Flexible ramping product (FRP) trading has emerged as a highly effective solution to cope with the volatility and uncertainty introduced by the increasing integration of renewable energy sources. This paper proposes a bidding method for electric vehicle aggregators (EVAs) in the FRP trading market. To effectively articulate the spatiotemporal operational characteristics intrinsic to EVAs, a charging and swapping flexibility aggregation model is formulated. The model is developed by accurately simulating the charging and swapping demands of plug-in electric vehicles and battery-swapping electric vehicles in different charging modes. A novel bilevel optimization model is developed to address the conflicting objectives in the FRP trading market between the EVAs and electric vehicles (EVs), aiming to optimize the incentive prices and charging strategies. The upper level optimizes the bidding profits of EVAs, whereas the lower level models the EV charging behavior using the charging and swapping flexibility aggregation model. To solve the high computational complexity of the high-dimensional nonconvex optimization problem owing to the vast number of EVs, a data-driven evolutionary algorithm incorporated with a zebra optimization algorithm is adopted. Owing to the limited data available for training high-quality agent models in real scenarios, a semi-supervised learning-based tri-training algorithm is adopted to enhance the efficiency of data utilization. Case studies validate the effectiveness of the proposed method.