Data Verification Method Based on Graph Attention Network

Abstract

In the context of the rapid development of autonomous driving, data security in the Internet of vehicles has an important impact on the actual road safety. Anomaly detection is the main means of data validity analysis in the Internet of vehicles. The existing methods do not make special use of the correlation of data features for data verification. In this paper, we propose a data verification algorithm based on graph attention network (GAT) model. In our graph attention network model, we combine the relationship between data features with the temporal relationship of the same feature, and use the graph attention network to combine different features under different time series through the graph relation. Later, based on the obtained feature correlation, our algorithm uses Gate Recurrent Unit (GRU) to verify data features. To verify the effectiveness of our algorithm, we use ablation studies. Experimental results show that the combination of relevant features and time series can verify data more effectively. At the same time, compared with other methods, our algorithm has better recall rate and accuracy, and can effectively detect wrong data.