A train gap mark adaptive compensation control system of an urban rail transit signal system

Abstract

The application discloses a kind of urban rail transit signal system train gap mark adaptive compensation control systems, it is related to urban rail transit train operation control technical field.System includes multi-source data acquisition unit, gap mark state identification unit, adaptive compensation controller, compensation instruction output unit, man-machine interaction unit and data record and self-learning unit.Multi-source data acquisition unit real-time acquisition train dynamic motion data, positioning calibration data, brake system data, vehicle state data, man-machine interaction data and line static data;Gap mark state identification unit calculates parking deviation and predicts final deviation, and uses classifier to identify the leading cause of gap mark online;Adaptive compensation controller executes differentiation compensation for different leading causes, including speed positioning compensation, electric-pneumatic brake coordination compensation, ATO parameter adaptive optimization and driver misoperation protection;Data record and self-learning unit uses historical data to update model parameters periodically offline.The application realizes parking deviation ≤±0.2m under all working conditions, without large-scale hardware modification, compatible with existing CBTC hardware platform, with the advantages of high control precision, strong adaptability, good engineering economy.

Description

Technical Field

[0001] This invention belongs to the field of urban rail transit train operation control technology, specifically relating to an adaptive compensation control system for the problem of over- or under-stopping during automatic train operation (ATO) mode in a signaling system based on vehicle-to-ground wireless communication (CBTC). Background Technology

[0002] The ATO subsystem of the CBTC signaling system in urban rail transit has stringent requirements for train stopping accuracy: a stopping probability of no less than 99.99% within ±0.3m and no less than 99.9998% within ±0.5m. As a typical moving block signaling system, the CBTC system obtains train position, speed, and movement authorization in real time through onboard controllers and trackside wireless communication, enabling automatic train operation. However, in actual operation, trains frequently experience "overshooting" (passing the stop marker) or "undershooting" (failing to reach the stop marker) when stopping at platforms, directly causing inaccurate alignment between the train doors and platform screen doors. In severe cases, manual intervention is required for secondary alignment, significantly reducing operational efficiency and affecting passenger comfort. In-depth analysis reveals that the root causes of the underperformance of bids can be mainly categorized into the following five types: (1) Speed ​​measurement and positioning error: The wheel axle speed sensor is susceptible to idling, coasting and wheel diameter wear; the odometer cumulative error gradually increases after long-distance operation; after the vehicle positioning equipment is initialized or reset, it needs to go through two consecutive transponders to complete the positioning recovery, during which there is position uncertainty. (2) Nonlinear response of braking system: There is a dead zone or response lag when switching between electric braking and air braking. Especially in the low speed range (usually below 5km / h), after the electric braking is disengaged, the air braking is not established in time or the pressure is over-over-rushed, resulting in fluctuations in braking force. (3) Fixed ATO control parameters: Traditional ATO uses a fixed braking start point and a fixed deceleration curve, which cannot adapt to changes in train load, wheel-rail adhesion coefficient and braking system performance degradation. (4) External environmental interference: Rain and snow weather can cause the track to become slippery, and the additional resistance of slopes and curves, as well as the piston wind in tunnels, can all affect the actual braking effect. (5) Human-machine interaction interference: In ATO mode, the driver accidentally touches the driver control device, causing the ATO command to be overridden by the manual command, resulting in an unexpected failure to meet the target. Existing related patents (such as CN202010712899, CN202311349880, etc.) are mostly focused on optimizing the parking accuracy of general CBTC systems, or adopt technical solutions such as PID / transitional algorithms and deep reinforcement learning, but they all have the following defects: ① They do not systematically improve the transponder positioning dependency mechanism, on-board-trackside cooperative control protocol, and electro-pneumatic braking matching characteristics unique to the signaling system; ② They do not provide an integrated solution that can identify the main causes of overshoot and undershoot online and dynamically switch compensation strategies, resulting in limited compensation effect and single applicable scenarios. Summary of the Invention

[0003] I. Purpose of the Invention This invention aims to overcome the problem of insufficient stopping accuracy in existing CBTC systems caused by factors such as speed measurement and positioning errors, nonlinear braking switching, fixed ATO parameters, and driver misoperation. It provides an adaptive compensation control system for train overshoot and undershoot in urban rail transit signaling systems, which can achieve a stopping deviation of ≤±0.2m under all working conditions, without the need for large-scale modification of existing signaling system hardware, thus taking into account both technical feasibility and engineering economy. II. Technical Solution Urban rail transit signaling system train overshoot / undershoot adaptive compensation control system An adaptive compensation control system for train overshoot / undershoot indicators in an urban rail transit signaling system includes a multi-source data acquisition unit, an overshoot / undershoot indicator status identification unit, an adaptive compensation controller, a compensation command output unit, a human-machine interaction unit, and a data recording and self-learning unit. The functions of each unit are as follows: 1. Multi-source data acquisition unit Communication connections are established with the onboard ATO unit, onboard ATP unit, wheel axle speed sensor, Doppler radar, accelerometer, transponder antenna, vehicle TCMS network, and driver controller status interface to collect the following data in real time: dynamic motion data (train wheel axle speed pulse, Doppler radar speed, three-axis acceleration); positioning calibration data (transponder ID and passing time); braking system data (traction / braking command level and actual feedback, including electric braking force and air brake cylinder pressure); vehicle status data (train load, wheel diameter wear, wheel flange lubrication status); human-machine interaction data (driver controller handle position and operation timestamp); and track static data (station kilometer markers, stop marker positions, gradient, curve radius, speed limit, pre-stored in the onboard database). 2. Over / Underpayment Status Identification Unit The system includes a parking deviation calculation subunit, a trend prediction subunit, and a cause classification subunit. Its specific functions are as follows: The parking deviation calculation subunit calculates the parking deviation ΔS = S_actual - S_target (positive values ​​indicate overshoot, negative values ​​indicate undershoot) based on the train's real-time positioning coordinates and the parking target coordinates; the trend prediction subunit predicts the final parking deviation ΔS_pred using a kinematic model (ΔS_pred = ΔS_curr + v² / (2a)) based on the current speed v, actual deceleration a, and remaining distance L; the cause classification subunit uses a pre-trained Bayesian classifier or decision tree to classify overshoot / undershoot events into one of the following five categories: ① speed measurement and positioning error-dominated; ② electro-pneumatic braking switching mismatch-dominated; ③ ATO parameter inappropriateness-dominated; ④ environmental interference-dominated; ⑤ driver misoperation-dominated. 3. Adaptive Compensation Controller The core execution unit of the system comprises four functional modules, each performing differentiated compensation for different primary causes: (1) Speed ​​and position compensation module: The unscented Kalman filter (UKF) is used to fuse wheel speed, Doppler radar speed and acceleration data to output the optimal speed / position estimate; when the freewheeling / coasting sign is detected, the radar speed or acceleration integral value is used as the reference for real-time correction; after passing a ground transponder, the odometer cumulative error is zeroed by using the absolute position of the transponder. (2) Electro-pneumatic braking coordination and compensation module: Real-time monitoring of the electric brake withdrawal time T_eb_off and the air brake establishment time T_ab_on, and calculation of the switching time difference Δt=T_ab_on-T_eb_off; If Δt> the set threshold (default 0.3s, which can be adjusted as needed), the air brake pre-charge command is sent to the vehicle TCMS Δt_pre (default 0.2s) in advance before the electric brake withdrawal, and the pre-charge time is dynamically adjusted according to the air brake cylinder pressure feedback to ensure continuous braking force. (3) ATO parameter adaptive module: Based on the historical parking deviation database and the current state vector (including train load, wheel diameter wear, wheel-rail adhesion coefficient, and track gradient), the radial basis function neural network (RBFNN) is used to optimize the ATO braking starting point distance D_brk and braking level curve coefficient K_brake online, and output the optimal control parameters adapted to the current working conditions. (4) Driver misoperation protection module: When the ATO mode is activated and the state of the driver controller handle changes, if the operation duration is less than 500ms and the change in level is less than 20% of the maximum commonly used braking level, it is determined to be a misoperation and the original ATO command output is maintained; if the operation duration is greater than 500ms or the change in level exceeds the threshold, it is determined to be a driver takeover and the switch to manual control mode is allowed. 4. Compensation command output unit Following the priority order of "emergency braking > ATP limiting command > ATO compensation command > driver operation command", the compensated traction / braking commands are sent to the train interface unit to ensure the safety and effectiveness of the control commands. 5. Human-Computer Interaction Unit (DMI) The system displays the following information to the driver in real time: current parking deviation (accurate to 0.01m), type of cause of misalignment, activation status of compensation strategy (activated / inactive), and suggested operation prompts (such as "Current braking parameters are being adaptively adjusted, please do not intervene"), thereby improving human-machine collaboration. 6. Data Recording and Self-Learning Unit The system stores complete data for each parking session (including time, station, train number, parking deviation, cause classification, compensation parameters, and actual control effect) to form a historical database. Periodically (every 500 parking sessions), the system uses a deep Q-network (DQN) or policy gradient method to retrain the parameters of the cause classifier and adaptive compensation controller offline, optimizes the model performance, and updates it to the onboard equipment after safety verification, thereby achieving continuous evolution of the system's control accuracy. III. Beneficial Effects 1. Comprehensive Cause Coverage and Intelligent Classification: For the first time, a feature-based classification-based online identification mechanism for the main causes of overshoot and undershoot is proposed for the unique mechanism of CBTC signal system. It accurately distinguishes five types of causes: speed measurement error, brake switching, parameter incompatibility, environmental interference, and misoperation, avoiding the problem of secondary effectiveness caused by "one-size-fits-all" compensation strategy, and making the compensation more targeted. 2. High positioning accuracy through multi-source fusion: By fusing wheel speed, radar and acceleration data through UKF and combining it with the absolute position of the transponder for periodic correction, the speed measurement and positioning error is reduced from the traditional meter level to within 0.05m, and is not affected by idling or coasting, providing a basic guarantee for high-precision parking. 3. Smooth transition between electric and air braking: The seamless transition between electric and air braking is achieved through dynamic adjustment of pre-charge pressure, completely eliminating low-speed undershoot caused by switching dead zone, which is especially suitable for line scenarios with low temperature in winter or aging braking system. 4. ATO parameter adaptive evolution: The braking curve is optimized online using RBFNN, which can automatically adapt to the decrease in train braking performance, changes in track adhesion and load fluctuations as the operating mileage increases, maintaining high-precision stopping over a long period of time without the need for frequent manual parameter calibration. 5. Intelligent protection against driver misoperation: Through a dual threshold judgment mechanism of "time + amplitude", it avoids parking deviations caused by unintentional touch of the driver's controller, while retaining the driver's emergency takeover operation authority, thus balancing operational safety and control efficiency. 6. Continuous learning capability: By periodically retraining offline using historical parking data, the system control accuracy continuously improves with the accumulation of operational data, achieving a self-evolving characteristic of "becoming more accurate with use". 7. Minimal engineering modifications: All improvements are concentrated at the vehicle software level, compatible with existing CBTC hardware platforms, requiring no replacement of hardware devices such as sensors and transponders, resulting in low promotion costs, short cycles, and easy engineering implementation. Detailed Implementation The technical solution of this invention will be described in detail below using a practical application example from a subway line: Application scenarios A certain subway line uses a CBTC signaling system. In ATO mode, the train continuously experienced substandard stopping accuracy at stations A, B, and C: before station A is a long downhill slope (gradient -30‰), with an average deviation of 0.4m over the standard; at station B, there is a lag in the switching between electric and air braking, with an average deviation of 0.28m under the standard; at station C, due to slippery tracks caused by rain and snow, the stopping deviation fluctuated significantly (±0.35m). Implementation process 1. System Deployment The adaptive compensation control software module of this invention is integrated into the on-board controller of the existing CBTC signaling system to complete the communication configuration between the multi-source data acquisition unit and the on-board ATO / ATP unit, TCMS network, and driver controller interface, and to load the line static database (including the location of each station stop sign, gradient curve, etc.). 2. A-site scenario (ATO parameter not being the dominant factor) The multi-source data acquisition unit acquires in real time: station A gradient -30‰, train load AW2 (fully loaded), wheel-rail adhesion coefficient estimate 0.25, and ATO default braking start point 75m; The under-caliber state identification unit calculates the current deviation ΔS_curr=+0.31m and predicts the final deviation ΔS_pred=+0.43m (exceeding the limit). The classifier outputs the dominant cause C=3 (ATO parameters are inappropriate). The ATO parameter adaptive module calls the RBFNN model, and outputs the optimal parameters after inputting the current state vector: braking start point D_brk_new=98m, braking level coefficient K_new=1.15 (15% higher than the default level). After the train applied the new braking curve, the actual stopping deviation was +0.06m, which meets the requirement of ≤±0.2m. 3. Bilibili scenario (electric-air handover mismatch-dominated type) The multi-source data acquisition unit monitored that: the electric brake withdrawal time T_eb_off=10.2s, the air brake establishment time T_ab_on=10.9s, and the switching time difference Δt=0.7s (exceeding the threshold of 0.3s). The final prediction deviation of the under-caliber state identification unit is ΔS_pred=-0.32m (exceeding the limit), and the main reason for the classifier output is C=2 (electric-air switching mismatch). The electric-air braking coordination and compensation module sends the air brake pre-charge command 0.4s in advance and dynamically adjusts the pre-charge time to keep the overlap time between electric braking and air braking at 0.2s. Actual braking effect: The braking force curve is uninterrupted, and the parking deficit has been improved from -0.28m to -0.04m. 4. C-station scenario (environmental interference-dominated) Data acquired by the multi-source data acquisition unit: wheel-rail adhesion coefficient of 0.18 and track gradient of 0‰ under rainy and snowy weather; The final prediction deviation of the under-caliber state identification unit is ΔS_pred = +0.38m (exceeding the limit), and the dominant cause of the classifier output is C = 4 (environmental interference). The adaptive compensation controller reduces the target braking deceleration to 0.7 m / s² and extends the braking initiation point to 85 m; The actual parking deviation was +0.12m, which meets the accuracy requirements. 5. Self-learning optimization After the data recording and self-learning unit has stored 1000 parking data points, the classifier and RBFNN controller are retrained offline using the DQN algorithm. After the updated model parameters passed the security verification, they were downloaded to the vehicle equipment. The average parking deviation across the entire line decreased from +0.32m before optimization to +0.05m, and the probability of parking within ±0.2m reached 99.96%. Implementation effect After three months of actual operation verification, the system of this invention effectively solved the problem of overshoot and undershoot in the CBTC signaling system of the line. The parking accuracy of each station was stably controlled within ±0.2m. There were no cases of misalignment between the train doors and the platform screen doors due to parking deviation, or manual recalibration. The operating efficiency was improved by 15%, and passenger satisfaction was significantly improved.

Claims

1. An adaptive compensation control system for train overshoot / undershoot in an urban rail transit signaling system, characterized in that, include: The multi-source data acquisition unit is used to acquire train speed measurement and positioning data, braking system status data, vehicle status data, driver control operation data, and track static data in real time. The overshoot / undershoot status identification unit is used to calculate real-time parking deviation, predict final parking deviation, and use a classifier to identify the dominant cause category of overshoot / undershoot online; An adaptive compensation controller, comprising at least a speed measurement and positioning compensation module, an electro-pneumatic braking coordination compensation module, an ATO parameter adaptive module, and a driver misoperation protection module, is used to perform differentiated compensation for different primary causes. The compensation command output unit is used to send the compensated control commands to the on-board ATO / ATP unit and vehicle TCMS of the signal system according to a preset priority. The human-computer interaction unit is used to display parking deviation, the type of the main cause, the compensation status, and suggested operation prompts in real time. The data recording and self-learning unit is used to store parking data and periodically update the classifier and controller parameters offline.

2. The system according to claim 1, characterized in that, The classifier in the overshoot / undershoot status identification unit uses a Bayesian classifier or a decision tree. Its input feature vector includes: wheel axle speed and Doppler radar speed deviation, electric braking and air braking switching time difference, ATO default braking start point, wheel-rail adhesion coefficient estimate, train load, track gradient, and driver control unit operation time. The output dominant cause category is one of the following five types: ① speed measurement and positioning error dominant type; ② electric-air braking switching mismatch dominant type; ③ ATO parameter inappropriateness dominant type; ④ environmental interference dominant type; ⑤ driver misoperation dominant type.

3. The system according to claim 1, characterized in that, The speed measurement and positioning compensation module uses an unscented Kalman filter (UKF) to fuse multi-source data from wheel axle speed sensors, Doppler radar, and accelerometers to output the optimal speed / position estimate. When a slip / tachycarding sign is detected, corrections are made based on the radar speed or acceleration integral value; after passing a ground transponder, the cumulative error is zeroed using the transponder's absolute position.

4. The system according to claim 1, characterized in that, The electro-pneumatic braking coordination and compensation module calculates the time difference Δt between the electric brake withdrawal time and the air brake establishment time in real time. When Δt > the preset threshold (0.3s), it outputs the air brake pre-charge command in advance, so that the electric brake and the air brake overlap by 0.1s to 0.3s in time, ensuring continuous braking force.

5. The system according to claim 1, characterized in that, The ATO parameter adaptive module uses a radial basis function neural network (RBFNN) with train load, wheel wear, wheel-rail adhesion coefficient, and track gradient as input vectors to output the optimal braking starting point distance and braking level curve coefficient online.

6. The system according to claim 1, characterized in that, The driver misoperation protection module filters the operation signal of the driver controller handle with a duration of <500ms and a level change of <20% of the maximum braking level, and maintains the ATO mode command output; when the operation duration is ≥500ms or the level change is ≥20% of the maximum braking level, it allows switching to manual control mode.