In the distribution network, the coordinated scheduling of flexible resources and distributed photovoltaics(PV) plays a crucial role in enhancing the consumption level of distributed renewable energy and promoting the realization of national "dual carbon" goals. However, traditional scheduling methods often require detailed topology and line impedance information of the distribution network, which is frequently unknown in actual distribution networks. To overcome this challenge, a non-parametric operation optimization method for distribution networks based on decision tree is proposed. This method first leverages unsupervised clustering to extract the topology status implied in the historical data to generate“pseudo labels”, and then train a decision tree to learn the relationship between operation strategies and the feasibility of flow constraints from historical operation data of the distribution network by training decision trees, and equivalently transforms the decision trees into mixed-integer linear forms that are easy to solve, thereby achieving a non-parametric reconstruction of flow constraints. Case studies show that this method can accurately depict the feasible domain of flow constraints under conditions where network topology and line impedance information are unknown, and its computational efficiency is significantly higher than that of traditional model-based methods.