Impact of Model and Forecast Uncertainties on the Performance of the Model Predictive Control of a PV-Battery-Heat Pump-Heat Storage System

Abstract

Recent research has shown that model predictive control (MPC) is a practical tool for the realization of an intelligent single-or multi-use energy management for both single and hybrid energy storage systems. Based on a system model and forecasts of external influences, such a controller will find the supposedly optimum decision to take in the immediate future. However, this decision will only be optimal for the given forecast and model. The inevitable model and forecast uncertainties may lead to decisions that are mathematically infeasible. Usually, underlying control loops ensure system stability and safety. However, uncertainties can be detrimental to the performance of the MPC, especially in multi-use applications, which have been shown to be preferable in practice due to a more economical usage of the storage devices.For this study, the authors carried out various analyses on the impact of both model and forecast uncertainties on the performance of the MPC in the case of a PV-Battery-Heat Pump-Heat Storage system in a single-family house providing self-consumption optimization and grid relief. Concerning the impact of model uncertainties, the use case was simulated repeatedly, varying both structure (linear and quadratic) and parameters of the optimization model. The impact of forecast uncertainties was investigated by simulating with real and ideal forecasts and identifying "typical" forecast errors that led to deviations in the system's behaviour using statistical methods. The results show that the influence of forecast uncertainties is usually higher than that of model uncertainties, but large model uncertainties may drastically alter the MPC's usage of a hybrid energy storage system. The identification of the most influential uncertainties forms the basis for developing a more robust MPC-based energy management technique.