A Machine Learning based Intelligent Propagation Model for RSRP prediction

Wireless propagation model modeling is of great significance for system design and base station deployment of 5G network-Traditional models are limited to a variety of propagation environments.A deterministic model based on ray tracing requires a great deal of computation.Therefore, we propose a deep learning-based fitting method.However, when the deep learning model is used for wireless transmission model modeling, it usually requires a lot of manual design features. In order to solve this problem, in the feature engineering stage, we proposed a feature generator based on the machine learning algorithm-Gradient Boosting Decision Tree to automatically characterize the combined features.Eight features of manual design based on HATA pathloss formula.In the model stage, an intelligent wireless propagation model based on BP neural network is established-Combined features, manual design features, original engineering parameters, geographic parameters and other data are used as inputs of neural networks for radio wave propagation model modeling and RSRP(Reference Signal Receiving Power) regression prediction.We quantitatively compare the performance of several machine learning algorithms in modeling wireless channels.Our results show that the overall performance of the deep neural network algorithm using the GBDT(Gradient Boosting Decison Tree) feature generator as auxiliary is better than other algorithms.