Solar Irradiance Forecasting Using Deep Learning: Impact of Hyperparameter Optimization

The balance of supply and demand is pivotal in ensuring efficient and reliable power grid utilization. With the growth of demand, the integration of renewable energy resources (RERs) into the power grid is increasing phenomenally. Because of the random nature of RERs, the power grid is vulnerable to frequency and voltage stability issues. Enhancing the performance of generation forecasts is one of the efficient ways to deal with these issues. An accurate generation forecast will facilitate better planning to reduce the impacts caused by the unpredictable nature of RERS. Implementing deep learning (DL) methods in forecasting is becoming more popular as it has been proven to generate forecasts with high accuracy. This paper implements a DL method for solar irradiance forecasting using hyperparameter optimization (HPO). Solcast dataset is utilized to validate the efficacy of the proposed method. The model performance with activation functions, rectified linear unit, Leaky <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{ReLu}$</tex>, and exponential linear unit with and without HPO is compared. Additionally, Bayesian and random search optimization methods are implemented, and their accuracy is also compared. The result shows the significant enhancement of the model performance after HPO, proving its importance in deep learning applications. Finally, the activation function and the optimization method based on accuracy and time of execution are suggested.