基于人工神经网络的电源变换器大信号建模
Large Signal Modeling of Power Converters Based on Artificial Neural Networks
摘要
针对传统基于小信号模型的机理建模方法应用于加速器磁铁电源时面临的参数不准确性以及开关电路非线性问题,为提高实际应用中对电源变换器的建模精度与预测能力,本文提出一种适用于加速器磁铁稳流电源的基于人工神经网络(artificial neural networks,ANN)的数据驱动建模方法.首先,在Matlab/Simu-link中搭建电路模型,并设计多种形式的输入电压与占空比激励信号,采集系统在不同激励工况下的输入输出数据,来构建带有历史延迟结构的输入特征集.其次,通过Levenberg-Marquardt算法整定网络权值,并利用与训练集不相关的测试集完成了模型验证.最后,在一台电源上验证了该建模方法在大信号、非线性系统建模中的有效性.
To address the issues of parameter inaccuracy and nonlinearity in switch-mode circuits when applying tradi-tional small-signal model-based mechanistic modeling methods to accelerator magnet power supplies,a data-driven mod-eling method is proposed based on artificial neural networks(ANN)for accelerator magnet constant-current power sup-plies.This method aims to improve the modeling accuracy and predictive capability of power converters in practical ap-plications.Firstly,a circuit model is established in Matlab/Simulink,and various forms of input voltage and duty cycle excitation signals are designed.Input-output data under different excitation conditions are collected to construct an input feature set with a historical delay structure.Next,the network weights are tuned using the Levenberg-Marquardt algo-rithm,and the model is validated using a test set unrelated to the training set.Finally,the effectiveness of the proposed modeling method in large-signal and nonlinear system modeling is verified on a power supply.
年,卷(期):
2026, 60(10)
分类号:
TM46
机标分类号:
TN702
TP183
TN302
在线出版日期:
2026年10月26日