Quantitative Assessment of Resilience in Complex Systems

Abstract

Complex systems, such as financial systems and infrastructure systems, are facing an increasing number of disruptions, either external (e.g., financial crisis and natural disasters) or internal (e.g., degradation, aging, and ill performance). Many previous works have dedicated to the question of how to measure the resilience in these complex systems. To date, system resilience can be assessed in various ways, i.e., qualitative or quantitative, summative or formative, and target-based or purpose-based. This thesis focuses on quantitative methods and makes contributions on filling research gaps and improving the up-to-date toolkits in three categories of resilience assessment tools, namely (1) Performance-based metrics; (2) Network-based approaches; And (3) Probability-based graphic models. For performance-based metrics, critical system functions, such as tolerance thresholds for different level of degradation, are absent in the construction of these tools, and the metrics' applicability in complex performance is less developed. Therefore, Chapter 2 aims to develop a generic resilience metric for quantitative assessment based on system functions and test its strength in the complex performance of stock markets. The proposed metric satisfactorily characterized the markets’ resilient behaviors and had comparative advantages throughout the analysis. In network-based studies, one of the significant missing links in network resilience (specifically in temporal networks) is the relationship between the resilience of individual components (nodes or edges) and the dynamic interdependence of networks (dynamic changes in the topology of temporal networks). Thus, Chapter 3 and 4 was designed in a progressive manner where Chapter 3 acts as a pilot study, aiming to characterize individual resilience using complex network approach and explore the descriptive strength of multiple associated factors. Chapter 4 is a further study, aiming to develop a set of statistical models to identify individuals' resilient performance in a networked environment and perform in-depth analyses and forecasts. Taking London stock exchange as study objective, Chapter 3 found that the survivability resilience of individual stocks was correlated with node degree and node strength. This was further confirmed by Chapter 4, whose results showed that the survivability resilience could be described and approximated by degree-related centrality measures. In addition, the statistical models proposed in Chapter 4 offers an effective tool that can be used to predict different individual stock's resilient performance in the networked market. Lastly, Chapter 5 proposes a hierarchical Bayesian network model with ontologically identified interdependence among resilience functions and system qualities. Based on current literature, the ontology-oriented Bayesian networks have been rarely applied to model system resilience, and the investigations on the dynamic resilience are still needed. Therefore, Chapter 5 aims to develop a probability-based graphic model to quantify the dynamic resilience. The chapter takes Beijing's road transportation system as a case study and studies the dynamic resilience of the system from 1997 to 2016 by fusing multi-source and heterogeneous urban data. The analysis found that Beijing’s road system was not as resilient as expected, with the probability of being resilient between 50% and 70%. Moreover, the critical system qualities that mostly affect its dynamic resilience have been identified as well. The model proposed in this chapter is a promising tool for resilience assessments. The main value of this thesis is to improve our understandings about how to effectively measure and quantify resilience in complex systems with complex performance, dynamic interdependence, massive system topology, and probability caused by uncertainties. The thesis enriches the state-of-art assessing methods in resilience research by exploring possible measures and methodologies in quantitative approaches. The specific findings of each chapter can be useful and heuristic for researchers, policy-makers, shareholders, and practitioners in the field. However, in the final chapter, the author has acknowledged some limitations and outlook of this thesis which can be addressed in future works.