As the core component in air supply system, the air compressor plays a key role in efficient operation of fuel cell system. In this research, accelerated endurance tests for air compressor with air bearings were conducted based on designed operating condition, and the Hybrid Variational Artificial Bee Colony (HyVABC) algorithm was proposed to achieve enhancement and balance of global optimization capability and local feature capture ability. On this basis, combinatorial prediction models were developed to realize accurate performance prediction of the compressor considering accelerated decay characteristics. The results indicate that the degradation degree of pressure ratio is the greatest at the flow rate of 133 g/s and the speed of 100000 rpm, with the degradation rate of 25.9% compared with that before break-in. The HyVABC algorithm obtains optimal values on 16 out of 20 test problems, which highlights its strong optimization performance, fast convergence speed and high robustness. The XGBoost-HyVABC model is chosen to describe mapping relationship between operating and performance parameters of the compressor, and absolute values of relative errors between predicted and test values are below 3%. All these provide efficient heuristic algorithm and accurate combinatorial model for accelerated decay performance prediction of fuel cell air compressors.