Drl-based Network Autonomous Optimization Simulation Platform And Its Application

With the rapid development of communication technology since the 20th century,the need for information exchange and transmission has led to an increasing number of network devices and a more complex network structure,while users’ demands on the network are also increasing,making network optimization a valuable and difficult research direction in the information age.Between the massive number of nodes and cross-scene demands,traditional optimization methods gradually show a declining trend,while at the same time,network optimization methods based on deep reinforcement learning stand out among traditional algorithms and conventional machine learning algorithms because they do not need to collect data in advance and have certain adaptability to the environment.In this paper,based on the characteristics of network optimization objectives and reinforcement learning algorithms,a general-purpose network autonomous optimization platform is developed and applied in two classic areas of network optimization,namely,intelligent routing and power allocation,as follows.(1)A network optimization platform is designed and implemented based on the characteristics of network optimization.In engineering,certain code needs to be written to combine reinforcement learning algorithms with specific application environments.In the scenario of network optimization,there is a certain commonality in the logic of interaction between the algorithm and the simulation environment.The autonomous network optimization platform(hereafter referred to as RL4Net)takes advantage of this commonality to design its interaction logic between the algorithm and the environment,making it unnecessary to redesign and implement the interaction logic for different specific network optimization tasks,reducing the coupling between modules and increasing the reusability and expandability of the modules.A good design lays the foundation for subsequent work and increases development efficiency.(2)Applying the RL4Net framework to the intelligent routing domain and verifying the effectiveness of reinforcement learning algorithms.This paper implements a simulation environment corresponding to the smart routing domain and integrates it into RL4Net,giving a common interface and training method for the application of reinforcement learning in the smart routing domain.Based on this,this paper successfully completes the training of reinforcement learning algorithms and verifies the effectiveness of reinforcement learning algorithms in the intelligent routing domain.(3)Apply RL4Net framework in the field of power allocation and verify the effect of reinforcement learning algorithm.In this paper,a finegrained simulation environment is implemented in the power allocation domain and integrated into RL4Net.The power allocation and intelligent routing environments are trained in a consistent manner to achieve the effect of RL4Net common interface.Based on this,this paper successfully completes the application of reinforcement learning in the power allocation domain and verifies the effectiveness of reinforcement learning in the power allocation domain.