针对硅锰合金埋弧冶炼过程中合金成分检测困难、离线化验滞后大、难以实时控制的问题,提出一种改进在线最小二乘支持向量机(IOLSSVM)的合金成分预测模型。该模型对每一个新增样本采用增量式学习,根据样本对模型贡献的不同删除样本集中对模型贡献最小的样本数据,利用递推计算增强模型的在线学习能力。将此模型应用于30MVA硅锰合金埋弧炉冶炼过程合金成分预测,实际生产运行数据表明了此方法的有效性。
To overcome the difficulty that the silicon-manganese alloy composition can not be effectively controlled in silico- manganese smelting process due to the lack of real-time on-line instrumentation,a prediction model based on Improved Online Least Squares Support Vector Machine(IOLSSVM) is proposed.According to the contribution of each data point to the prediction model,the least important data point is removed from the data set when adding new ones.The online learning ability is also improved by using the recursive algorithm.This method is applied to predict the silicon-manganese composition in a 30MVA submerged arc furnace smelting process and the results show its effectiveness.