基於演化式演算法與叢集運算之動態資料驅動預測模型

A dynamic data driven prediction model based on evolutionary algorithms and cluster computing

胡維哲(Wei-Che Hu)

指導教授 ： 林斯寅

中原大學/商學院/資訊管理研究所/碩士(2016年)

https://doi.org/10.6840/cycu201600926

摘要

近年來巨量資料已經成為了重要的研究課題，主要原因如以下特性：快速的即時資料變動、複雜的分散式資料來源、異質性資料的整合、以及資料量的快速成長，然而，在此複雜的巨量資料環境中，要達到有效率且準確的預測，是一個新的挑戰。動態資料驅動應用系統(Dynamic Data-Driven Application System, DDDAS)的概念是一種解決的方法，可提供模擬分析和預測的能力，並進一步擴展相關的應用模型。在DDDAS的概念中，找出資料的關聯性是重要的一環，即時的找出資料間資料驅動的相關性，是加強DDDAS框架效率的方法。在過去演化式演算法已被廣泛應用在規則啟發，已證明可以有效的解決在實際應用中最佳化的問題，但隨著資訊的快速發展和巨量資料的出現，這些問題的規模和複雜性持續擴大，傳統的演化式演算法，無法在合理的時間內給予答案。叢集運算是一種合作的平行運算架構，透過網路的整合，結合平行運算、高性能、分散式以及高可用性的能力，此運算框架可以用來解決在動態資料驅動概念下，演化式演算法在動態運算與資源分配上的需求。本研究將提出一個基於演化式演算法與叢集運算的動態資料驅動預測模型，在此模型中，將加入動態資料驅動應用系統的概念，基於叢集運算的架構，以分散式的演化式演算法來建置模型，在一個資料會隨時間變動的動態環境下，即時的找出預測目標與動態資料來源之間的資料關聯，並做出有效率且準確的預測。

關鍵字

叢集運算；動態資料驅動應用系統；演化式演算法；機器學習

並列摘要

In recent years, big data has become an important research topic. Such as the main reason for the following characteristics: high velocity in real-time data, distributed of complex data sources, integration of heterogeneous data and growth of data volumes. Therefore, in this complex information environment. It is a new challenge to achieve an efficient and accurate prediction. The concept of dynamic data-driven applications system (DDDAS) is a solution, to provide simulation and prediction capabilities, and expansion of the relevant application model. In the DDDAS framework, to find out the relationship between data instantly can help to improve the efficiency of DDDAS. In the past, the evolutionary algorithms have been widely used, it has been proven to be effective in solving the practical applications of optimization problems. But, with the advent of the rapid development of information and the big data. The size and complexity of these issues continues to expand. The traditional evolutionary algorithms can’t give a satisfactory answer within a reasonable time. Cluster computing is a parallel computing architecture. It combined with parallel computing, high-performance, distributed, and high availability capabilities through the network integration. In a dynamic data-driven concept. This architecture can be used to solve the operation with evolutionary algorithms on dynamic computing and dynamic resource allocation. In this study. We propose a dynamic data driven prediction model based on evolutionary algorithms and cluster computing. In this model, it will be added to the concept of dynamic data-driven application system. In a dynamic data environment, build a distributed evolutionary algorithms base on cluster computing architecture. To find out the relationship between the dynamic data and prediction target in time and make an efficient and accurate predictions.

並列關鍵字

cluster computing；dynamic data-driven applications system；evolutionary algorithms；machine learning

參考文獻

陳稼興, 謝佳倫, & 許芳誠. (2000). 以遺傳演算法為基礎的中文斷詞研究. 資訊管理研究, 2(2), 27–44.

林豐澤. (2005). 演化式計算上篇: 演化式演算法的三種理論模式. 智慧科技與應用統計學報, 3(1), 1–28.

張雅芳. (2012). 利用 MapReduce 實作分散式協同推薦系統. 清華大學資訊工程學系學位論文, 1–66.

Abraham, A. J., Kumar, K. R., Sridharan, R., & Singh, D. (2015). A Genetic Algorithm Approach for Integrated Production and Distribution Problem. Procedia - Social and Behavioral Sciences, 189, 184–192.

Antoniou, C., Koutsopoulos, H. N., & Yannis, G. (2013). Dynamic data-driven local traffic state estimation and prediction. Transportation Research Part C: Emerging Technologies, 34, 89–107.