The existing trajectory prediction algorithms focus on the mobility pattern of objects and simulate the traffic flow via mathematical models which are inaccurate at describing network-constraint objects. In order to cope with this problem, a self-adaptive parameter selection trajectory prediction model based on hidden Markov models (SATP) is proposed. The new model can efficiently cluster and partition location big data, and extract the hidden and observable states by using a density-based clustering approach in order to reduce the number of states in HMM. SATP can automatically select the parameters on the input trajectories and avoid the problems of discontinuous hidden states and state retention. Experimental results demonstrate that the SATP model has high prediction accuracy with less time overhead. The average prediction accuracy of SATP is 84.1% while the moving objects have a random changing speed, which is higher than the Na?ve algorithm with an average gap of 46.7%.