Multilayer Perceptron (MLP)-based Path Loss Prediction Modeling for LoRa Signals

With the rapid development of Internet of Things (IoT) technology, Low Power Wide Area Network (LPWAN) has become a key enabler for connecting large-scale devices. LoRa technology plays a pivotal role in IoT due to its excellent anti-interference ability and long-range transmission capabilities. However, environmental factors, terrain, and building occlusion significantly impact the propagation and loss of path of LoRa signals, which in turn affect network coverage and communication quality. Therefore, accurately predicting the path loss of LoRa signals is crucial for effective network optimization and troubleshooting. In this study, we propose a path loss prediction model based on a multilayer perceptron (MLP), integrating the Okumura model path loss as a reference feature and comparing it with traditional empirical models. The experimental results demonstrate that the MLP model effectively learns complex nonlinear relationships, improving the accuracy of prediction. Through 10-fold cross-validation and testing various neuron configurations, it is found that the MLP model with 20 neurons (MLP20) performs best, yielding the lowest root mean square error (RMSE). Compared to the traditional Okumura Hata model, the MLP model shows significant advantages in prediction accuracy and stability. This study not only takes advantage of the theoretical strengths of the Okumura model but also enhances the adaptability to complex environmental factors through MLP, providing a new solution for LoRa network optimization.