Power System Transient Stability Assessment Based On Semi-supervised Support Vector Machine

The traditional power system transient stability assessment methods based on the mathematical model can not meet the requirements of online assessment due to the increasingly complex dynamic behaviors of modern power system.Although the transient stability assessment methods based on the supervised support vector machine can achieve online assessment,problems of the existing models still remain including poor generalization and adaptation abilities to the actual system and lacking of online learning capacity.This paper studies the power system transient stability assessment methods based on semi-supervised support vector machine to overcome shortcomings of the existing methods.The main research contents are as follows:The transient stability assessment methods based on the supervised support vector machine and the concept of stability region only use offline time domain simulation to get samples.The generalization abilities of assessment models obtained from this way are limited.Besides,these methods treat all training samples equally in the objective function making the classification hyperplane susceptible to outliers.To solve the above problems,this paper proposes a power system transient stability assessment method based on semi-supervised fuzzy twin support vector machine.This method uses the steady-state power flow data before the fault occurred to form the original feature set.The feature set is filtered and compressed by Laplacian score and principal component analysis to obtain feature subsets which are strongly related to system stability.In order to fully exploit the information implicit in unlabeled samples and achieve semi-supervised learning,add Laplacian regular terms describing the distribution of the unlabeled samples to the objective function of the twin support vector machine.At the same time,a new fuzzy membership function which can correctly distinguish between support vector and outlier is constructed.On this basis,a fuzzy semi-supervised twin support vector machine model is formed and it is used to classify the reduced labeled samples and unlabeled samples simultaneously.Model parameters are optimized by whale optimization algorithm.The simulation results of WSCC-9 bus system and IEEE-39 bus system show that the proposed method has higher generalization ability and prediction accuracy than the traditional power system transient stability assessment methods based on supervised support vector machine.The existing methods based on the supervised support vector machine can not be updated online and their adaptability to the power system's complex dynamic environment is poor on the concept of transient stability assessment within stability domain.In addition,the model optimization of parameters is time consuming.This paper proposes a transient stability assessment method based on online adaptive transductive support vector machine by combining incremental learning and decrement learning with transductive learning.The proposed method can not only evaluate the consequence when the fault occurs,but also can use the new samples to realize the model self-renewal and correction.Besides,it can adaptively optimize the model through parameter perturbation.The feasibility and effectiveness of the proposed method are verified by IEEE-68 bus system.The simulation results show that the proposed method is more accurate than the traditional batch induction method.Evaluation time is in milliseconds and parameter optimization time is nearly 5 times better than full optimization.The proposed method can meet the needs of online transient stability assessment.