|
摘要: |
在应用混沌神经网络(CNN)进行同步发电机的建模过程中,对于CNN的学习,网络训练过程的收敛性很难控制。在研究了BP学习算法及其一些改进方法进行人工神经网络训练的轨迹收敛特性后,观测到运用梯度下降动量与自适应学习速率相结合的BP学习算法的神经网络训练轨迹的收敛特性良好。在用基于Aihara混沌神经元构成的3层反馈CNN进行同步发电机建模的应用中,用该BP学习算法对CNN进行了训练。结果表明:用该BP算法进行CNN发电机建模具有学习速度快和均方误差曲线轨迹收敛性好的特点,而且所建立的CNN同步发电机模型运行的动态过程误差小。 |
关键词: BP算法,混沌神经网络,发电机,建模 |
DOI: |
分类号:TM31 |
基金项目:上海市高等学校科学技术发展基金
,
上海海事大学校科研和教改项目 |
|
BP learning algorithm for CNN generator modeling |
SHI Wei-feng
|
Abstract: |
In the application of CNN(Chaotic Neural Networks) to synchronous generator modeling,the training process convergence of network learning is difficult to control.Based on the characteristic study of trajectory convergence in artificial neural network training using BP(Back Propagation) learning algorithm and its improved algorithms,it is found that the improved BP algorithm combining gradient descent momentum and adaptive learning rate is better,which is then used to train the CNN in the synchronous generator modeling based on a three-layer feed back CNN with Aihara chaotic neuron.Results indicate that,the learning speed is fast,the trajectory convergence of mean square error is better and the dynamic process error of synchronous generator operation is small. |
Key words: BP algorithm,chaotic neural network,generator,modeling |