基于GA优化LSSVM高炉铁水含硅量预报

Prediction of blast furnace hot metal silicon content based on LSSVM optimized by GA

摘要:

铁水含硅量是表征高炉生铁质量的重要指标，也是反映高炉内部热状态的重要参数。为了提高铁水含硅量测量精确度，保证高炉顺行，提出一种基于最小二乘支持向量机(LSSVM)的铁水含硅量预报模型，采用遗传算法(GA)确定模型参数的优化组合，以改善模型性能。将某钢管厂高炉的实际运行数据经过预处理后作为模型的训练和测试样本，进行模型预报实验，并与神经网络模型和时间序列分析模型的预报性能进行了比较。基于GA优化参数的LSSVM模型对铁水含硅量预报的最大相对误差为5.8%，相关系数为0.926 375，预报精确度比直接LSSVM模型提高了2.1%，比前向神经网络模型提高了4.3%。

Abstract:

Hot metal silicon content has long been used as one of the most important indices to represent the hot metal quality and the thermal state of a blast furnace. In order to improve the measurement precision and attain stable operation of the blast furnace, a novel model for predicting silicon content by using Least Square Support Vector Machine(LSSVM) is presented. It adopts Genetic Algorithm(GA) to determine the optimum parameter set and therefore improves the model performance. By training and testing the operational data from blast furnace at a steel tube plant, the experimental results indicate that the proposed model can predict silicon content in hot metal with a maximum relative error of 5.8 % and correlation coefficient of 0.926 375, whose accuracy can be improved by 2.1% and 4.3% than that of the direct LSSVM and the feed-forward network with the same data set, respectively.