4D Trajectory Prediction For Arrival Flights In Terminal Area Based On Improved LSTM

At present,with the expanding scale of passenger transportation and route network,China is bound to have the problem of shortage of airspace resources,which will directly affect the development of civil aviation industry and the safe operation of flights.To solve this problem,ICAO has proposed a next-generation air traffic management operation concept based on trajectory operation(TBO).4D trajectory is the core component of TBO operation,which can accurately manage and control aircraft operation.Therefore,improving the accuracy of 4D trajectory prediction has become the core problem that needs to be solved urgently.The large amount of historical track data and the rich spatial and temporal characteristics and abrupt changes in track location data make the 4D track prediction full of uncertainty and complexity.In addition,the use of the common basic long and short-term memory neural network(LSTM)for 4D track prediction is prone to the problems of missing prediction dimensions,insufficient processing of raw data,data mutation,and the difficulty of extracting spatio-temporal features from the track data at the same time,resulting in low accuracy of track prediction and unstable prediction results.To solve the above problems,this paper combines Convolutional Neural Networks(CNN),Bidirectional Gated Recurrent Unit(BIGRU),attention mechanism and Autoregressive Integrated Moving Average(ARIMA)to improve the basic LSTM neural network,and propose a 4D trajectory prediction model based on CNN-SBIGRU-ARIMA-AM hybrid network.The model can extract the spatial dimensional features of the track data using a one-dimensional CNN network,extract the temporal dimensional features of the track data using a BIGRU network,solve the problem of abrupt changes in the track data using an ARIMA model,and reasonably assign weights with an attention mechanism to solve the problems of missing prediction dimensions,abrupt changes in the data,and the inability to extract spatio-temporal features at the same time.The prediction accuracy is reduced due to data problems.In addition,a clustering sample set,a prediction sample set and three evaluation indexes are established to help the 4D trajectory prediction in the later stage.Finally,the real ADS-B trajectory data from Zhuhai Jinwan Airport is used for example validation,and the three comparison models established in this paper are compared with each other for analysis.The experimental results show that the prediction accuracy of the CNN-SBIGRU-ARIMA-AM hybrid prediction model is higher than that of the comparison model,and its prediction error is lower than that of the comparison model in all dimensions.In addition,this paper uses the prediction results to generate 4D trajectory planning maps,which provide a reference for the actual operation.