Abstract: To reduce the capacity redundancy caused by constant-power peak shaving, a composite load-stability index is established and an envelope-tracking variable-power control method is proposed. A dynamically adaptive particle swarm optimization algorithm coordinates the power trajectory and energy capacity. For a measured daily load profile, the required rated capacities in two operating scenarios are reduced by 29.57% and 29.87%, respectively, while all power, state-of-charge and energy-balance constraints are satisfied.