Accurate regulation of superheating in heat pump systems plays a critical role in ensuring the proper operation of the refrigeration cycle, protecting the integrity of the compressor and op- timizing the overall energy efficiency of the plant. This thesis is part of an industrial research project developed in collaboration with Swegon Operations SRL, with the aim of designing and implementing an intelligent system for superheating prediction and optimized control of the electronic expansion valve (EEV), based on Machine Learning techniques and optimization strategies. After an in-depth data collection and preprocessing phase, based on time series acquired from sensors installed across multiple units, supervised regression models were developed, with a fo- cus on boosting algorithms. In particular, the CatBoost algorithm was selected due to its efficient handling of categorical variables, robustness to missing data and good generalization capabili- ties. Several modeling strategies were explored, both with and without the use of lagged features and model performance was assessed using metrics such as Mean Absolute Error (MAE) and the Coefficient of Determination (R2). The predictive models were then integrated into an optimization algorithm driven by domain- specific decision rules, with the aim of dynamically suggesting the optimal EEV opening under varying operational conditions. Furthermore, uncertainty estimation was addressed using quan- tile regression techniques to evaluate the robustness of the model in real-world environments affected by noise and dynamic variability. The experimental results obtained from real-world datasets collected from in-field units confirm the validity of the proposed approach. Improvements in control stability, superheating predic- tion accuracy and potential energy efficiency were observed. The proposed techniques open perspectives for the deployment of advanced control systems in industrial HVAC and refrigeration applications.
La regolazione accurata del surriscaldamento (superheating) nelle pompe di calore riveste un ruolo fondamentale per garantire il corretto funzionamento del ciclo frigorifero, preservare l’integrità del compressore e ottimizzare l’efficienza energetica complessiva dell’impianto. In questo contesto, la presente tesi si inserisce all’interno di un progetto di ricerca industriale sviluppato in collaborazione con Swegon Operations SRL, con l’obiettivo di progettare e implementare un sistema intelligente per la previsione del superheating e il controllo ottimizzato dell’apertura della valvola di espansione elettronica (EEV), basato su tecniche di Machine Learning e ottimizzazione. Dopo un’approfondita fase di raccolta e preprocessing dei dati, provenienti da sensori installati su diverse unità, sono stati sviluppati modelli di regressione supervisionata, con particolare at- tenzione agli algoritmi di boosting. In particolare, l’algoritmo CatBoost è stato selezionato per la sua efficacia nella gestione di feature categoriche, robustezza alla presenza di dati mancanti e buone capacità di generalizzazione. Sono stati esplorati diversi approcci di modellazione, sia con che senza l’utilizzo di feature temporali ritardate (lagged features) ed è stata valutata la qualità delle previsioni attraverso metriche come Mean Absolute Error (MAE) e il Coefficiente di Determinazione (R2). Successivamente, i modelli predittivi sono stati integrati in un algoritmo di controllo ottimizzato basato su logiche decisionali derivate da esperti di dominio, con l’obiettivo di suggerire l’apertura ottimale della EEV in funzione delle condizioni operative rilevate. È stata inoltre effettuata un’analisi dell’incertezza predittiva mediante regressione quantile, al fine di valutare la robustezza del sistema in scenari reali caratterizzati da rumore e variabilità operativa. I risultati sperimentali, ottenuti su dataset provenienti da macchine installate in campo, di- mostrano la validità dell’approccio proposto, evidenziando una maggiore stabilità del controllo, una riduzione del margine di errore nella stima del superheating e un potenziale miglioramento dell’efficienza energetica. Le tecniche presentate aprono prospettive per l’adozione di sistemi di controllo avanzati nei settori della climatizzazione e della refrigerazione industriale.
EEV optimization and superheating prediction in HVAC systems using Machine Learning
FONSATO, ANDREA
2025/2026
Abstract
Accurate regulation of superheating in heat pump systems plays a critical role in ensuring the proper operation of the refrigeration cycle, protecting the integrity of the compressor and op- timizing the overall energy efficiency of the plant. This thesis is part of an industrial research project developed in collaboration with Swegon Operations SRL, with the aim of designing and implementing an intelligent system for superheating prediction and optimized control of the electronic expansion valve (EEV), based on Machine Learning techniques and optimization strategies. After an in-depth data collection and preprocessing phase, based on time series acquired from sensors installed across multiple units, supervised regression models were developed, with a fo- cus on boosting algorithms. In particular, the CatBoost algorithm was selected due to its efficient handling of categorical variables, robustness to missing data and good generalization capabili- ties. Several modeling strategies were explored, both with and without the use of lagged features and model performance was assessed using metrics such as Mean Absolute Error (MAE) and the Coefficient of Determination (R2). The predictive models were then integrated into an optimization algorithm driven by domain- specific decision rules, with the aim of dynamically suggesting the optimal EEV opening under varying operational conditions. Furthermore, uncertainty estimation was addressed using quan- tile regression techniques to evaluate the robustness of the model in real-world environments affected by noise and dynamic variability. The experimental results obtained from real-world datasets collected from in-field units confirm the validity of the proposed approach. Improvements in control stability, superheating predic- tion accuracy and potential energy efficiency were observed. The proposed techniques open perspectives for the deployment of advanced control systems in industrial HVAC and refrigeration applications.| File | Dimensione | Formato | |
|---|---|---|---|
|
Fonsato_Andrea.pdf
accesso aperto
Dimensione
3.29 MB
Formato
Adobe PDF
|
3.29 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/110134