基于神经网络的逆变器约束自适应PI控制

CONSTRAINED ADAPTIVE PI CONTROL OF INVERTER BASED ON NEURAL NETWORK

针对负载扰动以及输入输出约束条件下逆变器的电压跟踪控制问题,根据同步旋转(dq)坐标系下的电压电流时域数学模型,提出一种基于神经网络的约束自适应PI控制策略。首先,利用状态转换函数将带约束的系统转换为等价的无约束系统,同时引入动态辅助系统处理输入约束问题;其次,通过反步法进行控制器设计,并为解决传统反步法“微分爆炸”问题,运用神经网络逼近包括虚拟控制律导数在内的非线性项;最后,所设计的虚拟控制律和控制律由PI项和前馈项组成,且PI控制器的增益可在线调整;理论分析表明,所提控制策略能保证闭环系统的所有误差信号最终一致有界,Matlab/Simulink平台仿真结果则进一步验证了该方案的可行性和有效性。

According to the voltage and current time domain model in the synchronous rotating dq reference frame,a constrained adaptive PI control strategy is proposed to address the inverter voltage tracking problem with input and output constraints under load disturbance.Firstly,a state translation function is used to transform the system with output constraints into an equivalent unconstrained one.At the same time,an auxiliary dynamic system is introduced to tackle the effects of the input constraints.Secondly,the controller is designed by using back-stepping method.Wherein,neural network is utilized to approximate the nonlinear term including the time derivative of the virtual control input,and then“explosion of complexity”in the traditional backstepping method is avoided.Finally,the designed virtual control law and control law are both composed of PI term and feedforward term,and the gain of the PI controller is continuously adjusted along with the adaptive law.The Lyapunov stability theory is used to prove that all the error signals of the closed loop system are uniformly ultimately bounded under the proposed control strategy.Besides,simulation results in Matlab/Simulink platform demonstrate the feasibility and effectiveness of this method.