- 學位論文
整合案例式推理與倒傳遞類神經網路於新產品單機製造成本之預測~以行動電話為例
New Product Manufacturing Unit Cost Forecasting Using Integrated Case-Based Reasoning and Back Propagation Network approach ― The Case of the Mobile Phone
摘要
全球行動電話市場規模近年來有快速成長的趨勢，並呈現出已開發國家穩定成長而開發中國家倍數成長的現象。高價毛利率款式高和低價量多之銷售需求型態也成為目前各大行動電話代工廠主要爭取訂單及利潤所在，其中如何提出合理的報價以增加接單的機會也為目前業界極力研究的目標。 本研究之研究方法分為兩個階段：第一階段為影響單機製造成本因素的分析，利用研究個案公司之單機製造成本計算公式和專家意見法得到相關之定性和定量因子。第二階段為將上述所得到的因子代入整合案例式推理和類神經網路之預測模型，此階段可分為兩個部份，第一部份為利用定性因子代入案例式推理後預測個別的定量因子，並計算MAPE值以確認其誤差值在合理值內。 第二部份再利用上述所求得的定量因子代入倒傳遞類神經網路預測單機製造成本，先經過訓練後並計算RMSE值後確認該網路之收斂值為合理值後再進行測試並計算MAPE值後確認誤差值在合理值內，經實驗結果證實，整合案例式推理和類神經網路之預測模式預測得到的單機製造成本在合理的誤差值內。 最後將此模型和其他五種案例式推理和類神經網路搭配預測所求得的MAPE值進行比較，可得到本研究所設計之整合案例式推理和類神經網路之單機製造成本預測模式並代入本研究所得到的因子中可以得到較低誤差值，可提供相關人員在行動電話單機製造成本有效的預測參考方式。
並列摘要
The global market scale of mobile phone is tending fast growth in recent years, which demonstrates stable upward growth in developed countries and significant growth in developing countries. High gross profit rate and the demand feature of selling low-price and high-volume products simultaneously have stimulated well-known mobile communication equipment manufacturers to pursue orders and profits. The know-how strategy pending on reasonable prices to win more order opportunities also becomes the desirable focus and study for the present industrial field. In this work, the research method is divided into two steps. The first step is to focus on the analysis of unit production cost by studying specified cases to obtain unit production cost formula and expert advice about the relative nature and quantitative factor. The second step is to take the above factors to integrated Case-Based Reasoning and Back Propagation Network forecast model. This process can be divided into two parts. The first one is to estimate particular quantitative factor by using nature factor to integrated Case-Based Reasoning and calculate MAPE value to confirm its erroneous value in the reasonable value. The second part is to take nature factor to Back Propagation Network forecast model to estimate its unit production cost. After the training, RMSE value has to be calculated to confirm its convergence value in the reasonable value and then calculate MAPE value in order to ensure its erroneous value in the reasonable value. The conclusion of the experiment proves that Case-Based Reasoning and Back Propagation Network forecast model can be used to estimate the erroneous value of unit production cost in the reasonable value. Finally, comparing MAPE values of this model and that of other five types of Case-Based Reasoning and Back Propagation Network, the conclusion indicates that this experimental research design obtains unit production cost in Case-Based Reasoning and Back Propagation Network. Take the research factor to the calculation and the outcome is lower erroneous value, which provides the correlation personnel with estimated references about mobile phone unit production cost.