摘 要
本文在分析現(xiàn)代配電網(wǎng)拓?fù)浣Y(jié)構(gòu)以及傳統(tǒng)故障定位算法的基礎(chǔ)上，采用一種基于改進(jìn)BP神經(jīng)網(wǎng)絡(luò)算法、遺傳優(yōu)化神經(jīng)網(wǎng)絡(luò)GA-BP算法以及RBF徑向基神經(jīng)網(wǎng)絡(luò)算法在配電網(wǎng)故障定位中的應(yīng)用。對(duì)三種神經(jīng)網(wǎng)絡(luò)在配電網(wǎng)故障定位中的應(yīng)用進(jìn)行系統(tǒng)比較分析，從而為實(shí)現(xiàn)了配電網(wǎng)的故障診斷、隔離故障區(qū)域以及恢復(fù)非故障區(qū)域供電。配電網(wǎng)故障定位算法的研究及其改進(jìn)成為本論文的工作方向和重點(diǎn)內(nèi)容。
本文對(duì)故障定位算法做了系統(tǒng)深入的研究，主要研究工作如下：
1、深入研究了配電網(wǎng)拓?fù)浣Y(jié)構(gòu)。算法的實(shí)現(xiàn)正是基于配電網(wǎng)的拓?fù)浣Y(jié)構(gòu)，本文采用了現(xiàn)代電網(wǎng)手拉手的環(huán)狀結(jié)構(gòu)，具有正常時(shí)閉環(huán)結(jié)構(gòu)，開(kāi)環(huán)運(yùn)行，呈輻射狀向用戶供電的特點(diǎn)。配電網(wǎng)發(fā)生故障后，安裝在各分段開(kāi)關(guān)處的FTU會(huì)檢測(cè)到故障信息(如故障過(guò)電流)，上傳到控制中心SCADA系統(tǒng)，系統(tǒng)經(jīng)過(guò)故障診斷算法綜合分析，判別故障點(diǎn)位置，實(shí)現(xiàn)了故障定位，并下達(dá)命令遙控FTU斷開(kāi)故障點(diǎn)兩側(cè)的分段開(kāi)關(guān)，進(jìn)而隔離故障區(qū)域。
2、針對(duì)BP神經(jīng)網(wǎng)絡(luò)未考慮前一次調(diào)整時(shí)的誤差梯度方向以及最佳學(xué)習(xí)率的問(wèn)題，使得網(wǎng)絡(luò)訓(xùn)練過(guò)程發(fā)生振蕩，收斂緩慢，本文采用一種改進(jìn)BP神經(jīng)網(wǎng)絡(luò)算法在配電網(wǎng)故障定位中的應(yīng)用，解決了上述問(wèn)題，并通過(guò)了算例仿真驗(yàn)證。
3、針對(duì)神經(jīng)網(wǎng)絡(luò)收斂速度慢和容易陷入局部極小值的問(wèn)題，本文提出將遺傳神經(jīng)網(wǎng)絡(luò)算法應(yīng)用于配電網(wǎng)故障定位。用遺傳算法優(yōu)化BP神經(jīng)網(wǎng)絡(luò)解決了BP網(wǎng)絡(luò)的最優(yōu)初始權(quán)值和閾值問(wèn)題，加快了網(wǎng)絡(luò)的收斂。利用遺傳算法的全局搜索能力，進(jìn)一步提高了BP神經(jīng)網(wǎng)絡(luò)故障定位的準(zhǔn)確性和快速性，并通過(guò)了算例仿真驗(yàn)證。
4、針對(duì)遺傳算法訓(xùn)練時(shí)間長(zhǎng)、BP神經(jīng)網(wǎng)絡(luò)容錯(cuò)性能不佳、BP隱含層神經(jīng)元個(gè)數(shù)難以確定、收斂速度慢和容易陷入局部最優(yōu)的問(wèn)題，本文將RBF神經(jīng)網(wǎng)絡(luò)算法引入配電網(wǎng)故障定位中。RBF網(wǎng)絡(luò)采用隱含層為高斯函數(shù)，是局部逼近網(wǎng)絡(luò)，有效的加快了收斂速度和避免局部最優(yōu)。在實(shí)際配電網(wǎng)故障診斷中，采用RBF神經(jīng)網(wǎng)絡(luò)，實(shí)現(xiàn)了故障點(diǎn)的準(zhǔn)確定位，并利用Visual C++工具開(kāi)發(fā)故障定位主站，實(shí)現(xiàn)了配電網(wǎng)自動(dòng)化目標(biāo)，對(duì)故障判斷準(zhǔn)確，反應(yīng)迅速，完全達(dá)到了實(shí)時(shí)監(jiān)測(cè)的要求。

關(guān)鍵詞 配電網(wǎng)；故障定位；神經(jīng)網(wǎng)絡(luò)；遺傳算法；RBF；C++













































Abstract
Based on the analysis of modern distribution network topological structure and traditional fault location algorithm, this paper propose an improved BP neural network algorithm, genetic optimization of neural network GA-BP algorithm and RBF neural network algorithm in the application of distribution network fault section location. Through a comparative analysis of three types of neural network, so as to achieve distribution network fault location, isolation and restoration of power for fault region. Distribution network fault location algorithm and its improved become the main research contents.
In this paper, it studies distribution network fault location algorithm. The main research work is as follows:
1. In-depth study of the topological structure for distribution network. Algorithm is based on the topological structure of distribution network. This paper adopts the modern network hand in hand ring structure, with normal closed-loop structure, open loop operation, radially to user. After Distribution network faults, the FTU which installed in the switch detects the fault information (such as overcurrent), upload to the control center which called SCADA system. Based on fault diagnosis algorithm, the system analysis fault location.Then remote order the FTU disconnection switches on both sides of the fault section, and then isolating the fault area.
2. According to the BP neural network does not take into account the previous adjustment error gradient direction as well as the best learning rate problem, making the network training process occurs oscillation and converges slowly. This paper presents an improved BP neural network algorithm in fault location for distribution network, which solves the problem and through simulation.
3. According to the BP neural network converges slowly and easily falling into local minimum problem, this paper puts forward the genetic neural network algorithm is applied to fault location in distribution network. Using genetic algorithm to optimize BP neural network to solve the BP network optimal initial weights and thresholds, accelerate the network convergence. Using the global search ability of genetic algorithm, further improve the BP neural network fault positioning accuracy and rapidity, and through the example simulation.
4. According to Genetic algorithm training for a long time, the BP neural network fault tolerant performance, BP hidden layer neuron number is difficult to determine, converges slowly and easily to fall into local optimal problem. This paper adopts the RBF neural network algorithm for fault location. RBF networks using implicit layer for the Gauss function, which is local approximation network, effectively accelerates the convergence speed and avoid local optimum. In the practical fault diagnosis of distribution network, the RBF neural network, realizes the accurate fault location, and makes use of Visual C++ development tool for master station in distribution network automation, which achieves the goal..