A self-adaptive correction method for sensor data deviation

Abstract

The present invention provides a sensor data deviation self-adaptive correction method in the technical field of sensor data analysis, including the following steps: Step S10, acquiring the monitoring data of each sensor, and initializing the failure of each sensor based on the change trend of the monitoring data Judgment; step S20, create a mathematical model of each sensor based on the monitoring data, and use the mathematical model to perform online analysis and self-adaptive correction on the monitoring data of each sensor deviation, so as to realize accurate judgment; step S30, the corrected Check the monitoring data. The invention has the advantages of realizing real-time analysis and correction of sensor data deviation, greatly improving the detection rate and detection efficiency of sensor abnormalities, and greatly reducing the false alarm rate of sensors.

Description

technical field [0001] The invention relates to the technical field of sensor data analysis, in particular to a sensor data deviation self-adaptive correction method. Background technique [0002] With the development of information technology and sensor technology, the degree of informatization, automation, and intelligence continues to deepen. As an important support for on-site information acquisition and processing, sensors are increasingly used in various fields such as manufacturing, environmental monitoring, national defense, and logistics. . The quality of data collected by sensors directly affects the level of automation and intelligence. However, due to the long-term exposure of the sensor to harsh environments such as the open air, environmental factors, sudden node failures, network structure damage and other factors cause the sensor to experience zero-point drift and other faults, resulting in false positives or false positives of the sensor, reducing the relia...

AI Technical Summary

Problems solved by technology

[0004]Nearest neighbor based methods use several well-defined distance concepts to calculate the distance between each other (similarity distance), if the distance between data and its neighbors If it is too large, it will be marked as an outlier, but for multi-mode and high-dimensional data sets, the computational complexity of the distance between each data will increase geometrically, which is expensive

[0005] Classification-based methods use a set of data samples to learn a classification model, and then divide the test samples into the learned classification model according to the classification model, but the computational complexity And choosing an appropriate kernel function is the main bottleneck of the model's learning

[0006] Clustering based approach is to group similar samples into classes with similar behaviour, if samples do not belong to a class or their clusters are smaller than other clusters If there are many, mark them as outliers. This method can be used in the incremental model, and the scalability is relatively good. When new samples are input into the system, outliers can be found in time. However, it depends on the choice of cluster width, and in multivariate Calculating the distance between samples in the data is expensive

[0007]The method based on spectrum analysis is to use the principal components to find the normal behavior pattern in the data. It is necessary to construct the combination of the first few main components first. Those that do not conform to this structure samples are treated as outliers, but the analysis relies on the correlation matrix of the normal pattern, and accurate estimation of the correlation matrix is ​​computationally expensive

[0008] To sum up, the traditional sensor data deviation correction method has high computational complexity, and it is easy to cause data disaster under large-scale data, which seriously affects the detection efficiency of sensor nodes , unable to meet the real-time analysis and correction requirements of sensor data

Images