Comparison of three artificial neural networks for predict the electrodeposition of nano-silver film

Abstract

The back-propagation neural network (BP), genetic algorithm based on BP (GA-BP), and extreme learning machine (ELM) were established to predict the glossiness of silver film and current efficiency for the electrodeposition of nano-silver film in this paper. The artificial neural networks (ANNs) were built to predict the results and contrast with experimental results. The prediction results demonstrated that GA-BP exhibited a relatively higher accuracy than simple BP. ELM demonstrated an efficient rate and the time required was far less than BP and GA-BP, which also had a high accuracy if the sample data were sufficient. GA-BP only needs a few sample data as the basis to obtain high accuracy, which has a low demand for sample data. In conclusion, for the electrodeposition of nano-silver film, the order of the accuracy of learning process is ELM > GA-BP > BP, the order of analogical ability is GA-BP ≈ BP > ELM and the order of time required is GA-BP > BP > ELM. Mean impact value (MIV) analysis was utilized to explore the impact of each parameter for three kinds of ANNs. The results display that glossiness, and current efficiency are mainly controlled by current density and substrate glossiness, but which parameter plays the crucial role is not manifest.