Intelligent Optimization Design Of LCL-Type Three-phase Grid-Connected Converter

High-power grid connected converter(GCC) is widely used in the smart grid systems such as the photovoltaic(pv), fan power generation, fuel cell, the energy storage power station and so on. The optimization design of main circuit parameters and current loop performance has an important theoretical and practical engineering value. This thesis will discuss the above problem of three-phase GCC based on LCL filter.The mathematical model of LCL-type GCC is established and its completely decoupled state space expression under the rotating coordinate can be obtained because the new type complex vector operator is put forward. Using a complex vector PI current regulator the voltage coupling problem introduced by the coordinate transformation can be eliminated completely. In order to solve the stability deterioration problem due to the LCL filter inherent resonance peak, the delay effect brought by the digital control is used in this paper. The discrete system can be stable by changing the phase frequency characteristic curve so as to the system has a certain margin of amplitude and phase. In addition, the optimization resonance scope can be derived according to the frequency domain stability criterion. Within this range, the reasonable parameters of LCL filter can be designed using the LCL filter design criterion. In order to improve the dynamic response performance and the parameters robustness, a neural network adaptive PI controller is designed because of its ability of the infinite approximation nonlinear system. The BP neural network estimates the parameters of the PI controller online according to the external environment. This novel controller can improve the response performance under different operation conditions.A simulation model of LCL-type GCC based on the Saber platform is designed to validate the correctness of the theoretical analysis about undamping control; meanwhile, the reasonable of parameters design and the feasibility of the neural network algorithm are verified by the half-physical model simulation.