建構加權演化式模糊類神經網路於PCB生產預測之研究

Construct A Weighted Evolving Fuzzy Neural Network For PCB Sales Forecasting Problem

洪國智(Guo-Chih hung)

指導教授 ： 張百棧

元智大學/工程學院/工業工程與管理學系/碩士(2004年)

摘要

印刷電路板產業為我國每年帶來可觀的經濟貢獻，業者不斷的擴充產能，卻比不上需求的快速變化，造成供給與需求的失衡，普遍存在庫存堆積與部份停工待料的問題。因此，透過一套準確的生產預測模型，可以提供未來的需求趨勢，進而達到後續的物料準備與生產排程等工作，可減少庫存成本與停工的損失。本研究方法共分成四大部份，首先收集總體經濟指標、下游產品需求指標與工業生產指標等共15項因子，利用灰關聯分析篩選出對於PCB需求預測影響性較高的關鍵因子組合；第二部份加入時間序列因子，採用溫氏指數平滑法來考量預測資料的趨勢性與季節性因子的影響性；第三部份以演化式模糊類神經網路為基礎，考量不同的因子權重與採用歐基理德得距離計算模糊規則相似度，以進行模糊規則的分群與規則萃取的訓練階段，並進行加權演化式模糊類神經網路的回想過程；最後，藉由平均絕對百分比誤差(MAPE)、平均絕對偏差(MAD)、誤差均方根差(RMSE)與總成本差(TCD)四項衡量指標，與演化式模糊類神經網路、傳統倒傳遞網路、迴歸分析與基因神經網路進行評比。經實證結果可發現，加權演化式模糊類神經網路皆較其它預測模型效果佳，故本研究所提出之預測模型可作為業者進行需求預測之最佳參考依據。

關鍵字

生產預測 ； 加權演化式模糊類神經網路 ； 灰關聯分析 ； 溫氏指數平滑法 ； 印刷電路板

並列摘要

Printed Circuit Board (PCB) brings a bright and considerable economic contribution to our country every year. Businessmen keep expanding the capacity; however, they cannot keep up with the fast change of demands. Consequently, it causes the imbalance of the supply and demand and it is common to have the problems of severe inventory stacking and material-lacking. Therefore, through an set of accurate production-forecasting model, it can provide a trend of demands in the near future to reach the following material-preparation and production-schedule, etc. It can decrease the stock’s cost and the loss of pending works. The method of this research is divided into four parts in total. First of all, collect overall Economic Index, Lower Product-Demands Index, Industrial Product Indexes and Comparison, and so on. There are fifteen factors in total. Then, use Grey Relation to select out the higher effective key factors’ combination for PCB demand-prediction. In the second part, add time series factor and then use Winter’s Exponential Smoothing to consider the trend of predicted information and the effectiveness of seasonal factors. In the third part, it will proceed the training of WEFuNN, and also work the recall process of the WEFuNN. Finally, via the four measure indexes: MAPE, MAD, RMSE and TCD to compare with the EFuNN, Back-propagation, Regression Analysis,. Through the evidence and result, it shows the WEFuNN gets better effect. As a result, the brought-out predicted model in this research can give the businessmen the best reference to proceed the demand-prediction.

並列關鍵字

Forecast ； Weighted Evolving Fuzzy Neural Network ； Grey Relation Analysis ； Winter’s Exponential Smoothing

參考文獻

49.楊東翰，「整合基因演算法及類神經網路於印刷電路板生產預測之研究」，元智大學工業工程與管理研究所，碩士論文，2004。

43.周湘蘭，類神經網路在多重產品需求預測上之應用，元智大學工業工程管理研究所，碩士論文，2001。

2.Abdel-Aal, R.E. Z. Al-Garni, “Forecasting monthly electric energy consumption in eastern Saudi Arabia using univariate time series analysis”, Energy, Vol.22,pp. 1059-1069 , 1997.

3.Abraham, A. and N. Baikunth, “A neuro-fuzzy approach for modeling electricity demand in Victoria”, Applied Soft Computing , pp. 127-138. ,2001.

4.Abraham, A., N. Baikunth, and P.K. Mahanti, “Hybrid Intelligent Systems for Stock Market Analysis”, Lecture Notes in Computer Science, vol.2074,p.p.337-345,2001.