| 引用本文: | 马草原,孙富华,朱蓓蓓,等.神经网络算法的改进及其在有源电力滤波器中的应用[J].电力系统保护与控制,2015,43(24):142-148. |
| MA Caoyuan,SUN Fuhua,ZHU Beibei,et al.Study on algorithm improvement of BP neural networks and its application in active power filter[J].Power System Protection and Control,2015,43(24):142-148 |
|
| 摘要: |
| 针对有源电力滤波器的电流跟踪控制问题，设计了一种基于改进梯度算法的BP神经网络自适应PI控制器。该控制器将神经网络技术与PI参数设计相结合，与传统的PI控制器相比，该控制器具有结构简单、易于在线调整等优点。同时，为了克服采用神经网络算法修正权值系数时，会存在局部极小、收敛速度慢的问题，对BP神经网络采用的梯度算法进行改进。利用代数法代替梯度下降法，从而解决了易出现局部极小问题，且使收敛速度更快。仿真实验表明，改进后的神经网络自适应PI控制器较传统的PI控制器有更快的响应速度和更高的补偿精度，从而使系统更稳定，而且电网电流的谐波畸变率更低。 |
| 关键词: 有源电力滤波器 电流跟踪控制 BP神经网络 代数算法 梯度算法 |
| DOI：10.7667/j.issn.1674-3415.2015.24.022 |
| 分类号: |
| 基金项目:江苏省自然科学基金项目(BK20130187) |
|
| Study on algorithm improvement of BP neural networks and its application in active power filter |
|
MA Caoyuan, SUN Fuhua, ZHU Beibei, YIN Zhichao
|
|
School of Information and Electrical Engineering, China University of Mining & Technology, Xuzhou 221008, China
|
| Abstract: |
| For current tracking control problems in active power filter (APF), a BP neural network adaptive PI controller based on improved gradient algorithm is designed. It combines the neural network technology with PI controller structure. Compared with the traditional PI controller, it has a simple structure, and easy to on-line adjustment. Meanwhile, in order to overcome the local minimum and slow convergence problem when using neural network algorithm to weight correction coefficient, the gradient algorithm is improved and the algebraic method instead of gradient descent method is used to solve the problem of the local minimum arise, and makes convergence faster. Simulation experiments show that the improved adaptive neural network PI controller has faster response and higher compensation accuracy, thus to make the system more stable, and the harmonic distortion of grid current is lower. |
| Key words: active power filter current tracking control BP neural network algebraic algorithm gradient algorithm |
