基于遗传算法的神经网络在发电机定子超高频局部放电模式识别中的应用

介绍了发电机定子局部放电高频在线监测系统的结构特点及对局部放电特征的提取，应用BP算法、自适应遗传算法AGA和AGA—BP神经网络对发电机定子超高频局部放电的三种放电类型进行了模式识别。结合AGA和BP算法各自的优点，构造了AGA—BP混合算法作为神经网络的学习算法。实验结果表明，AGA—BP神经网络解决了BP神经网络对初始权值敏感和容易局部收敛的问题，提高了AGA神经网络的收敛速度、稳定性和求解质量。

Application of genetic algorithm based neural network in UHF PD pattern recognition of generator stator

The structural features of on - line high - frequency monitoring system for generator stator PD(Partial Discharge) is introduced,and the PD characteristics are picked up.By using BP -NN (Neural Network),AGA(Adaptive Genetic Algorithm) -NN and AGA - BP - NN,three types of ultra- high - frequency PD pattern appeared in generator stator are recognized:interior discharge,end discharge and slot space discharge.An AGA - BP hybrid algorithm based on AGA and BP algo -rithm is presented to train NN.Tests show that,it is not as sensitive to initial weights and easy in premature convergence as BP - NN algorithm,and enhances the convergence speed,stability and solution quality of AGA-NN algorithm.