Online Stream-Driven Energy Management in Microgrids Using Recurrent Neural Networks and SustainaBoost Augmentation

Abstract

In recent years, the operation of microgrids (MG) has faced increasing challenges due to the growing penetration of renewable energy sources (RES) and the integration of electric vehicles (EVs), which introduce significant uncertainties in power supply and demand dynamics. In response, neural network-based approaches emerge as promising solutions, adept at handling vast databases and learning diverse patterns for real-time decision-making. This paper proposes an online stream-driven energy management strategy for efficient grid-connected MG power management and cost minimization. The strategy considers the presence of EVs and RES, while also addressing the impact of noisy data. The strategy incorporates a recurrent neural network (RNN) to learn from time-series data and make real-time decisions. Additionally, an augmentation technique called SustainaBoost (SB) is introduced, designed to boost system sustainability and enhance the training quality of neural networks. The proposed RNN achieves 98.7% optimality in minimizing the operational costs of the MG on the test dataset.