PENINGKATAN AKURASI PREDIKSI POSISI KAPAL MENGGUNAKAN METODE PSO-BIGRU BERBASIS DATA AIS MIKRO SATELIT

Dicka Ariptian Rahayu, Dr.Eng. Ir. Igi Ardiyanto, S.T., M.Eng., IPM., SMIEEE.;Widyawan, S.T., M.Sc., Ph.D.

2025 | Tesis | S2 Teknologi Informasi

Keselamatan, keamanan, dan pemantauan maritim merupakan aspek krusial dalam transportasi laut di Indonesia, negara kepulauan dengan jalur pelayaran strategis. Sistem Automatic Identification System (AIS) konvensional terbatas pada jangkauan line-of-sight (40–70 km), sedangkan satelit orbit rendah seperti LAPAN-A2 dan LAPAN-A3 menghasilkan data temporal yang tidak kontinu akibat jeda waktu pemantauan (4–12 jam), menyebabkan periode tanpa pantauan posisi kapal. Penelitian ini bertujuan mengembangkan model deep learning berbasis Bidirectional Gated Recurrent Unit (BiGRU) untuk prediksi single-step posisi kapal secara akurat menggunakan data historis AIS dari LAPAN-A2 dan LAPAN-A3 di perairan timur Indonesia, khususnya sekitar Pulau Papua. Optimasi hyperparameter model BiGRU, meliputi jumlah neuron, layer, learning rate, dropout, dan optimizer, dilakukan dengan Particle Swarm Optimization (PSO) untuk meningkatkan akurasi. Hasil pengujian menunjukkan bahwa model PSO-BiGRU 2 mencapai Mean Squared Error (MSE) terendah sebesar 0.0472230 pada data uji 20%, mengungguli model RNN, LSTM, GRU, BiRNN, BiLSTM, Transformer, dan PSO-Transformer. Pada pengujian sembilan kapal terpilih, seperti Kapal F (MMSI 553111621), model menghasilkan MSE 0.00003 deg2, Fréchet Distance (FD) 0.00749 deg, dan Average Euclidean Distance (AED) 0.83238 km, menunjukkan keandalan untuk lintasan bervariasi. Pendekatan ini diharapkan meningkatkan efektivitas pemantauan maritim, mendukung navigasi otomatis dan deteksi aktivitas ilegal di Indonesia timur.  

Maritime safety, security, and monitoring are critical aspects of sea transportation in Indonesia, an archipelagic nation with strategic shipping lanes. Conventional Automatic Identification System (AIS) is limited to a line-of-sight range of 40–70 km, while low-earth orbit satellites such as LAPAN-A2 and LAPAN-A3 produce discontinuous temporal data due to monitoring gaps of 4–12 hours, resulting in periods without vessel position tracking. This study aims to develop a deep learning model based on Bidirectional Gated Recurrent Unit (BiGRU) for accurate single-step vessel position prediction using historical AIS data from LAPAN-A2 and LAPAN-A3 in eastern Indonesia, particularly around Papua waters. Hyperparameter optimization of the BiGRU model, including the number of neurons, layers, learning rate, dropout, and optimizer, was performed using Particle Swarm Optimization (PSO) to enhance accuracy. Testing results show that the PSO-BiGRU 2 model achieved the lowest Mean Squared Error (MSE) of 0.0472230 on 20% test data, outperforming RNN, LSTM, GRU, BiRNN, BiLSTM, Transformer, and PSO-Transformer models. For nine selected vessels, such as Vessel F (MMSI 553111621), the model yielded an MSE of 0.00003 deg², Fréchet Distance (FD) of 0.00749 deg, and Average Euclidean Distance (AED) of 0.83238 km, demonstrating reliability across varied trajectories. This approach is expected to enhance maritime monitoring effectiveness, supporting autonomous navigation and detection of illegal activities in eastern Indonesia.

Kata Kunci : Navigasi Maritim, Prediksi Posisi Kapal, AIS Satelit, Particle Swarm Optimization, Deep learning, Indonesia Timur.