The new method of historical sensor data integration using neural networks

Abstract

The main feature of a neural network used for accuracy improvement of physical quantities (for example, temperature, humidity, pressure etc.) measurement by data acquisition systems is the insufficient volume of input data for predicting neural network training at an initial exploitation period of sensors. The authors propose the technique of data volume increasing for predicting neural network training using: (i) an additional approximating neural network; (ii) method of "historical" data integration (fusion). The authors propose the advanced method of "historical" data integration and present simulation results on mathematical models of sensor drift using a single-layer perceptron.