<![CDATA[The distribution of 3 kg LPG plays an important role in meeting the energy needs of subsidized households and micro businesses. Therefore, proper distribution planning is needed to maintain supply availability. Fluctuations in distribution volume influenced by trend and seasonal patterns present challenges in estimating future distribution needs. This study aims to analyze the characteristics of 3 kg LPG distribution data, compare the performance of the Seasonal Autoregressive Integrated Moving Average (SARIMA) and Holt-Winters Exponential Smoothing (HWES) models, and implement the forecasting models in a web-based system. The data consist of historical 3 kg LPG distribution records from January 2020 to December 2025, comprising 72 monthly observations, with a 70:30 training and testing data split. The results show that the data exhibit an upward trend and annual seasonal patterns. Based on the testing data evaluation, the SARIMA(1,1,0)(0,1,0)₁₂ model achieved an MAE of 28,140 kg, RMSE of 31,885 kg, and MAPE of 2.35%, while the Holt-Winters Exponential Smoothing Additive model achieved an MAE of 60,938 kg, RMSE of 70,529 kg, and MAPE of 5.17%. SARIMA demonstrated better forecasting performance than Holt-Winters Exponential Smoothing and was used to generate the 2026 distribution forecast. The forecasting results indicate that the highest distribution is expected in December, while the lowest is expected in February. Both models were subsequently implemented in the PollyCast system to support the analysis and planning of 3 kg LPG distribution.]]>
Copyrights © 2026
