Knowledge Discovery Method from Abnormal Monitoring Data

Abnormal monitoring data contains a wealth of information and is also the concern object of people. For the time series characteristics of monitoring data, using the time series data mining techniques to discover the regularity knowledge from the abnormal sensor monitoring data is feasible method to help the supervisors identify the reason causing the exceptional fluctuation automatically and make the correct decisions promptly. Abnormal time series clustering method based on DTW distance is proposed firstly, thus the typical time series patterns can be obtained. From which the important shape indexes, such as gradient K, regression coefficients b and mean square deviation, can be extracted and filtered from about fifteen parameters based on piecewise shape measure method. At last, the knowledge used to recognize the exceptional pattern can be abstracted from the shape feature table and represented with the first order predicate logic language. As an example, this set of knowledge discovery method has been used in one high gas coal mine and proved the important promotion application value in the sensor monitoring field.