Landscape pattern is closely related to many local ecological processes. Study on the future evolution of landscape pattern in the arid area of Northwest China is of great significance to local prevention and controlling of desertification and water and soil conservation. Therefore, taking Dengkou County, Bayannaoer City, Inner Mongolia as study area, a AES—LPI—CA model was built based on the LPI—CA—Markov model to simulate the landscape pattern of Dengkou County in 2014 by using the remote sensing image interpretation data of 2000 and 2007. Firstly, the landscape pattern transfer appropriate atlas was built and artificial endocrine system (AES) was used to adjust the probabilities of the CA center cell transfer into different landscape types, the cellular automata neighborhood rule was taken into consideration, and the transfer direction of the center cell was settled. Then the un-transition probability (UTP) map which was built based on the quantitative relation between landscape index (LPI) and UTP was used to define the occurrence probability of the transfer, and the landscape transition probability matrix which was generated by using Markov model was used to make the final decision of transfer. The simulation result of the model was compared with the results of LPI—CA—Markov model and CA—Markov model. The Kappa index of agreement (KIA) of simulation results of the three models were 0.8236, 0.7855 and 0.7682, respectively, AES—LPI—CA model had a higher simulation precision. The research result had referential values for the study on future evolution of landscape and formulation of ecological policy.