This work addresses the task of road hazard state evaluation in traffic scenarios with MEC(multiple access edge computing) available information. Here we concentrate on inter-vehicle energy relationship, and devote to propose a prediction method, which could predict both regional overall hazard state and hazard distribution characteristics. A classical learning network, BP neutral network, is employed here for corresponding prediction process, while Ising model is used to explain road hazard state. The network training progress is discussed in detail. A validation data set is used to verify the effectiveness of proposed approach. Simulation results show that proposed method is able to make accuracy prediction of road hazard state. Resulting hazard values could be utilized to help driver avoid potential dangers and generate valid cruising trajectory.