基于概率补偿模糊神经网络的MMC-APF建模与控制研究

模块化多电平有源电力滤波器(MMC-APF)不仅可有效抑制谐波、改善电能质量，且满足高压大容量需求，具有广阔的应用前景。目前制约其发展的因素包括结构复杂难以精确建模、负荷扰动变化大恶化控制系统性能、传感器数量多降低系统可靠性等。本项目围绕MMC-APF数学模型构建、抗干扰电流控制和无传感器控制等方面展开研究，实现系统低建模误差、强动态性能及高可靠性。具体研究内容为：(1)构造概率补偿模糊神经网络用于MMC-APF数据驱动建模；(2)集成Hermit激活函数设计基于递归概率补偿模糊神经网络的有限时间连续电流控制器；(3)开发递归概率补偿模糊神经网络无传感器控制算法；(4)搭建实验平台并进行算法验证。本项目受认知行为启发，融合概率方法和补偿机制为MMC-APF建模与控制这一前沿科学问题提供独特的解决方案，不仅可丰富和发展模糊神经网络基础理论，且对智能电网领域的技术提升将有积极的现实意义。

Modular multilevel converter based active power filter (MMC-APF) can not only effectively suppress harmonics and improve power quality, but also meet the demand of high voltage and large capacity. However, it is difficult to obtain the accurate model for MMC-APF due to its complex structure, load disturbance changes greatly deteriorating control system performance, and the large number of sensors reduces the reliability of the system, which restricts its development. To achieve low modeling error, strong dynamic performance and high reliability, this project focuses on the construction of the complex mathematical model for MMC-APF, the design of anti-interference current control and sensorless control algorithm. The main research contents are listed as follows: (1) a probabilistic compensatory fuzzy neural network is constructed and applied to the data-driven modeling of MMC-APF; (2) integrating Hermit activation function, a finite time continuous current controller based on recurrent probabilistic compensatory fuzzy neural network is developed; (3) a sensorless control algorithm based on recurrent probabilistic compensatory fuzzy neural network is presented;(4) the experimental platform is built to verify the proposed algorithms. Inspired by the cognitive behavior, this project combines the probabilistic method and compensatory mechanism to provide a unique solution for the modeling and control of MMC-APF, which not only develops fuzzy neural network theory, but also possesses significance for the development of smart gird.