Application of artificial neural networks for predicting parameters of commercial vacuum cooling process of baby cos lettuce

Abstract

Artificial neural networks (ANNs) demonstrated sensitive results in predicting final temperature and weight loss percentage of commercial vacuum cooling process. According to the results for final temperature, ANNs showed better prediction performance than multiple linear regression in all criteria, including an adjusted R‐squared (R2adj) of .932 and root mean square error (RMSE) of 0.579. In addition, the predicted values of weight loss percentage from ANN models were in good agreement with all experimental data (Radj2 = .82 and RMSE = 0.286). The process parameters from proper ANN model was subsequently used to investigate the effect of vacuum cooling on the qualities of baby cos lettuce during storage compared with the non‐precooled samples. The results suggested that vacuum cooling was an effective method for extending shelf life of baby cos lettuce from 9 to 16 days at 4°C. Qualities of fresh lettuce vacuum cooled using selected process parameters simulated from proper ANN model were significantly better than the non‐precooled sample during storage (p ≤ .05). Practical Applications Vacuum cooling widely considered the best precooling technique for horticultural produce. However, vacuum cooling also has some disadvantages including weight loss and freezing injury occurring during the vacuum cooling process, due to the setting of inapplicable parameters before the vacuum cooling process such as final pressure and holding time. Experimental study of the optimum vacuum cooling parameters for baby cos lettuce in each season (winter or summer) takes a long time and is expensive. Therefore, the present study emphasized the effects of vacuum cooling parameters setting on final temperature and weight loss percentage of baby cos lettuce. Prediction methods using artificial neural network (ANN) allow for vacuum cooling processes based on experimental research to be recommended as feasible at an industrial scale.

Current Trends in Intelligent Control Neural Networks for Thermal Processing (Foods): Systematic Literature Review

“…ANN models contain a superficial computational layer of nodes operating as nonlinear summing tools. These nodes are interconnected with line-weighed connections; weights are adjusted as data is presented to the network [63], [64], [65], [66]. ANN can produce artificial neurons that perform tasks such as predicting output values, classifying objects, approximating functions, recognizing patterns in multifactorial data, and solving known problems (Fig.…”