Resident load is one of the important components of seasonal peak load and has huge demand response (DR) potential. But its randomness and decentralization limit the ability to flexibly participate in DR interactions. In view of the characteristics of resident load and the uncertainty of response behavior, the resident load model is established with the characteristics of load curve, historical DR participation and response degree as parameters. And the distribution parameters are identified through actual resident data sets. Furthermore, a method of forming a priority queue based on historical response effects is proposed. On this basis, the resident DR cost model is established, and the optimal resident DR strategy is obtained with the goal of cost minimization, so that the DR cost can be reasonably controlled under the premise of accurately achieving the load reduction target. The priority queue is updated and corrected successively in multiple DR events to realize automatic optimization of response strategies. Finally, the calculation example verifies the effectiveness of the proposed resident load model and the resident DR strategy self-optimization method.