M-Factors Fuzzy Time Series for Forecasting Moving Holiday Electricity Load Demand in Malaysia (S/O 14589)

• Main Authors: Mansor, Rosnalini, Mat Kasim, Maznah, Othman, Mahmod, Zaini, Bahtiar Jamili
• Format: Monograph
• Language: English
• Published: UUM
• Subjects: QA Mathematics
• Online Access: https://repo.uum.edu.my/id/eprint/30213/1/14589.pdf; https://repo.uum.edu.my/id/eprint/30213/
• Institution: Universiti Utara Malaysia

Moving holiday is a non-fixed holiday according to the Gregorian calendar. Most of the electricity load demand studies showed that this event affects the accuracy of load forecasting. It is due to a limited recent historical data about moving holiday, and a longer time series is acquired to reveal the pattern. Besides, different characteristics of each moving holiday and existence of a great amount of irregularities in the load data also contribute to the forecasting inaccuracy and uncertainty. Even though fuzzy time series (FTS) algorithm is able to overcome moving holiday electricity load demand (MH-ELD) forecasting problem, the current FTS algorithm lacks final model interpretation, is less interpretability of fuzzy logical relationship strength, and does not provide a complete FTS forecasting process. These will provide less information about the relationship that naturally represents how humans make judgments and decisions, and less guide to conduct complete FTS forecasting process. The FTS algorithm should be improved to overcome the cons of previous FTS algorithms. Therefore, the aim of this study is to modify the conventional FTS algorithm by applying weighted subsethood in FTS algorithm on 2017-2018 segmented Malaysia electricity load demand time series data. The modified algorithm, Weighted Subsethood Segmented Fuzzy Time Series (WeSuSFTS) consists of four main phases; data pre-processing, model development, model implementation and model evaluation. The modified algorithm uses the min-max operator for fuzzy reasoning and peak-point defuzzification which make the process simpler. Two types of WeSuSFTS: 1-factor and M-factors WeSuSFTS are also executed. The results show that the WeSuSFTS model has higher accuracy compared to the classical FTS models. As one of the alternative forecasting methods to forecast MH-ELD, 1-factor WeSuSFTS model has less mean absolute percentage error than M-factors WeSuSFTS model and two other conventional FTS models. Hence, the WeSuSFTS algorithm succeeds to improve the MH-ELD forecasting accuracy.