Journal article 94 views
Optimally configured Gated Recurrent Unit using Hyperband for the long-term forecasting of photovoltaic plant
Renewable Energy Focus, Volume: 39, Pages: 49 - 58
Swansea University Author: Shuai Li
Full text not available from this repository: check for access using links below.
The Photovoltaic generation inherits the instability due to the variability and non-availability of solar irradiation at times. Such unstable generation will cause grid management, planning, and operation issues. Researchers have proposed several classical and advanced algorithms to forecast the pow...
|Published in:||Renewable Energy Focus|
Check full text
No Tags, Be the first to tag this record!
The Photovoltaic generation inherits the instability due to the variability and non-availability of solar irradiation at times. Such unstable generation will cause grid management, planning, and operation issues. Researchers have proposed several classical and advanced algorithms to forecast the power generation of photovoltaic plants to avoid such unsuitability issues. Artificial Neural Networks advancement has pushed them in power forecasting, ranging from yearly to hourly prediction of power generation. These networks’ performance depends on the choice of hyperparameters, which includes the number of layers, neurons in each layer, activation function, and learning rate. In power forecasting, these parameters are selected through trial and error, which is inefficient and does not ensure the optimal selection. In this paper, we have presented the Hyperband Gated Recurrent Unit model for power, voltage, and current forecasting of the photovoltaic power plant. The model has a monthly prediction horizon with a temporal resolution of a day. We used the Hyperband technique for the optimal selection of the hyper-parameters. It is an iterative process where several configurations are tried on each trial, and only n best configurations qualify for the subsequent trial. The inputs to the model are irradiation, temperature, and wind speed, and the model's output includes current, voltage, and power. We trained our model on 11 months of data and predicated the outputs for the 12-th month. The network is tested with other variants of Recurrent Neural Networks based on several evaluation metrics. The proposed model achieved promising results with minimum error.
PV Forecasting, Recurrent Neural, HP-GRU, Hyperband, Machine Learning
College of Engineering