to the acceleration instruction generation module (20225); The instruction control module (20224) is used for transmitting the suitable instruction acceleration value generated by the prediction module (20222) from the prediction module (20222) when the working condition of the train switches, and providing the acceleration instruction generation module (20225); The acceleration instruction generation module (20225) is used for realizing the function of transmitting the real-time instruction acceleration value to the train traction / braking execution module (203).

2. The scenario-adaptive ATO control system for rail transit according to claim 1, wherein, In the same control cycle, only one of the instruction control module (20224) and the sliding mode PID control module (20223) will send the instruction acceleration, and the sending of the instruction acceleration is determined by the prediction module (20222). When the prediction module (20222) predicts that the working condition of the train is about to switch, the use of the instruction acceleration sent by the sliding mode PID control module (20223) will cause the train to overshoot when the working condition switches. The target curve acceleration value after the working condition switches will be sent to the instruction control module (20224) in advance to suppress the speed overshoot of the train in the upcoming working condition switching stage, and the output of the instruction acceleration of the sliding mode PID control module in this control cycle will be stopped. The instruction acceleration output by the acceleration instruction generation module (20225) will undergo an acceleration transmission and response process before being transmitted to the train traction / braking execution module (203).

3. The scenario-adaptive ATO control system for rail transit according to claim 1, wherein, The sliding mode PID control module includes a sliding mode controller and a PID controller, i.e. a sliding mode control model and a PID model; the neural network control algorithm mainly includes sliding mode control, PID control, intelligent learning algorithm and simulation running, and runs in the sliding mode PID control module and the machine learning module: The system running control model needs to give the initial speed gain parameter kv, acceleration gain parameter ka, proportional link parameter kp, integral link parameter ki, and differential link parameter kd value to initialize and set the sliding mode controller and the PID controller; I. The main parameters of the sliding mode control part are kv and ka. When the two parameters are determined, the controller of the sliding mode control model is determined; II. The main parameters of the PID control part are kp, ki and kd. When the three parameters are determined, the effect of the PID model controller is determined; III. Intelligent learning algorithm part: through the difference between the simulation acceleration of the train in the current simulation result and the target acceleration given in the ATO target speed curve, the PID controller kp, ki and kd values are iterated through optimization learning; IV. The system running control model needs to read the existing ATO target speed curve data, line environment condition data, etc. in advance to support the simulation running part; the relevant part will update the running state of the train according to the kinematics and dynamics calculation of the model corresponding to the current train control parameter and the line and environment conditions; when the sliding mode controller and the PID controller are determined, the running control model of the train at this moment is determined, and the simulation calculation and control of the train are realized through the simulation running part.

4. The scenario-adaptive ATO control system for rail transit according to claim 3, wherein, The specific process is as follows: At the initial moment, the output is the control acceleration given by the sliding mode controller , and the actual running acceleration of the controlled train at the previous moment , are input together into the PID controller, and a new control acceleration is obtained via the PID control , the two control accelerations are weighted to obtain the actual output command acceleration ; The actual running acceleration of the train is obtained through linear superposition of line conditions and environmental factors and nonlinear control of the controlled train and substituted into the closed-loop control of the next stage; also, This will also serve as input to the intelligent control algorithm, determining whether the control parameters of the PID controller need optimization and iteration to ensure the train maintains good operating performance; if the target acceleration in the ATO target speed curve data... Compared with actual running acceleration If the deviation does not exceed the set threshold, the calculation will proceed directly to the next time window; otherwise, before the next control, the PID controller needs to be optimized and learned to generate a completely new set of kp, ki, and kd for subsequent calculation and control.

5. The scenario-adaptive ATO control system for rail transit according to claim 4, wherein, The network structure parameter setting and iteration rule of the neural network are as follows: The BP neural network of 4-5-3 structure is selected, the inputs of the input layer are three indexes of the PID controller , , and constant term, the output of the output layer is the index value of the new PID controller; the activation function of the hidden layer and the activation function of the output layer are formula (1) and formula (2) respectively, and formula (3) is the performance index parameter calculation mode, wherein: Ak represents the actual output acceleration of the train at time k; denotes the actual running acceleration of the train: a target acceleration that the train should reach according to a given ATO profile; represents the selected performance indicator parameter value at time k; The output layer connection weight update mode is formula (4) and formula (5), the hidden layer connection weight update mode is formula (6) and formula (7), and denotes the model learning rate; represents the coefficient of inertia; is the output of the i-th neuron of the hidden layer at time k; output of the output layer for time k; output layer for time k; and Yk= (Yk1, Yk2,..., Ykl)T, where Ykiis the output of the i-th neuron of output of the input layer jthneuron at time k; for k+1, the input layer and the hidden layer weight change amount; for k+1, the weight change of the hidden layer and the output layer;

Images