Virtual Ion Selective Electrode for Online Measurement of Nutrient Solution Components
Virtual Ion Selective Electrode for Online Measurement of Nutrient Solution Components
复制标题
用于在线测量营养液成分的虚拟离子选择电极
DOI：
10.1109/jsen.2010.2060479
复制
发表时间：
2011-02-01
影响因子：
4.3
通讯作者：
Tang, Yongning
中科院分区：
文献类型：
--
作者：
Chen, Feng;Wei, Dali;Tang, Yongning
营养液中多种成分的测量是营养液优化控制的前提。目前营养液的测量方法是根据pH值和电导率（EC）值来估算营养液中各成分的浓度，误差较大。本文提出了一种虚拟离子选择电极（VISE）方法来在线测量营养液中难以测量的成分，并具有显着提高的性能（例如精度和速度）。为了有效地建模 VISE，必须分析营养液成分之间的相关性，由于其不确定性和复杂性，这具有很大的挑战性。本研究通过实验研究了营养液成分的变化规律。根据蔬菜生长的内在分析，找到营养液成分之间的相关性。在我们的方法中，采用最小二乘支持向量机（LS-SVM）来融合传感器数据，以实现快速计算和全局最优。此外，为了提高LS-SVM的估计精度并降低计算复杂度，根据离子选择电极（ISE）的特性引入了一个公式来表示正则化参数，这对于确定模型复杂度和拟合误差之间的权衡至关重要。实验结果表明，所提出的VISE模型是有效的，为多组分测量提供了有益的参考。
The measurement of multiple components in nutrient solution is prerequisite for optimal control of nutrient solution. The current measurement methods of nutrient solution estimate the concentrations of components in nutrient solution based on pH and electronic conductivity (EC) values, which lead to large errors. In this paper, a virtual ion selective electrode (VISE) approach is proposed to online measure the hard-to-measure components in nutrient solution with highly improved performance (e.g., accuracy and speed). In order to effectively model VISE, the correlation among nutrient solution components has to be analyzed, which is significantly challenging due to its uncertainty and complexity. In this study, the variation regularities of the nutrient solution components are experimentally investigated. The correlation among the nutrient solution components is found according to the intrinsic analysis of vegetable growth. In our approach, least squares support vector machine (LS-SVM) is adopted to fuse the sensor data to achieve fast computing and global optimum. In addition, to improve the estimation accuracy and reduce the computational complexity of LS-SVM, a formula is introduced based on the characteristics of ion selective electrode (ISE) to represent the regularization parameter, which is critical in determining the tradeoff between the model complexity and fitting errors. The experimental results show that the proposed VISE model is effective and offers a beneficial reference for multiple component measurement.