摘要
激励函数对深度神经网络的非线性逼近性能具有重要影响,其选择与任务相关。针对这一问题,提出基于自适应选择算法的深度置信回声状态网络模型,并应用于海洋环境多因素时间序列预测。该模型集成了14种激励函数,通过预测性能比较实现自适应选择功能。仿真结果表明,该模型能够正确选择出最优激励函数,具有良好的海洋数据多因素预测能力。
Activation function(AF)has an important effect on the nonlinear approximation performance of deep neural networks.The choice of AFs is task-related.For this problem,a deep belief echo-state network based on self-adaptive selection(SAS-DBEN)is proposed for ocean-related multi-factor time series prediction.In this model,14 activation functions are integrated,and the self-adaptive selection is implemented by comparing the prediction performance.Experimental results show that SAS-DBEN can select the optimal AF correctly and has a good multi-factor prediction ability of ocean data.
作者
王嘉琳
金宇悦
李志刚
WANG Jia-lin;JIN Yu-yue;LI Zhi-gang(North China University of Science and Technology,Tangshan 063210,China)
出处
《电脑知识与技术》
2021年第22期1-2,19,共3页
Computer Knowledge and Technology
基金
唐山市科技计划项目(19150230E)。
关键词
海洋环境数据
时间序列预测
深度置信回声状态网络
自适应选择算法
激励函数
ocean environment data
time series prediction
deep belief echo-state network
self-adaptive selection algorithm
activation function
