To address the impact of photovoltaic (PV) output randomness on the economic performance of active distribution networks, an adaptive robust optimization dispatch strategy based on Gaussian mixture model (GMM) is proposed to reduce system operating costs. First, PV output is categorized into two scenarios: sufficient illumination and insufficient illumination. GMM is used to cluster historical PV output data, generating the mean and standard deviation of PV output uncertainty sets for different time periods under varying illumination conditions. Based on the PauTa criterion, accurate uncertainty sets are constructed for each lighting condition. Next, an adaptive robust optimization dispatch model is established with the objective of minimizing the total scheduling cost of the distribution network. The model fully considers the uncertainty of PV output and uses an affine decision rule for solving, enhancing its adaptability to PV fluctuations. Finally, simulations are conducted on an improved IEEE 33-node distribution network system. The results show that the proposed model ensures system security while effectively reducing operating costs compared to traditional interval and polyhedral sets, with lower conservatism in the optimization results.