Advanced wind/photovoltaic power smoothing using LSTM neural networks and machine learning

The integration of stochastic renewable energy sources, like wind turbines and photovoltaic systems, into electrical grids introduces challenges to grid stability and reliability, leading to voltage and frequency deviations. This study addresses the fluctuation issue by examining hybrid energy storage systems combining batteries and supercapacitors. A novel power
smoothing approach is proposed, involving two strategies: employing LSTM neural networks for short-term prediction of RES power profiles and optimizing HESS through charge/discharge cycle control using a machine learning-based algorithm. This paper also introduces the synergy of vanadium redox flow batteries and supercapacitor for efficient energy storage. The proposed approach is validated through experimental testing in a controlled microgrid setting. The evaluation demonstrates significant improvements, including a 74.2% reduction in power fluctuations and an enhanced smoothing quality evaluation index by up to 40%, surpassing conventional methods like moving average, ramp rate, and low pass filter. The contributions of this research encompass an advanced energy smoothing methodology, streamlined storage
integration, and an enhanced energy quality framework for hybrid renewable energy systems.