Skripsi

DEEP REINFORCEMENT LEARNING FOR RELIABLE AND STABLE CONTROL OF A HYBRID RENEWABLE ENERGY SYSTEM IN A MICROGRID

---

Penilaian

---

The integration of renewable energy sources in microgrids presents significant challenges to ensuring stable and reliable control, particularly under dynamic and uncertain conditions. Such problems occur due to integration of inverter-based generation that dominantly controls renewable energy response to the grid. Inverted-based generation typically have low-inertia that induces instability to the system whenever faced with sudden disturbances, such as unplanned circuit trip, transient islanded event, etc. Deep Reinforcement Learning (DRL), with its adaptive and data-driven capabilities, emerges as a promising solution to address these complexities. This dissertation investigates the potential of DRL-based control systems to enhance voltage and frequency stability in microgrids, focusing on scenarios involving fluctuating loads and unplanned transitions between grid-tied and islanded modes. This study follows a multi-stage methodology. First, a systematic literature review (SLR) is conducted to map the state of the art in DRL for microgrid control and to distill open gaps that inform the design choices. Next, a single-inverter microgrid is modeled and a TD3-based DRL controller is formulated in the dq0 frame with a continuous action space; the observation vector includes bus-coupler status, V and f, their errors, derivatives, and integrals, while the reward penalizes deviation beyond operational bands (e.g., 49.5–50.5 Hz, 0.98–1.02 pu) and excessive control effort. The agent is trained under certain operating scenarios—blank start, unplanned grid-tied, and unplanned islanded transitions—with stochastic loads and scheduled exploration–exploitation profiles; FER is systematically swept (1–5%) to study policy robustness. For benchmarking, a tuned Proportional-Integral-Derivative (PID) controller is implemented on the same plant. Performance is evaluated via rise time, peak time, settling time, settling min/max, MAD and constraint violations (voltage/frequency bands), using multiple randomized seeds and unseen test episodes to assess generalization. Finally, implementation details (sampling time, episode length, hyperparameters) are fixed a priori to ensure reproducibility, and all simulations follow an identical measurement and metering pipeline. Simulation results demonstrate that a Final Exploitation Rate (FER) of 2% yields optimal performance during transient conditions, outperforming other DRL configurations in limiting voltage and frequency deviations. However, comparative analysis with conventional PID controllers reveals that PID maintains superior stability in steady-state conditions due to its passive control nature. The study also validates the DRL agent on a low-cost, single-phase inverter in an islanded microgrid, maintaining 220 V, 50 Hz, and THD below 8%, offering a cost-effective alternative to commercial solutions. Scalability is tested using the 2024 Sumatra blackout, simulating a 600 MW generation loss with and without DRL-controlled DRE units. Results show that DRL, though trained on local microgrid dynamics, still achieve less than 2% frequency deviation for system recovery, demonstrating its potential for scalability and generalization to regional scale in enhancing power system resilience. The findings highlight DRL's efficacy in managing transient instability and suggest its potential as a next-generation control strategy in distributed energy systems. The combination of simulation-based performance analysis and experimental validation underscores the feasibility of DRL for real-world microgrid applications, while identifying areas where traditional controllers remain advantageous.

---

Availability

Inventory Code	Barcode	Call Number	Location	Status
2507006281	T185806	T1858062025	Central Library (Reference)	Available but not for loan - Not for Loan

Detail Information

Series Title

-

Call Number

T1858062025

Publisher

Indralaya : Prodi Doktor Ilmu Teknik, Fakultas Teknik Universitas Sriwijaya., 2025

Collation

x, 168 hlm.; ilus.; tab.; 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

333.707

Content Type

Text

Media Type

unmediated

Carrier Type

other (computer)

Edition

-

Subject(s)

Prodi Doktor Ilmu Teknik
Hybrid Renewable Energy

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment