Optimal Sizing and Placement of Energy Storage Systems in Hybrid Energy Environment

Abstract

The study suggests a method for setting up a photovoltaic (PV) system's hybrid energy storage system (HESS), which consists of batteries and supercapacitors, to have the best possible capacity. The objective is to reduce overall expenses, including investment and running costs, while minimizing variability in PV production and maximizing system performance.

The optimization model has two layers. In order to reduce the overall PV-HESS system cost, the higher layer chooses the battery and supercapacitor capacity and power ratings. Using deep reinforcement learning, the lower layer optimizes the PV-HESS system's performance over the course of a year.

The investment expenses for the battery and supercapacitor are taken into account in the top layer, together with the annual running costs determined by the bottom layer model. By regulating the HESS charging and discharging and lowering PV curtailment and fluctuation penalties, the lower layer seeks to reduce operational costs. Both layers' limitations and objective functions are presented.

The lower layer operation optimization problem is proposed to be solved via the deep deterministic policy gradient (DDPG) algorithm. The inputs include the PV output, load demand, electricity price, and the HESS state of charge. The control actions are the battery/supercapacitor power outputs.

In Shandong, China, a 1 MW PV system is being researched. The ideal HESS capacity arrangement, according to the results, reduces the overall cost as compared to no storage by 7.7%. To smooth out variations and move PV generation to times of high prices, the HESS works in concert with the battery and supercapacitor. It is demonstrated that the DDPG algorithm efficiently optimizes system performance.

 In order to configure battery and supercapacitor capacities for a PV-HESS system while taking operating and investment costs into account, the study presents an optimisation technique. PV variability is managed by a reinforcement learning algorithm and a two-layer model to keep costs down.