一类支持向量机的设备状态自适应报警方法

Self-Adaptive Alarm Method for Equipment Condition Based on One-Class Support Vector Machine

为了提高对异常状态识别的适应性和有效性,提出了一种基于一类支持向量机的设备状态自适应报警方法.该方法使用一类支持向量机的在线算法,动态估计监测参数在高维特征空间中的最优分布区域,将新数据与上一时刻分布区域的相对距离作为异常指标,描述监测参数的统计特征变化,辨识出设备的异常状态.通过对仿真数据的报警效果分析,以及将该方法应用于对加热炉风机的振动监测中,得到的异常报警结果能够满足实际监测的需要,证明该方法具有异常的识别敏感性、缓慢劣化包容性和状态迁移适应性的特点.

To improve the adaptability and effectiveness of recognition on abnormal condition, a self-adaptive alarm method for equipment condition based on one-class support vector machine （OC-SVM） is proposed. The optimum distribution area of monitoring parameters in high-dimensional feature space is dynamically estimated with on-line algorithm of OC-SVM. The abnormal index, determined by the relative distance between the new data and the distribution area at the previous moment, describes the statistical feature variation of monitoring parameters and identifies the abnormal condition of the equipments. The characteristics, such as the sensitivity in abnormal condition recognition, the toleration in slow deterioration and the adaptability in condition alternation, are verified by simulation data. The present method is further applied to the vibration monitoring of heating furnace fan. The alarms under the actual abnormal condition meet the demand of equipment monitoring.