本文選題：過(guò)程神經(jīng)網(wǎng)絡(luò) + 極限學(xué)習(xí)　； 參考：《控制與決策》2017年04期

【摘要】：為了提高太陽(yáng)黑子預(yù)測(cè)預(yù)報(bào)的精度,提出固定型極限學(xué)習(xí)過(guò)程神經(jīng)網(wǎng)絡(luò)(FELM-PNN)和增量型極限學(xué)習(xí)過(guò)程神經(jīng)網(wǎng)絡(luò)(IELM-PNN)兩種學(xué)習(xí)算法.FELM-PNN的隱層節(jié)點(diǎn)數(shù)目固定,使用SVD求解隱層輸出矩陣的Moore-Penrose廣義逆,通過(guò)最小二乘法計(jì)算隱層輸出權(quán)值;IELM-PNN逐次增加隱層節(jié)點(diǎn),根據(jù)隱層輸出矩陣和網(wǎng)絡(luò)誤差計(jì)算增加節(jié)點(diǎn)的輸出權(quán)值.通過(guò)Henon時(shí)間序列預(yù)測(cè)驗(yàn)證了兩種方法的有效性,并實(shí)際應(yīng)用于第24周太陽(yáng)黑子平滑月均值的中長(zhǎng)期預(yù)測(cè)預(yù)報(bào)中.實(shí)驗(yàn)結(jié)果表明,兩種方法的預(yù)測(cè)精度均有一定程度的提高,IELM-PNN的訓(xùn)練收斂性優(yōu)于FELM-PNN.
[Abstract]:In order to improve the accuracy of sunspot prediction, two learning algorithms, fixed limit learning process neural network (FELM-PNN) and incremental limit learning process neural network (IELM-PNN), are proposed. The number of hidden layer nodes of FELM-PNN is fixed. The Moore-Penrose generalized inverse of the hidden layer output matrix is solved by SVD, and the output weight of the hidden layer is calculated by the least square method. IELM-PNN increases the hidden layer node step by step, and the output weight value of the node is calculated according to the hidden layer output matrix and network error. The effectiveness of the two methods is verified by Henon time series prediction and is applied to the medium-long term prediction of sunspot smoothing monthly mean in the 24th cycle. The experimental results show that the prediction accuracy of the two methods is better than that of FELM-PNN to a certain extent, and the training convergence of IELM-PNN is better than that of FELM-PNN.
【作者單位】： 東北石油大學(xué)計(jì)算機(jī)與信息技術(shù)學(xué)院;山東科技大學(xué)信息科學(xué)與工程學(xué)院;
【基金】：國(guó)家自然科學(xué)基金項(xiàng)目(61170132) 黑龍江省自然科學(xué)基金項(xiàng)目(F2015021)
【分類號(hào)】：P182;TP183
，

本文編號(hào)：2053985