基于改进神经网络算法的交变磁化下磁致伸缩模型研究

发布时间：2019-04-15 12:10

【摘要】：电力变压器铁心和电机的定子铁心分别由取向硅钢片和无取向硅钢片堆叠而成,而这两种硅钢片的磁致伸缩会引起铁心的振动噪声,从而减少电工设备的使用寿命并影响人们的生活健康。因此,准确的测量与模拟硅钢片的磁致伸缩特性是研究变压器和电机铁心振动噪声必不可少的先行工作。考虑到现有的磁致伸缩模型的研究大多基于复杂多项式拟合的方法,其不足是公式的推导过程较为复杂且需要拟合的参数过多。本文对交变磁场下取向和无取向硅钢片的磁致伸缩特性进行测量,并以取向硅钢片的磁致伸缩测量数据为基础,建立了改进型神经网络算法的磁致伸缩模型,通过单相变压器铁心实验验证了模型的准确性。具体的研究内容包括:首先,在交变磁场下,利用实验室现有的基于三轴应变片的磁致伸缩测量系统,对取向和无取向硅钢片磁致伸缩特性进行测量,并计算了磁致伸缩主应变的大小和方向。对比分析了两种硅钢片在交变磁场下磁致伸缩的共性和异性,并对两种硅钢片的磁致伸缩的各向异性进行了验证。其次,以交变磁场下测量的取向硅钢片的磁致伸缩数据为基础,建立了基于梯度下降法的传统Back Propagation(BP)神经网络的磁致伸缩模型,模型的输入为x和y两个方向上磁通密度的完整波形,输出为磁致伸缩的主应变大小及其方向波形。并指出传统的基于梯度下降法的BP算法的神经网络在建立的磁致伸缩模型应用中的缺点,其缺陷为网络的计算速度过慢且容易陷入局部最优解。再次,针对基于梯度下降法传统的BP算法进行改进。使用粒子群算法(PSO)优化了BP神经网络(BPNN)的初始权阈值,从而避免了网络在训练过程中容易陷入局部最优解的问题;利用Levenberg Marquardt(LM)算法代替梯度下降法对网络进行训练,大幅度提高了网络的计算速度。之后利用改进型BP神经网络重新对磁致伸缩进行建模,并以不在训练集中的15个磁致伸缩测量样本给出了模型的精度范围。最后,设计并制作了单相变压器铁心实验模型,测量了铁心上三个不同位置上的磁致伸缩特性,并用测量值与模型计算值进行对比,验证了建立的磁致伸缩模型在交变磁场下的准确性,为以后研究变压器铁心的振动噪声问题提供了参考。
[Abstract]:The stator cores of power transformer and motor are stacked with oriented silicon steel sheet and non-oriented silicon steel sheet respectively, and the magnetostriction of these two kinds of silicon steel sheets will cause vibration and noise of the iron core. Thus reduce the service life of electrical equipment and affect the health of people's lives. Therefore, the accurate measurement and simulation of the magnetostrictive characteristics of silicon steel sheet is necessary to study the vibration and noise of transformer and motor core. Considering that the existing magnetostrictive models are mostly based on the method of complex polynomial fitting, its deficiency is that the derivation process of the formula is more complex and there are too many parameters to be fitted. In this paper, the magnetostrictive characteristics of oriented and non-oriented silicon steel sheets under alternating magnetic field are measured. Based on the magnetostrictive measurement data of oriented silicon steel sheets, the magnetostrictive model of the improved neural network algorithm is established. The accuracy of the model is verified by the core experiment of single-phase transformer. The main contents are as follows: firstly, under alternating magnetic field, the magnetostrictive characteristics of oriented and non-oriented silicon steel sheets are measured by using the existing magnetostrictive measurement system based on triaxial strain gages in the laboratory. The magnitude and direction of the magnetostrictive principal strain are calculated. The generality and anisotropy of magnetostrictive properties of two kinds of silicon steel sheets under alternating magnetic field are compared and analyzed, and the anisotropy of magnetostrictive properties of two kinds of silicon steel sheets is verified. Secondly, based on the magnetostrictive data of oriented silicon steel sheet measured under alternating magnetic field, the magnetostrictive model of traditional Back Propagation (BP) neural network based on gradient descent method is established. The input of the model is the complete waveform of flux density in x and y directions, and the output is the main strain size and direction waveform of magnetostriction. The disadvantages of the traditional BP neural network based on gradient descent method in the application of the magnetostrictive model are pointed out, which is that the computation speed of the network is too slow and it is easy to fall into the local optimal solution. Thirdly, the traditional BP algorithm based on gradient descent method is improved. Particle swarm optimization algorithm (PSO) is used to optimize the initial weight threshold of BP neural network (BPNN) so as to avoid the problem that the network can easily fall into the local optimal solution in the course of training. The Levenberg Marquardt (LM) algorithm is used to train the network instead of the gradient descent method, which greatly improves the computing speed of the network. Then the modified BP neural network is used to re-model the magnetostriction and the precision range of the model is given with 15 magnetostrictive measurement samples which are not in the training set. Finally, an experimental model of single-phase transformer core is designed and made. The magnetostriction characteristics of three different positions on the core are measured, and the measured values are compared with those calculated by the model. The veracity of the established magnetostrictive model under alternating magnetic field is verified, which provides a reference for further study on the vibration and noise of transformer core.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM275

【相似文献】