An Incremental BP Neural Network Based Spurious Message Filter for VANET

In order to protect legitimate vehicles from cheating by spurious alert messages, we propose a general filter model for Vehicular Ad-hoc Network (VANET) to distinguish spurious messages from valid ones. It is a two-layer filter, the coarse filter is responsible for rapid filtration and the fine filter is for accurate filtration. The data flow should pass through them to get the classification results. The coarse filter makes a judgment by combining several sources of information such as timeliness of the report and correlation of the accident location while the fine filter is based on Back Propagation Neural Network (BPNN) which includes an incremental learning part. The BPNN module refers to vehicles´ reputations and behaviors in response to an event, and the support from neighbors will also be a great help. In this paper, we compare the filtering effect of incremental BPNN with several commonly used decision logics including majority voting, weighted voting and Bayesian method. The simulation results show that our scheme performs better both in filtering reliability and stability.