Solar Irradiance Forecast using Long Short-Term Memory: A Comparative Analysis of Different Activation Functions

Microgrid consists of multiple Distribution Generations (DGs) such as solar PV and wind turbines which are weather dependent. As DGs are weather dependent, a collection of weather parameters from the Singapore weather station is used as the dataset for solar irradiance forecasting. It is essential to have an accurate solar irradiance tool to efficiently manage microgrid renewable power generation. In this paper, a vanilla Long Short-Term Memory model is implemented using different activation functions, including ReLU, ELU, Leaky-ReLU, SELU, and GELU, in the dense layer to forecast short-term solar irradiance. Activation functions introduce the non-linearity and learn the relationship between the input and output values. The importance of the activation function is to support on the model learning and execution of difficult tasks. With the activation function, the model creates stacking capabilities of multiple layers of neurons to develop the deep neural networks to learn complex datasets. The impact of the proposed methods is evaluated in terms of efficiency, robustness, and accuracy of each activation function.