Parameter prediction of oilfield gathering station reservoir based on feature selection and long short-term memory network

As an essential part of the oil industry chain, oilfield united station needs the modeling and prediction of production parameters to avoid potential risks. In this study, the oil transfer temperature of an oilfield united station in China is modeled using long and short-term memory network (LSTM) with feature selection to attenuate "curse of dimensionality", including spearman's rank correlation coefficient-LSTM(SRCC-LSTM), R-type clustering-LSTM(R-LSTM) and transfer entropy-LSTM(TE-LSTM). Performance of these models is evaluated by four indicators. The contribution of the main control variables to the transfer temperature is determined based on the mean impact value method. The results show that the accuracy of the models reaches >95 %, which is better than the classical machine learning models. The computational efficiency is improved by 8.93 %similar to 13.66 %, indicating that the proposed models are reliable. In the future, the method in this study can also be used for determining the tendency of other sensor variables.