INTELLIGENT DEMAND FORECASTING APPROACHES FOR SPARE PARTS IN THE ENERGY INDUSTRY

In this study, two intelligent demand forecasting approaches are used to forecast the spare parts demand in the energy industry. In the first approach, a stacked generalization-based demand forecasting technique that combines traditional time-series forecasting and machine-learning methods is developed. In the second approach, external information (EI) is incorporated into the first one. Thus, the stacked generalization-based demand forecasting technique is used as a base model, and the EI is used to focus on predicting the peak demand; consequently, significant forecast errors can be minimized. A case study of a natural gas liquefaction company is then considered to test the performance of these methodologies. Our results show that the proposed techniques perform significantly better than the previous methods. Compared with the mean absolute scaled error and relative geometric root mean squared error of the company's forecasts, our intelligent demand forecasting approaches yield a 40.07% (36.81%) and 2.07% (3.40%) increase in butterfly-valve demand forecasting (spiral wound gasket) when no EI is used, and a 57.78% (60.41%) and 5.73% (7.36%) improvement with EI, respectively.