Track panel fine adjustment algorithm for double-block ballastless track

Abstract

The invention discloses a track panel fine adjustment algorithm for a double-block ballastless track, which belongs to the technical field of track panel fine adjustment, and comprises the following steps: S1, establishing a physical model and a virtual adjustment system of a pair of support legs of a track panel, S2, establishing a motion planner model, S3, randomly initializing a state; s4, continuing to output a new action through a reinforcement learning algorithm; s5, repeating S4 until convergence; S6, obtaining initial states and real-time states of a cross beam inclination angle and a lead screw deflection angle of the field control system through an inclination angle sensor; s7, the moving distance of each connecting rod on the straight line where the lead screws are located is calculated, the kinematics data of the steel rail are converted into kinematics data of each lead screw through a lead screw motion planner, and the track panel is controlled to complete adjustment; and S8, the steps from S6 to S7 are repeated till the steel rail reaches the error range allowed by the target point. Accurate track panel superelevation adjustment and horizontal adjustment can be carried out according to the steel rail parameters, calculation is simple, and the steps are few.

Description

Technical Field

[0001] The invention belongs to the technical field of twin-block ballastless track laying, and in particular relates to a track arrangement fine-tuning algorithm for twin-block ballastless track. Background Art

[0002] The double-block ballastless track structure has now become the main track structure for high-speed railways. The track precision adjustment during high-speed rail laying will directly affect the safety of high-speed train operation.

[0003] In the field of railway construction, a number of technologies are used to solve the problem of efficient adjustment of rail positioning. For example, a drive device is introduced in the process of rail adjustment to achieve the adjustment of rail height. This technology achieves precise control of rail height by adjusting the joint changes of the screw. However, this method has limitations in the adjustment direction and lacks flexibility, which to a certain extent limits its scope of application and efficiency. In addition, the position of the rail is adjusted in multiple directions by using motor adjustment and position comparison item by item. This technology can theoretically provide higher flexibility and a wider adjustment range, but in actual operation, multi-step adjustment is required, which not only increases the complexity of the operation, but also leads to an extension of the construction period, thereby affecting the overall construction efficiency. Summary of the invention

[0004] In view of the above problems existing in the prior art, the purpose of the present invention is to provide a track fine-tuning algorithm for a dual-block ballastless track, which can accurately adjust the track superelevation and level according to the rail parameters, meeting the requirements for precise track adjustment of the dual-block ballastless track. The calculation is simple and the steps are few, which reduces the complexity of track adjustment and improves the construction efficiency of track laying. The introduction of a reinforcement learning algorithm can adapt to the changes in different construction environments of track laying, reduce on-site debugging costs, and improve laying accuracy and efficiency.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: A track arrangement fine-tuning algorithm for a double-block ballastless track comprises the following steps: S1: Establish a physical model of a pair of legs of the rail track, and construct a virtual adjustment system based on the physical model, wherein the physical model of the legs includes a crossbeam and two rails, the two rails are arranged in parallel and supported on the crossbeam, and the virtual adjustment system includes four connecting rods and four screw rods, each end of the crossbeam is connected to two connecting rods, the four screw rods respectively drive one end of the four connecting rods to reciprocate linearly, and the four connecting rods drive the crossbeam hinged to the other end thereof to rotate and / or move; S2: Establishing a motion planner model for the outrigger screw, wherein the planner model includes a motion target position and a motion speed; S3: Randomly initialize the starting position state of each connecting rod on the lead screw and the starting deflection state of the connecting rod, and perform random processing on the environmental observation data; S4: Through the reinforcement learning algorithm, the rail motion trajectory is selected according to the environmental observation data and reward prediction, and the screw motion planner is used to convert the rail kinematic data into motor control instructions for controlling the screw movement, drive the outrigger adjustment, and update the reward data based on the new environmental observation data using the reward function, and continue to output new actions; S5: Repeat S4, iteratively update the strategy learning and value estimation network and store the experience in the experience pool until convergence, end the training and save the strategy as the rail motion trajectory; S6: Obtain the initial state and real-time state of the crossbeam inclination angle and the screw rod deflection angle of the field control system through the inclination sensor; S7: Calculate the moving distance of each connecting rod on the straight line where the screw rod is located, convert the kinematic data of the rail into the kinematic data of each screw rod through the screw rod motion planner, and control the rail arrangement to complete the adjustment; S8: Repeat S6 to S7 until the rail reaches the target point within the allowable error range.

[0006] As a further improvement of the above technical solution.

[0007] In step S1, the four connecting rods are respectively the first right connecting rod, the second right connecting rod, the first left connecting rod, and the second left connecting rod. The four screw rods are arranged in parallel and at intervals, namely the left screw rod, the second left screw rod, the second right screw rod, and the right screw rod arranged in sequence in one direction. The first left connecting rod, the second left connecting rod, the first right connecting rod, and the second right connecting rod are respectively connected to the left screw rod, the second left screw rod, the right screw rod, and the second right screw rod. The upper end of the first right connecting rod is connected to the right screw rod, and the lower end is connected to the crossbeam. The lower end of the second right connecting rod is connected to the right screw rod, and the upper end is connected to the crossbeam. The upper end of the first left connecting rod is connected to the left screw rod, and the lower end is connected to the crossbeam. The lower end of the second left connecting rod is connected to the left screw rod, and the upper end is connected to the crossbeam. There is a center line. The left screw rod and the right screw rod are symmetrical with respect to the center line, and the left and right screw rods are also symmetrical with respect to the center line. The first left connecting rod and the first right connecting rod are symmetrical with respect to the center line, and the left and right connecting rods are also symmetrical with respect to the center line.

[0008] The environmental observation data during the track adjustment process includes: sensor parameters, device status and target parameters, among which the sensor parameters include: the distance L between the left wire rod and the right wire rod 1 , the distance between the left and right screw rods L 2 , the distance between the two rails L 3, the height difference between the two rails, the offset distance, and the inclination angle, wherein the offset distance refers to the distance that the center point between the two rails deviates from the center line, and the inclination angle refers to the angle between the crossbeam and the horizontal plane; the device status includes: the displacement and speed of each connecting rod on the straight line where the corresponding screw rod is located, and the connecting rod deflection angle; the target parameters include: the target superelevation value and the target offset distance, wherein the superelevation value refers to the height difference between the two rails.

[0009] Step S4 includes the following steps: S4-1: Initialization state: obtain the initial state feature vector of the track, including sensor parameters, device state, and target parameters; S4-2: Action generation: Generate actions according to the current state through the current network strategy function in strategy learning, where the actions include rail target position and movement speed instructions; S4-3: Action execution and observation: Send the action to the screw motion planner, convert it into motor control execution instructions, and get the new state; S4-4: Experience storage: The five-tuple consisting of the current state, action, immediate reward, new state, and termination flag is stored in the experience replay pool, wherein the immediate reward is calculated by the reward function to reflect the adjustment accuracy, and the termination flag is used to determine whether the target error range is reached; S4-5: Network training: sampling multiple samples from the experience replay pool; S4-6: Target network synchronization: copy the current network parameters of value estimation or policy learning to the target network every C steps, where C represents the update frequency of the target network parameters; S4-7: Loop termination condition: If the new state is the termination state, that is, the track reaches the target accuracy, the current training round ends; otherwise, return to step S4-2 to continue iteration.

[0010] Step S7 includes the following steps: S7-1: perform super-height adjustment, after which the super-height value changes; S7-2: Perform horizontal adjustment with superelevation, after which the center point between the two rails is offset.

[0011] In step S7-1, the inclination angle of the beam that needs to be adjusted is calculated according to the superelevation value that needs to be adjusted and the distance between the two rails, and the displacement that each connecting rod needs to complete on the corresponding screw rod is calculated according to the relationship between the linear movement speeds of the two connecting rods on the same side of the center line.

[0012] Step S7-2 includes the following steps: S7-2-1: Calculate the deflection angle of each connecting rod after superelevation adjustment; S7-2-2: Calculate the displacement that each connecting rod needs to complete on the corresponding screw rod according to the target deflection angle of each connecting rod, the target offset distance, and the deflection angle obtained in step S7-2-1.

[0013] In step S7-1: ; ; ; ; ; ; Among them: A is the ultra-high value that needs to be adjusted, The inclination angle of the beam that needs to be adjusted. is the speed of the left threaded rod, is the speed of the right threaded rod, is the movement speed of the left second screw rod, is the movement speed of the second right screw, is the displacement that the left and right screws need to move, is the displacement that the left and right two screw rods need to move, and T is the time for super-high adjustment.

[0014] In step S7-2-2: ; ; in i =1,2,3,4, represents the length of each connecting rod, Represents the displacement of each connecting rod on the corresponding screw rod, It indicates the deflection angle of each connecting rod relative to the initial state after the horizontal adjustment with superelevation is completed. Indicates the offset distance between the center points of the two rails. Indicates the deflection angle after superelevation adjustment relative to the superelevation adjustment.

[0015] The beneficial effects of the present invention are: (1) It can accurately adjust the track height and level according to the rail parameters, meeting the requirements for precise track adjustment of double-block ballastless track.

[0016] (2) The calculation is simple and the steps are few, which reduces the complexity of track adjustment and improves the construction efficiency of track laying.

[0017] (3) The introduction of reinforcement learning algorithm can adapt to the different construction environment changes of track laying, reduce on-site debugging costs, and improve laying accuracy and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the track arrangement fine-tuning algorithm of the present invention.

[0019] Figure 2 This is a schematic diagram of the mechanism before the track is adjusted.

[0020] Figure 3 This is a schematic diagram of the mechanism after the track has been adjusted for super-height.

[0021] Figure 4 This is a schematic diagram of the mechanism after the track has been adjusted horizontally with superelevation.

[0022] Figure numerals: 1, crossbeam, 2, rail, ab, right first connecting rod, cd, right second connecting rod, ef, left first connecting rod, gh, left second connecting rod, 11, left first screw rod, 12, left second screw rod, 21, right first screw rod, 22, right second screw rod. DETAILED DESCRIPTION

[0023] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0024] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0025] A track fine-tuning algorithm for twin-block ballastless track, such as Figure 1 As shown, the following steps are included: S1: Establish a physical model of a pair of legs of the track to simplify the structural analysis, and build a virtual adjustment system based on the physical model to simulate the influence of the movement of the lead screw (see description below) on the track parameters; the legs are the unit structures that make up the track.

[0026] The physical model of the support leg includes a crossbeam 1 and two rails 2. The two rails 2 are arranged in parallel and supported on the crossbeam 1. The length of the rails 2 is substantially perpendicular to the length of the crossbeam 1.

[0027] The virtual adjustment system includes four connecting rods and four screw rods. The four connecting rods are respectively connected to the four screw rods. One end of each connecting rod is hinged on the crossbeam 1, and the other end is connected to the screw rod and can move back and forth on the straight line where the corresponding screw rod is located. Specifically, the screw rod can have two movement modes: First, the screw rod can only move back and forth along the straight line where it is located, so as to drive one end of the connecting rod connected to it to move synchronously, and the motor drives the screw rod to move in a straight line. Second, the screw rod itself can only rotate by itself, and one end of the connecting rod is screwed on the screw rod. When the screw rod rotates, one end of the connecting rod connected to it moves in a straight line along the length direction of the screw rod, and the motor drives the screw rod to rotate. For the convenience of description, the displacement and speed of one end of the connecting rod connected to the screw rod moving on the straight line where the screw rod is located are referred to as the displacement and speed of the screw rod.

[0028] The four connecting rods are respectively the first right connecting rod ab, the second right connecting rod cd, the first left connecting rod ef, and the second left connecting rod gh. The four screw rods are arranged in parallel and spaced apart, and are respectively the left screw rod 11, the second left screw rod 12, the second right screw rod 22, and the right screw rod 21 arranged in sequence in one direction. There is a center line, and the left screw rod 11 and the right screw rod 21 are symmetrical relative to the center line, and the left second screw rod 12 and the right second screw rod 22 are also symmetrical relative to the center line.

[0029] The two end points of the right connecting rod ab are point a and point b, wherein point b is hinged to one end of the beam 1, and point a is hinged to the right threaded rod 21, so that the right connecting rod ab can move up and down and deflect relative to the right threaded rod 21, and point b is located below point a.

[0030] The two end points of the right second connecting rod cd are point c and point d, wherein point d is hinged to one end of the beam 1, and point c is hinged to the right second screw rod 22, so that the right second connecting rod cd can move up and down and the right second connecting rod cd can deflect relative to the right second screw rod 22, and point c is located below point d.

[0031] The two end points of the left connecting rod ef are point e and point f, wherein point f is hinged at one end of the beam 1, and point e is hinged on the left threaded rod 11, so that the left connecting rod ef can move up and down and deflect relative to the left threaded rod 11, and point f is located below point e.

[0032] The two end points of the left second connecting rod gh are point g and point h, wherein point g is hinged to one end of the crossbeam 1, and point h is hinged to the left second screw rod 12, so that the left second connecting rod gh can move up and down and the left second connecting rod gh can deflect relative to the left second screw rod 12, and point h is located below point g.

[0033] The left end of the beam 1 connects point f of the left first connecting rod ef and point g of the left second connecting rod gh, and the right end connects point d of the right second connecting rod cd and point b of the right first connecting rod ab. Points f, g, d, and b are on a straight line, which is parallel to the length direction of the beam 1. When the length direction of the beam 1 is perpendicular to the length direction of each screw rod, the straight lines where the right first connecting rod ab, the right second connecting rod cd, the left first connecting rod ef, and the left second connecting rod gh are respectively parallel to the right screw rod 21, the right second screw rod 22, the left first connecting rod 11, and the left second screw rod 12. The distance from point f to the center line is equal to the distance from point b to the center line, and the distance from point g to the center line is equal to the distance from point d to the center line.

[0034] Based on the above structure, each screw rod can drive the corresponding connecting rod to move, and the connecting rod drives the crossbeam to tilt and / or move horizontally.

[0035] The physical model of the legs and the virtual adjustment system constitute the mechanism of this solution.

[0036] S2: Establish the outrigger screw motion planner model, which includes the screw motion target position and motion speed.

[0037] The model established in steps S1 and S2 is a model-based reinforcement learning method, which uses existing models to learn and make decisions, and models the track adjustment process as a Markov decision process (MDP).

[0038] The state space S in the track adjustment process is the environmental observation data, which includes sensor parameters, device status and target parameters, among which: The sensor parameters include: the distance L between the left wire 11 and the right wire 21 1 , the distance L between the left two screw rods 12 and the right two screw rods 22 2 , the distance L between the two rails 2 3 , the height difference A between the two rails 2, the offset distance ε (the distance that the center point between the two rails 2 deviates from the center line), the inclination angle θ (the angle between the beam 1 and the horizontal plane); The device status includes: screw displacement s 1 、s 2 ,speed , , connecting rod deflection angle ~ ; Target parameters include: Target superelevation value , Target Offset .

[0039] S3: Randomly initialize the starting position state of the screw and the starting deflection state of the connecting rod, and perform random processing on the environmental observation data. By introducing a parameter randomization strategy, the generalization ability of the reinforcement learning model to the real construction environment is enhanced, effectively solving the gap problem between the simulation environment and the actual engineering.

[0040] S4: Through the reinforcement learning (RL) algorithm, the rail motion trajectory is selected according to the environmental observation data and reward prediction, and the screw motion planner is used to convert the rail kinematic data into motor control instructions to drive the outrigger adjustment. The reward data is updated based on the new environmental observation data using the reward function, and new actions are continuously output.

[0041] The reinforcement learning algorithm specifically adopts the deep deterministic policy gradient (DDPG) algorithm as the benchmark algorithm, and uses the experience replay mechanism. During the interaction between the intelligent agent and the environment, the experienced state, action, reward, next state and other information are stored in a replay buffer, and then small batches of data are randomly sampled from the experience pool for sampling training.

[0042] In step S4, the Actor-Critic architecture is used to separate the strategy learning (Actor) and the value estimation (Critic), and a deep neural network is used to approximate the strategy function and the value function. The specific steps are as follows: S4-1: Initialization state: Get the track initial state feature vector , including sensor parameters: , Screw state: displacement ,speed , , connecting rod angle , target parameters , .

[0043] S4-2: Action generation: Current network policy function through policy learning and according to the status Generate Action , where action Includes rail target position and movement speed instructions.

[0044] S4-3: Action execution and observation: Sent to the screw motion planner, converted into motor control execution instructions, and obtained: new state , that is, updated track parameters, instant rewards ,in Calculated by the reward function, it reflects the adjustment accuracy and the termination flag is_end. is_end is used to determine whether the target error range has been reached.

[0045] S4-4: Experience storage: convert five-tuple ) is stored in the experience replay pool .

[0046] S4-5: Network training: from Sample N samples in , where j represents the index of the jth sample randomly sampled from the experience replay pool. Update according to the following rules: Critic Network Update: Calculate target value :

[0047] Where Q' is the Critic target network, ω' is the Critic target network parameter, is the discount factor, generally ranging from 0.9 to 0.99, represents the next state of the jth sample, Indicates the next state feature extraction, is the policy function of the Actor target network.

[0048] By minimizing the mean square error loss Update the current parameters of Critic . represents the current state of the jth sample, Indicates the current state Feature extraction. Q is the current network.

[0049] Actor Network Updates: Using Policy Gradients Update the current network parameters of the Actor .in Indicates policy parameters The gradient of represents the gradient of action a. represents the policy gradient, According to the status Generated network policy function.

[0050] S4-6: Target network synchronization: Every C steps, the current network parameters of Critic or Actor are synchronized. Copy to target network ). Where C represents the update frequency of the target network parameters.

[0051] S4-7: Loop termination condition: If It is the termination state, that is, the track reaches the target accuracy, and the current training round ends; otherwise, it returns to step S4-2 to continue iteration.

[0052] S5: Repeat S4, iteratively update the Actor and Critic networks and store the experience in the experience pool until convergence, end the training and save the strategy as the rail motion trajectory.

[0053] S6, obtain the inclination angle θ of the beam 1 and the deflection angle of the screw rod of the field control system through the inclination sensor ~ The initial state and real-time state of the rail are obtained through the track inspection car. The parameters required for adjusting the elevation and level of the rail from the current state to the final state. Including the required offset distance of the rail center and Figure 2 The distance L between the left thread rod 11 and the right thread rod 21 1 The distance between the left two screw rods 12 and the right two screw rods 22 is L 2 ; The distance between the vertices of the two rails is L 3 ; The height difference between the two rails 2 on the left and right sides of the track is the superelevation value of the rail. The above data obtained by the track inspection trolley and the inclination sensor are used as the environmental data of the field control system. The pre-trained model in S5 is loaded into the field control system. The experience pool is updated according to the environmental data and the strategy is optimized online. The optimal strategy is saved and the saved strategy is used to select the rail motion trajectory.

[0054] S7: Calculate the moving distance of each connecting rod on the straight line where the screw is located, convert the kinematic data of the rail into the kinematic data of the four motors of the left and right screws through the screw motion planner, and control the rail arrangement to complete the superelevation adjustment and the horizontal adjustment with superelevation.

[0055] Among them, horizontal adjustment with superelevation refers to horizontal adjustment, because the superelevation value also changes during the horizontal adjustment process.

[0056] In step S7, the control rail arrangement completes superelevation adjustment and level adjustment with superelevation, including the following steps: S7-1: Calculate the moving distance of the left and right screw rods to complete the superelevation adjustment. After the adjustment, the superelevation value changes: In this step, this example assumes Figure 2 is the initial state, at which beam 1 is parallel to the horizontal plane and the superelevation value is 0. The target state is Figure 3 At this time, the left rail 2 is higher than the right rail 2, and the superelevation value is A, that is, the superelevation value of the two rails 2 that needs to be adjusted is Amm. The four screw rods of the rail row need to be adjusted to achieve superelevation adjustment. First, calculate the inclination angle of the beam 1 that needs to be adjusted: ; Among them, A is the superelevation value of the left and right rails, is the distance between the two rail vertices.

[0057] The track is deflected around the center line, and the speed planning needs to meet the following requirements: ; in is the movement speed of the left thread rod 11, is the movement speed of the left second screw rod 12, is the movement speed of the right thread rod 21, is the movement speed of the right second screw rod 22, is the distance between the left thread rod 11 and the right thread rod 21, L 2 It is the distance between the left two screw rods 12 and the right two screw rods 22.

[0058] The displacement of the left screw 11 and the right screw 21 for: ;

[0059] Wherein T is the time of super-high adjustment, and the left thread rod 11 and the right thread rod 21 move in opposite directions.

[0060] The displacement of the left two screw rods 12 and the right two screw rods 22 for: ; The left two screw rods 12 and the right two screw rods 22 move in opposite directions.

[0061] After completing the super-high adjustment, the track mechanism Figure 3 shown.

[0062] S7-2: Calculate the moving distance of each connecting rod on the straight line where the lead screw is located to adjust the level of the belt superelevation. This step includes the following steps: S7-2-1: Calculate the deflection angle of the connecting rod after super-height adjustment: Calculate the deflection angles of the right first connecting rod ab, the right second connecting rod cd, the left first connecting rod ef, and the left second connecting rod gh after the super-height adjustment is completed, respectively. , , , .

[0063] beg :Depend on Figure 3 It can be seen that: ; ; Therefore: ; in, is the vertical distance from point b in the right connecting rod ab to the right thread 21, It represents the distance from point b in the right connecting rod ab to the center point Z of the track, where point Z is the midpoint of the straight line bf. To complete the deflection angle of the right connecting rod ab after super-high adjustment, Indicates that the right connecting rod ab is long. Line Zp is parallel to the horizontal plane, and line bp is perpendicular to line Zp.

[0064] beg :Depend on Figure 3 It can be seen that: ; ; Therefore: ; in, is the vertical distance from point d in the right second connecting rod cd to the right second screw rod 22, It represents the distance from point d in the right second link cd to the center point Z of the track. To complete the deflection angle of the right second link cd after super-high adjustment, Point t is a point on the right second screw rod 22 , and the straight line dt is perpendicular to the right second screw rod 22 .

[0065] beg : Wherein, point i is a point on the left thread 11, and the straight line fi is perpendicular to the left thread 11. Since the length of the left connecting rod ef is equal to that of the right connecting rod ab, and the vertical distance from point f in the left connecting rod ef to the left thread 11 is equal to the vertical distance from point b in the right connecting rod ab to the right thread 21. Similarly, we can get: ; beg , where point k is a point on the left second screw rod 12, and the straight line gk is perpendicular to the left second screw rod 12. Since the left second connecting rod gh and the right second connecting rod cd are of equal length, and the vertical distance from point g in the left second connecting rod gh to the left second screw rod 12 is equal to the vertical distance from point d in the right second connecting rod cd to the right second screw rod 22. Similarly, we can get: ; S7-2-2: Perform horizontal adjustment with superelevation. After the adjustment, the center points of the two rails are offset: like Figure 3 As shown, the relative superelevation value has been adjusted. In this implementation, assuming that the center position of the rail is biased to the left, the rail needs to be moved rightward as a whole along the figure, and the distance of the translation is At this time, the right connecting rod ab needs to move upward along the right thread rod 21. After completion, the vertical distance H between the right rail 2 and the ground remains unchanged.

[0066] Calculate the distance D that each connecting rod moves on the straight line where the screw rod is located1 , D 2 , D 3 , D 4 And the connecting rod deflection angle of the connecting rod during horizontal adjustment , , , : like Figure 4 In order to complete the track mechanism diagram with superelevation horizontal adjustment, for the right connecting rod ab:

[0067] By combining equations (1) and (2), we can obtain: ; ; Similarly, we can generalize the distance D that the right first connecting rod ab, the right second connecting rod cd, the left first connecting rod ef, and the left second connecting rod gh move on each screw rod respectively. 1 , D 2 , D 3 , D 4 And the deflection angle during horizontal adjustment , , , : ; ; in i =1,2,3,4. Indicates the length of each connecting rod ab, cd, ef, gh, Indicates the deflection angle of each connecting rod after the level adjustment with superelevation is completed. Indicates the offset distance of the center of rail 2.

[0068] S8, repeat S6 to S7 until the rail reaches the target point within the allowable error range.

[0069] Finally, it is necessary to explain here that the above embodiments are only used to further illustrate the technical solution of the present invention in detail and cannot be understood as limiting the scope of protection of the present invention. Some non-essential improvements and adjustments made by technicians in this field based on the above content of the present invention all fall within the scope of protection of the present invention.

Claims

1. A track arrangement fine-tuning algorithm for a twin-block ballastless track, characterized in that: The steps include: S1: Establishing a physical model of a pair of legs of a rail track, and constructing a virtual adjustment system based on the physical model, wherein the physical model of the legs comprises a crossbeam (1) and two steel rails (2), the two steel rails (2) are arranged in parallel and supported on the crossbeam (1), and the virtual adjustment system comprises four connecting rods and four screw rods, each end of the crossbeam (1) is connected to two connecting rods, the four screw rods respectively drive one end of the four connecting rods to linearly reciprocate, and the four connecting rods drive the crossbeam (1) hinged to the other end thereof to rotate and / or move; S2: Establishing a motion planner model for the outrigger screw, wherein the planner model includes a motion target position and a motion speed; S3: Randomly initialize the starting position state of each connecting rod on the lead screw and the starting deflection state of the connecting rod, and perform random processing on the environmental observation data; S4: Through the reinforcement learning algorithm, the rail motion trajectory is selected according to the environmental observation data and reward prediction, and the screw motion planner is used to convert the rail kinematic data into motor control instructions for controlling the screw movement, drive the outrigger adjustment, and update the reward data based on the new environmental observation data using the reward function, and continue to output new actions; S5: Repeat step S4, iteratively update the strategy learning and value estimation network and store the experience in the experience pool until convergence, end the training and save the strategy as the rail motion trajectory; S6: Obtain the initial state and real-time state of the beam tilt angle and the screw deflection angle of the field control system through the tilt sensor; S7: Calculate the moving distance of each connecting rod on the straight line where the screw rod is located, convert the kinematic data of the rail into the kinematic data of each screw rod through the screw rod motion planner, and control the rail arrangement to complete the adjustment; S8: Repeat steps S6 to S7 until the rail reaches the allowable error range of the target point.

2. The track arrangement fine-tuning algorithm according to claim 1, characterized in that: In step S1, the four connecting rods are respectively a right first connecting rod (ab), a right second connecting rod (cd), a left first connecting rod (ef), and a left second connecting rod (gh). The four screw rods are arranged in parallel and spaced apart, and are respectively a left first connecting rod (11), a left second connecting rod (12), a right second connecting rod (22), and a right first connecting rod (21) arranged in sequence in one direction. The left first connecting rod (ef), the left second connecting rod (gh), the right first connecting rod (ab), and the right second connecting rod (cd) are respectively connected to the left first connecting rod (11), the left second connecting rod (12), the right first connecting rod (21), and the right second connecting rod (22). The upper end of the right first connecting rod (ab) is connected to the right first connecting rod (21), and the lower end is connected to the crossbeam (1 ), the lower end of the right second connecting rod (cd) is connected to the right second screw rod (22), and the upper end is connected to the cross beam (1), the upper end of the left first connecting rod (ef) is connected to the left first screw rod (11), and the lower end is connected to the cross beam (1), the lower end of the left second connecting rod (gh) is connected to the left second screw rod (12), and the upper end is connected to the cross beam (1), there is a center line, the left first screw rod (11) and the right first screw rod (21) are symmetrical with respect to the center line, the left second screw rod (12) and the right second screw rod (22) are also symmetrical with respect to the center line, the left first connecting rod (ef) and the right first connecting rod (ab) are symmetrical with respect to the center line, and the left second connecting rod (gh) and the right second connecting rod (cd) are also symmetrical with respect to the center line.

3. The track arrangement fine-tuning algorithm according to claim 2, characterized in that: The environmental observation data during the track adjustment process include: sensor parameters, device status and target parameters, wherein the sensor parameters include: the spacing L1 between the left threaded rod (11) and the right threaded rod (21), the spacing L2 between the left second threaded rod (12) and the right second threaded rod (22), the spacing L3 between the two rails (2), the height difference, offset distance and tilt angle between the two rails (2), wherein the offset distance refers to the distance that the center point between the two rails (2) deviates from the center line, and the tilt angle refers to the angle between the crossbeam (1) and the horizontal plane; the device status includes: the displacement and speed of each connecting rod on the straight line where the corresponding threaded rod is located, and the connecting rod deflection angle; the target parameters include: the target superelevation value and the target offset distance, wherein the superelevation value refers to the height difference between the two rails (2).

4. The track arrangement fine-tuning algorithm according to claim 3 is characterized in that: Step S4 includes the following steps: S4-1: Initialization state: obtain the initial state feature vector of the track, including sensor parameters, device state, and target parameters; S4-2: Action generation: Generate actions according to the current state through the current network strategy function in strategy learning, where the actions include rail target position and movement speed instructions; S4-3: Action execution and observation: Send the action to the screw motion planner, convert it into motor control execution instructions, and get the new state; S4-4: Experience storage: The five-tuple consisting of the current state, action, immediate reward, new state, and termination flag is stored in the experience replay pool. The immediate reward is calculated by the reward function to reflect the adjustment accuracy, and the termination flag is used to determine whether the target error range has been reached; S4-5: Network training: sampling multiple samples from the experience replay pool; S4-6: Target network synchronization: copy the current network parameters of value estimation or policy learning to the target network every C steps, where C represents the update frequency of the target network parameters; S4-7: Loop termination condition: If the new state is the termination state, that is, the track reaches the target accuracy, the current training round ends; otherwise, return to step S4-2 to continue iteration.

5. The track arrangement fine-tuning algorithm according to claim 3 is characterized in that: Step S7 includes the following steps: S7-1: perform super-height adjustment, after which the super-height value changes; S7-2: Perform horizontal adjustment with superelevation, after which the center point between the two rails is offset.

6. The track arrangement fine-tuning algorithm according to claim 5, characterized in that: In step S7-1, the inclination angle of the beam (1) that needs to be adjusted is calculated based on the superelevation value that needs to be adjusted and the distance between the two rails, and the displacement that each connecting rod needs to complete on the corresponding screw rod is calculated based on the relationship between the linear moving speeds of the two connecting rods on the same side of the center line.

7. The track arrangement fine-tuning algorithm according to claim 5, characterized in that: Step S7-2 includes the following steps: S7-2-1: Calculate the deflection angle of each connecting rod after superelevation adjustment; S7-2-2: Calculate the displacement that each connecting rod needs to complete on the corresponding screw rod according to the target deflection angle of each connecting rod, the target offset distance, and the deflection angle obtained in step S7-2-1.

8. The track arrangement fine-tuning algorithm according to claim 6, characterized in that: In step S7-1: ； ； ； ； ； ； Among them: A is the ultra-high value that needs to be adjusted, is the inclination angle of the beam (1) to be adjusted, is the movement speed of the left threaded rod (11), is the movement speed of the right threaded rod (21), is the movement speed of the left second screw rod (12), is the movement speed of the right second screw rod (22), is the displacement that the left thread rod (11) and the right thread rod (21) need to move, is the displacement that the left second lead screw (12) and the right second lead screw (22) need to move, and T is the time for super-high adjustment.

9. The track arrangement fine-tuning algorithm according to claim 7, characterized in that: In step S7-2-2: ； ； in i =1,2,3,4, represents the length of each connecting rod, Represents the displacement of each connecting rod on the corresponding screw rod, It indicates the deflection angle of each connecting rod relative to the initial state after the horizontal adjustment with superelevation is completed. represents the offset distance between the center points of the two rails (2), Indicates the deflection angle after superelevation adjustment relative to the superelevation adjustment.

Images