Single-Well Yield Prediction Based on LSTM and MA Combination Model

On the basis of the characteristics of oil single-well production and oil well abnormal events, it is proposed to use the combination of long short-term memory network (LSTM) and moving average (MA) to predict oil single-well production, to use time series model with LSTM long-term memory ability to predict oil single-well production and to use MA to predict output after abnormal events fluctuation, and MA can therefore be used to predict output after abnormal events, and then to use LSTM to predict subsequent output after actual output is stable. The experimental oil and gas production data shows that the combination model of LSTM and MA is compared with the BP neural network and the single LSTM. The prediction results of combination model of LSTM and MA are significantly better than individual LSTM and BP neural networks for the situation after abnormal events in oil wells. The combined model can obviously reduce the yield prediction error within a period of time after the occurrence of abnormal events, make it more real close to the actual yield value and improve the prediction accuracy, which is of great significance in practical application.