The increasing number of take-out orders has caused major food delivery platforms to face enormous delivery pressure and operational challenges. There is an urgent need to improve the overall efficiency of food delivery schemes, to further reduce costs and improve the system service level. In this paper, the On-Demand Food Delivery Problem (OFDP) is studied by establishing a food delivery optimization model with the objective of minimizing the sum of traveling cost and overtime penalty cost. Considering the potential intensive advantage of order consolidation, an order group generation and assignment algorithm based on order matching degree and order overtime degree is proposed to solve the problem. Compared with the Gurobi solver, it is verified that the proposed algorithm can solve practical scale problems in a short time and obtain high-quality solutions. Moreover, the feasibility of the algorithm, total cost and order delivery overtime indicators under a limited number of drivers are further studied by generating test instances with different order sizes. Finally, a decision-making method for dynamically dispatching orders is proposed with the use of batching rule based on hybrid time window. The impact of several parameters including the number of drivers and overtime penalty coefficient on the operation strategy and service level of the food delivery system is explored, using the data from the real-world delivery platform. And relevant managerial insights are further summarized. The relevant research and conclusions in this paper can provide decision basis for order delivery practice of on-demand food delivery platforms.