Lifelong Learning for Complementary Generation Control of Interconnected Power Grids With High-Penetration Renewables and EVs

This paper proposes a lifelong learning (LL) based complementary generation control (CGC) of interconnected power grids with high-penetration renewable energy sources and electric vehicles (EVs). The wind farms (WFs), photovoltaic stations (PVs), and EVs are aggregated as a wide-area virtual power plant (WVPP) for automatic generation control (AGC), which can significantly accelerate the system response and reduce the regulation costs of balancing unexpected power mismatches between generation side and demand side. Under such framework, CGC is decomposed into a multi-layer generation command dispatch to rapidly compute an optimal solution. LL is first employed for the primary layer CGC between conventional power plants and a WVPP. Then, the secondary layer is implemented according to the ascending order of regulation costs of all reserve sources. Finally, the tertiary layer is accomplished by a coordinated control in each WFs or PVs, whereas an online optimization of EVs is adopted by considering the charging demands. The imitation learning is introduced to improve the learning efficiency of agents with transfer learning, thus an online optimization of CGC can be satisfied. Case studies are carried out to evaluate the performance of LL for multi-layer CGC of AGC on a practical Hainan power grid of southern China.