%0 Journal Article
%T Research on Outlier Detection Algorithm for Evaluation of Battery System Safety
%A Changhao Piao
%A Zhi Huang
%A Ling Su
%A Sheng Lu
%J Advances in Mechanical Engineering
%D 2014
%I SAGE Publications
%R 10.1155/2014/830402
%X Battery system is the key part of the electric vehicle. To realize outlier detection in the running process of battery system effectively, a new high-dimensional data stream outlier detection algorithm (DSOD) based on angle distribution is proposed. First, in order to improve the algorithm stability in high-dimensional space, the method of angle distribution-based outlier detection algorithm is employed. Second, to reduce the computational complexity, a small-scale calculation set of data stream is established, which is composed of normal set and border set. For the purpose of solving the problem of concept drift, an update mechanism for the normal set and border set is developed in this paper. By this way, these hidden abnormal points will be rapidly detected. The experimental results on real data sets and battery system simulation data sets demonstrate that DSOD is more efficient than Simple variance of angles (Simple VOA) and angle-based outlier detection (ABOD) and is very suitable for the evaluation of battery system safety. 1. Introduction In a world where environment protection and energy conservation are receiving extensive concerns, the development of electric vehicles has taken on an accelerated pace [1, 2]. Battery system technology is one of the critical technologies in electric vehicles, which is widely adopted in modern battery management system [3, 4]. With the popularization of electric vehicles, the safety performance of battery system has been attracting more and more attention. Outlier detection is an important branch of data mining area. As the technology advances, a lot of applications will produce time sequenced, massive, and vertiginous data streams, such as E-commerce, network flow monitor, and wireless communication. The running condition of battery system will also produce such data. However, there are some additional characteristics in the battery system data, like high dimension, concept drift, and so on [5]. This makes it more difficult to evaluate the safety performance of the battery system and disturbs the normal running of electric vehicles. Therefore, finding effective methods of mining unsafe factors from massive data is a very important need. Since the rise of outlier detection research, an increasing number of studies have been carried out in many famous research organizations and academic units, and they have achieved fruitful research results. In terms of statistic-based method, an outlier detection approach based on statistic for wireless sensor networks is proposed [6]. An improved online outlier detection method
%U http://www.hindawi.com/journals/ame/2014/830402/