Research Of ARIMA Prediction Model-Based Cluster Algorithm In Ad Hoc Networks

Ad Hoc networks formed by a collection of mobile nodes through radio links are multi-hop and temporary autonomous systems.They are typically characterized by their un-center, self-orgnization, dynamically reconfigurable network topology and limited resource. The architecture of Ad Hoc networks has two structures:plane and cluster. In the architecture of plane, with the expansion of network and the enhancement of mobility, the information of maintance exchanges frequently, leading to increased network latency and network congestion.The main method to solve the problem is to use cluster structure, but the cluster structure of the formation and maintenance depend on the particular clustering algorithm, so the clustering aigorithm become a very important research field in Ad Hoc networks.The paper analyzes and compares serval typical clustering algorithms, especially on-demand Weighted Clustering Algorithm (WCA). On the base of WCA, we propose a Location-based Weighted Clustering Algorithm based on ARIMA Prediction Model (ARP-LWCA).The algorithm adopts the accumulated holding time of node, the deviation of the actual degree and ideal degree, node residual energy to select the cluster head, and the accumulated holding time of node is as the main indicators to measure the stability of cluster head. The node location information provided by GPS system and the geographical position the next moment of the node can be predicted according to the established ARIMA model, and then the predicted value of accumulated holding time of node can be computed. With the classics routing protocol CBRP, and in the route maintenance stage the accumulated holding time of cluster head is predicted using the ARIMA prediction model. The predicted value of accumulated holding time is compared to the warning threshold, if the link predicted will be unstable, then the pre-restoration process is started before the link breaks off to look for the suitable routing, to avoid the frequent fracture of the network link, to reduce the dynamic changes of network topology on the impact of routing protocols, to maintain the stability of cluster structure, then the network will run well and the network performance is greatly improved.By the network simulator2(NS2), ARP-LWCA, RLWCA (Restrictive Location-based Weighted Clustering Algorithm) and LOWID are simulated and compared respectively from three different view, such as changing the node maximum speed, changing the pause time and changing the number of the nodes. Simulation results show that, compared to LOWID and RLWCA, ARP-LWCA greatly improved the packet delivery rate and reduce the network overhead normalized, the number of the routing interruption is significantly reduced, and then the network performance is improved.