传感器网络中异常数据实时检测算法

Real-time Detection Algorithm for Anomaly Data in Sensor Networks

如何实时检测传感器网络中异常数据是一项非常重要的工作。通过对线性自回归的分析法,给出传感器数据流的预测模型及其预测机制。当预测出现失败时,给出了一种预测模型自动调整策略,以降低预测误差。基于该预测模型,提出了一种异常数据检测方法,通过计算当前时刻的预测误差与平均预测误差的比值,比较该比值与预先设定的阈值的大小,以检测该时刻数据是否为异常数据。基于该方法,提出了异常事件检测和数据压缩处理的算法。仿真结果验证了预测模型的正确性和有效性,表明该模型能够实时检测异常事件和压缩数据处理。

Anomaly data detection is of great practical application meaning for sensor networks. Variable-self linear regression was adopted and a predictive model including its＇ predictive method over data stream in sensor networks was proposed. An adaptive strategy for the predictive model was proposed in order to decrease the rate of predictive errors when prediction was defected. A method of detecting anomaly data was give based on this model, which was used to detect anomaly events and compress data via computing the specific value between current prediction error and the average of prediction error. This specific value and threshold was compared for the sake of detecting whether the current data was anomaly or not .On the basis of it, an anomaly events detection and data compression algorithm was proposed. Analytical and experimental results show that this algorithm is effective in anomaly events detection and data compression.