A Study On Optimal Scheduling In Wireless Networks Based On Machine Learning

The resource allocation problem in wireless networks is an essential issue in designing Internet of Things systems.In this paper,the capacity optimization problem in wireless networks is considered,which involves flow allocation,link scheduling and power control.The computational complexity of this problem increases exponentially with the size of the network.In order to deal with such large-scale optimization problems,traditional algorithms usually use heuristic or iterative algorithms,which usually only achieve suboptimal solutions.In addition,traditional algorithms require long convergence time,which results in transmission delay or performance degradation under dynamic traffic demands.Different from most traditional iterative methods,we directly calculate the approximate solution based on machine learning.Firstly,an independent set learning algorithm based on deep belief network is proposed.Then we propose a learning system that consists of both support vector machine and deep belief network,using collaborative scheduling and power allocation algorithm to achieve capacity optimization.The main contributions of this paper can be summarized as follows:1.We first introduce the research background and difficulties of wireless network capacity optimization,and analyze its related resource allocation problems.It focuses on how to improve the efficiency of solving large-scale optimization problems under the premise of maintaining performance.The related work and the shortcomings in the current research work are discussed.2.The optimal scheduling model in wireless networks is proposed and the capacity optimization problem is decomposed into a linear program and a nonlinear weighted sum-rate maximization problem for power allocation.3.The independent set learning algorithm for capacity optimization in wireless networks is proposed.Since the optimal resource allocation problem in wireless networks usually concentrates on the scheduling problem of network independent sets,we define the concept of group independent sets.Then we propose the evaluation methods,utilize deep belief network to evaluate group independent sets under dynamic traffic demands.Simulation results indicate that the learning method based on deep belief network is superior to the traditional iterative algorithm in improving network capacity and computing efficiency.4.We propose a cooperative optimization algorithm of network transmission scheduling and power control.The machine-learning framework consisting of support vector machines and deep belief network is proposed.The approximate solution of nonlinear program can be obtained by the combination of classification of support vector machine and prediction results of deep belief network.Simulation results demonstrate the effectiveness of the proposed machine learning based algorithm.5.We summarize our work and discuss the future research work.