System Optimisation for Multi-Access Edge Computing Based on Deep Reinforcement Learning