An Interval Prediction Method for Imbalanced Fuel Consumption Data

Abstract

Fuel consumption data is imbalanced, which leads to the lower quality prediction interval. Aiming at this problem, an interval prediction model based on RU-SMOTE-XGBoost algorithm is proposed. Random undersampling(RU) algorithm is used to reduce the number of majority samples, and SMOTE algorithm is used to increase the number of minority samples in the training set, so that the imbalance of data is eliminated. For the interval prediction task, the quantile loss function is used as the loss function of XGBoost algorithm. At the same time, by smoothing the small area around the origin of its first derivative, the quantile loss function is improved to solve the problem that the quantile loss function causes the tree in the XGBoost algorithm to not split. Based on the above work, XGBoost algorithm, RU algorithm and SMOTE algorithm are combined to train the interval prediction model, and finally the upper and lower bound of the prediction interval are obtained respectively. Conducting experiments based on the Quick Access Recorder(QAR) data set, the experiment results indicate that compared with other methods, this method makes the prediction interval have higher interval coverage and narrower interval width, which improves the quality of the prediction interval.