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

Application of Neural Network Based on Genetic Algorithm to the Recognition of UHF PD Patterns of Generator Stator

应用BP(误差反向传播算法)、AGA(自适应遗传算法)和AGA-BP神经网络对发电机定子超高频局部放电的三种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神经网络既解决了BP神经网络对初始权值敏感和容易局部收敛的问题,又提高了AGA神经网络的收敛速度、稳定性和求解质量。

Using BP-NN, AGA-NN and AGA-BP-NN, we distinguished between three types of partial discharge pattern appearing in generator stator. By combining adaptive genetic algorithm（AGA） with BP algorithm, this paper presents AGA-BP hybrid algorithm to train neural network（NN）. Tests gave satisfactory results of the classification process. Compared with BP-NN and AGA-NN, AGA-BP-NN can overcome the entrapment in local optimum of BP-NN and the premature of AGA-NN. Thus, the convergence, discrimination and generalization ability of AGA-BP-NN is improved remarkably.