本文選題：故障診斷 + 教學(xué)算法(TLBO)�。� 參考：《華東理工大學(xué)》2015年碩士論文

【摘要】：支持向量機(jī)作為機(jī)器學(xué)習(xí)方法中一種比較成熟的理論已經(jīng)在實(shí)際化工生產(chǎn)中得到了越來(lái)越廣泛的應(yīng)用。本文主要提出了一種加權(quán)精英教學(xué)算法(WETLBO)并將其應(yīng)用到在支持向量機(jī)的參數(shù)優(yōu)化過(guò)程中,最后使用上述優(yōu)化后的模式識(shí)別方法解決Tennessee Eastman實(shí)際化工故障診斷問(wèn)題。主要工作如下：1.本文提出了一種改進(jìn)的加權(quán)精英教學(xué)算法。為了進(jìn)一步提高該算法在參數(shù)尋優(yōu)中的運(yùn)算效率和收斂速度,將精英思想引入到教學(xué)算法中,精英個(gè)體在整個(gè)算法中作為每次迭代過(guò)程中適應(yīng)度值最高的若干個(gè)最優(yōu)解將被直接保留至下一次的迭代運(yùn)算中。并根據(jù)精英個(gè)體和普通個(gè)體不同的特性,在種群更新時(shí)對(duì)精英個(gè)體和普通個(gè)體分別采用不同的慣性權(quán)重變化策略。對(duì)測(cè)試函數(shù)尋優(yōu)的結(jié)果驗(yàn)證了改進(jìn)后算法的可行性及一定的全局搜索能力。2.支持向量機(jī)中的懲罰參數(shù)和核函數(shù)參數(shù)往往決定著分類(lèi)器的性能。本文將支持向量機(jī)在訓(xùn)練過(guò)程中的參數(shù)選擇問(wèn)題轉(zhuǎn)化為與適應(yīng)度值相關(guān)的最優(yōu)化問(wèn)題,并使用加權(quán)精英教學(xué)算法用于最優(yōu)解的計(jì)算,從而得到最優(yōu)的支持向量機(jī)參數(shù)。3.針對(duì)美國(guó)Tennessee Eastman工廠的實(shí)際化工過(guò)程模型所產(chǎn)生的歷史故障數(shù)據(jù),本文使用支持向量機(jī)結(jié)合加權(quán)精英教學(xué)TLBO算法給出了具體的故障診斷和分析過(guò)程。結(jié)果表明,采用改進(jìn)后的WETLBO算法優(yōu)化懲罰參數(shù)和核函數(shù)參數(shù)后的支持向量分類(lèi)機(jī)應(yīng)用到故障數(shù)據(jù)分類(lèi)中,相比于其他常見(jiàn)的參數(shù)優(yōu)化算法方法而言具有較理想的分類(lèi)效果。
[Abstract]:As a mature theory in machine learning, support vector machine (SVM) has been widely used in practical chemical production. In this paper, a weighted elite teaching algorithm (WETLBOA) is proposed and applied to the parameter optimization of support vector machine (SVM). Finally, the above optimized pattern recognition method is used to solve the problem of practical chemical fault diagnosis in Tennessee Eastman. The main work is as follows: 1. This paper presents an improved weighted elite teaching algorithm. In order to further improve the efficiency and convergence speed of the algorithm in parameter optimization, the elite idea is introduced into the teaching algorithm. Elite individuals in the whole algorithm as the highest fitness in each iteration process of several optimal solutions will be directly retained in the next iteration operation. According to the different characteristics of elite individuals and ordinary individuals, different inertial weight changing strategies are adopted for elite individuals and ordinary individuals respectively during population renewal. The result of optimizing test function verifies the feasibility of the improved algorithm and its global search ability. 2. The performance of classifier is often determined by penalty parameters and kernel function parameters in SVM. In this paper, the parameter selection problem of support vector machine in training process is transformed into an optimization problem related to fitness, and the weighted elite teaching algorithm is used to calculate the optimal solution, and the optimal support vector machine parameter .3 is obtained. In view of the historical fault data generated by the actual chemical process model of the Tennessee Eastman factory in the United States, this paper presents a specific fault diagnosis and analysis process using support vector machine (SVM) combined with the weighted elite teaching TLBO algorithm. The results show that the improved WETLBO algorithm can be applied to the classification of fault data by optimizing the penalty parameters and kernel function parameters, and it has a better classification effect than other common parameter optimization algorithms.
【學(xué)位授予單位】：華東理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2015
【分類(lèi)號(hào)】：TQ050.7;TP181

【參考文獻(xiàn)】

相關(guān)期刊論文 前9條

1 蔣東翔;張禮勇;徐世昌;黃文虎;;基于知識(shí)的傳感器故障在線診斷系統(tǒng)[J];電測(cè)與儀表;1993年11期

2 呂干云,程浩忠,董立新,翟海保;基于多級(jí)支持向量機(jī)分類(lèi)器的電力變壓器故障識(shí)別[J];電力系統(tǒng)及其自動(dòng)化學(xué)報(bào);2005年01期

3 薄翠梅;張n\;張廣明;王執(zhí)銓;;基于特征樣本核主元分析的TE過(guò)程快速故障辨識(shí)方法(英文)[J];化工學(xué)報(bào);2008年07期

4 劉志剛,李德仁,秦前清,史文中;支持向量機(jī)在多類(lèi)分類(lèi)問(wèn)題中的推廣[J];計(jì)算機(jī)工程與應(yīng)用;2004年07期

5 羅躍綱,曾海泉,聞邦椿;機(jī)械故障診斷的遺傳BP算法應(yīng)用研究[J];機(jī)械科學(xué)與技術(shù);2002年04期

6 張學(xué)工;關(guān)于統(tǒng)計(jì)學(xué)習(xí)理論與支持向量機(jī)[J];自動(dòng)化學(xué)報(bào);2000年01期

7 羅剛,張n\,牛彥杰;基于TE過(guò)程的仿真系統(tǒng)的實(shí)現(xiàn)[J];南京工業(yè)大學(xué)學(xué)報(bào)(自然科學(xué)版);2005年03期

8 于坤杰;王昕;王振雷;;基于反饋的精英教學(xué)優(yōu)化算法[J];自動(dòng)化學(xué)報(bào);2014年09期

9 陳貴敏;賈建援;韓琪;;粒子群優(yōu)化算法的慣性權(quán)值遞減策略研究[J];西安交通大學(xué)學(xué)報(bào);2006年01期

，

本文編號(hào)：1952254