In this paper, a measurement system using absorption spectroscopy was designed and realized. Using alternativemeasurement method, the system could perform with minimum calibration requirement. Features of sucrosesolutions with Brix percentage from 12 oBrix to 16 oBrix were measured and extracted. Moreover, three artifi-cial intelligent methods, namely artificial neural networks, artificial neural networks with principal componentanalysis, and genetic algorithm, were applied to predict unknown concentrations. The key difference betweenmethods was the assumption on residual error's statistical characteristics. Experimental results showed thatthe correlation and collinearity in structure of measured data were essential for higher prediction accuracy inartificial intelligent methods. In our experiments, genetic algorithm provided the most accurate predictions.
展开▼
