深度学习增强的智能矿用电铲挖掘轨迹跟踪控制  

Deep Learning-enhanced Intelligent Electric Shovel Digging Trajectory Tracking Control

作　　者：付涛 张天赐 崔允浩 宋学官 FU Tao;ZHANG Tianci;CUI Yunhao;SONG Xueguan(School of Mechanical Engineering,Dalian University of Technology,Dalian 116023)

机构地区：[1]大连理工大学机械工程学院,大连116023

出　　处：《机械工程学报》2024年第16期357-366,共10页Journal of Mechanical Engineering

基　　金：国家自然科学基金(52075068);山西省科技重大专项(20191101014)资助项目。

摘　　要：矿用电铲是露天开采中重要的采装设备之一,用于地表覆盖物的剥离、矿石物料的装载。随着智慧矿山建设的不断推进,矿用电铲无人化、智能化成为发展趋势。不同于传统的人工操作,智能电铲采用自主作业方式,工作流程主要包括环境感知、挖掘轨迹规划与轨迹跟踪。挖掘轨迹跟踪控制的精度直接决定了作业效率和质量,对于智能化矿用电铲的开发具有重要意义。矿用电铲具有机身结构庞大、运行惯量大、作业环境复杂,挖掘过程中外部载荷剧烈变化等特点,使得模型存在较强的非线性和不确定性,基于传统线性反馈控制器开展智能矿用电铲挖掘轨迹跟踪存在控制滞后、精度低等问题。因此,提出一种深度学习增强的跟踪控制策略,利用深度长短时记忆神经网络建模电铲系统固有的逆动态响应特性,将规划的最优挖掘轨迹变换为可以减少跟踪误差的参考轨迹作为控制系统的输入。深度学习增强的智能矿用电铲轨迹跟踪算法不需要访问电铲内部控制回路,也不需要显式建模电铲动力学,在实践中具有可行性。为了验证控制策略的有效性,在由真实矿场物料搭建的试验场地下,利用1∶7智能电铲缩比样机开展对比试验研究。结果表明深度学习增强的挖掘轨迹跟踪精度更优,其中MAE降低了28.67%,RMSE降低了12.9%,MAPE降低了7.83%。Electric shovel(ES)is one of the most important production equipment used in the open-pit mining to peel off the surface cover and load ore materials.With the advancement of the construction of smart mines,unmanned and intelligent has become the development trend of the ES.Different from the traditional manual operation,the intelligent ES adopts an autonomous operation method,and the work process mainly includes environment perception,trajectory planning and trajectory tracking.And the accuracy of digging trajectory tracking control directly determines the efficiency and quality of the operation,which is of great significance to the development of intelligent ES.Due to the huge structure,the large operating inertia,the complex working environment,and the drastic change of external load in the excavation process,etc.,the model of ES shows strong nonlinearity and uncertainty.And there are some problems,such as control lag and low precision in carrying out digging trajectory tracking based on traditional linear feedback controller.Thus,in this study,a deep learning-enhanced tracking control strategy,which uses deep long and short-term memory neural networks(LSTM)to model the inherent inverse dynamic response characteristics of the ES system,and transforms the planned optimal digging trajectory into a reference trajectory that can reduce tracking errors as a control system input,is presented.The proposed approach does not need to access the internal control loop,nor does it need to explicitly model the dynamics of the ES,which is feasible in practice.To verify the effectiveness of the control strategy,a 1∶7 intelligent ES scaled prototype was used to carry out a comparative experimental study in an experimental site constructed from real mine materials.And experimental results show that the control strategy enhanced by deep learning improves the accuracy of digging trajectory tracking,in which MAE is reduced by 28.67%,RMSE is reduced by 12.9%,and MAPE is reduced by 7.83%.

关 键 词：智能矿用电铲 挖掘轨迹 跟踪控制 长短时记忆网络 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]