In order to develop a quick method for predicting fatty acid in rice storage, gas chromatography-ion mobility spectrometry (GC-IMS) was applied to detect and analyze volatile organic compounds (VOCs) at different rice storage stages, and partial least squares regression (PLSR) algorithm was used to establish a linear regression model between fatty acid values and characteristic VOCs. The results showed that rice fatty acid values increased gradually with extension of storage time. Odor components of rice mainly included alcohols and aldehydes. Except for 1-octene-3-alcohol, the content of other VOCs showed an overall downward trend during the storage period. After variable optimization using two different algorithms, the correlation coefficient of the PLSR cross validation model could reach 0.9544, and the corresponding root mean square error was 2.4093. In conclusion, fatty acid values of rice with different storage periods could be accurately predicted by using characteristic VOCs variables and chemometric tools, which would provide a rapid and nondestructive detection method for rice quality during storage based on odor information.

To study the feasibility of evaluating the quality characteristics of banana based on the browning area. The texture characteristics, total soluble solids (TSS), ascorbic acid, malondialdehyde (MDA) concentrations, relative conductivity, polyphenol oxidase, peroxidase, and phenylalanine ammonia-lyase (PAL) activities in banana peels were detected during storage. A linear model was made by principal component analysis and multiple linear regression between the banana browning area and characteristic indices. The results showed that the changes in the physiological characteristics of bananas were significantly different during different storage periods. The main factors that affected the banana browning area were relative conductivity, PAL, TSS, and MDA, indicating that lipid peroxidation, respiration, and metabolism of phenylpropanoids had significant influence on the banana browning area during storage. Thus, it is feasible to predict banana quality based on changes in browning area, which could be a rapid and non-destructive detection of banana quality during storage.

The estimation of construction costs for shield tunneling projects is typically based on a standard quota, which fails to consider the variation of geological parameters and often results in significant differences in unit cost. To address this issue, we propose a novel model based on a random forest machine learning procedure for analyzing the construction cost of shield tunnelling in complex geological conditions. We focus specifically on the unit consumption of grease, grouting, labor, water, and electricity. Using a dataset of geotechnical parameters and consumption quantities from a shield tunneling project, we employ KNN and correlation analysis to reduce the input dataset dimension from 17 to 6 for improved model accuracy and efficiency. Our proposed approach is applied to a shield tunneling project, with results showing that the compressive strength of geomaterial is the most influential parameter for grease, labor, water, and electricity, while it is the second most influential for grouting quantity. Based on these findings, we calculate the unit consumption and cost of the tunnelling project, which we classify into three geological categories: soil, soft rock, and hard rock. Comparing our results to the standard quota value, it is found that the unit cost of shield tunneling in soil is slightly lower (6%), while that in soft rock is very close to the standard value. However, the cost in the hard rock region is significantly greater (38%), which cannot be ignored in project budgeting. Ultimately, our results support the use of compressive strength as a classification index for shield tunneling in complex geological conditions, representing a valuable contribution to the field of tunneling cost prediction.