Study on wine quality evaluation based on extreme learning machine improved by whale optimization algorithm

Abstract

[Objective] In order to solve the issue of excessive redundant information in near-infrared spectroscopy, enhance the accuracy of wine quality evaluation models, a rapid and non-destructive method was established for wine quality evaluation. [Methods] A wine quality evaluation model was proposed based on competitive adaptive reweighting sampling method for feature wavelength screening and extreme learning machine improved by whale optimization algorithm. Various feature wavelength screening methods such as competitive adaptive reweighting sampling was used, and the most suitable method for wine spectral feature wavelength screening was determined. In response to the problem of initial value and hidden layer bias in ELM, the whale optimization method was used to optimize the initial value and hidden layer bias of ELM, and an wine quality evaluation model based on extreme learning machine improved by whale optimization algorithm was constructed. [Results] Compared with GA-ELM, PSO-ELM, and the traditional ELM model, the accuracy of WOA-ELM was the highest, reaching 0.944 5, which was better than GA-ELM (0.929 0), PSO-ELM (0.906 1) and traditional ELM (0.817 7). [Conclusion] The parameters of the ELM model optimized by intelligent algorithms can effectively improve the accuracy of wine quality evaluation.

Keywords

• near infrared spectroscopy
• extreme learning machine
• whale optimization algorithm
• characteristic wavelength
• competitive adaptive reweighted sampling