基于CNN-LSTM的挖掘机作业阶段感知

Working Stage Perception of Excavator Based on CNN-LSTM

挖掘机是工程建设项目中最常用的设备。挖掘机通常以重复循环的形式进行作业,精确识别挖掘机所处的不同作业阶段在挖掘机分阶段功率匹配中具有重要的作用。针对挖掘机循环作业阶段识别精度较低的问题,构建了一种基于卷积神经网络(Convolutional Neural Network, CNN)和长短期记忆神经网络(Long Short Term Memory,LSTM)相结合,并利用从挖掘机CAN总线中读取多路阀的先导压力信号作为输入特征向量的循环作业阶段识别模型。CNN可以提取各种信号中的高级特征,LSTM可结合信号不同时期的关联性,提高阶段识别的准确率。仿真结果表明,相较LSTM,CNN-LSTM的准确率由93.39%提升至96.09%,且各阶段的F1分数均有所提高。

Excavators are the most commonly used equipment in engineering construction projects.The excavator usually works in the form of repeating cycle.The accurate identification of different working stages of the excavator plays an important role in the power matching of the excavator in different stages.Aiming at the problem of low recognition accuracy in the cyclic operation stage of excavators,a Convolutional Neural Network(CNN)and Long Short Term Memory(LSTM)are integrated to solve the problem.The pilot pressure signal of multiway valve is read from CAN bus of excavator as input feature vector.CNN can extract advanced features in various signals,and LSTM can combine the correlation of signals in different periods to improve the accuracy of stage recognition.The simulation results show that compared with LSTM,the accuracy of CNN-LSTM is improved from 93.39%to 96.09%,and the F1 scores of each stage are improved.