A Multi-Task Learning Model for Trajectory and Cellular Signal Power Prediction

The prediction of user trajectories based on contextual information has emerged as a critical concern for Human-Centric Services (HCS), as they strive to effectively manage the rapid proliferation of mobile terminals and cater to diverse traffic demands. However, the increasing number of mobile users exacerbates the dynamic fluctuations in wireless services, posing challenges for Mobile Network Operators (MNOs) in efficiently overseeing and allocating network resources. In this paper, we propose a Multi-Task Learning (MTL) model for trajectory prediction and cellular signal power estimation. This model enables the anticipation of signal power levels while simultaneously providing location predictions, thereby contributing to the optimization of network resource management. Utilizing real-world measured data, the prediction input comprises historical trajectories coupled with corresponding Reference Signal Receiving Power (RSRP) and Reference Signal Receiving Quality (RSRQ) values, enhancing the feature set and consequently mitigating errors. Specifically, substantial volumes of measurement reports (MRs) that delineate the movement trajectories and radio network performance of mobile users have been gathered through drive tests in urban traffic scenarios. Subsequently, an MTL model based on Long Short-Term Memory (LSTM) architecture is constructed to capture patterns and dependencies within the position and cellular signal sequences in the MRs. This approach to joint prediction enhances prediction performance. As comparative experiments shown, the proposed model attains a low prediction error in both location and signal power prediction tasks, outperforming baseline methods (TCN, LSTM, Sequential LSTM, BiLSTM). Moreover, the proposed MTL model introduces multi-step prediction, thereby extending the scope of prediction. The balance between prediction error and prediction range is thoughtfully assessed across different scales, facilitating the practical application of the multi-task prediction model in real-world traffic scenarios.