Proactive Application Deployment for MEC: An Adaptive Optimization Based on Imperfect Multi-Dimensional Prediction

This paper explores proactive application deployment in Multi-Access Edge Computing (MEC) systems through dual-dimensional prediction of user mobility and computing resource demands. We reveal a fundamental trade-off between prediction accuracy and temporal scope: Shorter-term mobility predictions enhance accuracy but increase deployment costs, while longer-term predictions reduce costs at the expense of higher service delays. Dynamic computing demands and constrained edge resources further complicate application instance allocation. To address these challenges, we propose an adaptive deployment framework leveraging multi-user multi-period predictions. This scheme simultaneously determines optimal application placement across edge nodes, deployment timing, and instance quantity. We employ a residual Long Short-Term Memory (LSTM) framework for joint mobility and computing resource requirement prediction. Based on this, we develop an advanced branch-and-bound deployment algorithm to obtain near-optimal solutions and a low-complexity greedy algorithm for rapid decision-making. Simulations demonstrate the feasibility and superiority of our proposed scheme over counterparts.