Research On Network State Prediction Based On Multi Source Data Mining

With the widespread use of virtualization technology in the campus network, running status of network is the main concern indicator of network management system, if one or a few network nodes appear congestion or downtime will cause that the whole virtual network can not communicate, then causing the network crash or paralysis. Therefore, to predict in advance network delay and network congestion of network fault state, enables network operation and maintenance system can timely response and process, which is very important to ensure the stable operation of the network.To predict the running state of the network needs collection network performance data in the key nodes of the network, the paper based on SNMP completes data collection of network running status. Dividing the network operation data set into four group, such as ip group, tcp group, interface group and icmp group to predict network operation status. The collected data will be remove redundant data, clean up the error and data normalization and so on.In this paper, the design of the running state of the network prediction method is that according to the collected monitoring network running state data, usage ARMA model to predict the next phase data of the network operation and reuse data prediction the monitoring points of the running state of the network by using learning neural network to forecast. Finally,calculate the importance degree of each network monitoring using the energy migration of Markov chain, and make network status fusion using weighted neural network, Then the good fusion results are obtained.Using the virtual network environment for the operation of the network state prediction experiment, the experimental results show that ARMA model prediction obtained running state of network data and the real data contrast found that prediction accuracy meets the requirements of the study, and the accuracy of 87% calculated by using neural network node state data can meet the experimental requirements, the final data fusion conforms to the research target, so as to achieve the purpose of the study.