Aiming at the problem of high-dimensional solving and coexisting discrete and continuous variables in the site selection and capacity planning of distributed power generation, a method is proposed to first determine candidate locations and then a heuristic algorithm is used to optimize the capacity and location. Firstly, the nodes are screened according to the three indicators: active loss improvement rate, betweenness centrality and closeness centrality, and adjacent nodes are merged according to the node power load conditions to form a set of candidate nodes. Subsequently, an optimization model with the goal of minimizing total active loss, maximizing voltage stability, and minimizing total capacity of distributed power generation is constructed. In order to effectively solve the model, dung beetle algorithm is improved based on chaotic mapping, adaptive weights and Levy flight strategy to optimize the locations and capacities of distributed generation in the candidate nodes. Finally, simulation verifications on IEEE-33 and IEEE-69 node systems show that the proposed method performs well in reducing total active power loss and node voltage deviation.