Probabilistic linguistic multi-attribute group decision making with unknown attribute weights and expert weights is studied. Firstly, aiming at the deficiency of the distance measure of the traditional probabilistic linguistic term sets, an improved distance measure is proposed, and its properties and advantages are proved. Secondly, based on the new distance measure, the average similarity of decision makers is defined, and the comprehensive weight of decision makers under each attribute is calculated by combining the trust matrix between experts. A group consensus adjustment model based on similarity-trust analysis is constructed to retain the opinions of authoritative experts under each attribute as much as possible. Considering the correlation between attributes and the importance of each attribute, this paper constructs a subjective and objective comprehensive weighting model based on the generalized Choquet integral and deviation maximization method. Then, based on the new distance measure, a probabilistic linguistic multi-attribute group decision framework is constructed by combining the TODIM method to rank multiple alternatives. Finally, the site selection of photovoltaic power station is taken as an example to verify the effectiveness and rationality of the proposed method.