Balancing High-Load Scenarios with Next Cell Predictions and Mobility Pattern Recognition

Abstract

Knowing where a mobile user will be next can deliver a tremendous increase in network performance under high load, as this knowledge enables pro-active load balancing. To derive this information, sequences of traversed cells are fed into pattern detection algorithms. After the training phase the learned model predicts each user’s next cell. Even for complex scenarios, the prediction accuracy can exceed 90%. Predictions are used to rearrange mobile connections in a simulated high- load scenario centered around an event at a soccer stadium. To prevent call drops for mobile users targeting the stadium, apropriate resources in the predicted next cell are reserved. The results exceed 20% in improvements for throughput and call drop rates, enabling the network to bear a much higher load before stalling.