Abstract: To address the issues of large response delay and large overshoot when thermal power units follow automatic generation control (AGC) commands, a capacity optimal allocation method for hybrid energy storage systems based on robust empirical mode decomposition (REMD) was proposed. First, the power required by the hybrid energy storage system was obtained from the power curve of the thermal power unit following the AGC command. Then, the required power was decomposed by the REMD method to obtain power decomposition curves at different frequencies, and power allocation was conducted considering the operating characteristics of two different energy storage devices, namely lithium iron phosphate batteries and supercapacitors. Considering constraints such as the state of charge, capacity, and charge-discharge power of the energy storage system, a capacity optimal allocation model aiming at minimizing the total cost was developed. Joint optimization of the power decomposition results and the capacity allocation model was performed to obtain the optimal energy storage allocation scheme. The results of engineering case analysis show that the proposed method can effectively improve the delayed power response of thermal power units following AGC commands, and enhance the power supply reliability and economic benefits of the units.