Abstract: The trust model for massive terminals has the problems of coarse computing granularity and high resource consumption. A new dynamic trust evaluation method for power terminals based on graph model and similarity calculation(GASTEM) is proposed. Firstly, by establishing a graph model and combining fuzzy reasoning to calculate the direct trust of nodes, using similarity measurement method to calculate indirect trust, and the size of the comprehensive trust value through weighted summation is determined. Then, the AMRA model is used to predict the comprehensive trust value. The properties of the node are determined by comparing the residuals of the predicted value, actual value, and historical threshold. Finally, the experimental results demonstrate that this trust evaluation method can timely and accurately discover and detect malicious terminal nodes.