Anticipatory resource optimization based on model predictive control in virtual mobile networks

We developed a method that predicts the future location of human. The proposed method employs a sequential pattern mining algorithm called BIDE method to extract frequent trajectory patterns from a large amount of human trajectory data and calculates a score of frequent trajectory pattern to predict the future location of human. We evaluate the proposed method using about 18,000 trajectory data collected from 200 users during five and a half years and demonstrate that the proposed method yields 70% accuracy of prediction. Conventional resource allocation methods that use only traffic data could perform poorly when the numbers of users in the areas fluctuate. We develop a model predictive control for resource allocation method that uses not only traffic data but also predicted number of human in areas. We demonstrate that the proposed method, with help of exponential smoothing, allocates sufficient resource even when short-term fluctuation exists.