基于凸组合自适应滤波的变压器非线性有源噪声控制研究

发布时间：2019-06-15 22:02

【摘要】：随着中国经济飞速发展,人们越来越关注生活的品质。降低变电站噪声,可以减轻噪声对周围群众生活的干扰,大幅提升其生活质量。有源噪声控制系统能有针对性的降低变电站低频噪声,因此,简单、经济、有效的有源降噪系统在电力变压器噪声控制领域有着广阔应用前景。目前,大部分研究以线性系统为基础,该系统收敛速度与稳态误差之间存在不可调和的矛盾,而且在实际运行中,控制系统中不可避免的存在非线性因素,使这类线性系统的控制性能大打折扣。如果能解决系统中的非线性问题,有源降噪系统就能够选用含有非线性失真的低价电声器件,不仅能提高降噪性能,对降低系统的成本同样具有特殊意义。因此,论文以国家自然科学基金项目《基于内模控制的阵列式电力变压器有源消声技术研究》为依托,对电力变压器非线性有源噪声控制技术及系统收敛速度与稳态误差的协调方法进行了研究。论文针对系统收敛速度与稳态误差之间的矛盾,将凸组合滤波器引入有源噪声控制系统,对该结构所采用的组合算法及系统稳定性、收敛速度、稳态误差进行了详细的理论推导,理论证明组合算法能够兼顾稳态误差与收敛速度取得良好的综合性能。分别对目前非线性噪声控制系统中采用的两种主流方法进行了研究,一种是基于函数连接型神经网络(FLANN)的方法,另一种是基于Volterra滤波器的方法。分别阐述了它们的控制系统基本结构及经典算法,并做了可行性分析。分别将这两种滤波器的组合结构引入非线性有源噪声控制系统,并对它们采用的组合算法进行了理论推导,针对传统组合算法在收敛过程中出现的停滞现象,采用附加瞬时转移结构对算法进行优化。研究发现FLANN结构具有结构简单、计算量较小等优点,主要应用于非线性程度较弱的情况。而Volterra滤波器具有较强的非线性处理能力,但核函数的数目和计算量都随输入信号长度呈指数式增加。因此,将这两种滤波器通过凸组合后引入有源噪声控制系统,以结合它们在非线性处理能力,计算复杂度方面的优势,对该组合结构及算法进行了推导。为降低算法计算复杂度,采用修正箕舌线函数对联合系数进行选取,采用sign函数对混合系数进行更新,降低计算量的同时,能够避免联合参数趋向0或1时,混合系数更新缓慢甚至停滞的现象。最后通过大量仿真实验对文中研究内容进行了验证。
[Abstract]:With the rapid development of Chinese economy, people pay more and more attention to the quality of life. Reducing substation noise can reduce the interference of noise to the life of the surrounding people and greatly improve its quality of life. The active noise control system can reduce the low frequency noise of substation. Therefore, the simple, economical and effective active noise reduction system has a broad application prospect in the field of power transformer noise control. At present, most of the research is based on linear system, and there is an irreconcilable contradiction between the convergence rate and steady-state error of the system, and in the actual operation, there are inevitable nonlinear factors in the control system, which greatly reduces the control performance of this kind of linear system. If the nonlinear problems in the system can be solved, the active noise reduction system can select low-cost electroacoustic devices with nonlinear distortion, which can not only improve the noise reduction performance, but also has special significance to reduce the cost of the system. Therefore, based on the National Natural Science Foundation of China project "Research on active noise suppression Technology of Array Power Transformer based on Internal Model Control", the nonlinear active noise control technology of power transformer and the coordination method between convergence speed and steady state error of power transformer are studied in this paper. In view of the contradiction between the convergence speed and the steady-state error of the system, the convex combination filter is introduced into the active noise control system. The combined algorithm, the system stability, the convergence speed and the steady-state error of the structure are deduced in detail. The theory proves that the combined algorithm can achieve good comprehensive performance taking into account the steady-state error and the convergence speed. Two mainstream methods used in nonlinear noise control systems are studied, one is based on functional connection neural network (FLANN), the other is based on Volterra filter. The basic structure and classical algorithm of their control system are described respectively, and the feasibility analysis is made. The combined structures of the two filters are introduced into the nonlinear active noise control system, and the combined algorithms are deduced theoretically. Aiming at the stagnation phenomenon of the traditional combination algorithms in the process of convergence, the additional instantaneous transfer structure is used to optimize the algorithm. It is found that FLANN structure has the advantages of simple structure and small amount of calculation, and is mainly used in the case of weak nonlinear degree. However, Volterra filter has strong nonlinear processing ability, but the number and computation of kernel function increase with the length of input signal. Therefore, the two filters are introduced into the active noise control system after convex combination, in order to deduce the combined structure and algorithm according to their advantages in nonlinear processing ability and computational complexity. In order to reduce the computational complexity of the algorithm, the modified skip tongue line function is used to select the joint coefficient, and the sign function is used to update the mixing coefficient. While the computation is reduced, the phenomenon that the joint parameter tends to 0 or 1 and the mixing coefficient updates slowly or even stagnates can be avoided. Finally, a large number of simulation experiments are used to verify the research content of this paper.
【学位授予单位】：武汉大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM41

【参考文献】