Nuclear Instruments and Methods in Physics Research A 502 (2003) 523–525 Use of neural network based auto-associative memory as a data compressor for pre-processing optical emission spectra in gas thermometry with the help of neural network S.A. Dolenko a, *, A.V. Filippov b , A.F. Pal a , I.G. Persiantsev a , A.O. Serov a a Skobeltsyn Institute of Nuclear Physics, Moscow State University, Moscow 119992, Russia b Troitsk Institute of Innovation and Fusion Research, Troitsk, Moscow Reg., 142092, Russia Abstract Determination of temperature from optical emission spectra is an inverse problem that is often very difficult to solve, especially when substantial noise is present. One of the means that can be used to solve such a problem is a neural network trained on the results of modeling of spectra at different temperatures (Dolenko, et al., in: I.C. Parmee (Ed.), Adaptive Computing in Design and Manufacture, Springer, London, 1998, p. 345). Reducing the dimensionality of the input data prior to application of neural network can increase the accuracy and stability of temperature determination. In this study, such pre-processing is performed with another neural network working as an auto-associative memory with a narrow bottleneck in the hidden layer. The improvement in the accuracy and stability of temperature determination in presence of noise is demonstrated on model spectra similar to those recorded in a DC-discharge CVD reactor. r 2003 Elsevier Science B.V. All rights reserved. PACS: 84.35; 07.20.D; 33.20; 33.50; 07.05.K Keywords: Neural networks; Gas thermometry; Emission spectroscopy; Data compression 1. Previous study The problem of temperature determination from optical emission spectra is usually solved by fitting theoretically calculated spectra to measured spec- tra with a least-squares routine. However, this approach encounters some difficulties. Presence of a large number of local minima hampers the procedure of minimization of error. Under condi- tions of high temperature, when thermal equili- brium emission is considerable and when emission intensities of molecular band systems are affected by fluctuations, this method turns out to be inapplicable for determination of gas temperature in real time. In our previous studies [1], we have used multi- layer perceptron (MLP) [2] and Group Method of Data Handling (GMDH) [3] for temperature determination from model CO spectra similar to those obtained in a CVD reactor used for diamond film deposition, with different level of noise. The following statistics were used to estimate quality of the obtained nets: SD—standard *Corresponding author. E-mail address: dolenko@srd.sinp.msu.ru (S.A. Dolenko). 0168-9002/03/$-see front matter r 2003 Elsevier Science B.V. All rights reserved. doi:10.1016/S0168-9002(03)00489-3