Applications of Statistical and Machine Learning to Civil Infrastructure

Abstract

Roadway, buildings, and electrical infrastructure in the United States are triumphs of modern engineering and represent enormous societal investments. With the proliferation of emerging technologies---from solar panels to autonomous and electric vehicles---these systems will begin to be utilized in ways they were not originally designed for. The advent of open source, as well as municipal and government data sources on the usage of these systems, however, provides an opportunity to adapt these systems to emerging technologies without rebuilding national infrastructure from scratch. Many works in recent years have utilized statistical and machine learning to analyze these data sets in an effort to improve the operational efficiency of and adapt existing civil infrastructure to these emerging technologies. These works, however, often conduct superficial studies of the potential usage of statistical and machine learning techniques and ignore the physical and engineering design constraints on these systems. This produces models with predictive power exhibiting limited flexibility to changes in external factors driving the system, or control frameworks with limited guarantees security constraints are met. Addressing these shortcomings often requires carefully tailored combinations of domain-aware system models, available data, and statistical and machine learning techniques. To demonstrate the effectiveness of conscientiously combining these elements, this work conducts three in-depth case studies at the national, municipal, and local scales: by focusing on specific applications in a) municipal transportation networks, b) power grids and electrical markets, and at c) HVAC systems in buildings. The work is concluded with a discussion on patterns arising in applying statistical and machine learning techniques to these types of civil infrastructure effectively given available data sources and standing, domain-specific engineering questions.