An OD Market Air Passenger Flow Prediction Method Based on Spatio-Temporal Convolutional Network

Abstract

The present invention discloses an OD market air passenger flow prediction method based on a spatio-temporal convolutional network, belonging to the field of big data technology. It includes statistically organizing and storing the air passenger flow data from each departure airport to the destination airport within the regional multi-airport system, building a prediction model for the OD market air passenger flow based on the spatio-temporal convolutional network, determining the optimal hyperparameter settings of the prediction model under different data sets, and predicting the air passenger flow of several OD markets according to the historical air passenger flow data. It solves the technical problem of using the spatio-temporal convolutional network to predict the air passenger flow and realizing the simultaneous prediction of the passenger flows of multiple OD markets from multiple departure airports in the same region to the same destination airport. The present invention first applies the spatio-temporal convolutional network to the field of air passenger flow prediction, realizes the simultaneous prediction of the passenger flows of multiple OD markets from multiple departure airports in the same region to the same destination airport, and provides a new idea for air passenger flow prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of big data, and particularly relates to a method for predicting the air passenger flow in the OD market based on a spatio-temporal convolutional network. Background Art

[0002] For a long time, accurately predicting the air passenger flow has always been the primary task for all entities in the air transportation industry to carry out overall planning. Airlines can judge whether the OD market can make a profit or adjust the existing capacity supply based on this, and airports can timely evaluate the matching situation between the current infrastructure capacity and the passenger flow. However, the passenger flow is affected by various factors in the air transportation market and has uncertainty and unpredictability. Therefore, how to scientifically and accurately predict the air passenger flow in the OD market has always been one of the research hotspots in the field of civil aviation transportation.

[0003] According to the existing research at home and abroad, based on the model construction mechanism, the traffic flow prediction methods in the traffic field can be divided into parametric methods and non-parametric methods. The model structure of the parametric method needs to be determined according to corresponding theoretical assumptions and statistical analysis, and the model parameters are estimated using relevant historical time series data. The relevant models of this method include: least squares regression model, autoregressive moving average model, autoregressive distributed lag model, time-varying parameter model, vector autoregression, grey prediction model, etc. In addition, from the perspective of system dynamics, dynamic sub-models of various factors affecting the change of passenger flow can be constructed, and the air passenger flow prediction results can be obtained through system simulation. Since the parametric method prediction model starts from the idea of econometric modeling and attempts to establish and fit a specific model that can accurately reflect the change of passenger flow, but it is difficult for such models to characterize the non-linear relationship of causal variables and the generalization ability for dealing with data with large fluctuations is insufficient.

[0004] Non-parametric methods based on machine learning have also been widely used in market passenger flow prediction. The relevant application means of this method include: BP neural network, support vector regression, random forest regression, adaptive fuzzy neural network, wavelet neural network, and combined prediction model, etc. Although the results obtained by the combined prediction model are better than those of a single prediction model, most of the established models are shallow structures and cannot accurately mine the deep information behind large-scale data, and machine learning methods still have limitations in dealing with high-dimensional non-linear traffic data, so it restricts the further expansion of machine learning prediction.

[0005] In recent years, against the backdrop of an explosive growth in traffic data, deep learning, which has evolved from the research on artificial neural networks, has provided a new approach to further improving the accuracy of passenger flow prediction. In particular, convolutional neural networks (CNNs), with their unique network structure characteristics, have been widely applied in the fields of computer vision and image recognition. Traditional neural networks take data in the form of one-dimensional vectors as input, while the difference of CNNs lies in taking data in the form of multi-dimensional matrices as input and replacing the original neuron connection method with convolutional operations. Such multi-layer convolutional operations can automatically extract the correlation features between data with spatial characteristics, thereby reducing the errors caused by artificially constructed features. Therefore, it can be migrated and applied to the prediction problem of air passenger flow data with spatio-temporal characteristics.

[0006] In summary, previous methods for predicting air passenger flow have mostly focused on parametric methods, shallow machine learning, and combined prediction. Among them, the construction of the model in the parametric method is largely affected by subjective human factors, and at the same time, the accuracy of parameter estimation will affect the final prediction accuracy of the model; while in shallow machine learning and combined prediction, there will be situations such as overfitting in the prediction process. In addition, the prediction of air passenger flow only considers the historical data of a single OD market in the time dimension, ignoring the impact of adjacent airports with linkage effects in the spatial position on the passenger flow of the predicted OD market, and failing to predict the passenger flow of the OD market from the perspective of multiple airports in the region. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for predicting air passenger flow in OD markets based on a spatio-temporal convolutional network, which solves the technical problem of using a spatio-temporal convolutional network to predict air passenger flow and realizing the simultaneous prediction of passenger flows in multiple OD markets from multiple departure airports to the same destination airport in the same region.

[0008] To achieve the above purpose, the present invention adopts the following technical solutions:

[0009] A method for predicting air passenger flow in OD markets based on a spatio-temporal convolutional network, comprising:

[0010] Select a multi-airport system in a certain region through a client server, and count, organize and store the air passenger flow data from each departure airport to the destination airport in the multi-airport system in the region;

[0011] The prediction model server obtains the air passenger flow data, preprocesses the air passenger flow data, constructs an OD passenger flow grid map and an external influence factor feature vector, builds a prediction model for air passenger flow in OD markets based on a spatio-temporal convolutional network, and forms a corresponding training set for the prediction model;

[0012] The OD market refers to the air passenger transportation market from the Origin Airport to the Destination Airport.

[0013] By adjusting the data characteristics and the network structure of the prediction model, the optimal hyperparameter settings of the prediction model under different datasets are determined; the hyperparameters include the hyperparameters of the network structure and the hyperparameters of the data characteristics. The hyperparameters of the network structure include the number of convolutional layers, the size of the convolutional kernel, and the number of convolutional kernels. The hyperparameters of the data characteristics include the sample length of the selected trend segments and the sample length of the selected periodic segments.

[0014] The central server predicts the air passenger flow of several OD markets based on the historical data of air passenger flow.

[0015] Preferably, the management personnel input the three-letter codes of the departure airports, the three-letter codes of the destination airports, and the monthly passenger flow corresponding to each OD market through the client server.

[0016] The client server automatically arranges and generates the passenger flow time series data of each OD market according to the time, and the client server transmits the passenger flow time series data to the database module for storage.

[0017] Preferably, the air passenger flow data is preprocessed, including wavelet threshold denoising and data normalization processing. Among them, wavelet threshold denoising adopts the method of two-layer decomposition with soft threshold.

[0018] Preferably, an OD passenger flow grid map and an external influence factor feature vector are constructed, specifically including:

[0019] Step A1: Define O = {o 1 , o 2 , …, o N} to represent the set of departure airports, where N is the total number of departure airports; d represents the destination airport; <o, d> represents an OD market, and o ∈ O.

[0020] The prediction model server maps the N departure airports to the grid map R of I×J according to the longitude, latitude and relative geographical location distribution of the departure airports in each regional multi-airport system. The grid (i, j) represents any grid in the grid map R, where i represents the row number of the grid and j represents the column number of the grid. The grid (i, j) represents the passenger flow from the departure airport o in the system at this position to the destination airport d outside the system in the t-th month. Use a two-dimensional tensor X t ∈ R I×J to represent the OD passenger flow grid map from the departure airports in the system to the same destination airport outside the system in the t-th month. Among them,

[0021] Step A2: Define the external influencing factors including two features: month attribute and whether it includes holidays. Process the external influencing factors into a 0-1 vector, that is, use 0 or 1 to represent the two features of the predicted month attribute and whether it includes holidays. The first 12 bits correspond to January to December, and the 13th bit represents whether it includes holidays;

[0022] Preferably, the prediction model server uses the functional module in the Keras neural network library of Python to build a prediction model for the air passenger flow of the OD market based on the spatio-temporal convolutional network. Through the known m historical observation values X M ={X t |t = 1, 2, …, m}, predict the air passenger flow X K ={X t |t = m + 1, m + 2, …, m + k};

[0023] By changing the two parts of hyperparameters, namely the network structure and data characteristics, to determine the most suitable hyperparameter settings for the prediction model, use the Adam optimizer to train and optimize the prediction model, and adopt an early stopping strategy during the training process to avoid model overfitting.

[0024] Preferably, the specific steps for the central server to predict the air passenger flow of several OD markets include:

[0025] Step B1: Extract fragments of the OD passenger flow grid map. According to the dependence of the passenger flow on the time dimension, extract two time fragments at different time intervals for the prediction time points respectively: the trend grid picture fragment X Tre and the periodic grid picture fragment X Per , and the specific extraction form is as follows:

[0026]

[0027]

[0028] Among them, l tre and l per As adjustable data characteristic hyperparameters in the prediction model, they represent the sample length of the trend fragment and the sample length of the periodic fragment respectively;

[0029] Step B2: Build a spatio-temporal convolutional network. Based on the two grid picture fragments extracted at different time intervals, build spatio-temporal convolutional network branches with the same structure to capture the spatio-temporal characteristics of the OD passenger flow. The spatio-temporal convolutional network branch is based on S + 1 (S ≥ 1) convolutional layers;

[0030] Taking the spatio-temporal convolutional network branch of the trend part as an example, use Represents a trend grid picture segment, through the first convolutional layer C 1 Convert (X Tre ) (0) into a new tensor (X Tre ) (1) , and the conversion formula is as follows:

[0031]

[0032] Among them, and are the learning parameters of the first convolutional layer. After the operation of convolutional layer C 1 , continue to add (S - 1) convolutional layers according to the above formula. After the Sth convolutional layer, then pass through a convolutional layer C S+1 that only contains one convolutional kernel, and finally obtain the output result (X Tre ) (S+1) of the trend part; Similarly, use the same operation above to construct the spatio-temporal convolutional network branch of the periodic part, and obtain the output result (X Per ) (S+1) ;

[0033] Step B3: Construct an external influence factor network. The external influence factors considered by the prediction model include two aspects: month attribute and whether there is a holiday. Based on the tth prediction time point, obtain the corresponding external influence factor feature vector E t , and use a two-layer fully connected neural network branch as the external influence factor network. The first layer can be regarded as an embedding layer, mainly to quantitatively add external factors to the prediction model. The second layer is to map the features obtained by the first layer into a high-dimensional tensor, and its size should be the same as X t to facilitate fusion with the output result of the spatio-temporal convolutional network, and obtain the output result X Ext ;

[0034] Step B4: Fusion to obtain the prediction result. Assign different weight matrices to the output results (X Tre ) (S+1) and (X Per ) (S+1) of the spatio-temporal convolutional network in the form of learning parameters, and perform aggregation to obtain the weighted output result. The calculation formula is as follows:

[0035] X Con = W Tre * (X Tre ) (S+1) + W Per * (X Per ) (S+1)

[0036] Among them, * represents the Hadamard product; W Tre and W Perrespectively represent the weights of the trend and periodic parts, that is, the influence degrees of the outputs of the two parts on the final prediction result. Further, the calculation result X is passed through the tanh function Con and the output result X of the external influence factor network Ext are mapped to the range of [-1, 1] to obtain the final prediction result X K , and the calculation formula is as follows:

[0037] X K =tanh(X con +X Ext )

[0038] Taking the mean square error between the predicted value matrix X K and the true value matrix as the objective to train the prediction model:

[0039]

[0040] where θ represents all the learning parameters of the model.

[0041] A method for predicting the air passenger flow of OD markets based on a spatio-temporal convolutional network according to the present invention solves the technical problem of using a spatio-temporal convolutional network to predict the air passenger flow and realizing the simultaneous prediction of the passenger flows of multiple OD markets from multiple departure airports in the same region to the same destination airport. The present invention first applies the spatio-temporal convolutional network to the field of air passenger flow prediction to realize the simultaneous prediction of the passenger flows of multiple OD markets from multiple departure airports in the same region to the same destination airport, providing a new idea for air passenger flow prediction; the present invention regards the convolutional kernels in the convolutional network as the medium for capturing the spatial correlation of OD passenger flows, so that multiple convolutional kernels in a convolutional layer can accurately capture the complex potential feature relationships of OD passenger flows in the spatial dimension, improving the prediction effect; the present invention comprehensively considers the time dependence, spatial correlation of OD passenger flows and the role of external influence factors, uses RMSE as the evaluation criterion of the model, and takes the OD passenger flow data from 16 major airports in the Yangtze River Delta to Guangzhou Baiyun International Airport as an example. The prediction model is compared with ARIMA (Autoregressive Integrated Moving Average Model), SVR (Support Vector Regression), Elman neural network and LSTM (Long Short-Term Memory Network), and the fitting effect is better. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is the overall flowchart of the method for predicting the air passenger flow of OD markets based on a spatio-temporal convolutional network according to the present invention;

[0043] Figure 2 is the framework of the prediction model for the air passenger flow of OD markets based on a spatio-temporal convolutional network according to the present invention;

[0044] Figure 3It is a schematic diagram of the structure of the OD passenger flow grid map in the embodiment of the present invention;

[0045] Figure 4 It is the passenger flow fitting result of the model method proposed in the embodiment of the present invention and other models in different OD markets;

[0046] Figure 5 It is the passenger flow fitting result of the model method proposed in the embodiment of the present invention and other models at different prediction intervals. Detailed implementation manners

[0047] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0048] As Figures 1 - 5 shown, a method for predicting the air passenger flow in the OD market based on a spatio-temporal convolutional network includes:

[0049] Select a multi-airport system in a certain area through the client server, and count, organize and store the air passenger flow data from each departure airport to the destination airport in the multi-airport system in the area;

[0050] The prediction model server obtains the air passenger flow data, preprocesses the air passenger flow data, constructs an OD passenger flow grid map and an external influence factor feature vector, builds an OD market air passenger flow prediction model based on the spatio-temporal convolutional network, and forms a training set corresponding to the prediction model;

[0051] The OD market refers to the air passenger transportation market from the Origin Airport to the Destination Airport;

[0052] The destination airport is the destination airport outside the multi-airport system.

[0053] By adjusting the data characteristics and the network structure of the prediction model, determine the optimal hyperparameter settings of the prediction model under different data sets; the hyperparameters include hyperparameters of the network structure and hyperparameters of data characteristics. The hyperparameters of the network structure include the number of convolutional layers, the size of the convolutional kernel, and the number of convolutional kernels. The hyperparameters of data characteristics include the sample length of the selected trend segment and the sample length of the selected periodic segment;

[0054] The data set refers to the air passenger flow data from each departure airport to the destination airport in the multi-airport system in the area (that is, the air passenger flow in multiple OD markets); different multi-airport systems in the area will form different data sets, but the data type is all passenger flow.

[0055] Hyperparameters refer to the model parameters that are set manually before starting to train a prediction model.

[0056] Based on the historical data of air passenger flow, the central server predicts the air passenger flow of several OD markets simultaneously.

[0057] Preferably, the management personnel input the three-letter codes of the departure airports, the three-letter codes of the destination airports, and the monthly passenger flow corresponding to each OD market through the client server;

[0058] "Departure airport" refers to all or part of the airports within the selected "regional multi-airport system", such as all the airports in the Yangtze River Delta region.

[0059] "Destination airport" refers to an airport outside the selected "regional multi-airport system", and only one can be selected, such as Beijing Capital International Airport.

[0060] The client server automatically arranges and generates the passenger flow time series data of each OD market according to time, and the client server transmits the passenger flow time series data to the database module for storage.

[0061] Preferably, preprocess the air passenger flow data, including wavelet threshold denoising and data normalization processing, where wavelet threshold denoising uses the method of soft threshold two-layer decomposition.

[0062] Preferably, construct an OD passenger flow grid map and an external influence factor feature vector, specifically including:

[0063] Step A1: Define O = {o 1 , o 2 , …, o N} to represent the set of departure airports, N is the total number of departure airports; d represents the destination airport; use <o, d> to represent an OD market, o ∈ O;

[0064] The prediction model server maps the N departure airports to the I×J grid map R according to the longitude, latitude and relative geographical location distribution of the departure airports within each regional multi-airport system. The grid (i, j) represents any grid in the grid map R, i represents the row number of the grid, j represents the column number of the grid, and the grid (i, j) represents the passenger flow from the departure airport o at this location to the destination airport d outside the system in the t-th month Use a two-dimensional tensor X t ∈R I×J to represent the OD passenger flow grid map from the departure airport within the system to the same destination airport outside the system in the t-th month, where,

[0065] For example Figure 3As shown, each grid is pre-corresponded to a departure airport o, and the grid (i, j) corresponds to the passenger flow from the departure airport o represented by the grid in a certain month to the destination airport d; the grid map is equivalent to a flow matrix and serves as the input of the model.

[0066] In this embodiment, according to the longitude, latitude and relative geographical location distribution of each departure airport in the multi-airport system of the Yangtze River Delta region, 16 departure airports are mapped to a 4×4 grid map, and the mapping positions of each departure airport in the grid map are as Figure 3 shown in the lower left corner.

[0067] Step A2: Define the external influencing factors including two features: month attribute and whether it includes holidays, and process the external influencing factors into a 0-1 vector, that is, use 0 or 1 to represent the two features of the predicted month attribute and whether it includes holidays. The first 12 bits correspond to January to December, and the 13th bit represents whether it includes holidays. For example, if predicting the passenger flow in October 2018, it is converted to [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1];

[0068] Preferably, the prediction model server uses the functional module in the Keras neural network library of Python to build a prediction model for the OD market air passenger flow based on the spatio-temporal convolutional network, as Figure 2 shown, which is the framework of the prediction model;

[0069] Through the known m historical observation values X M ={X t |t = 1, 2, …, m}, predict the air passenger flow X K ={X t |t = m + 1, m + 2, …, m + k} of several OD markets in the next k months;

[0070] By changing two parts of hyperparameters, namely the network structure and data characteristics, to determine the most suitable hyperparameter settings for the prediction model, use the Adam optimizer to train and optimize the prediction model, and adopt an early stopping strategy during the training process to avoid model overfitting.

[0071] Early stopping strategy: 1) Divide the original training data set into a training set and a validation set; 2) The model is trained on the training set and the error of the model on the validation set is calculated in each epoch; 3) Stop training when the error of the model on the validation set is worse than the previous training result; 4) Use the parameters in the previous iteration result as the final parameters of the model.

[0072] That is, by weighing the training period and generalization error to achieve the best model training effect and avoid model overfitting.

[0073] In this embodiment, the fixed learning rate is 0.002, the number of samples processed per batch is 3, and the number of training epochs is 100. Taking the OD passenger flow data from 16 major airports in the Yangtze River Delta to Guangzhou Baiyun International Airport as an example, the number of convolutional layers in the network structure is set to ∈{1, 2, 3}, the convolutional kernel size is ∈{(2, 2), (3, 3)}, the number of convolutional kernels is ∈{16, 32, 64, 128}, and the sample lengths of the trend segments and periodic segments selected in the data characteristics are ∈{2, 3, 4}; the data from January 2010 to December 2017 is selected as the training data in the embodiment, and the data from January 2018 to December 2018 is used as the test data, with a prediction time interval of 1 month.

[0074] Taking the absolute error Δ, MAE (Mean Absolute Error), and RMSE (Root Mean Square Error) as the evaluation criteria, the comparison of aviation passenger flow predictions by different algorithm models is shown in Table 1, and the passenger flow fitting results with other models in different OD markets are as Figure 4 shown, and the passenger flow fitting results with other models at different prediction intervals are as Figure 5 shown:

[0075]

[0076] Table 1

[0077] Based on the historical data of aviation passenger flow, using the prediction model with the determined hyperparameter settings, predict the aviation passenger flow in multiple OD markets for a period of time.

[0078] Preferably, the specific steps for the central server to predict the aviation passenger flow in several OD markets include:

[0079] Step B1: Extract the segments of the OD passenger flow grid map. According to the dependence of the passenger flow on the time dimension, two time segments are extracted at different time intervals for the prediction time points respectively: namely, the trend grid picture segment X Tre and the periodic grid picture segment X Per , and the specific extraction form is as follows:

[0080]

[0081]

[0082] where l tre and l per , as adjustable data characteristic hyperparameters in the prediction model, respectively represent the sample length of the selected trend segment and the sample length of the selected periodic segment;

[0083] Step B2: Construct a spatio-temporal convolutional network. Based on the extracted grid image segments at two different time intervals, construct spatio-temporal convolutional network branches with the same structure to capture the spatio-temporal characteristics of OD passenger flow. The spatio-temporal convolutional network branches are based on S + 1 (S ≥ 1) convolutional layers;

[0084] Taking the spatio-temporal convolutional network branch of the trend part as an example, use to represent the trend grid image segment. Through the first convolutional layer C 1 transform (X Tre ) (0) into a new tensor (X Tre ) (1) , and the transformation formula is as follows:

[0085]

[0086] where and are the learning parameters of the first convolutional layer. After the operation of convolutional layer C 1 , continue to add (S - 1) convolutional layers according to the above formula. After the S-th convolutional layer, pass through a convolutional layer C S+1 that only contains one convolutional kernel, and finally obtain the output result (X Tre ) (S+1) of the trend part; Similarly, use the same operation above to construct the spatio-temporal convolutional network branch of the periodic part and obtain the output result (X Per ) (S+1) ;

[0087] Step B3: Construct an external influence factor network. The external influence factors considered by the prediction model include two aspects: month attribute and whether it includes holidays. Based on the t-th prediction time point, obtain the corresponding external influence factor feature vector E t , and use a two-layer fully connected neural network branch as the external influence factor network. The first layer can be regarded as an embedding layer, mainly to quantitatively add external factors to the prediction model. The second layer is to map the features obtained in the first layer into a high-dimensional tensor, and its size should be the same as X t so as to fuse with the output result of the spatio-temporal convolutional network to obtain the output result X Ext ;

[0088] Step B4: Fuse to obtain the prediction result. Assign different weight matrices to the output results (X Tre ) (S+1) and (X Per ) (S+1) of the spatio-temporal convolutional network in the form of learning parameters, and perform aggregation to obtain the weighted output result. The calculation formula is as follows:

[0089] XCon = W Tre *(X Tre ) (S+1) + W Per *(X Per ) (S+1)

[0090] where * represents the Hadamard product; W Tre and W Per respectively represent the weights of the trend and periodic parts, that is, the influence degrees of the outputs of the two parts on the final prediction result. Further, the calculation result X Con and the output result X Ext of the external influence factor network are mapped to the interval [-1, 1] through the tanh function to obtain the final prediction result X K , and the calculation formula is as follows:

[0091] X K = tanh(X Con + X Ext )

[0092] Taking the mean square error between the predicted value matrix X K and the true value matrix as the objective to train the prediction model:

[0093]

[0094] where θ represents all the learning parameters of the model.

[0095] A method for predicting the air passenger flow of OD markets based on a spatio-temporal convolutional network according to the present invention solves the technical problem of using a spatio-temporal convolutional network to predict the air passenger flow and realizing the simultaneous prediction of the passenger flows of multiple OD markets from multiple departure airports in the same region to the same destination airport. The present invention first applies the spatio-temporal convolutional network to the field of air passenger flow prediction to realize the simultaneous prediction of the passenger flows of multiple OD markets from multiple departure airports in the same region to the same destination airport, providing a new idea for air passenger flow prediction; the present invention regards the convolutional kernels in the convolutional network as the medium for capturing the spatial correlation of OD passenger flows, and multiple convolutional kernels in a convolutional layer can accurately capture the complex potential feature relationships of OD passenger flows in the spatial dimension, improving the prediction effect; the present invention comprehensively considers the time dependence, spatial correlation of OD passenger flows and the role of external influence factors, takes RMSE as the evaluation criterion of the model, and takes the OD passenger flow data from 16 major airports in the Yangtze River Delta to Guangzhou Baiyun International Airport as an example. The prediction model is compared with ARIMA (Autoregressive Integrated Moving Average Model), SVR (Support Vector Regression), Elman neural network and LSTM (Long Short-Term Memory Network), and the fitting effect is better.

[0096] In the present invention, any process or method description represented in a flowchart or otherwise described herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of the present invention includes additional implementations, where functions may be executed not in the order shown or discussed, including in substantially simultaneous manners according to the involved functions or in reverse order, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0097] The logic and / or steps represented in a flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing a logical function, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing as appropriate, and then storing it in a computer memory.

[0098] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0099] Those of ordinary skill in the art can understand that all or part of the steps carried out in implementing the above-described embodiment methods can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0100] In addition, in each of the embodiments of the present invention, the functional units can be integrated into a processing module, or each unit can exist physically alone, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0101] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting the air passenger flow in the OD market based on a spatio-temporal convolutional network, characterized in that: It includes: Select a multi-airport system in a certain area through the client server, and count, organize and store the air passenger flow data from each departure airport to the destination airport in the multi-airport system; The prediction model server obtains the air passenger flow data, preprocesses the air passenger flow data, constructs an OD passenger flow grid map and an external influence factor feature vector, builds a prediction model for the air passenger flow in the OD market based on the spatio-temporal convolutional network, and forms a corresponding training set for the prediction model; The OD market refers to the air passenger transportation market from the Origin Airport to the Destination Airport; By adjusting the data characteristics and the network structure of the prediction model, determine the optimal hyperparameter settings of the prediction model under different data sets; the hyperparameters include the hyperparameters of the network structure and the hyperparameters of the data characteristics. The hyperparameters of the network structure include the number of convolutional layers, the size of the convolutional kernel, and the number of convolutional kernels. The hyperparameters of the data characteristics include the sample length of the trend segment selection and the sample length of the periodic segment selection; The central server predicts the air passenger flow of several OD markets according to the historical air passenger flow data; The prediction model server uses the functional module in the Keras neural network library of Python to build a prediction model for the air passenger flow of OD markets based on a spatio-temporal convolutional network. Through the known m historical observation values X M ={X t | t = 1, 2, …, m}, it predicts the air passenger flow X K ={X t | t = m + 1, m + 2, …, m + k}; By changing the two parts of hyperparameters, namely the network structure and the data characteristics, to determine the most suitable hyperparameter settings of the prediction model, use the Adam optimizer to train and optimize the prediction model, and adopt an early stopping strategy during the training process to avoid model overfitting; The specific steps for the central server to predict the air passenger flow of several OD markets include: Step B1: Extract segments of the OD passenger flow grid map. According to the dependence of passenger flow on the time dimension, extract two time segments at different time intervals for each prediction time point: namely, the trend grid map segment X Tre and the periodic grid map segment X Per , and the specific extraction form is as follows: Among them, l tre and l per As adjustable data characteristic hyperparameters in the prediction model, they respectively represent the sample length of the selected trend segments and the sample length of the selected periodic segments; Step B2: Construct a spatio-temporal convolutional network. Based on the extracted grid picture segments with two different time intervals, respectively construct spatio-temporal convolutional network branches with the same structure to capture the spatio-temporal characteristics of the OD passenger flow. The spatio-temporal convolutional network branches are based on S + 1 convolutional layers, S≥1; Taking the spatio-temporal convolutional network branch of the trend part as an example, use to represent the trend grid picture segment. Through the first convolutional layer C 1 transform (X Tre ) (0) into a new tensor (X Tre ) (1) . The conversion formula is as follows: Among them, and are the learning parameters of the first convolutional layer. After the convolutional layer C 1 operation, (S - 1) convolutional layers are added successively according to the above formula. After the S-th convolutional layer, it passes through a convolutional layer C S+1 with only one convolutional kernel, and finally the output result (X Tre ) (S+1) of the trend part is obtained; similarly, the spatio-temporal convolutional network branch of the periodic part is constructed using the same operation as above to obtain the output result (X Per ) (S+1) ; Step B3: Construct an external influencing factor network. The external influencing factors considered by the prediction model include two aspects: month attribute and whether it includes holidays. Based on the t-th prediction time point, the corresponding external influencing factor feature vector E is obtained. t , and use a two-layer fully connected neural network branch as the external influencing factor network. The first layer can be regarded as an embedding layer, which quantitatively adds external factors to the prediction model. The second layer maps the features obtained from the first layer into a high-dimensional tensor, and its size should be the same as that of X t to facilitate the fusion with the output result of the spatio-temporal convolutional network, and obtain the output result X Ext ; Step B4: Fuse to obtain the prediction result, and assign the output result (X Tre ) (S+1) and (X Per ) (S+1) with different weight matrices, and perform aggregation to obtain the weighted output result. The calculation formula is as follows: X Con = W Tre *(X Tre ) (S+1) + W Per *(X Per ) (S+1) ; where * represents the Hadamard product; W Tre and W Per respectively represent the weights of the trend and periodic parts, that is, the influence degrees of the outputs of the two parts on the final prediction result. Further, the calculation result X Con and the output result X Ext of the external influence factor network are mapped to the interval [-1, 1] to obtain the final prediction result X K , and the calculation formula is as follows: X K =tanh(X Con +X Ext ); To minimize the predicted value matrix X K and the true value matrix The prediction model is trained with the mean square error between them as the objective: Among them, θ represents all the learning parameters of the model.

2. A method for predicting the air passenger flow in the OD market based on a spatio-temporal convolutional network according to claim 1, characterized in that: Managers input the three-letter codes of the departure airport, the three-letter codes of the destination airport, and the monthly passenger flow corresponding to each OD market through the client server; The client server automatically arranges and generates the passenger flow time series data of each OD market according to the time, and the client server transmits the passenger flow time series data to the database module for storage.

3. A method for predicting the air passenger flow in the OD market based on a spatio-temporal convolutional network according to claim 1, characterized in that: Preprocess the air passenger flow data, including wavelet threshold denoising and data normalization processing. Among them, wavelet threshold denoising adopts the method of soft threshold two-layer decomposition.

4. A method for predicting the air passenger flow in the OD market based on a spatio-temporal convolutional network according to claim 1, characterized in that: Construct an OD passenger flow grid map and an external influence factor feature vector, specifically including: Step A1: Define \(O = \{o 1 , o 2 , \ldots, o N \}\) to represent the set of departure airports, where \(N\) is the total number of departure airports; \(d\) represents the destination airport; use \(\langle o, d\rangle\) to represent an OD market, where \(o\in O\); The prediction model server maps N departure airports to a grid graph R of I×J according to the longitude, latitude, and relative geographical location distribution of the departure airports in the multi-airport system in each region. The grid (i, j) represents any grid in the grid graph R, where i represents the row number of the grid and j represents the column number of the grid. The grid (i, j) represents the passenger flow from the departure airport o to the destination airport d outside the system corresponding to this location in the t-th month. Use a two-dimensional tensor X t ∈R I×J to represent the OD passenger flow grid graph from the departure airport within the system to the same destination airport outside the system in the t-th month. Among them, Step A2: Define the external influencing factors to include two features, namely the month attribute and whether it includes holidays. Process the external influencing factors into a 0-1 vector, that is, use 0 or 1 to represent the two features of the predicted month attribute and whether it includes holidays. The first 12 bits correspond to representing January to December, and the 13th bit represents whether it includes holidays.

Images