Over the past few decades, air pollution has caused serious damage to public health. Therefore, making accurate predictions of PM2.5 is a crucial task. Due to the transportation of air pollutants among areas, the PM2.5 concentration is strongly spatiotemporal correlated. However, the distribution of air pollution monitoring sites is not even making the spatiotemporal correlation between the central site and surrounding sites vary with different density of sites, and this was neglected by previous methods. To this end, this study proposes a weighted long short-term memory neural network extended model (WLSTME), which addressed the issue that how to consider the effect of the density of sites and wind conditions on the spatiotemporal correlation of air pollution concentration. First, a number of nearest surrounding sites were chosen as the neighbor sites to the central site, and their distance, as well as their air pollution concentration and wind condition, were input to multilayer perception (MLP) to generate weighted historical PM2.5 time series data. Second, historical PM2.5 concentration of the central site and weighted PM2.5 series data of neighbor sites were input into a long short-term memory (LSTM) to address spatiotemporal dependency simultaneously and extract spatiotemporal features. Finally, another MLP was utilized to integrate spatiotemporal features extracted above with the meteorological data of the central site to generate the forecasts future PM2.5 concentration of the central site. Daily PM2.5 concentration and meteorological data on Beijing-Tianjin-Hebei from 2015 to 2017 were collected to train models and to evaluate its performance. Experimental results with three existing methods showed that the proposed WLSTME model has the lowest RMSE (40.67) and MAE (26.10) and the highest p (0.59). Further experiments showed that in all seasons and regions, WLSTME performed the best. This finding confirms that WLSTME can significantly improve PM2.5 prediction accuracy.