In this paper, an energy scheduling problem is formulated for the prosumer-based urban area, where prosumers are regarded as the drone charging stations for urban air mobility (UAM). Particularly, since electric vertical take-off and landing aircraft (eVTOL) is regarded as the anticipated technique for future UAM, we consider eVTOL drone taxis for transporting passengers. The objective is to minimize the overall energy supply-demand imbalance cost. This problem covers two aspects: 1) association between passengers and eVTOLs, and 2) energy balance strategy determination through power grid energy scheduling for each prosumer. For the first aspect, a destination collision-aware Gale-Shapely matching game (DC-MG) approach is proposed, where the distance concern of passengers, the remaining energy of eVTOLs, and the destination collision are comprehensively considered. Subsequently, hierarchical agglomerative clustering (HAGC)-based multi-agent dueling double deep Q network (MA3DQN) with a multi-step bootstrapping (MSB) approach (CMA3DQN) is proposed, where the input (i.e., energy demand) depends on the output of the first aspect. Particularly, the HAGC approach is adopted to group all prosumers into several agents to reduce the input feature size of each agent. Then the MA3DQN with MSB approach is applied to achieve the best grid energy balance strategy per prosumer. Finally, the experimental results demonstrate the effectiveness of the proposed method. Particularly, the imbalance cost achieved by the proposed joint method is separately $128.71\times$, $12.57\times$, and $11.72\times$ less than the random energy scheduling approach, the independent multi-agent dueling DQN approach, and the approach of employing the double deep Q network per cluster.