Abstract: In order to improve the prediction accuracy of comprehensive energy consumption models for oil and gas production process, we apply the Gaussian mixture model (GMM) to the error compensation of these models. However, when the traditional expectation-maximization algorithm (EM) is used for GMM parameter estimation, it may cause problems with falling into a local minimum and overfitting. In order to address these problems, we optimize the structure and parameters of the GMM as a whole. In addition, we improve the EM algorithm as follows: first, we introduce a gradient operator into the genetic algorithm (GA) to form a gradient-based hybrid genetic algorithm (GGA), and then combine the GGA with the EM algorithm to form a GGA-EM algorithm. Next, we propose a GMM model error compensation method based on the GGA-EM algorithm. Finally, we apply the proposed GMM model error compensation method to the oil and gas production process in which there is a blockage in the oil-recovery operation area. The results show that this method can effectively improve model prediction accuracy, and lays a solid foundation for the optimal control of oil production process.