A method and system for protecting the safe operation of a virtual consist train

Abstract

The application discloses a kind of virtual marshalling train's tracking operation safety protection method and system, it is related to urban rail transit management technical field.The application includes: obtaining the historical data feature set of leading train and the actual acceleration of leading train in the historical operation process of virtual marshalling train.The historical data feature set is input into acceleration prediction model, and the predicted acceleration of leading train is obtained.The error of the predicted acceleration of leading train and the actual acceleration of leading train is minimized as the goal, and the acceleration prediction model is trained, and the trained acceleration prediction model is obtained.Real-time data feature set is input into the trained acceleration prediction model, and the predicted acceleration and predicted speed of leading train at next time step are obtained, and then the safety protection distance between following train and leading train at next time step is obtained.The application can significantly improve the accuracy of safety protection distance, and guarantee the safe and efficient operation of virtual marshalling train.

Description

A method and system for tracking and protecting the safety of virtual train formations. Technical Field

[0001] This invention relates to the field of urban rail transit management technology, and in particular to a method and system for tracking and operating safety protection of virtual train formations. Background Technology

[0002] With the development of urban rail transit, residents' travel demand continues to rise, posing a severe challenge to rail transit operation organization, especially during morning and evening rush hours when large passenger flows accumulate at stations, leading to continuous train overload. In recent years, virtual train formation operation has been considered an effective means to solve or alleviate capacity shortages on certain sections of lines during peak periods. Virtual train formation, based on wireless communication information interaction and automatic control technologies, combines multiple trains into a large formation, which can be disassembled and restored to multiple independently operating trains when needed, meeting the complex and ever-changing needs of urban rail transit operation. As an emerging technology, virtual train formation requires appropriate tracking and safety protection methods to ensure its safe and efficient operation; this is a necessary condition for its widespread application.

[0003] In urban rail transit moving block systems, existing train operation safety protection mainly employs two methods: the absolute braking distance method and the relative braking distance method. During train operation, the distance between trains must be greater than the calculated safety protection distance. In the absolute braking distance method, the safety protection distance is calculated assuming the leader train's braking distance is 0, with following trains operating under the most unfavorable braking conditions. In the relative braking distance method, the leader train's braking distance is subtracted when calculating the safety protection distance, rather than assuming it to be 0.

[0004] The absolute braking distance method assumes the preceding train's braking distance is zero, meaning the following train calculates its safety protection distance based solely on its own emergency braking capability. However, in virtual train formations, the lead train and following trains are dynamically coupled through wireless communication and coordinated control. The actual braking distance of the lead train cannot be zero, leading to an underestimation of the calculated safety protection distance and increasing the risk of rear-end collisions. The relative braking distance method, under complex conditions such as slopes and curves, requires train deceleration, affecting the braking distance and resulting in inaccurate safety protection distance calculations. Therefore, existing train operation safety measures cannot accurately determine the safety protection distance between the lead train and following trains, leading to increased train intervals and reduced track capacity and operational efficiency. Summary of the Invention

[0005] Therefore, it is necessary to provide a method and system for ensuring the safe operation of virtual train formations in response to the aforementioned technical problems.

[0006] This invention provides a method for ensuring the safe operation of virtual train formations during tracking, comprising:

[0007] Acquire the historical data feature set of the lead train and the actual acceleration of the lead train during the historical operation of the virtual train formation;

[0008] Historical data feature sets are input into the acceleration prediction model, which consists of a temporal convolutional network, a long short-term memory network, and a multi-head self-attention mechanism connected in sequence. The temporal convolutional network captures the long-distance temporal dependencies of the historical data feature set to obtain temporal context features. The long short-term memory network models the temporal dynamic changes of the temporal context features to generate temporal dynamic features. The multi-head self-attention mechanism weights the temporal dynamic features to obtain attention-weighted features. The predicted acceleration of the leader train is obtained based on the attention-weighted features. The acceleration prediction model is trained with the goal of minimizing the error between the predicted acceleration of the leader train and the actual acceleration of the leader train.

[0009] The system acquires the real-time data feature set of the leader train at the current time step during virtual train formation tracking. It inputs the real-time data feature set into the trained acceleration prediction model to obtain the predicted acceleration of the leader train at the next time step, and calculates the predicted speed of the leader train at the next time step based on the predicted acceleration. Based on the predicted acceleration and predicted speed of the leader train at the next time step, the system obtains the safe protection distance between the following train and the leader train at the next time step.

[0010] Optionally, the data feature set includes acceleration state data and track condition data;

[0011] Acceleration status data includes: train speed at the previous time step, train acceleration at the previous time step, and train traction and braking level at the previous time step;

[0012] The route condition data includes: the route gradient and the radius of the vertical curve.

[0013] Optionally, based on the predicted acceleration and predicted velocity of the leading train in the next time step, the safe protection distance between the following train and the leading train in the next time step is obtained, specifically including:

[0014] The formula for calculating the safe protection distance is determined by the following formula:

[0015] ;

[0016] in, For safe protection distance; To meet the operational space requirements of the train in the next time step; To follow the train in time speed; The time step length; The desired interval between adjacent trains when they stop at the platform; To follow the train in time The dynamic distance margin term;

[0017] By inputting the predicted acceleration and predicted speed of the leading train at the next time step into the safety protection distance calculation formula, the safety protection distance between the following train and the leading train at the next time step is obtained.

[0018] Optionally, the dynamic distance margin term is determined, specifically including:

[0019] The potential proximity margin term is determined based on the following formula:

[0020] ;

[0021] ;

[0022] ;

[0023] in, The distance traveled to adjust the speed of the train at maximum acceleration in the next time step; The distance traveled by which the train adjusts its speed in the next time step based on the predicted acceleration of the train in the next time step; To predict the acceleration of the train in the next time step;

[0024] The communication delay margin term is determined based on the following formula:

[0025] ;

[0026] in, To follow the train in time The communication delay time, its time-varying range is ; This is the maximum communication delay time for the train; To guide the train's predicted speed at the next time step;

[0027] The control delay margin term is determined based on the following formula:

[0028] ;

[0029] in, To follow the train in time The control delay time, whose time-varying range is... ; To lead the train in time The control delay time, whose time-varying range is... ; This is the maximum control delay time for the train;

[0030] The dynamic range margin term is determined based on the following formula, using the potential approach range margin term, communication delay margin term, and control delay margin term:

[0031] ;

[0032] in, For potential proximity margin terms, For communication delay margin, To control the delay margin term.

[0033] Optionally, the acceleration prediction model also includes an input layer, which normalizes the data feature set to obtain preprocessed features, specifically including:

[0034] Outliers in the data feature set are removed to obtain the preprocessed data feature set;

[0035] The preprocessed data feature set is globally normalized to scale the data range of the preprocessed data feature set to [-1, 1], eliminate the dimensional differences between different features, and obtain the preprocessed features.

[0036] Optionally, a multi-head self-attention mechanism is used to fuse the weights of variables affecting train acceleration in the temporal dynamic features to obtain attention-weighted features, specifically including:

[0037] Temporal dynamic features are reconstructed into vector form through a multi-head self-attention mechanism;

[0038] The query, key, and value matrices are generated in vector form through linear transformation, and the attention weights of the query, key, and value matrices are determined respectively.

[0039] The attention weights of the query, key, and value matrices are weighted and fused to obtain attention-weighted features.

[0040] Optionally, the acceleration prediction model also includes an output layer, which obtains the predicted acceleration of the leading train based on attention-weighted features, specifically including:

[0041] By inversely normalizing the attention-weighted features through the output layer, the attention-weighted features are restored to acceleration values, thus obtaining the predicted acceleration of the leader train.

[0042] This invention also provides a tracking and safety protection system for virtual train formations, comprising:

[0043] The data acquisition module is used to acquire the historical data feature set of the leading train and the actual acceleration of the leading train during the historical operation of the virtual train formation.

[0044] The data processing module is used to input historical data feature sets into the acceleration prediction model. The acceleration prediction model includes a temporal convolutional network, a long short-term memory network, and a multi-head self-attention mechanism connected in sequence. The temporal convolutional network captures the long-distance temporal dependencies of the historical data feature set to obtain temporal context features. The long short-term memory network models the temporal dynamic changes of the temporal context features to generate temporal dynamic features. The multi-head self-attention mechanism weights the temporal dynamic features to obtain attention-weighted features. The predicted acceleration of the leader train is obtained based on the attention-weighted features. The acceleration prediction model is trained with the goal of minimizing the error between the predicted acceleration of the leader train and the actual acceleration of the leader train to obtain the trained acceleration prediction model.

[0045] The safety protection distance output module is used to acquire the real-time data feature set of the leader train at the current time step during virtual train formation tracking. The real-time data feature set is input into the trained acceleration prediction model to obtain the predicted acceleration of the leader train at the next time step, and the predicted speed of the leader train at the next time step is calculated based on the predicted acceleration. Based on the predicted acceleration and predicted speed of the leader train at the next time step, the safety protection distance between the following train and the leader train at the next time step is obtained.

[0046] The tracking and operation safety protection method and system for virtual train formations provided in this invention have the following advantages compared with the prior art:

[0047] The acceleration prediction model proposed in this invention captures long-distance temporal dependencies through temporal convolutional networks, models temporal dynamic changes using long short-term memory networks, and strengthens the weight fusion of key variables through a multi-head self-attention mechanism. It comprehensively captures long-term dependencies, dynamic changes, and the impact of key variables in the time series, achieving high-precision prediction of the acceleration of the lead train. Based on the acceleration prediction results of the lead train, the safe protection distance is determined, which can dynamically adapt to complex working conditions and changes in operating status. It overcomes the calculation deviations caused by traditional methods that assume the braking distance of the preceding train is 0 or do not fully consider complex working conditions, significantly improving the accuracy of the safe protection distance and ensuring the safe and efficient operation of virtual train formations. Attached Figure Description

[0048] Figure 1 is a flowchart illustrating a method for tracking and ensuring the safety of virtual train formations in one embodiment.

[0049] Figure 2 is a hybrid TLMA model diagram of a virtual train formation tracking operation safety protection method provided in one embodiment;

[0050] Figure 3 is a train acceleration curve of a virtual train tracking operation safety protection method provided in one embodiment;

[0051] Figure 4 is a train speed curve of a tracking operation safety protection method for virtual train formations provided in one embodiment;

[0052] Figure 5 is a schematic diagram of the train trajectory with a potential approach distance margin in a tracking operation safety protection method for a virtual train formation provided in one embodiment;

[0053] Figure 6 is a comparative diagram of a method for tracking and protecting the safety of virtual train formations provided in one embodiment. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] First Embodiment

[0056] This invention provides a method for ensuring safe tracking operation of virtual train formations. Figure 1 shows a schematic flowchart of the method. As shown in Figure 1, the method includes the following steps:

[0057] Acquire the historical data feature set of the lead train and the actual acceleration of the lead train during the historical operation of the virtual train formation;

[0058] Historical data feature sets are input into the acceleration prediction model, which consists of a temporal convolutional network, a long short-term memory network, and a multi-head self-attention mechanism connected in sequence. The temporal convolutional network captures the long-distance temporal dependencies of the historical data feature set to obtain temporal context features. The long short-term memory network models the temporal dynamic changes of the temporal context features to generate temporal dynamic features. The multi-head self-attention mechanism weights the temporal dynamic features to obtain attention-weighted features. The predicted acceleration of the leader train is obtained based on the attention-weighted features. The acceleration prediction model is trained with the goal of minimizing the error between the predicted acceleration of the leader train and the actual acceleration of the leader train.

[0059] The system acquires the real-time data feature set of the leader train at the current time step during virtual train formation tracking. It inputs the real-time data feature set into the trained acceleration prediction model to obtain the predicted acceleration of the leader train at the next time step, and calculates the predicted speed of the leader train at the next time step based on the predicted acceleration. Based on the predicted acceleration and predicted speed of the leader train at the next time step, the system obtains the safe protection distance between the following train and the leader train at the next time step.

[0060] The acceleration prediction model also includes an input layer, which normalizes the data feature set to obtain preprocessed features. Specifically, this includes: removing outliers from the data feature set to obtain a preprocessed data feature set; and then performing global normalization on the preprocessed data feature set to scale its data range to [-1, 1], eliminating dimensional differences between different features to obtain the preprocessed features.

[0061] The acceleration prediction model also includes an output layer, which performs inverse normalization on the attention-weighted features to restore the attention-weighted features to the actual values, thereby obtaining the predicted acceleration of the leading train.

[0062] The implementation process includes:

[0063] Step S1: Obtain the real-time data feature set for real-time prediction of train acceleration.

[0064] Step S2: Construct an acceleration prediction model (TLMA) based on Temporal Convolutional Network (TCN), Long Short-Term Memory (LSTM), and Multi-Head Attention (MATT).

[0065] Before using TLMA for prediction, model parameter training is required. The collected historical data feature set is standardized and input into the TLMA. During training, 30 seconds of historical data are used to predict the acceleration for the next second. The Adam optimizer is used, and the training cycle is 300 rounds. The trained TLMA model is then deployed on the following train. Real-time data feature sets are collected in real-time via input sensors and wireless communication technology to obtain the predicted acceleration and velocity of the lead train in the next time step.

[0066] Step S3 proposes a formula for calculating the safe protection distance, which takes into account the operating space requirements of the following train in the next time step, the safety requirements when the train stops, and the safety requirements for following train operation under adverse conditions.

[0067] Step S4: Substitute the predicted acceleration and predicted speed of the leader train into the safety protection distance calculation formula to calculate the safety protection distance between the following train and the leader train in the next time step.

[0068] Furthermore, in step S1, the acquired train running ATO time is subjected to basic analysis and screening to remove outliers. Based on the characteristics of the dataset required for train acceleration prediction, the acceleration state data variable in the data feature set of the prediction model is determined to be the train speed of the previous time step. The previous time step train acceleration The previous time step of the train traction and braking level .

[0069] During the operation of a virtual train formation, in addition to the train's own state, it is also affected by track conditions. Based on the train dynamics model, the track condition data variable selected from the data feature set of the prediction model is the track gradient value. Vertical curve radius of the line .

[0070] The values ​​of some variables input into the prediction model are shown in Table 1.

[0071] Table 1 Input dataset for the prediction model

[0072]

[0073] Furthermore, in step S2, an acceleration prediction model TLMA is proposed, which combines a temporal convolutional network, a long short-term memory network, and a multi-head self-attention mechanism. The main advantage of this model is that it can overcome the limitations of traditional prediction models in handling long-term dependencies and capturing nonlinear relationships, while enhancing the model's focus on key information.

[0074] Figure 2 shows the structure of TLMA. In the input layer, outliers in the data feature set are removed to obtain a preprocessed data feature set. This preprocessed data feature set is then globally normalized to scale its range to [-1, 1], eliminating dimensional differences between features and obtaining preprocessed features. In TCN, the time window is expanded by adjusting the dilation factor in the dilated convolution to increase the receptive range of the convolution kernel, thereby capturing long-distance temporal dependencies. Causal convolution is also used to avoid the leakage of future information, combined with residual connections to prevent gradient vanishing. Then, the high-order features output by TCN are input into LSTM. LSTM uses gating mechanisms and cell state design to model the temporal dynamic changes of the temporal context features, obtaining temporal dynamic features, which can more effectively handle temporal dynamic data, making the model more effective when processing long sequence data. Furthermore, MATT reconstructs the temporal dynamic features into vector form. Query, key, and value matrices are generated through linear transformation. Attention weights are calculated for the query, key, and value matrices respectively, and the results are weighted and fused to obtain attention-weighted features. This reflects the relative importance of relevant variables and optimizes the model's prediction results. Finally, the attention-weighted features are denormalized in the output layer to restore them to their actual values, thus obtaining the predicted leader train acceleration.

[0075] To evaluate the effectiveness of the proposed TLMA, four evaluation metrics—mean squared error, root mean square error, mean absolute error, and coefficient of determination—were used to compare the prediction results. Table 2 shows the comparison values ​​with existing prediction models.

[0076] Table 2 Comparison of Prediction Models

[0077]

[0078] The proposed TLMA has a mean square error of 0.0404, a root mean square error of 0.0053, a mean absolute error of 0.0730, and a coefficient of determination of 0.9653. These results demonstrate that the TLMA-predicted acceleration values ​​are in good agreement with the actual acceleration values. Therefore, TLMA is an effective and accurate short-term prediction model for train acceleration.

[0079] Figures 3 and 4 show the acceleration curve of the lead train predicted using TLMA and the speed curve of the lead train calculated further, respectively. The figures show that the predicted acceleration is close to the actual acceleration, and the predicted speed calculated from the predicted acceleration matches the actual speed very well. The predicted acceleration and speed results will be used for the safety protection distance calculation in step S4.

[0080] Furthermore, in step S3, a train tracking safety protection method based on acceleration prediction is proposed. Existing safety protection methods require the distance between trains to be greater than the braking distance difference between adjacent trains. However, when the real-time operating status of trains can be predicted, ensuring that the distance between trains is greater than the safety protection distance of the next time step is sufficient to ensure safe train operation, and it is not necessary to calculate the braking distance difference.

[0081] The safety protection method proposed in this invention mainly comprises three parts. The first part is the interval distance between trains running at the current speed in the next time step, which basically meets the operating space requirements of the train in the next time step. The second part is the expected stopping interval distance between adjacent trains within the platform, to ensure safety requirements when the train stops. The third part is a dynamic distance margin term, which ensures the safety requirements of train operation under unfavorable conditions. The calculation formula for the train safety protection method is:

[0082] (1)

[0083] in, The safe protection distance calculated for the proposed operational safety protection method; To meet the operational space requirements of the train in the next time step; To follow the train in time speed; The time step length; The desired interval between adjacent trains when they stop at the platform is used to ensure the safety requirements when trains stop at the station; To follow the train in time The dynamic distance margin term is used to ensure the safety requirements of following train operation under adverse conditions.

[0084] The dynamic distance margin term consists of three parts: potential approach distance margin, communication delay margin, and control delay margin. The potential approach distance margin refers to the reduced interval between the following train and the lead train when the following train accelerates to its maximum acceleration in the next time step. This margin term mitigates the impact of extreme acceleration by the following train. Furthermore, wireless communication and automatic driving, as key technologies for virtual train formations, may experience communication and control delays due to different train operating environments. These delays can cause fluctuations in the interval between trains. Therefore, the impact of communication and control delays is considered when calculating the safety protection distance, and corresponding margin terms are added. The calculation method for the dynamic distance margin term is as follows:

[0085] (2)

[0086] in, This is a potential proximity margin term; This is the communication delay margin term; To control the delay margin term.

[0087] Figure 5 shows a schematic diagram of the train trajectory for the potential approach distance margin term. In the next time step, if the distance traveled by the following train accelerating at maximum acceleration is still less than the predicted distance traveled by the leading train, then the potential approach distance margin is 0. Otherwise, the extreme acceleration behavior of the following train will shorten the interval between trains. Therefore, the potential approach distance needs to be compensated. The calculation method for the potential approach distance margin term is as follows:

[0088] (3)

[0089] (4)

[0090] (5)

[0091] in, To follow the train at maximum acceleration in the next time step The distance traveled to adjust the speed; The distance traveled by which the train adjusts its speed in the next time step based on the predicted acceleration of the train in the next time step; To predict the acceleration of the train in the next time step.

[0092] If the speed of the lead train is greater than the speed of the following train, the communication delay will not affect the safety guard distance. However, if the speed of the lead train is less than the speed of the following train, the communication delay will shorten the interval between trains. In this case, an additional distance margin is needed to compensate for the extra distance traveled by the following train during the communication delay. The communication delay margin term is calculated as follows:

[0093] (6)

[0094] in, To follow the train in time The communication delay time, its time-varying range is ; This is the maximum communication delay time for the train; To guide the train's predicted speed at the next time step.

[0095] Control delay refers to the time it takes for a train to calculate and execute the appropriate instructions required for its operation. If the control delay of the lead train is longer than that of the following train, the control delay does not affect the safe distance between trains. However, if the control delay of the following train is longer than that of the lead train, the interval between trains will be shortened within the control delay time, requiring an increase in distance margin. The calculation method for the control delay margin is as follows:

[0096] (7)

[0097] in, To follow the train in time The control delay time, whose time-varying range is... ; To lead the train in time The control delay time, whose time-varying range is... ; This is the maximum control delay time for the train.

[0098] Furthermore, in step S4, the acceleration and speed results of the leader train predicted by TLMA are substituted into the safety protection distance calculation formula to obtain the safety protection distance between the following train and the leader train in the next time step.

[0099] Figure 6 shows a comparison of the safety protection distance calculation results of the proposed safety protection method with existing absolute braking distance methods and relative braking distance methods. As shown in the figure, under the absolute braking distance method, the average safety protection distance between the following train and the lead train is 110.68 m, and the maximum safety protection distance is 325.22 m. Under the relative braking distance method, the average safety protection distance between the following train and the lead train is 26.87 m, and the maximum safety protection distance is 78.63 m. Under the acceleration prediction-based safety protection method proposed in this invention, the average safety protection distance between the following train and the lead train is 17.02 m, and the maximum safety protection distance is 36.53 m. Under the condition of meeting safe operation requirements, the average safety protection distance calculated by the proposed safety protection method is reduced by 84.6% and 36.7% compared to the absolute braking distance method and the relative braking distance method, respectively.

[0100] Second Embodiment

[0101] This embodiment provides a tracking and operation safety protection system for virtual train formations. The system includes: a data acquisition module, a data processing module, and a safety protection distance output module.

[0102] The data acquisition module is used to acquire the historical data feature set of the leading train and the actual acceleration of the leading train during the historical operation of the virtual train formation.

[0103] The data processing module is used to input historical data feature sets into the acceleration prediction model. The acceleration prediction model includes a temporal convolutional network, a long short-term memory network, and a multi-head self-attention mechanism connected in sequence. The temporal convolutional network captures the long-distance temporal dependencies of the historical data feature set to obtain temporal context features. The long short-term memory network models the temporal dynamic changes of the temporal context features to generate temporal dynamic features. The multi-head self-attention mechanism weights the temporal dynamic features to obtain attention-weighted features. The predicted acceleration of the leader train is obtained based on the attention-weighted features. The acceleration prediction model is trained with the goal of minimizing the error between the predicted acceleration of the leader train and the actual acceleration of the leader train to obtain the trained acceleration prediction model.

[0104] The safety protection distance output module is used to acquire the real-time data feature set of the leader train at the current time step during virtual train formation tracking. The real-time data feature set is input into the trained acceleration prediction model to obtain the predicted acceleration of the leader train at the next time step, and the predicted speed of the leader train at the next time step is calculated based on the predicted acceleration. Based on the predicted acceleration and predicted speed of the leader train at the next time step, the safety protection distance between the following train and the leader train at the next time step is obtained.

[0105] The aforementioned virtual train tracking and operation safety protection system is used to implement the virtual train tracking and operation safety protection method described in the first embodiment. The method and the method part have a one-to-one correspondence. The above description and limitations of the method also apply to the virtual train tracking and operation safety protection system of this embodiment, and will not be repeated here.

[0106] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for ensuring safe operation of virtual train formations during tracking, characterized in that, include: The process involves acquiring historical data feature sets and actual acceleration of the leading train during the historical operation of a virtual train formation; inputting the historical data feature set into an acceleration prediction model, which includes a temporal convolutional network, a long short-term memory network, and a multi-head self-attention mechanism connected in sequence; capturing the long-distance temporal dependencies of the historical data feature set through the temporal convolutional network to obtain temporal context features; and modeling the temporal dynamic changes of the temporal context features through the long short-term memory network to generate temporal dynamic features. The temporal dynamic features are weighted using a multi-head self-attention mechanism to obtain attention-weighted features; the predicted acceleration of the leader train is obtained based on the attention-weighted features. To minimize the error between the predicted acceleration and the actual acceleration of the lead train, an acceleration prediction model is trained to obtain the trained model. Real-time data feature sets of the lead train at the current time step during virtual train tracking are acquired and input into the trained acceleration prediction model to obtain the predicted acceleration of the lead train at the next time step. The predicted velocity of the lead train at the next time step is then calculated based on the predicted acceleration. The safe protection distance between the following train and the lead train at the next time step is obtained based on the predicted acceleration and velocity of the lead train at the next time step. Specifically, obtaining the safe protection distance between the following train and the lead train at the next time step involves determining the calculation formula for the safe protection distance using the following formula: ;in, For safe protection distance; To meet the operational space requirements of the train in the next time step; To follow the train in time speed; The time step length; The desired interval between adjacent trains when they stop at the platform; To follow the train in time The dynamic distance margin term; the predicted acceleration and predicted speed of the leading train in the next time step are input into the safety protection distance calculation formula to obtain the safety protection distance between the following train and the leading train in the next time step.

2. The method for tracking and ensuring safe operation of a virtual train formation as described in claim 1, characterized in that, The data feature set includes acceleration state data and line condition data; The acceleration status data includes: train speed at the previous time step, train acceleration at the previous time step, and train traction and braking level at the previous time step; the track condition data includes: track gradient and track vertical curve radius.

3. The method for tracking and ensuring safe operation of a virtual train formation as described in claim 1, characterized in that, Determining the dynamic distance margin term specifically includes: determining the potential approach distance margin term based on the following formula: ； ； ;in, The distance traveled to adjust the speed of the train at maximum acceleration in the next time step; The distance traveled by which the train adjusts its speed in the next time step based on the predicted acceleration of the train in the next time step; To predict the train's acceleration at the next time step, the communication delay margin term is determined based on the following formula: ;in, To follow the train in time The communication delay time, its time-varying range is ； This is the maximum communication delay time for the train; To guide the predicted speed of the train at the next time step, the control delay margin term is determined based on the following formula: ;in, To follow the train in time The control delay time, whose time-varying range is... ； To lead the train in time The control delay time, whose time-varying range is... ； The maximum control delay time for the train; the dynamic distance margin term is determined based on the following formula using the potential approach distance margin term, communication delay margin term, and control delay margin term: ;in, For potential proximity margin terms, For communication delay margin, To control the delay margin term.

4. The method for tracking and ensuring safe operation of a virtual train formation as described in claim 1, characterized in that, The acceleration prediction model also includes an input layer, which normalizes the data feature set to obtain preprocessed features. Specifically, this includes: removing outliers from the data feature set to obtain a preprocessed data feature set; and performing global normalization on the preprocessed data feature set to scale the data range of the preprocessed data feature set to [-1, 1] to eliminate the dimensional differences between different features, thus obtaining the preprocessed features.

5. The method for tracking and ensuring safe operation of a virtual train formation as described in claim 1, characterized in that, The step of using a multi-head self-attention mechanism to perform weighted fusion of variables affecting train acceleration in temporal dynamic features to obtain attention-weighted features specifically includes: reconstructing the temporal dynamic features into vector form using a multi-head self-attention mechanism; generating query, key, and value matrices in vector form through linear transformation, and determining the attention weights of the query, key, and value matrices respectively; and weighted fusion of the attention weights of the query, key, and value matrices to obtain attention-weighted features.

6. The method for tracking and ensuring safe operation of a virtual train formation as described in claim 1, characterized in that, The acceleration prediction model also includes an output layer. The step of obtaining the predicted acceleration of the leader train based on the attention-weighted features specifically includes: performing inverse normalization on the attention-weighted features through the output layer to restore the attention-weighted features to acceleration values, thereby obtaining the predicted acceleration of the leader train.

7. A tracking and operation safety protection system for virtual train formations, characterized in that, include: The data acquisition module is used to acquire the historical data feature set of the leading train and the actual acceleration of the leading train during the historical operation of the virtual train formation. The data processing module is used to input the historical data feature set into the acceleration prediction model, which includes a temporal convolutional network, a long short-term memory network, and a multi-head self-attention mechanism connected in sequence; the temporal convolutional network captures the long-distance temporal dependencies of the historical data feature set to obtain temporal context features; the long short-term memory network models the temporal dynamic changes of the temporal context features to generate temporal dynamic features. The temporal dynamic features are weighted using a multi-head self-attention mechanism to obtain attention-weighted features; the predicted acceleration of the leader train is obtained based on the attention-weighted features. With the goal of minimizing the error between the predicted acceleration and the actual acceleration of the lead train, an acceleration prediction model is trained to obtain the trained acceleration prediction model. A safety protection distance output module is used to acquire the real-time data feature set of the lead train at the current time step during virtual train tracking. This real-time data feature set is input into the trained acceleration prediction model to obtain the predicted acceleration of the lead train at the next time step, and the predicted velocity of the lead train at the next time step is calculated based on the predicted acceleration. Based on the predicted acceleration and velocity of the lead train at the next time step, the safety protection distance between the following train and the lead train at the next time step is obtained. Specifically, obtaining the safety protection distance between the following train and the lead train at the next time step, based on the predicted acceleration and velocity of the lead train at the next time step, includes determining the safety protection distance calculation formula using the following formula: ;in, For safe protection distance; To meet the operational space requirements of the train in the next time step; To follow the train in time speed; The time step length; The desired interval between adjacent trains when they stop at the platform; To follow the train in time The dynamic distance margin term; the predicted acceleration and predicted speed of the leading train in the next time step are input into the safety protection distance calculation formula to obtain the safety protection distance between the following train and the leading train in the next time step.

Images