Long-term Electricity Consumption Forecasting: How Different Models Forecast into the Future

Time series forecasting is one of the most frequent predictive tasks in various domains, including power consumption and electrical load forecasting. While short-term forecasting of univariate time series data is an extensively studied area, with many existing methods available, long-term forecasting still presents many challenges. An experiment has been designed and conducted to empirically evaluate how different forecasting methods perform on a household electricity consumption dataset. Four machine learning models were trained on historical consumption data and used to predict future consumption for different forecasting horizons into the future. The results show the superiority of deep learning methods over traditional machine learning algorithms for short-term forecasting. However, with the extension of the forecasting horizon, their accuracy drops significantly, and the aggregated errors even exceed those of simpler methods.