Penentuan Nilai Estimasi Pemakaian Gas Bumi Pelanggan Rumah Tangga Menggunakan Metode Exponential Smoothing

Suminar, Ratna Dian (2024) Penentuan Nilai Estimasi Pemakaian Gas Bumi Pelanggan Rumah Tangga Menggunakan Metode Exponential Smoothing. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032212133_Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 2 February 2026. Download (13MB) | Request a copy

Abstract

Nilai tagihan gas pelanggan rumah tangga dapat diperoleh dari pencatatan meter aktual, catat meter mandiri, dan estimasi. Peran estimasi krusial dalam proses bisnis perusahaan karena pencatatan meter aktual hanya dilakukan setiap tiga bulan dan rendahnya partisipasi catat meter mandiri. Nilai estimasi eksisting dihitung dari rata-rata pemakaian gas 12 (dua belas) bulan sebelumnya. Banyaknya keluhan pelanggan terkait lonjakan tagihan menjadi salah satu indikator perlunya peningkatan akurasi estimasi. Ketiadaan analisa pola pemakaian pelanggan, penggunaan data historis yang tidak sesuai, pengambilan keputusan secara manual dan subyektif, serta keterbatasan penggunaan metode eksisting bagi pelanggan dengan masa berlangganan gas kurang dari 1 (satu) tahun, menjadi latar belakang pelaksanaan penelitian ini. Metode Exponential Smoothing dipilih karena dapat menyesuaikan diri dengan tren dan musiman yang terjadi, cocok untuk data historis yang relatif pendek, langkah kerja sederhana, dan memberikan akurasi yang baik untuk peramalan jangka pendek. Penelitian menggunakan data pemakaian gas bumi dari bulan Juli 2022 hingga September 2023 pada 1993 pelanggan rumah tangga di Area Sidoarjo. Estimasi dihitung menggunakan metode eksisting dan metode Exponential Smoothing yang dioptimalisasi pada tiga variasi data training: dua belas bulan, sembilan bulan dan enam bulan. Variasi data training terbaik ditentukan berdasarkan nilai Sum Square Error (SSE) terkecil serta akurasi hasil peramalan dari metode ekstisting dan Exponential Smoothing dihitung menggunakan nilai Mean Absolute Error (MAE) dan Mean Square Error (MSE). Penelitian menghasilkan parameter pemulusan level dan tren yang spesifik untuk setiap pelanggan, tanpa parameter pemulusan musiman untuk seluruh dataset, yang berarti bahwa metode Exponential Smoothing optimal adalah antara Single Exponential Smoothing dan Double Exponential Smoothing. Variasi data training enam bulan memiliki nilai SSE terkecil dan penambahan data tidak memberikan dampak signifikan pada peningkatan akurasi model. Secara akumulatif klaster, metode Exponential Smoothing memberikan nilai MAE dan MSE yang lebih kecil daripada metode eksisting dan secara individual memberikan akurasi peramalan yang terbaik pada hampir 70% pelanggan.
==================================================================================================================================
The household customers gas bills are derived from actual meter reading, self-recorded reading by customer and estimations. Estimation plays a crucial role in the business process due to the infrequency of actual meter readings (every three months and low customer participation in self-recorded readings. The existing estimates are based on the average gas usage over the previous twelve months. The number of customer complaints about bill spikes indicating the necessity for improved accuracy in estimations. This research addresses the lack of customer usage pattern analysis, the use of inappropriate historical data, manual decision-making, subjectivity, and limitations in existing methods for customers with subscriptions under one year. The choice of the Exponential Smoothing method is justified by its ability to accommodate trends and seasonality, suitability for short historical data, simplicity in workflow, and accuracy in short-term forecasting. Gas usage data from July 2022 to September 2023 for 1993 households in Sidoarjo were utilized. Estimations were performed using both existing methods and Exponential Smoothing, optimized with three data training variations: twelve months, nine months, and six months. The best data training was determined based on the smallest Sum Square Error (SSE) value. Subsequently, accuracy comparison of forecasting results using Mean Absolute Error (MAE) and Mean Squared Error (MSE) values were conducted between the two methods. The research yielded specific smoothing parameters for level and trend for each customer, with no seasonal smoothing parameters for the entire dataset, which implies that the optimal Exponential Smoothing method lies between Single Exponential Smoothing and Double Exponential Smoothing. The six-month data training had the smallest SSE value, and additional data had minimal impact on the model accuracy improvement. In cumulative cluster analysis, Exponential Smoothing method outperformed the existing method, providing smaller MAE and MSE values and individually enhancing forecasting accuracy for almost 70% of customers.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	data time series, estimasi, Exponential Smoothing, pelanggan gas bumi rumah tangga, peramalan, estimation, Exponential Smoothing, forecasting, household natural gas customer, time series
Subjects:	T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Ratna Dian Suminar
Date Deposited:	12 Feb 2024 00:35
Last Modified:	12 Feb 2024 00:35
URI:	http://repository.its.ac.id/id/eprint/106895

Actions (login required)

View Item