摘要
针对电力电子变换器采用有限集模型预测控制(MPC)时,其成本函数中权重系数难以整定的问题,设计了一种基于人工神经网络(ANN)的新型MPC权重系数整定方案。基于仿真平台搭建了变换器电路仿真模型并代入不同权重系数进行测试以获取对应的如总谐波失真等关键性能指标,然后利用这些数据对ANN进行训练,使得ANN具备为任意权重系数组合快速准确估计性能指标的能力。从而,对于任意结合输出指标的用户自定义适应度函数,可快速准确地找到权重系数优化组合。利用不间断电源系统开展了对MPC权重系数设计的ANN方法的实验,实验结果为对于指定的示例性适应度函数,采用AAN设计权重系数后可产生预期的控制性能,并对负载变化具有一定的鲁棒性,同时与仿真模型的误差小于3%,与实测误差小于10%。实验结果验证了所设计的ANN方法整定电力电子变换器MPC权重系数的有效性。
Aiming at the problem that selection of the weighting factors in cost function is difficult when the power electronic converter adopts the finite set model predictive control(MPC),a novel MPC weighting factors adjustment scheme based on artificial neural network(ANN)is designed.The converter circuit simulation model is built based on the simulation platform,then different weighting factors are substituted into i for testing to obtain corresponding key performance metrics such as total harmonic distortion.This data is then used to the ANN,so that the ANN has the ability to provide fast and accurate estimates of the performance metrics for any combination of weighting factors.Therefore,for any user-defined fitness function that combines output metrics,the combination of weighting factors that optimize the given function can be quickly and accurately found.The uninterruptible power supply system was used to carry out the experiments of the ANN-based method for MPC weighting factors design.The experimental result is that for the specified exemplary fitness functions,the AAN design weighting factors can produce the expected control performance and robust to load variations,and the error with the simulation model is less than 3%,and the error with the actual measurement is less than 10%.So,the effectiveness of the designed ANN method to set the MPC weight coefficients for power electronic converters is verified by test results.
作者
毕长飞
BI Changfei(Liaoning Geology Engineering Vocational College,Liaoning Dandong 118008,China)
出处
《机械设计与制造》
北大核心
2025年第9期221-224,共4页
Machinery Design & Manufacture
基金
2022年度教育厅基本科研项目(面上项目)(LJKMZ20222185)。
关键词
模型预测控制
电力电子变换器
成本函数
权重系数
人工神经网络
Model Predictive Control
Power Electronic Converter
Cost Function
Weighting Factor
Artificial Neural Network
作者简介
毕长飞(1978-),女,辽宁丹东人,研究生,教授,主要研究方向:机电一体化、机械制造。
