摘要
针对传统神经网络算法在投资银行风险预测的应用中表现出预测准确性不高的问题,提出了一种基于动态参数优化神经网络的投资银行风险预测模型。首先根据动态合并与删减规则,对参数进行自适应动态调整,得到最为合适的神经网络模型,然后为了加速收敛和防止振荡,引入一个动量因子,最后修改误差函数,在保证网络训练误差尽可能小的情况下使网络具有较小的权值。仿真试验结果表明,本文提出的基于动态参数优化神经网络的投资银行风险预测模型相比较标准神经网络算法,具有更高的预测准确性。
Traditional neural network algorithm to predict the risk in the investment banking applications exhibit predictive accuracy is not high, risk prediction model proposed in this paper a dynamic parameter optimization of investment banking based on neural network, the first under the dynamic consolidation and deletion rules adaptive dynamic adjustment of parameters to obtain the most appropriate neural network model, then in order to accelerate convergence and prevent oscillation, the introduction of a momentum factor, last modified error function, to ensure the network training error as small as possible so that the network has a smaller case weights. The simulation results showed that the bank's risk prediction model based on neural network optimized dynamic parameters proposed investment compared to standard neural network algorithm, has higher prediction accuracy.
出处
《科技通报》
北大核心
2015年第9期192-195,共4页
Bulletin of Science and Technology
关键词
神经网络算法
投资银行
风险预测
动量因子
动态参数
neural network algorithm
investment banking
risk prediction
momentum factor
dynamic parameters
