Performance analysis of driving style identification by using wolf-inspired evolutionary clustering

Driving style identification is an essential task in vehicle technology to improve security, driving experience, and customise targeted energy management strategies. Commonly used clustering algorithms, like K-means plus, occasionally suffer from convergence to local optimum and unreasonable setting of initial centroid placement. This paper proposed a neural wolf-inspired evolutionary clustering method for driving style identification. Driving styles show distinctive features in a time series, thus the fast Fourier transform (FFT) is used to convert time series data into energy data corresponding to the main frequency in the frequency domain. Then grey wolf optimisation (GWO) algorithm as a bio-inspired optimisation algorithm is formulated to search the global optimal clustering centroids. The proposed algorithm is compared with K-means plus, Gaussian mixture model (GMM), and PSO-inspired clustering method to identify three distinct driving styles. The data investigated were collected on a driver-in-the-loop intelligent simulation platform. Silhouette coefficient was selected to evaluate the clustering effect of the test dataset in the clustering model implemented by the proposed algorithm, its five times average reached 0.9531, which is 0.0280 higher than K-means plus, 0.0124 higher than GMM, and 0.0106 higher than PSO-inspired clustering method.