One of the central questions in science is forecasting: based on the past history, how well can we predict the future? In many domains with complex multivariate correlation structures and nonlinear dynamics, forecasting is highly challenging. In the oil and gas industry, forecasting decline curves is an important component for E&P companies in business planning, asset evaluation, and decision making. Here we introduce a machine learning approach to tackle the problem, and to be more specific, an LSTM approach (LSTM stands for Long Short Term Memory, which is one kind of recurrent neural network). Compared with the hyperbolic approach, the LSTM model is more dynamic and has a better chance of capturing non-linear events. We build the prediction model from the accumulated curve domain, and eventually ensemble multiple models to reduce the variance.  Given the fact that the model is only trained on the first 3 months of data , the prediction for the first 2 years shows great promise.