Research On Control For Single Phase Inverter Based On Neural Network

Single phase inverter source is mainly used for uninterruptible power supply, airborne power, locomotive auxiliary power, new energy power generation and other applications. Many factors effect the performance of single-phase inverter output voltage, while non-linear load is the main reason causing the increase of output voltage harmonic components. The thesis focuses on single phase inverter power supply problems with nonlinear loads, making a detailed study of control strategy for single-phase inverter power.First, this thesis analyzes the flaws of the single closed-loop PI control strategy:PI controller can not perfectly track the given sinusoidal signal and the steady-state error of the amplitude and phase angle can not be eliminated. Focusing on these problems, this thesis puts forward a control strategy of the single-phase inverter based on the Hopfield neural network.Second, from the perspective of energy function of Hopfield neural network, the thesis uses CHNN for optimizing calculation. On this basis, instantaneous harmonic estimation that takes advantage of HNN estimating online each harmonic from non-sinusoidal periodic signals is proposed.Third, this thesis extracts fundamental wave from output voltage enriched with harmonics using instantaneous harmonic estimation. And fundamental AC components are transformed to DC components by amplitude and phase decomposition. Then the PI control can eliminate the amplitude and phase error between reference fundamental wave and output fundamental wave. Finally, amplitude and phase control quantity is transformed to fundamental component by amplitude-phase synthesis. Thus, the fundamental voltage closed-loop tracking control of zero steady state error is realized by amplitude and phase decomposition. For the distortion in the output voltage caused by the harmonics in the filter inductor, this thesis proposes using neural network model of harmonic detection to obtain voltage drop in the filter inductor and superimpose on the fundamental control component as feed forward to eliminate harmonic voltage drop in filtering inductance. All the harmonics can be assumed to be equivalent to external noise signal. If a control loop could be built around the noise signal, with a high feed back gain for the noise alone, the closed loop gain for the noise becomes smaller to realize the harmonic suppression. The simulation model of control strategy built in Matlab/Simulink can verify the validity of algorithm.Finally, the main power circuit, control board circuit and monitoring unit circuit are introduced, while describing the control algorithm of the software design. The experiment proves the validity of single-phase inverter control strategy based on neural network model of harmonic detection proposed in this thesis.