TY - JOUR
T1 - Bayesian optimisation algorithm based optimised deep bidirectional long short term memory for global horizontal irradiance prediction in long-term horizon
AU - Madhiarasan, Manoharan
PY - 2025
Y1 - 2025
N2 - With the continued development and progress of industrialisation, modernisation, and smart cities, global energy demand continues to increase. Photovoltaic systems are used to control CO2 emissions and manage global energy demand. Photovoltaic (PV) system public utility, effective planning, control, and operation compels accurate Global Horizontal Irradiance (GHI) prediction. This paper is ardent about designing a novel hybrid GHI prediction method: Bayesian Optimisation algorithm-based Optimized Deep Bidirectional Long Short Term Memory (BOA-D-BiLSTM). This work attempts to fine-tune the Deep Bidirectional Long Short Term Memory hyperparameters employing Bayesian optimisation. Globally ranked fifth in solar photovoltaic deployment, the INDIA Two Region Solar Irradiance Dataset from the NOAA-National Oceanic and Atmospheric Administration was used to assess the proposed BOA-D-BiLSTM approach for the long-term prediction horizon. The superior prediction performance of the proposed BOA-D-BiLSTM is highlighted with the help of experimental results and comparative analysis with grid search and random search. Furthermore, the forecasting effectiveness is compared with other models, namely, the Persistence Model, ARIMA, BPN, RNN, SVR, Boosted Tree, LSTM, and BiLSTM. Compared to other forecasting models according to the resulting evaluation error metrics, the suggested BOA-D-BiLSTM model has minor evaluation error metrics, namely, Root Mean Squared Error: 0.0026 and 0.0030, Mean Absolute Error:0.0016 and 0.0018, Mean-Squared Error: 6.6852 × 10−06 and 8.8628 × 10−06 and R-squared: 0.9994 and 0.9988 on both dataset 1 and 2 respectively. The proposed BOA-D-BiLSTM model outperforms other baseline models. Thus, the proposed BOA-D-BiLSTM is a viable and novel potential forecasting model for effective distributed generation planning and control.
AB - With the continued development and progress of industrialisation, modernisation, and smart cities, global energy demand continues to increase. Photovoltaic systems are used to control CO2 emissions and manage global energy demand. Photovoltaic (PV) system public utility, effective planning, control, and operation compels accurate Global Horizontal Irradiance (GHI) prediction. This paper is ardent about designing a novel hybrid GHI prediction method: Bayesian Optimisation algorithm-based Optimized Deep Bidirectional Long Short Term Memory (BOA-D-BiLSTM). This work attempts to fine-tune the Deep Bidirectional Long Short Term Memory hyperparameters employing Bayesian optimisation. Globally ranked fifth in solar photovoltaic deployment, the INDIA Two Region Solar Irradiance Dataset from the NOAA-National Oceanic and Atmospheric Administration was used to assess the proposed BOA-D-BiLSTM approach for the long-term prediction horizon. The superior prediction performance of the proposed BOA-D-BiLSTM is highlighted with the help of experimental results and comparative analysis with grid search and random search. Furthermore, the forecasting effectiveness is compared with other models, namely, the Persistence Model, ARIMA, BPN, RNN, SVR, Boosted Tree, LSTM, and BiLSTM. Compared to other forecasting models according to the resulting evaluation error metrics, the suggested BOA-D-BiLSTM model has minor evaluation error metrics, namely, Root Mean Squared Error: 0.0026 and 0.0030, Mean Absolute Error:0.0016 and 0.0018, Mean-Squared Error: 6.6852 × 10−06 and 8.8628 × 10−06 and R-squared: 0.9994 and 0.9988 on both dataset 1 and 2 respectively. The proposed BOA-D-BiLSTM model outperforms other baseline models. Thus, the proposed BOA-D-BiLSTM is a viable and novel potential forecasting model for effective distributed generation planning and control.
KW - and global horizontal irradiance
KW - bayesian optimisation algorithm
KW - bidirectional long short term memory
KW - deep learning
KW - hyperparameters
KW - long-term horizon
KW - prediction
UR - http://www.scopus.com/inward/record.url?scp=86000290650&partnerID=8YFLogxK
U2 - 10.3389/fenrg.2025.1499751
DO - 10.3389/fenrg.2025.1499751
M3 - Journal article
SN - 2296-598X
VL - 13
SP - 1499751
JO - Frontiers in Energy Research
JF - Frontiers in Energy Research
M1 - 1499751
ER -