Anomaly prediction for IoT equipment based on correlation of multidimensional indicators

DOI:	10.11809/bqzbgcxb2024.01.010
Keywords:	correlation of multidimensional indicators; principal components analysis; long term short term memory neural network; density based spatial clustering of applications with noise; anomaly prediction
Abstract:	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.
Published:	2024-01-28
PDF