Energy Consumption Forecasting in Smart Grids with Time Series Models

Abstract

As global energy demand surges, smart grids have emerged as a critical innovation to efficiently manage energy resources and optimize consumption. The ability to accurately forecast energy consumption in smart grids not only ensures reliable energy distribution but also supports sustainability efforts by reducing wastage and integrating renewable energy sources. Time series models have gained prominence for their efficacy in predicting energy consumption trends due to their ability to account for temporal dependencies and seasonality in energy data. This paper investigates the application of various time series models—including Auto-Regressive Integrated Moving Average (ARIMA), Seasonal Decomposition of Time Series (STL), Long Short-Term Memory (LSTM), and Prophet—in forecasting energy consumption in smart grids. The performance of these models is evaluated on real-world datasets, considering metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE).

The results indicate that while traditional models like ARIMA excel in short-term forecasting with stationary data, advanced deep learning models such as LSTM demonstrate superior performance in capturing long-term dependencies and nonlinear patterns. Prophet’s ease of implementation and flexibility also make it a viable choice for energy forecasting in dynamic grid environments. Furthermore, the study highlights the challenges in data preprocessing, including handling missing data, outliers, and data normalization, which are critical for improving forecasting accuracy. This work contributes to the growing body of knowledge by offering practical insights into selecting and optimizing time series models for smart grid applications.

DOI - https://doi.org/10.65525/SVUP.9788199565432.2026.107-111