PPO CVaR Multisource Fusion Model for Power Allocation in Hybrid Energy Storage Systems

Hybrid energy storage systems must divide power between energy type and power type devices while renewable output, load demand, electricity price, and converter limits change at different time scales. This paper develops a PPO CVaR multisource fusion model for risk aware power allocation between lithium-ion batteries and supercapacitors in a grid connected renewable microgrid. The method fuses photovoltaic power, wind power, load, price, state of charge, temperature, and ramp indicators into a compact decision state, and then trains a proximal policy optimization agent with a conditional value at risk penalty for high loss operating tails. Experiments are conducted on a 96-interval daily scheduling dataset with four seasonal profiles and six operating stress cases. Compared with rule-based filtering, deterministic PPO, and DDPG allocation, the proposed model reduces mean tracking error from 4.82 kW to 2.93 kW, lowers the 95 percent tail loss by 31.6 percent, and decreases equivalent battery throughput by 18.4 percent while keeping supercapacitor utilization within the rated transient band. The best case achieves 97.8 percent state of charge feasibility and limits grid exchange variance to 7.4 kW squared under fast renewable ramps. The results show that CVaR guided reinforcement learning can improve allocation reliability when multisource information is available and uncertainty is concentrated in short high impact intervals.