It is an important problem that the oilfield can accurately predict the monthly oil production in the production process. The traditional linear prediction method has poor adaptability, and the machine learning method can’t be well applied to the time series data, and there are some important data missing in the monthly oil production data. Therefore, this paper proposes to use the random forest algorithm to fill in the missing values, and uses the long-term memory neural network (LSTM) to solve the time correlation in the data, and proposes to use the attention mechanism to enhance the influence of the important features of the data on the model, and adjusts the parameters from three aspects of time lag, learning rate attenuation and neuron random deactivation to select the optimal parameters The prediction model of monthly oil production is established. Taking an oilfield as an example, this paper uses attention mechanism and long-term and short-term memory neural network to model the monthly oil production of the oilfield, and compares it with linear regression, random forest, support vector regression, artificial neural network and single long-term and long-term memory neural network. The results show that compared with the single LSTM, the new network structure reduces MAPE of three wells by 15.5% - 51.9%, and predicts The trend is close to the original value. The model can be applied to practical development.