基于多维指标关联的物联网装备异常预测方法

Anomaly prediction for IoT equipment based on correlation of multidimensional indicators

针对高动态物联网装备多维指标相互作用使得装备状态时变性极强,难以实现装备状态精准的评估与异常检测的问题,提出一种面向多维指标关联的装备状态异常预测方法。该方法通过计算物联网装备状态海量指标的斯皮尔曼相关系数,得到多维指标间的相关性,利用主成分分析对与目的检测指标强相关的其他指标进行特征提取,将提取结果和目的指标本身历史数据作为基于长短期记忆神经网络的装备状态感知模型的输入,进而对目的指标未来状态趋势进行精准预测;在此基础上,利用无监督的DBSCAN算法对装备状态感知模型的输出结果进行分析,定位目的指标未来可能出现的异常,实现了装备状态的评估。实验研究结果表明:该方案能够高精度预测物联网装备未来异常的发生,保护物联网装备免受潜在异常的影响,增强物联网装备的稳定性。

In highly dynamic IoT devices,the multi-dimensional indicators that interact with each other and change frequently over time bring about intensely time-varying devices’state,making it difficult to accurately evaluate the status of the equipment and detect anomalies.An anomaly prediction method for IOT equipment based on correlation of multidimensional indicators was proposed.The method obtained the correlation between multidimensional indicators by calculating the Spearman correlation coefficient of the massive indicators in IoT equipment.The principal components analysis was used to extract features from other indicators that are strongly related to the target indicator.The extraction results and the historical data of the target indicator itself were used as the inputs of the equipment state-aware model based on long-term short-term memory(LSTM)neural network,then the future status trend of the target indicator could be accurately predicted.On this basis,the unsupervised Density-Based Spatial Clustering of Applications with Noise(DBSCAN)algorithm was used to analyze the output results of the equipment state-aware model and locate the future anomalies of the target indicator,realizing equipment status assessment.The performance of the anomaly prediction method is verified by numerical simulations,and the result indicates that the method can predict future abnormalities of IoT equipment status with high precision,thus protecting equipment from potential abnormalities and enhancing the stability of equipment.