A cohesive structure of Bi-directional long-short-term memory (BiLSTM) -GRU for predicting hourly solar radiation

• Published in: Renewable energy Vol. 222; p. 119943
• Main Authors: Michael, Neethu Elizabeth, Bansal, Ramesh C., Ismail, Ali Ahmed Adam, Elnady, A., Hasan, Shazia
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2024
• Subjects: Bi-directional long short-term memory; Deep learning; Gated recurrent unit; Solar forecasting; Solar irradiance; Stacked LSTM; Deep learning; Solar irradiance; Gated recurrent unit; Solar forecasting; Stacked LSTM; Bi-directional long short-term memory
• ISSN: 0960-1481; 1879-0682
• DOI: 10.1016/j.renene.2024.119943

Uncertain weather scenarios have an impact on the output of solar farms and therefore affect the security of the grid. It is advantageous for power system operators to forecast solar energy to balance the load generation and for optimal power scheduling. The most promising deep-learning techniques to combine weather variables with precise measurements of solar irradiance are not widely discussed. To close this research gap and produce better prediction results, this article aims to formulate and compare two distinctive deep learning algorithms for using time series forecasting approaches to predict solar irradiance. For multivariate data, the forecasting technique Bi-Directional Long Short-Term Memory (BiLSTM), and BiLSTM-GRU (Gated Recurrent Unit) Dropout, are examined in this study. The output results from the proposed model are compared with other benchmark models based on performance error measurements, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Coefficient of Determination (R2) and Mean Absolute Percentage Error (MAPE). It was found that the proposed hybrid method, BiLSTM-GRU with dropout outperformed the other methods in terms of solar irradiance predicting accuracy. The analysis presented the best RMSE of 1.55 and MAE of 1.13 for BiLSTM and RMSE of 1.40 and MAE of 0.91 for BiLSTM-GRU architecture using hyperparameter tuning. The comparison results show that the prediction accuracy is improved by tuning the hyperparameters.