The Design Of Liquid Identifier Based On RBF Neutron Network

Liquid analyzer is an equipment to classify and recognize different kinds of liquid by fiber capacity liquid analysis.The main principle is to convert the physical change when liquid forming droplets into fiber and capacity voltage signals, protract droplet fingerprint with these two signals, extract features from the fingerprint, and then design and train the classifier to classify and recognize liquid.It can be known from the whole procedure that the design of liquid identifier is the core of liquid analyzer, as its performance can decide the whole performance of the equipment. Now kinds of algorithm are being used to design classifier, such as BP neutron network and fuzzy pattern recognition. BP network is widely used and has high recognition rates, but its convergence is quite slow; fuzzy pattern recognition can classify samples with obscurity boundary, but it is quite new. So this article lead RBF neutron network into the design of liquid identifier, which can reduce network training time by a large margin without bringing down the recognition rate.This article introduces the basis of RBF neutron network and the process to design, train and test RBF network with MATLAB neural network toolbox.This article takes Gaussian kernel function as the radbasis function of the network, and at the same time, use k-means clustering algorithm to calculate the core of the function to improve the recognition rate.This article also puts forward the combination of genetic algorithm and k means to select and calculate the core of Gaussian kernel function, then improve the recognition rate and at last get the best RBF network identifier. It comes out that RBF network can be used to design liquid identifier and the output can meet the demands. At last, this article puts forward the design of RBF neutron network liquid identifier experiment platform with MATLAB. This platform can be used to load samples from text, design and train network with these samples.