Aiming at the challenge of enhancing throughput and maintaining high fairness in heterogeneous network access selection, a proximal policy optimization （PPO） algorithm based on reinforcement learning is proposed. This algorithm interacts withthe environment and samples data, aiming to maximize users' long-term throughput and satisfaction. It randomly simulates userlocations, collects user attribute data for model training, and acquires the optimal network access point allocation strategy. Thesimulation results show that compared with traditional algorithms, when the number of access users reaches the maximum, thePPO algorithm can increase throughput by 40% to 70%, while the average user satisfaction can still exceed 30%, and the userfairness index can reach 0.82.