Simulation of construction land expansion considering the difference of population density level

China is in the middle and late stage of urbanization, which requires more rational planning of cities to avoid blind expansion of construction land. Cellular Automata (CA) is widely used in the simulation of land use change because it is easy to be coupled with other models, and the description of cellular state through different methods can realize the simulation and prediction for different purposes, and then provide scientific guidance for spatial planning. In this paper, we firstly adopted the Percolation-based City Clustering Algorithm (PCCA), which considers the critical characteristics of the system, to divide the population density distribution into high population density areas and low population density areas; and then according to the built environment distribution in the land use and cover data, the construction land distribution pattern taking into account the difference of population density level was obtained, including high population density urban areas, low population density settlements and non-settlement areas; and then we used Artificial Neural Network (ANN) to construct a CA model (PCCA-ANN-CA) to simulate the expansion of the above categories of construction land. Taking the Beijing-Tianjin-Hebei urban agglomeration as an example, in this paper we found that the threshold for the division of high/low population density areas in 2020 is lower than that in 2010, and the regional population density decreases. By analyzing the proportion distribution of counties, it was found that the hot spots of low population density settlements are concentrated in Hengshui, Cangzhou, Tianjin, Tangshan and the south of Qinhuangdao in 2010, and the heat value of Tianjin decreases by 2020. In 2010, the hot spots in high population density urban areas were concentrated in Beijing and Tianjin, and the sub-hot spots were concentrated near Shijiazhuang. In 2020, the locations of hot spots and sub-hot spots did not change, but the calorific value of Baoding, Zhangjiakou and Chengde decreased. PCCA-ANN-CA can not only simulate the conversion between non-construction land and construction land, but also simulate the conversion between different types of construction land, and using the difference of population density level to subdivide the attributes of construction land can avoid the excessive concentration of simulated new construction land in the area with more existing construction land, so that the simulation results are more similar to the actual situation.