Abstract:

The development of microgrid with high proportion of renewable energy is one of the important means to construct new modern power systems so as to achieve energy security and low carbon emissions. However, amid the analysis of the dynamic characteristics of microgrid-integrated power system, the current equivalent models appear to be not robust enough. Specifically, these models can well reproduce the behaviors of actual system under the faults in training set, they may not be able to reflect actual system responses under other unknown faults (non-training faults). In regard to this, k-means++ is introduced first to effectively distinguish the typical operation condition of microgrid such that the randomness and time-varying characteristics of the system can be represented. Next, key parameter selection-based parameter identification method is applied to avoid the issue of multiple solutions in parameter identification process. Then, the convolutional neural network is used to generalize the model parameters with respect to different typical system operation conditions. Additionally, online matching of equivalent model parameters is achieved by virtue of Fisher discriminant analysis. Finally, the effectiveness of the proposed method has been verified in a real microgrid system in China.

Key words: microgrid, equivalent modeling, robustness, k-means++ clustering, convolutional neural network, Fisher discriminant analysis