The time series classification algorithm based on Shapelet has the characteristics of interpretability, high classifica-tion accuracy and fast classification speed. Among these Shapelet-based algorithms, learning Shapelet algorithm does not rely on a single classifier, and Shapelet that is not in the original time series can be learned, which can achieve a high classification accuracy and ensure that Shapelet discovery and classifier construction are completed at the same time. However, if too many Shapelets are generated, it will increase the dependent parameters, resulting in too long training time, low classification speed, and difficult dynamic updates. And similar redundancy Shapelets will reduce the interpretability of the classification. This study proposes a new selective extraction algorithm to select Shapelet candidate set and change the learning method to accelerate the learning process of Shapelet and puts forward two optimization strategies. By using time series clustering for the original training set, Shapelets not in the original time series can be obtained. Meanwhile, a voting mechanism is added into the selective extraction algorithm to solve the problem of excessive Shapelet generation. Experiments show that the proposed algorithm can improve the training speed while maintaining high accuracy.