Pattern Recognition-based Predictive Under-Frequency Load Shedding

This paper presents the simple but effective idea of improving under-frequency load shedding. The main disadvantage of conventional under-frequency load shedding schemes is their lack of adaptability, as they do not distinguish between different frequency gradients. Since they are designed for the worst-case scenario, their inflexibility usually leads to over-shedding. To make them more precise, it is necessary to improve the situational awareness of the load shedding relays. A large number of solutions indicate that the rate of change of frequency is the key factor. However, not many implementations could be found in the literature. Wide-area measurement system has brought numerous new innovative solutions, but their complexity and communication needs make them less robust and therefore less interesting for actual use. In this paper we present a supplement to the conventional scheme that eliminates all these shortcomings. In fact, it retains the current relay settings and introduces an additional stage whose settings can be automatically adjusted according to the expected future frequency trend. The prediction is based on the use of a system frequency response model instead of the commonly used polynomials of different degrees. To use this model for prediction, any potentially critical frequency deviation must be detected. It turns out that a principal component analysis can be used to successfully identify such events. Since the proposed method is capable of detecting a less severe frequency deviations, it will in such a situation disconnect fewer consumers than the conventional under-frequency load shedding approach.