利用資料探勘及模糊推論技術預測海水溫度與鹽度變化之研究

Predicting Ocean Salinity and Temperature Variations Using Data Mining and Fuzzy Inference

高立仁(Li-Jen Kao)

指導教授 ： 黃有評

大同大學/電機資訊學院/資訊工程學系所/博士(2008年)

摘要

由於全球氣候變遷，專家學者發現海洋在氣候變遷中佔有極重要角色，海水溫度與鹽度變化研究已引起人們廣泛注意與討論。本研究將分析Argo海水溫度與鹽度資料並有效地找出其中溫度與鹽度樣式。傳統關聯法則演算法在尋找關聯法則時，僅能找出交易中項目與項目間的關係，因此若利用傳統關聯法則尋找時空樣式時，其所尋找出樣式無法顯示出時間與空間上關係。例如：若臺灣東北東部近距離地區海水鹽度上升0.15psu到0.25psu，則下個月臺灣東北東部遠距離地區海水溫度將會上升0℃到1.2℃。為了能找出前述的時空樣式，本研究提出一個可將Argo資料轉化成交易型態資料集的方法，接著再設計一個量化跨界性關聯法則探勘模型，能有效地從被轉化後的Argo資料中找出具時空變化關係之溫度與鹽度樣式。另一方面，我們結合FITI及PrefixSpan技術於量化跨界性關聯法則探勘模型中，以增進資料探勘的效率。最後則是設計模糊推論系統來預測海水溫度與鹽度變化。模糊法則庫取自於所發掘之溫度與鹽度樣式，這將使預測兼具準確性與彈性。本研究以資料探勘技術與模糊推論來分析臺灣附近海域海水溫度與鹽度資料集，實驗亦證明本研究所提之跨界性關聯法則演算法比其他研究所提之跨界性關聯法則演算法來得更有效率。

關鍵字

跨界性關聯法則 ； 資料探勘 ； 海水溫度 ； 海水鹽度 ； 模糊推論

並列摘要

Global ocean salinity and temperature variations are attracting increasing attention, due to its influence on global climate change. This research presents an efficient technique for analyzing Argo ocean data comprising time series of salinity and temperature measurements where informative salinity and temperature patterns are extracted. Most traditional mining techniques focus on finding associations among items within one transaction and are therefore unable to discover rich contextual patterns related to location and time. In order to show the associated salinity and temperature variations among different locations and time intervals, for example, “if the salinity rose from 0.15psu to 0.25psu in the area that is in the east-northeast direction and is near Taiwan, then the temperature will rise from 0℃ to 1.2℃ in the area that is in the east-northeast direction and is far away from Taiwan next month”, the research designs a transformation method to convert Argo spatial-temporal data to market-basket type data and then a quantitative inter-transaction association rules mining algorithm is proposed to apply to the transformed data set to get salinity and temperature variation patterns. The FITI and the PrefixSpan algorithms are adopted to maximize the mining efficiency. Next, a fuzzy inference model that employs the discovered salinity and temperature patterns as its rule base is designed to predict salinity and temperature variations. The strategy is applied to ocean salinity and temperature measurements obtained from the waters surrounding Taiwan. These experimental evaluations show that the proposed algorithm achieves better performance than other inter-transaction association rule mining algorithms.

並列關鍵字

inter-transaction association rules ； data mining ； ocean temperature ； ocean salinity ； fuzzy inference

參考文獻

[1] R. Agrawal, T. Imielinski and A. Swami, “Mining association rules between sets of items in large database,” Proc. ACM SIGMOD Int’l Conf. on Management of Data, Washington D.C., U.S.A., pp.207-216, May 1993.

[3] R. Agrawal and R. Srikant, “Mining sequential patterns,” Proc. Int’l Conf. on Data Engineer, Taipei, Taiwan, pp.3-14, Mar. 1995.

[4] R.A. Angryk and F.E. Petry, “Mining multi-level associations with fuzzy hierarchies,” Proc. IEEE Int’l Conf. on Fuzzy Systems, Reno, NV, U.S.A., pp.785-790, May 2005.

[6] S.D. Bay, “Multivariate discretization of continuous variables for set mining,” Proc. ACM SIGKDD Int’l Conf. on Knowledge Discovery and Data Mining, Boston, U.S.A., pp.315-319, Aug. 2000.

[7] U. Bhaskar, D. Swain and M. Ravichandran, “Inferring mixed-layer depth variability from Argo observations in the western Indian Ocean,” Journal of Marine Research, vol. 64, no. 3, pp.393-406, May 2006.