Statistical Methods for System Monitoring and Its Applications

系統監測的統計方法及其應用

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date18 Jun 2019

Abstract

The data generated from the accelerating advancement of information, communication, sensing, and characterization technologies is expected to grow exponentially, resulting in so-called “big data.” One focus of big data analytics is to explore the massive amount of data and extract useful information or knowledge for future actions. System monitoring poses a new research opportunity for data analytics in the era of big data and refers to the framework of analysis and interpretation of related data and continuous monitoring for decision-making and strategic planning. This framework is essential to ensure that an entire system is stable and in control. It is solidly founded on engineering sciences, an array of systems-theoretical methods, and computer-aided tools. System monitoring bridges the gap to established disciplines in engineering and science dealing with systems problems and offers sensible applications to solve the problems of those disciplines. A fundamental problem in system monitoring is how to take advantage of big data to address systems problems such as estimation and monitoring.

System monitoring has reached out into new application domains such as human and network systems, and contributes to the solution of these “non-traditional” systems problems. Its applications will continue to broaden in the future with the expansion of the boundaries that define the systems. Research into system monitoring has focused on novel methods and tools and faces application challenges in emerging fields. The successful application of system monitoring methods and tools requires at least the tailoring of existing methods and tools, and even the development of completely new ones, to effectively address the specific problems of new domains.

In this thesis, we develop tailored statistical solutions to two emerging system problems. The first part of the thesis focuses on estimation and prediction problems in the human system. To alleviate the heavy burden of the aging population, we develop a personalized health monitoring system of elderly wellness that aims to facilitate diagnosis, treatment, and care based on an individual’s lifestyle. We also propose a solution to integrate continuous activity data and discrete physiological data for quantitative assessments of health-related risks on an individual-specific basis. Besides, we develop an automatic surrogate approach to assess the functional balance and mobility of the elderly without the involvement of health care professionals. The automatic estimation of functional balance using a computerized system and a wearable sensor can provide more sensitive, specific, and responsive biomarkers for long-term monitoring in clinical practice. Both proposed health monitoring system and surrogate assessment can provide early warnings and timely feedback regarding the effectiveness of administered interventions, thus enabling intervention strategies to be modified or changed if they are found to be ineffective, that is particularly important for long-term monitoring.

The second part of the thesis focuses on monitoring and surveillance problems in the network system. Network monitoring concerns security and management issues, with the aim of providing a secure and persistent network. In the first stage of this part, we propose a probability-based approach to model anomalous nodes that arise from different mechanisms in a dynamic network with a community structure. We also develop a multivariate surveillance plan that is flexible enough to detect different types of node propensity change. In the second stage, we propose a simulation-based strategy to systematically compare the performance of network monitoring methods over a variety of dynamic network changes. We further compare the performance of several state-of-the-art social network monitoring methods using our family of simulated dynamic networks. Both proposed strategy contributes significantly to future research on providing guidelines for selecting appropriate network monitoring methods on specific circumstances and evaluating the utility of the new monitoring method.

This thesis contributes to the research on system monitoring by developing several novel statistical approaches for system problems in different application domains based on the big data environment. This thesis also demonstrates numerous emerging applications of system monitoring require synergize expertise from various domains to obtain high-quality solutions to their complex system problems. The analysis included in this thesis provides an extensive and valuable knowledge base that is of particular relevance to the use of system monitoring for real-life field applications.