Prediction Method of Heavy Oil Horizontal Well Cycle Oil Production Based on PCA and Gradient Boosting Decision Tree

Abstract

With the increasing production of heavy oil reservoirs, it is very important to predict the periodic oil production of horizontal well steam stimulation. In view of the difficulties and limitations of conventional forecasting methods, a PCA-GBDT production forecasting method was established based on Principal Component Analysis (PCA) and Gradient Boosting Decision Tree (GBDT). Firstly, On the basis of the existing geological data and field production data, partial data are fused and processed. And then based on grey correlation analysis to all kinds of factors affecting the relevance, we can obtain the main controlling factors of cycle oil production. Next, principal component analysis is used to eliminate the correlation of influencing factors, and a sample library for machine learning is established. Finally, Gradient Boosting Decision Tree was used to train the established machine learning sample library, and grid search was used to optimize the maximum tree depth of model parameters and the number of decision trees to form the yield prediction model. The results show that the PCA-GBDT production prediction model can accurately predict cycle oil production, and the average absolute error is less than 10%, which meets the requirements of field and engineering. This method is of great practical significance for the production and development adjustment of heavy oil horizontal well steam huff and puff and the application of machine learning in unconventional resources.