Advanced machine learning algorithms for reactive power forecasting in electric distribution systems

Due to the rising penetration of distributed generators into the current microgrids, reactive power management has become a crucial concern in terms of voltage stability and resilience of smart grids. In this regard, reactive power forecasting (RPF) is an essential tool for maintaining the reactive power management and planning of active electric distribution systems in which power flow is bidirectional. Machine learning (ML)-based algorithms are frequently applied to electric load forecasting owing to the fact that these methods achieve more accurate results in the short-term horizon. RPF is one of the challenging implementations of electric load forecasting and it can be characterised as a nonlinear problem with a variety of explanatory variables such as active and lagging reactive power values. In this paper, a real-time short-term RPF using ML-based algorithms including long short-term memory (LSTM) networks, random forest (RF), and extreme gradient boosted decision trees (XGBoost) were employed for an electric distribution system located in the North of England, UK. The study also incorporated convolutional neural network (CNN), gated recurrent unit (GRU) networks, and light gradient boosting machine (LightGBM) for benchmarking with the main selected methods. The experimental results demonstrated that LightGBM outperformed other models by achieving the highest accuracy with an R2 of 95.37% and the lowest root mean squared scaled error (RMSSE) of 0.541 while maintaining the shortest computation time of 0.396 s. These findings highlighted the potential of ML-based RPF techniques for improving voltage stability, optimising reactive power compensation, and enhancing energy efficiency in modern smart grids. To the best of our knowledge, there is a lack in the current literature for real-time applications of RPF and this paper is considered to fill this deficiency to create a path for aspiring researchers in the field.