Efficient Method of Finding the Best Time Interval for Predicting Short Term Solar Radiation Using CNN-LSTM Model

Even though the application of solar energy for electric power production is rapidly growing throughout the world, its unpredictability continues to provide significant difficulties. The primary source of this issue is the fluctuation of the solar radiative power, which the Photovoltaic (PV) cells convert into electrical energy at the power plants. In determining the best time interval or horizon for solar irradiance forecasting, the CNN-LSTM hybrid model was used. The input data consisted of historical solar irradiance obtained at five different time intervals over two years period. The dataset was created using historical meteorological data for Cape Town for two years. Eighty percent of the whole dataset was used to train the model for up to 1,000 epochs. The metric deployed to assess the model’s performance was Root Mean Squared Error (RMSE). Results from this experiment were compared with those from the Support Vector Regression (SVR) model that was fitted independently using a similar volume of data. From the performance metrics analyzed, the CNN-LSTM achieved better results than the SVR model. It recorded an RMSE of 6.2 percent using training data collected at 5-minute intervals. This result was best when contrasted with others obtained when the models were trained with data obtained from other different horizons. Adopting the data produced by the CNN-LSTM hybrid model at the five-minute horizon in Cape Town is suggested to improve control over the issues caused by fluctuating solar radiative power on the power system smart grid.