论文部分内容阅读
采用主成分－人工神经网络对不同馏程柴油的近红外光谱进行校正，预测其闪点。采用监控集控制网络训练过程，以避免过训练。探讨了人工神经网络（ＡＮＮ）、直接线性连接人工神经网络（ＬＡＮＮ）的校正效果，并与局部权重回归（ＬＷＲ）、主成分回归（ＰＣＲ）及偏最小二乘（ＰＬＳ）等校正方法进行了比较，认为人工神经网络及直接线性连接人工神经网络具有较好的准确性及抗干扰性，可以用于较宽的样品范围，是解决非线性关联的较好手段。
The main component - artificial neural network was used to correct the NIR spectra of different distillate diesels and the flash point was predicted. Monitoring set control network training process to avoid over training. The correction effects of artificial neural network (ANN) and direct linear connected artificial neural network (LANN) were discussed. Corrections were also carried out with local weight regression (LWR), principal component analysis (PCR) and partial least squares (PLS) It is considered that artificial neural network and direct linear connection artificial neural network have good accuracy and anti-interference ability, which can be used in a wide range of samples. It is a better way to solve nonlinear correlation.