基于改进Elman神经网络的电子元件性能预测

Prediction for Electronic Component Performance Based on Modified Elman Neural Network

为提高传统Elman神经网络的动态性能,通过增加输出层与承接层之间的反馈环节,提出了一种新的改进的Elman神经网络模型,利用梯度下降原理对其学习算法进行了推导。同时引入附加动量和变学习率算法,建立了基于改进Elman神经网络的预测方法,并将其应用于电子元件性能参数的预测中。仿真实验证明,相比于BP和传统Elman神经网络,改进后的Elman神经网络训练速率快,预测精度高,具有良好的动态性能。由此可见,改进的Elman神经网络模型在对具有非线性时序特征参数的预测中,具有良好的应用前景。

In order to improve the dynamic performance of traditional Elman neural network, a modified Elman neural network is proposed by adding the output feedback link between output layer and context layer, and its learning algorithm is deduced by using the theory of gradient descent. A prediction method based on modified Elman neural network is built with the additional momentum and variable learning speed algorithm, and it is applied in the prediction of electronic component performance parameters. The simulation experiments show that the modified model has good dynamic performance compared with back propagation （BP） and traditional Elman neural network, and it has faster training speed and higher precision. Thus, the modified Elman neural network model has a good application prospect in the prediction of nonlinear and time series parameters.