Transient Stability Assessment In Bulk Power Grid Based On Improved Support Vector Machine

Transient Stability Assessment(TSA)is closely related to the security and stability operation of power system.It becomes more and more difficult of TSA with the rapid development of renewable energy because of the randomness and volatility.It is more important to propose a method of TSA which is used in power system with renewable energy.In this paper,the feature selection method which is based on maximal relevance and minimal redundancy(mRMR)is proposed to determine the optimal feature quantity,and then the refined pinball loss support vector machines(PinSVM)method is proposed to improve the computing speed on the premise of the stability of evaluation.The main contents are as follows.Adopt the feature selection method of transient stability assessment based on improved maximum correlation minimum redundancy criterion.Based on the standard mRMR method,a weight factor is introduced into the maximum correlation minimum redundancy criterion to refine the measurement of feature correlation and redundancy.The original feature set of transient stability represented by system index is constructed with onsidering the real post-fault information provided by the phasor measurement unit.The improved mRMR is applied on feature selection.A set of nested candidate feature subset is obtained by incremental search algorithm,and the classification performance of each candidate feature subset is verified by support vector machine classifier,and the feature subset with the maximum classification accuracy is selected.A novel transient stability assessment method is proposed using sequential minimal optimization support vector machine with pinball loss in bulk power grid with renewable energy.The concept of quantile is introduced in the loss function of the traditional support vector machine to improve the stability of transient stability assessment.Sequential minimum optimization strategy is used to solve Pin-SVM to reduce computing time.Firstly,a group of system-level and renewable energy classification features are first extracted from the power system operation parameters to build the original feature set.These features can reflect the influence of renewable energy access on the transient stability of power systems.Furthermore,a feature selection approach is employed to evaluate the classification capability of the original features for feature selection.Then,feature sets are mapped to a higher dimensional space and the TSA problem is then transformed into a linear classification problem.The definition of the nearest point between the two classes is changed by the concept of quantile.It can effectively reduce the impact of critically stable interference samples due to increased uncertainty in renewable energy.At the same time,the method improves the stability of transient stability assessment with renewable energy.Furthermore,the sequential minimum optimization strategy is introduced to transform the high-dimensional binomial optimization problem of PinSVM into multiple low-dimensional binomial optimization problems,which can effectively improve the calculation speed.