結合時間序列分析與類神經網路建構可靠度預測模式

Abstract

精確的可靠度預測模式是可靠度管理者作決策時的重要依據，並可依 此模式減少成本。本研究的目的是結合ARIMA分析與倒傳遞類神經網路兩 種預測技術，提出新的可靠度預測模式之構建方法，稱之為〝結合輸入倒 傳遞類神經網路〞。此技術的作法是利用ARIMA分析所得的結果，當作倒 傳遞類神經網路訓練與測試時的輸入變數。這種結合的作法不僅考慮到預 估誤差的資訊，而且先以ARIMA進行分析對問題會有一初步的概括瞭解。 本研究以ARIMA的分析結果當作BPNN的輸入變數，不僅可以提昇可靠度預 測模式的預測能力，也可增快網路收斂的速度。 A precise reliability forecasting model offers manager an important referencewhen making decisions and reducing warranty cost. The purpose of this study isto integrate the two forecasting techniques , Autoregressive Integrated andMoving Average Analysis (ARIMA) and Back Propagation Neural Network (BPNN) , andpropose a new modeling technique for building a reliability forecasting modelwhich is called Input-Combined Back Propagation Neural Network (ICBPNN). The technique uses the result of ARIMA as an input of variables for trainingand testing BPNN . It can not only reduces the estimated error but also givesusers a basic concept of the problem by pre-analyzing with ARIMA. Moreover ,using the analysis result of ARIMA as the input variables for BPNN canefficiently improve the forecasting ability of reliability forecasting modeland speed the convergence of network. Two cases studies are presented , demonstrating the proposed technique ,ICBPNN , has better consistency and forecasting ability than that of ARIMAor BPNN .