Dynamic Multi-Modal Multi-Objective Intersection Signal Priority Optimization
-
2013-12-31
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
TRIS Online Accession Number:01537940
-
Corporate Publisher:
-
NTL Classification:NTL-OPERATIONS AND TRAFFIC CONTROLS-Traffic Control Devices
-
Abstract:In recent years travelers have shown an increased interest in multi-modal transportation including transit, bike, and pedestrian modes. Past research have studied various aspects of multi-modal traffic signal strategies, including the assessment of relative mode importance, and how to provide more equitable service for all modes by optimizing signal settings at intersections and along corridors. Most studies on the subject show multi-modal signal control is limited to at most two modes, and are based on traditional approaches, which are very restricting in nature compared to cycle-free strategies such as the one proposed in this study. This project takes into account some of the concepts used in previous research, and applies multi-attribute decision-making (MADM) methods to combine the effects of four modes of transportation (automobiles, buses, pedestrians, and bicycles) in selecting the most appropriate signal timing settings at an intersection. Three MADM methods were used: SAW (Simple Average Weighting), AHP (Analytic Hierarchy Process), and TOPSIS (Technique for Order Preference by Similarity to Ideal Solution). The results from the MADM methods are compared for different scenarios, including scenarios with weights to specify the relative importance of the four modes. A case study involving an intersection with the option of servicing pedestrians using standard parallel crossings or a pedestrian scramble phase is evaluated. In addition to the MADM methods, a multi-agent approach based on reinforcement learning was applied to optimize signal timings using a computer simulation package and real-time decision making based on inputs from virtual detectors. This resulted in a signal timing operation that is cycle-free and adaptive. The agent-based approach uses model-free reinforcement learning to optimize the operation of the signals through a multi-objective reward function. The agents make decisions, observe, and learn from the behavior of the system, evolving the knowledge about the scenario presented to the agent and thus, improving future decisions. The microscopic simulator VISSIM was selected for this study because it is capable of simulating all four modes of transportation: pedestrians, bicycles, motor-vehicles, and transit, and it also has the capabilities of using external controllers (i.e. reinforcement learning agents with a multi-objective reward function) for manipulating the traffic signals in running time.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type: