Air Quality Prediction for Healthier Urban Cycling
Mentor: Grzegorz Chmaj
Air quality monitoring and forecasting have become an essential component of smart city planning due to the increasing health risks associated with urban air pollution, especially for cyclists who are often more heavily exposed to such pollution. This project will focus on building a data‑driven air‑quality prediction system using publicly available IoT sensor datasets and applying machine learning techniques. Using real-world datasets such as OpenAQ and the UCI Air Quality dataset, the project will examine how key variables, including temperature, humidity, atmospheric pressure, and particulate matter (PM) concentrations – related to air-quality indicators such as PM2.5 and the Air Quality Index (AQI). The project will aim to explore both regression model(s) and time‑series forecasting (with focus on Random Forest with LSTM) to predict short‑term pollution levels in urban environments. By comparing these approaches the project will identify effective ways for predicting the risk of air pollution at various locations. The final outcome will include working predictive model, an evaluation of model performance and optionally visualization dashboard. As an optional extension, population density data may be incorporated to explore exposure risk and public health implications. Tools expected to be used: Python, libraries, JSON, ML frameworks and API access.
Required skills: programming (preferably Python); preferred skills: machine learning
Research Areas: Optimization, Machine Learning, Internet-of-Things Systems
Scalable Energy System for Smart Cities
Mentor: Marie-Odile Fortier
The high electricity demands of cities are projected to rise substantially in coming years, due in part to increasing electrification of transportation and of households that rely on both natural gas and electricity to meet energy needs. However, cities have limited space for new installations of conventional renewable energy systems like utility-scale wind turbines. Novel, smaller scale, decentralized energy systems that are integrated into the infrastructure of cities may be a solution, but they have not previously been modeled for location-specific conditions within highly urbanized spaces. For example, vertical axis micro-wind turbines on highway medians and piezoelectric energy harvesting devices in roads could recover some lost energy from traffic flows, but their electricity generation potential and the carbon footprint of the electricity generated from these small-scale systems have not been calculated using location-specific and time-series data within cities. Similarly, other scalable energy systems, like solar roadways and solar parking canopies, also have highly geospatially dependent electricity generation and potential to be deployed at various scales. The designs and installation locations of such emerging energy systems could be optimized towards high electricity generation and low carbon footprints using life cycle assessment (LCA) models, generating plans for individual cities to transition towards renewable energy that is locally supplied and scalable over time. The student researcher will develop parametric LCA models of emerging energy systems with site-specific data for the Southwest US, and optimize system designs, scales, and installation locations throughout the region to match nearby projected hourly demands (e.g., for electric vehicle charging at office buildings) while minimizing the carbon footprint..
Required skills:
Research Areas: Energy systems, Life-Cycle Assessment, Decarbonization
Traffic Prediction and Visualization
Mentor: Brendan Morris
A major concern for traffic operators is the early detection of congestion and incidents. With early detection, resources can be allocated and the incident can be addressed more quickly resulting in significant delay savings, which can improve safety by providing critical moments for emergency personnel. While many tools are currently available, such as traffic maps by companies like Google or reports from Waze, their data is designed for public consumption rather than for traffic operations. The goal of this project is to further the development of a traffic prediction platform which combines traffic sensors with machine learning algorithms for prediction up to one hour into the future with a web-based visualization dashboard for traffic operator use. This project combines data science through use of databases and data preprocessing, machine learning with deep learning-based traffic prediction models, and web-based visualization frameworks.
Required skills: programming (preferably Python); preferred skills: machine learning
Research Areas: Machine Learning
Smart Intersection Sensor for Unsafe/Illegal Pedestrian Crossing
Mentor: Brendan Morris
Pedestrian safety continues to be a major concern for Las Vegas. Over the past few years an alarming number of pedestrian fatalities have occurred. Many accidents have occurred in places such as Boulder Highway where pedestrians risk crossing many lanes of high-speed traffic with low-illumination to avoid walking the extra mile to a marked crosswalk. The State of Nevada has been examining technologies to help prevent such issues such as LiDAR sensors. While LiDAR is effective, the sensors are expensive, which limits wide-spread deployment. In contrast, camera-based systems are inexpensive and, with modern deep-learning techniques, can accurately detect pedestrians. REU students will develop a complete low-light camera system for detection of unsafe and illegal pedestrian crossings. The students will work with convolutional neural networks (CNNs) to detect and track pedestrians and count the number of illegal crossings. Deep trajectory forecasting techniques will be used to provide a prediction of the intent to cross illegally for advanced warning.
Required skills: programming (preferably Python); preferred skills: computer vision/image processing, machine learning
Research Areas: Computer Vision, Machine Learning
Crash Report Analysis
Mentor: Jay Park
Analyze crash reports using manual reading and test natural language processing (NLP) models to see their prediction capabilities of finding information from crash reports.
Required skills: –
Research Areas: Transportation, Safety, Construction