Optimization tool for operating isolated diesel-photovoltaic-battery hybrid power systems using day-ahead power forecasts

Abstract

This paper presents a computational tool based on a genetic algorithm and artificial neural network for optimizing the operation of isolated diesel-photovoltaic-battery hybrid power systems using day-ahead power forecasts obtained with quantile random forests. The optimization tool was conceived to be flexible, i.e., it can be used to operate isolated power systems with multiple configurations of diesel generator sets (DGS), to work with a reduced number of input data, and to be as simple as possible to be used. The optimization relies on combining valley-filling and peak-shaving strategies using battery energy storage systems while considering the combined forecast of demand and photovoltaic (PV) generation. The tool also simulates the behavior of the DGS to define the optimum arrangement of diesel generators considering the variability of both demand and PV generation. The output consists of hourly values of energy storage power dispatch, DGS arrangement, and, if necessary, load shedding and/or PV curtailment. The algorithm that implements the optimization tool, which is currently in the phase of field-test in the isolated diesel-photovoltaic-battery hybrid power system of Fernando de Noronha, Brazil, demonstrated a good performance in computer simulations validated with real measured data.