改进型柔性神经网络及在开关磁阻电机磁化曲线建模中的应用

针对传统神经网络建模的不足提出了一种改进型的柔性神经网络。阐述该网络在学习、训练过程中不仅可以调节连接权,而且加强了对网络非线性函数参数的实时修改,通过多自由度的训练与调整,使所建网络达到最佳的性能。给出了所建网络的结构与学习算法,并通过算例的形式将其与传统BP神经网络及传统已有柔性神经网络进行了全方位比较。结果表明,改进型网络由于其三自由度调节参数的能力,具有比传统BP网络及已有柔性神经网络更强的学习能力,它以最少的迭代循环次数实现了期望精度。

An Improved Flexible Neural Network and Application to Magnetization Curves Modeling of Switched Reluctance Motor

An improved flexible neural network is proposed in the light of the disadvantages of traditional neural networks. The proposed neural network could not only adjust the connection right during the process of study and train, but improve the real-time modification of network non-linear function parameters. The built network could be optimized through training and adjustment with multi- freedom. The structure and algorithm of the built network is given and is compared with the traditional BP neural network as well as the existing traditional flexible neural network through examples. The results show that the improved network features stronger capability in learning, due to its capability in triple-freedom adjustment parameter. The expected accuracy is realized with least times of iterativeness cycling.