Abstract: Considering the risk of voltage violation induced by the large-scale distributed PV access to the distribution networks, an evaluation method of PV hosting capacity based on the improved holomorphic embedding method is carried out in this paper. Firstly, the discrete-clustering method is introduced to process the time series data, and the extreme scenarios generated by the method are utilized instead of the original series for the subsequent evaluation. Secondly, a self-defined direction model is constructed according to the traditional holomorphic embedding model, and the feasible region range satisfying the voltage constraint is further deduced. In each of the calculations, the intersection of the bus Sigma trajectory and the boundary of the feasible region is defined as the criterion of the maximum PV hosting capacity. Afterwards, the Monte Carlo simulation is employed to generate some mutually independent PV access scenarios, and the hosting capacity corresponding to the extreme sets under each of the access scenarios is determined using the proposed criterion. Consequently, this study counts the calculation results and solves the cumulative distribution curve of the PV hosting capacity, thus making sure of the hosting capacity of the system under the risk of voltage overrun. Moreover, the effectiveness and feasibility of the method proposed are verified in the IEEE 33-bus distribution network system.