摘 要:提出通過人工神經網絡擬合極限狀態函數的方法來解決結構可靠性問題。根據多層神經網絡映射存在定理,對于任何在閉區間內的一個連續函數都可以用含有一個隱含層的BP網絡來逼近。應用此定理,通過人工神經網絡擬合極限狀態方程,借助神經網絡的函數映射關系產生大量的極限狀態函數值,作為下一步的分析數據。此過程并不像Monte Carlo法對每一點都做確定性計算,因而達到減少計算工作量的目的。該方法僅采用Monte Carlo法隨機抽樣的思路,對大范圍的數據進行概率分析,通過概率分析得到極限狀態函數值的均值和標準差,以便求得結構系統的可靠性指標,進行結構系統可靠性分析。
關鍵詞:BP神經網絡; Monte Carlo法; 結構可靠性; 極限狀態函數
中圖分類號:TB114.3;TP183 文獻標識碼:A
文章編號:1004-373X(2010)12-0059-03
Structure Reliability Analysis Based on BP Neural Network Monte Carlo Method
ZHANG Liang1, ZHAO Na2
(1. Propaganda Department, China University of Petroleum, Dongying 257061, China;
2. Safety Supervision Station, Jiaonan Municipal and Rural Construction Bureau, Jiaonan 266400, China)
Abstract:The method of fitting the limit state functions through an artificial neural network is put forward to solve the problem of structure reliability. According to the existence theorem of multilayer neural network mapping, any continious function in the closed interval can be approached with BP network containing a hidden layer.Many limit state function values, which are acted as the analysis data in the next step, are generated with the theorem and the fitting of the limit state equations by the aid of the function mapping relationship of neural network. The probability analyses for a wide range of data are performed only by the thought of random sampling with Monte Carlo method. carry on to the data of the large range, receive the mean value and standard deviation of the limit state function values are obtained by the probability analyses to derive the reliability index of the structure system to carry on the reliability analysis of the structure system.
Keywords:BP neural network; Monte Carlo method; structure reliability;limit state function
0 引 言
在進行結構可靠性分析中,常遇到某些結構的極限狀態函數不能明確給出,或者有些更為復雜結構的極限狀態函數根本不能寫出來,此時,直接應用大多數可靠性方法都會遇到困難。因此,可以利用數值模擬或實驗等手段得到結構的多組輸入及響應,以此作為神經網絡的訓練數據對神經網絡進行訓練。經適當訓練的神經網絡能夠較好地逼近結構的極限狀態函數,在此基礎上可以十分方便地利用Monte Carlo法模擬結構的可靠度。
1 基本原理
1.1 Monte Carlo法
Monte Carlo法[1]是通過隨機模擬和統計實驗來求解結構可靠度的近似數值方法。
根據大數定律,設X1,X2,…,Xn是n個獨立的隨機變量……