Train behavior planning method suitable for train-train communication train control system

Abstract

The invention discloses a train behavior planning method suitable for a train-train communication train control system, and relates to the technical field of high-speed railway train operation control. The output of a neural network trajectory prediction model is changed into a base trajectory of a train operation trajectory from a train operation state value within a certain time; a base trajectory set is constructed based on historical train operation data of the same line, the neural network trajectory prediction model is matched in the base trajectory set, a base trajectory corresponding to a preceding train is output, the operation state of the preceding train in a period of time in the future is obtained, the prediction dimension of the neural network trajectory prediction model is reduced, and the prediction efficiency is improved. Complexity of calculation of the neural network trajectory prediction model is effectively reduced, trajectory prediction efficiency is improved, the neural network trajectory prediction model is adopted for prediction control, the distance between the rear train and the front train is shortened, meanwhile, the running dynamic state of the front train is tracked, and safe train following at shorter intervals is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of high-speed railway train operation control, and particularly relates to a train behavior planning method suitable for a vehicle-to-vehicle communication train control system. Background Art

[0002] A train operation control system refers to a control system composed of ground equipment and on-vehicle equipment, which is used to control the running speed of trains, ensure the safe and efficient operation of trains, and is one of the important components of the railway signal control system. The train operation control system is a rail transit signal system that develops with the development of train technology and the train-ground information transmission system, integrating advanced control technology, communication technology, computer technology and railway signal technology, and controlling the running direction, running interval and running speed of trains, which is the core to ensure train operation safety and improve transportation efficiency.

[0003] The high-speed railway train control system based on vehicle-to-vehicle communication (i.e., the train control system based on vehicle-to-vehicle communication) takes the on-vehicle mobile body as the core to realize train operation control, breaks through the traditional train control mode based on routes for high-speed railways, and through two-way, large-capacity, high-speed vehicle-ground wireless communication, multi-sensor fusion train speed measurement and positioning, and train autonomous cooperative moving block control technology, relies on its own perception and autonomous decision-making to achieve intelligent operation control. Compared with the China Train Control System (CTCS) with more trackside equipment and complex interfaces and the Communication-based Train Control (CBTC), the train control system based on vehicle-to-vehicle communication changes the communication data flow from "vehicle-ground-vehicle" to "vehicle-vehicle", streamlines the system equipment, has richer information interaction, including position, speed, movement authorization, etc., considers the movement trend of trains, and adopts the method of "hitting a soft wall" for safety protection, shortening the tracking interval.

[0004] The on-vehicle equipment of the high-speed railway train control system based on vehicle-to-vehicle communication undertakes functions such as mobile authorization calculation and information interaction of the Radio Block Center (RBC), and also has functions such as interlocking, target-distance mode curve calculation, automatic driving and train integrity check. Compared with the traditional RBC, the on-vehicle equipment of the high-speed railway train control system based on vehicle-to-vehicle communication has obvious similarities and differences. Both mainly complete two-way communication and train operation permit calculation functions in terms of functions, but from the perspective of the design concept of the whole system, the specific differences are reflected in:

[0005] Structural differences: RBCs are arranged at fixed physical locations, and the train needs to switch when passing through the ranges of two RBCs; the RBC generates control commands sent to the train by exchanging information with external ground equipment and on-vehicle equipment through vehicle-ground two-way communication. The main function is to provide a movement authority to enable the train to operate safely on the lines within the jurisdiction of the RBC, and to complete train spacing control and train protection. The train control system based on vehicle-to-vehicle communication realizes two-way wireless communication functions with on-vehicle equipment of the preceding / following trains, the ground system data server, trackside controllers, and train end equipment through wireless communication units and GPRS radio stations. Moreover, the on-vehicle main control unit has the function of calculating the movement authority based on information such as station routes and section directions, electronic maps, the position and speed of the train itself, and the position and speed of the preceding train.

[0006] Train following operation differences: Traditional high-speed railway train control systems generally operate using the absolute braking moving block system. The ground radio block center generates a movement authority based on the train operation conditions and line occupancy within the block section and sends it to the following train. The end point of the following train's tracking is the position of the tail of the preceding train plus a certain safety protection distance. In the train control system based on vehicle-to-vehicle communication, the following train is controlled according to the dynamics of the preceding train, operates in a relative braking mode, predicts the future trajectory of the preceding train based on its operation dynamics, and the following train controls its operation state to be close to that of the preceding train, and maintains a certain distance from the preceding train, so that the following train can use normal braking to stop at a safe distance behind the tail of the preceding train after the preceding train makes an emergency brake, thereby shortening the train following distance and increasing the train operation density.

[0007] Therefore, for the on-vehicle equipment of the communication-based high-speed railway train control system, considering the above differences, designing a train following behavior planning method that meets the requirements in the train operation control system based on vehicle-to-vehicle communication is an urgent problem to be solved. Summary of the Invention

[0008] Aiming at the above deficiencies in the prior art, the train behavior planning method provided by the present invention for a vehicle-to-vehicle communication train control system can effectively plan the train behavior and meet the train following behavior.

[0009] To achieve the above invention purpose, the technical solution adopted by the present invention is as follows:

[0010] Provide a train behavior planning method suitable for a vehicle-to-vehicle communication train control system, which includes the following steps:

[0011] Obtain the train operation trajectory based on the historical operation data of trains on the same line and construct a trajectory distance table; where the train operation trajectories form an operation trajectory data set;

[0012] By setting a distance threshold between trajectories, a trajectory coverage table is formed based on the trajectory distance table; a set of base trajectories that cover all running trajectories and minimize the total distance between the base trajectory and other running trajectories is obtained from the trajectory coverage table;

[0013] After numbering the base trajectories and running trajectories respectively, they are used as the horizontal and vertical columns of the table to form a corresponding table of base trajectories; when the running trajectory corresponding to the table is covered by the corresponding base trajectory, the table is marked as 1; otherwise, it is marked as 0;

[0014] Extract the train following scenario vectors corresponding to each running trajectory in the same line to obtain a scenario vector table; the train following scenario vector includes the speed of the leading vehicle, the speed of the following vehicle, and the distance between the two vehicles at each sampling time point;

[0015] Construct a neural network trajectory prediction model with the train following scenario vector as the input and the base trajectory as the output; input the train following scenario vector of the target vehicle into the neural network trajectory prediction model, obtain the corresponding base trajectory and use it as the state change amount of the leading vehicle;

[0016] Based on the current running state of the leading vehicle, convert the state change amount of the leading vehicle into the running state trajectory of the leading vehicle;

[0017] Set the safety protection distance between the two vehicles, set the control objective function and constraint conditions for the following vehicle to track the leading vehicle, and obtain the control data of the following vehicle based on the running state trajectory of the leading vehicle.

[0018] Furthermore, the specific method for constructing the trajectory distance table includes:

[0019] Extract the continuous train absolute position and train running speed in the train operation data in the same line, that is, obtain the running state trajectory of the train;

[0020] Extract the train running state trajectory continuously for 10 seconds starting from each moment at a set sampling interval, calculate the speed change amount and position change amount at adjacent moments, obtain the train running trajectory, and further obtain the running trajectory data set;

[0021] Calculate the distance between all running trajectories in the running trajectory data set through the atomic norm to form a trajectory distance table, and its expression is:

[0022]

[0023] where d mn is the distance between running trajectory m and running trajectory n; I is the total number of sampling points in a single running trajectory; Δp mi and Δp ni are the position change amounts of running trajectory m and running trajectory n at the i-th sampling respectively; Δv miThey are the velocity change amounts of the running trajectories m and n at the i-th sampling respectively.

[0024] Furthermore, the setting method of the distance threshold between trajectories is as follows:

[0025] Calculate the coverage effect of the base trajectory set on all trajectories under different distance thresholds between trajectories. Its expression is:

[0026]

[0027] Where is the average distance between the base trajectory and its covered trajectories, that is, the coverage effect; H is the dimension of the base trajectory set; h is the base trajectory number; g h is the number of running trajectories covered by the h-th base trajectory; q is the running trajectory number covered by the base trajectory; is the distance between the h-th base trajectory and the q-th running trajectory it covers;

[0028] Taking the coverage dimension and the average distance between trajectories as indicators, calculate the score of the base trajectory. Its expression is:

[0029]

[0030] Where F is the score of the base trajectory; both α and β are weight parameters;

[0031] Select the distance threshold between trajectories corresponding to the lowest base trajectory score.

[0032] Furthermore, the neural network trajectory prediction model includes an input layer, a hidden layer, and an output layer connected in sequence; where the number of neurons in the input layer is 3, the number of layers in the hidden layer is 2, and the number of neurons in each layer of the input layer and the hidden layer is 64; the number of neurons in the output layer is equal to the number of base trajectories, that is, one neuron in the output layer corresponds to one base trajectory; the activation function of the hidden layer is the relu function, and the activation function of the output layer is the linear activation function; the neurons in the output layer are used to calculate the probability that a certain running trajectory is covered by each base trajectory, and set the output value of the neuron corresponding to the highest probability value to 1, and set the output values of the remaining neurons in the output layer to 0; the base trajectory corresponding to the neuron whose output value is set to 1 is the base trajectory output by the neural network trajectory prediction model, that is, the neural network trajectory prediction model outputs a binary vector with the same dimension as the base trajectory set.

[0033] Furthermore, the expression of the activation function of the hidden layer of the neural network trajectory prediction model is:

[0034]

[0035] Where relu(.) represents the relu function; x is the activation object.

[0036] Furthermore, the loss function of the neural network trajectory prediction model during training is as follows:

[0037]

[0038] where \(J(y, z)\) is the loss value during the training of a sample; \(A\) is the dimension of the binary vector output by the neural network trajectory prediction model; \(y\) a is the \(a\)-th element in the binary vector output by the neural network trajectory prediction model; \(z\) a is the category of the \(a\)-th base trajectory in the true base trajectory set. When the base trajectory corresponds to the input train following scenario vector, the category is 1, otherwise the category is 0; \(\omega_1\) and \(\omega_0\) are both weight coefficients; \(\gamma_1\) and \(\gamma_2\) are constants greater than or equal to 0 and less than or equal to 1; is the indicator function, \(\chi\) is the condition of the true value \(z\) a ; relu(.) represents the relu function.

[0039] Furthermore, the specific method for converting the state change amount of the leading vehicle into the running state trajectory of the leading vehicle based on the current running state of the leading vehicle is as follows:

[0040] Add the position in the current running state of the leading vehicle to the first position change amount in the base trajectory prediction result to obtain the first predicted position of the leading vehicle; add the speed in the current running state of the leading vehicle to the first speed change amount in the base trajectory prediction result to obtain the first predicted speed of the leading vehicle;

[0041] Add the \((j - 1)\)-th predicted position of the leading vehicle to the \(j\)-th position change amount in the base trajectory prediction result to obtain the \(j\)-th predicted position of the leading vehicle; add the \((j - 1)\)-th predicted speed of the leading vehicle to the \(j\)-th speed change amount in the base trajectory prediction result to obtain the \(j\)-th predicted speed of the leading vehicle; obtain the running state trajectory of the leading vehicle, and its expression is:

[0042] \(p\) b,1 = \(p\) b + \(\Delta p_1, v\) b,1 = \(v\) b + \(\Delta v_1\)

[0043] \(p\) b,2 = \(p\) b,1 + \(\Delta p_2, v\) b,2 = \(v\) b,1 + \(\Delta v_2\)

[0044] …

[0045] \(p\) b,J = \(p\) b,J-1 + \(\Delta p\) J , \(v\) b,J = \(v\) b,J-1 + \(\Delta v\) J

[0046] where p b and v b are respectively the positions in the current running state of the leading vehicle; Δp1 and Δv1 are respectively the first position change amount and the first position speed change amount in the basic trajectory prediction result; p b,1 and p b,2 are respectively the first predicted position and the second predicted position of the leading vehicle; v b,1 and v b,2 are respectively the first predicted speed and the second predicted speed of the leading vehicle; Δp2 and Δv2 are respectively the second position change amount and the second position speed change amount in the basic trajectory prediction result; p b,J and v b,J are respectively the Jth predicted position and the Jth predicted speed of the leading vehicle; p b,J-1 and v b,J-1 are respectively the (J - 1)th predicted position and the (J - 1)th predicted speed of the leading vehicle; Δp J and Δv J are respectively the Jth position change amount and the Jth position speed change amount in the basic trajectory prediction result.

[0047] Furthermore, set the safety protection distance between the two vehicles, set the control objective function and constraint conditions for the following vehicle to track the leading vehicle, and the specific method for obtaining the control data of the following vehicle based on the running state trajectory of the leading vehicle includes:

[0048] Set the safety protection distance d safe ;

[0049] Set the control objective function for the following vehicle to track the leading vehicle as:

[0050]

[0051] where x k is the position of the following vehicle at time k; is the position of the leading vehicle at time k, obtained from the running state trajectory of the leading vehicle; u k is the acceleration of the following vehicle at time k; u k-1 is the acceleration of the following vehicle at time k - 1; means to find the value of u k when it is the smallest;

[0052] Set the constraint conditions for the following vehicle to track the leading vehicle as:

[0053]

[0054] 0 ≤ v k ≤ v max

[0055] d ≥ ds

[0056]

[0057] where s.t. represents the constraint condition; x k+1 is the position of the following vehicle at the (k + 1)-th moment; v k is the speed of the following vehicle at the k-th moment; v max is the speed limit value of the road; represents the universal quantifier; K is the total number of data points in the running state trajectory of the leading vehicle, that is, the total number of predicted time points; u represents the acceleration interval, d s is the safe distance between the two vehicles; d is the distance that the following vehicle tracks the leading vehicle, d = p b - p f - d safe - l, p b is the position in the current running state of the leading vehicle, p f is the position in the current running state of the following vehicle, and l is the vehicle length;

[0058] On the basis of satisfying the constraint conditions, the acceleration control sequence [u1,..., u K of the following vehicle is calculated through the control objective function for the following vehicle to track the leading vehicle, that is, the control data of the following vehicle is obtained.

[0059] The beneficial effects of the present invention are as follows: The output of the neural network trajectory prediction model of the present invention is changed from the train running state values within a certain period of time to the base trajectory of the train running trajectory. A base trajectory set is constructed based on the train historical running data of the same line, and the neural network trajectory prediction model is matched in the base trajectory set to output the corresponding base trajectory of the leading vehicle and then obtain its running state within a future period of time, reducing the prediction dimension of the neural network trajectory prediction model, effectively reducing the calculation complexity of the neural network trajectory prediction model, improving the trajectory prediction efficiency, and using the neural network trajectory prediction model for prediction control to enable the following vehicle to track the running dynamics of the leading vehicle while shortening the distance from the leading vehicle, realizing the safe following of trains with shorter intervals. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a schematic flow chart of this method;

[0061] Figure 2 is a schematic diagram showing the distance between trajectories and the mutual coverage of train running trajectories;

[0062] Figure 3 is a schematic structural diagram of the neural network trajectory prediction model;

[0063] Figure 4 is a schematic flow chart of the control in the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0064] The specific embodiments of the present invention will be described below to facilitate those skilled in the art to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. The step numbers are only used to number each operation and do not correspond to a clear sequence. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions made using the concept of the present invention are within the scope of protection.

[0065] As Figure 1 shown, the train behavior planning method suitable for the vehicle-to-vehicle communication train control system includes the following steps:

[0066] S1. Obtain the running trajectory of the train based on the historical running data of the trains on the same line and construct a trajectory distance table; wherein the running trajectories of the trains constitute a running trajectory data set;

[0067] S2. By setting a distance threshold between trajectories, form a trajectory coverage table based on the trajectory distance table; obtain a set of base trajectories that cover all running trajectories and minimize the sum of the distances between the base trajectories and other running trajectories from the trajectory coverage table;

[0068] S3. After numbering the base trajectories and running trajectories respectively, use them as the horizontal rows (for example, the base trajectories can be numbered as b1, b2,..., b M ) and vertical columns (for example, the running trajectories can be numbered as a1, a2,..., a R ) of a table to form a base trajectory correspondence table; wherein when the running trajectory corresponding to the table is covered by the corresponding base trajectory, the table is marked as 1; otherwise, it is marked as 0; the trajectory correspondence table is a statistical manifestation of the base trajectory covering other running trajectories;

[0069] S4. Extract the train following scenario vectors corresponding to each running trajectory on the same line to obtain a scenario vector table; wherein the train following scenario vector includes the speed of the leading vehicle, the speed of the following vehicle, and the distance between the two vehicles at each sampling time point; the train following scenario vector formed by q sampling time points can be expressed as [v b , v f , d] q ;

[0070] S5. Construct a neural network trajectory prediction model with the train following scenario vector as the input and the base trajectory as the output; input the train following scenario vector of the target vehicle into the neural network trajectory prediction model, obtain the corresponding base trajectory and use it as the state change amount of the leading vehicle;

[0071] S6. Based on the current running state of the leading vehicle, convert the state change amount of the leading vehicle into the running state trajectory of the leading vehicle;

[0072] S7. Set the safety protection distance between the two vehicles, set the control objective function and constraint conditions for the rear vehicle to track the front vehicle, and obtain the control data of the rear vehicle based on the running state trajectory of the front vehicle.

[0073] In this embodiment, the specific method for constructing the trajectory distance table in step S1 includes the following sub-steps:

[0074] S1-1. Extract the continuous train absolute positions and train running speeds in the train operation data on the same line, that is, obtain the running state trajectory of the train, which can be expressed as [p0, v0, p1, v1, …, p 50 , v 50 ;

[0075] S1-2. Extract the train running state trajectories within 10 seconds continuously starting from each moment at a set sampling interval, calculate the speed change amount and position change amount between adjacent moments, obtain the running trajectory of the train, and further obtain the running trajectory data set; where the sampling interval is 0.2 seconds, and a single group of sampling obtains a total of 50 data samples; where the running trajectory of the train can be expressed as [p1 - p0, v1 - v0, …, p 50 - p 49 , v 50 - v 49 or [Δp1, Δv1, …, Δp 50 , Δv 50 ;

[0076] S1-3. Calculate the distances between all running trajectories in the running trajectory data set through atomic norm to form a trajectory distance table, and its expression is:

[0077]

[0078] where d mn is the distance between running trajectory m and running trajectory n; I is the total number of sampling points in a single running trajectory, taking the value of 50 when the sampling interval is 0.2 seconds; Δp mi and Δp ni are the position change amounts of running trajectory m and running trajectory n at the i-th sampling respectively; Δv mi are the speed change amounts of running trajectory m and running trajectory n at the i-th sampling respectively.

[0079] In this embodiment, the setting method of the distance threshold between trajectories in step S2 includes the following sub-steps:

[0080] S2-1. Calculate the coverage effect of the base trajectory set on all trajectories under different distance thresholds between trajectories, and its expression is:

[0081]

[0082] Among them is the average distance between the base trajectory and its covered trajectory, that is, the coverage effect; H is the dimension of the base trajectory set; h is the base trajectory number; g h is the number of running trajectories covered by the h-th base trajectory; q is the running trajectory number covered by the base trajectory; is the distance between the h-th base trajectory and the q-th running trajectory it covers;

[0083] S2-2. Taking the coverage dimension and the average distance between trajectories as indicators, calculate the score of the base trajectory, and its expression is:

[0084]

[0085] Among them, F is the score of the base trajectory; both α and β are weight parameters;

[0086] S2-3. Select the distance threshold between trajectories corresponding to the lowest score of the base trajectory.

[0087] When selecting the base trajectory, in this embodiment, the greedy algorithm can be used to select the base trajectories in descending order of the trajectory coverage ability. When the number of covered trajectories of two running trajectories is the same in one round of selection, compare the sum of their distances to the covered running trajectories, and select the smaller one as the base trajectory of this round. In the next round of selection, the covered running trajectories are excluded.

[0088] In the specific implementation process, as Figure 2 shown, where the left figure is the table of the distance between all running trajectories obtained according to the distance calculation formula between trajectories, and the right figure is the mutual coverage situation of all running trajectories obtained according to the distance values in the left figure and the distance threshold ∈ between trajectories. If the distance between trajectories is less than ∈, it is marked as 1, otherwise it is 0. If the value of the table (n, 1) is 1, then trajectory 1 and trajectory n can represent (or cover) each other. The last column in the table represents the coverage ability of each trajectory, which is represented by the number of trajectories that can be covered, and is used as the basis for judging whether a trajectory can be used as a base trajectory.

[0089] In this embodiment, as Figure 3As shown, the neural network trajectory prediction model includes an input layer, a hidden layer, and an output layer connected in sequence; among them, the number of neurons in the input layer is 3, the number of layers in the hidden layer is 2, and the number of neurons in each layer of the input layer and the hidden layer is 64; the number of neurons in the output layer is equal to the number of base trajectories, that is, one neuron in the output layer corresponds to one base trajectory; the activation function of the hidden layer is the relu function, and the activation function of the output layer is a linear activation function; the neurons in the output layer are used to calculate the probability that a certain running trajectory is covered by each base trajectory, and set the output value of the neuron corresponding to the highest probability value to 1, and set the output values of the remaining neurons in the output layer to 0; the base trajectory corresponding to the neuron whose output value is set to 1 is the base trajectory output by the neural network trajectory prediction model, that is, the neural network trajectory prediction model outputs a binary (0-1) vector with the same dimension as the base trajectory set.

[0090] The expression of the activation function of the hidden layer of the neural network trajectory prediction model is:

[0091]

[0092] where relu(.) represents the relu function; x is the activation object;

[0093] The loss function of the neural network trajectory prediction model during training is:

[0094]

[0095] where J(y,z) is the loss value during the training of a sample; A is the dimension of the binary vector output by the neural network trajectory prediction model; y a is the a-th element in the binary vector output by the neural network trajectory prediction model; z a is the category of the a-th base trajectory in the true base trajectory set. When the base trajectory corresponds to the input train following scenario vector, the category is 1, otherwise the category is 0; ω1 and ω0 are both weight coefficients; γ1 and γ2 are constants greater than or equal to 0 and less than or equal to 1, which are used to increase the robustness of the neural network trajectory prediction model's prediction classification; is the indicator function, χ is the condition of the true value z a ; relu(.) represents the relu function.

[0096] In the specific implementation process, the neural network trajectory prediction model uses the gradient descent method to find the model parameters, that is, the weight coefficients, that minimize the loss function during training. The update formula is as follows:

[0097]

[0098] where g represents the number of iterations; J(w) is the loss value; η is the learning rate; w (g)is the weight coefficient obtained after the g-th iteration; w (g+1) is the weight coefficient obtained after the (g + 1)-th iteration; is the gradient of J(w) at w = w (g) at that point.

[0099] The specific method for converting the state change amount of the leading vehicle into the running state trajectory of the leading vehicle based on the current running state of the leading vehicle in step S6 includes the following sub-steps:

[0100] S6-1. Add the position in the current running state of the leading vehicle to the first position change amount in the base trajectory prediction result to obtain the first predicted position of the leading vehicle; add the speed in the current running state of the leading vehicle to the first speed change amount in the base trajectory prediction result to obtain the first predicted speed of the leading vehicle;

[0101] S6-2. Add the (j - 1)-th predicted position of the leading vehicle to the j-th position change amount in the base trajectory prediction result to obtain the j-th predicted position of the leading vehicle; add the (j - 1)-th predicted speed of the leading vehicle to the j-th speed change amount in the base trajectory prediction result to obtain the j-th predicted speed of the leading vehicle; obtain the running state trajectory of the leading vehicle, and its expression is:

[0102] p b,1 = p b + Δp1, v b,1 = v b + Δv1

[0103] p b,2 = p b,1 + Δp2, v b,2 = v b,1 + Δv2

[0104] …

[0105] p b,J = p b,J-1 + Δp J , v b,J = v b,J-1 + Δv J

[0106] where p b and v b are respectively the position in the current running state of the leading vehicle; Δp1 and Δv1 are respectively the first position change amount and the first position speed change amount in the base trajectory prediction result; p b,1 and p b,2 are respectively the first predicted position and the second predicted position of the leading vehicle; v b,1 and v b,2The first predicted speed and the second predicted speed of the leading vehicle respectively; Δp2 and Δv2 are the second position change amount and the second position speed change amount in the basic trajectory prediction result; p b,J and v b,J are the Jth predicted position and the Jth predicted speed of the leading vehicle respectively; p b,J-1 and v b,J-1 are the (J - 1)th predicted position and the (J - 1)th predicted speed of the leading vehicle respectively; Δp J and Δv J are the Jth position change amount and the Jth position speed change amount in the basic trajectory prediction result respectively.

[0107] In step S7, set the safety protection distance between the two vehicles, set the control objective function and constraint conditions for the following vehicle to track the leading vehicle, and the specific method for obtaining the control data of the following vehicle based on the running state trajectory of the leading vehicle includes the following sub-steps:

[0108] S7-1. Set the safety protection distance d safe ;

[0109] S7-2. Set the control objective function for the following vehicle to track the leading vehicle as:

[0110]

[0111] where x k is the position of the following vehicle at time k; is the position of the leading vehicle at time k, obtained from the running state trajectory of the leading vehicle; u k is the acceleration of the following vehicle at time k; u k-1 is the acceleration of the following vehicle at time k - 1; represents to find the value of u k when it is the smallest;

[0112] S7-3. Set the constraint conditions for the following vehicle to track the leading vehicle as:

[0113]

[0114] 0 ≤ v k ≤ v max

[0115] d ≥ d s

[0116]

[0117] where s.t. represents the constraint conditions; x k+1 is the position of the following vehicle at time k + 1; v k is the speed of the following vehicle at time k; v max is the road speed limit value; represents the universal quantifier; K is the total number of data points in the running state trajectory of the leading vehicle, that is, the total number of predicted time points; u represents the acceleration interval, d s is the safe distance between the two vehicles; d is the distance of the following vehicle tracking the leading vehicle, d = p b -p f -d safe -l, p b is the position in the current running state of the leading vehicle, p f is the position in the current running state of the following vehicle, l is the vehicle length;

[0118] S7-4. On the basis of satisfying the constraint conditions, calculate the acceleration control sequence [u1,..., u K of the following vehicle through the control objective function of the following vehicle tracking the leading vehicle, that is, obtain the control data of the following vehicle.

[0119] In the specific implementation process, as Figure 4 shown, combining the actual running state y of the following vehicle and the running state input of the leading vehicle, through the neural network trajectory prediction model, predict the states of the leading vehicle at the next N moments as the control reference trajectory y r . y m represents the predicted states of the position and speed of the following vehicle at the next N moments under the acceleration u control evolved according to the train dynamic simplified dynamics model. The controller iteratively generates the optimal control acceleration sequence at the next N moments to satisfy the control constraints and minimize the objective function value.

[0120] To sum up, the present invention changes the output of the neural network trajectory prediction model from the train running state values within a certain time to the base trajectory of the train running trajectory, constructs a base trajectory set based on the train historical running data of the same line, and makes the neural network trajectory prediction model match in the base trajectory set, outputs the corresponding base trajectory of the leading vehicle and then obtains its running state within a period of time in the future, reduces the prediction dimension of the neural network trajectory prediction model, effectively reduces the calculation complexity of the neural network trajectory prediction model, improves the trajectory prediction efficiency, and uses the neural network trajectory prediction model for predictive control to enable the following vehicle to track the running dynamics of the leading vehicle while shortening the distance from the leading vehicle, realizing the safe following of trains with shorter intervals.

Claims

1. A train behavior planning method suitable for a train-to-train communication train control system, characterized in that: The following steps are involved: Obtaining the running track of the train based on the historical running data of the train on the same line and constructing a track distance table; The running track of the train constitutes the running track data set; By setting the distance threshold between trajectories, a trajectory coverage table is formed based on the trajectory distance table; Obtain a base trajectory set that covers all running trajectories and minimizes the sum of distances between the base trajectory and other running trajectories from the trajectory coverage table; The base trajectory and the running trajectory are numbered respectively as the horizontal row and vertical column of the table, forming a base trajectory correspondence table; when the running trajectory corresponding to the table is covered by the corresponding base trajectory, the table is marked as 1; otherwise, it is marked as 0; Extract the train following scene vector corresponding to each running track on the same line to obtain a scene vector table; The train following scenario vector includes the speed of the leading vehicle, the speed of the following vehicle, and the distance between the two vehicles at each sampling time point; Construct a neural network trajectory prediction model with a train-following scenario vector as input and a base trajectory as output; input the train-following scenario vector of the target vehicle into the neural network trajectory prediction model, obtain the corresponding base trajectory and use it as the state change of the leading vehicle; Based on the current running state of the front vehicle, the state change of the front vehicle is converted into the running state trajectory of the front vehicle; Set the safety protection distance between the two vehicles, set the control objective function and constraints for the rear vehicle to track the front vehicle, and obtain the control data of the rear vehicle based on the operating status trajectory of the front vehicle.

2. The train behavior planning method suitable for a train-to-train communication train control system according to claim 1, characterized in that: The specific method of constructing the trajectory distance table includes: Extracting the continuous absolute position and speed of trains from the train running data on the same line, that is, obtaining the running status trajectory of the train; The train running status trajectory within 10 seconds from each moment is extracted at the set sampling interval, and the speed change and position change at adjacent moments are calculated to obtain the train running trajectory, and then the running trajectory data set is obtained; The distances between all running trajectories in the running trajectory data set are calculated by the atomic norm to form a trajectory distance table, which is expressed as: where d mn is the distance between running track m and running track n; I is the total number of sampling points in a single running track; Δp mi and Δp ni are the position changes of running tracks m and running tracks n at the i-th sampling time; Δv mi are the speed changes of running trajectory m and running trajectory n at the i-th sampling time respectively.

3. The train behavior planning method suitable for a train-to-train communication train control system according to claim 2, characterized in that: The threshold of the distance between tracks is set as follows: Calculate the coverage effect of the base trajectory set on all trajectories under different distance thresholds between trajectories. The expression is: in is the mean distance between the base trajectory and its coverage trajectory, i.e., the coverage effect; H is the dimension of the base trajectory set; h is the base trajectory number; g h is the number of running tracks covered by the h-th basic track; q is the running track number covered by the basic track; is the distance between the h-th base trajectory and the q-th running trajectory it covers; The coverage dimension and the mean distance between trajectories are used as indicators to calculate the score of the base trajectory, which is expressed as: Where F is the score of the base trajectory; α and β are weight parameters; The inter-trajectory distance threshold corresponding to the lowest base trajectory score is selected.

4. The train behavior planning method suitable for a train-to-train communication train control system according to claim 1, characterized in that: The neural network trajectory prediction model includes an input layer, a hidden layer and an output layer connected in sequence; the number of neurons in the input layer is 3, the number of hidden layers is 2, and the number of neurons in each layer of the input layer and the hidden layer is 64; the number of neurons in the output layer is equal to the number of basis trajectories, that is, one neuron in the output layer corresponds to one basis trajectory; the activation function of the hidden layer is the relu function, and the activation function of the output layer is the linear activation function; the neurons in the output layer are used to calculate the probability that a certain running trajectory is covered by each basis trajectory, and the output value of the neuron corresponding to the highest probability value is set to 1, and the output values ​​of the remaining neurons in the output layer are set to 0; the basis trajectory corresponding to the neuron whose output value is set to 1 is the basis trajectory output by the neural network trajectory prediction model, that is, the output of the neural network trajectory prediction model is a binary vector with the same dimension as the basis trajectory set.

5. The train behavior planning method suitable for a train-to-train communication train control system according to claim 4, characterized in that: The expression of the activation function of the hidden layer of the neural network trajectory prediction model is: Where relu(.) represents the relu function; x is the activation object.

6. The train behavior planning method suitable for a train-to-train communication train control system according to claim 4, characterized in that: The loss function of the neural network trajectory prediction model during training is: Where J(y,z) is the loss value of a sample during training; A is the dimension of the binary vector output by the neural network trajectory prediction model; y a Output the ath element in the binary vector of the neural network trajectory prediction model; z a is the category of the ath basis trajectory in the real basis trajectory set. When the basis trajectory corresponds to the input train following scenario vector, the category is 1, otherwise the category is 0; ω1 and ω0 are both weight coefficients; γ1 and γ2 are constants greater than or equal to 0 and less than or equal to 1; is the indicative function, χ is the true value z a Condition; relu(.) represents the relu function.

7. The train behavior planning method suitable for a train-to-train communication train control system according to claim 2, characterized in that: The specific method of converting the state change of the preceding vehicle into the running state trajectory of the preceding vehicle based on the current running state of the preceding vehicle is: The position of the preceding vehicle in the current running state is added to the first position change in the base trajectory prediction result to obtain the first predicted position of the preceding vehicle; the speed of the preceding vehicle in the current running state is added to the first speed change in the base trajectory prediction result to obtain the first predicted speed of the preceding vehicle; The j-1th predicted position of the preceding vehicle is added to the j-th position change in the prediction result of the base trajectory to obtain the j-th predicted position of the preceding vehicle; the j-1th predicted speed of the preceding vehicle is added to the j-th speed change in the prediction result of the base trajectory to obtain the j-th predicted speed of the preceding vehicle; the running state trajectory of the preceding vehicle is obtained, and its expression is: p b,1 =p b +Δp1,v b,1 =v b +Δv1 p b,2 =p b,1 +Δp2,v b,2 =v b,1 +Δv2 … p b,J =p b,J-1 +Δp J ,v b,J =v b,J-1 +Δv J where p b and v b are the positions of the preceding vehicle in its current running state; Δp1 and Δv1 are the first position change and the first position speed change in the basic trajectory prediction results; p b,1 and p b,2 are the first predicted position and the second predicted position of the front vehicle respectively; v b,1 and v b,2 are the first predicted speed and the second predicted speed of the preceding vehicle respectively; Δp2 and Δv2 are the second position change and the second position speed change in the prediction result of the base trajectory respectively; p b,J and v b,J are the J-th predicted position and J-th predicted speed of the preceding vehicle respectively; p b,J-1 and v b,J-1 are the J-1th predicted position and J-1th predicted speed of the preceding vehicle respectively; Δp J and Δv J are the J-th position change and the J-th position speed change in the basic trajectory prediction results respectively.

8. The train behavior planning method suitable for a train-to-train communication train control system according to claim 1, characterized in that: The specific method of setting the safety protection distance between the two vehicles, setting the control objective function and constraint conditions for the rear vehicle to track the front vehicle, and obtaining the control data of the rear vehicle based on the running state trajectory of the front vehicle includes: Set the safety protection distance d between the two vehicles safe ; The control objective function of setting the rear vehicle to track the front vehicle is: where x k is the position of the following car at time k; is the position of the preceding vehicle at time k, obtained from the running state trajectory of the preceding vehicle; u k is the acceleration of the following vehicle at time k; u k-1 is the acceleration of the following vehicle at time k-1; Express request The smallest u k value; The constraints for the rear vehicle to track the front vehicle are set as follows: 0≤v k ≤v max d≥d s Where st represents the constraint condition; x k+1 is the position of the following vehicle at time k+1; v k is the speed of the following vehicle at time k; v max is the road speed limit; represents the universal quantifier; K is the total number of data points in the running status trajectory of the preceding vehicle, that is, the total number of predicted time points; represents the acceleration range, d s is the safe distance between the two vehicles; d is the distance that the rear vehicle follows the front vehicle, d = p b -p f -d safe -l,p b is the position of the preceding vehicle in its current operating state, p f is the position of the following vehicle in the current running state, l is the vehicle length; On the basis of satisfying the constraint conditions, the acceleration control sequence [u1,…,u K ], that is, obtain the control data of the following vehicle.

Images