Modeling Of Blast Furnace Temperature Based On Improved Particle Swarm Optimizer And Support Vector Machine

The modeling and optimization problems of complex industrial system always cannot be described by traditional mathematical models. Artificial intelligence is an effective way to solve these problems. For large blast furnace, temperature is an important performance indicator to measure the condition of blast furnace and guide the experienced foremen to determine control strategy. This paper aims at predicting the temperature of blast furnace through translating the temperature into the silicon content in hot metal. An improved particle swarm optimizer(PSO) is proposed, and experimental results are shown to test its performance. The improved algorithm is used to optimize the support vector regression(SVR) model of hot metal silicon content. Finally, the single-model and multi-model are built. This paper has theoretic significances and application values, the main research results of this paper can be summarized as follows:Firstly, based on the deficiency of particle swarm optimizer, DMS-PSO-CLS is proposed. DMS-PSO-CLS is the improvement of DMS-PSO. A new cooperative learning strategy is added to DMS-PSO so that more information can be exchanged among sub-swarms. Finally, the balance between the global exploration and the local exploitation can be achieved. Simulation experiments are conducted on 18 benchmark functions. The experimental results of DMS-PSO-CLS are compared with six other improved PSOs. Experiments show that DMS-PSO-CLS can achieve a significant performance.Secondly, DMS-PSO-CLS is used to optimize the parameters in SVR and the optimized SVR is used to build the sing-model of hot metal silicon content. The penalty factor and Kernel function parameters are optimized by DMS-PSO-CLS. Then the optimized single SVR model is used to predict the silicon content. It can be observed from the experiments that the optimized single SVR model has a better performance.Finally, the optimized SVR is used to build the multi-model of hot metal silicon content. Considering blast furnace has some different working conditions, clustering algorithm is used to classify the dataset of blast furnace. Each category is approximated to one working condition. For each category, a SVR prediction model is built and finally the multi-model of silicon content is built. Through the comparison of experiments, it can be observed that the DMS-PSO-CLS based multi-model perform better. At the same time, we can observe that the multi-model of blast furnace temperature, which combines the mechanism and data, has a better performance.