Big Data and Deep Learning Analytics for Robust PV Power Forecast in Smart Grids

Abstract

Photovoltaic (PV) power generation is intermittent and volatile in nature, rendering its large-scale deployment a challenge for the smart electricity grid’s operation safety, stability, and economic efficiency. Ultra-fast and accurate prediction of PV power helps effectively adjusting the dispatch schedules during different operating states the power grid may undergo. This chapter proposes a deep learning-based PV power forecasting approach, the so-called Chaotic-LSTM, which ensembles the principles of the long short-term memory (LSTM) neural network and chaos theory. The LSTM neural network is used to construct a nonlinear mapping between input and output variables, while the phase space reconstruction technology in chaos theory is used to analyze the nonlinear time series of PV power generation, and extract the intrinsic dynamic characteristics of the PV arrays. Finally, a correlation analysis is applied to extract the external factors influencing the PV arrays. The effectiveness of the Chaotic-LSTM technology is demonstrated by comparing with three state-of-the-art neural network models: back-propagation, radial basis function, and simple recursive Elman neural networks. The accuracy of the proposed method was assessed using four different forecasting time horizons (i.e., one-hour, four-hour, one-day, and four-day-ahead) and three evaluation metrics. Additional tests are conducted with seven levels of signal-noise ratio to provide a measure of model robustness and effectiveness. Numerical results will demonstrate that the proposed Chaotic-LSTM method can significantly improve the prediction accuracy of the short-term PV power generation.