Improved Hourly Prediction of BIPV Photovoltaic Power Building Using Artificial Learning Machine: A Case Study

Abstract

In the energy transition, controlling energy consumption is a challenge for everyone, especially for BIPV (Building Integrated Photovoltaics) buildings. Artificial Intelligence is an efficient tool to analyze fine prediction with a better accuracy. Intelligent sensors are implemented on the different equipments of a BIPV building to collect information and to take decision about the energy in order to reduce its consumption. This paper presents the implementation of a machine learning model of short and medium term hourly energy production of photovoltaic panels in BIPV buildings on several sites. We selected the data influencing the energy efficiency of the PV panels, with the measurement of variable importance score for each model. Indeed, we have developed and compared several machine learning models of hourly prediction independently of the building location taking into account the weather forecast data on site such as DHI, DNI and GHI and the same in clear sky condition. Five methods are tested and evaluated to determine the best prediction: Random Forest (RF), Artificial Neural Networks (ANN), Support Vector Regression (SVR), Decision Trees regression (DTR), and linear regression. The methods are evaluated based on their ability to predict photovoltaic energy production at hourly and daily resolution.