Abstract: It proposes a new load-balancing routing protocol based on neural network model for traffic prediction—NNTP-LBRP.In this protocol，the RBF neural network model is introduced to predict the traffic load of nodes according which the nodes with heavy traffic load will not be selected as intermediate nodes of a valid route，thus avoiding the congestion of nodes and improving the network performance.The traffic load is measured by the length of the queue at interface in MAC layer.In addition，the delay response of destination node，i.e.，the response to the route with lightest traffic load among the several routes，also improves the network performance to some extent.Simulation is done on a ns2-based platform for comparing the properties of the new protocol presented and conventional AODV protocol，and the results show that the packet delivery ratio increases approximately by 10%，the average end-to-end delay reduces by 27% on average，and the normalized routing overhead decreases by 26% on average.

Key words: Ad hoc networks, load balance, neural network, prediction

摘要： 提出了一种基于神经网络预测模型对网络流量负载进行预测的负载均衡协议NNP-LBRP（Load-Balanced Routing Protocol based on Neural Network Prediction model），该协议利用RBF神经网络预测模型对Ad hoc网络中的节点流量负载进行预测，根据预测到的下一时刻的流量负载状况，在节点出现拥塞之前提前更换路径，避免中间节点出现拥塞，以此来提高网络的性能。协议中的流量值是以MAC层接口队列长度来衡量，负载均衡中的负载是流量负载。另外，协议在目的节点处采用了延迟应答策略，即在多路径中选择负载最轻的路径应答，对改善网络的性能也有一定作用。仿真结果与AODV路由协议进行比较，数据包投递率提高了约10%;平均端到端延时平均降低了27%;网络开销平均降低了26%。

关键词: Ad hoc网络, 负载均衡, 神经网络, 预测