Contextual Multi-Armed Bandit Learning for Freshness-aware Cache Update in Vehicular Edge Networks

Abstract

Vehicular edge caching (VEC) is expected to support real-time intelligent transportation systems by providing low-latency data services. However, dynamic vehicular environment, such as time-varying data freshness and dynamic data preference, may result in low cache efficiency. Based on the above motivation, this paper designs a system model of freshness-aware VEC system. Accordingly, we formulate the problem of Vehicular Edge Cache Update (VECU) by exploiting the concept of AoI and data heterogeneity for evaluating data freshness, which aims at maximizing the edge cache benefit. On this basis, the Contextual Multi-Armed Bandit for Caching Update (CMAB-CU) algorithm is designed to determine cache update decision by online estimating reward of each arm based on a linear function of dynamic vehicular features and historical observations. Finally, we modeling a simulation model and conduct simulation results, which demonstrates the effectiveness of the proposed algorithm in various service scenarios.