传统页岩气井产量预测方法难以对储层参数、压裂参数与产量的关系做出有效分析，而机器学习方法具有解决这一问题的能力。提出了基于物理意义和随机组合的方法构建特征参数，并采用小批量梯度下降法（MBGD）作为训练函数，建立了针对页岩气井产量预测的改进人工神经网络预测模型。然后结合实例，利用改进后的人工神经网络模型对页岩气井产量进行预测，并通过计算均方误差（MSE）和修正决定系数（T）的值对模型的优劣程度和预测精度进行评价。结果表明，建立的改进神经网络模型预测产量结果与实际产量值吻合度较高，且相比传统的BP（误差反向传播算法）神经网络模型，在预测精度和稳定性方面具有明显优势。该模型能为页岩气储层压裂优化设计以及产能评价提供重要支持。

Traditional methods for predicting shale gas well production often struggle to effectively analyze the complex relationship between reservoir parameters, fracturing parameters and production. To address these challenges, a novel approach is introduced, involving the construction of characteristic parameters based on physical meaning and random combination. The small batch gradient descent method（MBGD） is adopted as the training function to develop an improved artificial neural network prediction model for shale gas well production. An example is utilized to demonstrate the effectiveness of the improved artificial neural network model in predicting shale gas well production. The model’s performance is evaluated using the mean squared error（MSE） and the modified determination coefficient（T）. The results indicate that the predictions from the improved network model align well with the actual production data. Moreover, the model exhibits superior prediction accuracy and stability compared to the traditional BP（error backpropagation algorithm） neural network model. With its high accuracy and reliability, the proposed model can provide valuable support for fracturing optimization design and productivity evaluation in shale gas reservoirs.