论文部分内容阅读
为避免权值反复迭代修正的冗长BP训练过程,避免传统方法陷入局部极小点,根据多项式理论,构造了一种新型前向神经网络模型,推导了基于最速下降法的误差反传算法和基于伪逆的直接确定法.仿真结果显示,迭代方法和伪逆直接确定法都能达到比较高的工作精度(10-6).
In order to avoid the lengthy BP training process of iterative iterative weight correction and avoid the traditional method falling into local minimum, a new type of forward neural network model is constructed based on polynomial theory, and the error back propagation algorithm based on the steepest descent method is deduced. The direct determination method of pseudo-inverse.The simulation results show that both the iterative method and the pseudo-inverse direct determination method can achieve higher working accuracy (10-6).