吳姵萱

Wu, Pei-Hsuan

碩士論文(2022)

具持續改善與知識管理機制之異常管理模式與技術研究

Research on an Abnormal Management Model and Technologies with Continuous Improvement and Knowledge Management Mechanisms

關鍵字 Keywords

機器學習、異常處理、知識管理、持續改善

Machine Learning、Exception Handling、Knowledge Management、Continuous Improvement

摘要

傳統產業在異常發生時，多依靠人工方式進行處理和分析，且未能留下處理之經驗以及不具改善機制，以致常發生問題無法解決或處理時間過久之情事，更難以藉由異常事件進行改善。

隨著物聯網的普及機器學習技術的發展，資料蒐集與分析技術隨之進步，也利於異常處理與管理之改善。本研究針對異常處理之經驗擷取與持續改善之需求，提出一「具持續改善與知識管理機制之異常管理方法」與開發實現技術，並依此建立一「持續改善平台」，供產業參考應用，提升異常管理之效益。

針對研究目標，本研究之研究項目包括:具持續改善與知識管理機制之異常管理模式設計、異常處理程序設計、異常處理相關方法設計及知識管理模型設計，並針對異常管理模式中之重點功能，重要影響因素分析、經驗挖掘與經驗分享等技術以及持續改善機制進行開發。

本研究使用台中某一糕餅業者的異常處理案例與數據集進行實驗，驗證所提之方法與技術的可行性與有效性。經驗分享機制評量之準確率為96.73%、召回率為93.88%，若利用持續改善方式，使用固定長度之數據，定期更新重要影響因素，則準確率可達97.25%、召回率為95.33%，驗證持續改善機制之方法可行且有效。

Abstract

When anomalies occur, traditional industries mostly rely on manual methods for processing and analysis, and there is no experience in handling and no improvement mechanism, so that problems often occur that cannot be solved or take too long to handle, and cannot be improved by abnormal events. .

With the popularization of the Internet of Things and the development of machine learning technology, data collection and analysis technology has also improved, which is also conducive to the improvement of exception handling and management. Aiming at the needs of experience acquisition and continuous improvement in exception handling, this research proposes an "abnormal management method with continuous improvement and knowledge management mechanism", and establishes a "continuous improvement platform" based on this for industry reference and application to improve exception management. benefit.

For the research objectives, the research projects of this research include: Exception management model design with continuous improvement and knowledge management mechanism, exception handling program design, exception handling related method design and knowledge management model design. And for the key functions in the exception management mode, analysis of important influencing factors, knowledge mining of exception rules, exception handling technology and continuous improvement mechanism development.

This study uses an exception handling case and dataset of a pastry business in Taichung to conduct experiments to verify the feasibility and effectiveness of the proposed methods and techniques. In the prediction of the exception handling model, the model prediction accuracy rate is 96.73% and the recall rate is 93.88%. However, if the concept of continuous improvement is used, fixed-length data is used, and important influencing factors are regularly updated, the model prediction accuracy rate can reach 97.25% , the recall rate is 95.33%, the experimental results of the method with continuous improvement mechanism are better than the method of directly using incremental data, so the method with continuous improvement mechanism is feasible and effective.