Research On Network Traffic Perception And Prediction Based On Machine Learning Under SDN

Prediction is very important in real production and life.People can generate prediction model according to the real data,which can be used to prevent and prepare corresponding events.Compared with the traditional network,the Software Defined Network has more flexible structures and wider applications,aiming at the low prediction accuracy and slow convergence speed of the existing prediction models,we propose a prediction model based on Fireworks Algorithm(FWA)and Bi-Long Short-Term Memory(Bi-LSTM)to predict time-related data.Firstly,the interconnection structure model of Long Short-Term Memory(LSTM)is established.Secondly,the structure of the LSTM is improved to have a two-layer structure.The Bi-LSTM not only has a front-to-back sequence,but also has a back-to-front sequence,and it can better perceive the two-way timing characteristics of time-related data.Then,considering the diversity and concurrency of the group,the swarm intelligence algorithm-FWA is used to optimize the hyperparameter combination.Finally,the FWA is improved by adding three mutation operators(Gaussian mutation operator,Cauchy mutation operator,and Discrete mutation operator).We use enhanced FWA to optimize the hyperparameter combination of BiLSTM,the addition of mutation operators can increase the diversity of hyperparameter combination,and the optimization result will be better.The simulation experiment is carried out on the MATLAB.The experimental data is software-defined network traffic data.Three mutation operators are added to improve the Fireworks algorithm.The enhanced Firework algorithm is used to optimize the Bi-Long ShortTerm Memory,and the optimized Bi-Long Short-Term Memory is used to predict softwaredefined network traffic;the evaluation indicators include mean absolute error(MAE),root mean squared error(RMSE)and mean absolute percentage error(MAPE).Experimental result shows that compared with the existing LSTM and LSTM-GS models,the performance of the proposed Bi-LSTM-enhanced FWA model has been significantly improved.The MAE are reduced by 54.55%and 38.73%;the RMSE are reduced by 42.16%and 32.79%;the MAPE are reduced by 50.82%and 40%.