Analysis and Modeling of a Heating, Ventilation and Air Conditioning (HVAC) system using Machine Learning

The main objective of this thesis is the identification of a machine learning model able to reproduce the behavior of a Heating, Ventilation and Air Conditioning (HVAC) system, in collaboration with the company "DENSO Thermal Systems". Precisely, the developed model will must be able to predict volumetric airflows of the outlets, taking as inputs the speed of the blower motor and the outlets percentage apertures. The model will be the plant of a control system (the controller development is outside the scope of this work). The system under analysis is an innovative four-zone HVAC system for a luxury vehicle. The innovation lies in the zonal independence, which makes the object of this work able to produce many more operating modes, with respect to conventional automotive HVAC systems. The characteristics of the system, such as the presence of a lot of components and the high nonlinearity of their behaviors, make it very complex. So, it was considered as a black box and modeled only using experimental data. The main objective of this thesis was achieved, organizing the work into two phases: first, an activity of analysis of the datasets provided by the company, aimed at achieving a better comprehension of the HVAC system, identifying critical behaviors and correlation between the variables, performing attempts of dimensionality reduction and investigating the performances of the system. Second, the exploration and the comparison of the peformances of different machine learning approaches, in particular Support Vector Regression (SVR) and Neural Networks (NN) in order to find the model that best reproduce the behavior of the system. Finally, the adoption of Bayesian Optimization technique for tuning the hyperparameters of the developed machine learning models.