Short-Term Solar PV Generation Forecast Using Neural Networks and Deep Learning Models

Accurate short-term forecasting of renewable-based generating sources is an essential tool for plants owner and system operator in order to take informed operational decisions. This chapter presents various neural network (NN) and deep learning (DL)-based approaches to forecast the solar PV generation. The random vector functional link (RVFL) model, a variant of NN and long short-term memory (LSTM), a DL-based model are developed to forecast solar PV generation considering a realistic dataset of solar PV plants at IIT Gandhinagar campus. Application of DL-based LSTM and NN-based variants of RVFL models for forecasting solar PV generation for both clear sky and cloudy day is carried out, and a performance comparison between LSTM, RVFL, and wavelet decomposition (WD)-RVFL-based solar PV forecasting algorithms for clear sky is presented. On analysis, it was found that DL-based LSTM model performs better than variants of NN-based RVFL models. As a result, the drawback of shallow NN models over DL models are highlighted. At last, the need for sophisticated DL models over shallow NN models to solve the problem of renewable energy forecasting is stressed.