Narrowband Internet of Things (NB-IoT) is a fundamental technology for massive Machine Type Communications (mMTC) usage scenarios. Before uplink communication, a user equipment (UE) needs to establish connection to the base station by performing a random access (RA) procedure. In this paper, focusing on the contention-based random access of a heterogeneous NB-IoT network, we provide a model based on Markov chains to obtain the steady-state distribution of each UE type in the network, and then further evaluate the system performance. Based on the analysis of the model, we adopt an optimization strategy that configures parameters for each UE type individually. Considering the fairness of the system, we propose an optimization algorithm based on simulated annealing. Simulation results show that the proposed model is valid, and the proposed optimization algorithm can improve the overall throughput while taking count of the fairness among users with an acceptable complexity.