کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

LSTM and Markov-Based Mobility Prediction for Multi-access Edge Computing

نویسندگان :

Hadi Ghavaminejad¹ Nasser Yazdani² Golboo Rashidi³

1- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر 2- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر 3- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر

کلمات کلیدی :

multi-access edge computing،base station prediction،long short-term memory networks

چکیده :

With the implementation of the fifth-generation (5G) telecommunication networks and research on the sixth-generation (6G), new requirements and services have been introduced. Multi-access edge computing (MEC), as one enabler, enhances users' quality of service (QoS) in urban mobility environments through innovative solutions. In cities, where vehicles are mostly moving, ensuring seamless connectivity with minimal delays is crucial for providing a smooth user experience. This research aims to solve vehicle movements challenge in such dynamic environments by refining the present solutions, which consider an always-established radio connection and data delivery and utilizing long short-term memory (LSTM) networks to predict the next base station to which a vehicle will likely be connected. In addition, a Markov chain-based model addresses situations where connection loss or data unavailability occurs. Unlike previous works focusing on mobility or trajectory prediction alone, this approach integrates it with outage handling. The provided framework can be used for different challenges in MEC, such as resource allocation, service migration, and load balancing. In this way, real-world mobility data for Luxembourg is used from a SUMO-simulated 24-hour urban scenario. The results demonstrate that this approach achieves a prediction accuracy of 89.01% and a loss function value of 0.2943, which is 1.5% more accurate than the baseline method with 5 times less memory occupancy. Also, for a probable outage scenario, the proposed method reached an accuracy of 84.15% and a loss value of 0.4345.