基于偏最小二乘回归的焦炭热性质非线性预测模型

摘要

Based on the fact that the coal quality indicators may be multi-collinear, which will make the coke's thermal properties prediction difficult, an idea was put forward that the partial least-squares (PLS) regression technology was applied to find the coke's thermal properties prediction model.Taking into account the complex non-linear relationship between the coal quality indicators and the coke's thermal properties, a quasi-linear method was adopted in the modeling that the inputs of the model not only included the linear effect of coal quality indicators, but also the quadratic effects and interaction effects.The practicing example was predicted applying non-linear PLS regression model using quasi-linear method and linear PLS regression model respectively.The results show that it is efficient and available using the PLS regression to predict the coke's thermal properties.The non-linear PLS regression model is more accurate than the linear PLS regression model.%针对基于煤质指标预测焦炭热性质建模过程中易出现的多重共线性问题,提出应用偏最小二乘回归对焦炭热性质进行预测的建模思路.考虑到煤质指标与焦炭热性质之间复杂的非线性关系,采用拟线性化处理的方法,将煤质指标的一次效应、二次效应及交互效应作为模型输入,建立焦炭热性质预测的偏最小二乘回归模型;基于拟线性化处理的非线性偏最小二乘回归和线性偏最小二乘回归对焦炭热性质预测实例进行分析.研究结果表明:基于偏最小二乘回归方法建立的焦炭热性质预测模型是有效可行的:非线性偏最小二乘回归模型的预测精度明显比线性偏最小二乘回归模犁的预测精度高.