A Study of Artificial Neural Networks for Electrochemical Data Analysis

Abstract

Artificial neural network has been demonstrated the capability of identifying the chemical composition from various analytical methods. This study tested the feasibility of using a single network to calculate data obtained from polarography, linear scanning voltammetry (LSV), and electrochemical impedance spectroscopy (EIS). Computer generated data were used for network learning and testing. The network property of a single layer network was calculated at various learning cycles, learning rates, and β values of the neuron activation function. This study also evaluated two-layer, three-layer, and four-layer networks. Test results conclude that a single artificial neural network could analyze various analytical data by using corresponding weight matrices. To produce accurate results, the network requires high training cycle, small β value of sigmoid function, and small learning rate. Network accuracy increase as the number of neuron increased to certain extents. Network structure plays minimum effects on the accuracy of network for a given number of neurons.