Sistem Prediksi Tinggi Muka Air Guna Peringatan Dini pada Sungai Kalibokor dengan Metode Long Short Term Memory

Kautsari, Kafka (2026) Sistem Prediksi Tinggi Muka Air Guna Peringatan Dini pada Sungai Kalibokor dengan Metode Long Short Term Memory. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221013-Undergrduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (49MB) | Request a copy

Abstract

Surabaya memiliki jaringan saluran air yang luas, salah satunya adalah Sungai Kalibokor yang berfungsi sebagai saluran primer di wilayah timur kota. Sungai ini kerap mengalami luapan air dengan genangan mencapai 0–15 cm akibat curah hujan tinggi dan sistem drainase pasif, yang menyebabkan banjir lokal khususnya di wilayah hilir seperti kelurahan Keputih. Hingga saat ini, belum tersedia sistem pemantauan tinggi muka air yang dapat memberikan informasi kondisi sungai secara langsung dan berkelanjutan, sehingga potensi banjir sulit diketahui sejak dini dan hanya bergantung pada data curah hujan dari BMKG tanpa memperhitungkan parameter lingkungan lain secara realtime. Penelitian ini bertujuan merancang sistem monitoring dan prediksi tinggi muka air berbasis Internet of Things (IoT) yang mampu memberikan data kondisi sungai secara realtime sekaligus memperkirakan potensi banjir secara dini. Sistem memanfaatkan beberapa sensor untuk mengukur parameter lingkungan seperti temperatur, intensitas cahaya, curah hujan, ketinggian air, arah serta kecepatan angin. Pada tahap pemodelan prediksi, digunakan dua parameter utama yaitu tinggi muka air dan curah hujan yang diperoleh dari stasiun tengah sebagai indikator dini bagi wilayah hilir, sedangkan parameter lingkungan lainnya digunakan sebagai data monitoring pendukung. Data diolah menggunakan metode Long Short Term Memory (LSTM) sebagai model utama dan Polynomial Features Linear Regression sebagai pembanding, dengan evaluasi pada tiga horizon prediksi yaitu 1 jam (H+1), 6 jam (H+6), dan 24 jam (H+24) ke depan. Hasil pengujian menunjukkan akurasi tertinggi pada prediksi jangka pendek dan menurun seiring bertambahnya horizon. Pada H+1, metode Polynomial mencapai R² 0,99 dengan MAE 0,75 cm dan RMSE 4,13 cm, mengungguli LSTM yang mencapai R² 0,98 dengan MAE 2,41 cm. Pada H+6, R² menurun ke kisaran 0,91–0,93, sedangkan pada H+24 akurasi kedua metode menurun signifikan hingga R² bernilai negatif (−0,08 hingga −0,20). Metode Polynomial Features Linear Regression secara konsisten mengungguli LSTM pada prediksi tinggi muka air, sementara prediksi curah hujan menghasilkan R² rendah hingga negatif pada seluruh horizon. Dengan demikian, sistem yang dikembangkan mampu berfungsi sebagai pendukung peringatan dini banjir di Sungai Kalibokor, dengan keandalan tertinggi pada prediksi tinggi muka air jangka pendek.
======================================================================================================================================
Surabaya has an extensive network of waterways, one of which is the Kalibokor River, which serves as the primary waterway in the eastern part of the city. This river frequently overflows, with water levels reaching 0–15 cm due to heavy rainfall and a passive drainage system, causing localized flooding particularly in downstream areas such as the Keputih neighborhood. To date, there is no water level monitoring system available that can provide realtime and continuous information on river conditions, making it difficult to identify potential flooding early on and forcing reliance solely on rainfall data from the BMKG without accounting for other environmental parameters in realtime. This study aims to design an Internet of Things (IoT) based water level monitoring and prediction system capable of providing realtime data on river conditions while also forecasting potential flooding early on. The system utilizes several sensors to measure environmental parameters such as temperature, light intensity, rainfall, water level, and wind direction and speed. In the prediction modeling stage, two main parameters water level and rainfall obtained from the midstream station are used as early indicators for downstream areas, while other environmental parameters serve as supporting monitoring data. The data were processed using the Long Short Term Memory (LSTM) method as the primary model and Polynomial Features Linear Regression as a comparison, with evaluation conducted across three forecast horizons: 1 hour (H+1), 6 hours (H+6), and 24 hours (H+24) into the future. The test results showed the highest accuracy in short-term predictions, which decreased as the forecast horizon increased. At H+1, the Polynomial method achieved an R² of 0.99 with an MAE of 0.75 cm and an RMSE of 4.13 cm, outperforming LSTM, which achieved an R² of 0.98 with an MAE of 2.41 cm. At H+6, R² decreased to the range of 0.91–0.93, while at H+24, the accuracy of both methods declined significantly to negative R² values (−0.08 to −0.20). The Polynomial Features Linear Regression method consistently outperformed the LSTM method in water level predictions, whereas rainfall predictions yielded low to negative R² values across all time horizons. Thus, the developed system is capable of supporting early flood warnings on the Kalibokor River, with the highest reliability in short-term water level predictions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Internet of Things, Long Short Term Memory, Polynomial Features Linear Regression, prediksi kebanjiran, sistem peringatan dini. early warning system, flood prediction, Internet of Things, Long Short Term Memory, Polynomial Features Linear Regression.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Kafka Kautsari
Date Deposited:	01 Aug 2026 02:58
Last Modified:	01 Aug 2026 02:58
URI:	http://repository.its.ac.id/id/eprint/141248

Actions (login required)

View Item