Uporaba metod spodbujevanega učenja za optimizacijo krmiljenja prometa na cestni arteriji

doktorska disertacija

Rok Marsetič (Author), Marijan Žura (Mentor), Matjaž Mikoš (Thesis defence commission member), Peter Lipar (Thesis defence commission member), Tomaž Maher (Thesis defence commission member), Drago Sever (Thesis defence commission member)

Abstract

Današnja družba se zlasti v mestnih središčih pogosteje srečuje s prometnimi zastoji, višjimi stroški, nižjo prometno varnostjo in večjim onesnaženjem okolja. Prometne obremenitve vedno bolj presegajo kapaciteto cestne infrastrukture, posebno v koničnih urah. V mestih je velik delež križišč opremljenih s svetlobnosignalnimi napravami, ki imajo ključno vlogo pri varnem in učinkovitem vodenju prometa. Pri zagotavljanju čim krajših zamud je pomembna hitra prilagoditev krmilnih programov v realnem času. Z omejitvijo širjenja cestne infrastrukture in naraščanjem prometnih obremenitev čedalje težje ohranjamo zadovoljiv prometni nivo uslug s klasičnimi tehnikami krmiljenja. To še posebno velja ob nastanku nepredvidenih prometnih dogodkov. V takšnih situacijah so pomanjkljivosti zdajšnjih krmilnih sistemov še opaznejše. Promet je stohastične narave in spremembe v prometnem toku zahtevajo, da se krmilna strategija nenehno posodablja. Za reševanje tako zahtevnih in kompleksnih problemov optimizacije krmiljenja svetlobnosignalnih naprav potrebujemo sistem, ki se nenehno prilagaja in uči. S pomočjo umetne inteligence je mogoče razviti sisteme, ki so sposobni samostojnega učenja. V doktorski disertaciji smo predstavili sodoben pristop krmiljenja svetlobnosignalnih naprav na cestni arteriji, in sicer z uporabo algoritma spodbujevanega učenja. Predlagani algoritem omogoča hitro in samostojno učenje optimalne strategije krmiljenja v najrazličnejših prometnih razmerah. Uspešnost predlaganega algoritma smo preverili s pomočjo mikrosimulacijskega orodja, s katerim lahko z veliko natančnostjo simuliramo realne prometne razmere. Dokazali smo, da je učinkovitost predlaganega krmiljenja boljša v primerjavi s klasičnim prometno odvisnim krmiljenjem.

Keywords

Grajeno okolje;gradbeništvo;disertacije;spodbujevano učenje;cestna arterija;krmiljenje prometa;svetlobnosignalne naprave;

Data

Language:	Slovenian
Year of publishing:	2016
Typology:	2.08 - Doctoral Dissertation
Organization:	UL FGG - Faculty of Civil and Geodetic Engineering
Publisher:	[R. Marsetič]
UDC:	656.05: 004.451.26:519.8:(043)
COBISS:	7394401
Views:	1775
Downloads:	507
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Application of reinforcement learning methods for optimization of traffic control on arterial roads
Secondary abstract:	Nowadays, society faces several traffic related problems, such as traffic jams, time loss, lower traffic safety, increased pollution, etc., especially in urban areas. This is caused by high traffic volumes, which often exceed the capacity of the road infrastructure, particularly in peak hours. A common way of managing traffic in urban areas is traffic light control, which plays a key role in traffic safety and efficiency. To reduce delays the traffic light controllers should adjust to changing traffic volumes continuously and rapidly. Limited possibilities for road infrastructure extensions and growing traffic volumes represent a challenge for existent control techniques with increasing problem of maintaining suitable level of service. When unexpected events occur, the disadvantage of current traffic control system is even more evident. Stochastic nature of traffic and constant changes in traffic flow requires continuous adaption of traffic light controller. For solving complex problem of traffic lights optimization the system that continuously adapts and learns should be employed. Artificial intelligence approaches enable development of self-learning systems. The thesis presents a novel approach for solving problems of traffic light controller optimization with use of the reinforcement learning. The proposed algorithm enables fast and self-learning optimal strategy of traffic control in different traffic conditions. The efficiency of proposed algorithm was tested using a micro simulation tool, which simulates traffic conditions with great accuracy. The results of the performed experiments show that proposed algorithm outperforms the actuated signal controllers.
Secondary keywords:	Built Environment;civil engineering;doctoral thesis;reinforcement learning;road artery;traffic control;traffic lights;
URN:	URN:NBN:SI
File type:	application/pdf
Type (COBISS):	Doctoral dissertation
Thesis comment:	Univ. v Ljubljani, Fak. za gradbeništvo in geodezijo
Pages:	XVIII, 109 str.
ID:	9127679