A Fusion Model Based On Optimized Neural Network For Wireless Network Traffic Prediction

Wireless network is acting an increasingly important role in the information society. Wireless networks can easily and effectively carry out high-speed communication, which not only facilitated people’s lives, but also brought new development opportunities for the country’s economy, politics and military. As more and more wireless broadband network access points being deployed in industries and public areas, wireless networks has become increasingly large-scale in complex circumstances. Network operators suffered lacking effective means to ensure the wireless network QoS. Besides, network operators also demanded effective statistical model study on wireless network traffic characteristics and reliability, which makes achieving guaranteed network QoS, maintaining network security and network fault detection difficult to be carried out in practice. In order to solve the problem, modeling the network traffic has become a major tool. In this paper, we proposed a systematic study on wireless network traffic and its modeling based on optimized artificial neural networks.In order to study wireless network traffic prediction method, the wireless network traffic is studied by analyzing its statistical properties, relevant characteristics, self-similarity and chaos characteristics. In comparison with the wire-line network traffic, it is proved that wireless network traffic showed more dispersion, randomness and chaos characteristics.Then, we proposed several researches on traditional and chaotic time series analysis. Also we studied on ARIMA model prediction and chaos-RBF neural network model prediction. The results showed that both model did not perform well in wireless network prediction.Additionally, we focused on BP neural network, quantum genetic algorithms and wavelet transform theory. Specific research on BP neural network of its structure, advantages and drawbacks was taken, according to which, an optimization of BP neural network based on quantum genetic algorithm was proposed in theory. Furthermore, based on the optimization of a neural network with strong robustness and non-linear processing capabilities, combining the stationary wavelet transform, a fusion model named SWT-QGA-BP model was developed for wireless network traffic prediction.Four experimental simulations of wireless network traffic including a one-step prediction and multi-step predictions were taken to evaluate the SWT-QGA-BP model by several statistical criteria. In comparison to ARIMA model and Chaos-RBF model, the new model is verified adaptive and precise prediction superiority. Therefore, SWT-QGA-BP model is verified as an accurate and efficient wireless network traffic prediction method for wireless network traffic, it’s capable of providing adequate evidence for QoS guarantee, network resource management and network security maintenance.