基于深度回归网络的航煤水分离指数分析方法研究

Analysis of jet fuel water separation index by using deep regression network

水分离指数是航空煤油的重要评价指标,然而窄带吸收光谱法进行航煤水分离指数测量时存在精度低、稳定性差等问题。针对此问题,搭建一个航煤水分离指数分析系统平台,并提出一种基于深度回归网络的航煤水分离指数的分析方法。该方法首先对采集到的紫外吸收光谱进行预处理,然后通过深度信念网络来对光谱数据进行深度特征提取,最后使用极限学习机来对所提取的特征进行训练,构建一个可实现高精度、稳定测量的航煤水分离指数分析模型。结果表明:水分离指数标准值为65、70、89时,模型预测平均值为64.99、69.59、88.96,且最大相对误差控制在2%以内。该算法在精度和稳定性上具有明显优势,适合航空煤油水分离指数测量的实际需求。

The water separation index is an important evaluation index of jet fuel,the absorption spectrometry by narrowband light is used for the measurement of water separation index,but it is easy lead to low accuracy and poor stability due to insufficient absorption.To solve these problems,an analysis method of jet fuel water separation index was proposed based on deep regression network.The method first preprocesses the collected ultraviolet absorption spectrum,and then it uses the deep belief network to perform deep feature extraction on the spectral data.Finally,the extreme learning machine is trained by the extracted deep feature to realize the measurement of jet fuel water separation index.Experimental results show that the average prediction values of model are 64.99,69.59 and 88.96 when the standard values of water separation index are 65,70 and 89,and the maximum relative error is controlled within 2%. The proposed deep regression network algorithm is superior toconventional algorithms in terms of accuracy and stability, it is suitable for the actual demand of waterseparation index measurement of jet fuel.