In order to fully exploit the response potential of electric vehicles (EVs) as well as to improve the profitability of EV aggregators (EVA), we construct an dynamic incentive price optimization model to improve EVA’ adjustable response capacity. Current research on incentive price optimization mainly focuses on the static market-side revenue, leaving a gap in the study of dynamic pricing under user-side response uncertainty. To characterize the response uncertainty of different EV users, this paper divides EVs entry into three modes, corresponding to different compensation prices with battery losses. Then, an adjustable capacity portraying method of EVA under different modes is proposed and analyzed based on the optimal working conditions. The evaluation indexes of the aggregation results are proposed based on the TRN-AHP method. The modeled is solved by a proposed data-driven evolutionary algorithm based on semi-supervised learning. The solution algorithm adopts tri-training radial basis function network fitting instead of the aggregation model, which overcomes the difficulty of obtaining users’ private information while avoiding the prediction error caused by the sparse dataset. The solution accuracy is consequently improved by expanding the optimal range of the incentive price based on the Snow ablation optimizer (SAO). The simulations on the actual charging/discharging prices verify the accuracy and feasibility of the proposed algorithm, providing a theoretical decision basis for the EVA to improve the adjustable capacity by dynamically setting the incentive price.