In recent years, mini-station monitoring of air quality (AQ) has gradually become an important part of the AQ monitoring network. With the continuous economic development and increasing acceleration of the urbanization process, monitoring stations appear redundant and not as representative as they used to be. Due to the weak ability of AQ monitoring mini-stations against sudden environmental factors, monitoring data missing can easily occur, which not only greatly increases the complexity and difficulty of data analysis but also leads to the deviation of optimization and distribution results. In view of the above problems, we propose a site optimization method that combines the BiLSTM neural network with clustering, of which the BiLSTM neural network is applied to complete the missing data, and the condensed hierarchical clustering method is employed for the clustering of restored data. On the basis of the feedback on the AQ level with as few and accurate sites as possible, the clustering accuracy is greatly improved. Finally, this study uses the monitoring data from 18 AQ monitoring mini-stations in the Hunnan District of Shenyang for verification. The results show that the performance of the proposed algorithm is improved to a certain extent compared with that of general clustering algorithms, which provides a new method for the optimization of AQ monitoring sites.