Data-driven Modeling Method For Photovoltaic Grid Connected Inverters

With the development of human society,energy crisis and environmental problems are one of the major challenges facing the world today.The ”carbon peaking” and ”carbon neutral” strategies require an increase in the proportion of new energy in the energy mix,and clean energy,with a focus on photovoltaic power generation,will gradually take a dominant position.The increased penetration of new energy will lead to further uncertainty between the source and load of the grid.As the key equipment for grid-connected PV power generation system,grid-connected PV inverter plays a crucial role in the performance and stable operation of PV power generation system.In this paper,the circuit parameter identification and black-box modeling of PV grid-connected inverters are investigated based on mechanistic modeling and data-driven modeling theory.First,the mechanistic modeling method and data-driven modeling method of gridconnected inverters are introduced in detail.In terms of basic theory,the selection method of inverter circuit parameters,maximum power point tracking technology,inverter gridconnected control strategy and grid-connected requirements are highlighted.Based on the basic theory of grid-connected PV inverter,a simulation model of single-phase,two-stage PV power generation system is built to support the subsequent experiments.Then,a mechanism-based modeling and data-driven method for circuit parameter identification of single-phase LCL-type grid-connected inverters is proposed based on a single-phase LCL grid-connected inverter.Based on the simulation model of a singlephase two-stage PV power generation system to collect the necessary data for the experiment,the relationship between the circuit parameters and the adaptation function of the optimization algorithm is established by building the harmonic state space model of the LCL filter of the single-phase grid-connected inverter and solving the harmonic state space equation under the steady state condition.The identification results of IWOA are compared with those of WOA and PSO algorithms.Although the initial convergence speed of IWOA is not significantly improved compared with that of WOA,the final identification accuracy is higher than that of WOA and PSO algorithms.The parameter identification results are substituted into the two-stage PV power generation system simulation model,and the effect of parameter errors on the inverter input and output is illustrated by comparing the simulation results of the original parameters.Finally,a PV grid-connected inverter modeling method based on LSTM and migration learning is proposed for single-phase PV grid-connected inverters and three-phase PV grid-connected inverters.A three-phase two-stage PV power generation system simulation model is built,and the LSTM model is used instead of the nonlinear module in the traditional nonlinear system identification.The two model structures are determined by analyzing the input and output characteristics of the inverters,and the equivalent dynamic models of the single-phase and three-phase grid-connected PV inverters are established.The results show that the models can well fit the output current characteristics of the gridconnected PV inverter,which proves the effectiveness of this modeling method.Based on the migration learning theory,the three-phase and single-phase models are migrated to each other,and the experimental results illustrate the ability to obtain a highly accurate target domain model based on the inverter model trained in the source domain with a small amount of target domain data for training.