5G Network Reference Signal Receiving Power Prediction Based on Multilayer Perceptron

Building an accurate and efficient wireless propagation model is important for predicting radio wave propagation signals in the target area of the communication area. Different from the traditional empirical model relying on empirical values, this paper proposes a wireless propagation intelligent prediction model based on multilayer perceptron. Firstly, based on the analysis of the Cost231-Hata model, 50 features are constructed, containing 4 types of categories: Transmitter features, gird-to-site features, site geophysical features and gird geophysical features. Then, the redundant features with zero variance are also eliminated, the other features are normalized to eliminate the influence of the dimension. Finally, a Mish-function-activated multilayer perceptron, termed as MLP-Cost231-Hata, for reference signal receiving power (RSRP) prediction, is proposed. Besides, the improved gradient descent algorithm DiffGrad is adopted for optimizing the training of MLP-Cost231-Hata. The experimental results show that MLP-Cost231-Hata has high accuracy in RSRP prediction and shows good performance in four metrics, such as poor coverage recognition rate, R- Squared, averaged training time and root mean square error, laying the foundation for optimizing the automated siting of 5G base stations.