Stochastic multi-carrier energy management in the smart islands using reinforcement learning and unscented transform

Abstract

This article investigates the optimal management of multi-carrier water and energy system(MCWES) considering the high penetration of renewable energy sources as non-dispatchable units and the seawater desalinization mechanism for serving water demand in the target area. The proposed model encompasses several demand layers including power energy, natural gas and water layer which supplies the electricity, thermal and drinkable water demands in the smart island. In order to capture the uncertainty effects in the technical decisions of optimal scheduling, a stochastic approach based on unscented transform (UT) is developed to handle the forecast error in the electrical and thermal energy demands, market energy prices related to the different energy layers and the output power forecast error in the renewable energy sources. Solving the proposed coordinated scheduling problem in an hourly timescale requires heavy calculations that make it impractical. Therefore, a novel reinforcement learning (RL) based approach is devised for finding a near optimal solution and facilitates the searching process with a trivial computational burden. The simulation results indicate that the proposed cooperation approach minimizes both the operation and investment costs substantially with an efficient computational burden based on the advanced features coming out of the proposed RL approach. Last but not least, the simulation results on a practical smart island advocate the effectiveness and high efficacy of the proposed model. Also it was seen that the RL approach could properly solve the optimization model and the selection of the sizes of the components was highly linked to the hourly values of demands and prices of the energy.