基于二次指数平滑的BP神经网络模型研究及应用

The Study and Application of Improved BP Neural Networks

在分析标准BP算法的基础上,针对BP神经网络暴露的易于陷入局部最优和训练时学习新样本有遗忘旧样本的内在缺陷,结合二次指数平滑的思想,提出一种基于二次指数平滑的BP神经网络算法,对网络输入的原始数据进行二次平滑处理,提高BP网络的学习速度和预测精度。同时将研究结果应用到全国人均发电量的预测建模中,仿真结果表明所提出的方法具有逼近能力强、收敛速度快的优点。

Based on the analysis of classical BP algorithm, aiming at the internal bugs of BP net such as local convergence and leaving the old samples while learning new samples, a new PB algorithm based on double smoothing is put forward by combining BP net and double smoothing, which improves the studying speed and predictive precision through dealing with the original data by double smoothing. At last use the investigate result to the prediction model based on national generated energy per capita. The simulation suggests that it has a good rapidity of convergence and perfect predictive precision.