Pengembangan Sistem Peramalan Jangka Pendek Untuk Memprediksi Beban Pendinginan Bangunan Komersial Dengan Metode Quantile Regression Forest

Hanafi, Reynaldi Rahmad Syahputra (2024) Pengembangan Sistem Peramalan Jangka Pendek Untuk Memprediksi Beban Pendinginan Bangunan Komersial Dengan Metode Quantile Regression Forest. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 02311940000080-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 July 2026. Download (2MB) | Request a copy

Abstract

Tugas akhir ini fokus pada penerapan teknik Machine Learning dalam memprediksi beban pendinginan pada bangunan komersial. Metode ini memungkinkan sistem untuk mempelajari pola dari data masa lalu dan menghasilkan prediksi akurat untuk kejadian di masa depan tanpa instruksi langsung. Dalam konteks konsumsi energi bangunan komersial, yang dapat mencapai 40%-60% dari total, peramalan beban pendinginan menjadi krusial untuk mengoptimalkan sistem HVAC. Pendekatan data-driven, terutama menggunakan metode Quantile Regression Forest (QRF), digunakan untuk meningkatkan akurasi prediksi. Tinjauan terhadap penelitian terdahulu menunjukkan bahwa model non-linier, seperti QRF, mampu memberikan hasil prediksi yang lebih akurat dibandingkan model linier. Namun, sebelumnya, banyak penelitian hanya mempertimbangkan data beban pendinginan historis dan temperatur saat itu tanpa memasukkan fitur humidity, yang ternyata memiliki pengaruh signifikan pada beban pendinginan. Penelitian ini mengambil langkah lebih lanjut dengan mengintegrasikan fitur humidity dalam model peramalan. Tingkat akurasi dari model yang digunakan dapat mencapai persentase error kisaran 15.92% pada kuantil 50%, 17.48% pada kuantil 10%, dan 17.34% pada kuantil 90%, yang mana ini berarti model sudah memiliki tingkat akurasi yang cukup baik.
=====================================================================================================================================
This final project focuses on the application of Machine Learning techniques to predict cooling loads in commercial buildings. This method allows the system to learn patterns from historical data and generate accurate predictions for future events without direct instructions. In the context of commercial building energy consumption, which can account for 40%-60% of the total, forecasting cooling loads becomes crucial for optimizing HVAC systems. A data-driven approach, particularly using the Quantile Regression Forest (QRF) method, is employed to enhance prediction accuracy. A review of previous research indicates that non-linear models, such as QRF, can provide more accurate predictions compared to linear models. However, earlier studies often considered only historical cooling load data and current temperature, neglecting humidity features, which, as it turns out, significantly influence cooling loads. This research takes a step further by integrating humidity features into the forecasting model. The accuracy level of the model used can reach a percentage error of around 15.92% at the 50% quantile, 17.48% at the 10% quantile, and 17.34% at the 90% quantile, which means the model already has a fairly good level of accuracy.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Machine Learning, Cooling Load, Quantile Regression Forest, Interval Forecast; Machine Learning, Beban Pendinginan, Quantile Regression Forest, Interval Forecast.
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Reynaldi Rahmad Syahputra Hanafi
Date Deposited:	06 Feb 2024 03:26
Last Modified:	06 Feb 2024 03:26
URI:	http://repository.its.ac.id/id/eprint/106219

Actions (login required)

View Item