In the last years, the significance of improving efficiency of urban Public Transport Systems (PTSs) is taking a key role in the development of modern society. In an ideal world, urban PTSs have no particular problems in providing effective and efficient services that improve the Quality of Life (QoL) of people living in big urban areas. An efficient service would also help to solve some additional problems, such as air pollution and traffic congestions that often afflict high density urban areas. The reality, however, is that urban transportation systems in most big cities are far from ideal. Managers of PTSs services are experiencing ever greater difficulties in maintaining high levels of efficiency due to the ever-increasing congestion affecting the major urban centers. In order to face these issues and, at the same time, improve the efficiency of PTSs services, three different proposals are introduced. First, a framework, aimed to ease the design of a monitoring system in the public transport domain, has been realized by adopting the European standard Transmodel as reference model for a generic PTS system. The proposed framework is built around the definition of a knowledge base that includes a conceptualization of the public transportation domain, on the top of which a set of logic-based functionalities are developed. The second improvement proposed, consists in the implementation of a forecasting algorithm in order to predict arrival time at bus stop in urban areas that allows to improve the travelers' perception of the Quality of Service offered. In literature, several type of algorithms have been tested in order to predict arrival time at bus stops. However, in recent years, in addition to models formed by a single algorithm (called Simple Models), models formed by the composition of more algorithms (called Hybrid Models) have emerged. An overview of Hybrid Models has been performed by comparing them with Simple Models in a real-world case study from which resulted that the Hybrid Models outperform Simple Models in every experiment performed. The third improvement proposed regards the study of the impact of an "umbalanced" methodology in vision systems able to solve various issues related to traffic management such as parking discovery and occupational turnaround in order to develop a software application to optimize their employment and, in the same time, minimize traffic caused by parking lot search. Unfortunately PluService company, which provided the data to develop the entire work, had no dataset that can be used for the proposed purpose, and, for this reason, in order to assess the quality of the proposed methodology, it has been tested with building detection problem from LIDAR aerial data. Building Detection from aerial data, in literature, represents a notoriously "difficult" application domain to face, cause the imbalance that characterizes the datasets used. The proposed method takes advantage of the Baesyan Vector Quantizer (BVQ) algorithm and, in order to verify its efficiency in strongly umbalanced application domains, it has been compared with other well known methodologies, like Adaboost or Metacost, in a real-world case study. The datasets used are formed by four strongly imbalanced classes (building, high vegetation, low vegetation and streets) and the obtained results demonstrated that BVQ outperform the other methodologies applied in the totality of experiments performed.
Nelle ultime decadi, l'incremento dell'efficienza di un Sistema di Trasporto Pubblico urbano (STP) sta assumendo un ruolo chiave nello sviluppo della società moderna. Ciononostante, la realtà è piuttosto lontana da un "mondo STP ideale", in cui i Trasporti Pubblici urbani non hanno particolari problemi nel fornire servizi efficaci ed efficienti nelle grandi citta. La realtà ci racconta che i gestori dei sistemi di trasporto urbano, nella maggior parte delle grandi città, stanno incontrando difficoltà sempre crescenti a mantenere alti livelli di efficienza. Per rispondere a questi problemi, in questo lavoro, vengono presentate tre proposte differenti. Innanzitutto è stato realizzato un framework volto a facilitare la progettazione di un sistema di monitoraggio nel settore dei trasporti pubblici, adottando lo standard europeo Transmodel come modello di riferimento. Il sistema proposto viene realizzato partendo dalla definizione di una "Knowledge Base" rappresentante una concettualizzazione del dominio del trasporto pubblico, al si sopra della quale vengono sviluppate una serie di funzionalità "logic-based". La seconda soluzione proposta consiste nell'implementazione di un algoritmo di previsione utilizzato per predire il tempo di arrivo alla fermate dell'autobus nelle aree urbane, permettendo di migliorare la percezione dei viaggiatori sulla qualità del servizio offerto. Negli ultimi anni, oltre ai modelli formati da un singolo algoritmo (chiamati "Modelli semplici"), sono emersi modelli formati dalla composizione di più algoritmi (chiamati "Modelli ibridi"). In questo lavoro viene fornita una panoramica di "Modelli ibridi" confrontandone, allo stesso tempo, le prestazioni con "Modelli semplici" applicandoli ad un caso di studio reale. Da tale studio emerge che i "Modelli ibridi" superano le prestazioni "Modelli semplici" in ogni esperimento eseguito. Il terzo modello proposto si focalizza sullo studio di una metodologia "umbalanced" applicabile a sistemi di visione per risolvere le problematiche legate alla gestione del traffico, come, ad esempio, la ricerca di parcheggi. A tale scopo viene sviluppata un'applicazione software per ottimizzare la fruizione dei parcheggi minimizzando, allo stesso tempo, il traffico causato dalla ricerca degli stessi. A causa della mancanza di dati utilizzabili, al fine di valutare comunque la qualità della metodologia proposta, tale modello è stato testato con il problema di rilevamento automatico di edifici a partire da dati aerei LIDAR. Il problema della Building Detection automatica a partire da dati aerei, in letteratura, rappresenta un dominio applicativo notoriamente "difficile" da affrontare, a causa lo squilibrio che caratterizza i set di dati utilizzati. Il metodo proposto sfrutta l'algoritmo di Baesyan Vector Quantizer (BVQ) ed, al fine di verificarne l'efficienza in domini applicativi fortemente "umbalanced", è stato confrontato con altre metodologie note, come Adaboost o Metacost, in un vero e proprio caso di studio reale. I set di dati utilizzati sono formati da quattro classi fortemente sbilanciate (edificio, vegetazione alta, vegetazione bassa e strade) ed i risultati ottenuti hanno dimostrato che BVQ supera le prestazioni delle altre metodologie nella totalità degli esperimenti eseguiti.
Intelligent Decision Support Systems in Public Transport / Benvenuti, Filippo. - (2018 Mar 27).
Intelligent Decision Support Systems in Public Transport
BENVENUTI, FILIPPO
2018-03-27
Abstract
In the last years, the significance of improving efficiency of urban Public Transport Systems (PTSs) is taking a key role in the development of modern society. In an ideal world, urban PTSs have no particular problems in providing effective and efficient services that improve the Quality of Life (QoL) of people living in big urban areas. An efficient service would also help to solve some additional problems, such as air pollution and traffic congestions that often afflict high density urban areas. The reality, however, is that urban transportation systems in most big cities are far from ideal. Managers of PTSs services are experiencing ever greater difficulties in maintaining high levels of efficiency due to the ever-increasing congestion affecting the major urban centers. In order to face these issues and, at the same time, improve the efficiency of PTSs services, three different proposals are introduced. First, a framework, aimed to ease the design of a monitoring system in the public transport domain, has been realized by adopting the European standard Transmodel as reference model for a generic PTS system. The proposed framework is built around the definition of a knowledge base that includes a conceptualization of the public transportation domain, on the top of which a set of logic-based functionalities are developed. The second improvement proposed, consists in the implementation of a forecasting algorithm in order to predict arrival time at bus stop in urban areas that allows to improve the travelers' perception of the Quality of Service offered. In literature, several type of algorithms have been tested in order to predict arrival time at bus stops. However, in recent years, in addition to models formed by a single algorithm (called Simple Models), models formed by the composition of more algorithms (called Hybrid Models) have emerged. An overview of Hybrid Models has been performed by comparing them with Simple Models in a real-world case study from which resulted that the Hybrid Models outperform Simple Models in every experiment performed. The third improvement proposed regards the study of the impact of an "umbalanced" methodology in vision systems able to solve various issues related to traffic management such as parking discovery and occupational turnaround in order to develop a software application to optimize their employment and, in the same time, minimize traffic caused by parking lot search. Unfortunately PluService company, which provided the data to develop the entire work, had no dataset that can be used for the proposed purpose, and, for this reason, in order to assess the quality of the proposed methodology, it has been tested with building detection problem from LIDAR aerial data. Building Detection from aerial data, in literature, represents a notoriously "difficult" application domain to face, cause the imbalance that characterizes the datasets used. The proposed method takes advantage of the Baesyan Vector Quantizer (BVQ) algorithm and, in order to verify its efficiency in strongly umbalanced application domains, it has been compared with other well known methodologies, like Adaboost or Metacost, in a real-world case study. The datasets used are formed by four strongly imbalanced classes (building, high vegetation, low vegetation and streets) and the obtained results demonstrated that BVQ outperform the other methodologies applied in the totality of experiments performed.File | Dimensione | Formato | |
---|---|---|---|
Tesi_Benvenuti.pdf
accesso aperto
Descrizione: Tesi_Benvenuti.pdf
Tipologia:
Tesi di dottorato
Licenza d'uso:
Creative commons
Dimensione
5.4 MB
Formato
Adobe PDF
|
5.4 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.