No abstract

Driving style recognition plays a key role in ensuring driving safety and improving vehicle traffic efficiency. With the development of sensing technology, data-driven methods are more widely uesd to recognize driving style. However, adequately labeling data is difficult for supervised learning methods, while the classification accuracy is not sufficiently approved for unsupervised learning methods. This paper proposes a new driving style recognition method based on Tri-CatBoost, which takes CatBoost as base classifier and effectively utilizes the semi-supervised learning mechanism to reduce the dependency on data labels and improve the recognition ability. First, statistical features were extracted from the velocity, acceleration and jerk signals to fully characterize the driving style. The kernel principal component analysis was used to perform nonlinear feature dimension reduction to eliminate feature coupling. CatBoost is an ensemble of symmetric decision trees whose symmetry structure endows it fewer parameters, faster training and testing, and a higher accuracy. Then, a Tri-Training strategy is employed to integrate the base CatBoost classifiers and fully exploit the unlabeled data to generate pseudo-labels, by which the base CatBoost classifiers are optimized. To verify the effectiveness of the proposed method, a large number of experiments are performed on the UAH DriveSet. When the labeling ratio is 50%, the macro precision of Tri-CatBoost is 0.721, which is 15.7% higher than that of unsupervised K-means, 1.6% higher than that of supervised GBDT, 3.7% higher than that of Self-Training, 0.7% higher than that of Co-training, 1.5% higher than that of random forest, 6.7% higher than that of decision tree, and 4.0% higher than that of multilayer perceptron. The macro recall of Tri-CatBoost is 0.744, which is also higher than other methods. The experimental results fully demonstrate the superiority of this work in reducing label dependency and improving recognition performance, which indicates that the proposed method has broad application prospects.

Data-driven methods are widely applied to predict the remaining useful life (RUL) of lithium-ion batteries, but they generally suffer from two limitations: (i) the potentials of features are not fully exploited, and (ii) the parameters of the prediction model are difficult to determine. To address this challenge, this paper proposes a new data-driven method using feature enhancement and adaptive optimization. First, the features of battery aging are extracted online. Then, the feature enhancement technologies, including the box-cox transformation and the time window processing, are used to fully exploit the potential of features. The box-cox transformation can improve the correlation between the features and the aging status of the battery, and the time window processing can effectively exploit the time information hidden in the historical features sequence. Based on this, gradient boosting decision trees are used to establish the RUL prediction model, and the particle swarm optimization is used to adaptively optimize the model parameters. This method was applied on actual lithium-ion battery degradation data, and the experimental results show that the proposed model is superior to traditional prediction methods in terms of accuracy.

No abstract

No abstract