Converter-based Adaptive Neural Network Fuzzy Inference System, The End Of Forecast

A new method to predict end-point of basic oxygen furnace (BOF) is presented. This method is on the basis of the furnace flame spectrum integrated with adaptive neuro-network-based fuzzy inference system(ANFIS).The experimental system based fiber optic spectrometer and Laboratory Virtual Instrumentation Engineering Workbench (LabVIEW) is established, including spectrum data acquisition system and data analysis and processing platform.Through the program debugging and the experiment vertification of BOF,Data acquisition and analysis system designed above is suitable for use.After deliberating on the steel-making spectral data available,the experimental results show that corresponding light intensity for three characters wavelengths (589nm,765nm, 769nm)exist obvious evolution during converting process.Furthermore,characteristic spectrum data sets and progression values set according to steelmaker experience are served as training data, the grid partition initial FIS model,the subtractive clustering initial fis model and the fuzzy c-mean clustering initial FIS model are conducted with the help of fuzzy logic toolbox in MATLAB.After training model by the function named anfis, three different kinds of BOF end-point prediction models are put forward corresponding to these initial FIS models.In conclusion,21 series of BOF spectrum data are used to testify end-point prediction models available.The experiments and simulation results suggest that the forecast errors of the endpoint-prediction models all are less than 6 seconds,which can realise model required accuracy within expectation.