Comparative Assessment of the Patterns of Solar Irradiance from Multiple Locations Using Deep Learning Methods

Accurate forecasting is critical in mitigating the negative effects of intermittent solar radiation when producing stable electricity from photovoltaic panels at a solar plant. Recently, artificial intelligence-based algorithms to predict solar radiation have dramatically improved solar power production efficiency. Predicting solar radiation with any degree of certainty depends on a number of variables, including the model trained with the data, the location of the data collection, the time horizon for data collection, and the quality of the training data. Solar radiation projections based on data obtained from a single place at specific time intervals predominate in the existing literature. The effect of data collection sites on model accuracy has been disregarded in this practice. Some locations are more susceptible to lower solar radiation and a more frequent cycle of intermittent solar radiation than others. As a result, gathering training data from multiple locations is required to determine the best location for establishing a solar power farm. Historical weather data from Cape Town, Durban, Limpopo, and Johannesburg were used in this study to forecast solar irradiance at 10-minute intervals. Based on the results of the experiments in this study, Limpopo is regarded as the best location for establishing a solar plant. In Limpopo, the MLP model had the lowest Root Mean Square Error (RMSE) of 18.87. This was the best result from the study and is considered to be above expectations. The RF model produced an RMSE value of 22.51 in Cape Town, the second-best location. Limpopo > Cape Town > Durban > Johannesburg is the order of performance. In these cities, the MLP model produced an RMSE of 18.87 with Limpopo data, the RF model produced an RMSE of 22.51 with Cape Town data, the MLP produced the lowest RMSE of 23.52 with Durban training data, and the ConvLSTM produced an RMSE of 24.85 with Johannesburg training data.