基于改进MSVR的鲜食葡萄运输过程中环境因子与感官品质建模

挖掘运输过程中环境因子和生鲜果蔬感官品质之间的关系，实现基于环境因子的感官品质评估和预测，对于保持生鲜果蔬物流品质具有重要意义。该研究以鲜食葡萄为研究对象，通过对实际运输过程的跟踪监测，在实验室开展了鲜食葡萄运输模拟试验和感官试验，构建了鲜食葡萄运输感官品质数据集。在建模方法层面，该研究提出了一种基于多输出支持向量回归（Multiple Output Support Vector Regression，MSVR）模型的运输环境因子（温度、相对湿度）与感官品质（外观、香气、质地和风味品质）的预测模型，并利用粒子群（Particle Swarm Optimization，PSO）算法和遗传算法（Genetic Algorithm，GA）对模型进行优化（PSOGA-MSVR）。结果表明，PSOGA联合优化算法有效提高了MSVR模型的调参效率，且在常温运输、保冷运输和冷链运输3种不同的运输模式下，PSOGA-MSVR模型的预测效果均更优，决定系数R²高于0.985且各项误差指标低于其他模型；研究结果可为运输过程中合理调控环境因子，减缓生鲜水果感官品质的下降提供参考。

Modeling of environmental factors and sensory quality during transportation of table grapes using improved MSVR

Fresh fruits and vegetables need transportation and storage after harvesting, due to the regional and seasonal production in recent years. It is a high demand to clarify the relationship between environmental factors and the sensory quality of fresh fruits and vegetables during transportation. Accurate and rapid evaluation and prediction can be of great significance to maintain the logistics and sensory quality of fresh fruits and vegetables using environmental factors. Taking the table grape as the research object, the transportation simulation and sensory experiments were carried out in the laboratory, according to the tracking and monitoring actual transportation process. A data set of sensory quality was then constructed for the table grape during transportation. A prediction model was established for the environmental factors (temperature, relative humidity) during transportation and sensory quality (appearance, aroma, texture and flavor) using the Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) with the Multiple Output Support Vector Regression (PSOGA-MSVR) model. Specifically, the SVR model was widely used to identify the small sample, nonlinear, and high-dimensional pattern in the field of food quality prediction. The accuracy and generalization of SVR model depended largely on the kernel function and hyperparameters. The Radial Basis Function (RBF) was selected as the kernel function in this case. The results show that the PSOGA joint optimization effectively improved the parameter adjustment efficiency of the MSVR model. The improved PSOGA-MSVR model performed better prediction under three transportation modes, including normal, cold temperature, and cold chain transportation. Among them, the Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and coefficient of determination (R2) of the model were 0.083, 0.029, 0.172, and 0.982, respectively, under the normal temperature transportation. The MAE, MSE, RMSE, and R2 were 0.077, 0.022, 0.148, and 0.981, respectively, under cold storage and transportation. The MAE, MSE, RMSE, and R2 of cold chain transportation were 0.063, 0.018, 0.138, and 0.985, respectively. Therefore, cold chain transportation achieved the highest prediction accuracy for the sensory quality of table grapes. The better-fitting model was attributed to the longest time series and the most abundant data in the cold chain data set. There was a significant nonlinear relationship between transportation environmental factors and grape sensory quality (appearance, aroma, texture and flavor). Consequently, reasonable regulation of transportation environmental factors can be expected to effectively maintain the sensory quality of fresh fruits. The SVR model can also be used to simulate the logistics environmental factors and grape sensory quality of fresh table grapes. The quality evaluation and prediction can provide a theoretical, technical, and practical solution to improve the logistics process and quality of fresh table grapes.