定義
![極值分布](/img/7/efc/wZwpmLxYTM0YDN4cTOzkDN1UTM1QDN5MjM5ADMwAjMwUzL3kzL2czLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
設 為從總體F抽出的獨立同分布樣本,且
![極值分布](/img/0/47b/wZwpmLxMTOzEDM4ATN2YjN1UTM1QDN5MjM5ADMwAjMwUzLwUzL1YzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
![極值分布](/img/4/a01/wZwpmL4YzMwUzNwAjN2YjN1UTM1QDN5MjM5ADMwAjMwUzLwYzL2AzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/f/88b/wZwpmL0MzN4YDM2gTN2UzM1UTM1QDN5MjM5ADMwAjMwUzL4UzL0QzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/3/b6d/wZwpmLyMjM0AzNzgzN2YjN1UTM1QDN5MjM5ADMwAjMwUzL4czL3UzLt92YucmbvRWdo5Cd0FmLyE2LvoDc0RHa.jpg)
如果存在常數 及 ,使 依分布收斂於G(x),則稱G(x)為一極大值分布;類似地定義極小值分布。它們統稱為 極值分布,而分布F稱為“底分布”。
![極值分布](/img/e/8dc/wZwpmL2ETO1cTNwAzNwMzM1UTM1QDN5MjM5ADMwAjMwUzLwczLzEzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/f/871/wZwpmLygDN2IDNwMTOwMzM1UTM1QDN5MjM5ADMwAjMwUzLzkzL4YzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/f/e1d/wZwpmL3QTM1EzM3EzN2YjN1UTM1QDN5MjM5ADMwAjMwUzLxczLwAzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/6/77e/wZwpmLzUDN3kDNwgzN2YjN1UTM1QDN5MjM5ADMwAjMwUzL4czL0MzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
兩個分布函式 和 稱為是同類的,若存在常數a>0及b,使 ,並記為 。
顯然,這種關係具有自反、對稱和傳遞性。
極值分布的三大類型(Fisher—Tippett Theorem):若G(x)為一連續極值分布,則G必與下列三個分布函式之一同類:
![極值分布](/img/9/ef1/wZwpmLyYzNzcTOyQzN2YjN1UTM1QDN5MjM5ADMwAjMwUzL0czL2EzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
![極值分布](/img/4/52d/wZwpmL0IDO0kTMykjN2YjN1UTM1QDN5MjM5ADMwAjMwUzL5YzLzMzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/2/52e/wZwpmL2ADN0AjM3gTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL4UzLwAzLt92YucmbvRWdo5Cd0FmLyE2LvoDc0RHa.jpg)
![極值分布](/img/e/5a9/wZwpmL4UTO1ETN5cTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL3UzL3AzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
分別稱為第Ⅰ、Ⅱ、Ⅲ型極值分布,也分別稱為Gumbel、Fr6cht、Weibull型極值分布。
一般的Gumbel型極值分布為
![極值分布](/img/0/6e9/wZwpmLwgTMwgDO4gTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL4UzL0IzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
相應的生存函式為
![極值分布](/img/f/2a0/wZwpmL3MzMxAjN4ATN2YjN1UTM1QDN5MjM5ADMwAjMwUzLwUzL3AzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
![極值分布](/img/f/6b7/wZwpmL1MTMxgDM5MjN2YjN1UTM1QDN5MjM5ADMwAjMwUzLzYzL4IzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/c/227/wZwpmLwIzM4gzMxATN2YjN1UTM1QDN5MjM5ADMwAjMwUzLwUzL2czLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/9/8cc/wZwpmL0YTOzEDN5gDN2YjN1UTM1QDN5MjM5ADMwAjMwUzL4QzL1czLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
當T服從威布爾分布且有密度函式式一般的Gumbel型極值分布時, 就服從 和眾數為 的一般Gumbel型極值分布。
Gumbel型極值分布
極小值分布
最小極值Ⅰ型分布簡稱極小值分布,其分布密度函式和分布函式分別為
![極值分布](/img/c/4bc/wZwpmL4UDM0IDO5YTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL2UzL2IzLt92YucmbvRWdo5Cd0FmLyE2LvoDc0RHa.jpg)
及
![極值分布](/img/2/855/wZwpmL0ITN4ITN4QjN2YjN1UTM1QDN5MjM5ADMwAjMwUzL0YzL3EzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![圖1](/img/d/c4a/wZwpmL1gzN1MTM2UzN2YjN1UTM1QDN5MjM5ADMwAjMwUzL1czL0YzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/0/588/wZwpmL3gDM3MDM4MDMwEDN0UTMyITNykTO0EDMwAjMwUzLzAzLzMzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/7/9d5/wZwpmL1EDO4kTNwMTMzEDN0UTMyITNykTO0EDMwAjMwUzLzEzLwMzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
式中 ——位置參數,實際上是分布的眾數; ——尺度參數,與分布的離散性有關。
![極值分布](/img/0/588/wZwpmL3gDM3MDM4MDMwEDN0UTMyITNykTO0EDMwAjMwUzLzAzLzMzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/7/9d5/wZwpmL1EDO4kTNwMTMzEDN0UTMyITNykTO0EDMwAjMwUzLzEzLwMzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
![極值分布](/img/d/3aa/wZwpmL3gjMwYzN0cTNxYjN1UTM1QDN5MjM5ADMwAjMwUzL3UzLwAzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
必須注意, 和 不是分布的均值及標準差,但與它們有關,分布密度函式式的圖形見圖1。圖中曲線為 的情況,由圖可知,極小值分布為一偏態分布(右偏)。
1.標準極小值分布,
![極值分布](/img/d/d8d/wZwpmLzQjMzYTOyUDM5YzM1UTM1QDN5MjM5ADMwAjMwUzL1AzL4AzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/6/7f9/wZwpmL0cDM5ETOwATN2YjN1UTM1QDN5MjM5ADMwAjMwUzLwUzLxUzLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
令 ,則 ,代入上述分布密度函式和分布函式式子中得到Z的密度函式及分布函式分別為
![極值分布](/img/3/9af/wZwpmL0EjM5cjNyUTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL1UzL0MzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/2/0d7/wZwpmL4AjN4IDMxkDN2YjN1UTM1QDN5MjM5ADMwAjMwUzL5QzLwYzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/0/588/wZwpmL3gDM3MDM4MDMwEDN0UTMyITNykTO0EDMwAjMwUzLzAzLzMzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/7/9d5/wZwpmL1EDO4kTNwMTMzEDN0UTMyITNykTO0EDMwAjMwUzLzEzLwMzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
上兩式稱為標準極小值分布,並且與分布參數 及 無關。
2.標準極小值分布的期望值及方差,
![極值分布](/img/9/b79/wZwpmLzQDOxcDNyMzN2YjN1UTM1QDN5MjM5ADMwAjMwUzLzczLwczLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
![極值分布](/img/7/c55/wZwpmL2QDN4cTM5YDN2YjN1UTM1QDN5MjM5ADMwAjMwUzL2QzL4UzLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
令 ,代入上式得
![極值分布](/img/1/2da/wZwpmL4ITMxUTO3AzN2YjN1UTM1QDN5MjM5ADMwAjMwUzLwczL2UzLt92YucmbvRWdo5Cd0FmLyE2LvoDc0RHa.jpg)
![極值分布](/img/5/0b3/wZwpmL2gDO3YTN3ADO3EDN0UTMyITNykTO0EDMwAjMwUzLwgzLyQzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
上式積分為一常數,稱作歐拉(Euler)常數。通常記為“ ”即
![極值分布](/img/0/d9d/wZwpmLzATM5MDM4YTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL2UzL0AzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
又
![極值分布](/img/1/9e9/wZwpmLygjMzczM3gTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL4UzLyIzLt92YucmbvRWdo5Cd0FmLyE2LvoDc0RHa.jpg)
所以
![極值分布](/img/1/ee0/wZwpmL1EDO5gjMzITN2YjN1UTM1QDN5MjM5ADMwAjMwUzLyUzL3czLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
3.極小值分布的期望值及方差,
因為
![極值分布](/img/f/48a/wZwpmLwczMyYjMwkTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL5UzLxczLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
所以
![極值分布](/img/5/d11/wZwpmL4UzM1YTMycDN2YjN1UTM1QDN5MjM5ADMwAjMwUzL3QzL4EzLt92YucmbvRWdo5Cd0FmLyE2LvoDc0RHa.jpg)
及
![極值分布](/img/1/d7b/wZwpmLycDN5kDN0MzN2YjN1UTM1QDN5MjM5ADMwAjMwUzLzczLyMzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/4/a02/wZwpmLyAzNwcDN5czM5MTN0UTMyITNykTO0EDMwAjMwUzL3MzL2IzLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
![極值分布](/img/7/72f/wZwpmL0QjM1ITMwITN2IDN0UTMyITNykTO0EDMwAjMwUzLyUzL2MzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
![極值分布](/img/e/6d8/wZwpmLzEDO0kTMyAjN2YjN1UTM1QDN5MjM5ADMwAjMwUzLwYzL4YzLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
![極值分布](/img/0/588/wZwpmL3gDM3MDM4MDMwEDN0UTMyITNykTO0EDMwAjMwUzLzAzLzMzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/7/9d5/wZwpmL1EDO4kTNwMTMzEDN0UTMyITNykTO0EDMwAjMwUzLzEzLwMzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
如果已知樣本的試驗數據,則可以計算總體的均值及標準差的估計值 及s,再由 和 的等式可以得到極小值分布的位置參數 及尺度參數 的估計值:
![極值分布](/img/3/47c/wZwpmL4YzN1ETOwcDN2YjN1UTM1QDN5MjM5ADMwAjMwUzL3QzLzAzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
極大值分布
最大極值Ⅰ型漸近分布密度函式和分布函式分別為
![極值分布](/img/9/448/wZwpmLxQDMwMDO5MjN2YjN1UTM1QDN5MjM5ADMwAjMwUzLzYzL2YzLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
![圖2](/img/5/0f2/wZwpmLwEDN2EDN3czN2YjN1UTM1QDN5MjM5ADMwAjMwUzL3czL4czLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
![極值分布](/img/0/288/wZwpmL2ATO4YTOxcDN2YjN1UTM1QDN5MjM5ADMwAjMwUzL3QzL2YzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/0/588/wZwpmL3gDM3MDM4MDMwEDN0UTMyITNykTO0EDMwAjMwUzLzAzLzMzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/7/9d5/wZwpmL1EDO4kTNwMTMzEDN0UTMyITNykTO0EDMwAjMwUzLzEzLwMzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
式中, ——位置參數; ——尺度參數。
極大值分布密度函式的圖形如圖2所示。
1.標準極大值分布
![極值分布](/img/d/d8d/wZwpmLzQjMzYTOyUDM5YzM1UTM1QDN5MjM5ADMwAjMwUzL1AzL4AzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/6/7f9/wZwpmL0cDM5ETOwATN2YjN1UTM1QDN5MjM5ADMwAjMwUzLwUzLxUzLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
令,則 ,代入最大極值Ⅰ型漸近分布密度函式和分布函式兩式中,得到
![極值分布](/img/b/e77/wZwpmL2MTM1IDM3QjN2YjN1UTM1QDN5MjM5ADMwAjMwUzL0YzL0EzLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
及
![極值分布](/img/9/803/wZwpmL2cjNwgjM4gTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL4UzL4AzLt92YucmbvRWdo5Cd0FmLzE2LvoDc0RHa.jpg)
![極值分布](/img/9/043/wZwpmL4cTN1YDN0IjM3QTN1UTM1QDN5MjM5ADMwAjMwUzLyIzL0AzLt92YucmbvRWdo5Cd0FmLwE2LvoDc0RHa.jpg)
上兩式稱為標準極大值分布密度函式及分布函式,它們與分布參數 無關。
標準極大值分布的期望值及方差分別為
![極值分布](/img/4/9bd/wZwpmL1QDOzMDNzATO2YjN1UTM1QDN5MjM5ADMwAjMwUzLwkzL4YzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/2/576/wZwpmL1EjM4kTN5cTN2YjN1UTM1QDN5MjM5ADMwAjMwUzL3UzLyYzLt92YucmbvRWdo5Cd0FmL0E2LvoDc0RHa.jpg)
![極值分布](/img/4/577/wZwpmLwMTM0YDNzUjN2YjN1UTM1QDN5MjM5ADMwAjMwUzL1YzL2YzLt92YucmbvRWdo5Cd0FmLyE2LvoDc0RHa.jpg)
2.極大值分布的期望值和方差
![極值分布](/img/7/c68/wZwpmL3AzNyIzMxUjN2YjN1UTM1QDN5MjM5ADMwAjMwUzL1YzL3UzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)
![極值分布](/img/1/5ab/wZwpmL0ATMyYTOwITN2YjN1UTM1QDN5MjM5ADMwAjMwUzLyUzLwAzLt92YucmbvRWdo5Cd0FmLxE2LvoDc0RHa.jpg)