AN INTELLIGENT MEASUREMENT METHOD FOR SENSOR NETWORKS IN LABORATORY TESTING

Abstract

The paper presents a developed method of intelligent measurement in sensor networks during laboratory tests. The method provides an opportunity to improve the accuracy and stability of intelligent measurements in conditions of metrological noise and anomalies that arise during intelligent measurements in laboratory tests. A two-level model for performing intelligent measurements is proposed, with preliminary data processing using edge computing and subsequent final evaluation of input parameters before forming the final measurement result. The reduction of the influence of noise and anomalies on local and global weight coefficients is ensured by the use of bootstrap aggregation, improved by sample prediction. A neural network model that implements the principles of intelligent measurements is proposed. A simulation experiment demonstrated an increase in the accuracy of intelligent measurements in the developed method by 9.98% compared to the average accuracy of similar approaches. The reliability of the method in conditions of noise and anomalies in the sensor network was increased by 35.9%. Bootstrap aggregation ensured the reproducibility of measurement results, which is an important condition for laboratory testing. The limitation of the method is the need to stabilize the prediction accuracy at a value higher than the measurement accuracy of the sensors. Stabilization is proposed to be achieved through adaptive weighted regression and forecast coordination. The developed method can be applied to self-driven laboratories and classical research laboratories, the automation of which is a priority task. An important direction for further research is the processing and interpretation of parameters obtained during intelligent measurement for the purpose of adaptive control of measurement processes.