Research On Prediction And Early Warning Of Coal Shearer State Based On Deep Learning

As one of the three fully mechanized mining machines,the shearer is a large and complex machine that integrates mechanical,electrical and hydraulic pressure.Realizing the timely monitoring and early warning of the shearer status can not only effectively improve the maintenance efficiency of the shearer and in crease the economic benefits,but also has great significance for the safety production of the coal mine.In this paper,the data of the MG400/930-WD electric traction shearer at a fully mechanized coal mining face in Shaanxi Coal Group is used as the research object,and the prediction and early warning of the shearer status based on the deep learning method are studied.First,based on the basic structure and principle of the shearer,the common faults of the shearer are analyzed.From the four aspects of the cutting unit,the traction unit,th e electrical control unit and,other devices,ten types of monitoring data are selected to characterize the operating status of the shearer:the cutting motor temperature and the cutting motor current of the cutting unit;Traction motor current,traction motor temperature,traction motor speed;inverter current and transformer temperature of the electrical control unit;cooling water pressure of the hydraulic unit,hydraulic pressure increase system working pressure,increase pump motor-speed.Secondly,in view of the problem that traditional pretreatment methods cannot effectively distinguish between abnormal values that need cleaning and abnormal values that do not need cleaning,established a pretreatment combined cleaning model based on ARIMA.Data verification experiments show that the accuracy of the model is higher than the traditional moving average method,verifies the practicability of the established pretreatment cleaning model in cleaning outliers.After the data is cleaned by the pre-processing model,it is imported into 3 types of model recurrent neural network(RNN),long short-term memory network(LSTM)and gated recurrent unit(GRU)for tuning and training respectively,and the best performing GRU was selected through experiments.Based on the threshold setting of the state data of the shearer,the classification and early warning of the state data of the shearer are realized.The practicability of the GRU model is verified by the three evaluation indexes of the model:Mean Absolute Error(MAE),Root Mean Square Error(RMSE),Goodness of Fit(R2)and early warning accuracy.Finally,combining the established pre-processing cleaning model and GRU predictive and early warning model,based on the Storm platform,a distributed real-time prediction framework for the safety status of the shearer is established.Realize real-time distributed processing of ten types of monitoring data,the efficiency of the Storm framework is verified from three aspects:the prediction results of the prediction model,the accuracy of early warning and the processing efficiency based on the Storm platform.The experimental results show that the established distributed real-time prediction framework based on Storm has an accuracy of over 90%for the early warning of each state data of the shearer,and the early warning time can also meet the requirements of measurement points,which can provide timely guidance for safe production in mines.