Enhanced Short-term Reactive Energy Demand Forecasting by Employing Seasonal Decomposition and Multi-Model Approach

The forecasting of reactive energy demand is an essential task for power system operators as it enables effective planning and management of reactive energy sources. This work proposes a short-term reactive energy forecasting method using seasonal decomposition and a multi-model approach. The main concept of seasonal decomposition is to decompose the time series into trend, seasonal, and residual components. Subsequently, the multi-model approach is used to forecast different components, and each model will be selected based on the components' characteristics. Different models are proposed, including linear regression (LR), convolutional neural network (CNN), and eXtreme Gradient Boosting (XGB) to forecast the reactive energy components. The LR, CNN, and XGB are employed to forecast the trend, seasonality, and residual components, respectively. Moreover, a real-time dataset consisting of the transformer's actual interval meter readings and performance metrics validated the proposed method's effectiveness.