Oktaviano, Akhrim (2026) Desain Kendali Adaptif Inersia Virtual Berbasis Reinforcement Learning untuk Peningkatan Stabilitas Frekuensi Akibat Penetrasi Pembangkit Listrik Tenaga Surya. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6022241131-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6022241131-Master_Thesis.pdf Restricted to Repository staff only Download (11MB) | Request a copy |
Abstract
Perkembangan teknologi sumber energi terbarukan (EBT), terutama sistem Pembangkit Listrik Tenaga Surya (PLTS) dengan kapasitas yang bervariasi telah menimbulkan tantangan baru bagi sistem tenaga listrik, terutama pada sistem mikrogrid terisolasi yang dapat mengganggu stabilitas frekuensi akibat keterbatasan inersia dalam menanggulnagi ketidakpastian yang dihasilkan oleh PLTS. Untuk mengatasi masalah ini, mekanisme kendali inersia virtual (VIC) adaptif perlu diintegrasi pada sistem penyimpanan energi berbasis baterai untuk mengatasi ketidakpastian yang berasal dari PLTS dan permintaan beban. Oleh karena itu, penelitian ini bertujuan untuk merancang algoritma reinforcement learning (RL) berupa Twin Delayed Deep Deterministic Policy Gradient (TD3) yang tangguh dalam berbagai kondisi ketidakpastian untuk meningkatkan stabilitas frekuensi dalam sistem mikrogrid terisolasi. Uji simulasi dilakukan pada variasi penetrasi sistem PLTS sebesar 20%, 40%, dan 60% dengan beban listrik yang bervariasi. Hasil menunjukkan bahwa TD3 lebih unggul Deep Deterministic Policy Gradient (DDPG) maupun Fuzzy Logic Controller (FLC) dengan rata-rata peningkatan performa respon frekuensi dan rata-rata peningkatan kemampuan peredaman overshoot masing-masing sebesar 24,75% dan 16,65% pada penetrasi PLTS sebesar 20%, 23,39% dan 18,46% pada penetrasi PLTS sebesar 40%, serta 16,19% dan 20,27% pada penetrasi PLTS sebesar 60%. Penelitian ini juga memperlihatkan keunggulan mekanisme TD3 dalam mengatasi permasalahan estimasi bias secara berlebihan dari metode DDPG.
=======================================================================================================================================
The development of renewable energy technology, especially photovoltaic (PV) systems with varying capabilities, has created new challenges for power system stability, particularly in isolated microgrid systems, which can disrupt frequency stability due to inertia limitations in responding to the uncertainty generated by PV systems. To overcome this problem, an adaptive virtual inertia control (VIC) mechanism must be integrated into battery-based energy storage systems to address uncertainties arising from solar power uncertainty and load demand. Therefore, this study proposes a reinforcement learning (RL) algorithm, based on the Twin Delayed Deep Deterministic Policy Gradient (TD3) architecture, designed to be robust to various sources of uncertainty and to improve frequency stability in isolated microgrid systems. Simulation tests were conducted to assess variations in the PV system's penetration of 20%, 40%, and 60% under varying electrical loads. The results show that TD3 outperforms Deep Deterministic Policy Gradient (DDPG) and Fuzzy Logic Controller (FLC) with an average increase in frequency response performance and an average increase in overshoot damping capability of 24,75% and 16,65% at a PV systems penetration of 20%, 23,39% and 18,46% at 40% PV systems penetration, lastly 16,19% and 20,27% at 60% PLTS systems penetration. This study also demonstrates the superiority of the TD3 mechanism in mitigating the excessive bias estimation problem in the DDPG method.
Actions (login required)
 |
View Item |