A Study Of Wavelet Ant Colony Optimization Based Neural Network Prediction Model

Nowadays, with the increasing usage and emerging types of the Internet, The amount of information on the network has increased largely and has shown it’s varied dynamic characteristics. How to predict the network traffic accurately in order to achieve effective management of the network is becoming a increasingly important issue. In the issue, how to build a prediction model which can predict the network traffic effectively and efficiently is the key problem and it is of great significance for the test of network performance, the control of network traffic, the design of network structure and the quality guarantee of network.This thesis focuses on the research and design for the network traffic prediction model. It includes three aspects:(1)Describe the origin of network traffic prediction, the development and meaning of the research. analyze and describe the characteristics of network traffic.(2)Highlight the Wavelet Decomposition Theory, the operating principle of BP Neural Network and the Wavelet Neural Network(WNN) which is combined of Wavelet Decomposition Theory and BP Neural Network. Meanw hile, emphisize the WNN prediction model based on the Genetic Algorithm(GA).(3)Based on the flaws which exists in the GA-based WNN prediction model, this thesis proposes to replace GA with Ant Colony Optimization(ACO) which has a feedback mechanism and heuristic learning characteristics and proposes the WACONN prediction model which is combined of WNN and ACO.In order to overcome the flaws that BP algorithm is sensitive to initial weights and thresholds of the network and it may fall into local optimal solution in training process, ACO-BP algorithm is proposed which is combined of ACO and BP algorithm. Firstly the global optimization ability of ACO was used to provide more appropriate initial weights and thresholds for BP Neural Network, which can overcome the initial flaws. Then BP algorithm was used to seek for global optimal solution iteratively. During the simulation experiment, firstly the network traffic data was decomposed into high frequency compo nent and low frequency component based on Wavelet Decomposition Theory which can describe the characteristics of traffic data in both frequency domain and time domain. Then, input the prediction part of both high frequency component data and low frequency component data into the prediction model. At last superimpose the high frequency component with low frequency component of predictions,get the final predictions.The experimental data showed that compared with the GA-based WNN prediction model, The ACO-based WNN prediction model which is proposed in this paper has better prediction accuracy and higher network convergence speed. It is proved that the ACO-based WNN prediction model proposed in this paper is a more effective predictive model.