Reinforcement learning based Dynamic Network Self-optimization for heterogeneous networks

The coexistence of different heterogeneous Radio Access Technologies (RATs) is a significant feature of current wireless networks. Thus, it is important for network elements, such as the Base Stations (BSs) of cellular networks or access points (APs) of wireless local area networks (WLANs) to be reconfigurable according to the real-time network environment. This will enable interconnection between different networks. In this paper, we propose an efficient distributed reconfiguration algorithm: the Dynamic Network Self-optimization Algorithm (DNSA). This algorithm is based on the widely used Q-learning algorithm and regards the network self-optimization entity of each radio access network as an independent intelligent agent. Multiple agents perform the optimization cooperatively to reduce the system blocking rate and improve network revenue. In the proposed algorithm, the dynamic network self-optimization problem is transformed into a multi-agent reinforcement learning problem, which has much lower complexity and better performance.