A Multiple Optimizer Based Long Short-Term Memory Models for the Efficient Prediction of Solar Radiation

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

To ensure energy security and environmental sustainability, transition to renewable energy sources is required, one of the most important and sustainable renewable energy is solar energy, However, in developing predicting system, accurate solar radiation is essential for optimizing solar energy systems particularly in regions with limited ground based infrastructures including Northwest Nigeria, To address this challenge a time series dataset of daily solar radiation measurements from available meteorological parameter, several previous studies have used various methods including ARIMA, SARIMA, Machine learning and Deep leaning models with static hyperparameter optimization approach, Meanwhile optimization of forecasting systems is necessary for developing accurate energy predicting systems. Deep learning is effective in solar radiation forecasting. To evaluate the performance of deep learning method for daily solar prediction, an optimized Long Short-Term Memory (GWO-LSTM and BO-LSTM) based deep learning models were developed in the study, The optimized models were created using daily solar radiation data from Nigerian Meteorological Agency for a period of 20 years, in three states in Northwest Nigeria: Kaduna, kano and katsina, This models were evaluated using two forecasting performance metrics: R2 coefficient of determination and Root Mean Square Error (RMSE), The models results shows that both optimized models outperformed baseline LSTM model, The LSTM-BO indicates performance with training R2 = 0.8500, RMSE=0.3876 and testing of R2 = 0.8205, RMSE = 0.4309 while GWO-LSTM achieves training of R2 = 0.8358, RMSE = 0.4054 and testing of R2 = 0.8190, RMSE = 0.4340 respectively. The LSTM-BO achieves slightly better predicting accuracy than LSTM-GWO.

Links and sources

• PDF from original source

Comments