Abstract: The wireless propagation model,which can accurately predict the path loss of radio waves,plays an important role in the estimation of communication rate,coverage and interference.It plays a fundamental role in the design of communication systems in civil and military fields.With the advance in artificial intelligence,there appears a significant trend to develop intelligent wireless propagation model that replaces the empirical formula with machine learning algorithms to fit the path loss.The intelligent wireless propagation model effectively extends the applicability of the propagation model and reduces the error in predicting path loss.However,because the optimal input features set of the intelligent wireless propagation model may be different in diffe-rent propagation environments,it is important to optimally design and select the input features for different scenarios.Therefore,this paper proposes a self-adaptive intelligent wireless propagation model(SAIWP).Firstly,inspired by the processing methods of empirical model for features in different scenarios,the SAIWP model extends the input features set of the intelligent wireless propagation model.And then,the SAIWP model uses the simulated annealing algorithm to self-adaptively select the optimal input feature subset to reduce the error in the prediction of path loss.Finally,the SAIWP model exploits the optimal input feature subset in the optimization process and all data set to train the intelligent wireless propagation model.Simulation results show that,in the LTE networks and the smart campus,compared with traditional empirical models and intelligent wireless propagation models,the SAIWP model predict accurately in various terrains and distances,and effectively reduces the error in the prediction of path loss.

Key words: Deep learning, Empirical model, Simulated annealing algorithm, Wireless propagation model