Urban rail train ATO control performance indoor simulation test method and system

Abstract

The application discloses a kind of urban rail train ATO control performance indoor simulation test method and system, it is related to urban rail train control technical field, method includes: train data set and the slope data of line where train is input to the preset ATO control performance indoor simulation test model, to obtain train state prediction data;Train state prediction data and the acceleration control analog quantity of ATO output are associated, and are marked to be generated as simulation test result;The preset ATO control performance indoor simulation test model includes train dynamics model and train motion model;Train dynamics model is based on train data set and the slope data of line where train is, determines train current cycle prediction actual acceleration;Train motion model is based on train current cycle prediction actual acceleration and train line, calculates train in next cycle speed and position.The application improves the correctness of train state prediction for automatic driving.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of urban rail train control technology, in particular to an urban rail train ATO control performance indoor simulation test method and system. BACKGROUND

[0002] When developing a new train control algorithm, a large amount of simulation verification needs to be performed on the control algorithm to adjust the parameters of the control system. It is too costly to adjust the parameters of the control algorithm on an actual line, and there is no relatively accurate indoor simulation platform for ATO indoor simulation testing before this. This is because traditional train dynamics modeling is mostly based on dynamic modeling. There are control delays, environmental influences and complex wheel-rail relationships in the actual operation of the train, and there is still a non-negligible error between the traditional dynamic model and the real train.

[0003] In recent years, the field of artificial intelligence has been developing, from Bayesian networks to deep learning and reinforcement learning, and more and more research has applied artificial intelligence technology to the field of rail transit. At present, the method of deep learning has shown good results in predicting regression problems. Recurrent neural network (RNN) has good performance in predicting problems with time series data as input. Long short-term memory (LSTM) solves the problem of short-term memory on the basis of RNN. At the same time, since convolutional neural network (CNN) can efficiently extract features from data, many studies combine it with RNN to complete more complex tasks.

[0004] Since the deep learning model is essentially a data-driven modeling, the prediction effect of the model is closely related to the training data. In practical applications, it is difficult to obtain data in various scenarios, especially in the case of abnormal train operation. Therefore, the current research is difficult to make predictions for abnormal train operation scenarios, i.e. scenarios that do not exist in the training set. And before the train is officially put into operation, it is difficult to obtain a large amount of real train operation data or operation data on the entire line, i.e. there is a problem of small and unbalanced data in the model training stage. SUMMARY

[0005] The purpose of the present application is to provide an urban rail train ATO control performance indoor simulation test method and system, which improves the correctness of the state prediction of the automatic driving train.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] An urban rail train ATO control performance indoor simulation test method, comprising:

[0008] obtain a train data set and slope data of a line on which the train is located in a current period; the train data set comprises train speed, acceleration control analog quantity output by the ATO, and actual acceleration of the train in a previous period;

[0009] input the train data set and the slope data of the line on which the train is located into a preset ATO control performance indoor simulation test model to obtain train state prediction data; the train state prediction data comprises train speed and position in a next period;

[0010] associate the train state prediction data and the acceleration control analog quantity output by the ATO, and mark the association as a simulation test result;

[0011] The preset ATO control performance indoor simulation test model comprises a train dynamics model and a train motion model; the train dynamics model is configured to determine train current period predicted actual acceleration based on the train data set and the slope data of the line on which the train is located; and the train motion model is configured to calculate train speed and position in a next period based on the train current period predicted actual acceleration and the line on which the train is located.

[0012] The train dynamics model is obtained by training a preset neural network based on a training sample set; each training sample in the training sample set comprises train corresponding train speed, acceleration control analog quantity output by the ATO, actual acceleration of the train in a previous historical period, slope data of the line on which the train is located, and train actual acceleration in any historical period; and the preset neural network is a CNN-BiLSTM neural network model based on an attention mechanism.

[0013] Optionally, the preset neural network comprises an input layer, a hidden layer, and an output layer connected in sequence;

[0014] The hidden layer comprises a convolution layer, a pooling layer, a Dropout layer, a BiLSTM network layer, and an attention mechanism layer connected in sequence;

[0015] In the training process of the preset neural network, the attention mechanism layer outputs initial predicted acceleration features, a preset loss function is used to calculate a loss value between the initial predicted acceleration features and train actual acceleration in the training sample; based on the loss value, a gradient of the preset loss function with respect to each layer parameter in the hidden layer is calculated by using back propagation through the chain rule of derivatives, and each layer parameter in the hidden layer is updated according to the gradient.

[0016] Optionally, the BiLSTM network layer comprises a plurality of LSTM units, and the BiLSTM network layer is configured to:

[0017] The feature vector output by the Dropout layer is taken as an input feature, and is input into the forward network and the backward network in a forward order and a reverse order respectively;

[0018] Based on the LSTM unit, a hidden state value at time t is calculated according to the input feature, a hidden state value and a cell state value of the LSTM unit at time t-1 through an activation function;

[0019] The hidden state value of the LSTM unit of the forward network corresponding to the same input feature is spliced with the hidden state value of the LSTM unit of the corresponding backward network to obtain an output feature vector.

[0020] Optionally, the attention mechanism layer is used for:

[0021] The feature vector output by the BiLSTM network layer is normalized into a probability distribution vector based on a softmax activation function, and the sum of each probability is 1;

[0022] The feature vector output by the BiLSTM network layer is multiplied by the probability distribution vector and summed to obtain an initial predicted acceleration feature.

[0023] Optionally, the slope data of the line on which the train is located includes a slope value of the line on which the train is located, a tilting direction of the slope section of the line on which the train is located relative to the logical direction of the line, and a vertical curve radius of the line on which the train is located.

[0024] Optionally, in the slope data of the line on which the train is located, the tilting direction of the slope section of the line on which the train is located relative to the logical direction of the line is a 0-1 value.

[0025] When the tilting direction of the slope section of the line on which the train is located relative to the logical direction of the line is a 0 value, it indicates that the slope section of the line on which the train is located is an uphill; and when the tilting direction of the slope section of the line on which the train is located relative to the logical direction of the line is a 1 value, it indicates that the slope section of the line on which the train is located is a downhill.

[0026] Optionally, the method further comprises:

[0027] Obtaining train operation line map data; the train operation line map data includes slope data and speed limit data corresponding to each operation line;

[0028] The train operation line map data, the current period predicted actual acceleration of the train, and the speed and position of the train in the next period are associated and fed back to the ATO to obtain an acceleration control analog quantity output by the ATO corresponding to the next period.

[0029] To achieve the above purpose, the application also provides the following technical solutions:

[0030] The application discloses an indoor simulation test system for ATO control performance of a city rail train.

[0031] A train data acquisition module is configured to acquire train data sets and slope data of a line on which the train is located in a current period, wherein the train data sets include train speed, acceleration control analog quantity output by ATO and actual acceleration of the train in a previous period;

[0032] A train state prediction module is configured to input the train data sets and the slope data of the line on which the train is located into a preset indoor simulation test model of ATO control performance, so as to obtain train state prediction data, wherein the train state prediction data include train speed and position in a next period;

[0033] A simulation result output module is configured to associate the train state prediction data and the acceleration control analog quantity output by ATO, and mark the association as a simulation test result.

[0034] The preset indoor simulation test model of ATO control performance includes a train dynamics model and a train motion model, wherein the train dynamics model is configured to determine predicted actual acceleration of the train in the current period based on the train data sets and the slope data of the line on which the train is located, and the train motion model is configured to calculate the train speed and position in the next period based on the predicted actual acceleration of the train in the current period and the line on which the train is located.

[0035] The train dynamics model is obtained by training a preset neural network based on a training sample set, wherein each training sample in the training sample set includes corresponding train speed, acceleration control analog quantity output by ATO, actual acceleration of the train in a previous historical period, slope data of the line on which the train is located and actual acceleration of the train in any historical period, and the preset neural network is a CNN-BiLSTM neural network model based on an attention mechanism.

[0036] According to the embodiments of the application, the following technical effects are achieved.

[0037] The application discloses a kind of urban rail train ATO control performance indoor simulation test method and system, on the basis of CNN (convolutional neural networks, convolutional neural network) and Bi-directional LSTM (Bi-directional LSTM, BiLSTM, Bi-directional long short-term memory network), attention mechanism is added to construct train dynamics model, then train current cycle prediction actual acceleration is calculated based on train dynamics model. Since attention mechanism can make model have stronger generalization ability, to establish more perfect accurate train dynamics model, so that model can be in different scenes different section under the premise of small data and unbalanced, different state prediction is carried out to train, i.e. It can improve the correctness of train state prediction. Then, using train motion model, based on train current cycle prediction actual acceleration and train line, the speed and position of train in next cycle are calculated, and the simulation of train running process is realized. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0039] Figure 1 The flowchart of the urban rail train ATO control performance indoor simulation test method of the present application is shown in the figure.

[0040] Figure 2 The structure diagram of the preset neural network of the present application is shown in the figure.

[0041] Figure 3 The structure diagram of the convolutional neural network composed of convolutional layer and pooling layer of the present application is shown in the figure.

[0042] Figure 4 The principle diagram of the Dropout layer of the present application is shown in the figure.

[0043] Figure 5 The neural network structure diagram of the BiLSTM layer of the present application is shown in the figure.

[0044] Figure 6 The principle diagram of the attention mechanism layer of the present application is shown in the figure.

[0045] Figure 7 The flowchart of indoor simulation test in the specific example of the present application is shown in the figure.

[0046] Figure 8 The structure diagram of the urban rail train ATO control performance indoor simulation test system of the present application is shown in the figure. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0048] In the development of the urban rail train control system, the parameters of the ATO (automatic train operation) control system need to be designed and adjusted according to the characteristics of the line and the vehicle, so as to improve the ATO control performance indicators including the parking accuracy, the average inter-station travel time, the control impact rate, the control adjustment frequency and the like.

[0049] Based on this, the present application provides an urban rail train ATO control performance indoor simulation test method and system, which trains a CNN-BiLSTM model based on an attention mechanism by using real line data and train operation data, so as to obtain an accurate train dynamics model. Then, the to-be-tested ATO control algorithm is connected with the train dynamics model, the train dynamics model periodically receives the control command of the control algorithm, and feeds back the train state to the control algorithm, so as to realize the simulation test of the control performance of the ATO control algorithm in the indoor environment.

[0050] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0051] Embodiment one

[0052] As shown in the figure, the present application provides an urban rail train ATO control performance indoor simulation test method, which comprises: Figure 1

[0053] Step 100, obtaining a train data set and slope data of a line on which the train is located in a current period; the train data set comprises train speed, acceleration control analog quantity output by the ATO and actual acceleration of the train in the last period.

[0054] Step 200, inputting the train data set and the slope data of the line on which the train is located into a preset ATO control performance indoor simulation test model, so as to obtain train state prediction data; the train state prediction data comprises speed and position of the train in the next period.

[0055] Step 300, associating the train state prediction data and the acceleration control analog quantity output by the ATO, and marking the generated simulation test result.​

[0056] The preset ATO control performance indoor simulation test model comprises a train dynamics model and a train motion model. The train dynamics model is configured to determine a train current period predicted actual acceleration based on the train data set and slope data of a line on which the train is running. The train motion model is configured to calculate a train speed and position in a next period based on the train current period predicted actual acceleration and the line on which the train is running. Specifically, the train dynamics model determines a train speed and position in a current period according to the train current period predicted actual acceleration and the line on which the train is running, and then calculates a train speed and position in a next period according to the train speed and position in the current period.

[0057] The train dynamics model is obtained by training a preset neural network based on a training sample set. Each training sample in the training sample set comprises train speed corresponding to the train, acceleration control analog quantity (representing ATO expected acceleration) output by the ATO, actual acceleration of the train in a previous historical period, slope data of the line on which the train is running, and actual acceleration of the train in any historical period. Each training sample is real line data, which is historical running data extracted from the on-board equipment of the train. If necessary, the actual acceleration of the train is calculated according to the speed difference between the current time and the previous time.

[0058] Generally, after obtaining the training sample set, the data in the training sample set needs to be preprocessed to eliminate the error data.

[0059] The slope data of the line on which the train is running comprises a slope value of the line on which the train is running, an inclination direction of a slope section of the line on which the train is running relative to a logical direction of the line, and a vertical curve radius of the line on which the train is running. The logical direction of the line refers to the direction in which the train advances along the line.

[0060] When each training sample in the training sample set is input to the preset neural network, the unit of the train speed is m / s, the unit of the slope value is ten-thousandth of a degree, the unit of the vertical curve radius is cm, and the unit of the acceleration is m / s 2 . The inclination direction of the slope section of the line on which the train is running relative to the logical direction of the line is a 0-1 value. When the inclination direction of the slope section of the line on which the train is running relative to the logical direction of the line is 0, it indicates that the slope section of the line on which the train is running is uphill. When the inclination direction of the slope section of the line on which the train is running relative to the logical direction of the line is 1, it indicates that the slope section of the line on which the train is running is downhill.

[0061] In addition, to ensure that the CNN-BiLSTM model based on the attention mechanism can sufficiently learn each dimension feature, the slope value needs to be multiplied by 10 -3 , and the vertical curve radius needs to be multiplied by 10-6 .

[0062] The preset neural network is a CNN-BiLSTM neural network model with an attention mechanism. For example... Figure 2 As shown, the preset neural network includes an input layer, a hidden layer, and an output layer connected in sequence; the hidden layer includes a convolutional layer, a pooling layer, a Dropout layer, a BiLSTM network layer, and an attention mechanism layer connected in sequence. Due to control delays, environmental influences, and complex wheel-rail relationships during actual train operation, there is still a significant error between the modeling based on dynamic principles and the real train. However, the CNN-BiLSTM model with an attention mechanism utilizes data-driven modeling, which can reduce errors through continuous learning from historical data, resulting in higher accuracy.

[0063] Training sample F t After the input layer enters the convolutional layer, it undergoes a convolution operation with the convolutional kernel. Then, the pooling layer selects the maximum value within the pooling window to obtain F. t 1 A convolutional neural network structure consisting of convolutional layers and pooling layers is as follows: Figure 3 As shown, m×n is the size of the input vector, y is the number of convolution kernels, x is the size of the convolution kernels, and k is the size of the pooling window. The values ​​of x, y, and k can be adjusted as needed.

[0064] F after processing by convolutional layers and pooling layers t 1 Entering the Dropout layer, we get F t 2 The purpose of this is to prevent the model from overfitting. The principle of the Dropout layer is as follows: Figure 4 As shown. The essence of the Dropout layer is to make each neuron stop working with probability p, i.e., F t 1 Each dimension of the data has a probability p becoming 0, which can reduce the complex co-adaptation relationship between neurons and improve the generalization ability of the model.

[0065] F t 2 Entering BiLSTM to obtain H t .like Figure 5 As shown, the BiLSTM network layer includes multiple LSTM units, each consisting of an input gate, a forget gate, and an output gate. The BiLSTM network layer is used to: 1) process the feature vector x output by the Dropout layer... t As input features, they are fed into the forward and backward networks in ascending and descending order, respectively. 2) Based on the LSTM unit, according to the input features and the hidden state value h of the LSTM unit at time t-1...t-1 and the cell state value C t-1 The hidden state value h at time t is calculated by an activation function t . 3) The hidden state value of the LSTM unit of the forward network corresponding to the same input feature is spliced with the hidden state value of the LSTM unit of the corresponding backward network to obtain an output feature vector.

[0066] H t The output a of the CNN-BiLSTM model with attention mechanism is obtained by entering the attention mechanism layer t As shown in Figure 6 , the attention mechanism layer is used to: 1) normalize the feature vector output by the BiLSTM network layer into a probability distribution vector based on a softmax activation function, and the sum of each probability is 1; 2) multiply and sum the feature vector output by the BiLSTM network layer and the probability distribution vector to obtain an initial predicted acceleration feature. Figure 6 In the formula, x t is input information, q is a query vector for judging the importance of information, s is a scoring function for calculating the similarity between input information and a query vector, and a t is an attention weight. In the present application, an attention mechanism is added, which can increase the generalization ability of the model and give correct prediction results when the input is not in the training set.

[0067] In the training process of the preset neural network, the attention mechanism layer outputs an initial predicted acceleration feature, a preset loss function is used to calculate the loss value between the initial predicted acceleration feature and the actual acceleration of the train in the training sample; based on the loss value, the gradient of the preset loss function with respect to each layer parameter in the hidden layer is calculated by the derivative chain rule through back propagation, and the update of each layer parameter in the hidden layer is performed according to the gradient, so that the loss function is continuously reduced, and finally the train dynamics model M with precision meeting the requirements of the test platform is obtained.

[0068] In one specific embodiment, as shown in Figure 7 , the preset ATO control performance indoor simulation test model further includes an electronic map and a train state fusion model, and Figure 7 the CNN-BiLSTM train dynamics model based on attention mechanism in the formula corresponds to the train dynamics model in the foregoing; and Figure 7 In the formula, the CNN-BiLSTM train dynamics model based on attention mechanism corresponds to the train dynamics model in the foregoing, which is used to calculate the predicted actual acceleration of the current period, and the actual acceleration input from the train state fusion model to the CNN-BiLSTM train dynamics model based on attention mechanism corresponds to the train dynamics model in the foregoing is the actual acceleration of the train in the last period.

[0069] The electronic map records train operation line map data, which is static line data, including line speed limit, line slope size and direction, and line vertical curve radius. The train state fusion model is used to associate the static line data with the train speed and position in the next period output by the train motion model, and then send to the measured ATO to calculate the control command in the next period. At the same time, the associated train speed and position in the next period, the current period acceleration control module output by the ATO control system are input into the test result module for storage and output. Repeat the above steps until the test is completed. After the test is completed, the test platform will package the train state data of this test to generate an Excel document, and calculate the train running time in this test.

[0070] Based on the above, the indoor simulation test method for ATO control performance of the urban rail train further includes: 1) obtaining train operation line map data; the train operation line map data includes slope data and speed limit data corresponding to each operation line; 2) associating the train operation line map data, the train current period predicted actual acceleration and the train speed and position in the next period, and feeding back to the ATO to obtain the ATO output acceleration control analog quantity corresponding to the next period.

[0071] Embodiment two

[0072] As Figure 8 shown, in order to realize the technical solutions in embodiment one and achieve the corresponding functions and technical effects, the embodiment further provides an indoor simulation test system for ATO control performance of an urban rail train, which includes:

[0073] A train data acquisition module 101 is configured to acquire a train data set and slope data of a line on which the train is located in a current period; the train data set includes train speed, ATO output acceleration control analog quantity and actual acceleration of the train in a previous period.

[0074] A train state prediction module 201 is configured to input the train data set and the slope data of the line on which the train is located into a preset ATO control performance indoor simulation test model to obtain train state prediction data; the train state prediction data includes train speed and position in a next period.

[0075] A simulation result output module 301 is configured to associate the train state prediction data and the ATO output acceleration control analog quantity, and mark the generated data as simulation test results.

[0076] The preset ATO control performance indoor simulation test model comprises a train dynamics model and a train motion model; the train dynamics model is configured to determine a train current period predicted actual acceleration based on the train data set and slope data of a line on which the train is running; and the train motion model is configured to calculate a train speed and position in a next period based on the train current period predicted actual acceleration and the line on which the train is running.

[0077] The train dynamics model is obtained by training a preset neural network based on a training sample set; each training sample in the training sample set comprises, in any historical period, a corresponding train speed of a train, an acceleration control analog quantity output by ATO, an actual acceleration of the train in a previous historical period, slope data of a line on which the train is running, and a train actual acceleration; and the preset neural network is a CNN-BiLSTM neural network model based on an attention mechanism.

[0078] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the system disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0079] The principles and implementation manners of the present application are described by using specific examples in the specification, and the above embodiment description is only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range can be changed. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A kind of urban rail train ATO control performance indoor simulation test method, it is characterized in that, The method comprises: acquiring train data set and slope data of a line on which the train runs in a current period; the train data set comprises train speed, acceleration control analog quantity output by ATO and actual acceleration of the train in a previous period; inputting the train data set and the slope data of the line on which the train runs into a preset ATO control performance indoor simulation test model to obtain train state prediction data; the train state prediction data comprises train speed and position in a next period; associating the train state prediction data and the acceleration control analog quantity output by ATO, and marking to generate simulation test results; the preset ATO control performance indoor simulation test model comprises a train dynamics model and a train motion model; the train dynamics model is used to determine train current period prediction actual acceleration based on the train data set and the slope data of the line on which the train runs; and the train motion model is used to calculate train speed and position in a next period based on the train current period prediction actual acceleration and the line on which the train runs; the train dynamics model is obtained by training a preset neural network based on a training sample set; each training sample in the training sample set comprises train corresponding train speed, acceleration control analog quantity output by ATO, actual acceleration of the train in a previous historical period, slope data of the line on which the train runs and train actual acceleration in any historical period; and the preset neural network is a CNN-BiLSTM neural network model based on an attention mechanism.

2. The urban train ATO control performance indoor simulation test method according to claim 1, characterized in that, the preset neural network comprises an input layer, a hidden layer and an output layer connected in sequence; the hidden layer comprises a convolution layer, a pooling layer, a Dropout layer, a BiLSTM network layer and an attention mechanism layer connected in sequence; in the training process of the preset neural network, the attention mechanism layer outputs initial prediction acceleration features, a preset loss function is used to calculate a loss value between the initial prediction acceleration features and train actual acceleration in the training sample; based on the loss value, the gradient of the preset loss function to each layer parameter in the hidden layer is calculated by using back propagation through the derivative chain rule, and each layer parameter in the hidden layer is updated according to the gradient.

3. The urban train ATO control performance indoor simulation test method according to claim 2, characterized in that, the BiLSTM network layer comprises a plurality of LSTM units, and is used to: input the feature vector output by the Dropout layer as input features into a forward network and a backward network in a forward order and a reverse order respectively; based on the LSTM unit, the hidden state value and the cell state value of the LSTM unit at t-1 time are calculated through an activation function to obtain the hidden state value at t time according to the input features; the hidden state value of the LSTM unit of the forward network corresponding to the same input features is spliced with the hidden state value of the LSTM unit of the corresponding backward network to obtain an output feature vector.

4. The urban train ATO control performance indoor simulation test method according to claim 2, characterized in that, the attention mechanism layer is used to: normalize the feature vector output by the BiLSTM network layer into a probability distribution vector based on a softmax activation function, and the sum of each probability is 1. The feature vector output by the BiLSTM network layer is multiplied with a probability distribution vector and summed to obtain an initial predicted acceleration feature.

5. The urban train ATO control performance indoor simulation test method according to claim 1, characterized in that, The slope data of the line on which the train is located includes a slope value of the line on which the train is located, a tilt direction of the slope section of the line on which the train is located relative to a logical direction of the line, and a vertical curve radius of the line on which the train is located.

6. The urban train ATO control performance indoor simulation test method according to claim 5, characterized in that, In the slope data of the line on which the train is located, the tilt direction of the slope section of the line on which the train is located relative to the logical direction of the line is a 0-1 value. Wherein, when the tilt direction of the slope section of the line on which the train is located relative to the logical direction of the line is 0, it indicates that the slope section of the line on which the train is located is an uphill; when the tilt direction of the slope section of the line on which the train is located relative to the logical direction of the line is 1, it indicates that the slope section of the line on which the train is located is a downhill.

7. The urban train ATO control performance indoor simulation test method according to claim 1, characterized in that, The method further comprises: Obtaining train operation line map data; the train operation line map data includes slope data and speed limit data corresponding to each operation line; Correlating the train operation line map data, the train current period predicted actual acceleration, and the train speed and position in the next period, and feeding back to the ATO to obtain the ATO output acceleration control analog quantity corresponding to the next period.

8. A kind indoor simulation test system of ATO control performance of city rail train, it is characterized in that, The system comprises: A train data acquisition module for acquiring train data set and slope data of the line on which the train is located in the current period; the train data set includes train speed, ATO output acceleration control analog quantity, and train actual acceleration in the last period; A train state prediction module for inputting the train data set and the slope data of the line on which the train is located into a preset ATO control performance indoor simulation test model to obtain train state prediction data; the train state prediction data includes train speed and position in the next period; A simulation result output module for correlating the train state prediction data and the ATO output acceleration control analog quantity, and marking the generated simulation test result; The preset ATO control performance indoor simulation test model includes a train dynamics model and a train motion model; the train dynamics model is used to determine the train current period predicted actual acceleration based on the train data set and the slope data of the line on which the train is located; the train motion model is used to calculate the train speed and position in the next period based on the train current period predicted actual acceleration and the line on which the train is located; The train dynamics model is obtained by training a preset neural network based on a training sample set; each training sample in the training sample set includes train corresponding train speed, ATO output acceleration control analog quantity, train actual acceleration in the last historical period, slope data of the line on which the train is located, and train actual acceleration in any historical period; the preset neural network is a CNN-BiLSTM neural network model based on attention mechanism.

Images