針對(duì)傳統(tǒng)的時(shí)間序列分析方法預(yù)測(cè)科學(xué)數(shù)據(jù)效果較差的特點(diǎn)，提出了一種結(jié)合自組織神經(jīng)網(wǎng)絡(luò)和灰色理論的時(shí)間序列預(yù)測(cè)方法。該方法利用度量時(shí)間序列相似性距離函數(shù)，將時(shí)間序列按照其變化規(guī)律分成不同的類(lèi)別，并在GM算法中針對(duì)白化參數(shù)進(jìn)行優(yōu)化，對(duì)科學(xué)數(shù)據(jù)時(shí)間序列進(jìn)行自組織聚類(lèi)，針對(duì)各類(lèi)別采用灰色理論建立預(yù)測(cè)模型。試驗(yàn)表明，該模型適合科學(xué)數(shù)據(jù)的變化特點(diǎn)，提高了預(yù)測(cè)精度。
關(guān) 鍵 詞 神經(jīng)網(wǎng)絡(luò); 灰色理論; 時(shí)間序列; 預(yù)測(cè)

Traditional methods have poor efficiency and effect to deal with the scientific data series forecasting. In this paper, a forecasting algorithm based on grey theory and self-organized map neural networks is proposed. Firstly, the scientific data time series cluster in self-organized mannar. Then the forecast model is established with grey theory. In clustering, a distance criterion is proposed to scale the difference between series. In grey theory, the whiten parameter is optimized. The experiments show that this algorithm surpasses those traditional forecasting methods in precision and time efficiency.
Key words neural network; grey theory; time series; forecasting

大規(guī)?？茖W(xué)計(jì)算是對(duì)連續(xù)變化的科學(xué)現(xiàn)象進(jìn)行離散抽樣的數(shù)值模擬，計(jì)算過(guò)程除需花費(fèi)大量的計(jì)算資源外，還必然會(huì)產(chǎn)生大規(guī)模的數(shù)值模擬數(shù)據(jù)。仿真物理現(xiàn)象的數(shù)值模擬數(shù)據(jù)由數(shù)值計(jì)算程序通過(guò)一系列時(shí)間步產(chǎn)生，是對(duì)連續(xù)空間數(shù)據(jù)離散采樣、帶有網(wǎng)格結(jié)構(gòu)的時(shí)間序列[1-2]。當(dāng)數(shù)據(jù)規(guī)模增大時(shí)，時(shí)間序列的分析和處理都因?yàn)閿?shù)據(jù)特征復(fù)雜、規(guī)模龐大而難以進(jìn)行。數(shù)值模擬應(yīng)用中，大規(guī)?？茖W(xué)數(shù)據(jù)的時(shí)間序列分析[3-5]，已經(jīng)成為影響科學(xué)數(shù)據(jù)挖掘的一個(gè)重要瓶頸問(wèn)題。
科學(xué)數(shù)據(jù)具有幾何特性、拓?fù)浣Y(jié)構(gòu)、物理屬性和時(shí)變性四個(gè)基本特征。特別是對(duì)于時(shí)變性來(lái)講，存在對(duì)時(shí)間規(guī)則的網(wǎng)格數(shù)據(jù)和對(duì)時(shí)間步間隔變化的網(wǎng)格數(shù)據(jù)。常用的時(shí)間序列分析方法假定數(shù)據(jù)時(shí)間間隔相等，因此不適于科學(xué)數(shù)據(jù)時(shí)間序列分析。影響科學(xué)現(xiàn)象的物理因素是多種多樣的，各種各樣的物理因素對(duì)物理現(xiàn)象的具體影響仍需進(jìn)一步研究，因此對(duì)科學(xué)數(shù)據(jù)時(shí)間序列的預(yù)測(cè)常具有灰色性特征。本文建立灰色預(yù)測(cè)模型，即以自組織神經(jīng)網(wǎng)絡(luò)聚類(lèi)后的時(shí)間序列為基礎(chǔ)，運(yùn)用灰色模型預(yù)測(cè)科學(xué)數(shù)據(jù)時(shí)間序列的變化。