粗略集合論於預測電力需求之分析與應用

Rough Set Theory in Electricity Load Forecasting

陳泰吉(Tai-Ji Chen)

指導教授 ： 白炳豐

國立暨南國際大學/管理學院/資訊管理學系/碩士(2007年)

摘要

本文是以粗略集合論(Rough set theory, RST)建立一電力負載之預測模型，並藉此預測未來用電增減程度之百分比。對於電力調度人員而言，電力負載一直是個重要的課題，如何有效地調度，增加電力輸送之效率，降低廠房的反應時間，以提高電力使用品質，顯得極為重要。一個良好的預測系統不但可以提供有效的預測，更可以替國人帶來更高品質的用電環境。
特徵擷取(feature selection)在資料分析中，是一項非常有價值的技術，在資料縮減的領域中，用以保有原始資料之資訊顯得極為重要，由原始資料的部份子集合來呈現整個資料所擁有的資訊，對使用者而言有著非常重要的意義。粗略集合論在折減推演時，是一NP-hard的問題，在此我們提出新的方法來取代傳統求算折減的步驟。
傳統電力負載預測主要著重於多變量之經濟模型以及單變量之時間序列(time series)模型，雖然這些模型在電力負載預測領域中，有一定程度的貢獻，但這些模型主要卻是建立在數學函式及數值上，這些以數學函式為基礎的模型最大缺點在於不能處理非數值型態的資料，而粗略集合論最大的好處在於可同時處理數值型與非數值型的混合式資料，因此本文採用粗略集合論來解決這項問題。
本文所始用之資料為民國85年至民國92年每月資料作為訓練資料，以民國93、94年資料來評估預測模型之準確性。研究結果顯示，當決策屬性兩類時，能提供100%的正確率，分成三類時也能達到87.5%，因此，以粗略集合論結合線性判別分析(Linear Discriminant Analysis, LDA)之模型應用於電力需求，確實能提供不錯的預測。

關鍵字

電力預測 ； 粗略集合論 ； 特徵擷取

並列摘要

In this paper, we built an electric power loading forecasting system to predict the percentages of electric consumptions in the future by using rough set theory (RST). How to dispatch electricity to appropriate place effectively to increase the efficiency of electric transportation and reduce reaction time of equipments will be an important object for workmen in electric factories. A superior forecasting system can provide not only precise results but a high quality environment of electric consumptions for compatriots.
Feature selection is a valuable technique in data analysis. It is very important to preserve information of raw data in the domain of data reduction, and has significance for users by using subset of raw data to present information or knowledge. Hence, we propose a novel approach to extract important attributes instead of traditional steps because of an NP-hard problem for reducts induction by using rough set.
Traditional electric loading forecasting models concentrate multivariate econometric models and univariate time series. These models are primly based on mathematical functions and numeric in nature although these models had some degrees of contributions in electric loading forecasting. A major drawback of these models based on mathematical functions is their inability of dealing with non-numeric data. On the other hand, a major advantage of rough set is its ability to handle numeric and non-numeric data simultaneously. Hence, we employ rough set to solve this problem.
In our research, we obtain data monthly during the period of 1996-2003 for the usage of training data set, while the data during 2004-2005 were used to evaluate the forecasting model. Experiment results indicated a 100 percent accuracy while the decision attributes were divided into two-classes and also a 87.5 percent accuracy regarding three. It successfully forecasted by using combined model of rough set and linear discriminant analysis (LDA).

並列關鍵字

electricity load forecasting ； rough set ； feature selection

參考文獻

參考文獻

Ahn, B. S., S. S. Cho & C. Y. Kim, "The integrated methodology of rough set theory and artificial neural network for business failure prediction", Expert Systems with Applications Vol.18, pp. 65-74, 2000.

Al-Kandari, A. M., S. A. Soliman & M.E. El-Hawary, "Fuzzy short-term electric load forecasting", Electrical Power and Energy Systems, Vol.26, pp. 111-122, 2004.

An, A., N. Shan, C. Chan, N. Cercone & W. Ziarko "Discovering rules for water demand prediction: An enhanced rough-set approach", Engineering Application and Artificial Intelligence, Vol.9, pp. 645-653, 1996.

Batainch, S., A. A. Anbuky & S. A. Aqtash, "Power demand prediction using fuzzy logic", Control Eng. Practice, Vol.3, pp. 1291-1298, 1995.