LSTM-based Passenger Traffic And Flight Fare Prediction Research On Civil Aviation Routes

Deep learning technology has made great development and successful application in sequence prediction,such as machine translation.Based on this kind of sequence prediction,many researchers have applied it to time series prediction and made great progress.This kind of time series prediction method based on deep learning is better than the traditional time series prediction method in prediction effect,and it can be effective dealing with the non-linear and non-stationary relationship of multi-dimensional time series,the traditional time series prediction method is difficult to achieve this.There is a complex non-linear and non-stationary relationship between the passenger volume data and the air ticket price data in the time dimension.Therefore,this paper will use deep learning technology to predict the passenger volume and ticket price of civil aviation based on time series.After understanding the current situation of the research on the prediction of civil aviation passenger volume and ticket price,this paper first processes and analyzes the civil aviation data of Yunnan small and medium-sized Aviation Division,then uses the time series prediction method based on deep learning to predict the passenger volume and ticket price of the route,and finally carries out the experimental analysis.The specific research work is as follows:(1)In view of the large and complex data of Yunnan small and medium-sized Aviation Division,this paper carries out ETL processing and multi-dimensional analysis,portraits passengers and flights,fully understands the civil aviation data of Yunnan small and medium-sized Aviation Division,and makes data preparation for passenger volume prediction and flight ticket price prediction.(2)In this paper,the single dimension and multi-step prediction of civil aviation passenger volume is carried out.Based on long and short-term memory and attention mechanism,which can effectively capture the time-series dependence of time series,a passenger volume prediction model(PVPM)is proposed to predict the passenger volume data of Kunming-Xishuangbanna(KMG-JHG),Kunming-Lijiang(KMG-LJG)and KunmingChengdu(KMG-CTU)routes.Firstly,the definition and formal representation of passenger volume prediction are given.Then,the PVPM model and its training and optimization are introduced.Finally,experiments are carried out and compared with the benchmark models such as gated recurrent unit and temporal convolutional network.The results show that PVPM is relatively better in RMSE and MAE evaluation indexes.(3)In this paper,a multi-step prediction model(FFFM)is proposed to predict the ticket prices of 11 flights on the Kunming-Xishuangbanna route.FFFM uses a two-layer one-dimensional convolutional neural network to extract the local pattern of the fare series and obtain its key local feature information.It uses a multi-layer LSTM to obtain the complex time series correlation of the fare series.At the same time,the attention mechanism is used to calculate the weight of each time step on the second layer LSTM,which makes the model pay more attention to the time step that plays a role in the prediction target,and effectively obtains the dependence information of the target output on each time step.First,it analyzes the forecast of flight ticket price,and then describes the FFFM model.Finally,the experimental analysis is carried out.The ablation experiments show that the one-dimensional convolutional neural network and attention mechanism can effectively improve the prediction performance of FFFM,and the accuracy of flight ticket price prediction can be improved by considering the information of airlines,aircraft types,week attributes,etc.FFFM is compared with the standard one-dimensional convolutional neural network,TCN and other benchmark models on RMSE and MAE evaluation indexes to verify the prediction performance.The results show that FFFM has a better prediction effect on flight ticket price.