From Data to Energy Management: Evaluating and Analysis of Univariate and Multivariate AI Models for Photovoltaic Systems in Smart Grids

Protecting our environment necessitates a significant shift towards renewable energy sources. Among these, photovoltaic (PV) energy is one of the most widely used. However, the dependence of this energy on sunlight presents challenges in predicting energy production. So, it is crucial to have methods that allow us to predict the PV energy production of our PV systems. This will enable us to better manage this energy, anticipate energy needs, and optimize the energy supply for Buildings-Integrated PV (BIPV). The data generated by PV systems are time series data (TSD), and tools like Artificial Intelligence (AI) are widely used to predict this type of data. The objective of this study is to explore the performance of various AI models applied to TSD analysis, using both univariate and multivariate approaches. We will then examine their integration into Smart Grids (SG), specifically within a PV system connected to the electrical grid of a BIPV house in Switzerland. The goal is to determine which of these two approaches offers the best efficiency in terms of energy prediction and management within SG. We worked with four machine learning and deep learning models: XGBoost, Random Forest, Long Short-Term Memory (LSTM), and a hybrid model (CNN-LSTM) combining LSTM with a Convolutional Neural Network (CNN). Our proposed SG consists of a PV system connected to the grid and a battery for energy storage. In our future work, we will develop this SG by automating energy management system based on the energy predictions made by our AI model.