The high penetration of wind energy sources in power systems has substantially increased the demand for faster-ramping thermal units participating in the frequency regulation service. To fulfill the automatic generation control (AGC) and compensate the influence of wind power fluctuations simultaneously, ramping capacity should be considered in the dispatch model of thermals. Meanwhile, conventional methods in this area do not take the impact of transmission loss into the dispatch model, or rely on offline network model and parameters, failing to reflect the real relationships between the wind farms and thermal generators. This paper proposes an online approach for AGC dispatch units considering the above issues. Firstly, the power loss sensitivity is online identified using recursive least square method based on the real-time data of phasor measurement units. It sets up power balance constraint and results in a more accurate dispatch model. Then, an improved multi-objective optimization model of dispatch is proposed and a connection is established between the thermal units with fast ramping capacity and the wind farms with rapid fluctuations. Genetic algorithm is used to solve the dispatch model. The proposed method is compared with conventional methods in simulation case in the IEEE 30-bus system. Finally, simulation results verify the validity and the feasibility of identification method and optimization model.