Use Connected Vehicle Event Data Together with Machine Learning and Artificial Intelligence Analysis Methods to Inform Traffic Management Centers for Making Real-Time Adjustments to Alleviate Non-Recurring Congestion.
Texas Based Study Exploring Artificial Intelligence and Machine Learning Usage in Transportation Revealed Lessons Learned for Successful Data Acquisition.
Austin, Texas, United States
El Paso
An Exploration of the Use of Artificial Intelligence for Enhanced Traffic Management, Operations and Safety
Summary Information
In recent years, it has been widely recognized that data analytics, automation, Artificial Intelligence (AI), and Machine Learning (ML) help transportation agencies improve their operations, service delivery, and accuracy of collected data. This study explored the feasibility of using AI and ML in transportation, within the framework of Integrated Corridor Management (ICM), first by using a survey to gain insights into successful data procurement strategies, completed by 25 Texas DOT employees. Then, the study developed three prototype ML models for four use cases, and then conducted a field test for one of the prototype models.
The three prototype ML models considered were supervised, unsupervised, and reinforcement learning techniques. The study applied the first two techniques on a Connected Vehicle (CV) crash and trip event dataset obtained from a connected vehicle company to identify safety hotspots before and after COVID-19 through AI cluster analysis to predict trip generation patterns through regression analysis, considering the road segments along I-35 and census block- groups in Austin, Texas. The third ML technique was implemented in a microscopic traffic simulation model to support optimization of signal timing plans in response to real-time traffic conditions, using speed and traffic volume data comparable to that collected by ITS sensors. Finally, the study used supervised learning techniques (time series and recurrent neural networks) to generate a ML model that could predict short term travel times using traffic volume data from traffic sensors, and probe-based speed data provided by a commercial traffic data company for a 23-mile section of I-35 in Austin, Texas. The study field-tested and assessed the performance of the developed ML prototypes by training them on two new sites in Austin and El Paso, Texas, using segment-level travel times and traffic volume data provided by the El Paso District from a center-to center (C2C) data feed.
- Use CV event data together with ML and AI analyses methods to inform Traffic Management Centers (TMC) for making real-time adjustments to alleviate non-recurring congestion. This study pointed out that data from hard-braking events, adverse weather, crowdsourced data, and queue warnings could be used to detect real-time traffic information and notify TMCs of anomalies.
- Help decision makers in short term travel time predictions with properly trained ML models using real data. This study demonstrated the use of the proposed metrics to evaluate deployed models using a testing dataset from Austin and El Paso suggested that trained models have good predictive power and lead to experienced travel time estimates that are within one minute (and three percent) of measured travel times.
- Put efforts into centralization and standardization of data. A centralized repository can reduce redundant storage of frequently accessed data and make it easier to find. Data streams should have defined formatting to ensure consistency across efforts.
- Assess the trade-offs between different data storage options. For example, this study suggested that localized storage hardware could be costly but would allow greater control over data management, whereas more feasible cloud storage options could pose security risks with increased latency to access data. The study also suggested that an ideal solution could combine local storage for data requiring higher security and low latency with cloud storage for other needs.
- Provide better access to training and coordination of communications between divisions and districts for improved personnel management. This would guide the employees in navigating tasks, and encourage collaboration and growth.
