论文部分内容阅读
分析化学中传统的多元校正通常采用线性回归或人工神经网络算法。但线性回归不能适应实测数据或多或少的非线性,而人工神经网络又有过拟合弊病造成误差。为此我们提出用新发展的既能处理非线性数据,又能限制过拟合的支持向量机算法。本文首次提出导数光谱-支持向量回归法。该法用于NO_3~--NO_2~-体系的同时测定解得的浓度平均相对误差在±82％,明显好于ANN法(±9.15％)和线性回归法(±11.5％)。这表明支持向量机算法在分析化学的校正技术中是有用的。
Traditional multivariate calibration in analytical chemistry usually uses linear regression or artificial neural network algorithms. However, linear regression can not adapt to the measured data more or less nonlinear, and artificial neural network has over-fitting errors caused by error. To this end, we propose a newly developed algorithm that can handle both nonlinear data and over-fit support vector machine. This paper presents the first derivative spectroscopy - support vector regression. The average relative error of the method for the simultaneous determination of NO_3 ~ --NO_2 ~ - system was ± 82%, which was significantly better than ANN method (± 9.15%) and linear regression method (± 11.5%). This shows that support vector machine algorithm is useful in analytical chemistry calibration techniques.