請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/21360
標題: | 利用人工智慧之消費者偏好學習與動態訂價方法 An Artificial Intelligence Approach for Consumer Preference Learning and Dynamic Pricing |
作者: | Wan-Ling Chu 朱琬琳 |
指導教授: | 吳政鴻 |
關鍵字: | 需求學習,動態規劃,深度學習,長短期記憶神經網路,貝氏學習, Demand learning,Dynamic programming,Deep learning,Long short-term memory,Bayesian learning, |
出版年 : | 2019 |
學位: | 碩士 |
摘要: | 本研究嘗試結合動態規劃與深度學習技術應用在動態定價與需求學習中,開發出能夠在規劃時程前可以做出良好的初始決策,並且隨著環境回饋進行即時學習與參數優化之模型。
需求學習可以幫助業者理解消費者喜好來達到滿足市場需求之目的,但受限於資料不足及環境不確定性,無法達到最佳的效果。動態規劃雖然以被證實為最佳解的模型,卻因維度詛咒、模型假設與專一性的限制,無法普及於實際應用中。我們提出的方法將解決上述提及之困難。 本研究中,使用販賣一種時效性商品定價模型,以最大化利潤為目標。使用少量歷史銷售資料作為訓練樣本,建構長短期記憶神經網路,用於識別未知的市場環境。再使用動態規劃求解出所有狀態集合下最佳的定價策略資料,訓練出深層神經網路,可重複求解各種模型下的最佳決策。最後結合貝氏學習,隨著環境回饋進行即時參數優化,達到最大化預期利潤之目的。並透過離散事件模擬的方式,驗證在各式各樣的市場環境下,我們的模型與最佳利潤的差異,結果證實,本研究提供的模型可以在未知的環境下也能夠擁有良好的效能。 This study attempts to combine dynamic programming with deep learning method in dynamic pricing and demand learning to develop a model that can make good initial decisions before planning horizon begins and conduct online learning and decision optimization. Demand learning can help the business understand consumer preferences and meet demand, but it is limited by insufficient data and environmental uncertainty to achieve the best results. Dynamic programming, although proven to be the best solution, is not universally applicable due to dimensional curses, model assumptions, and specificity limitations. The method we propose will solve the difficulties mentioned above. In this research, we consider the problem of a dynamic pricing problem for a perishable product with a multiple period lifetime. Use a small amount of historical sales data, we construct a Long short-term memory neural network to identify unknown environments. Then we train a deep neuron network with the optimal pricing strategy to make the pricing policy. Finally, combined with Bayesian learning to improve the ability to adapt to uncertainty. We use discrete simulation to verify the cost different between our model and optimal policies from dynamic programming in a wide variety of market environment. The result shows that our model performs well in unknown environment. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/21360 |
DOI: | 10.6342/NTU201903148 |
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-108-1.pdf 目前未授權公開取用 | 6.77 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。