【摘要】：電力變壓器是電力系統(tǒng)中最重要的設(shè)備之一,它的運(yùn)行狀態(tài)直接影響到整個(gè)電網(wǎng)的輸變電狀態(tài)。在電力行業(yè)不斷發(fā)展的今天,由于諸多因素的影響,傳統(tǒng)的DGA方法已經(jīng)無法準(zhǔn)確地判別出變壓器的故障類型,滿足不了現(xiàn)今對(duì)變壓器故障判別準(zhǔn)確率的要求。因此,DGA與智能方法結(jié)合的組合模型己成為變壓器故障診斷的一種必然發(fā)展趨勢(shì)。目前,最常用的診斷方法是DGA與BP神經(jīng)網(wǎng)絡(luò)的組合模型。在這一模型中,BP網(wǎng)絡(luò)本身存在的自適應(yīng)學(xué)習(xí)、并行處理、聯(lián)想記憶和非線性映射等特性可以完善普通DGA方法存在的缺陷。然而,若收集到的故障樣本數(shù)目過于龐大且對(duì)故障診斷的檢測(cè)精度要求較高時(shí),BP網(wǎng)絡(luò)本身的缺陷將延長網(wǎng)絡(luò)達(dá)到收斂時(shí)所需要的時(shí)間,甚至使得網(wǎng)絡(luò)不收斂,易將局部最小值當(dāng)作全局最優(yōu)值,從而導(dǎo)致故障診斷的準(zhǔn)確率降低。所以,DGA與BP神經(jīng)網(wǎng)絡(luò)的組合模型在變壓器故障診斷方面仍然存在不足,為了進(jìn)一步完善該方法,有必要利用其它優(yōu)化方法對(duì)BP神經(jīng)網(wǎng)絡(luò)進(jìn)行改進(jìn)。蟻群算法(ACA)是一種新型的仿生態(tài)算法,它具有全局優(yōu)化能力和啟發(fā)式搜索特性,將其與BP網(wǎng)絡(luò)結(jié)合可以改進(jìn)BP網(wǎng)絡(luò)的性能。本文中提出利用ACA調(diào)節(jié)BP神經(jīng)網(wǎng)絡(luò)權(quán)值,達(dá)到提高BP網(wǎng)絡(luò)性能的要求,并利用改進(jìn)后的BP網(wǎng)絡(luò)對(duì)變壓器故障進(jìn)行再診斷,以此驗(yàn)證新方法的優(yōu)越性。首先構(gòu)建出結(jié)構(gòu)為5-8-5的BP神經(jīng)網(wǎng)絡(luò)對(duì)變壓器進(jìn)行仿真與故障識(shí)別,利用MATLAB編寫程序,得出結(jié)果,證明單純的BP神經(jīng)網(wǎng)絡(luò)能夠?qū)ψ儔浩鞴收线M(jìn)行識(shí)別,但準(zhǔn)確率不高。其次闡述利用ACA優(yōu)化BP網(wǎng)絡(luò)的基本原理,說明其核心思想就是利用ACA優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的權(quán)值參數(shù),提出基于ACA優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的變壓器故障診斷方法(ACA-BP方法),利用ACA-BP方法對(duì)變壓器進(jìn)行仿真分析和故障識(shí)別,對(duì)比單純BP網(wǎng)絡(luò)診斷變壓器故障的結(jié)果,證明ACA-BP方法能夠有效防止訓(xùn)練過程中診斷結(jié)果陷入局部最優(yōu),加快網(wǎng)絡(luò)收斂速度,縮短學(xué)習(xí)時(shí)間,提高故障辨識(shí)的準(zhǔn)確度,更準(zhǔn)確地反應(yīng)出變壓器的實(shí)際故障。文章最后總結(jié)了ACA-BP方法的優(yōu)越性及進(jìn)一步改進(jìn)的方向。
[Abstract]:Power transformer is one of the most important equipments in power system, its running state directly affects the transmission and transformation state of the whole power network. With the continuous development of electric power industry, due to the influence of many factors, the traditional DGA method has been unable to accurately identify the fault types of transformers, and can not meet the requirements of the accuracy of transformer fault discrimination. Therefore, the combination model of DGA and intelligent method has become an inevitable trend of transformer fault diagnosis. At present, the most commonly used diagnostic method is the combination model of DGA and BP neural network. In this model, the self-adaptive learning, parallel processing, associative memory and nonlinear mapping of BP networks can perfect the defects of ordinary DGA methods. However, if the number of fault samples collected is too large and the detection accuracy of fault diagnosis is high, the defect of the BP network itself will prolong the time required for the network to reach convergence, even make the network not converge. The local minimum value is easy to be regarded as the global optimal value, which leads to the reduction of the accuracy of fault diagnosis. Therefore, the combination model of DGA and BP neural network still has shortcomings in transformer fault diagnosis. In order to further improve the method, it is necessary to use other optimization methods to improve the BP neural network. Ant colony algorithm (ACA) is a new ecological-like algorithm, which has global optimization ability and heuristic search characteristics. Combining it with BP network can improve the performance of BP network. In this paper, ACA is used to adjust the weights of BP neural network to improve the performance of BP network, and the improved BP network is used to re-diagnose transformer faults, so as to verify the superiority of the new method. Firstly, the BP neural network with the structure of 5 ~ 8 ~ 5 is constructed to simulate and identify the transformer fault. The program is programmed by MATLAB, and the result shows that the simple BP neural network can identify the transformer fault, but the accuracy is not high. Secondly, the basic principle of optimizing BP network by ACA is described, and the core idea is to optimize the weight parameters of BP neural network by ACA. A fault diagnosis method for transformer based on ACA optimized BP neural network (ACA-BP method) is put forward. The ACA-BP method is used to simulate the transformer and identify the faults. Compared with the results of simple BP network diagnosis, it is proved that the ACA-BP method can effectively prevent the diagnosis results from falling into the local optimum in the course of training. The network convergence speed is accelerated, the learning time is shortened, the accuracy of fault identification is improved, and the actual fault of transformer is reflected more accurately. Finally, the advantages of ACA-BP method and the direction of further improvement are summarized.
【學(xué)位授予單位】：長沙理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2014
【分類號(hào)】：TM407;TP18
，

本文編號(hào)：2472262