應用機器學習於投影機噪音預估之研究

The Development of Noise Predictive Models for Projector with Machine Learning

黃誌浩(Chih-Hao Huang)

指導教授 ： 項衛中

中原大學/電機資訊學院/工業與系統工程研究所/碩士(2010年)

https://doi.org/10.6840/cycu201000816

摘要

投影機運作時噪音的大小是很重要的品質指標。而投影機噪音形成的因素有許多，可以採用不同的方法來分析和探討，本技術報告希望由量測所得到的工程數據，來建立投影機噪音的預估模式。以便將來投影機設計時可用量測到的工程數據資料，預估噪音是否符合標準。

本研究使用的方法是運用機器學習的決策樹與類神經網路技術來建構預估模式。將工程數據分成60筆以訓練而得適當的參數，再以40筆資料測試預估模式的準確性。本報告用WEKA軟體來建構決策樹，也利用Qnet 2000軟體來建構類神經網路，得到最佳的決策樹是J48，最佳的類神經網路是三層式，節點各為15, 8及1。最後結果發現決策樹的正確率為79%，而類神經網路正確率為82.5%。

從結果顯示，使用類神經網路所得到的正確率是所有預估模式中最佳的，因此建議以類神經網路做為未來預估噪音的模式，對於個案公司投影機後續研發降低噪音可以提供顯著的改善。

關鍵字

決策樹 ； 類神經網路 ； 投影機

並列摘要

The noise level of projector operations is an important quality indicator. The operation noise may be generated by many factors and they can be analyzed and discussed with different methods. The purpose of this report is to build a practical model to estimate noise levels by using engineering data collected in experiments. Thus, engineering data can be used to forecast whether the noise level will meet the noise standard in the design phase.

The Decision Tree and Neural Network techniques, two popular methods in the field of Machine Learning, were adopted for this study to construct predictive models. The engineering data were separated into two parts, one with 60 records to obtain appropriate parameters of the models, and the other with 40 records to test the accuracy of the models. WEKA software was adopted to construct the Decision Tree while Qnet 2000 was utilized to build the Neural Network, it was found that J48 was the best Decision Tree, and the best Neural Network is a three layer one. The node numbers in three layers were 15, 8, and 1 individually. The result indicated that the accuracy for the Decision Tree was 79%, while the accuracy level for the Neural Network was 82.5%.

The results from both techniques were acceptable, but among all predictive models, the Neural Network technique had the highest level of accuracy. Thus, the Neural Network technique is recommended for forecasting the noise level of projectors in future designs and this kind of model will improve noise level of future products.

並列關鍵字

Neural Network ； Decision Tree ； Projector.

參考文獻

[19]郭一聰，「應用決策樹與類神經網路於應收帳款之呆帳預警模式研究」，中原大學資訊管

[18]劉近興，「以類神經網路研究半導體封裝廠銲線機台選擇問題」，中原大學工業工程學系碩士論文，2005。

[10]王美慧，運用類神經網路於證卷業網路下單服務品質之研究，顧客滿意期 刊，民國九十四年。

[21]陳旗明，「應用資料探勘技術分析車輛燃料之消費行為」，國立成功大學工業與資訊管

[22]葉榕真，「在資料對映中的應用機器學習得初探」，中原大學工業工程學系碩士論文，