Owing to the ever-increasing concerns regarding greenhouse gas emissions and global warming, electric vehicles (EVs) have become increasingly popular in both personal and public transportation solutions. Consequently, this increased popularity has resulted in increased electricity demand by the consumers. This issue prompted researchers to focus on electric load and charger occupancy predictions as a promising research field. Such studies can be particularly useful to deal with the increased demand brought on the electric grid by the ever-increasing number of EVs that circulate in traffic. Accordingly, this study serves as a literature review so that best practices are identified and implemented on a case study, which aimed to predict the load profile of a given EV charger and its occupancy. The goal of this study was to understand how state-of-the-art machine learning practices are implemented in this domain. Subsequently, Long-Short Term Memory Network (LSTM) and eXtreme Gradient Boosting (XGBoost) frameworks were considered to train machine learning models leveraging a "time series-based supervised learning" approach. Owing to their computational efficiency and high accuracy, XGBoost models were used to obtain the final results reported in this study. The XGBoost models achieved an average validation accuracy of 81.1% and mean-absolute-error (MAE) of 4.595 for the classifier and regressor, respectively. In summary, this study functions as a literature review and case study of electric charger load and occupancy prediction. The results reported herein demonstrate the applicability of a common and powerful machine learning tool to day-ahead load profile estimation, which is useful for energy providers to manage demand and mitigate undesired effect of transport electrification on the power grid.
A causa delle crescenti preoccupazioni relative alle emissioni di gas serra e al riscaldamento globale, i veicoli elettrici (EV) sono diventati sempre più popolari nelle soluzioni di trasporto sia personale che pubblico. Di conseguenza, questa crescente popolarità ha comportato un aumento della domanda di elettricità da parte dei consumatori. Questo problema ha spinto i ricercatori a concentrarsi sulle previsioni relative al carico elettrico e all’occupazione del caricatore come campo di ricerca promettente. Tali studi possono essere particolarmente utili per far fronte alla crescente domanda portata sulla rete elettrica dal numero sempre crescente di veicoli elettrici che circolano nel traffico. Di conseguenza, questo studio funge da revisione della letteratura in modo da identificare e implementare le migliori pratiche in un caso di studio, che mirava a prevedere il profilo di carico di un determinato caricabatterie per veicoli elettrici e la sua occupazione. L’obiettivo di questo studio era capire come vengono implementate le pratiche di machine learning all’avanguardia in questo ambito. Successivamente, sono stati presi in considerazione i framework Long-Short Term Memory Network (LSTM) ed eXtreme Gradient Boosting (XGBoost) per addestrare modelli di apprendimento automatico sfruttando un approccio di "apprendimento supervisionato basato su serie temporali". Grazie alla loro efficienza computazionale e all’elevata precisione, i modelli XGBoost sono stati utilizzati per ottenere i risultati finali riportati in questo studio. I modelli XGBoost hanno raggiunto un'accuratezza media di validazione dell'81,1\% e un errore medio assoluto (MAE) di 4,595 rispettivamente per il classificatore e il regressore. In sintesi, questo studio funziona come una revisione della letteratura e un caso di studio sul carico dei caricabatterie elettrici e sulla previsione dell’occupazione. I risultati qui riportati dimostrano l’applicabilità di un comune e potente strumento di machine learning per la stima del profilo di carico del giorno prima, utile per i fornitori di energia per gestire la domanda e mitigare gli effetti indesiderati dell’elettrificazione dei trasporti sulla rete elettrica.
Daily occupancy and load profile estimation for an electric vehicle charger
KORUTURK, YUKSEL UTKU
2022/2023
Abstract
Owing to the ever-increasing concerns regarding greenhouse gas emissions and global warming, electric vehicles (EVs) have become increasingly popular in both personal and public transportation solutions. Consequently, this increased popularity has resulted in increased electricity demand by the consumers. This issue prompted researchers to focus on electric load and charger occupancy predictions as a promising research field. Such studies can be particularly useful to deal with the increased demand brought on the electric grid by the ever-increasing number of EVs that circulate in traffic. Accordingly, this study serves as a literature review so that best practices are identified and implemented on a case study, which aimed to predict the load profile of a given EV charger and its occupancy. The goal of this study was to understand how state-of-the-art machine learning practices are implemented in this domain. Subsequently, Long-Short Term Memory Network (LSTM) and eXtreme Gradient Boosting (XGBoost) frameworks were considered to train machine learning models leveraging a "time series-based supervised learning" approach. Owing to their computational efficiency and high accuracy, XGBoost models were used to obtain the final results reported in this study. The XGBoost models achieved an average validation accuracy of 81.1% and mean-absolute-error (MAE) of 4.595 for the classifier and regressor, respectively. In summary, this study functions as a literature review and case study of electric charger load and occupancy prediction. The results reported herein demonstrate the applicability of a common and powerful machine learning tool to day-ahead load profile estimation, which is useful for energy providers to manage demand and mitigate undesired effect of transport electrification on the power grid.| File | Dimensione | Formato | |
|---|---|---|---|
|
Thesis_Project_with_Template_v2.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
3.34 MB
Formato
Adobe PDF
|
3.34 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/214620