Prediction model of goaf coal temperature based on PSO-GRU deep neural network

Abstract

The accurate determination of coal temperature in hidden space such as goaf has always been a worldwide problem that needs to be solved. The research of coal temperature prediction model has important practical significance for the accurate detection of loose coal temperature. Based on the coal natural ignition experiment and coal spontaneous combustion gas characterization index as the initial data set GRU neurons are used to mine the nonlinear relationship between the index gas and temperature, and the GRU model parameters are optimized by PSO to obtain the predicted value of coal body temperature. The results show that the predicted MAE value of the PSO-GRU model is 1.37 °C, 6.51 °C, 11.40 °C, 15.90 °C, 20.20 °C lower than that of the PSO-SVM, PSO-BP, BP, RF and SVM prediction models respectively. The RMSE value decreased by 0.45 °C, 4.44 °C, 10.33 °C, 15.71 °C and 24.24 °C respectively. The judgment coefficient R2 on the test set and the training set of the PSO-GRU model is 0.99, and the generalization, prediction accuracy and robustness are all good. The experimental results show that the inversion temperature error of the model is within the range of 3.87 %, the maximum temperature difference is 5.69 °C, and the average temperature difference is 2.83 °C, which can meet the accuracy requirements of the field temperature measurement.

Keywords

• Coal spontaneous combustion
• Temperature prediction
• Indicator gas
• PSO-GRU model
• Recurrent neural network