Abstract: To address the deterioration of coordinated control quality of ultra-supercritical units caused by high-frequency flipping load commands in AGC-R mode, this paper proposes a coordinated control optimization strategy based on multi-mode decomposition of command signals, in which a scale-to-channel mapping criterion is established. After analyzing the evolution of AGC dispatching modes and extracting the statistical characteristics of the commands, the variational mode decomposition is employed to identify their flipping-scale and trend-scale structures, based on which the scale-to-channel mapping criterion is established according to the responsiveness of the milling-combustion-heat transfer process: the trend scale feeds the feedforward baseline, the flipping scale is attenuated, and the residual scale enters a flexible dead zone. Accordingly, three improvement blocks are designed: online identification of R mode with S-curve trajectory planning to soften the flipping commands, a double-inertia differential feedforward to extract the trend rate and attenuate flipping excitation, and a water-to-fuel incremental feedforward with four-quadrant fuzzy gain to suppress temperature-pressure coupling disturbance. Simulation results driven by real DCS data of an ultra-supercritical unit under AGC-R conditions show that, compared with conventional coordinated control, the maximum deviation of main steam pressure is reduced by 18.4%, the maximum deviation of power by 12.0%, and the high-frequency feedforward oscillation is attenuated by 83.5%, while the maximum deviation of the intermediate point temperature increases from 11.60 °C to 15.06 °C as a trade-off for the pressure and power improvement and remains below the field measured level, which verifies the effectiveness and engineering applicability of the proposed strategy.