Electricity consumption forecasting using ARIMA and LSTM

Abstract

Accurate load forecasting is essential for the reliable and efficient operation of modern power systems. This study presents a comparative analysis of two prominent forecasting models-Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM)-to assess their effectiveness in predicting electricity consumption. Both models were developed and fine-tuned through hyperparameter optimization to ensure fair and optimal performance. The evaluation considered predictive accuracy, computational efficiency, and resource usage. While ARIMA demonstrated advantages in inference speed and model simplicity, the LSTM model consistently outperformed it in terms of forecasting accuracy and its ability to capture complex temporal dependencies. These findings underscore the importance of selecting appropriate models and tuning strategies for specific forecasting scenarios. The study highlights LSTM as a more suitable approach for applications that demand high accuracy and adaptability, and it provides a foundation for future research involving advanced or hybrid methods.Precizno prognoziranje potrošnje električne energije ključno je za pouzdan i efikasan rad savremenih elektroenergetskih sistema. Ova studija prikazuje uporednu analizu dva istaknuta modela za prognozu - Autoregressive Integrated Moving Average (ARIMA) i Long Short-Term Memory (LSTM) - sa ciljem procene njihove efikasnosti u predviđanju potrošnje električne energije. Oba modela su razvijena i optimizovana radi pravičnog poređenja i najboljih mogućih performansi. Procena modela je obuhvatila tačnost predviđanja, računarsku efikasnost i upotrebu memorijskih resursa. Iako je ARIMA pokazala prednosti u brzini predviđanja i jednostavnosti modela, LSTM je dosledno davao preciznije prognoze i bolje prepoznavao složene vremenske obrasce. Rezultati ukazuju na značaj pažljivog izbora modela i strategije podešavanja za konkretne scenarije prognoziranja. Studija ističe LSTM kao pogodniji pristup za primene koje zahtevaju visoku tačnost i prilagodljivost, te pruža osnovu za buduća istraživanja naprednih ili hibridnih metoda.