Dynamic Traffic Predicton Model And Its Application Based On Recurrent Neural Network

Along with the rapid development of the mobile Internet and the Internet of Things,new types of services and numbers have increased dramatically.Problems such as declining service quality due to inadequate monitoring of network indicators and declining network stability have occurred.In modern networks,network traffic is an important indicator to measure the overall network operation.Modeling network traffic char-acteristics is the premise of network state analysis.It also has significant impact on network management,new network protocol development,and improving network operation service quality.The research of network traffic prediction model has far-reaching significance for better understanding the performance and regularity of various services on the network,assisting in planning network design,determining network congestion control,QoS guarantee,and improving service quality.From the early Poisson process to the use of long correlation models to describe network traffic,new research methods and research tools are constantly being presented,and researchers have been exploring ways to find better modeling methods.Since the new century,with the introduction of new research models such as neural networks and chaos theory,Related research on network traffic has also been promoted.However,these methods have problems such as theoretical complexity,hyperparameter debugging,and complex model fitting.The introduction of these theories into network traffic has presented new challenges for researchers.In view of the difficulty in debugging hyper-parameters in neural networks,This paper proposes an improved firefly algorithm.The characteristics of population diversity were introduced,the position update was adjusted by the population diversity index,and the adaptive step size factor was introduced to improve the iterative step size.Experiments show that the improved firefly algorithm has better search performance and can better serve hyperparameter selection.Combined with the improved firefly algorithm,this paper combines LSTM(long-term and short-term memory network)to construct a traffic prediction model,using LSTM for historical memory of time series and neural network for complex nonlinearity.The system's fit,learning and memory network traffic characteristics,using this model to predict the future time flow sequence.The experimental results show that the model has higher prediction accuracy and accuracy than the common time series prediction model.For the application of network traffic,this paper designs and implements a set of traffic prediction and warning system.The system implements real-time traffic detection,traffic prediction and warning functions,and the prediction model periodically uses the latest sampling data to update and ensure prediction accuracy.Through this system,users can monitor the network traffic in real time and implement network traffic prediction and warning through specific configurations.