Moving toward environmental sustainability in the fruit production systems needs accurate monitoring of the systems in terms of energy flow. In this regard, using regression models (i.e., Cobb–Douglas), Multiple Linear Regression (MLR)), and Artificial Neural Network (ANN), this paper examined the energy flow in the pomegranate orchards in Iran. Unlike other similar studies, in the current research, the cross-validation technique was employed to evaluate how choosing datasets would modify the behavior of the models. The results demonstrated that the total energy consumption and energy efficiency of the pomegranate production were 13,634.13 MJ ha⁻¹ and 1.01, respectively. The results also indicated that among investigated MLR models, the pure-quadratic model had the best performance. Besides, among 13 evaluated ANN-training algorithms, the Bayesian regulation algorithm had the highest accuracy in prediction and the best model consisted of one hidden layer with three neurons (5–3–1 topology) with logarithm sigmoid activation function. The results highlighted that the ANN model (R² = 0.91) outweighs the MLR model (R² = 0.86). The results also approved that the proposed ANN model was capable to use different datasets with high generalizability. The results of sensitivity analysis claimed the significant role of water and fuel consumption and consequently, the importance of establishing management policies to optimizing the use of these inputs in the pomegranate orchards.

在水果生产系统中实现环境可持续性需要在能量流方面对系统进行准确监测。在这方面，本文使用回归模型（即 Cobb-Douglas）、多元线性回归（MLR）和人工神经网络（ANN），研究了伊朗石榴园的能量流。与其他类似的研究不同，在当前的研究中，交叉验证技术被用来评估选择数据集将如何修改模型的行为。结果表明，石榴生产的总能耗和能源效率为13,634.13 MJ ha ^-1和 1.01，分别。结果还表明，在研究的 MLR 模型中，纯二次模型的性能最好。此外，在 13 种评估的 ANN 训练算法中，贝叶斯调节算法的预测精度最高，最好的模型由一个隐藏层和三个神经元（5-3-1 拓扑）和对数 sigmoid 激活函数组成。结果突出显示，ANN 模型 ( R ²  = 0.91) 超过了 MLR 模型 ( R ² = 0.86）。结果还证实，所提出的 ANN 模型能够使用具有高泛化性的不同数据集。敏感性分析的结果表明水和燃料消耗的重要作用，因此，建立管理政策以优化石榴园中这些投入物的使用的重要性。