Prediction model of post-harvest peach pre-cooling effectiveness based on GA-BPNN

This study aims to rapidly and accurately monitor the pre-cooling performance of post-harvest peach in Forced-Air Cooling (FAC) for a better commercial and economic potential of fruits. Taking the Okubo peach as a research object, a single-box multi-layer Computational Fluid Dynamics-With Internal Heat Source (CFD-WIHS) heat and mass transfer numerical model was established to consider the internal heat source term from the respiration and evaporative latent heat. The transient mathematical model consisted of an unsteady Shear Stress Transport (SST) k-w turbulence model, the pressure-velocity coupling, and the semi-implicit for pressure-linked equations (SIMPLE). A pressure-based split solver was also utilized to select the second-order upwind scheme for the discrete format of momentum, energy, turbulent kinetic energy, and diffusivity. The outlet of the computational domain was set as a pressure-outflow boundary condition, whereas, the inlet was set as a velocity-inlet boundary condition before simulating. The result showed that the maximum Root Mean Square Error (RMSE) of the fruit temperature was 1.668 °C, the Mean Absolute Percentage Error (MAPE) was 13.65%, and the maximum relative error of predicted Seven-Eight Cooling Time (SECT) was 19.23 %, indicating the excellent feasibility and versatility of the model. Correspondingly, a Genetic Algorithm-Back Propagation Neural Network (GA-BPNN) prediction model was constructed to adopt a large amount of high-precision three-dimensional space-time temperature and flow field distribution data. Specifically, the velocity of air inflow, the temperature of pre-cooling air, the initial temperature of peach, the vent number, and opening diameter were taken as the input variables, while the peach Seven-Eight Cooling Time (SECT) and the Overall Heterogeneity Index (OHI) were used as the output variables. After that, 100 simulated datasets were divided into the BP neural network training set and test set at a ratio of 4:1, in order to determine the optimal number of hidden nodes and GA setting parameters. It was found that the correlation coefficients between the predicted and test value of the GA-BPNN model were 0.962 5 and 0.863 8, respectively, for the Seven-Eight Cooling Time (SECT) and the Overall Heterogeneity Index (OHI). Compared with the BP, the RMSE and MAPE of GA-BPNN model decreased by 30.22 % and 32.84 %, respectively, for the Seven-Eight Cooling Time. In pre-cooling uniformity, the RMSE and MAPE of GA-BPNN model decreased by 21.91 % and 39.45 %, respectively. Consequently, the GA-BPNN model demonstrated a better non-linear relationship between the pre-cooling performance of post-harvest peach and the main control factors, compared with the BP neural network model without optimization. The findings can also provide practical support to the real-time prediction on the pre-cooling performance of post-harvest peach for better pre-cooling quality in small- and medium-sized orchards.