Intelligent Propagation Model Method for RSRP Prediction Based on Machine Learning

The radio propagation model predicts the propagation characteristics of radio waves in the objective communication overlay area, estimates the coverage area of cells, network interference between cells, and other indicators, which is of important meaning for precise 5G wireless network planning. The existing wireless propagation models are generally divided into the empirical model, theoretical model, and improved empirical model. Due to the diversity and complexity of planning scenarios, these three types of propagation models need to conduct propagation model correction and prediction of road loss according to the actual scene data to measure the network coverage, and the calculation process is complicated. For the past few years, data-driven artificial intelligence technology has made great strides in 5G intelligent propagation model fitting. Based on artificial intelligence technology and with enough measured data, we propose a deep learning fitting method to help fit any function with hidden features to predict the coverage prediction of the target region. Basic features are built by measured data, and the combination features and Gradient Enhanced Decision Tree (GBDT) feature generator are designed manually to help build higher-order features as the input of the neural network. Perform regression predictions for Reference Signal Receiving Power (RSRP). The output results of other machine learning algorithms are compared and analyzed. The research shows that data pre-processing of sample data can improve the prediction accuracy of the deep learning model, and the deep neural network algorithm assisted by the GBDT feature generator is superior to other algorithms.