Sistem Rekonstruksi Data Time Series Resolusi Tinggi Pada Data Cuaca Sparse Berbasis IoT dengan Metode Generative Adversarial Imputation Network

Prasetya, Rangga Arya (2026) Sistem Rekonstruksi Data Time Series Resolusi Tinggi Pada Data Cuaca Sparse Berbasis IoT dengan Metode Generative Adversarial Imputation Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221072-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Sistem weather monitoring berbasis IoT memberikan data cuaca time series dengan resolusi temporal yang tinggi, seperti pada interval setiap 30 detik guna mempresentasikan dinamika lingkungan secara akurat. Namun, dengan pengiriman data secara kontinu dengan interval yang rapat menyebabkan redudansi informasi serta peningkatan beban komunikasi dan konsumsi daya. Dengan mengurangi frekuensi transmisi menjadi setiap 2 menit dapat menurunkan beban komunikasi sistem, tetapi menyebabkan lebih dari 50% data hilang dan kontinuitas informasi menurun. Penelitian ini bertujuan untuk mempertahankan resolusi informasi agar tetap tinggi pada kondisi data jarang melalui pendekatan rekonstruksi data dengan metode GAIN time series. Dengan pengambilan data resolusi tinggi sebagai data acuan, simulasi pengurangan transmisi, penerapan metode rekonstruksi untuk mengestimasi data yang hilang, serta melakukan evaluasi akurasi hasil menggunakan metrik kesalahan terhadap data asli. Didapatkan model terbaik yaitu GAIN (Generative Adversarial Imputation Network) yang merupakan metode generalisasi dari GAN, dengan tambahan komponen “petunjuk” agar proses adversarial sesuai target. GAIN dengan transformasi arah angin sinus dan cosinus menghasilkan validasi error RMSE terkecil pada missing 75% yaitu, 0,223% dibandingkan dengan GAIN dasar 0,240%, GAIN time series 0,232% dan GAIN dengan pretraining 0,504%. Dengan split data 70% training, 20% validation dan 10% testing, didapatkan hasil terbaik interval pengiriman data adalah 2 menit atau dengan missing rate 75% dari sampling 30 detik. Dengan interval ini, konsumsi energi yang awalnya 77,7 Wh berkurang menjadi 64,2 Wh dalam 10 jam.
==============================================================================================================================
IoT based weather monitoring systems provide time series weather data with high temporal resolution such as at 30 second intervals to accurately represent environmental dynamics. However, continuous data transmission at such frequent intervals leads to information redundancy as well as increased communication load and power consumption. Reducing the transmission frequency to every 2 minutes can lower the system’s communication load, but results in more than 50% data loss and reduced information continuity. This study aims to maintain high information resolution under sparse data conditions through a data reconstruction approach using the time series GAIN method. Using high resolution data as a reference, the study simulates transmission reduction, applies reconstruction methods to estimate missing data, and evaluates the accuracy of the results using error metrics against the original data. The best model identified was GAIN (Generative Adversarial Imputation Network), a generalized version of GAN that incorporates an additional “hint” component to ensure the adversarial process aligns with the target. GAIN with sine and cosine wind direction transformations produced the smallest RMSE validation error on the 75% missing data set, namely 0.223%, compared to the baseline GAIN of 0.240%. the time series GAIN at 0.232%, and the GAIN with pretraining at 0.504%. With a data split of 70% for training, 20% for validation, and 10% for testing, the best results were obtained with a data transmission interval of 2 minutes or a 75% missing rate from 30 second sampling. With this interval, energy consumption, which was initially 77.7 Wh, decreased to 64.2 Wh over 10 hours.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Data Reconstruction, Generative Adversarial Imputation Network, Time Series, Weather Monitoring.
Subjects:	Q Science Q Science > Q Science (General) Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Rangga Arya Prasetya
Date Deposited:	05 Aug 2026 02:42
Last Modified:	05 Aug 2026 02:50
URI:	http://repository.its.ac.id/id/eprint/143486

Actions (login required)

View Item