Explainable machine learning methods for predicting electricity consumption in a long distance crude oil pipeline.

Accurate prediction of electricity consumption in crude oil pipeline transportation is of significant importance for optimizing energy utilization, and controlling pipeline transportation costs. Currently, traditional machine learning algorithms exhibit several limitations in predicting electricity consumption. For example, these traditional algorithms have insufficient consideration of the factors affecting the electricity consumption of crude oil pipelines, limited ability to extract the nonlinear features of the electricity consumption-related factors, insufficient prediction accuracy, lack of deployment in real pipeline settings, and lack of interpretability of the prediction model. To address these issues, this study proposes a novel electricity consumption prediction model based on the integration of Grid Search (GS) and Extreme Gradient Boosting (XGBoost). Compared to other hyperparameter optimization methods, the GS approach enables exploration of a globally optimal solution by exhaustively evaluating all hyperparameter combinations, thereby enhancing the model's predictive performance. To improve the interpretability of the model's predictions, SHapley Additive exPlanations (SHAP) is employed to quantify the contribution of input parameters to the model's predictive outcomes. Experimental results demonstrate that the GS-XGBoost hybrid model outperforms Multilayer Perceptron, Support Vector Machine, Extreme Learning Machine, Random Forest, and Gradient Boosting Regression Tree, achieving the best predictive performance with mean absolute percentage error (MAPE) and coefficient of determination (R