Page 35 - SMILESENG

Intl. Summer School on Search- and Machine Learning-based Software Engineering
SmartTLC: Towards Smart Traffic Light Systems
Jose´ R. Lozano-Pinilla
QSEG, Universidad de Extremadura, Spain Email: joserralp@unex.es
Abstract—The increasing number of fuel-based vehicles has several negative impacts on the environment, the economy and the citizen’s daily life, being the largest contributor to Green- House Gas (GHG) emissions, mostly due to traffic congestion. Actually, more and more cities are deploying ICT-based infrastructures to monitor the traffic and its environmental impact (air pollution, noise, etc.). In this line, this paper describes SmartTLC: a software framework, aimed at enabling the simulation and comparison of different traffic light adaptive control algorithms based on traffic data (either historical, real-time or both). This framework allows designers to select the best traffic light control strategy for different situations, indicating which one achieves better results in terms of reducing traffic congestion. The experimental results obtained so far demonstrate that the adoption of a context-aware adaptive approach significantly improves traffic fluidity, reducing vehicle waiting time, in particular, in roads with a higher traffic demand.
I. INTRODUCTION
The increasingly growing number of vehicles, nowadays mostly fuel-based, has important negative impacts not only in the environment and in the personal and global economy, but also in our daily life [1]. In 2019, the transportation sector was the largest contributor to Green-House Gas (GHG) emissions, accounting up to 23% in Europe (29% in the USA) of the total GHG emissions. Up to 82% of these emissions were produced by light-duty vehicles (58%) and medium and heavy trucks (24%) [2], although these are not homogeneously distributed across all continents, countries and regions [3].
According to the International Energy Outlook 2021 [4], electric vehicles currently make up only 30% of the 1.446 billion cars estimated on Earth in 2022 [3]. Furthermore, the report seriously warns that emissions from the transportation sector are expected to increase through 2050 unless world leaders establish legal and regulatory changes. In this context, the implementation of new mobility management policies becomes an urgent must.
II. HYPOTHESIS
In this work we propose a smart traffic light control system aimed at helping designers analyze how different traffic light strategies behave in different situations and select the one that provides better results in terms of alleviating traffic congestion. Addressing this issue is a highly complex challenge that involves (1) context-awareness, enabling real- time mobility monitoring (road demand, average waiting time, GHG emissions, etc.); (2) the identification of (eventually changing) traffic patterns; and (3) the adequacy of the road
C. Vicente-Chicote
QSEG, Universidad de Extremadura, Spain Email: cristinav@unex.es
infrastructures and of the traffic lights that control the vehicular (and pedestrian) flows in urban environments.
Appropriately processing the context data provided by road- located IoT sensors can be useful (1) to predict future road demands and adequately plan and design the required infras- tructures; (2) to identify relevant traffic patterns (e.g., daily peak hours) in order to schedule the best (predefined) traffic light control policies; and (3) to react to unforeseen situations (e.g., an unexpected increase of road demand), enabling the dynamic adaptation of the traffic light control strategy.
III. OBJECTIVES
The SmartTLC framework is targeted at helping designers simulate different traffic conditions and analyze which traffic light control algorithm performs better in each of them. To achieve this goal, SmartTLC can use either (1) the information about the real-time traffic conditions provided by (real or simulated) IoT sensors; (2) predictions based on the traffic patterns learnt from a historical traffic dataset; or (3) both. Based on this information, the framework allows designers to simulate different traffic light control algorithms to find out which one achieves a higher vehicle waiting time reduction. It is worth noting that reducing vehicle waiting time can also indirectly improve other metrics such as air and noise pollution, fuel consumption, etc.
In this work, both the traffic conditions and the results of applying four different traffic light control algorithms were simulated using the SUMO framework [5].
IV. SCENARIO
The scenarios defined to test the SmartTLC framework are based on four different concepts: (1) the network topol- ogy, representing roads, lanes, signals and traffic regulations; (2) the vehicular traffic types, representing different traffic intensities (vehicles per hour), along with a numeric range to support traffic uncertainty; (3) the traffic light algorithms, where the green phase can be calculated in different ways; and (4) the traffic time patterns, representing how the traffic evolves during a day. The scenario defined for this preliminary study considered: (1) a topology with a single junction, where vehicles could only travel north-south (NS) or east-west (EW), i.e., turns were not supported; (2) four different traffic types, each one defining a range of vehicles per hour from 1-5 (very low) to 350-650 (high); (3) four different traffic light control approaches, including a basic fix-cycle one and three adaptive algorithms based, in turn, on historical traffic data, real-time
23