Oil Well Production Prediction Based on Informer Neural Network Model

Abstract The prediction of oil well production is of great significance for oilfield development and management. However, traditional oil well production prediction methods face complex geological conditions due to strong reservoir heterogeneity and non-linear data characteristics, resulting in unclear decline patterns and difficulty in achieving satisfactory prediction results. The article proposes a prediction method of oil well production prediction based on the Informer neural network model. By fully utilizing its advantages in processing long sequence data, a prediction model was established that comprehensively considers the correlation characteristics between production and geological, development, and other influencing factors, as well as the temporal changes in production itself. The main controlling factors of are selected through correlation analysis. Production data for 4 years are used as the dataset, the effectiveness of the production prediction model was demonstrated through model training, validation, and testing. The influence of different parameters on the prediction results of the model are studied, the results show that the method has good predictive performance, and the new model overcomes the traditional decreasing model in describing the correlation of production time series data. The ability to comprehensively consider the influence of multiple factors such as daily oil production, daily water production, back pressure and wellhead temperature on production has effectively improved the accuracy of production prediction. The model constructed in this article effectively captures the long-term dependency relationship between production and various parameters through self-attention mechanism.