Umjeravanje mikroskopskog prometnog modela primjenom genetskog algoritma

master's thesis

• Title (english): Microscopic Traffic Model Calibration Using the Genetic Algorithm
• Author: Vjeko Glibota
• Mentor: Edouard Ivanjko (mentor)
• Committee member: Luka Novačko (predsjednik povjerenstva)
• Committee member: Martin Gregurić (član povjerenstva)
• Committee member: Miroslav Vujić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Transport and Traffic Sciences; (Division of Intelligent Transport Systems and Logistics); (Department of Intelligent Transport Systems); Zagreb
• Defense date and country: 2021-09-22, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Traffic and Transport Technology; Intelligent Transport Systems and Logistics

Abstract

Primjena prometnih modela u današnje vrijeme je postala učestala praksa, a sve češće i poželjna prije bilo kakve implementacije novog prometnog rješenja na terenu, te naknadnog testiranja. Izgradnja dodatne prometne infrastrukture ili bilo kojeg skupog infrastrukturnog zahvata, bez prethodne simulacije budućeg prometnog rješenja može se smatrati arhaičnim i neefikasnim pristupom. Uzimajući u obzir konstantno povećanje zahtjeva cestovnog prometa i nemogućnost predvidanja optimalnog iskorištavanja trenutačnih i budućih kapaciteta. Uz pomoć inteligentnih transportnih sustava se prikupljaju vrijednosti parametara s terena, koji se koriste za umjeravanje prometnog modela. Razvoj računalnih tehnologija i aplikacija su omogućili upotrebu složenih alogritama, poput genetskog algoritma, koji je korišten u ovom radu, prilikom umjeravanja prometnih simulacijskih modela. Točnije, genetski algoritam je korišten za umjeravanje prometnog toka svakog pojedinog privoza unutar raskrižja, a za to je korišten PTV VISSIM simulacijski program unutar kojeg se vršila simulacija, povezana s programskim jezikom Python putem COM veze. Umjeravanje je uspjelo po ciklusu od jednog sata optimizirati prometni tok, ovisno koliko se razlikuju ulazne vrijednosti od izlaznih prometnih tokova odabranog izoliranog raskrižja, koje bi trebale biti približno iste ili iste. Kod umjeravanja modela u ciklusu od 24 sata, umjeravanje simulacije je ovisilo o omjeru ulaznih i izlaznih vrijednosti vozila za svaki pojedini sat. Strojno učenje koristeći genetski algoritam je trajalo u periodu od par sati za jedan sat, do četiri dana za period od 24 sata. Kod perioda od 24 sata, svaki veći omjer broja vozila za pojedini sat, za svaki privoz može rezultirati neprihvatljivim vrijednostima funkcije dobrote, što na kraju generira lošija rješenja. Prilikom manjeg volumena prometnog toka po privozu i kraćih ciklusa, poput jednog sata, modul genetskog algoritma generira optimalna rješenja za taj ciklus, u odnosu za period od cijelog dana, gdje koeficijenti privoza odstupaju izvan granica prihvatljivih okvira.

Abstract (english)

Application of traffic models nowadays has become a common practice, and is preferable before any implementation of new traffic solutions on the field, and subsequent testing. The construction of additional transport infrastructure or any expensive infrastructure intervention, without prior simulation of a future transport solution can be considered and archaic and inefficient approach. Taking into account the constant increase of road traffic demands, the impossibility of predicting the optimal use of current and future capacities. With the help of intelligent transport systems, the values of parameters from the terrain are collected, which are then used to calibrate the traffic model. The development of computer technologies and applications has enabled the use of complex algorithms while calibrating traffic simulation models. Genetic algorithm is used in this master’s thesis to calibrate the traffic flow of each individual approach within the intersection, and for this, PTV VISSIM simulation program was used. Within VISSIM the simulation was performed, associated with Python programming language via COM connection. The calibration was able to optimize the traffic flow in one hour interval, depending on how different the input values are from the output traffic flows of the selected isolated intersection, which should be approximately the same. When calibrating the model in a 24-hour cycle, the calibration of the simulation depended on the ratio of vehicle input and output values, for each individual hour. Initiated machine learning using a genetic algorithm lasted for a couple of hours, for an one hour simulation, to four days for an 24-hour simulation. Within a period of 24 hours, any higher ratio of vehicle numbers per hour, for each approach can result in a unacceptable values for the fitness function, which ultimately generated poorer solutions. With a smaller volume of traffic flow per approach and shorter cycles, such as one hour, the genetic algorithm module generates optimal solutions, compared to the period for the the whole day, where the coefficients for every approach deviated beyond acceptable limits.

• Keywords: inteligentni transportni sustavi; umjeravanje prometnog modela; genetski algoritam; mikroskopska simulacija prometa; izolirana raskrižja; strojno učenje
• Keywords (english): intelligent transport systems; traffic model calibration; genetic algorithm; microscopic traffic simulation; isolated intersection; machine learning
• Language: croatian
• URN:NBN: urn:nbn:hr:119:811277
• Study programme: Title: Intelligent Transport Systems and Logistics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka prometa (magistar/magistra inženjer/inženjerka prometa)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2021-10-25 15:27:53