Shifting Classification of Automatic Transmission Based on Shapelet Transform
Wujun Zou, Ye Wang, Jiarong Pu, Chaojie Zhong · 2022 4th International Conference on Communications, Information System and Computer Engineering (CISCE) · 2022
During the driving process of the vehicle, a large amount of time series containing shifting information was continuously generated. It is a laborious process to accurately detect the factory quality inspection and shifting characteristics of automatic transmissions through these data. In this context, this paper proposed a method to classify shifting data using shapelet transform, which is a local shape similarity classification method based on the most discriminative subsequence in a time series. In consideration of the low efficiency of traditional shapelet extraction methods, we introduce the idea of mathematical statistical probabilistic models into the extraction of shapelets. More specifically, we first build a probabilistic model of the training dataset and sample some time series from the training dataset by the utilization of distribution estimation algorithm (EDA). Then we use the probabilistic model update method to find discriminative subsequences in the time series as shapelet candidates. We next prune highly similar shapelet candidates through shapelet filters. Finally, we use the best k shapelets to convert each time series in the dataset into k corresponding features, which can be used in conjunction with traditional classifiers to classify the data. The automatic transmission shifting classification experiment shows that the shapelet discovery algorithm based on the probability model proposed in this paper can greatly improve the efficiency of the algorithm while ensuring the classification accuracy.