面向电网故障诊断的BP神经网络优化算法研究

发布时间：2018-01-05 10:14

本文关键词：面向电网故障诊断的BP神经网络优化算法研究　出处：《昆明理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着经济的蓬勃发展,我国的电力行业发展向着更大容量、自动化水平更高的方向转变。对于电网运行稳定性的要求越来越高,电网运行中出现的故障给供电公司以及用户带来了巨大的损失。为了减少电网的故障时间,同时加强电网供电的可靠性,电网在发生故障后,应该准确及时地发现故障位置,隔离故障元件,消除故障隐患,从而提升系统的安全可靠性,同时采取相应的办法恢复电网运行。BP神经网络是目前应用于电网故障诊断中最广泛的神经网络模型之一,BP神经网络具有良好的自学习能力以及自适应和泛化能力,但是BP神经网络算法是基于梯度的方法,存在运算过程中容易陷入局部极小值的不足,同时当学习样本数目比较多、输入与输出关系比较复杂的时候,网络会出现收敛速度缓慢,收敛精度不高,甚至不收敛的问题。萤火虫算法具有全局寻优能力,利用萤火虫优化BP神经网络中的初始权值阈值,可以避免BP神经网络陷入局部极小的问题,有效地提高故障诊断的准确性。但是,传统的萤火虫算法本身也有缺陷,在全局寻优的过程中,容易得到局部最优解,出现“早熟”的问题。为了提高萤火虫算法的全局寻优能力,克服“早熟”问题。我们在传统萤火虫算法的基础上引入了变步长因子,当萤火虫个体在感知范围内没有找到优秀个体时,通过移动一个步长,从而改变邻域范围,在新范围内寻找局部最优。从而提高了算法的全局寻优能力。本论文提出了改进萤火虫算法优化BP神经网络。当萤火虫个体在感知范围内没有发现优秀个体时,通过引入变步长因子,随机移动一个步长,更新邻域半径,从而提高了算法的局部寻优能力。进而提高了全局寻优能力。将结合后的算法应用于电网区域的故障诊断,得到了较好的诊断效果。
[Abstract]:With the vigorous development of economy, China's power industry development toward larger capacity, higher automation level change direction. And the increasingly high demand for the stability of power grid, the fault occurred in power grid operation has brought huge losses to the power supply companies and users. In order to reduce the fault time of the grid, while enhancing the reliability of power supply the grid after a failure, should be timely and accurately find the fault location, fault isolation components, the failure to eliminate hidden dangers, so as to enhance the safety and reliability of the system, and take corresponding measures to restore the electric network running.BP neural network is one of the neural network model for fault diagnosis of the most widely used BP neural network with self adaptive learning ability and generalization ability and good, but the BP neural network algorithm is based on the gradient method, the operation process is easy to fall into existence Local minimum problem, at the same time when the number of samples to learn more, when the relationship between input and output is more complex, the network will appear slow convergence speed, convergence precision is not high, even no convergence problem. Firefly algorithm has the ability of global optimization, the initial weights and threshold using the firefly to optimize the BP neural network. To avoid the BP neural network into local minima problems, effectively improve the accuracy of fault diagnosis. However, the traditional firefly algorithm itself has a flaw in the global optimization process, easy to get the local optimal solution, the premature problem. In order to improve the global searching ability of the firefly algorithm, overcome the "premature" problem. We introduce a variable step size based on traditional firefly algorithm, when the firefly individual did not find the best individual in the perception range, by moving a step to Change the neighborhood range, find the local optima in the new range. In order to improve the ability of global optimization of the algorithm. This paper presents the improvement of firefly algorithm to optimize BP neural network. When the firefly individual found no outstanding individuals in the perception range, the variable step factor, random moving one step update neighborhood radius thus, to improve the local searching ability of the algorithm. And then improve the ability of global optimization. The fault diagnosis combined algorithm is applied to the regional power grid, get a good diagnosis effect.

【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TM711

【参考文献】