Research On Health Management Methods Of Lead-acid Batteries In Data Center

Valve-Regulated Lead-Acid(VRLA)batteries are widely used in data center distribution systems as the main energy storage equipment for uninterrupted power supply(UPS).Battery failure or abnormality will directly affect the safe and stable operation of the data center.Therefore,it is necessary to detect and analyze the health status of the battery.However,in this type of precision data center,there are few abnormalities in the battery,which is characterized by high sample imbalance;In data center,batteries are usually in a floating charge state,with fewer charging and discharging times,and the obtained battery data is under a single working condition.These problems have brought great difficulties to the health status detection of the battery.(1)A hierarchical clustering anomaly detection algorithm based on K-shape is proposed based on the trend of battery voltage variation.The algorithm realizes the anomaly detection of the battery by grouping the battery and clustering on a monthly time scale.Then use the average voltage sorting table of the charging cycle to determine the abnormal battery,and finally determine whether the battery is abnormal.(2)Combining battery voltage and resistance monitoring data,statistical methods are used to obtain characteristics that characterize changes in battery health.The Layda criterion is used to identify battery health characteristics,achieving real-time detection of abnormal batteries.After verification,the algorithm detects abnormal batteries at least 7 days earlier than the above K-shape-based hierarchical clustering anomaly detection algorithm.At the same time,according to the different performance requirements of different data centers for VRLA batteries,the corresponding anomaly thresholds can be set to realize the anomaly detection of VRLA batteries.(3)In order to detect battery anomalies as early as possible,a feature enhancement mothed based anomaly prediction model for VRLA batteries was proposed.After verification,the model detects impending battery anomalies at least one day earlier than the real-time detection method described above.Compared to the VRLA battery replacement strategy recommended in the Recommended Practice for Maintenance,Testing,and Updating of VRLA Battery Packs(IEEE STD 1188-2005),detect impending abnormalities in the battery at least 3 days in advance.