A Novel k-MPSO Clustering Algorithm for the Construction of Typical Driving Cycles

Abstract

The practical driving cycle is of great significance in studying the control strategy of vehicles, and effective clustering of micro-trips is the key to obtaining the typical driving cycle. A novel and efficient method for constructing typical driving cycles is presented in this paper. First, by combining the preying behavior and random behavior of the artificial fish swarm algorithm (AFSA) with particle swarm optimization (PSO), a modified particle swarm optimization (MPSO) is proposed. By comparing the means and standard deviations of the optima, MPSO is verified as much more accurate and stable than PSO, AFSA, select particle swarm optimization (SPSO) and cross particle swarm optimization (CPSO) in the optimization calculation of four typical multi-modal benchmark functions. Second, by applying MPSO to optimize the k-means algorithm, the k-MPSO clustering algorithm is obtained. In the case of clustering the Iris standard data set, the average error rates of the k-means algorithm and k-MPSO clustering algorithm are 11.6% and 7.8%, respectively, which means that the k-MPSO clustering algorithm has a stronger searching ability. Finally, with the ECAN Tools software, real-world driving data that include thousands of micro-trips in Jinan are collected, and 19 representative characteristic parameters are selected to fully describe the driving conditions. After principal component analysis (PCA), the k-MPSO clustering algorithm method is applied to cluster the micro-trips into three classes and construct the typical driving cycle in Jinan.

Keywords

• Artificial fish swarm algorithm
• driving cycle
• k-MPSO clustering algorithm
• particle swarm optimization