断路器数据在线异常点检测算法研究

Algorithm for online data outlier detection of circuit breaker

针对常规断路器异常数据检测存在的准确度低、时间复杂度高等问题,引入数据流挖掘技术,提出了一种在滑动时间窗口上的基于局部异常因子的异常点检测算法。该算法分为四个步骤:首先,将时间轴划分为一个个连续的时间窗口;其次,当前时间窗口满了以后,对当前时间窗口内的数据运用滑动平均过滤的方法进行筛选,以此减少检测数据的规模,降低算法的时间复杂度;然后,计算当前时间窗口内可能存在异常的每个数据点的局部异常因子(local outlier factor,LOF),在计算的过程中对部分计算结果进行了优化存储,以此减少重复计算;最后,对当前时间窗口内的局部异常因子值排序,输出LOF>1的数据点。实验表明,该算法较好地提高了断路器异常点在线检测效率。

Aiming at the inaccuracy and high time complexity of outlier data of circuit breaker,this paper introduced the data stream mining technology and proposed a novel algorithm of outlier detection which based on local outlier factor in the sliding window.This algorithm consisted of four distinct steps.Firstly,it divided the time axis into a sequence of continuous timewindows.Secondly,it used the moving average method to filter out candidate outliers in order to reduce the size of the testing data and reduce the time complexity of the algorithm,when the current time window was full.Then,it calculated the local outlier factor （LOF） for each data point which might be candidate oufliers in the current time window-,and saved some calculation results partly to remove the repeated calculations.Finally,these data points whose value of LOF were greater than 1 were output by sorting the values of LOF.The results of experiments show that the new algorithm has greatly improved the efficiency of online outlier detection for circuit breaker.