增强偏最小二乘回归算法在近红外光谱法啤酒度数软测量建模中的应用

Application of Boosting Partial Least Square Regression to Soft-sensor Modeling for NIRS Determination of Alcohol Degree of Beer

软测量技术广泛应用于工业过程,其核心是建立一个可靠的软测量模型。常规的软测量都是基于建立单个的数学模型,常难达到需要的精确和稳健性。基于机器学习的集成思想,给出了增强偏最小二乘回归(boosting-PLS)算法,并将其用于一个基于近红外光谱法啤酒度数软测量中,试验结果表明:应用boosting-PLS算法所建模型是一种精确、稳健、有应用潜力的软测量方法,特别适合于类似涉及高维光谱数据的软测量。

Soft sensors have been widely used in industrial process,the core of which is the establishment of a reliable soft-sensor model.Conventionally the application of soft-sensor is based on the establishment of a single mathematical model,which is often difficult to achieve at the required accuracy and robustness.Base on the idea of ensemble from machine learning,the algorithm of boosting ensemble partial least square regression（boosting-PLS） was proposed and applied to soft-sensor modelling,which was used in NIRS determination of alcohol degree of beer.As shown by experimental results,the soft-sensor model established on the base of the algorithm of boosting-PLS was proved to be accurate and robust and suitable especially for soft-sensor involving high-dimensional spectral data.