Analisis Spatio-Temporal Prediksi Pasang Surut Menggunakan Nonlinear Autoregressive With Exogenous Input (NARX) Neural Network

Arnia, Satya (2026) Analisis Spatio-Temporal Prediksi Pasang Surut Menggunakan Nonlinear Autoregressive With Exogenous Input (NARX) Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5016221054-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Pasang surut merupakan fenomena naik turunnya permukaan air laut yang terjadi secara periodik akibat pengaruh faktor astronomis dan non-astronomis, seperti angin, curah hujan, dan tekanan atmosfer. Metode analisis komponen harmonik yang umum digunakan dalam prediksi pasang surut memiliki keterbatasan karena hanya memperhitungkan aspek astronomis tanpa mempertimbangkan faktor non-astronomis. Penelitian ini menerapkan pendekatan machine learning berbasis Nonlinear Autoregressive with Exogenous Input (NARX) Neural Network untuk memprediksi pasang surut dengan mempertimbangkan variasi spatio-temporal. Penelitian ini bertujuan untuk menganalisis karakteristik spatio-temporal dua stasiun pasang surut yang berdekatan, mengevaluasi akurasi model NARX, serta membandingkannya dengan metode analisis harmonik U-Tide. Data yang digunakan meliputi data pasang surut stasiun Cilacap (CLCP) sebagai output, serta data pasang surut stasiun Cilacap UHSLC (CILI), data angin, curah hujan, dan tekanan atmosfer sebagai input eksogen. Model NARX dikonfigurasi dengan input delays 1:12, feedback delays 1:12, dan 8 neuron pada hidden layer. Hasil penelitian menunjukkan bahwa kedua stasiun memiliki karakteristik spatio-temporal yang serupa dengan tipe pasang surut semidiurnal (bilangan Formzahl CLCP = 0,224; CILI = 0,233), selisih amplitudo komponen harmonik antarstasiun kurang dari ±0,006 m, dan pola residu yang konsisten, sehingga data CILI layak digunakan sebagai variabel input eksogen. Pada fase pelatihan, model mencapai nMAE 1,047%, RMSE 0,031 m, dan R² = 0,995. Pada fase prediksi bulan Desember, model menghasilkan nMAE di bawah 3% pada seluruh horizon prediksi (7 hari: nMAE 2,557%, R² 0,985; 15 hari: nMAE 2,378%, R² 0,985; 29 hari: nMAE 2,186%, R² 0,985). Perbandingan dengan metode U-Tide menunjukkan keunggulan NARX secara konsisten, dengan nMAE NARX sebesar 2,186% dibandingkan U-Tide sebesar 5,216% pada periode prediksi 29 hari. Hasil ini mengkonfirmasi bahwa model NARX mampu menangkap komponen non-astronomis yang tidak dapat dimodelkan oleh metode harmonik konvensional
================================================================================================================================
Tides are a periodic rise and fall of sea level driven by astronomical and non-astronomical factors, including wind, rainfall, and atmospheric pressure. The harmonic analysis method, which is widely used for tidal prediction, has a fundamental limitation in that it only accounts for astronomical components without incorporating non-astronomical influences. This study applies a Nonlinear Autoregressive with Exogenous Input (NARX) Neural Network to predict tides by considering spatio-temporal variations. The study aims to analyze the spatio-temporal characteristics of two nearby tidal stations, evaluate the accuracy of the NARX model, and compare its performance against the U-Tide harmonic analysis method. The dataset includes tidal observations from the Cilacap station (CLCP) as the output variable, and tidal data from the Cilacap UHSLC station (CILI), wind, rainfall, and atmospheric pressure as exogenous inputs. The NARX model was configured with input delays of 1:12, feedback delays of 1:12, and 8 neurons in the hidden layer. Results show that both stations share similar spatio-temporal haracteristics, classified as semidiurnal tide type (Formzahl number: CLCP = 0.224; CILI = 0.233), with inter-station amplitude differences of less than ±0.006 m and consistent residual patterns, confirming that CILI data is appropriate as an exogenous input variable. During the training phase, the model achieved nMAE of 1.047%, RMSE of 0.031 m, and R² = 0.995. In the December prediction phase, the model produced nMAE below 3% across all prediction horizons (7 days: nMAE 2.557%, R² 0.985; 15 days: nMAE 2.378%, R² 0.985; 29 days: nMAE 2.186%, R² 0.985). Compared to U-Tide, NARX demonstrated consistently superior performance, with nMAE of 2.186% versus 5.216% over the 29-day prediction period. These findings confirm that the NARX model effectively captures non-astronomical tidal components that conventional harmonic methods cannot model

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Prediksi Pasang Surut, Machine learning, NARX Neural Network Tide Forecasting, Machine learning, NARX Neural Network
Subjects:	G Geography. Anthropology. Recreation > GC Oceanography > GC89 Sea Level
Divisions:	Faculty of Civil, Planning, and Geo Engineering (CIVPLAN) > Geomatics Engineering > 29202-(S1) Undergraduate Thesis
Depositing User:	Satya Budi Arnia
Date Deposited:	29 Jun 2026 04:22
Last Modified:	29 Jun 2026 04:22
URI:	http://repository.its.ac.id/id/eprint/134090

Actions (login required)

View Item