A Mixture Model Parameters Estimation Algorithm for Inter-Contact Times in Internet of Vehicles

Communication opportunities among vehicles are important for data transmission over the Internet of Vehicles (IoV). Mixture models are appropriate to describe complex spatial-temporal data. By calculating the expectation of hidden variables in vehicle communication, Expectation Maximization (EM) algorithm solves the maximum likelihood estimation of parameters, and then obtains the mixture model of vehicle communication opportunities. However, the EM algorithm requires multiple iterations and each iteration needs to process all the data. Thus its computational complexity is high. A parameter estimation algorithm with low computational complexity based on Bin Count (BC) and Differential Evolution (DE) (PEBCDE) is proposed. It overcomes the disadvantages of the EM algorithm in solving mixture models for big data. In order to reduce the computational complexity of the mixture models in the IoV, massive data are divided into relatively few time intervals and then counted. According to these few counted values, the parameters of the mixture model are obtained by using DE algorithm. Through modeling and analysis of simulation data and instance data, the PEBCDE algorithm is verified and discussed from two aspects, i.e., accuracy and efficiency. The numerical solution of the probability distribution parameters is obtained, which further provides a more detailed statistical model for the distribution of the opportunity interval of the IoV.