Solar Power Forecasting Using Hybrid Model

The significance of forecasting has been increasing in power systems due to the more usage of renewable energy sources. Renewable energy sources such as solar energy, which is very intermittent in nature. Hence, to capture this nature, Deep learning (DL) based techniques are required. Because DL-based techniques learn quickly from patterns and deal with intermittent changes in load and renewable sources. This study proposes DL technique-based convolution neural network (CNN) - long short-term memory (LSTM) hybrid model to forecast solar energy. This allows the LSTM to learn features from the input data that have been learned by CNN. Further, 15 minutes of solar power generation data for one year from Visvesvaraya National Institute of Technology (VNIT), Nagpur is used as the benchmark data to evaluate the hybrid model. This study considers a novel meteorological factor and compares that with without considering a meteorological factor. Moreover, CNN and LSTM individual DL models were developed, analyzed, and tested for the same VNIT data for forecasting the 15-minute solar power generation. The proposed hybrid model is compared with individual LSTM and CNN models in terms of performance evaluation matrices i.e., The root mean squared error (RMSE), the mean squared error (MSE), the mean absolute error (MAE), the mean absolute percentage error (MAPE), and the coefficient of determination (R-Squared) are all measures of error. The percentage improvement from the individual model and the proposed hybrid model has been evaluated with and without considering meteorological factors. The proposed model improves the efficiency by 0.45% and 21.31%, as compared to LSTM and CNN by considering metrological factors respectively. Similarly, it improves without considering metrological factors 1.01% and 21.59% for LSTM and CNN respectively.