神经网络与激光诱导击穿光谱技术结合的烧结矿中硅元素定量分析方法探究

烧结矿中二氧化硅的含量对高炉炉渣产量以及冶炼能耗有重要影响,因此探索一种能够快速、准确地分析烧结矿中硅元素含量的方法具有重要的研究意义.拟采用激光诱导击穿光谱技术(LIBS)对30个烧结矿实际样品进行快速分析,收集其190～300 nm范围的光谱信号,先建立特征线(Si 288.16 nm)的标准曲线,分析特征线信号强...

Study on quantitative analysis method of silicon in sinter based on neural network and laser-induced breakdown spectroscopy

The content of silica in the sinter had an important influence on the output of slag and energy consumption in smelting.Therefore,it was of great research significance to explore a quick and accurate analysis method of the content of silicon in the sinter.30 actual samples of sinter ore was analyzed quickly by laser-induced breakdown spectroscopy (LIBS).After the spectral signals in the range of 190-300 nm was collected,a standard curve of the characteristic line (Si 288.16 nm) was established to analyze the relationship between the intensity and the element concentration.Then two neural network prediction models were constructed to analyze their prediction performance with different numbers of features as input.The experimental results showed that the standard curve method had poor prediction performance and the correlation coefficient was 0.230 9.The neural network prediction model built with 55 features had overfitting phenomenon and could not meet the detection needs.The neural network prediction model built with 5 features was optimum among the three models.For the test set,the correlation coefficient was 0.886 3,the root mean square error was 0.209 0,and the relative error was 1.42%.In addition,as the number of features decreased,the average training time of the model decreased from 11.9 s to 0.3 s.LIBS technology combined with neural network method could effectively analyze the content of silicon in sintered ore,and provide data support for real-time adjustment of ingredients in the metallurgical process.