Recently, with the advancement of communication technology, an increasing number of applications have placed requirements on the freshness of information, introducing Age of Information (AoI) to quantify the freshness of information at the application side since its generation. However, many applications do not directly provide services to users based on the information itself, but rather on the knowledge extracted after processing this information. This paper introduced the derivative concept of Age of Knowledge (AoK) from AoI to quantify the freshness of knowledge in Mobile Edge Computing (MEC) systems and formulated an optimization problem to minimize the long-term average AoK under limited bandwidth and energy consumption. We construct a constrained Markov Decision Process (CMDP) and transformed it into a Multi-Decision Partially Observable Markov Decision Process (MDPOMDP) to avoid the curse of dimensionality. Furthermore, we propose a learning-based scheduling algorithm to control sensor transmission and verify the effectiveness of our algorithm through simulation experiments.