Dealing with uncertainty in automated operational planning for residential fuel cell system: A comparative study of state-of-the-art approaches

Abstract

Polymer electrolyte fuel cell cogeneration systems (PEFC-CGSs) provide hot water by utilizing exhaust heat produced in electricity generation process. The energy saving potential of PEFC-CGSs can be maximized by optimal operational plans, and most state-of-the-art approaches implement operational planning function (OPF) based on energy demand time-series prediction by using machine learning techniques. In general, prediction of demand time-series with small expected average errors is regarded as the most important point in obtaining appropriate operational plans; however, several recent studies have revealed that other complex factors such as the direction and timing of forecast errors greatly affect the quality of operational plans in some cases. Core ideas proposed in these previous studies are broadly classified into seven types. The purpose of this study is to characterize these OPFs from the two aspects: the output form of prediction model and prediction target variable, and to clarify “what kind of uncertainty should be focused on” and “how this uncertainty should be handled” in designing OPF. The seven kinds of OPFs were comprehensively evaluated via numerical simulations using real-world data. The results show the significance of OPF based on prediction of expected operational cost surface using multiple output prediction model.