Feature Selection For Transient Stability Assessment Based On Genetic Simulated Annealing Algorithms

Artificial neural network has alluring prospect of application in the area of power system transient stability assessment, where selection of the input feature variables and the design of classifier are the two critical problems to the transient stability assessment based on neural network, which greatly influence the quality of assessment. For the problem of the input feature selection, a new method that combined the genetic algorithm with the simulated annealing algorithm, genetic simulated annealing algorithm, is proposed in this paper to find the optimal feature set that can correctly reflect the physical nature of power system transient stability directly or indirectly, and then give a better representation of the system dynamic characteristics. For the design of classifier, this paper proposed an improved particle swarm optimization algorithm, multi-phase particle swarm optimization, to train the neural network for power system transient stability assessment. It provides a way to solve the problems that the error back propagation algorithm or other algorithms take much training time and tend to get into local minimum in neural network training and also improved the classification accuracy rate of neural network greatly. The application results on the 8-machine 36-bus system and the 10-machine 39-bus New England system reveal the validity of the proposed approach.