Abstract: Cluster analysis of photovoltaic (PV) output scenarios is one of the effective ways to describe the uncertain typical output characteristics of PV systems. How to measure the similarity of complex and fluctuating PV power generation curves and generate representative PV output scenarios is currently a pressing issue. A PV typical output scenario clustering method considering comprehensive similarity measurement is proposed. The basic approach is to first consider the similarity in terms of the quantity, trend, and fluctuation position of PV power generation, in order to obtain a comprehensive similarity distance measurement suitable for the PV power generation curve. Secondly, the shape centroid is used as an optimization problem to obtain the actual centroid that balances both the amount of electricity and the shape by using the same multiple amplification method. To address the shortcomings of traditional clustering algorithms in determining initial centers, a PV typical scenario set generation model based on an improved K-means algorithm is proposed using the 24 solar terms as intervals. Finally, a PV power generation scenario set index evaluation system is constructed, and the Entropy-weighted Topsis method is used to comprehensively evaluate the typical output scenario set. The results of a PV power station with an installed capacity of 50MW in a certain area of Yunnan from 2018 to 2020 indicate that the proposed algorithm can accurately classify and extract typical PV output scenarios. The typical scenario set generated based on solar terms shows good performance in terms of fluctuations and electricity indicators, which proves the effectiveness of the algorithm.