Data-driven cloud-based intelligent gear-shift decision strategy of vehicle considering driving behavior and environment

Abstract

The gear-shift strategy plays a decisive role in reducing fuel consumption and improving drivability. Based on naturalistic driving data and a machine learning method, this study proposes an intelligent gear-shift decision-making system that considers fuel consumption, long short-term driving behavior, and driving environment, as well as an incremental updating strategy for the system. First, the unsupervised classification of long-term driving behavior is realized based on the Gaussian mixture model and KL-divergence, and the short-term driving behavior is modeled based on a long short-term memory neural network. Second, a slope estimation method based on multi-data fusion is proposed. Dynamic programming and a multi-objective genetic algorithm are used to solve the optimal gear-shift decision-making problem considering fuel consumption, long short-term driving behavior, and the driving environment. Finally, a gear-shift decision-making model based on the random forest is proposed. In a long-distance road condition, vehicles under the proposed mild, normal, and aggressive strategies consumed 0.79%, 1.17% and 4.21% less fuel than that of the standard original vehicle strategy. The power index is 35.78, 8.04, and 20.32%, and the auxiliary braking index is 35.78, 8.07 and 11.88% more than that of the original strategy, respectively. The number of shifts of the three strategies was reduced by 56.92, 48.74, and 50.36%, respectively. Moreover, the decision-making model can effectively self-learn and incrementally update the gear-shift decision-making strategy, continuously adapting to changes in the driving behavior and environment.