|
چکیده
|
Machine learning technologies have been applied to improve the real-time performance of small-signal stability assessment, while achieving a high accuracy requires numerous samples, and the error may go large if the model is not updated over time. Additionally, the single model usually learns the general features of the given samples and lacks analysis of the particular characteristics within it, which may lead to a high frequency of large errors, especially at operating points (OPs) where the eigenvalue trajectories have sudden changes. Facing such issues, this paper introduces the concept of reference points (RPs) to accurately track the rightmost eigenvalue (RE) online, as the information of RP reflect the characteristics among its surrounding OPs. The performance of this model is sensitive to RPs, affinity propagation clustering that can accommodate the different characteristics of OPs is employed to determine the number of RPs and generate the corresponding group. In this paper, the data-driven networks are generated for each group and they are combined into a multi-network for accurate RE prediction. Moreover, to alleviate the computational stress, this paper proposes an adaptive partial update strategy based on dynamic time warping algorithm, which avoids updating all the networks in each sliding time window. Numerical studies based on two test cases demonstrate that the proposed method can reduce the overall error of the data-driven model, decrease the frequency of large errors occurring on a single day, and accurately capture sudden changes in the RE trajectory.
|