mCity: using smart city monitoring data to characterize and improve urban mobility

• Autor(a) principal: Almeida, Ana Filipa Simão de
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/29686
• topic: Smart urban mobility,; Traffic flow,; Forecasting,; Deep learning,; Driving behavior

The sustainable growth of cities created the need for better informed decisions based on information and communication technologies to sense the city and quantify its pulse. An important part in this concept of \smart cities" is the characterization of the traffic flows. In this work, we aim at characterizing the urban mobility in two different cities, Porto and Aveiro. The structure and contents of the corresponding datasets is very different, enabling two case studies, with distinct use cases related to traffic analysis and forecasting. For the Porto use case, we had access to road-mounted traffic sensors and the buses tracking data. The first source was studied and was looked for patterns (e.g.: weekdays behavior). Historic traffic counters data was used to forecast future flows, using both statistical and deep learning methods. We found that it was not possible to find a clear relationship between (buses) speed and traffic intensity, however, when the speed was high, there was low intensity, and when there was high intensity, the velocity was low. There are daily and weekly patterns in the traffic flow data that enable forecasting. When the anomalies in traffic do happen, the methods for short-term forecasting perform better than those for long-term forecasting. In the Aveiro use case, the dataset includes bus traces, that were used to characterize the driving behavior, based on speed and acceleration. These data were mapped into the city to find problematic areas. Side-by-side visualizations help with the comparison of the traffic behavior in selected time periods. We observed that some roads often present the same problems, independently of the day or time of the day. In other parts of the city, the problems can be found more often in specifics periods. The datasets for Aveiro and Porto were sampled with different frequency (each second and each minute, respectively). We confirmed, with simulations, that the analysis made for Aveiro was not possible with the granularity of the Porto's data set (as some information would be lost). The computational pipeline to run the supporting analyses is fully implemented, as well the required integrations to programmatically obtain the data from the existing data sinks. For the driving behavior analysis, a web dashboard is deployed, enabling the relevant departments to study potential problematic areas in the city of Aveiro.