Data-driven model for mono and custom-built bi-facial PV output based on deep neural networks

Abstract

Accurate forecasting of photovoltaic output plays a critical role in designing PV plants. A data-driven model forecasting dynamic PV output, such as mono-facial and custom-built bi-facial, is proposed in this article. In this regard, an optimum model architecture is developed based on deep neural networks, where weather parameters are the input features. PV panels output as well as weather parameters, such as temperature, pressure, and humidity, were recorded from an experimental bench setup on the rooftop. The data was recorded during the winter for three months. The proposed model is compared with classical forecasting models such as linear regression, decision tree, and random forest. The optimization algorithm showed the proposed model had optimum training vs. validation loss. The proposed model's forecasting results showed excellent evaluation scores compared to classical models.