- 學位論文
DEA共同權重模式之研究—以IC設計產業為例
Common Weight Models In DEA For IC Design Industry
摘要
資料包絡分析法(Data Envelopment Analysis, DEA)為一效率評估工具，可處理多投入產出指標的效率評估問題，且客觀的評估各決策單位元(Decision Making Unit, DMU)的相對效率分數。然而，傳統DEA有幾個評估上的問題：(1)各個DMU採取對於自己最有利的權重，造成有效率DMU過多，降低DEA的鑑別能力。(2)各DMU採用不同權重，無法使各DMU在同一個基準上進行比較。 針對上述兩個問題，本研究利用妥協解的概念，以原始DEA模式為架構作為共同權重模式的計算基準，透過廣義距離衡量方式產生一組共同權重，如此，能使各個DMU在相同基準上做效率評估的衡量，並改善DEA鑑別度問題，降低有效率DMU個數。此外，共同權重模式為非線性規劃問題，在求解上無法求得全域最佳解，為了不使區域最佳解影響評估結果，本研究運用凸化策略與凸下界將模式轉換成凸規劃(convex programming)問題，並透過分支界定法，求得共同權重的全域最佳解，增加共同權重模式的應用價值。 最後，本研究以一個IC設計產業的例子，來說明並比較DEA -CCR模式與共同權重模式所得出的評估結果。
並列摘要
Data Envelopment Analysis is a tool of performance assessment measuring multiple inputs and outputs indices which can measure the relative efficiency score among all decision-making units objectively. However, conventional DEA models have some problems in evaluation, such as (1) DMUs obtain the most advantageous weights for themselves resulting in too many efficient DMUs and decreasing the discrimination ability. (2)Each DMU has different weights resulting in unable to compare DMUs on a common base. To deal with the problems, the study uses the concept of compromise solution and generalized distance measure for producing a set of common weights under the conventional DEA CCR model. Therefore, all DMUs can compare with each other base on a set of common weights. Besides, the presented common weight models can improve the discrimination of DMUs and decrease the number of efficient DMUs. In addition, the common weight models are nonlinear programming problems which cannot guarantee to get the globally optimal solution. To deal with the difficulty, this study utilizes the convexification strategies and convex underestimation method to convert the nonconvex model into a convex problem. Then the global solution of the transformed model can be found. An example of IC design industry is presented to illustrate the common weight models and the usefulness of optimization techniques. Comparison results between the DEA CCR model and the proposed common weight models are also listed and analyzed.