在現今產業全球化、產品多樣化的市場經營體系中,企業間的競爭使得訂單需求預測的準確性顯得更不容忽視,預測的準確與否往往會影響生產成本。當不同組件產品具有關連性時,個別的產品需求亦會交互影響到共同零組件的需求,再加上產品短期需求不確定的波動性更會降低需求預測的準確性降低。因此,依據多重產品間彼此的關連性設計一個真正適合此生產線的需求預測模式為一重要的研究課題。 本論文將利用類神經網路在預測上高準確之優點,來學習多重產品之特性進行需求預測模式之建構,以供決策者在進行生產計畫時之依據。論文中,利用AweSim模擬軟體模擬六個月的短期訂單模擬,收集訂單種類、到達時間、數量等資料並以BPN類神經網路進行需求預測模型之建立。模式中利用移動視窗法進行網路學習,設定的輸入及輸出變數主要分為到達之間隔時間、產品數量、訂單種類權重。然後再以平均方根誤差(MSE)作為評估指標。分析結果顯示利用BPN類神經網路所建立之需求預測模式在訂單分配率改變時皆有不錯的預測效果,其測試樣本之平均MSE值皆在0.15以下,本論文針對訂單種類預測所提出之訂單特徵向量也能有效降低預測誤差,可提供在未來進行多重產品需求預測之相關研究另一參考依據。
At present, market management system is challenged by industry globally and variety products. An accurate prediction of order demand will increase the competitive ability for an enterprise. The closeness of the forecasted amount to the real demand will influence the production cost. The individual product demand having interaction effect on the demand of manufacture components when the several modular products are produced, meanwhile, fluctuation of the short-term products demand will decrease the precision of demand forecasting. Therefore, a multi-products short-term demand-forecasting model is important and necessary. In this thesis, we take the merit of the neural network in high precision of the forecasting to the multi-products system and establish a demand-forecasting model. In this paper, we use the simulation software AweSim to simulate the orders data in production line for six months. Data for order types, arrival times, and numbers of product are collected and trained by the Back-Propagation neural network (BPN) to build a BPN multi-product demand-forecasting model. Simulation study shows that the BPN forecasting model has performance on the different order distributions, and the average mean square errors (MSEs) of the test samples are all under 0.15. In addition, this research brings up an order features vector also effective in reducing the forecasting error, which can take a reference for the future research.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。