改进的Elman模型与递归反传控制神经网络

An Improved Elman Model and Recurrent Back-Propagation Control Neural Networks

在Elman网络的基础上提出了两种改进网络:输出-输入反馈Elman网络和输出-隐层反馈Elman网络模型,并以前者作为误差反传的通道,建立了递归反向传播控制神经网络模型.在Lyapunov稳定性意义下分别给出了改进网络的稳定性证明,得到了保证网络稳定收敛的最佳自适应学习速率.分别用Elman网络及其改进网络对超声马达进行了模拟.利用改进的Elman网络模型,除了可以较好地模拟马达速度以外,还得到了一些有意义的结果,据此可以根据现场数据采样的情况,选用不同的网络模型.模拟实验结果表明,递归反向传播控制神经网络对多种形式的超声马达参考速度都有很好的控制效果.

Two improved Elman neural networks, output-input feedback Elman network and output-hidden feedback Elman network are presented based on the Elman neural network. By using the output-input feedback Elman network as a passageway of the error back propagation, a recurrent back propagation control neural network model is developed. The stability of the improved Elman neural networks is proved in the sense of Lyapunov stability theory. The optimal adaptive learning rates are obtained, which can guarantee the stable convergence of the improved Elman networks. The ultrasonic motor is simulated by using the Elman and improved Elman networks respectively. Besides simulating the speed of the ultrasonic motor successfully, some useful results are also obtained. According to the results, the different network models based on the sampling situation in the fieldwork can be chosen. Numerical results show that the recurrent back propagation control neural network controller has good effectiveness for various kinds of reference speeds of the ultrasonic motor.