Graduation Year

2022

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Civil and Environmental Engineering

Major Professor

Yu Zhang, Ph.D.

Co-Major Professor

Xiaopeng Li, Ph.D.

Committee Member

Zhenyu Wang, Ph.D.

Committee Member

Seckin Ozkul, Ph.D.

Committee Member

Yang Liu, Ph.D.

Keywords

Anomaly Detection, Graph Modeling, Symbol Recognition, Train Platforming and Routing

Abstract

The continued and substantial growth in railroad transportation in many countries inspires railroad operators to leverage advanced methods into better railroad operation decisions. Among all facilities, significant returns on investment can always be achieved by optimizing the operations of network nodes - junctions and stations, because stations usually form the capacity bottlenecks in the system.

There are thousands of decision-making problems in relation to the station operations and can be classified into different levels to achieve different goals. From top to bottom, high-level business strategies aim to make decisions to achieve long-term strategical benefits, such as optimizing local station functionalities to succeed in global network-wise goals. Mid-level management tactics focus on formulating mid-term optimization tactics to improve management capabilities on safety, efficiency, and operating costs. Ground-level operating decision supports are to support short-term or real-time operations such as rescheduling timetables after a disruption. Foundation-level technology capabilities provide various fundamental IT tools to support the implementation of upper levels, such as data collection, data preprocessing, and process automation.

Following this hierarchy, three problems draw the interest of this dissertation:

Firstly, to fill the gap of using scientific methods to understand station operations and identify target objects at the strategic level. A meaningful railroad station anomalies detection problem attracted the attention of this dissertation.

Detecting railroad station anomalies is a critical task prior to segmentation and making optimization decisions for each cluster. Three types of anomalies caused by the specialty of railroad operations bring the existing methods non-trivial challenges in detection accuracy and efficiency. To tackle these tasks, this dissertation proposes a novel anomaly detection method named Huffman Anomaly Detection Forest (HuffForest) to detect station anomalies in performance, which leverages Huffman encoding to measure abnormalities in certain railroad scenarios with high accuracy.

This method establishes a Huffman forest by constructing trees from the perspective of data points and subsequently computes anomaly scores of instances considering both local and global information. A sampling-based version is also developed to improve scalability for large datasets. Taking advantage of the encoding mechanism, the proposed method can effectively recognize the underlying patterns of railroad stations and detect outliers in various complicated scenarios where the conventional methods are not reliable.

Experiment results on both synthesized and public benchmarks are demonstrated to show the advances of the proposed method compared to the state-of-the-art isolation forest (iForest) and local outlier factor (LOF) methods on detection accuracy with an acceptable computational complexity.

Secondly, after the target stations are properly selected, this dissertation is motivated to drill down to the tactical level of decisions to maximize station capacity and scheduling reliability, where a train platforming and routing (TPR) problem with different interlocking modes is identified.

The TPR problem is to decide train operations within stations after the network-wise train schedules are determined. A feasible TPR plan requires both platform and route conflict-free, where the avoidance of route conflict is controlled according to three interlocking modes. Although the TPR problem is widely studied, none of them did a serious investigation on the optimality impacts of different interlocking modes in the TPR. Therefore, this dissertation introduced and formulated a space-time version of TPR considering three interlocking modes and subsequently conducted numerical experiments to analyze the optimality differences under each mode. Based on the experimental findings, engineering practical suggestions are also provided. In summary, the experiment results showed that both route-locking sectional-release and sectional-locking sectional-release modes significantly outperform the route-locking route-release mode. And among them, using the route-locking sectional-release mode can bring notable benefits to large stations with high-density volumes while using the sectional-locking sectional-release mode can always provide outstanding outcomes over various station and traffic settings.

In the meanwhile of addressing the above issues, a fundamental technical problem at the tech-foundation level arises - the manual modeling approach often costs engineers significant efforts and notably limits the generality and extensivity of many advanced methods.

However, creating a high-fidelity railroad station model to match the physical details of hundreds of tracks and switches is never a trivial task. With the widespread of CAD technology, many stations are drawn proportionally into two-dimensional DXF files. Thus, this dissertation proposed a framework to efficiently convert DXF files into meaningful station models. The proposed framework consists of two phases (1) converting graphic basic primitives without explicit engineering interpretations into recognizable railroad symbols and (2) modeling undirected railroad station graphs with necessary configurations such as endpoints and routes. Subsequently, the proposed framework is developed into a GUI application with minimal user interaction, and the validity, productivity, and applicability are tested at several real-world passenger stations in Asia.

Download

Share

COinS