Deep Learning-enhanced Intelligent Electric Shovel Digging Trajectory Tracking Control
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％.
intelligent electric shoveldigging trajectorytracking controllong-short term memory
万方数据
维普
