Sensor stream pattern mining for automatic anomaly recognition and intervention

该项目将开发一个通用框架，用于准确自动识别多变量传感器数据流中的有意义异常，这些异常需要采取行动以避免有害事件，并允许自动干预以有效缓解。现有的异常识别算法错过了许多模式，并且手动地将共同出现的流模式与异常相关联是低效且容易出错的。该项目预计将制定方法，拦截共同出现的模式的组合，以确定异常是什么，并确定异常及其阶段，表明干预的必要性。该项目将推进传感器流数据挖掘技术，并实现传感器监视和自动机械干预的一般应用。

This project will develop a general framework of accurate automatic recognition of meaningful anomalies in multivariate sensor data streams that require action to avoid detrimental events and allow automatic intervention for efficient mitigation. Existing anomaly recognition algorithms miss many patterns and manually relating co-occurring stream patterns to an anomaly is inefficient and error-prone. The project expects to develop methods for intercepting a combination of co-occurring patterns to ascertain what an anomaly is and identify the anomaly and its stages that indicate the necessity of intervention. This project will advance techniques for sensor stream data mining and enable general applications of sensor surveillance and automatic mechanical intervention.