Forecasting of Electricity Usage of Indonesian Consumer Using Transformer Time Series

The rapid growth of the Indonesian electricity market demands more accurate and efficient forecasting methods to ensure the reliability of the national grid and optimize resource allocation. In this study, historical electricity consumption data, along with various relevant features such as weather conditions, demographics, and economic indicators, are used to train deep learning models. The study leverages a rich dataset of historical usage of Indonesian consumer electricity, incorporating various socioeconomic and environmental factors that influence electricity usage fluctuations. Deep learning models, including Transformer Time Series, SVR and LSTM are employed to capture complex patterns and dependencies within the data. The results indicate that Transformer Time Series models outperform traditional time series forecasting methods with result Accuracy 0.92, MAPE is 6.88 and MSE 166.23, Transformer Time Series providing more accurate and reliable predictions of the usage of electricity Indonesian customer.