基于回声状态网络的柴油性质近红外光谱测量模型

Echo-state Network- based NIRS Model for Diesel Oil Property Measurement

针对柴油性质的近红外光谱测量问题,提出了一种基于回声状态网络(Echo State Network,ESN)的近红外光谱测量模型。该模型采用化学计量学方法,通过ESN对柴油性质的近红外光谱数据进行学习以建立光谱信息与柴油性质间的输入输出关系模型。为克服ESN生成过程产生的随机性对测量精度的干扰,采用集成学习方法对模型进行校正。对柴油的十六烷值和密度的近红外光谱测量表明,该模型在测量精度上优于偏最小二乘、主成分回归及多元线性回归等常规化学计量学方法建立的模型。

Considering the measurement of diesel oil properties with near-infrared spectroscopy （NIRS）, an echo-state network （ESN）-based NIRS measurement model was proposed, which adopts chemometrics method and learns NIRS data about diesel oil properties to build an input-output relationship between the NIRS data and the diesel oil properties. In order to overcome the measurement disturbance incurred by randomness in ESN generation process, an integration learning method was applied to calibrate the model. Measuring diesel oil＇s cetane number and density indicates that the ESN model outperforms other models built by conventional methods such as multiple linear regression （MLR）, principal components regression （PCR） and partial least squares （PLS） in measurement accuracy.