中文摘要：

无线传感网络经常被部署于条件恶劣、无人值守的环境中,受到恶劣天气、软硬件故障、能量不足或者恶意攻击等因素的影响,传感器节点感知数据的缺失或错误难以避免。因此,传感器数据流的离群检测对于提高系统可用性至关重要。提出一种基于约简策略与自适应SVDD（Support Vector Data Description）的离群检测方法（RASVDD）,该方法首先使用基于马氏距离标准的方法约简数据集,有效地减少了训练样本;然后利用基于数据分布密度准则和数据流时间相关性自适应更新决策模型。针对Intel Berkeley数据集及Sensor Scope System数据集的仿真实验表明,RASVDD的离群检测正确率TPR（True Positive Rate）平均达98%,误报率FPR（False Positive Rate）平均仅为1%,并且与传统的SVDD算法相比,RASVDD决策模型训练时间减少了20%以上。

英文摘要：

Wireless sensor networks are often deployed in the harsh and unattended environment,and the sensor data loss or error usually happens for the sake of bad weather,hardware or software fault,energy dissipation or the adverse attack. Outlier detection of the sensor data streams is critical for improving the system＇s availability. In this paper,an outlier detection method（ RASVDD） based on the data reduction and adaptive SVDD is proposed. RASVDD uses the Mahalanobis distance criterion to reduce the data set and the training samples,and then the data distribution density criterion and the temporal correlation of data stream are applied to update the training model adaptively.The simulation results for the Intel Berkeley dataset and the SensorS cope System dataset showed that,RASVDD had an average true positive rate of 98% and an average false positive rate of 1%,and reduced the model training time more than 20% compared to traditional SVDD.