Integrating physical and data-driven system frequency response modelling for wind-PV-thermal power systems

This paper presents an integrated system frequency response (SFR) modelling method for wind-PV-thermal power systems (WPTPSs) by combining both physical model-based and data-driven modelling methods. The SFR physical model is built and simplified by the balanced truncation (BT) method. Based on the physical model, an improved radial basis function neural networks (RBFNNs) is then employed to establish an off-line SFR model using source data. Following the transfer learning principle, the transferred data from the source data set is determined by the maximum mean discrepancy (MMD) criterion. The RBFNN-based SFR model is then fine-tuned using both the transferred source data and target data. Finally, the fine-tuned RBFNNs is applied to investigate real-time SFR of WPTPSs. Simulation results confirm the effectiveness of the proposed SFR modelling strategy with an illustrative WPTPS.