Intelligent Wireless Network Performance Evaluation And Resources Management

Due to the high-density deployment of APs in large enterprise park WLAN environment,the co-channel interference between APs has been greatly increased which may reduce network performance.However,enterprise users have higher requirements for the performance of the wireless network.Hence,it is essential to effectively evaluate WLAN performance in enterprise enviroment,and to tune network resources based on the performance evaluation results.This paper focuses on the high-density deployment of APs in large-scale enterprise campus WLAN environment,carrying out the research on the AP throughput prediction based on time series,the maximum throughput estimation based on network parameters,and the network throughput optimization based on dynamic power adjustment.The main work and results of this paper are as follows:1.As for the throughput forcasting of Wireless Access Point(AP)based on time series,the effectiveness of ARIMA algorithm and Gaussian Process Regression(GPR)algorithm model in the application of real-time throughput time-series forcasting are studied and compared.Meanwhile,based on the two algorithms' ex-isting issue and considering the need for reliability and accuracy in the actual application of the enterprise,GPR-ARIMA,combination of two model,is proposed.To calculate the weight coefficients,Mean Absolute Error(MAE)and the least square method are used to determine the coefficients jointly.In the real secenry,ex-periment redults have indicated that GPR-ARIMA joint model has better performance,which can effectively reduce the prediction error and raise the reliability of the prediction by offering the confidence interval of the prediction value.2.As for the problem of obtaining the maximum throughput data label,a test APP which simulates the maximum uplink and downlink is developed independently to collect the maximum throughput label.Aim-ing at the problem of too many network features in the collecting dataset,a feature selection based on neural network saliency detection method is proposed in the process of establishing the dataset?in the terminal max-imum throughput estimation research,focusing on the exisiting problems in Deep Neural Network(DNN),DNN with the history information for optimization and Stacking Neural Network by ensembling several models to boost the performance are proposed.By means of the terminal test application aiming at detecting the maximum throughput and the network features collected in the actual peak period scenery,the DNN with the history information for optimization and the Stacking model have better performance than the traditional DNN model.3.Aiming at the problem of WLAN resources management to optimize throughput,an optimization al-gorithm based on the gradient rise method is designed,which is able to optimize throughput by iteratively adjusting the power.Firstly,neural network is utilized to construct the fitting relationship between the net-work features and the output throughput.Then the gradient information of the network output throughput to the input power dimension will be calculated.Finally,the gradient rise method is utilized via iteratively ad-justing power to maximize output throughput.The experiment result reveals that the output throughput can be improved to a certain extend through the power dimension iteration.Hence,conclusion can be made that this algorithm can effectively promote the WLAN throughput.