在现代空管系统中，高效准确地识别和分类危险源文本数据对于保障飞行安全至关重要，空管危险源数据指的是那些可能影响航空安全的潜在因素、条件或事件的信息集合，然而现有的文本分类方法难以应对数据类别多样性和类别不平衡问题。当下迫切需要开发适用于空管系统的高效分类方法，以提高飞行安全水平。针对单一学习器用于空管危险源文本分类存在的类别分布较多，难以捕捉类别数据不平衡时的文本特征导致预测精度下降的问题，提出基于Stacking训练思想的、两次加权的改进集成模型。首先，参考双防机制对危险源和安全隐患完成类别划分；再采用词频-逆文档频率（term frequency-inverse document frequency, TF-IDF）算法提取预处理后的危险源文本特征完成向量化，并利用合成少数类过采样技术（synthetic minority over-sampling technique, SMOTE）和自适应合成过采样算法（adaptive synthetic sampling approach, ADASYN）分别随机生成向量化后的少数类文本，使文本数据集的类别分布趋于平衡；再从基学习器每折交叉验证的F₁分数加权和基学习器之间敏感性评估机制动态加权两方面改进Stacking集成模型，提高类别不平衡危险源文本的分类性能。在所构建的数据集上的实验结果表明：相较于SMOTE +改进集成模型，ADASYN +改进集成模型的精确率、召回率和F₁分数分别提升0.9、1.1和1.0个百分点，较好地抑制处理多数类别过拟合的问题，实验结果验证了所提算法的有效性。

Modern air traffic management systems necessitate efficient and accurate identification and classification of hazard-related text data to ensure flight safety. Air traffic control hazard data encompasses information on potential factors, conditions, or events that may adversely impact aviation safety. Existing text classification methods face challenges due to the diversity of data categories and imbalances within classes. An enhanced ensemble model based on the Stacking framework, incorporating a dual-weighting mechanism was proposed for improved performance. A dual-protection strategy was implemented to categorize hazards and safety risks systematically. The methodology employed the term frequency-inverse document frequency(TF-IDF) algorithm to extract and vectorize features from preprocessed hazard texts. To address class imbalance, the synthetic minority over-sampling technique(SMOTE) and adaptive synthetic sampling approach (ADASYN) algorithms were utilized to generate synthetic samples for minority classes. The Stacking ensemble model was refined by dynamically weighting the F₁ scores derived from cross-validation of base learners and integrating a sensitivity assessment mechanism across the ensemble. Experimental results on the constructed dataset demonstrate that the ADASYN-enhanced ensemble model achieves notable improvements in precision, recall, and F₁ scores by 0.9%, 1.1%, and 1.0%, respectively, effectively mitigating overfitting in majority classes. The proposed algorithm significantly enhances the classification performance of imbalanced hazard text categories, contributing to the advancement of safety risk management in air traffic control.