Abstract: The promotion of rural roof photovoltaic project has been strongly supported by the national policy, but because of the instability of light intensity, the output of photovoltaic is intermittent and volatile, and the instability of photovoltaic output causes great impact on the power quality of distribution network after photovoltaic access, so the power quality of distribution network after photovoltaic access becomes very important. The construction of judgment matrix is an important step in power quality evaluation. Because the judgment of the importance degree between two power quality indicators is a relatively vague concept, it is difficult to define clearly. Therefore, there are some differences in the judgment of the importance degree between the indicators from the experts. How to unify the opinions of the experts is the problem to be solved in the construction of judgment matrix. In this paper, the D-S evidence theory was used to fuse the different judgment opinions of experts on the importance of various electric energy indicators to form a judgment matrix, which could avoid the risk of inaccurate evaluation results caused by single expert's misjudgment. Then the weight distribution of each power quality index was obtained by analytic hierarchy process. In order to reduce the interference of subjective factors on the evaluation results, the entropy weight method was introduced to improve the analytic hierarchy process. The probability matrix of each power quality index was analyzed by the method of entropy weight, and the entropy weight distribution of each power quality index was obtained. The two weight allocations were synthesized. The composite weight coefficient reduced the interference of subjective factors on the evaluation results. At the same time, the paper improved the shortcomings of the analytic hierarchy process and the entropy weight method, which not only avoided the steps of consistency checking in the analytic hierarchy process, simplified the calculation, but also solved the disadvantage of the traditional entropy weight method that when the entropy value approached to a minimum, the difference of the entropy value would cause the double changes of the entropy weight. Finally, the final power quality evaluation results were obtained by using probability theory and combining the comprehensive weight and the probability matrix of each power quality index. The simulation results showed that when the three-phase unbalance index and voltage offset index changed dramatically, the weight of the two indexes could be increased from 0.067 and 0.183 to 0.164 and 0.192 by the proposed method, and the final evaluation result could also be increased from 2.323 to 2.679. From the weight coefficient, it showed that the drastic change of the two factors was the main factor affecting power quality. Therefore, the proposed method is more suitable for rural distribution network system with large fluctuation of power quality indicators than the traditional power quality assessment method.