Research On Data-Driven Grid-Connected Photovoltaic System Modeling Method

Recently,photovoltaic(PV)power generations have been widely installed around the world.With the increase of its penetration into the grid,challenges were brought to the stable operation of the power grid.The modelling of grid-connected PV system is the foundation to study its interaction with the power grid and other power electronic interface devices.In fact,during the planning and operation of power systems,it is necessary to model and simulate PV systems in order to study their characteristics and potential impact on the grid.As electric utilities do not usually develop their own PV models,instead,they tend to rely on proprietary or generalized models provided by manufacturers.While for the sake of confidentiality,the model provided by the manufacturers is information-limited.Besides,the controlling parameters of the models were often inaccurate,which ought to be adjusted according to the actual scenarios.These lead to the lack of available models in practical situations.Therefore,it is urgent to establish reliable and accurate PV models.To tackle the above problems,a 3kW commercial single-phase grid-connected photovoltaic inverter was taken as a modelling example.The modelling process was carried out without knowing its internal structure,control strategy or parameters.More details are as follows:1.For the PV system operating at unity power factor mode,solar irradiation,temperature,active power reference and grid voltage were taken as the inputs of the model,the AC current was taken as the output of the model.Transfer function was adopted as the model structure,which was identified through system identification.2.For the PV system operating at the Reactive Power/Voltage(Var/Volt)mode,the grid voltage was taken as the input,the AC current was taken as the output of the model.Neural networks were taken as the model structure.3.A RTDS based Power-Hardware-In-the-Loop(PHIL)experiment platform of gridconnected PV system was set up to conduct the modelling process,as well as validating the models.The results showed the developed models were of high accuracy,thus,these modeling processes were validated.As this modelling method requires no internal knowledge of the system undertest,it can be of reference when modelling other power electronic devices.