Short-term Forecasting Using A Combined Model Based On Time Series Decomposition And LSTM Neural Networks

With the rapid development of railway,more and more cities are facing a series of problems such as traffic congestion,so it is urgent to build rail transit.Not only that,for example,Beijing and Shanghai,which developed urban rail transit earlier,are also facing the complexity and difficulty of line operation.How to effectively provide effective data analysis for rail transit operation,and provide safe and high-quality services for passengers,has become an important topic in the field of rail transit.As an important collectable data in the whole operation system,the inbound passenger flow of railway is of great significance to analyze the spatial and temporal distribution characteristics and change rules of passenger flow.Firstly,the paper introduces AFC system used in the operation of railway,and analyzes the advantages and disadvantages of the system in the process of data transmission.On this basis,it puts forward the detection method of identifying data outliers,and corrects the outliers according to the characteristics of big data.At the same time,the time-space characteristics of the typical stations of rail transit are analyzed,which shows that the distribution characteristics of passenger flow are different in different surrounding environment and time.In the third chapter,aiming at the strong non-stationary characteristics of the inbound passenger flow data of rail transit,the paper uses STL time series decomposition to decompose the inbound passenger flow data of Beijing Railway Station and Tiananmenxi Station,and uses sample entropy and white noise detection to analyze the three components obtained from the decomposition.The residual component with effective information is decomposed twice by using the empirical mode decomposition algorithm,and the regular IMF component is obtained.In the fourth chapter,the advantages of long-term and short-term memory neural network in predicting time series are introduced.The key parameters of the neural network model,such as the activation function,the number of input nodes and so on,are adjusted,and the three super parameters of the model are optimized by particle swarm optimization algorithm,which makes the neural network improve its accuracy in the prediction of data to the greatest extent.In the third chapter,inbound passenger flow obtained by the second decomposition are used as input values in the model prediction,and the prediction results are compared with other models horizontally to verify the effectiveness of the combined model.Through comparison,it is proved that the combination model proposed in this paper has a significant reduction in prediction error compared with other models.