Research On High-dimensional Data Stream Outlier Detection Algorithm For Battery Safety Assessment

With the developing of science and technology and human society,there is a huge amount of data accumulated in human’s daily life,and a big data era is coming.Different from the static data sets,the real data set is often exists in the form of a data streams,and the data flow usually has the following characteristics: high dimensional,infinite changes,concept drift.While the battery condition data as a typical high dimensional data stream,and how to detect the potential outliers(such as the fault data hiding in the battery condition data stream)which contains the valuable information has become an urgent problem to be solved.Through the analysis of domestic and foreign outlier detection algorithm for high-dimensional data stream,and in order to detect potential outliers timely in the case of traditional outlier detection algorithm does not apply well in high-dimensional data stream.A robust preprocessing and feature extraction method for real time angle based outlier detection algorithm is proposed to improve performance and stability of traditional angle based method in high-dimensional space.First,k-nearest neighbor and similarity method are applied for data clustering analysis.Then use angle vector feature extraction method to lessen the data dimension for improving outlier detection algorithm computation speed.Finally,the angle based method is used to identify outliers in high-dimensional data streams.The experimental results in synthetic data set and UCI real-world data sets corroborated the suggested approach which can improve the robustness and lessen the running time of the algorithm in high-dimensional space under the condition of outlier detection,and provides a theoretical basis for the rapid detection of high dimensional outliers in the battery system model.In this paper,in order to apply the proposed algorithm to the battery system security evaluation,considering the distribution characteristics of the data flow,the proposed algorithm is improved and optimized.And the experimental data set of the improved algorithm is obtained through the battery in loop simulation platform.The experimental results in battery condition data set corroborated the improved algorithm can detect the outlier quickly and effectively,which contains the fault information,and providing a reference for safety assessment of the battery system.