Combined Prediction Model of Gasoline Octane Loss Value Based on XGboost-GBDT Combined Prediction

Establishing a gasoline octane loss value influence factor and prediction model can provide an optimization reference for industry to optimize the desulfurization and olefin reduction process flow, in addition to providing an efficient and low-cost octane loss prediction method can complement the current complex and high-cost octane measurement methods. In this paper, XGBoost and Random Forest models are applied to realize the screening of gasoline octane loss value problem. XGBoost and GBDT models are applied for weighted combination using inverse error method. The problems such as the large error of a single model are corrected, and the combined prediction model of XGBoost-GBDT finally obtains a MAPE of 0.0883 and an RMSE of 0.1660, which achieves further optimization of the prediction problem of gasoline octane loss value.