Electronic Theses and Dissertation

Internet of Drones (IoD) dalam skenario komunikasi multi-hop menghadapi tantangan kritis berupa keterbatasan energi, terutama pada node relay yang menangani beban transmisi data secara terus-menerus. Dinamika jaringan akibat mobilitas tinggi drone serta fluktuasi kanal pada standar IEEE 802.11ah menyebabkan metode pemilihan relay heuristik konvensional menjadi kurang adaptif, sehingga memicu ketimpangan konsumsi energi. Untuk mengatasi permasalahan krusial tersebut, penelitian ini mengusulkan sebuah mekanisme pemilihan node relay adaptif berbasis Deep Reinforcement Learning (DRL) menggunakan algoritma Proximal Policy Optimization (PPO) guna mengoptimalkan network lifetime. Eksperimen ini mengimplementasikan arsitektur co-simulation yang mengintegrasikan Network Simulator 3 (NS-3) dan agen Python secara real-time via TCP Socket, dengan mengevaluasi 68 drone di mana 10 di antaranya bertindak sebagai kandidat relay. Berdasarkan observasi terhadap sisa energi, posisi, RSSI, dan jarak ke Ground Station, agen AI dirancang menggunakan fungsi reward multi-objektif yang memberikan penalti ekstrem pada kondisi dead node dan apresiasi tinggi untuk efisiensi energi kumulatif. Hasil pengujian membuktikan bahwa agen PPO berhasil mencapai konvergensi kebijakan yang stabil dengan rata-rata sisa baterai akhir sebesar 75,48% mengungguli metode Random Baseline yang berada di angka 73,52%. Lebih lanjut, metode ini sukses meningkatkan kualitas pengiriman data dengan Packet Delivery Ratio (PDR) sebesar 64,2%, meningkat signifikan sebesar 12,6% dibandingkan metode acak yang hanya mencapai 51,6%. Dengan demikian, metode PPO terbukti sangat efisien dalam mendistribusikan beban energi secara merata,

The Internet of Drones (IoD) in multi-hop communication scenarios faces a critical challenge of energy constraints, particularly at relay nodes handling continuous data transmission loads. Network dynamics due to the high mobility of drones and channel fluctuations in the IEEE 802.11ah standard render conventional heuristic relay selection methods inadequately adaptive, thereby triggering imbalances in energy consumption. To overcome this crucial issue, this research proposes an adaptive relay node selection mechanism based on Deep Reinforcement Learning (DRL) using the Proximal Policy Optimization (PPO) algorithm to optimize network lifetime. This experiment implements a co-simulation architecture that integrates Network Simulator 3 (NS-3) and a Python agent in real-time via a TCP Socket, evaluating 68 drones, of which 10 act as relay candidates. Based on observations of residual energy, position, RSSI, and distance to the Ground Station, the AI agent is designed using a multi-objective reward function that applies extreme penalties for dead node conditions and high appreciation for cumulative energy efficiency. The test results prove that the PPO agent successfully achieved stable policy convergence with an average final residual battery of 75.48%—outperforming the Random Baseline method, which only reached 73.52%. Furthermore, this method successfully boosted data delivery quality with a Packet Delivery Ratio (PDR) of 64.2%, a significant 12.6% increase compared to the random method which only reached 51.6%. Thus, this method is proven to be highly efficient in evenly distributing energy loads, improving successful packet delivery, and extending the operational lifespan of the network.