因應相關環保法規(如歐盟WEEE)、綠色供應鏈風潮與企業永續發展，電子資訊廢棄物(electronic waste, e-waste)之回收需求分析與預測議題漸成為環保與廢棄物管理重要之研究方向。本研究進行電子資訊廢棄物回收需求分析與回收量預測；在回收需求分析方面，針對電子資訊廢棄物回收需求進行調查與分析，以了解電子資訊物品從生命終期廢棄物產生至實際形成廢棄物回收過程之關鍵影響因素。在回收量預測方面，本研究先藉由使用年限機率與回收機率之推估，透過電子資訊產品出貨量歷史資料推測未來電子資訊廢棄物之潛在回收量，再進一步整合應用類神經網路模式，藉以降低預測不確定性與隨機性因素提升預測精度，而建構電子資訊廢棄物回收量預測模式。最後，針對台灣地區電子資訊廢棄物之稽核認證回收量統計資料進行預測分析，預測結果顯示本研究所建構之電子資訊廢棄物回收量預測模式之預測能力均較機率推估型預測模式之使用年限法、及時間數列型預測模式之ARIMA與GM(1,1)模式、以及整合型模式之二元迴歸及GM(1,2)模式為佳，驗證本研究模式可行且具有較佳之預測與解釋能力。本研究成果不僅在學術上可作為廢棄物回收需求分析與回收量預測模式相關研究之參考，所發展之模式亦可提供實務上進行逆物流回收預測模組開發之模式基礎。 With the global eco-awareness, the European Union has claimed several regulations, such as the Directive on Waste Electrical and Electronic Equipment (WEEE) to regulate recycling items for end-of-life (EOL) electrical and electronic equipment. Demand analysis and forecasting for electronic waste (e-waste) recycling is a critical foundation in the field of environment and waste management. This study develops a series of models to analyze the demand factors, and to predict the return quantity for e-waste. The first part of this study conducted a demand survey and analysis of e-waste recycling. This study applies exploratory factor analysis to identify key demand for EOL electrical and electronic equipment recycling. This study proposes a binary logistic regression model to determine the return probability. In the second part of the study, this study combines probability estimation and time-series forecasting model to propose a hybrid forecasting model for e-waste return quantity forecasting. Considering useful life of electrical and electronic equipment, the production shipment volume, and the return probabilities, the potential return quantity is estimated. Furthermore, a neural network model is developed to improve the forecasting accuracy and eliminate the uncertainty and randomness surrounding the input data. Finally, an example with e-waste data in Taiwan was provided. The proposed model yields more accurate prediction results than ARIMA, GM(1,1), binary regression and GM(1,2) models. The numerical results verified that the proposed model is practicable, and provide a better prediction and explanation ability.
