請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/22093
標題: | 應用深度學習於需求與供應不確定之調度問題 Application of deep neural network to the supply and demand matching problem |
作者: | Yu-Chun Pan 潘昱均 |
指導教授: | 周雍強(Yon-Chun Chou) |
關鍵字: | 不確定供應與需求,供需調度,機器學習,深度神經網路, uncertain demand and uncertain supply,supply and demand matching,machine learning,deep neural network, |
出版年 : | 2018 |
學位: | 碩士 |
摘要: | 供應鏈與物流當中存在許多供應不穩或需求擾動的不確定因素。而供應未能配合需求的情況下就會產生利用率以及經濟的損失,像是原料短缺或是生產不穩造成無法完致使發生違約賠償,甚至商譽及市占率的損失,另外需求的擾動也會造成存貨堆積產生跌價損失。
本文將對於供需兩端同時進行建模處理,供需模型的概念是來自於共享單車的存貨調度。因為人潮的需求隨機性會使得存貨產生分布不均勻的現象,藉由站點的存量水準決定出供應需求兩種不同性質的類別,再將存量較多的供應性質站點調度一些存貨到存貨量較少的需求站點。 且利用深度學習來代替原本一連串的處理與計算過程,簡化中間的流程與增加模型的彈性,在面對不同的情境不同的調度情況出現時不必再經由一連串的步驟來得到最佳解。只需透過建立好的深度學習模型就可以得到預測的解,並對深度學習在此問題時的建模提出建模流程以及對不同資料格式與變數檢定得出一個完整的深度學習模型。最後提出一個衡量模型的指標讓使用者可以決定是否採用該模型,再藉由此指標對前面建立的深度學習模型的進行檢測表現,驗證該模型是可以接受來做為替代資料前處理與數學規劃求解的方法。 There are many uncertain supply and uncertain demand supply issues in supply chain management and logistics process, such as supply disturbance and demand instability; manufacturing process, material delivery, production quality and market order stochastic. Abundant of previous papers discuss about one side uncertainty, and the research work on both side uncertainties is not a lot. A bike-sharing concept of supply and demand, which is determine surplus side or deficit side by the stock level of the site. Surplus sites with more stocks supply some stocks to deficit station. The basic data means the setting of surplus deficit site using to generate sample probability, and then the supply and demand functions required by matching model are obtained through conversion. This matching uncertainty supply demand model can be separate into two stage model: discontinuous dispatching model and continuous dispatching model. Simulation method can generate many cases, and these cases can be used to explore the feature of random supply and demand cases. Parameter settings and model output discussion would give out reason to choose variable to use in machine learning model. Machine learning model can fit the function of the data pre-processing and scheduling logic behind the data. There are three issues about data format, variable in training data and data volume are researched. After the best model under this research been found, this model could be applied to obtain relocation information without data pre-processing and optimization calculating work. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/22093 |
DOI: | 10.6342/NTU201802535 |
全文授權: | 未授權 |
顯示於系所單位: | 工業工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-107-1.pdf 目前未授權公開取用 | 3.02 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。