A hybrid framework of short-duration simulation and ANN-based transient stability assessment for contingency screening

A hybrid framework of numerical simulation and ANN mapping is proposed for contingency screening of on-line dynamic security assessment in the paper. In the proposed framework, the three-layer feed-forward neural networks are employed as pattern classifiers to build fast relation mappings between the transient stability results and selected input attributes. The numerical simulation technique is employed to produce the input attributes for the ANNs by short-duration integration terminated at the fault clearing time. Each ANN is designed to deal with a single contingency scenario. After being trained using a semi-supervised back-propagation learning algorithm, the ANN can derive a continuous-spread stability index to indicate the relative stability degree for the specific contingency. Based on the derived stability index, a reasonably conservative classification threshold is set to avoid omission of insecure cases which is unacceptable to system operation. Therefore, the ANN classifiers can be more safely applied to on-line contingency screening in a practical environment. Applications to the 10-unit New England power system demonstrate the validity of the proposed approach.