基于增量学习的装载机铲装作业阻力预测技

Resistance Prediction Technology of Loader Shoveling Operation Based on Incremental Learning

中文关键词: 增量学习  装载机  作业阻力预测  Bi-LSTM-Inc模型

英文关键词: incremental learning  loader  prediction of operation resistance  Bi-LSTM-Inc model

中文摘要:

      针对装载机铲装过程中作业阻力影响因素众多且环境复杂多变，传统作业阻力预测模型的适用范围有限、动态预测效果不佳等问题，文中提出了一种基于增量学习的装载机铲装作业阻力预测技术。首先，建立装载机铲装作业阻力的双向长短期记忆（Bidirectional Long Short Term Memory, Bi-LSTM）模型作为增量学习的基础模型；然后，基于增量学习搭建模型动态更新框架，构建出能够适应新数据并持续更新的Bi-LSTM增量学习（Bi-LSTM-Incremental, Bi-LSTM-Inc）模型；最后，利用基于EDEM-RecurDyn软件的装载机铲装细沙物料联合仿真的数据进行模型算法的寻优，并利用装载机在相同工况下的在役运行数据进行实际测试验证。结果表明，基于增量学习的装载机铲装作业阻力预测技术能够有效降低预测误差，提高预测精度，作业阻力的动态预测效果好，为装载机无人自主作业提供了技术支撑。

英文摘要:

      The factors affecting the operation resistance of loaders is numerous and the environment is complex and everchanging, which limit the application scope of traditional resistance prediction model and cause its poor dynamic prediction effect. This paper proposes a new method for predicting the operation resistance of loaders based on incremental learning. Firstly, the Bi-LSTM model of resistance is established as the basic model of incremental learning. Secondly, a dynamic model update framework is constructed based on incremental learning to build a Bi-LSTM-Inc model that can adapt to new data and update continuously. Finally, the model algorithm is optimized by using the data of the co-simulation of loader shoveling fine sand materials based on EDEM-RecurDyn, and is tested and verified by using the in-service data of loader under the same operation conditions. The results show that the resistance prediction technology based on incremental learning can effectively reduce the prediction error and improve the prediction accuracy, and the dynamic resistance prediction has good effect, which provides technical support for the loader’s unmanned autonomous operation.