Combining Multi-objective Evolutionary Approach and Machine Learning to Optimize PCI Configuration in Large-scale LTE Networks

Wireless interference seriously affects the quality of service in mobile communication networks, and PCI planning and optimization is an effective method to reduce interference in 4G and 5G networks. Most of the existing PCI configuration optimization methods focus on solving basic problems, such as PCI collision, PCI confusion and PCI mod 3 interference, which can not meet the complex requirements of actual LTE network optimization. In this paper, we establish six objectives and six constraints based on the PCI optimization requirements of large-scale real LTE networks, and propose a decomposition-based multi-objective evolutionary algorithm combining community detection and reinforcement learning mechanism improvement. Specifically, community detection is used to improve the selection method of crossover segments and mutation points in the evolutionary algorithm so that good sub-region patterns can be inherited to the next generation, and Q-learning method is took to adaptively adjust the crossover and mutation probabilities according to the evolutionary iteration number, population diversity and average fitness to improve the diversity of the population. The PCI optimization results for 1231 optimized cells and 5169 associated cells in a city of China show that our algorithm has improvement in all six objectives with an average 1.38% increase in the optimization rate of the original solution compared to the baseline algorithm, and 21 % reduction in runtime for 1000 generations. Therefore, the proposed algorithm is an effective method to improve PCI configuration and reduce network interference.