Energy Efficient Multi-User Computation Offloading Based on Throughput Prediction

Mobile Edge Computing(MEC) pushes computation and data processing capabilities toward the network edge to meet the demands of fast and low-latency applications. With the increasing interest in mobile applications such as vehicular networks, channel prediction-based edge computing become a crucial task due to the time-varying channels. This paper proposes a multi-user, multi-task computation offloading model with uplink throughput prediction. First, the proposed scheme uses an exponential smoothing model to predict future Reference Signal Received Power (RSRP) accurately. Based on the collected RSRP and uplink throughput datasets in real network environments, machine learning is employed to train prediction models for RSRP and uplink throughput. Second, based on the predicted uplink throughput, we then formulate the offloading slot selection as an integer linear programming problem based on minimizing energy consumption. Since this problem is considered NP-hard, the paper proposes a polynomial time complexity computation offloading algorithm. Simulation results show that the algorithm can approximately achieve the minimum energy consumption and has a more efficient solution rate than branch and bound (B&B) algorithm.