Research On Multi-parameter Fusion Prediction Of Gas Concentration Based On RNN

Gas concentration is one of the important factors leading to gas disasters,and gas concentration prediction is to ensure the safety of coal mine production and workers' lives.Therefore,effective prevention of gas disasters can reduce the occurrence of gas accidents and the losses caused by accidents.To meet the needs of coal mine safety production,it is a very important research topic to accurately predict the gas concentration in coal mining face.Coal mine gas concentration is affected by many factors,such as flow rate,wind speed,extraction negative pressure,concentration,temperature and so on.Because of the complex nonlinear relationship among the influencing factors,the prediction accuracy of gas concentration by using a single sensor will be low,which can not effectively reflect the real environmental conditions of the mine.Therefore,it is necessary to improve the accuracy of gas concentration prediction through multi-parameter fusion prediction,and then build a multi-parameter gas concentration prediction model with higher accuracy.At first,based on Lasso regression algorithm,a gas concentration feature selection algorithm is constructed,and the optimal value of ? and L1 norm are selected based on grid search method,so as to determine the gas concentration time series feature with strong interpretability.The gas concentration prediction results before and after Lasso feature selection are compared experimentally,and the effectiveness of the algorithm is verified.Secondly,taking the gas concentration feature set selected by Lasso regression algorithm as the research object,a multi-parameter fusion prediction model of gas concentration in working face based on RNN is constructed.and the performance evaluation index of the prediction method of the model is determined.The grid search method is used to optimize the hatch size,the number of neurons.the learning rate,the discarding ratio and the depth of the network,and the early stop method is used to prevent over-fitting.Finally,taking MAPE as the performance evaluation index,the gas concentration prediction model based on RNN is compared with the gas concentration prediction model based on PSO-SVR and PSO-Adam-BP neural network.The experimental results show that the gas concentration time series selected by Lasso regression algorithm can effectively remove the characteristic variables that have little correlation with the target variables,and keep the variables that have great correlation with the target variables,thus improving the interpretability of the gas concentration time series.The three variables selected by Lasso feature include return gas concentration,upper corner gas concentration and temperature.Using grid search method to adjust the parameters of batch size,neuron number,learning rate,discarding ratio and network depth can effectively find the optimal combination of parameters,and the training error can be reduced to 0.0195;Compared with BP neural network and SVR,the RNN gas concentration prediction model optimized by Adam has higher accuracy and stability.MAE can be reduced to 0.0573,RMSE can be reduced to 0.0167,and MAPE can be reduced to 0.3384%in the prediction process.RNN gas concentration prediction model and parameter optimization method based on Adam optimization can effectively predict gas concentration,and this method has higher accuracy in gas concentration time series prediction,which can provide certain reference for mine gas control.