故障診斷論文：故障診斷人工免疫克隆選擇免疫抗體記憶分類思維進化

1、故障診斷論文：故障診斷人工免疫克隆選擇免疫抗體記憶 分類思維進化【中文摘要】電力變壓器是電力系統(tǒng)中的關(guān)鍵設(shè)備，其穩(wěn)定可靠 的運行是電力系統(tǒng)安全可靠運行的重要保障。一旦發(fā)生故障，將會影 響電力系統(tǒng)的穩(wěn)定運行，給人們的生產(chǎn)生活帶來極大地不便，因此電 力變壓器故障診斷一直為國內(nèi)外學(xué)者所關(guān)注。 本文簡要介紹了課題的 來源,背景及意義，通過參閱大量文獻分析國內(nèi)外研究現(xiàn)狀并對各種 診斷方法的優(yōu)劣做了比較。 本文詳細介紹了人工免疫算法原理及在變 壓器故障診斷中的應(yīng)用，針對傳統(tǒng)克隆免疫算法在變壓器故障診斷應(yīng) 用中學(xué)習(xí)速度慢、無分類能力及抗體空間小的限制提出了免疫抗體記 憶分類算法，該算法基于免疫空間抗體識別抗

2、原原理和抗體記憶功能 將克隆選擇與進化算法相結(jié)合，學(xué)習(xí)訓(xùn)練抗原及附屬類型信息，得到 不同類型檢測集組成的故障信息庫。并在抗體訓(xùn)練克隆過程中，對傳統(tǒng)的克隆算子進行改進，形成了免疫抗體記憶分類算法。該算法在保 證親和力的前提下提高了學(xué)習(xí)速度，擴大了免疫搜索空間，豐富了抗 體空間多樣性，提升了故障診斷的準(zhǔn)確率。同時以人工識別球（ARB） 為基本診斷單元，利用 ARB 與刺激水平線性關(guān)系對故障類型診斷，保 證了診斷的快速性。人工免疫算法本身具有動態(tài)學(xué)習(xí)，自適應(yīng),記憶，并行處理等優(yōu)點。但是在免疫算法中高頻變異算子可能造成本來具有 足夠親和力的抗體在通過變異后形成無功能的或是親和力較差的抗 體,減緩了訓(xùn)練

3、學(xué)習(xí)速度。針對以上不足本文引入思維進化算法，在抗 體的訓(xùn)練中將抗體群分為多個子抗體群，在趨同和異化操作的交替作 用下進行抗體親和力訓(xùn)練，在保證親和力穩(wěn)步增高的同時加快了親和 力訓(xùn)練的速度?！居⑽恼?Power transformer is the key equipment in the powersystem, the stable and reliableoperation of which is animportant guarantee for successful operation of power system. Oncefaulted, it will affect the

4、 stability operation of power system, and lead ing toa greatly inconvenience in the production and people s lives.As a result,power transformer fault diag no sis has bee n a concern to scholars. Thepaper brieflyin troduced the source,backgro und and sig nifica neeofthe topic , an alyze the domestic

5、and foreig n research status andcompared the quality of all kinds of diag no stic methods through reading alarge number of documents. It introduces the principle of artificial immunealgorithm and the application of tran sformer fault diag no sis. In order tomake up the shortages of traditional Cloni

6、ng immune algorithm for the transformer fault diag no sis applicati ons, such as the slowly learning speed,non-classified ability and the small antibody space. This paper proposes aclassification algorithm of antibody memory based on the principle that theantibodies recognize the antigens in the imm

7、unespace and antibodymemoryfunction.The algorithm, combining the clone selection and theevoluti onary algorithm, lear ns the training an tige n and subsidiary type informati on to build the fault in formati on database which is composed ofdiffere nt types of detecti on aggregate. The lear ning speed

8、 of thisalgorithm is very fast while the affinity is high eno ugh. At the same time, ithas an expanded immunization search space and improves the antibodiesspatial diversity. Therefore, this algorithm can improve the accuracy offault diag no sis. Besides, it use the artificial recognition ball (ARB)

9、 as thebasic diagnostic unit. As a result, the speed of diag no sis is greatlyimproved because of the lin ear relati on ship betwee n the ARB and thestimulati on level.The artificial immunealgorithm has the ability of dynamiclearning, adaptive, memory, parallel processing and so on. However, thehype

10、rmutati on mecha nism would lead to the non-fun cti onal or pooraffinity antibodies from sufficient affinity antibodies and a slower lear ningspeed. In order to make up the shortco min gs, this article in troduces in mind evoluti onary algorithm that the an tibody groups will be divided intomultiple sub-antibodi