Abstract: To solve the problems of low accuracy using single sensor for cutting loads of shearers and improve the sensing ability of shearer running safety states, based on the applications of DBN, BR and PSO algorithm, a multi-sensor shearer drum load identification and prediction method was proposed. BR-PSO-DBN prediction model was constructed using 22 sets of test data monitored by idler shaft sensors, connecting frame pin shaft sensors and rocker strain sensors as input samples to predict the three directional loads and torque of drum cutting. Based on the experimental system of cutting loads, the field test was carried out, and the measured results were compared with the predicted ones. The results show that the measured curves are basically the same as the predicted curves, and the peak points correspond to each other. The prediction accuracy of the model for the three-way cutting loads of the drum is more than 83%, and the prediction accuracy of the roller torque is 95%, which indicates that the prediction model has high prediction accuracy and may provide reference for field applications to guide safety productions.

Key words: shearer, load prediction, deep belief network(DBN), Bayesian regularization(BR), improved particle swarm optimization(PSO)

摘要： 为解决单一传感信息对采煤机截割载荷预测精度低的问题，提高采煤机运行安全状态的感知能力，在应用深度信念网络（DBN）的基础上，引入贝叶斯正则化（BR）方法和粒子群优化（PSO）算法，提出了一种基于多传感器的采煤机滚筒载荷辨识与预测方法。构建BR-PSO-DBN预测模型，以惰轮轴传感器、连接架销轴传感器和摇臂应变传感器监测的22组测试数据为输入样本，预测滚筒截割三向载荷与扭矩。依托截割载荷实验系统进行现场测试，并将实测结果与预测结果进行对比分析，结果表明：实测曲线与预测曲线变化规律基本相同，峰值点相对应，模型对滚筒三向截割载荷的预测精度达到83%以上，其中对滚筒扭矩预测精度达到95%，说明预测模型具有较高的预测精度，能够为现场应用指导安全生产提供参考。

关键词: 采煤机, 载荷预测, 深度信念网络, 贝叶斯正则化, 改进粒子群优化