| With the continuous development of social economy and the expanding frequency of human production activities,oil spill accidents on the sea surface often occur.At present,the causes of sea oil spill include oil leakage caused by production and transportation failures such as oil drilling platforms and oil leakage caused by collisions of maritime transport vessels.Once oil spills from the sea,it will cause environmental pollution,waste of oil resources and damage to the Marine economy and industry.Laser induced fluorescence technology has the advantages of non-contact detection,and can be carried by aircraft or ships.It is an important detection method used in sea oil spill detection.Therefore,quantitative analysis of oil samples obtained based on laserinduced fluorescence technology can provide reference for rapid detection of oil spill thickness and oil content of oil-in-water emulsion.Based on common petroleum products,this paper constructed a quantitative analysis model based on laser induced fluorescence technology and combined characteristic wavelength selection algorithm with machine learning algorithm to realize the evaluation of oil film thickness and oil content of oil-inwater emulsion.Specific research contents are divided into the following aspects:(1)As there are many variables without information in the fluorescence spectrum data obtained by laser induced fluorescence technology,which affects the accuracy of quantitative analysis model,it is proposed to use characteristic wavelength selection method and linear regression model to achieve the inversion prediction of relatively thick oil film thickness.Taking diesel oil film and white oil film as the research object,the linear quantitative analysis model of oil film thickness was established by using characteristic wavelength selection and partial least square method.The characteristic wavelength selection method can reduce the complexity of the quantitative model algorithm,and the combination of partial least square modeling can effectively improve the prediction accuracy of oil film thickness.(2)In view of the different quantitative analysis effects of different regression models,it is proposed to model and predict the thick oil film thickness based on the characteristic wavelength optimization combined with the gray Wolf algorithm to optimize the support vector regression algorithm.Taking diesel oil film and white oil film as the research object,grey Wolf algorithm was used to optimize the support vector regression algorithm to establish a nonlinear regression model to achieve effective inversion of oil film thickness,which has better prediction ability and stability than partial least squares linear regression algorithm.The model based on characteristic wavelength optimization combined with gray Wolf algorithm optimized support vector regression algorithm has better prediction effect than the model based on full spectrum.(3)In view of the complex fluorescence spectrum data of oil-in-water emulsion obtained by laser induced fluorescence technology,particle swarm optimization algorithm was proposed to optimize the extreme learning machine network for modeling and prediction.Taking the oil-in-water emulsion of diesel oil,white oil,ordinary kerosene and aviation kerosene as the research object,using particle swarm optimization algorithm to optimize the extreme learning machine network,the quantitative analysis model of oil-inwater emulsion of oil was established and the oil content was predicted.Compared with the original Extreme Learning machine network,particle swarm optimization support vector machine algorithm and Grey Wolf algorithm optimized support vector machine regression algorithm,the prediction accuracy and stability of the model are improved,and the quantitative prediction of oil content in oil-in-water emulsion can be achieved well. |
