Sensor Data Anomaly Detection And Correction Algorithm For Edge Computing

Sensing data inevitably occurs in the process of acquisition and transmission.Anomaly detection can effectively improve the overall quality of sensor data and tap potential information.The traditional anomaly detection cloud computing processing method has the problems of high delay and large bandwidth consumption.With the development of cloud computing,big data processing,Internet of Things,and information physics systems,manufacturing companies have put forward higher requirements for abnormal data detection in the manufacturing process.Edge calculation refers to the calculation task at the edge end,with high availability,low latency,high real-time and so on.The edge detection method is used to study the anomaly detection algorithm,and the data processing task of the anomaly detection is sunk to the data acquisition end,which can realize the high-efficiency and high-reality abnormal data detection.Using the advantages of edge calculation,the sensor data anomaly detection and estimation correction at the data acquisition end can optimize the overall quality of the data and improve the utilization efficiency under the premise of efficient processing,and has high application value.This thesis first analyzes the current edge computing reference architecture and main implementation modes,and combines data processing tasks with data acquisition to design a data acquisition and analysis system based on Ethernet mode for edge computing.The software and hardware architecture and business process design of the system were carried out.Then based on the characteristics of time series data,the single source sensor data is used.Based on the distance-based anomaly detection algorithm,an anomaly detection algorithm based on offset distance is proposed in this thesis.The offset distance of the sensor data is monitored by the timing continuity of the sensor data itself,and the possibility of abnormality of the sensor data is determined according to the abnormality level of the data.In order to effectively restore the actual value of the anomaly data,the single-layer linear network model is used for data prediction analysis and the historical data is used to train the single-layer linear network model,the actual value of theabnormal data is estimated according to the prediction of the model.Finally,an example of data acquisition and analysis system for edge computing is carried out for the semi-automatic production line of Huagong Science and Technology Internet of Things Research Center.The temperature and humidity data of hydrological system in a certain region is selected as experimental data for performance analysis of the algorithm.The experimental results show that the sensor data anomaly detection and correction algorithm proposed in this thesis has high detection rate and good effect.The curve after abnormal detection and correction is relatively smooth,which can reflect the actual dynamic trend of data more accurately.