With the development of 5th Generation Mobile Communication Technology (5G) and the evolution of network architecture, the analysis and optimization of network coverage need to consider comprehensive factors, including not only the link budget of the base station antenna system, but also the geographical conditions and time characteristics of the area covered by the base station. Therefore, a more accurate network coverage optimization plan should be designed. This paper proposes an adaptive network coverage optimization algorithm based on Q-learning. The method first adopts a cellular network coverage prediction model based on data mining, which can predict the coverage situation of the access terminal through the configuration of the antenna of the cell, and verify the accuracy of the prediction based on real data. Then, a network coverage optimization algorithm based on Q-learning is proposed which modifies the action selection strategy of the agent in reinforcement learning process. According to the coverage of each cell, different optimization priorities are set. Combined with the greedy strategy, a cell and its antenna parameters are decided by the agent in each iteration. This method effectively reduces the probability of falling into a local optimum during the iteration process, and the method also has a good performance in reducing the convergence time of the optimization process. The simulation result shows that the algorithm can increase the network coverage by up to 20%.