| In order to solve the problem of unbalanced development of airspace resources and airspace flow,both the new generation of air intelligent traffic system and ASBU take 4D flight path prediction as the key content of future air traffic control development.At present,China is faced with dual challenges of tight airspace environment and lagging control level.At present,the core of solving this problem is to use limited resources to achieve effective management of aircraft on air routes and routes.Based on this situation,the concept of Traiectory Based Operation(TBO)based on four-dimensional flight paths was first proposed by the International Civil Aviation Organization(ICAO),and listed as the core operation concept of the next generation air traffic management system.As the core support technology based on flight path operation,high precision 4-D flight path prediction technology determines the sustainability of the development of the next generation air traffic management system.TBO mainly takes accurate flight path prediction of aircraft in a period of time in the future as a shared resource within the air traffic control system,and directly controls the efficient operation of aircraft in the airspace.Therefore,to explore a high precision 4D flight path prediction technology is the core problem to be solved.Firstly,the advantages of ADS-B technology and data content are briefly summarized in this paper,and the acquired ADS-B data are cleaned and selected.The main work is to remove some noise point sets and redundant points and interpolate the track vacancy value.This part is the basic work of neural network construction,and the processing of data values directly relates to the performance of the prediction model.The processed flight path data needs to be saved and used as the input part of the neural network prediction model.Secondly,the second part of the systematic work is to use LSTM neural network to build the model.LSTM model is often used in time series prediction because of its simple structure and excellent prediction performance.However,with the further increase of the complexity of airspace application,the accuracy of the prediction model of a single LSTM neural network has been unable to meet the ideal accuracy requirements.Therefore,on the basis of LSTM neural network,this paper uses CNN module to extract the spatial dimension characteristics of the adjacent region of the track,and LSTM module to extract the dependency relationship of the time dimension of the track.Combining the advantages of Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM),the time and space features of the track are fully integrated.The CNN-LSTM track prediction model and CNN-Lstm-Attention track prediction model with higher prediction accuracy were constructed successively.Next,in order to further reduce the model errors caused by model parameter setting,initial weight and threshold,avoid the prediction algorithm falling into local optimization as soon as possible and better verify the model prediction performance,this thesis uses the intelligent bionic algorithm to optimize the parameters of the CNN-LSTM-attention model previously constructed.Two trajectory prediction models based on particle swarm optimization algorithm and gray Wolf optimization algorithm are constructed.Finally,the experimental environment was built to complete the simulation and verification of the prediction accuracy of the experimental model and the comparison model.Aiming at the normalization and standardization of real flight path data samples on the route,the results of flight path prediction experiment and comparison model are output according to model initialization and parameter setting.R2(coefficient of determination),MAE(mean absolute error),RMSE(root mean square error)and MAPE(mean absolute percentage error)were selected as the evaluation criteria of the model.The prediction performance of the model was evaluated through the four dimensions of longitude,dimension,height and velocity.The experimental results showed that,The selection of superparameters(training batches,number of neurons,number of hidden layers)has a certain degree of influence on the performance of the prediction model.The improved neural network prediction model basically fits the actual flight path in terms of prediction effect,and the accuracy is higher than the traditional neural network prediction model,and it has better robustness and stability. |
