Large array antennas are often assembled with several subarrays, and assembling errors containing position and orientation are the key factors determining the final radiation pattern. As an important feature of radiation patterns, the antenna gain significantly influences the design of the antenna. However, little research has been conducted on the accurate prediction and detailed compensation of gain with assembling errors. In this work, the authors propose an accurate gain prediction model using an improved extreme gradient boosting (XGBoost) algorithm and the transfer learning method. Knowledge from both the simulation data and experience is converted to weights to help train the improved XGBoost model. Experimental data are then used to modify the model for complex factors, such as mutual coupling and element type. Compensation methods are proposed to provide directions to limit the degradation of the gain within a range by controlling the assembling errors. Experiments are conducted on a platform with a 3 × 3 subarray antenna. The results indicate that the proposed gain prediction model is more accurate than the model developed using artificial neural network, support vector regression, and existing XGBoost algorithms. The steps of gain compensation are also reduced with the proposed compensation methods.

中文翻译：

---

基于改进的XGBoost和转移学习的具有装配误差的子阵列天线增益预测和补偿

大阵列天线通常与几个子阵列组装在一起，并且包含位置和方向的组装误差是确定最终辐射方向图的关键因素。作为辐射方向图的重要特征，天线增益会显着影响天线的设计。但是，关于装配误差的准确预测和详细补偿的研究很少。在这项工作中，作者提出了一种使用改进的极端梯度提升（XGBoost）算法和传递学习方法的准确增益预测模型。来自模拟数据和经验的知识都将转换为权重，以帮助训练改进的XGBoost模型。然后，将实验数据用于修改复杂因素的模型，例如相互耦合和元素类型。提出了补偿方法以提供方向，以通过控制组装误差来将增益的下降限制在一定范围内。实验是在带有3×3子阵列天线的平台上进行的。结果表明，提出的增益预测模型比使用人工神经网络，支持向量回归和现有XGBoost算法开发的模型更为准确。所提出的补偿方法也减少了增益补偿的步骤。和现有的XGBoost算法。所提出的补偿方法也减少了增益补偿的步骤。和现有的XGBoost算法。所提出的补偿方法也减少了增益补偿的步骤。