ON THE ROBUSTNESS, EFFICIENCY AND INFORMATION FRESHNESS IN NETWORKED SYSTEMS
Open Access
- Author:
- Wang, Boyu
- Graduate Program:
- Electrical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- September 09, 2019
- Committee Members:
- Jing Yang, Dissertation Advisor/Co-Advisor
Jing Yang, Committee Chair/Co-Chair
Nilanjan Ray Chaudhuri, Committee Member
Aylin Yener, Committee Member
Hui Yang, Outside Member
Kultegin Aydin, Program Head/Chair - Keywords:
- cyber physical systems
robustness
efficiency
information freshness - Abstract:
- The integration of physical network systems with computational capabilities has led to the new generation of systems, the cyber-physical systems (CPS). Applications of CPS range from manufacturing facilities, smart grid to robotics, smart transportation systems, etc. In contrast to traditional engineered systems, CPS emphasize the interaction between cyber space and physical environment, leading to significant improvement in system reliability, efficiency, and controllability. However, besides the benefits that CPS bring to us, the operation of large-scale CPS also poses severe challenges, among which robustness, efficiency and information freshness are three fundamental ones. Many CPS applications are safety-critical, e.g., smart power grid, which need to operate reliably and avoid abnormal working situations. Comparing to traditional engineered systems, CPS are more vulnerable to failures, faults and attacks, considering the data management and communication layer. For example, the fault in one of the networks can cause the failures of the dependent nodes in the other network, and further lead to a cascade of faults that will bring a catastrophic impact, examples include the 2003 blackout in the northeastern United States. How to guarantee the safety and robustness of CPS is of critical importance. Efficiency is another critical concern in CPS. Many CPS operates under given physical resources, and resource efficient operation is of great importance for the sustainability of CPS. For example, an electric vehicle (EV) charging network is a typical CPS, which includes a charging station that connected to the power grid and a large number of EVs. The charging station collects information and controls the charging procedure, where efficient charging scheduling is needed to reduce the electricity costs, to prevent the system from overloading, and to satisfy customers' demands. Besides, data freshness has been playing a critical role for real-time system status monitoring and control in CPS. However, the unprecedented high dimensionality and generation rate of the sensing and control data also impose severe challenges on its timely delivery. Therefore, how to maintain data freshness at the central controller under given resource constraints in the system has become another critical yet challenging problem. The goal of this study is to address these challenges in specific CPS, as outlined as follows. In Chapter 2, we consider the line outage detection and identification problem in power networks. We formulate this problem as a sparse binary-valued vector estimation problem, and leverage the cluster structure existing in most multiple line outages to solve it. We propose three low-complexity graph-based algorithms to identify clustered line outages. Simulated tests in IEEE-118 bus system and IEEE-2383 bus system confirm that the proposed algorithms can significantly improve the accuracy and efficiency of baseline algorithms that do not leverage the cluster structure of multiple line outages. In Chapter 3, we follow the problem in Chapter 2, but consider no prior knowledge of the power systems or model parameter. A pure data-driven approach is proposed to detect the events using only the frequency measurements from the power systems. Due to the complex and unknown statistic of the frequency measurements, the popular change point detection framework cannot work directly. To overcome this issue, we first use a recurrent neural network based model to predict the future measurements using the observed measurements, thus the prediction error can be monitored on-line. We note the prediction error follows to a normal distribution and the parameters can be learned by the predictor, thus makes this detection problem trackable. We assume the post change prediction error follows to a normal distribution with unknown parameters, and then this change point detection problem can be solved by many exists procedures. In this work, a Rao test based CUSUM procedure is used to recursively calculate the CUSUM statistic without complex maximum likelihood estimation calculation. We evaluate the performances through real PMU events data. In Chapter 4, we develop an admission control and scheduling mechanism to jointly consider the revenue of charging stations and the service requirements of customers in large-scale smart grid systems. We consider an offline setting where given a set of inflexible service demands from customers. Our objective is to design an admission control and scheduling scheme, so that the admitted demands can be satisfied while the revenue of the charging station is maximized. We first propose a calculus based scheduling algorithm, and show that if the system is underloaded, it maximizes the revenue of the charging station. We then consider the general overloaded case, and prove that it is NP-complete. A heuristic algorithm is developed to greedily decline a subset of demands until the remaining demands can be satisfied. An on-line algorithm is also proposed when the demands are unknown until them starts. We evaluate the performances of the proposed algorithms through simulations and real data set. In Chapter 5, we consider a scenario where a source continuously monitors an object and sends time-stamped status updates to a destination through a rate-limited link. In order to measure the ``freshness'' of the status information available at the destination, we adopt the metric called Age of Information (AoI). We first assume all updates are of the same size, and arrive randomly at the source according to a Bernoulli process. Due to the link capacity constraint, it takes d (d>=2) time slots for the source to complete the transmission of an update. Therefore, when a new update arrives at the source during the transmission of another update, the source needs to decide whether to skip the new arrival or to switch to it, in order to minimize the expected average AoI at the destination. We prove that within a broadly defined class of online policies, the optimal policy should be a renewal policy, and has a sequential switching property. We then show that the optimal decision of the source in any time slot has a multiple-threshold structure, and only depends on the age of the update being transmitted and the AoI in the system. We then consider the various update sizes, and similar results are derived. The thresholds are then numerically identified by formulating the problem as a Markov Decision Process (MDP).