CN-121756364-B - Train unhooking control dynamic optimization method and system

CN121756364BCN 121756364 BCN121756364 BCN 121756364BCN-121756364-B

Abstract

The invention provides a dynamic optimization method and a dynamic optimization system for train unhooking control, which relate to the technical field of robot control, and the method comprises the following steps: according to front end sensing collection of the unhooking robot, multi-mode sensing data are obtained and mapped to a high-dimensional state embedded manifold in the intelligent control unit, manifold positioning and similarity retrieval based on the high-dimensional state embedded manifold are executed, a unhooking control strategy is generated, a continuous control track of the unhooking control strategy is analyzed into a motor instruction sequence of a robot joint, and the motor instruction sequence is issued to a control end of the unhooking robot to perform automatic unhooking driving management. The technical problems of improving the automation level of unhooking operation and avoiding unsafe factors and low efficiency caused by human factors are solved. The technical effects of reducing manual intervention and improving unhooking operation efficiency, accuracy and safety through automatic intelligent analysis control are achieved.

Inventors

WU JINFENG
CHEN QI
SHI QUAN
ZHANG CHUNLEI
LIU CHANGXU
WANG GUOQIANG
BAI FEIYU
LU CHUNLIN
ZHANG HAIFENG

Assignees

大唐吉林发电有限公司辽源发电分公司

Dates

Publication Date: 20260508
Application Date: 20260303

Claims (9)

1. A method for dynamically optimizing train unhooking control, the method comprising: Acquiring multi-mode sensing data according to front end sensing acquisition of the unhooking robot; Transmitting the multi-mode sensing data to an intelligent control unit embedded in the central control of the unhooking robot, and mapping the multi-mode sensing data to a high-dimensional state embedded manifold in the intelligent control unit, wherein points in the high-dimensional state embedded manifold represent operation scenes, and geodesic distances among the points represent operation similarity among the scenes; The intelligent control unit executes manifold positioning and similarity retrieval based on the high-dimensional embedded manifold to generate an unhooking control strategy, wherein strategy generation is performed by solving a continuous track from a real-time positioning state to a successful unhooking target area; Analyzing the continuous control track of the unhooking control strategy into a motor instruction sequence of a robot joint, and sending the motor instruction sequence to a control end of the unhooking robot to perform automatic unhooking driving management; before mapping to a high-dimensional state embedded manifold in an intelligent control unit, the building of the high-dimensional state embedded manifold comprises: acquiring historical unhooking operation data, marking working condition categories and operation result labels by data sequences, and generating a triplet sample, wherein the triplet sample comprises an anchor point sample, a positive sample and a negative sample; constructing a first network layer and executing supervision training based on the triplet sample, wherein the first network layer takes multi-mode data class as input and takes low-dimensional embedded vector with fixed length as output; And inputting the historical unhooking operation data into a first network layer of training, obtaining sample embedding vectors, and forming the high-dimensional state embedding manifold, wherein the high-dimensional state embedding manifold characterizes a continuous distribution structure formed by the sample embedding vectors in a high-dimensional space.
2. The train unhooking control dynamic optimization method according to claim 1, wherein in the monitoring training process based on the triplet sample, a training convergence condition is set by minimizing a distance between an anchor sample and a positive sample embedded vector and taking a distance between the anchor sample and a negative sample embedded vector as a training convergence condition, wherein the distance between the anchor sample and the negative sample embedded vector is at least greater than a preset interval.
3. The method for dynamically optimizing train unhooking control according to claim 2, wherein performing manifold positioning and similarity retrieval based on the high-dimensional state embedded manifold comprises: the first network layer obtains a real-time state coordinate point by mapping the multi-mode sensing data to the high-dimensional state embedded manifold; Calculating the geodesic distance between the real-time state coordinate point and all the state points marked with successful unhooks on the high-dimensional state embedded manifold, and searching out K nearest successful points, wherein the coordinate safety area constraint is carried out according to the preset distance of the boundary of the failed cluster; And taking the control strategies corresponding to the K success points as an initial strategy reference set, wherein strategy elements of the initial strategy reference set comprise track information, force control parameters and execution time sequence characteristics.
4. A method of dynamically optimizing train unhooking control as claimed in claim 3, wherein generating the unhooking control strategy comprises: extracting statistical distribution of strategy elements aiming at the initial strategy reference set; Defining an optimization function, wherein the optimization function is defined based on whether the tail end of the track falls into a successful unhooking target area, track smoothness and the coincidence degree of the predicted contact force and the reference force profile; and executing track generation iteration under gradient optimization based on the statistical distribution and the optimization function, and determining the unhooking control strategy.
5. The dynamic optimization method for train unhooking control according to claim 4, wherein the unhooking task is decomposed into task semantic units, wherein the task semantic units at least comprise precisely positioning grabbing points, applying vertical unlocking force and executing rotation separation actions; Mapping each task semantic unit into a group of adjustable parameterized motion primitives and a force control template, determining a mapping relation, and taking the mapping relation as a constraint condition in the track generation iteration process.
6. The dynamic optimization method for train unhooking control according to claim 1, wherein the parsing the continuous control track of the unhooking control strategy into a motor instruction sequence of a robot joint comprises: Acquiring a kinematic mechanism of the unhooking robot, and converting the pose sequence of the end effector in the continuous control track in the unhooking control strategy into a driving time sequence of each joint angle through inverse kinematic analysis; And converting the driving time sequence into motor servo instructions and sending the motor servo instructions to a control end of the unhooking robot, wherein each motor servo instruction comprises a position, speed and moment instruction sequence of each joint servo motor.
7. A method for dynamically optimizing train unhooking control as recited in claim 1, wherein said method comprises: a data interface for data interaction with the vehicle dumper central control is established in the unhooking robot central control, and a real-time external signal is received, wherein the real-time external signal at least comprises a carriage positioning signal and a unhooking instruction window signal; Generating an unhooking operation starting instruction meeting multiple conditions by fusing real-time external signals and multi-mode sensing data according to a flow state machine, wherein the multiple conditions comprise that a unhooking instruction window is effective, a target carriage is positioned, and a unhooking robot is in a ready state; After unhooking is completed, the process state machine feeds back a unhooking result to the car dumper for control, wherein the unhooking result is marked with a timestamp.
8. The dynamic optimization method for train unhooking control according to claim 1, wherein the automatic unhooking driving management is performed, comprising: Synchronously carrying out unhooking control monitoring to obtain sensing feedback data; Mapping the sensing feedback data to a high-dimensional state embedded manifold, comparing the high-dimensional state embedded manifold with expected state coordinates based on the unhooking control strategy, and determining manifold drift amount; when the manifold drift exceeds the deviation threshold, repositioning the real-time state on the high-dimensional embedded manifold, triggering small-scale re-optimization, generating a corrected control track segment, converting the corrected control track segment into a motor servo instruction, and transmitting the motor servo instruction to a control end of the unhooking robot.
9. A train unhooking control dynamic optimization system characterized by the steps for implementing a train unhooking control dynamic optimization method according to any one of claims 1 to 8, said train unhooking control dynamic optimization system comprising: the data acquisition module is used for acquiring multi-mode sensing data according to front end sensing acquisition of the unhooking robot; the data mapping module is used for transmitting the multi-mode sensing data to an intelligent control unit embedded in the central control of the unhooking robot, and mapping the multi-mode sensing data to a high-dimensional state embedded manifold in the intelligent control unit, wherein points in the high-dimensional state embedded manifold represent operation scenes, and geodesic distances among the points represent operation similarity among the scenes; The intelligent control unit is used for performing manifold positioning and similarity retrieval based on the high-dimensional embedded manifold to generate an unhooking control strategy, wherein strategy generation is performed by solving a continuous track from a real-time positioning state to a successful unhooking target area; And the unhooking driving management module is used for analyzing the continuous control track of the unhooking control strategy into a motor instruction sequence of the robot joint, and sending the motor instruction sequence to a control end of the unhooking robot to perform automatic unhooking driving management.

Description

Train unhooking control dynamic optimization method and system Technical Field The invention relates to the technical field of robot control, in particular to a dynamic optimization method and system for train unhooking control. Background In a power plant, train transportation coal is always the main supply mode, and a dumper is used as important unloading equipment, so that the automation of processes such as dumping operation, wagon transfer and the like is realized, and the unloading efficiency is greatly improved. However, the unhooking operation of the tippler still depends on manual operation, which has a plurality of problems in practical application. The manual unhooking operation requires a certain skill and experience of the operator and a high degree of synchronization with the operation rhythm of the dumper. Although experienced staff can operate skillfully, new replacement staff or unskilled staff often have the problems of incomplete action, delay and the like in the unhooking process, so that the working efficiency is low. In addition, manual operation also brings potential safety hazard. Because unhooking operation is closely related to the overturning rhythm of the car dumper, once operators fail to timely finish unhooking, the car dumper can be stopped, and the overall operation efficiency is affected. More seriously, the manual operation has a large safety risk, and particularly, personnel injury or accident easily occurs when the operation is performed in a high-risk area. Therefore, how to improve the automation level of unhooking operation, avoiding unsafe factors and inefficiency caused by human factors is a key technical problem existing in the prior art. Disclosure of Invention The application aims to provide a dynamic optimizing method and a dynamic optimizing system for train unhooking control, which are used for solving the technical problems of improving the automation level of unhooking operation and avoiding unsafe factors and low efficiency caused by human factors. In view of the above problems, the application provides a dynamic optimization method and a dynamic optimization system for train unhooking control. The train unhooking control dynamic optimization method comprises the steps of acquiring multi-mode sensing data according to front end sensing acquisition of a unhooking robot, transmitting the multi-mode sensing data to an intelligent control unit embedded in a middle control of the unhooking robot, mapping the intelligent control unit to a high-dimensional state embedded manifold in the intelligent control unit, wherein points in the high-dimensional state embedded manifold represent operation scenes, geodesic distances among the points represent operation similarity among the scenes, performing manifold positioning and similarity retrieval based on the high-dimensional state embedded manifold by the intelligent control unit, generating a unhooking control strategy, analyzing the continuous control track of the unhooking control strategy into a motor instruction sequence of a robot joint, and sending the motor instruction sequence to a control end of the unhooking robot for automatic unhooking driving management. Optionally, acquiring historical unhooking operation data, marking working condition types and operation result labels according to data sequences, generating a triplet sample, constructing a first network layer, and performing supervision training based on the triplet sample, wherein the first network layer takes multi-mode data types as input and takes low-dimensional embedded vectors with fixed lengths as output, inputting the historical unhooking operation data into the first network layer for training, acquiring sample embedded vectors, and forming the high-dimensional state embedded manifold, and the high-dimensional state embedded manifold characterizes continuous distribution structures formed by the sample embedded vectors in a high-dimensional space. Optionally, in the supervised training process based on the triplet samples, the distance between the anchor point sample and the positive sample embedded vector is minimized, and the distance between the anchor point sample and the negative sample embedded vector is at least greater than a preset interval to be used as a training convergence condition. The method comprises the steps of obtaining a high-dimensional state embedded manifold, obtaining a real-time state coordinate point by mapping multi-mode sensing data to the high-dimensional state embedded manifold, calculating the geodesic distance between the real-time state coordinate point and all the state points marked with successful unhooks on the high-dimensional state embedded manifold, searching out K success points closest to the real-time state coordinate point, carrying out coordinate safety area constraint according to preset distances of failure cluster boundaries, and taking a control strategy corresponding to the K success points as