INTELLIGENT MICROGRID ENERGY MANAGEMENT USING DEEP Q-LEARNING METHOD BASED ON GRID FEED OPTIMIZATION

• Main Author: Mahardika P. Setiawan, Fadhila
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73210
• Institution: Institut Teknologi Bandung

Microgrid (MG) is one of the energy system entities in a smart electricity network that realizes the integration of renewable energy sources to answer the increasing demand for electrical energy along with the depletion of fossil energy reserves. However, the intermittent nature of renewable energy sources is an obstacle for MG, because it can cause a decrease in MG's performance which is marked by the non-maximum renewable fraction (RF). This can be overcome by implementing controls on MG components, with one of the components being the energy storage battery system (SBPE). By adding the grid feed action option to the SBPE, the charging and discharging range of the battery will increase and based on grid feed scheduling it can result in an increase in RF and PV Utilization (PU). In this research, optimization-based scheduling of SBPE actions was developed using a type of reinforcement learning method, namely deep Q-learning based on grid feed optimization. The control algorithm is designed to be compatible with the MG digital twin (MGDT) framework which models physical objects into digital objects. Based on the test results and analysis, it was found that the increase in RF and PU was 15.30% and 36.28%, respectively. Grid feed optimization reduced the LCOE (Levelized Cost of Electricity) value from IDR 3,102.09/kWh to IDR 139.10/kWh. The average value of the training reward is 3.93 after adding the grid feed.