DEPENDENCE OF TEMPERATURE VALUE PREDICTION ERROR BY NEURAL NETWORKS ON ADC RESOLUTION

2018;
: pp. 12-15
1
Lviv Polytechnic National University, Ukraine
2
Lviv Polytechnic National University

Current article describes the results of the study of the error of temperature values prediction using neural networks. In the introduction, the authors consider previous research pointing out problems that arise during measuring the high temperatures. To solve these problems the neural networks applies. The formula for temperature transition process is derived.

Dependence of the temperature value prediction error by neural networks on the ADC resolution is considered. Algorithm for creating and teaching neural networks is studied. Sequences modeling for the neural network training and the equation for calculating the absolute error of temperature prediction are given. Data used by the neural network are quantized by the level. The number of quantizing levels depends on the ADC resolution. Thus, while processing the results of measurements by the neural network, additional error rises caused by ADC resolution.

Results of the study of dependence of the temperature value prediction error on the number of network inputs and on the ADC resolution are presented. They envisage that the prediction error decreases with ADC resolution growth and the inputs amount in the neural network reduction. Also, lower predicting temperature values errors are located in the middle of the temperature range of the object of measurement.

Also the dependence of average error and absolute uncertainty of the temperature value prediction on the ADC resolution are studied. Equations for computing the mean temperature error, standard deviation and uncertainty are deduced. Table with results of the study of temperature value prediction error for quantized data and for double type of data is given. In result, we have defined the temperature prediction error dependence on the ADC resolution.

[1] O. Lopatko, I. Mykytyn, “Temperature value prediction errors using neural networks and ideal transition process”, Measuring equipment and metrology, Lviv, Ukraine: no.78, p.20–24, 2017.

[2] N. Yaryshev, Theoretical basis for measuring non-stationary temperatures, Leningrad, USSR: Energoatomizdat, 1990.

[3] N. Kovalchuk,E. Polischuk, I. Pytel, K. Semenistyi. Modern methods and means for determining the dynamic characteristics of converters, Overview information TS-6, CNIITEI Instrumentation, is.1, 1983.

[4] D. Kriesel. A Brief Introduction to Neural Networks, 2007, [On-line]. Available:  http://www.dkriesel. com/en/science/neural_networks.

[5] R. Bordawekar, B. Blainey, R. Puri, Analyzing Analytics, Morgan & Claypool Publishers, 2015.

[6] M. Dorozhovets, V. Motalo, B. Stadnyk. Fundamentals of metrology and measuring technique, vol.1, Lviv, Ukraine: Publ. house Lviv Nat. Polytechn. Univ., 2005.