梁小林 李靜 郭敏
摘 要 根據(jù)更新幾何過(guò)程的定義和性質(zhì),運(yùn)用lasso類(lèi)方法構(gòu)建模型獲取其參數(shù)估計(jì),并通過(guò)數(shù)值模擬進(jìn)行檢驗(yàn),驗(yàn)證了該方法是有效的.在具有不同更新幾何過(guò)程比率的條件下,比較了lasso和自適用 lasso兩種方法的估計(jì),結(jié)果表明自適用lasso方法更適合更新幾何過(guò)程的參數(shù)估計(jì).
關(guān)鍵詞 數(shù)理統(tǒng)計(jì);參數(shù)估計(jì);lasso類(lèi)方法;更新幾何過(guò)程
中圖分類(lèi)號(hào) O212;O213.2文獻(xiàn)標(biāo)識(shí)碼 A
Abstract According to the definition and the properties of renewal-geometric process ,the Lasso type method was used to establish a model for the parameter estimation.Some simulation experiments were performed in the test,and the results show that the proposed method is effective.This article compared the performance of Lasso and adaptive Lasso method in different rates,which shows that adaptive Lasso method is more suitable for renewal-geometric process in parametric estimations.
Key words mathematical statistics;parametric estimation;lasso-type method;renewal-geometric process
1 引 言
基于維修問(wèn)題中的“修復(fù)非新”現(xiàn)象,Lam(1988)[1]首次提出了一類(lèi)單調(diào)的隨機(jī)過(guò)程模型,即幾何過(guò)程模型.對(duì)于幾何過(guò)程的參數(shù)估計(jì),Lam(2007)[2]利用對(duì)數(shù)變換將幾何過(guò)程的參數(shù)估計(jì)問(wèn)題轉(zhuǎn)化為線性回歸參數(shù)估計(jì)問(wèn)題,獲得了幾何過(guò)程的非參數(shù)估計(jì).然而,系統(tǒng)多樣性和環(huán)境的隨機(jī)性決定了單調(diào)幾何過(guò)程的局限性.為了得到更貼近實(shí)際問(wèn)題的維修模型,梁小林(2015)等[3]提出了基于分階段退化思想的更新幾何過(guò)程模型,并對(duì)更新幾何過(guò)程的相關(guān)性質(zhì)進(jìn)行研究,Niu(2016)等[4]利用更新幾何過(guò)程研究了維修問(wèn)題中的最優(yōu)更換策略.更新幾何過(guò)程的應(yīng)用決定了參數(shù)估計(jì)的重要.根據(jù)更新幾何過(guò)程的特點(diǎn),以幾何過(guò)程的非參數(shù)估計(jì)為基礎(chǔ),構(gòu)建了分階段線性回歸的參數(shù)估計(jì)模型,并將其轉(zhuǎn)化為lasso型問(wèn)題,同時(shí)進(jìn)行變量選擇與參數(shù)估計(jì)從而得出相關(guān)參數(shù)的有效估計(jì).
2 lasso與自適用lasso
2.1 lasso
一般最小二乘估計(jì)是通過(guò)最小化殘差平方和得到的,但一般的最小二乘估計(jì)存在不足,首先是預(yù)測(cè)精度不夠,最小二乘具有低偏移和高方差性,其次是它的模型可解釋性不強(qiáng).Tibshirani(1996)[5]打破傳統(tǒng)模型選擇思維,提出了新的的變量選擇技術(shù)lasso.lasso是在一般線性最小二乘的前提下加了約束,使各系數(shù)的絕對(duì)值之和小于某一常數(shù),由于這個(gè)約束的自然屬性,使得該回歸模型得出的回歸系數(shù)有的可能為零,因此便于選擇變量與解釋模型.
6 結(jié)束語(yǔ)
對(duì)于更新幾何過(guò)程基于lasso和自適用lasso的參數(shù)估計(jì)方法,隨機(jī)模擬結(jié)果驗(yàn)證了此方法的可行性,進(jìn)一步表明自適用lasso方法更優(yōu).雖然lasso類(lèi)方法具有同時(shí)變量選擇和參數(shù)估計(jì)的優(yōu)良性質(zhì),但對(duì)于更新幾何過(guò)程模型,還存在著需要改進(jìn)的地方.如a=1的情況需要另做考慮,變量選擇的準(zhǔn)確性還有待提高,參數(shù)估計(jì)的性質(zhì)需要進(jìn)一步研究等.
參考文獻(xiàn)
[1] LYEH.Geometric processes and replacement problem[J].Acta Mathematicae Applicatae Sinica,1988,4(4):366-377.
[2] L YEH.Nonparametric inference for geometric processes[J].Communication in Statistics.-Theory and Methods,2007,21(7):2083-2105.
[3] 梁小林,牛彩云,田學(xué).更新幾何過(guò)程及相關(guān)性質(zhì)[J].應(yīng)用概率統(tǒng)計(jì),2015,31(4):384-394.
[4] CY NIU,X L LIANG,B GE,X TIAN,Y W CHEN .Optimal replacement policy for a repairable system with deterioration based on a renewal-geometric process[J]. Annals of Operations Research,2016:244(1):49-66.
[5] R TIBSHIRANI.Regression shrinkage and selection via the lasso: a retrospective[J].Journal of the Royal Statistical Society,2011,73(3):273–282.
[6] H ZOU .The adaptive LASSO ad its oracle properties[J].Journal of the American Statistical Association,2006,101(12):1418-1429.
[7] Z HARCHAOUI,C LEVY-LEDUC .Catching change points with Lasso[C].Advances in Neural Information Processing System,Vancouver:MIT Press,2007,22: 617-624.
[8] Z HARCHAOUI,C LEVY-LEDUC.Multiple change-point estimation with a total variation penalty[J].Journal of the American Statistical Association,2010,105(492): 1480-1493.
[9] G CIUPERCA .Model selection by Lasso methods in a change-point model [J].Statistical Papers,2014,52(2):349-374.