With the alteration of energy structures in power system, the allocation of automatic generation control (AGC) is facing new challenges. The emergence of high penetration of manufacturing sectors and renewable energy sources has increased the demand for faster-ramping resources to participate in the frequency regulation service. Additionally, the current regulation service does not properly arrange the output of the resources considering the actual performance while they follow the AGC allocation signals, which affects the accuracy of frequency regulation. The fast-ramping capacity and response accuracy of AGC units are supposed to be considered in the dispatch. Meanwhile, the power outputs of different governors have different impacts on the system frequency, which has important guiding significance for the AGC dispatch. With the purpose of improving frequency regulation service, this paper proposes a dynamic performance-based dispatch model considering the above issues. We first prove that there is a linear relation between the output of the generators and system frequency, defined as frequency sensitivity. Then, the frequency sensitivity of each generator can be identified using the least square (LS) method. Furthermore, a dynamic multi-objective optimization allocation model is established, which considers the units’ economy, ramping capacity and accuracy. Finally, the proposed identification method and allocation model are simulated in the IEEE-9 bus system, and the simulation results verify their validity and feasibility