Visibility Forecast For Airport Operations By LSTM Neural Network

Visibility forecast is a meteorological problems which has direct impact to daily lives.For instance,timely prediction of low visibility situations is very important for the safe operation in airports and highways.Especially for aviation,low visibility can cause visual obstacles to aircraft flight and endan-ger flight,safety.In this paper,we try to predict visibility(especially for low visibility data)through machine learning methods.We first analyse the visi-bility data of the Beijing Meteorological Observation 54511 through statistical characteristics and correlation analysis of the data,which will provide more priori information for our model.In addition,we also fix the missing data by spatially nearest neighbor interpolation.In our experiments,we investigate the use of Long Short-Term Mem-ory(LSTM)model to predict visibility.By adjusting the loss function and network structure,we optimize the original LSTM model to make it more suitable for practical applications,which is superior to previous models in short-term low visibility prediction.We predict visibility through regression and classification methods,respectively.For the regression model,we construct a weighted loss function to improve the prediction accuracy of low visibility data.Taking the Mean Absolute Error as evaluation criterion,we find that the prediction results of LSTM model are better than those of the commonly used random forest method and multi-layer perccptron model.In addition,there is a systematic" time delay problem"when LSTM model is used to solve the problem of time series prediction,which is mainly systematic error of the model due to the insufficient amount of data in the training set.For the classification model,we classify the data by the commonly used visibility standards for airports operations regarding safe aircraft taking-off and landing.Through the classification model,we find that although the ac-curate numerical prediction results of visibility are not available,the grading prediction results are more accurate,which can provide more reference infor-mation for aircraft taking-off and landing in practical applications.