Abstract: my country's energy production and energy consumption structure has long been characterized by a high proportion of coal and a low proportion of oil, and oil plays an irreplaceable role in economic and social development. This paper conducts an empirical study on the monthly data of crude oil production in my country from January 1986 to October 2021. Aiming at the non-stationary and strong volatility characteristics of the data, the wavelet signal decomposition processing technology and the model combination idea are used to establish a model from two perspectives. ARIMA-NAR neural network combined forecasting model, and then carried out a comparative study on the forecasting performance of the combined forecasting model. The study found that the prediction performance of the combined prediction method based on data preprocessing and model combination is significantly better than that of the single prediction model, and the ARIMA-NAR neural network combined prediction model based on model combination has higher prediction accuracy and can effectively predict my country's crude oil production provides a more scientific decision-making basis for my country's energy policy formulation.

Key Words: oil;wavelet decomposition; neural network ;difference integrated moving average autoregressive model