Predictive Modelling and Optimization of Multi-Energy Systems: A Machine Learning Approach

The objective of this thesis is to evaluate the significance and impact of applying machine learning models in the energy sector through a real-case example. A multi-energy plant located in Northern Italy, with production of electricity and heat through cogenerators, boilers and photovoltaic generation, connected to the energy networks and to a district heating system, has been analysed. Several regression models have been developed to predict specific system parameters, such as methane gas consumption and thermal energy production. Additionally, attention has been given to the development of the cost function related to the system's operational framework, leading to the understanding that certain strategies are required to improve the system performance. Furthermore, given that the plant includes a photovoltaic system, an analysis of its energy production forecast has been conducted by creating a regression model using a neural network. In the final part of the thesis, various considerations are made regarding the strong correlation between machine learning models and the development of renewable energy systems, known for their unpredictability and intermittency. The results obtained through this thesis are highly promising. The performance and accuracy of the developed regression models have been primarily assessed using the R² metric, which serves as a key indicator for evaluating model performance.