Satellite Network Load Balancing Routing Key Technology Research Based On Traffic Prediction

With the continuous development of LEO satellite communication,especially on the rising capacity of the satellite, coupled with the advantages that it has wide coverage, low effects of geographical environment and flexible networking, it has attracted wide attention from scholars. At present, there is not any effective method for fast and accurate traffic prediction for low-orbit satellites under high dynamic operation. At the same time, the real-time change of the delay, bandwidth and location information caused by the high-dynamic of the satellite makes the efficiency of the route ineffective without the unified indicator of the link state.To solve this issue, this thesis studies the load balancing routing algorithm for satellite networks based on traffic prediction with the support of national 863 project. The main contents are as follows:1 .SVM traffic prediction algorithm based on wavelet decomposition.Aiming at the disadvantages that the satellite network traffic prediction method is too simplex and cannot give attention to both efficiency and precision, this thesis deeply analyzes the similarity characteristics of satellite signal based on the basic theory of wavelet and the advantages of support vector machine (SVM). The algorithm eliminates the noise in the signal by the use of wavelet theory. Then Quantum-behaved Particle Swarm Optimization algorithm with non-linear expansion factor is used to optimize SVM, which completes the prediction and reconstruction of the signal after the decomposition and realize the accurate forecast of the whole network traffic. Simulation results show that the new algorithm reduces the error index e from the original 6.11×10-4 to 2.96×10-4 and obtains a better prediction result which provides a good reliability guarantee for the subsequent load balancing routing algorithm.2. Space-Time-Frequency Tight Coupling Routing Algorithm based on Delay-Tolerant Pheromone Dispersion Search. The algorithm uses the traffic prediction value as a protection amount in the routing algorithm to prevent the generation of positive feedback effect and improve the validity of the routing. In addition, difference of time, space, frequency domain among various satellite business are adopted to make quantitative description and dynamic learning towards the pheromone mentioned in the algorithm, thus the link status in the satellite network can be reflected dynamically. The new strategy can not only overcome the existing disadvantage of single reference value and low learning ability, but also solve the local tidal effects issue resulted from the high dynamic of network and low processing capacity of satellite nodes. Simulation shows that compared to the existing schemes, the algorithm can provide larger throughput and smaller time delay under the guarantee of the low packet drop rate. In the case that traffic rate is greater than 1Mbps, the delay is reduced by 6.23%, and the throughput of the system is increased by 6.85%over the whole interval.