Towards fault tolerance and attack resiliency in cyber-physical energy systems through learning from data streams under harsh learning conditions

点击翻译按钮获取中文摘要

The welfare and security of modern societies rely on the safe and secure operation of complex safety-critical cyber-physical systems (CPSs). With advancement of digitalization in modern industries, nowadays, CPSs are applied in various technical areas including energy, automotive, medicine, industries, transportation, and defense. CPSs are defined as the close interaction and seamless combination of physical processes and cyber components. These cyber modules monitor, make decisions and control the physical components, and adapt themselves to changes in non-stationary environments. Safety and security are major concerns for the CPS operation due to the great potential for the occurrence of faults, the broad attack surface, and the severe consequences of attacks and faults. CPSs dependency on digitalization, wireless communication, and remote control systems increases their vulnerabilities to malicious threats and cyber-attacks, which lead to the loss of system integrity and functionality. These widen the range of possible problems that cannot be properly addressed, unless under a unified view of safety and security characteristics. To maintain a high level of performance, safety, and security in CPSs; abnormal system operations and anomalies including faults (safety-related incidents) as well as malicious threats and cyber-attacks (security-related incidents) must be detected quickly. Although cyber-attacks and faults originate from different sources, may have similar signatures, and result in increased operating costs, the chance of line shutdown, and the possibility of detrimental environmental impacts. Nevertheless, the source and severity of each must be identified, so that corrective actions can be taken promptly. Therefore, early detection and diagnosis of cyber-attacks and faults are strategically essential for companies to remain competitive in world markets. In addition, from the unified view of safety and security, classifying cyber-attacks from faults is of paramount importance for assessing their possible effects on the system integrity and choosing an appropriate set of preventive and recovery actions for resilience. Furthermore, many cyber-attacks and system malfunctions do not have only safety, security, or life-threatening consequences but may seriously affect the ecology. To enhance the system resiliency, it is crucial to integrate the knowledge on machine learning, big data analytics, cybernetics, cyber security, and safety, to address potential failures and malicious threats. The objective is to focus on missing principal knowledge in detection, diagnosis, and prognosis along with machine learning, big data analytics, and cybernetics that would pave the way together towards attack-resilient and fault-tolerant CPSs. Although the proposed research can be applied to a wide range of applications, the focus of this proposal is toward cyber-physical energy and power systems, with applications to modern power grids.