一種基于增量加權(quán)平均的在線序貫極限學(xué)習(xí)機算法
發(fā)布時間:2019-01-28 07:45
【摘要】:針對在線序貫極限學(xué)習(xí)機(OS-ELM)對增量數(shù)據(jù)學(xué)習(xí)效率低、準(zhǔn)確性差的問題,提出一種基于增量加權(quán)平均的在線序貫極限學(xué)習(xí)機(WOS-ELM)算法.將算法的原始數(shù)據(jù)訓(xùn)練模型殘差與增量數(shù)據(jù)訓(xùn)練模型殘差進行加權(quán)作為代價函數(shù),推導(dǎo)出用于均衡原始數(shù)據(jù)與增量數(shù)據(jù)的訓(xùn)練模型,利用原始數(shù)據(jù)來弱化增量數(shù)據(jù)的波動,使在線極限學(xué)習(xí)機具有較好的穩(wěn)定性,從而提高算法的學(xué)習(xí)效率和準(zhǔn)確性.仿真實驗結(jié)果表明,所提出的WOS-ELM算法對增量數(shù)據(jù)具有較好的預(yù)測精度和泛化能力.
[Abstract]:In order to solve the problem of low efficiency and poor accuracy of online sequential extreme learning machine (OS-ELM) for incremental data, an on-line sequential extreme learning machine (WOS-ELM) algorithm based on incremental weighted average is proposed. The residual of the original data training model and the residual of the incremental data training model are weighted as the cost function, and the training model used to balance the original data and the incremental data is deduced, and the fluctuation of the increment data is weakened by the original data. It can improve the learning efficiency and accuracy of the algorithm by making the online extreme learning machine more stable. Simulation results show that the proposed WOS-ELM algorithm has good prediction accuracy and generalization ability for incremental data.
【作者單位】: 東北大學(xué)計算機科學(xué)與工程學(xué)院;東軟公司軟件架構(gòu)新技術(shù)國家重點實驗室;
【基金】:國家863計劃項目(2015AA016005) 國家自然科學(xué)基金項目(61402096,61173153,61300196)
【分類號】:TP18
本文編號:2416850
[Abstract]:In order to solve the problem of low efficiency and poor accuracy of online sequential extreme learning machine (OS-ELM) for incremental data, an on-line sequential extreme learning machine (WOS-ELM) algorithm based on incremental weighted average is proposed. The residual of the original data training model and the residual of the incremental data training model are weighted as the cost function, and the training model used to balance the original data and the incremental data is deduced, and the fluctuation of the increment data is weakened by the original data. It can improve the learning efficiency and accuracy of the algorithm by making the online extreme learning machine more stable. Simulation results show that the proposed WOS-ELM algorithm has good prediction accuracy and generalization ability for incremental data.
【作者單位】: 東北大學(xué)計算機科學(xué)與工程學(xué)院;東軟公司軟件架構(gòu)新技術(shù)國家重點實驗室;
【基金】:國家863計劃項目(2015AA016005) 國家自然科學(xué)基金項目(61402096,61173153,61300196)
【分類號】:TP18
【相似文獻】
相關(guān)期刊論文 前4條
1 呂鎮(zhèn)邦;周利華;;基于有序加權(quán)平均算子的概率模糊認(rèn)知圖[J];計算機科學(xué);2008年12期
2 彭會萍;曹曉軍;楊永旭;;基于決策距離的多傳感器信息融合加權(quán)平均算法[J];微計算機信息;2012年09期
3 張秋余;鄭魯寧;孫磊;;云理論和多傳感器加權(quán)平均的性能指標(biāo)描述[J];計算機工程與設(shè)計;2009年11期
4 陳梵;薛敏;葛蓬;李俊達;;基于云理論和多傳感器加權(quán)平均的關(guān)鍵性能指標(biāo)狀態(tài)描述[J];微電子學(xué)與計算機;2012年01期
相關(guān)會議論文 前1條
1 曹潔;徐微;;基于加權(quán)平均方法的交通流檢測技術(shù)的研究[A];2007通信理論與技術(shù)新發(fā)展——第十二屆全國青年通信學(xué)術(shù)會議論文集(下冊)[C];2007年
,本文編號:2416850
本文鏈接:http://sikaile.net/kejilunwen/zidonghuakongzhilunwen/2416850.html
最近更新
教材專著