TimeGAN-Informer长时机场能见度预测

Long-term prediction of visibility using TimeGAN Informer

能见度的预测对机场的业务决策、保障飞机的安全起降具有重要的意义。针对现有能见度预测模型预测时间较短的问题,提出一种基于TimeGAN Informer(Time Generative Adversarial Network-Informer)的机场能见度预测方法。利用2018—2022年气象和污染物数据,通过相关系数法和递归特征消除法提取出能见度的主要影响因素,使用TimeGAN时间序列生成对抗网络对数据进行扩充,并将Informer长时间序列预测模型应用于能见度预测。结果显示:当预测步长为1 d、2 d、3 d时,TimeGAN Informer的绝对误差(Mean Absolute Error,MAE)分别为2.42、3.13、3.57,比Informer分别降低了0.29、0.27、0.28,比长短时记忆网络(Long Short-Term Memory,LSTM)分别降低了0.28、0.49、0.63;均方根误差(Root Mean Square Error,RMSE)分别为3.03、3.7、4.09,比Informer分别降低了0.38、0.22、0.24,比长短时记忆网络(LSTM)分别降低了0.3、0.5、1.04;百分误差小于30%的分别占测试样本集的78.07%、70.68%、63.84%。尽管随着步长的增加预测效果变差,但在预测步长为3 d时,多数样本的预测误差仍小于30%,实现了对机场区域较为准确的长时能见度预测。

The accurate prediction of visibility is pivotal for airport operations,guiding crucial business decisions,and ensuring the safe takeoff and landing of aircraft.However,existing visibility prediction models often face limitations in providing sufficiently long-term forecasts.To address this challenge,this paper introduces a novel method for long-term visibility prediction at airports,leveraging the innovative combination of TimeGAN and Informer,termed TimeGAN Informer.When dealing with a dataset containing a limited number of samples,the risk of overfitting becomes a concern.Utilizing the Time series Generative Adversarial Network(TimeGAN)can effectively mitigate this issue by expanding the dataset,thereby enhancing the accuracy of the deep learning model.Consequently,for this study,meteorological and pollutant data from Tianjin Binhai Airport and Tianjin spanning from 2018 to 2022 were selected as research data.Through methods such as correlation coefficient analysis and recursive feature elimination,the main influencing factors on visibility were extracted.Subsequently,the visibility time series data were augmented using TimeGAN,followed by employing the long time series prediction model,Informer,for visibility prediction.The findings indicate that for prediction steps of 1 day,2 days,and 3 days,the Mean Absolute Error(MAE)of the TimeGAN Informer model were 2.42,3.13,and 3.57,respectively.Compared to the Informer model,this represented a reduction of 0.29,0.27,and 0.28,respectively.Similarly,compared to the LSTM model,reductions of 0.28,0.49,and 0.63 were observed.Moreover,the Root Mean Square Error(RMSE)values were 3.03,3.7,and 4.09,respectively.In comparison to the Informer model,this indicated reductions of 0.38,0.22,and 0.24,respectively.Likewise,compared to the LSTM model,reductions of 0.3,0.5,and 1.04 were observed.Additionally,the percentage of errors less than 30%accounted for 78.07%,70.68%,and 63.84%of the test sample set,respectively.Despite the decline in prediction performance as the prediction step size increases,the majority of samples still exhibit prediction errors of less than 30%when the prediction step size is extended to 3 days,indicating relatively accurate long-term predictions.In conclusion,the TimeGAN Informer model proposed in this study demonstrates good performance in long-term visibility prediction and holds potential for application in airport visibility forecasting.