为了有效检测和清洗飞行轨迹异常数据，研究了在数据采集过程中容易出现的重复、缺失、错误等问题，基于机器学习提出了一种准确性和鲁棒性较高的数据异常检测与清洗方法。首先，结合线性插值和最大最小值算法对数据进行去重、插值和归一化处理；其次，基于GRU构建飞行轨迹预测模型；最后，利用SVDD模型对飞行轨迹数据进行异常检测，当实际数据与预测数据偏差较大时使用预测数据进行替换，达到清洗效果。结果表明：相较于其他LSTM算法模型，本文方法得到清洗后的飞行轨迹数据具有更高的准确性，F₁分数平均能达到0.932，较好地逼近原始飞行轨迹数据；通过检验位置偏移（干扰）、高度偏差（篡改）、航路交叉（替换）3种攻击方法，证明该方法具有较高的鲁棒性。本文方法能够准确预测飞行轨迹，并对异常数据进行有效检测与清洗，提高数据质量，有助于准确分析航班运行情况。

To effectively detect and clean anomalous flight trajectory data, issues such as duplication, missing values, and errors that frequently occur during data collection were studied. A data anomaly detection and cleaning method with high accuracy and robustness was proposed based on machine learning. First, duplicate removal, interpolation, and normalization of the data were performed by combining linear interpolation and the min-max scaling algorithm. Next, a flight trajectory prediction model was constructed based on GRU. Finally, the SVDD model was utilized for anomaly detection of flight trajectory data. When the actual data significantly deviated from the predicted data, the predicted data was used for replacement to achieve the cleaning effect. The study results showed that compared to other LSTM algorithm models, the proposed method yielded flight trajectory data with higher accuracy after cleaning, with an average F₁ score of 0.932, closely approximating the original flight trajectory data. By examining three attack methods of position shift (tampering), altitude deviation (interference), and route crossing (replacement), the robustness of the proposed method was demonstrated. The method can accurately predict flight trajectories, effectively detect and clean anomalous data, improve data quality, and contribute to accurate analysis of flight operations.