Aiming at the problem that the existing ensemble classification methods have poor learning ability for unbalanced time series data, the idea of optimizing component algorithm performance and integration strategy was adopted, and based on the heterogeneous ensemble method Hierarchical Vote Collective of Transformation-based Ensembles (HIVE-COTE), an ensemble classification algorithm IMHIVE-COTE (Imbalanced Hierarchical Vote Collective of Transformation-based Ensembles) for unbalanced time series was proposed. The algorithm mainly contains two improvements:first, a new unbalanced classification component SBST-HESCA (SMOM ( K-NN-based Synthetic Minority Oversampling algorithm for Multiclass imbalance problems) & Boosting into ST-HESCA (Shapelet Transformation-Heterogeneous Ensembles of Standard Classification Algorithm) algorithm) was added, the idea of boosting combined with resampling was introduced, and the sample weights were updated through cross-validation prediction results, so as to make the re-sampling process of the dataset more conducive to improving the classification quality of minority samples; second, the HIVE-COTE calculation framework was improved by combining the SBST-HESCA component, and the weight of the component algorithm was optimized, so that the unbalanced time series classification algorithm had higher voting weight to the classification result, as a result, the overall classification quality of the ensemble algorithm was further improved. The experimental part verified and analyzed the performance of IMHIVE-COTE:compared with the comparison methods, IMHIVE-COTE had the highest overall classification evaluation, and the best, the best and third overall classification evaluation on three unbalanced classification indexes. It is proved that IMHIVE-COTE's ability to solve the problem of unbalanced time series classification is significantly better.

Focusing on low quality of u-shapelets (unsupervised shapelets) in time series clustering based on u-shapelets, a time series clustering method based on optimal u-shapelets named DivUshapCluster was proposed. Firstly, the influence of different subsequence quality assessment methods on time series clustering results based on u-shapelets was discussed. Secondly, the selected best subsequence quality assessment method was used to evaluate the quality of the u-shapelet candidates. Then, the diversified top- k query technology was used to remove redundant u-shapelets from the u-shapelet candidates and select the optimal u-shapelets. Finally, the optimal u-shapelets set was used to transform the original dataset, so as to improve the accuracy of the time series clustering. The experimental results show that the DivUshapCluster method is superior to the traditional time series clustering methods in terms of clustering accuracy. Compared with the BruteForce method and the SUSh method, the average clustering accuracy of DivUshapCluster method is increased by 18.80% and 19.38% on 22 datasets, respectively. The proposed method can effectively improve the clustering accuracy of time series in the case of ensuring the overall efficiency.

Shapelet is a kind of recognizable time series sub-sequence, by identifying the local characteristics to achieve the purpose of accurate classification of time series. The original shapelet discovery algorithm has low efficiency, and much work has focused on improving the efficiency of shapelet discovery. However, for the time series with trend change, the typical time series representation is used for shapelet discovery, which tends to cause the loss of trend information in the sequence. In order to solve this problem, a new trend-based diversified top- k shapelet classification method was proposed. Firstly, the method of trend feature symbolization was used to represent the trend information of time series. Then, the shapelet candidate set was obtained according to the trend signature of the sequence. Finally, the most representative k shapelets were selected from the candidate set by introducing the diversifying top- k query algorithm. Experimental results of time series classification show that compared with the traditional classification algorithms, the accuracy of the proposed method was improved on 11 experimental data sets; compared with FastShapelet algorithm, the efficiency was improved, the running time of the proposed method was shortened, specially for the data with obvious trend information. The experimental results indicate that the proposed method can effectively improve the accuracy and the effciency of time series classification.

Focusing on the issue that shapelets candidates can be very similar in time series classification by shapelets transform, a diversified top-k shapelets transform method named DivTopKShapelet was proposed. In DivTopKShapelet, the diversified top-k query method was used to filter similar shapelets and select the k most representative shapelets. Then the optimal shapelets was used to transform data, so as to improve the accuracy and time efficiency of typical time series classification method. Experimental results show that compared with clustering based shapelets classification method (ClusterShapelet) and coverage based shapelets classification method (ShapeletSelction), DivTopKShapelet method can not only improve the traditional time series classification method, but also increase the accuracy by 48.43% and 32.61% at most; at the same time, the proposed method can enhance the computational efficiency in 15 data sets, which is at least 1.09 times and at most 287.8 times.