基于机器学习的模块化多电平换流器的模型预测控制研究  

Research on Model Predictive Control of Modular Multilevel Converter Based on Machine Learning

作　　者：余瑜[1] 杨文康 徐岸非 汪健 YU Yu;YANG Wenkang;XU Anfei;WANG Jian(Hubei Collaborative Innovation Center for High-efficiency Utilization of Solar Energy,School of Electrical and Electronic Engineering,Hubei University of Technology,Wuhan 430068,China)

机构地区：[1]湖北工业大学电气与电子工程学院、太阳能高效利用及储能运行控制湖北省重点实验室,湖北武汉430068

出　　处：《控制工程》2023年第10期1775-1785,共11页Control Engineering of China

基　　金：湖北省技术创新重大项目(2019AAA018);襄阳湖北工业大学产业研究院重点资助项目(XYYJ2022C11)。

摘　　要：模型预测控制器(model predictive controller,MPC)因可实现多目标优化控制被广泛应用在模块化多电平换流器(modular multilevel converter,MMC)领域,但随着子模块数量增加,MPC的在线计算量成几何级数增长。因此,提出了基于机器学习的MMC的模型预测控制方法,首先利用MPC-MMC仿真平台收集数据并对数据进行预处理,再进行神经网络训练得到神经网络-MPC(neural network-MPC,NN-MPC)。为了提高神经网络训练效率,采用随机森林来优化神经网络的初始权值和阈值,得到随机森林-神经网络-MPC(random forest-neural network-MPC,RF-NN-MPC),将其用来模拟MPC。仿真结果表明,RF-NN-MPC在学习效率和学习精度方面都优于NN-MPC,在保持了良好的控制效果的同时,使MPC-MMC不受子模块数量约束,在线计算量始终为1次。Model predictive controller(MPC)is widely used in the field of modular multilevel converter(MMC)due to its ability to achieve multi-objective optimal control.But with the increase in the number of sub-modules,the on-line computation of MPC increases geometrically.Therefore,a model predictive control method for MMC based on machine learning is proposed.Firstly,the MPC-MMC simulation platform is used to collect data and the data are preprocessed.Then,neural network is trained to obtain the neural network-MPC(NN-MPC).In order to improve the efficiency of neural network training,random forest is used to optimize the initial weights and thresholds of the neural network to obtain random forest-neural network-MPC(RF-NN-MPC),which is used to simulate MPC.Simulation results show that RF-NN-MPC is better than NN-MPC in terms of learning efficiency and learning accuracy,and while maintaining a good control effect,MPC-MMC is not constrained by the number of sub-modules,and the online calculation amount is always one time.

关 键 词：神经网络 随机森林 模型预测控制 模块化多电平换流器 变分模态分解 

分 类 号：TM464[电气工程—电器] TM712