Short-Term Prognostication of Solar Power Generation Using Hybrid VAR-CNN-LSTM Network Topology

The modern microgrid contains various renewable energy sources that are intermittent. Due to the stochastic nature of the sources, it is essential to predict solar power for effective integration. The evolution of deep learning makes the prediction simple with complex and non-linear temporal data. In this paper, a hybrid Vector Auto Regressive (VAR) model, Convolutional neural network (CNN) and Long-short term memory (LSTM) are proposed to predict solar power using real-time data from Tamilnadu, India. The VAR-CNN-LSTM model employs the VAR model and deep learning network that combines CNN-LSTM layers to capture linear and non-linear data features. In this proposed hybrid method, the VAR model capture linear feature of the data, the CNN layer will extract the hierarchical structure from the various weather parameters, and LSTM will extract the long-term temporal characteristics of the data. The results obtained in this study for the proposed model is then compared with distinct VAR, CNN, LSTM, and hybrid CNN-LSTM models to verify the performance of the model.