根据UHF信号特征的GIS局部放电模式识别

综合自适应遗传算法和BP算法各自的优点,构造了基于两者混合训练的神经网络,应用到GIS局部放电超高频的模式识别。分别用基于自适应遗传算法的神经网络、基于BP算法的神经网络,以及基于自适应遗传算法和BP算法混合训练的神经网络对用局部放电超高频检测系统检测到的GIS中4种模式的局部放电进行了识别。实验结果表明,基于自适应遗传算法和BP算法混合训练的神经网络提高了神经网络训练的收敛速度,保证了收敛的可靠性,具有较高的识别率和较强的泛化能力。

Partial Discharge Pattern Recognition of GIS Based on the UHF Signal Characteristics

By combining the advantage of adaptive genetic algorithm and back-propagation algorithm, a neural network trained by a hybrid algorithm based on these two algorithms was presented in this paper. It was applied in pattern recognition of partial discharge UHF detection for GIS. Four patterns of partial discharge in GIS detected by UHF detection system were recognized by the neural network trained by adaptive genetic algorithm, back-propagation algorithm and hybrid algorithm respectively. The experimental results show that this hybrid algorithm can improve the rate of convergence and assure the reliability of convergence in process of training, so it has a remarkable discrimination and generalization ability.