應用類神經網路預測辦公用品物流中心之隨機出貨量

Applying Artificial Neural Networks to Forecast the Radom Delivery Quantity of Office Supplies Distribution Center

陳進福(Chin-Fu Chen)

指導教授 ： 車振華

國立臺北科技大學/管理學院/工業工程與管理系EMBA班/碩士(2008年)

https://doi.org/10.6841/NTUT.2008.00057

摘要

由於近年來流通產業的快速成長，使得具有連結上游製造商與下游零售商功能的物流中心在商業化的過程中扮演著整合者重要的角色。
物流中心為達到大量進貨統一分配的物流機能，以滿足多樣少量、多頻率配送的現代化消費需求特性，必須強化物流中心的管理與作業功能。因此精確有效的出貨預測是物流中心內部不可或缺的重要功能之一。
會影響物流中心出貨量的因素既多且複雜，除了一般性的時間序列可量化因素外，尚有夾雜著一些無法量化之因素，且不易取得，這些因素交錯著影響著物流中心的出貨量。
本研究主要提出以辦公用品物流中心的隨機出貨資料，應用類神經網路建構預測出貨產出量模式之程序，經由此程序可根據其出貨產出量的預測變數，包含量化與非量化數值，根據預測變數的型態進行資料前處理，並應用皮爾森相關係數分析進行預測變數篩選，以利類神經網路導入使用。
利用類神經網路的學習、回想與推論特性，經由比較各種參數組合訓練範例與測試範例的所得之預測出貨產出量，找出最佳的參數組合。最後以預測值比對個案公司物流中心出貨量資料驗證其方法的有效性，並協助個案公司發現其問題的所在，提出改善方向與建議。

關鍵字

類神經網路 ； 倒傳遞類神經網路 ； 出貨預測 ； 物流中心

並列摘要

Due to the rapid growth of distribution industry,distribution center which connects manufacturers and retailers plays more and more important role in the business process.
To meet distribution demands on mass stock and centralized distribution, and fulfills consumer’s requirements of few diverse and highly frequent dispatch, it is necessary to reinforce the management and operation of distribution center. Thus to conform to precisely and effective distribution forcast which being one of the essential functions of modern distribution center.
Factors which affect distribution quantity of distribution center are various and complicated. Except for general quantifiable time series factors, there still exists lots of unquantifiable factors which are not easy to access and surely will lead to interact its performance on distribution quantity.
This research randomly derives distribution data from an office supplies distribution center, and applys Artificial Neural Networks technics to construct a model to forcast the distribution quantity. By its processes of this model, such data could be pre-treated according to the distribution forcast variables including quantifiable and unquantifiable, and types of the forcast variables.
Furthermore the Pearson Correlation Analysis also be applied to filter those forcast variables, thus to benefit the introduction and application of Artificial Neural Networks.
By means of the characteristics of Artificial Neural Networks—learning, calling and generalization, and comparison of different forecasting distribution quantities produced from testing of different parameter-mix, the optimum parameter-mix so could be brought out. Eventually which could be compared with the actual statistics of the researched objective to test and verify its effectiveness, so as to find out possible problems and propose corresponding solutions.

並列關鍵字

Artificial Neural Network ； Back Propagation neural network ； Delivery Quantity Forecast ； Distribution Center

參考文獻

【19】楊清潭，應用類神經網路於健康檢查顧客忠誠度之研究，銘傳

配銷管理模式，中國工業工程學刊，1998年。

【11】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

【12】林照挺，以特徵式模型建構塑膠射出成形產品成本預估及產品

【14】康永成，印刷電路板製造資料模型建構與成本分析，元智大學