Soft Measurement Method Base On LS-SVM For Mixture Granularity Distribution

Sintering process is an important procedure of the procee before the iron and steel smelting iron, and sinter, the production of sintering, which has a direct impact on the output and quality of steel, is a vital raw material for blast furnace. Mixing granulation is a very important sub process of sintering to ensure the permeability of the sinter layer, achieve thick layer sintering and to improve the efficient use of energy. Due to the above situation, researching on the mixing granulation process is of great significance for steel enterprises to save energy and reduce emissions.The mixing granulation process is a very complex industial process with the features of non-linear, hysteresis, process parameters are difficult to measure. Base on the difficulties of granularity meansuring, combine the analysis of mechanism and grey relational method explores the factors influencing the process of mixing granulation, the model for soft meansure granularity distribution is proposed.By collecting production process data and preprocess the data, the model for indirect meansure granularity distribution is proposed base on least squares support vector machine(LS-SVM). The affection of model parameters(width factor σ and penalty factor c) on the granulity distribution soft measurement model is anlysised, and use intelligent optimization algorithms to optimize the model parameters.To deal with the problem that the model parameters selecting mostly on experience and trial and error method, quantum particle swarm optimization(QPSO) algorithm was presented to select the parameters of LS-SVM model. QPSO is the improvement base on standard particle swarm,make up the shortcomings that standard PSO easily fall into local optimal value. The simulation results shows that the method bosed on QPSO is fast and precise, A granulity distribution soft measurement model based on QPSO and LS-SVM was established. simulation results shows that the proposed model is more better than the model which based on traditional method about precise and optimization time.