Multi-scenario control of photovoltaic-solid oxide fuel cell hybrid energy system based on deep reinforcement learning technology

Abstract：Solid oxide fuel cell ( SOFC ) is quiet, clean and efficient, and has broad application prospects. SOFC can be flexibly coupled with other power generation devices and energy storage devices to form a hybrid energy system. The photovoltaic-solidoxidefuelcell ( PV-SOFC ) hybrid energy system studied in this paper converts solar energy into hydrogen energy for SOFC through power-to-gas ( P2G ) technology, which can effectively compensate for the shortcomings of large fluctuation and strong intermittency of solar energy. Aiming at the characteristics of multivariable and strong nonlinearity of SOFC hybrid energy system, this paper uses deep reinforcement learning ( DRL ) technology to study an intelligent operation control strategy. Firstly, the simulation model of PV-SOFC hybrid energy system is built. The model takes SOFC as the core, focuses on the power flow and hydrogen flow in the system, and considers the constraints of operating conditions within the system. Secondly, the system intelligent operation control strategy based on deep deterministic policy gradient ( DDPG ) is proposed in the networked operation mode and the island operation mode respectively. Finally, the effectiveness of the algorithm is verified by simulation. The results show that the DDPG algorithm in island operation mode can keep the hybrid energy system with low probability of power failure and abandonment under different environmental and load demand conditions, and effectively improve the operation reliability and economy of the system. In the networked operation mode, the DDPG algorithm can use the price difference and residual hydrogen storage in different time periods to trade flexibly with the power grid, so as to obtain higher long-term benefits of the system.