Energy Efficiency Optimization Research Based On Deep Learning In Ultra Dense Network

With the rapid development of mobile applications in modern society,people have higher and higher requirements for the data rate and service quality of wireless communication.As a key technology in 5G communication,ultra dense network(UDN)has received extensive research and attention from scholars in recent years.Flexible and dense deployment of small base stations can reduce the distance between user equipment and access nodes,effectively improving the spectral efficiency of the communication network.However,the large number of base stations and the floating network load lead to waste of energy and communication resources at the same time,making the need to improve the energy efficiency of ultra-dense networks more and more urgent.In this context,this paper investigates the problem of deep learning-based network energy efficiency optimization in ultra-dense networks.The main work and contributions of this paper are as follows:Firstly,aiming at the problems of high computational complexity and poor generalization performance of existing power allocation methods,a power allocation method based on sequential fractional programming and its deep neural network implementation and a power allocation method based on deep domain adaptation are proposed.Starting from the traditional mathematical optimization method,the sequential fractional programming method is used to solve the optimal power allocation problem,and then the data-augmented deep neural network is used to fit the mapping relationship between channel gain and power allocation.For the ultra dense network where the position of the obstacle changes,the migration of the data is further considered.Through the adversarial learning of the difference metric and weighted offset correction,the data domain before and after the change is deeply adapted,using a large amount of data from original communication network and a small amount of data from changed communication network.Simulation experiments show that the two deep learning algorithms proposed in this chapter are excellent and stable,and can significantly optimize the energy efficiency of ultra dense networks.Secondly,aiming at the difficulty of modeling and solving the joint optimization of power allocation technology and base station sleep technology,from the perspective of time series prediction,a joint optimization algorithm of power allocation and base station sleep based on long short term memory network is proposed.By using the idea of base station sleep with too low allocated power,the decision of base station sleep and power allocation is jointly processed,and the long and short term memory network is used to mine the channel gain information in the previous time to make decisions for base station sleep and power allocation at the current moment.Simulation experiments verify the performance of the algorithm on the task of improving energy efficiency.The results show that,compared with the method of using solo power allocation and the method of using long short term memory network to do sleep task and then separately doing the spower allocation task,the algorithm proposed in this chapter can reduce unnecessary state switching of base stations and improve the long term average energy efficiency of ultra-dense networks,which is more advantageous.Thirdly,aiming at the difficulty of modeling and solving the joint optimization of power allocation and base station sleep,from the perspective of sequential decision making,a power allocation and base station sleep joint optimization algorithm based on deep deterministic policy gradient is proposed.Using the markov property brought by the base station sleep technology,the joint optimization problem of power allocation and base station sleep is transformed into a markov decision process.Then the deep deterministic policy gradient method is used to continuously interact with the simulation environment of the ultra dense network.Power allocation and base station sleep adjustment actions that bring the greatest expected return for long term energy efficiency optimization.Simulation experiments show that the power allocation and base station sleep joint optimization algorithm based on the deep deterministic policy gradient can achieve lower base station sleep frequency and higher long term average energy efficiency than the method using solo power allocation and the method without base station sleep and power allocation.