A novel approach to improve model generalization ability in dynamic equivalent of active distribution network

With the development of renewable resources, large amounts of Distributed Generation (DG) units penetrated into distribution network. Instead of traditional passive PQ bus equivalence, DG characterized Active Distribution Network (ADN) need to be appropriate modeled as active components to represents its dynamic behaviors. Though existed ADN equivalent research considered the inverter-based DG units model, the uncertainties impacts, such as system faults or contingencies, between ADN and grid are not to be investigated on the model. Based on previous dynamic equivalent of ADN process, the equivalent model may not robust enough to reflect the correlated impacts between original ADN and transmission system. To be specific, equivalent model cannot predict the unknown fault based on historical analyzed faults: when the fault condition changed, the ADN model may not be utilized any more. This phenomenon terms as weak Model Generalization Ability (MGA). In order to overcome the issues, this paper presents a novel approach to improve model generalization ability in dynamic equivalent of ADN. An algorithm based on correlation and trajectory sensitivity analysis are introduced to screen out “Key Parameters”, then a fault information database which contains multiple faults is established. This multiple faults and “Key Parameter” based parameter identification scheme can eliminate the influence contaminations on modeling process effectively. The MGA of ADN equivalent model is able to increase significantly through the proposed approach. A simulation case on modified IEEE two-area four-machine power system with sample results are also provided to verify the improvement of MGA.