There are problems with trust initialization and low evaluation accuracy in the Internet of Vehicles (IoV). This paper proposes a trust management scheme for road prediction based on Bayesian network methods. Firstly, environmental factors are introduced as variables in the Bayesian network for road prediction in trust initialization. The predicted results are used to evaluate vehicle messages for obtain initial trust values and improve model evaluation accuracy. Secondly, in terms of trust processing, an adaptive forgetting factor and cosine similarity are introduced to refine the trust evaluation method, which enhance model evaluation accuracy and resistance to attacks. The experiment shows that the model has a high evaluation accuracy when the proportion of malicious nodes is 40%. The proposed model can effectively identify malicious nodes and has good resistance to switch attacks.