Recently, photovoltaic (PV) cell technology has witnessed significant developments and improvements. However, solar power is still unattractive to some consumers due to its unpredictability. Therefore, accurate prediction of solar irradiance continues to be critical for stable solar power. In this article, a proposed methodology to improve PV systems' reliability and stability, using state-of-the-art machine learning techniques, is introduced. Time-series long short-term memory (LSTM) network, convolutional LSTM (ConvLSTM), convolutional neural network (CNN), random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost) regression models were used to predict the city of Johannesburg's solar irradiance. These models were separately trained using the historical meteorological data of Johannesburg. Using the normalized root mean square error (nRMSE) as a performance indicator, the ConvLSTM model outperformed all the other models. When the models were individually trained with 80% of one-year input data, the results obtained showed that ConvLSTM recorded an nRMSE value of 4.05%. Results obtained using LSTM, XGBoost, and SVM models came next with nRMSE values of 4.72%, 6.63%, and 7.8%, respectively. The RF model gave an nRMSE value of 19.8%, whereas the CNN time-series model recorded an nRMSE value of 12.61%. The best experimental result obtained was an nRMSE value of 1.51% when the ConvLSTM model was trained at 500 epochs using 80% of ten years of actual historical solar irradiance as the dataset. These results suggest that the adoption of the ConvLSTM model for solar irradiance forecasts by solar farm operators in South Africa could improve electricity generation stability for PV connection to the power grid.

Globally, the consistent clamor by environmentalists for the need to mitigate the effects of climate change has necessitated the adoption of renewable energy sources (RES) for use by many developed/developing nations. The approach is geared towards gradually replacing or reducing the use of fossil fuels for electric power production. Solar power is among the major renewable energy sources in use today. But the use of solar energy is, unfortunately, characterized by fluctuations in its power generation due to the unpredictability of solar irradiance. Despite many methods in use already, accurate forecasting of solar irradiance has continued to be a great need both in the field of physical simulations and artificial intelligence. In this paper, an extreme gradient boosting (XGBoost) regression algorithm was deployed to successfully predict solar power with minimal error. Eighty percent of one-year and five years of historical hourly solar irradiance data of Johannesburg city were separately used as the training dataset. The results obtained using this algorithm were compared with the ones from Support Vector Machine (SVM), to determine the model with the least forecast errors. The results showed that the XGBoost model with an nRMSE value of 6.63% performed better than the SVM model with 6.81%. It is hoped that the implementation of the XGB algorithm for solar irradiance forecasts could greatly improve the stability of solar electric power generated for optimum connection to the power grid.

The increasing global demand for solar energy is a good indicator that it is a viable alternative to fossil energy. However, solar irradiance which is the principal component required for efficient power generation in solar plants is stochastic in nature. This is the reason why the prediction accuracy of solar irradiance for reliable electricity power output continues to be a very difficult task whether in the field of artificial intelligence (AI) or physical simulation. In this paper, Long Short-Term Memory (LSTM) Network, a variant of Recurrent Neural Network (RNN), was used to forecast hourly solar irradiance of Johannesburg city. LSTM is a deep learning network designed to overcome the vanishing and exploding gradient problems associated with a typical RNN. Ten years of historical meteorological data used in this work were obtained from Meteoblue. The simulation results obtained using the LSTM model were compared with the ones recorded using Support Vector Regression (SVR). From the results, it was observed that the LSTM network with normalized Root Mean Square Error (nRMSE) value of 3.2% performed much better than SVR using the same dataset.

Lately, the environmental impacts of fossil energy have forced many countries to begin advocating for renewable energy sources (RES). Solar energy is one of the RES that has gained worldwide acceptability because it is cheap and available. However, solar power is characteristically unpredictable. Accurate prediction of global solar radiation (GSR) continues to be a great need both in the field of physical simulations and artificial intelligence. Both statistical and machine learning methods have proved to be very useful, but researchers are still working on algorithms that can give the least possible error. In this paper, four deep learning algorithms were used to successfully to estimate GSR. Historical meteorological data for Johannesburg city were used as the dataset. The dataset was split into training and testing data. Results obtained showed that Convolutional Long Short-Term Memory (ConvLSTM) with nRMSE value of 1.62% performed better than the other models. Other models used were convolutional neural network (CNN), Long Short-Term Memory Network (LSTM), and a hybrid of CNN and LSTM known as CNN-LSTM network. Real-time implementation of these results to solar plant operations in South Africa could go long to generate a stable electric power for the smart grid.

Photovoltaic (PV) panels need to be exposed to sufficient solar radiation to produce the desired amount of electrical power. However, due to the stochastic nature of solar irradiance, smooth solar energy harvesting for power generation is challenging. Most of the available literature uses machine learning models trained with data gathered over a single time horizon from a location to forecast solar radiation. This study uses eight machine learning models trained with data gathered at various time horizons over two years in Limpopo, South Africa, to forecast solar irradiance. The goal was to study how the time intervals for forecasting the patterns of solar radiation affect the performance of the models in addition to determining their accuracy. The results of the experiments generally demonstrate that the models’ accuracy decreases as the prediction horizons get longer. Predictions were made at 5, 10, 15, 30, and 60 min intervals. In general, the deep learning models outperformed the conventional machine learning models. The Convolutional Long Short-Term Memory (ConvLSTM) model achieved the best Root Mean Square Error (RMSE) of 7.43 at a 5 min interval. The Multilayer Perceptron (MLP) model, however, outperformed other models in most of the prediction intervals.

Accurate prediction of solar irradiance for steady supply of electrical energy has always been a difficult task both in the field of physical simulation or artificial intelligence. If solar irradiance could be predicted with very high precision, it would be easier for power grid operators to take corrective actions in advance to ensure failsafe outcomes during periods of sudden or cascading drops in solar energy. In this paper, three machine learning techniques were used to forecast hourly solar irradiance of Johannesburg city using historical meteorological data of two consecutive years. The goal was to build a model that would readily utilize historical meteorological data to predict highly accurate solar irradiance information on an hourly basis. Results obtained showed that Support Vector Regression (SVR) with normalized Root Mean Square Error (nRMSE) of 7.2% gave the best overall performance. This was followed by Artificial Neural Network (ANN) and Random forest respectively.