基于RBF神经网络PI算法的T型三电平调光电源控制策略

为提高单相T型三电平助航调光电源的动态性能以及抑制输出电流谐波,提出基于RBF(RBF-RadialBasis Function)神经网络和梯度下降法相结合的新型控制策略。首先对T型三电平拓扑进行建模分析,利用根轨迹方法研究系统的动态特性,揭示了影响输出响应的关键因素。在此基础上,研究设计利用局部逼近的RBF神经网络来快速辨识输出相对于输入的变化率(Jacobian信息),并用梯度下降法动态在线调整T型三电平逆变器的PI控制参数,自适应调光电源工况变化。最后,针对系统调光、空载、短路等工况下的控制效果,采用MATLAB/Simulink软件与常规PI控制进行对比仿真,并搭建30 kW调光电源实验平台进行实验。结果表明所研究方法能有效改善分级调光动态性能和输出电流谐波。

Control strategy of T-type three level dimming power supply based on RBF PI algorithms

In order to improve the dynamic performance and suppress the harmonic of a single-phase three-level T-type inverter in navigation dimming power supply, a new method is presented based on the Radial Basis Function (RBF) neural networks combined with the gradient descent method. First, the mathematical model and analysis of three-level T-type inverter topology is established. The dynamic characteristics of the system are researched in detail using the root track method, and the key parameters which influence the output responses are revealed. On this basis, a new control technique is proposed, one which uses RBF neural networks to identify the Jacobian information, and dynamically adjust the control parameters by the gradient descent method. Finally, a comparative simulation analysis is carried out for level dimming, open circuit load, short circuit load based on MATLAB/Simulink, and experiments are carried out based on the experimental platform of a 30 kW dimming power supply. The results show that the proposed method can effectively improve dynamic performance and output current harmonic. This work is supported by Major Equipment Development Project of Sichuan Provincial Economic and Information Committee (No. 2018CD00113) and the Sichuan Provincial Science and Technology Department (No. 2019JDKP0049).