Data Science and Reinforcement Learning for Price Prediction and Procurement Decision of Petrochemical Raw Material

• Main Authors: Bai-JianChou, 周百建
• Other Authors: Chia-Yen Lee
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/67ab45

碩士 === 國立成功大學 === 製造資訊與系統研究所 === 106 === Petrochemical industry is one of the major industries in the world-wide economy. In general, ethylene, propylene and butadiene, which are associated with almost synthetic chemicals, are the main raw materials of this industry with around 70-80% cost structure. In particular, butadiene is one of the key materials for producing synthetic rubber and used for several daily commodities. However, the price of butadiene fluctuates along with the demand-supply mismatch or by the international economic fluctuations and political events. Therefore, a precise forecast of raw materials price would effectively support the company in reducing procurement costs and improving operation performances. This study proposes data science framework to predict the weekly price of butadiene by using the historical price of the butadiene industry supply chain, contract price, capacity supply rate, capacity demand rate, and downstream competitor information. Thus, this study includes two modules. One is the price prediction model with time series decomposition method, support vector regression and deep learning technique to predict weekly butadiene prices. The other module applies the analytic hierarchy process, Markov decision process and reinforcement learning technique to make a best procurement decision. An empirical study was conducted to validate the prediction and decision models, and the results show that the proposed model supports the company in raw materials procurement and provide some insights for practical decision-making process.