A Load Balancing Algorithm Based On Kohonen Network Combined Prediction Model In Congitive Networks

With the rapid development of modern network and the appearance of new network service,besides,people's telecommunication quality is more and more strict which brings a series of problems, such as the poor network compatibility.the serious network traffic delay and congestion, and the poor network load balancing effect. But the traditional traffic forecast model generally only deal with the short or long related characteristics of traffic which couldn't completely characterize the sudden, chaos, self-similarity, periodicity and other features of network traffic and result in the high prediction error.Meanwhile, the existing network load mechanism generally can't perceive the network situation in advance and self-adaptive adjust the network parameters that leads to a certain hysteresis and unable to guarantee the network services QoS. Therefore, establishing an effective network traffic prediction model and efficient load balancing mechanism now become a research hotspot domestic and overseas.Cognitive network has the ability of independent learning and reconfiguration.Through perceiving the environment in advance, it will intelligently, adaptively adjust the network parameters, so as to realize the network load balancing, improve the resource utilization rate and provide network Qos better. Therefore.aiming at the problems above and combining with cognitive network, in this paper we set up a combined network traffic prediciton model based on Kohonen neural network.The model uses the features of fast learning rate, high classification precision and strong anti-noise ability of the Kohonen neural network to improve the nonlinearity, multi-scale network traffic prediciton accurancy.Meanwhile.basing on the Kohonen neural network traffic prediction model,this paper improves the current widely used scheduling algorithm the weighted least connection algorithm and established a new load balancing model in cognitive network.Through perceiving the environment in advance,it configures the available network resources ahead of time, so as to realize the dynamic,real-time network business scheduling and achieve the network load balancing in the end.The specific research works and results are as follow:Firstly, this paper introduces the concept and model of cognitive network and do in-depth study on the cognitive network characteristics of self-learning and self-adaption.Besides,thesis breifly introduces the network traffic different features.Meanwhile,combining with these features,this paper carefully analyze the traffic prediction and load balance model at this stage and sums up their advantages and disadvantages in the corresponding.Secondly, thesis sets up a combined network traffic prediciton model based on Kohonen neural network.This paper chooses cognitive network as the main research direction, according to the cognitive network self-learning, self-adaption characteristics, specificlly bringing in Kohonen neural network.using its features of fast learning rate,high classificaion precision, strong anti-noise,self-organization,self-adaption,fast convergence in trainning and clustering, meanwhile,combining with the advantages of wavelet in multi-scale data decomposition, AR model in the linear data processing, BP neural network in data fitting and LMS, designs a combined network traffic prediciton model based on Kohonen neural network. The network traffic firstly is processed in LMS.Then it will be decomposized by wavelet transform into high frequency part which is dealtd by Kohonen model and low frequency part which is dealt by AR model to predict, In the end,it will be dealt by BP mode to match. Through the Matlab simulation,it proved that the the combined traffic prediction model based on the Kohonen effectively improve the nonlinearity, multi-scale data prediction precision compared with the traditional network prediciton model.Finally, thesis designs a new dynamic load balancing model in cognitive network based on the Kohonen network traffic prediction model and the improves the existing traffic scheduling algorithm.This paper combines the prediction model based on Kohonen neural network above with cognitive network and mproved the existing scheduling algorithm the weighted least connection scheduling algorithm. Due to the algorithm whithout self-learning, self-adaption and without considering the parameters of server performance this paperwhich results in a certain hysteresis, this paper proposes a cognitive network load balancing model based on Kohonen neural network prediction model and an improved scheduling algorithm.According to the features of self-cognitive and self-learning of congitive network,the model adaptively adjusts the parameters of the neural network in advance and achieve network load balancing in the end.In this paper, using OPNET to simulate the new load balancing model, respectively simulating the corresponding process model, node model, network model.And we anlyzed the effect of the server load and network delay compared with general network which only used unimproved the weighted least connection scheduling model.The results show that the prediction model in cognitive network based on Kohonen network and improved weighted scheduling algorithm make the server load balancing better and improve the network delay efficiency.By the configuring network resources in advance, the model effectively shorts the server idle time and queue delay time, ensures the business flow dynamic distributed into the network and transimits the network packets timely, meanwhile, making the server a better load balancing.