论文部分内容阅读
针对高炉炼铁过程的关键工艺指标——铁水硅含量[Si]难以直接在线检测且化验过程滞后的问题,提出一种基于稀疏化鲁棒最小二乘支持向量机(R-S-LS-SVR)与多目标遗传参数优化的铁水[Si]动态软测量建模方法.首先,针对标准最小二乘支持向量机(LS-SVR)的拉格朗日乘子与误差项成正比导致最终解缺少稀疏性的问题,提取样本数据在特征空间映射集的极大无关组来实现训练样本集的稀疏化,降低建模的计算复杂度;其次,标准最小二乘支持向量机的目标函数鲁棒性不足的问题将IGGIII加权函数引入稀疏化后的最小二乘支持向量机模型进行鲁棒性改进,得到鲁棒性较强的稀疏化鲁棒最小二乘支持向量机模型;最后,针对常规均方根误差评价模型性能的不足,提出从建模误差与估计趋势评价建模性能的多目标评价指标.在此基础上,利用非支配排序的带有精英策略的多目标遗传算法优化模型参数,从而获得具有最优参数的铁水[Si]在线软测量模型.工业实验及比较分析验证了所提方法的有效性和先进性.
Aiming at the problem that it is difficult to directly detect the silicon content of molten iron [Si] and the lag of the testing process, a new robust least square support vector machine (RS-LS-SVR) Multi-objective genetic parameter optimization for hot metal [Si] dynamic soft-sensing modeling method.Firstly, the Lagrange multiplier for the standard least squares support vector machine (LS-SVR) is proportional to the error term, resulting in the lack of sparsity Secondly, the objective function of the standard least square support vector machine is not robust enough. In the second part, the objective function of the standard least square support vector machine The problem is to introduce the IGGIII weighting function into the sparsified LS-SVM model to improve the robustness and obtain a robust robust sparsely-constrained LS-SVM model. Finally, according to the conventional root mean square error To evaluate the performance of the model, a multi-objective evaluation index is proposed to assess the modeling performance from the modeling error and the estimated trend. Based on this, a multi-objective genetic algorithm with non-dominated sorting Optimization model parameters to obtain molten iron [Si] with optimal parameters online soft measurement model of industrial experiments and comparative analysis of the validity and advantage of the proposed method.