摘 要
在配電網(wǎng)調(diào)度自動(dòng)化系統(tǒng)中，報(bào)警系統(tǒng)信息量巨大，其中往往含有大量的噪聲信息。傳統(tǒng)的報(bào)警系統(tǒng)只是單純地上傳告警信息，未經(jīng)處理且快速變化的海量報(bào)警信息容易造成調(diào)度員忽略真正重要的報(bào)警信息，延時(shí)處理故障，危害電網(wǎng)安全。
針對以上問題，本文將數(shù)據(jù)挖掘技術(shù)引入到配電網(wǎng)報(bào)警信息的噪聲處理中，應(yīng)用改進(jìn)后的ID3算法構(gòu)造對配電網(wǎng)中的報(bào)警信息進(jìn)行噪聲過濾的分類器。該分類器可以將報(bào)警信息分為噪聲信息和非噪聲信息兩類，很好的解決了報(bào)警信息不加識(shí)別的上傳給調(diào)度人員的缺點(diǎn)。本文所做的工作主要如下：
(1) 深入研究了數(shù)據(jù)挖掘技術(shù)，重點(diǎn)對數(shù)據(jù)挖掘中的決策樹算法進(jìn)行了詳細(xì)探討；概括總結(jié)了ID3算法、C4.5算法、CART算法、SLIQ算法及SPRINT算法的特點(diǎn)及優(yōu)缺點(diǎn)。
(2) 本文探討了經(jīng)典的ID3算法的兩個(gè)突出缺點(diǎn)：運(yùn)算復(fù)雜和在選擇屬性時(shí)偏向于取值較多的屬性，并對這兩個(gè)方面進(jìn)行改進(jìn)。ID3算法是以信息熵理論為基礎(chǔ)構(gòu)建決策樹的，每次在選擇分裂屬性時(shí)都要對其進(jìn)行多次對數(shù)運(yùn)算，在數(shù)據(jù)量較多的時(shí)候，運(yùn)算速度明顯變慢，本文針對這一問題，提出了相應(yīng)的簡化方法；再則，ID3應(yīng)用信息增益作為最佳屬性的選擇標(biāo)準(zhǔn)，這樣就會(huì)導(dǎo)致該算法偏向于選擇取值較多的屬性，而有時(shí)候取值較多的屬性并不是最優(yōu)屬性，針對這一問題，本文引入一乘積因子到簡化后的信息熵中，來克服ID3算法的取值的偏向性。
(3) 本文通過實(shí)例分析對改進(jìn)后的ID3算法的優(yōu)越性進(jìn)行實(shí)驗(yàn)驗(yàn)證，并實(shí)現(xiàn)了用MATLAB語言來構(gòu)造決策樹。
(4) 本文在研究配電網(wǎng)中報(bào)警信息特點(diǎn)的基礎(chǔ)上，實(shí)現(xiàn)了配電網(wǎng)中的報(bào)警信息的預(yù)處理及主要?dú)w納關(guān)系的形成，只有對形式復(fù)雜的報(bào)警信息進(jìn)行預(yù)處理并且得到主要?dú)w納關(guān)系后才可以將它們用于數(shù)據(jù)挖掘任務(wù)中。
(5) 用改進(jìn)后的ID3決策樹算法設(shè)計(jì)了用于配電網(wǎng)報(bào)警信息中噪聲識(shí)別的分類器。該分類器可以從報(bào)警信息中識(shí)別出噪聲信息。

關(guān)鍵詞 數(shù)據(jù)挖掘；決策樹；ID3算法；報(bào)警信息

Abstract
In dispatching automation system of distribution power grid, there is a large amount of information which usually contains a lot of noisy information. The traditional alarm system simply uploads alarm information, so the untreated and fast changing mass alarm information will result in ignoring really important information for dispatcher, and it will also delay the processing of the fault and endanger the safety of the power grid.
Aming at the problems above, this paper introduces data mining technology to the noise processing of alarm information in power grid，using the improved ID3 algorithm to construct noise filtering classifier dealing with alarm information in power grid. The classifer constructed in this paper can divide the alarm information into two kinds: noisy information and normal alarm information, and it can resolve the defect that the alarm information is uploaded to dispatcher without recognition. The main work done by this paper is as follows:
(1) This paper studies the data mining technology deeply, mainly discusses decision tree algorithm in detail; This paper sums up the characteristics, strengths and weaknesses of the follow algorithm: ID3 algorithm, C4.5 algorithm, CART algorithm, SLIQ algorithm, and SPRINT algorithm.
(2) This paper discusses two shortcomings of the classical ID3 algorithm: too complicated in calculating and leaning on the attribute which have more value while selecting the optimum attribute, aiming at these two aspects, this paper puts forward corresponding improved method. ID3 algorithm which bases on information entropy theory has multiple logarithm operation in choosing the splitting attribute when building the decision tree, so the computing speed will slows down especially when the data is large, aiming at this problem, this paper puts forward corresponding simplified method. Furthermore, the ID3 algorithm applicates the information gain as the criteria of seleting splitting attribute, so it often leans on the attribute which has more value, aiming at this problem, we introduce a divisor to the simplified ID3 algorithm to overcome the bias on attribute.
(3) Through the analysis of examples, the superiority of improved ID3 algorithm is verified and this paper carries out the construction of decision tree by MATLAB language.
(4) Based on the study of the alarm information’s characteristics, this paper realizes the preprocessing and relationship induction of the alarm information. The alarm information has complex forms, so it’s necessary to preprocess it and summarize relationship from it before data mining.
(5) Constructing noise information recognition classifier used in alarm information of distribution power grid with improved ID3 algorithm, this classifier can recognize noise information from alarm information.

Key words data mining; decision tree; ID3 algorithm; alarm information


目 錄
摘 要 I
Abstract III
第1章 緒 論 1
1.1 課題的研究背景 1
1.2 國內(nèi)外研究現(xiàn)狀 3
1.3 本文的主要工作 5
1.4 本文的組織結(jié)構(gòu) 5
第2章 數(shù)據(jù)挖掘技術(shù)及其在電力系統(tǒng)中的應(yīng)用 7
2.1 數(shù)據(jù)挖掘簡介 7
2.2 數(shù)據(jù)挖掘的任務(wù)及步驟 8
2.3 數(shù)據(jù)挖掘中常用的算法和技術(shù) 10
2.4 電力系統(tǒng)中的數(shù)據(jù)挖掘 12
2.5 配電網(wǎng)報(bào)警信息處理中的數(shù)據(jù)挖掘 13
2.5.1 遠(yuǎn)動(dòng)系統(tǒng)的特點(diǎn) 13
2.5.2 報(bào)警信息的分類 14
2.5.3 報(bào)警信息的特點(diǎn) 14
2.5.4配電網(wǎng)報(bào)警信息中的數(shù)據(jù)挖掘 15
2.6..