Memory cutting and optimization of shearer based on K-GRU neural network

AN Weipeng; YAN Penghao; ZHANG Wenbo; SUN Xuxu

doi:10.16186/j.cnki.1673-9787.2021090055

Abstract：Objective Aiming at the inaccurate memory cutting and the low degree of automation of shearer,Methods This paper proposed a shearer memory cutting algorithm based on K-GRU neural network.This al-gorithm was more suitable for processing long-time sequence data.Combining the algorithm with the memory cutting of shearer can reduce the damage of the drum during the coal mining process and protect the safety of workers'lives.The algorithm introduced the proportional factor K at the input end of the deep gated re-current unit(GRU),and used the proportional factor K to show the importance of data at different times and to strengthen the memory of the model for long-time sequence data,thereby improving the accuracy of memory cutting.In the model training stage,the random search algorithm(RS)was used to optimize the hyper-parameter selection of the deep K-GRU neural network to speed up the training speed of the model.Results In the experiment,Python was used to complete the construction of the K-GRU model and the optimization of hyperparameters.Using the random search algorithm,the optimal solution of the hyperparameter could be obtained in a shorter time.The optimal solution of the hyperparameter epochs of 317 and the batch_size of 70 costed a total of 154 s.In the case of the optimal solution,the error of the calculation model's prediction of the real coal mining data was 0.0467,R2 was 0.957 8,EVS was 0.965 6,and the ME was 0.083 3.Con-clusion Finally,it showed that the optimized deep K-GRU model was better than SVM,KNN,LSTM,RNN and ordinary GRU models in terms of interpretation of variance score,maximum error and decision coeffi-cient,which significantly improved the applicability and accuracy of shearer memory cutting.