PREDIKSI TOTAL TAGIHAN LISTRIK PELANGGAN TARIF P1 ULP SUKADANA MENGGUNAKAN METODE LONG SHORT-TERM MEMORY (LSTM)

Rida Fitriani (Program Studi Matematika, Fakultas Sains, Institut Teknologi Sumatera)
Werry Febrianti (Program Studi Matematika, Fakultas Sains, Institut Teknologi Sumatera)

MATHunesa: Jurnal Ilmiah Matematika
Vol. 14 No. 02 (2026)

This study aims to predict the total electricity bills of P1 tariff customers at the Sukadana Customer Service Unit (ULP) using the Long Short-Term Memory (LSTM) method. In this study, the data used is historical monthly data from January 2023 to June 2025, which includes energy consumption (kWh), hours of use, power, working hours, and total bills. The research process includes data collection, separation of data into training and test data, parameter determination, model training, tuning, and evaluation using Mean Absolute Percentage Error (MAPE). The tuning results showed the best configuration in 56 LSTM units, 12 Dense units, and a learning rate of 0.1. The model produced a MAPE value of 2.89% for training data and 5.81% for test data, which was categorized as highly accurate. This model was then used to predict the electricity bill for July 2025, with an estimated result of IDR 433,344,576.00. The results of this study show that the LSTM method is capable of effectively recognizing time series data patterns and can be used as a tool for more accurate electricity bill estimates, as well as supporting decision-making in operational planning at PLN UP3 Metro.