A Local Measurement Based Protection Methodology for Initiating Controlled Islanding in Power Systems

Despite layers of protection at work, power systems still have experienced major blackouts in the past due to cascading outages. Hence, preventing the cascading outage holds the key to the prevention of such blackouts in the future. To address this problem, researchers have proposed controlled islanding as a last resort that splits the power system into individual intentional islands. A controlled islanding methodology needs to (a) predict system instability “when” islanding is to be initiated and (b) find locations “where” islanding can be implemented. The focus of this research is mainly on the “when” aspect of islanding. Majority of the existing methods to solve this aspect are based on wide area measurement system (WAMS), which are either offline (Decision Tree) based that require time-consuming periodic offline simulation, or real-time based using a computationally intensive algorithm. The complexity in most of the real-time methods is due to the involved calculation of dynamic system equivalents after generator coherency determination followed by further sophisticated calculation. The proposed methodology simplifies the real-time procedure of initiating controlled islanding by using a local measurement-based generator out-of-step (OOS) protection along with the WAMS based coherency detection method. The equal area criterion (EAC) in the time domain method is utilized for the local OOS detection at each generator bus that requires only a single measurement of its active power output. In the proposed methodology, this EAC prediction result is communicated from each generator bus to the central “when” unit as a stable/OOS flag unlike the real-time phasor measurements used in the existing WAMS based methods. The proposed methodology is also independent of the system topology compared to the methods based on dynamic system equivalents. Other research work in the past has shown that the system-wide stability or spontaneous splitting of a system (blackouts) can be predicted by monitoring merely the generators in that system. The proposed scheme further simplifies this approach by monitoring only a set of generators that are swinging away from the reference bus and its pre-fault steady-state condition. This approach is valid because a generator experiences instability when a pole slip (OOS condition) occurs as the rotor deviates past 180 degrees phase angle. Therefore, when a disturbance or set of disturbances occurs, the proposed algorithm triggers a coherency identification process to identify two coherent group of generators based on the nature of their rotor swing. After the coherency detection, the time domain EAC status flags of generators swinging away from the pre-fault condition are used to decide “when to island” the system. The islanding is initiated when all these generators are predicted unstable. The status flags from the inbuilt generator OOS relays are also incorporated for the condition when these OOS results are faster. For the “where to island” block, an existing method based on spectral clustering with graph reduction technique is adopted, which is also triggered in parallel after the coherency detection. A graph representation of the power network is obtained in this method, which is reduced using different graph reduction techniques. The spectral clustering algorithm in the reduced graph provides clusters that represent the islands and also gives the locations ``where'' to separate the system if the decision to island is made. Testing of the proposed method is undertaken through two developed test cases of cascading outage scenario using the IEEE 39 bus system implemented in the real-time digital simulator (RTDS). These test cases are based on outages due to transmission line overload protection and generator OOS protection relays, which are modelled in RTDS. A commonly industry-used method, Quadrilateral blinder based OOS protection, is tested for the generator and transmission line OOS protection. The sequence of the cascading outage event and the operation of the proposed islanding methodology are presented in this thesis. In the first test case, the proposed method predicted system instability 0.8062 s before blackout (unintentional splitting) and 0.364 s before the first cascading event following an initial fault. In the second test case, system instability is predicted 2.1875 s before the blackout and 0.1265 s before the second cascading event following an initial fault. Comparison of lost components (generators, loads and transmission lines) with and without controlled islanding scheme also demonstrated the significant reduction in the area of the blackout