To address the capacity allocation challenge of energy storage-assisted thermal power units in Automatic Generation Control (AGC) load variation scenarios, this study first analyzes one-week AGC load variation data of a 660MW-rated thermal power unit across spring, summer, autumn, and winter seasons. The characteristics of AGC command magnitude, direction, and interval duration are systematically investigated. A novel capacity allocation methodology is proposed, incorporating both individual/short-interval continuous step changes of AGC commands and the self-recovery characteristics of energy storage State of Charge (SOC), along with an auxiliary load regulation control strategy. Through case studies using typical daily operational data with an Improved Particle Swarm Optimization (IPSO) algorithm, the results demonstrate that configuring a 20.082MW/2.75MWh power-type flywheel energy storage system under 2.45%Pe ramping rate achieves maximum return rate of 0.92. The profitability increases proportionally with frequency regulation price and mileage, exceeding 1 when the price reaches 11 CNY/MW or mileage attains 2800MW. This research provides valuable references for capacity allocation in diverse power unit types.