Multi-modal transport data integration and analytics platform

Abstract

Transportation is one of the crucial areas that needs to be optimized by the officials because of the increase in the demand for efficient travel and transportation due to the rapid urbanization. Integration of data from different sources has been explored and introduced as a method to address cross-domain problems like managing assets and resources efficiently. Several data integration methods have been proposed over the years, but the utilization of microservices architecture has been rare, especially in the transportation field. Microservices architecture, supported by container orchestration can be used to realize high availability, scalability, and low-cost operations. In this research, a microservices-based data integration platform was proposed to meet the demand for transportation related data integration. The proposed solution supports data importing, storing, processing, analysis and exporting of several data formats and types. A performance analysis was done to measure the scalability of the platform, accomplished utilizing the container orchestration. A real-world dataset and an experimental setup, hosted on a public cloud were employed for the analysis. The analysis demonstrates that the platform can manage around 500 RPS with a substantially low response time when auto-scaling is enabled. Finally, an approach for transportation mode detection, a use case scenario of the platform is briefly presented. As an analytics example, another research is done for short-term traffic volume forecasting. Accurate short-term traffic volume forecasting has become an important element in traffic management in intelligent transportation systems. A significant amount of literature can be found on short-term traffic forecasting based on traditional learning approaches, however deep learning based solutions have also produced substantial strides in recent years. In this paper, we propose several long-short term memory (LSTM) based deep learning models to extract inherent temporal and spatial features for traffic volume forecasting. A standard LSTM model, LSTM encoder-decoder model, CNN-LSTM model and a Conv-LSTM model were designed, optimized, and evaluated using a real-world traffic volume dataset for multiple prediction horizons. The experimental results shows that the Conv-LSTM model produced the best performance for the prediction horizon of 15 minutes with a MAPE of 9.03%. At the same time, one of the novelties of the research is the forecasting during the traffic volume anomalies due to the Covid-19.