A Hybrid Method for Electric Spring Control based on Data and Knowledge Integration

Electric spring (ES) is a relatively new power electronics device for voltage regulation in a distribution system. Some existing ES control methods are system-independent; however, their performances generally can be significantly improved when accurate system models are properly utilized. Unfortunately, accurate model of the concerned distribution system is not always available in practice. This paper proposes a hybrid method for ES control, based on the integration of data-driven and analytical models. The inaccurate analytical model provides a basic policy to generate system data. Three data-driven models based on the extreme learning machine (ELM) are built using these data to replace the power flow, active power, and reactive power models respectively. An ELM-based control model is proposed to generate the final control strategies. The hybrid method is tested using a 15-node distribution network. Simulation results show that the data-driven model is more accurate than the analytical model in predicting system states, while the proposed control method outperforms the original analytical method.