Demand Forecasting in Data-Scarce and Resource-Restricted Environments

With the growing integration of renewable energy technologies, forecasting demand in residential settings is becoming increasingly important. This research addresses the challenge of forecasting energy demand in data-scarce, resource-restricted environments, where traditional static load profiling proves inadequate due to the high variability of energy use. We present a lightweight average-based approach aimed at modeling dynamic user behavior in settings with limited historical data. This approach divides each day into evenly spaced time intervals and various average user demands, while forecasts are made and refined based on on-site data to account for short-term fluctuations. The proposed approach was compared with ARIMA and neural networks and tested on three publicly available datasets with differing temporal resolutions and data volumes. The results demonstrate that our approach is well suited in settings with little or no historical data, whereas ARIMA and neural networks outperform the average-based approach as the quantity of data increases. The proposed algorithm allows rapid integration into existing demand forecasting systems, providing adaptive demand forecasting for residential environments with restricted resources.