Intelligent TCP Congestion Control Policy Optimization

Network congestion control is an important means to improve network throughput and reduce data transmission delay. To further optimize the capability, this research suggests a proximal policy optimization-based intelligent TCP management method, creates proxy that can communicate with real-time environment, abstracts mechanism into partially observable Markov decision process. Changes in state of are fed back agent, agent makes action commands size window, which will produce new state, immediately receive feedback reward value. guarantee actions taken optimum, agent’s goal obtain highest The space characteristics should be designed so agents observe enough information make appropriate decisions. function through weighted algorithm enables balance latency. model parameters updated by optimization algorithm, truncation keeps within certain range, reducing possibility oscillation during gradient descent ensuring training process converge quickly. Compared traditional CUBIC results show TCP-PPO2 reduces latency 11.7–87.5%.