H-BILSTM: A Novel Bidirectional Long Short Term Memory Network Based Intelligent Early Warning Scheme in Mobile Edge Computing (MEC)

Abstract

Due to the limitations of edge server resources, an efficient offloading strategy is critical for reducing the service latency and energy consumption of terminal devices. However, the efficiency of the resource allocation relies heavily on an accurate evaluation of the load in the edge servers, which is usually unknown and changes periodically. In this paper, we analyse the performance of the resource allocations based on various load prediction algorithm allocations in a mobile edge computing (MEC) system and propose a neural network-based load alert scheme. First, we propose a deep neural network-based MEC load prediction algorithm inspired by a bidirectional long-term and short-term memory neural network (BILSTM) called H-BILSTM to predict the load utilization of MEC servers in future time slots. Second, we correspondingly propose an early warning mechanism based on the predicted load level that can quickly adjust the computing offloading strategy. The simulation results show that both the computing efficiency and prediction accuracy of H-BILSTM are better than those of several baseline prediction methods. In addition, the proposed warning scheme with the early warning unit effectively reduces the response time by 66.5 <inline-formula><tex-math notation="LaTeX">$\%$</tex-math></inline-formula> compared with the baseline.