Airport passenger flow multi-factor fusion prediction method based on causal inference

Abstract

The invention relates to an airport passenger flow multi-factor fusion prediction method based on causal inference, belongs to the technical field of airport passenger flow prediction, and solves the problem that airport resource allocation is difficult due to change of passenger volume in the prior art. Performing Granger causality inference test on the airport passenger flow volume; obtaining a multi-factor Granger causality inference result; dividing the time sequence of the input characteristics for predicting the airport passenger flow volume; establishing an attention mechanism in a time period and obtaining local time sequence information; establishing an attention mechanism between time periods and obtaining global time sequence information; fusing the obtained local time sequence information and global time sequence information with a Granger causality inference result to obtain a predicted value of the airport passenger flow volume; calculating an error between the predicted value and the true value; a trained airport passenger flow volume prediction model is obtained; and inputting the measured real-time data to the trained airport passenger flow volume prediction model to obtain airport passenger flow volume prediction information.

Description

Technical Field

[0001] The present invention relates to the technical field of airport passenger flow prediction, and particularly to a multi-factor fusion prediction method for airport passenger flow based on causal inference. Background Art

[0002] With the increase in travel demand, the civil aviation passenger volume has increased, which has exacerbated the imbalance between airport resources and airport passenger flow, bringing great pressure to the internal resource allocation of the airport and the transportation system around the airport. Predicting the passenger flow in the airport in advance is one of the most effective methods to achieve the balance between airport resources and passenger demand. Through the prediction of passenger flow, airport management can better schedule traffic resources, allocate and arrange resources in advance to cope with emergencies and passenger flow pressure during peak hours. For passengers, the waiting time is reduced and the travel experience is improved; for the airport, the operating cost is reduced and the resource utilization rate is increased. Therefore, predicting the airport passenger flow in advance has practical significance.

[0003] If the predicted value of the airport passenger flow is too low, it will cause passengers to stay in the airport, leading to various problems such as public opinion and paralysis of the surrounding traffic. If the predicted value of the airport passenger flow is too high, it will result in waste of resources and economic losses, such as idle equipment resources and low investment returns. Therefore, it is necessary to accurately predict the airport passenger flow. At the same time, predicting the airport passenger flow itself also faces a series of challenges. The airport passenger flow is affected by multiple related factors, such as flight delays, meteorological conditions, sudden control, etc., and the influencing factors are complex. Moreover, these influencing factors interact and are interrelated, making the airport passenger flow data prone to sudden fluctuations in a short period of time. These reasons make the airport passenger flow prediction problem complex and difficult.

[0004] However, in the existing airport passenger flow prediction methods, the characteristic of the coupling of multiple influencing factors of passenger flow is ignored. In order to make the airport passenger flow prediction method show better performance and better meet the actual needs, an improved prediction method is needed in this technical field, which can fully master the multi-factor coupling intensity of passenger flow data and thus improve the prediction accuracy. Summary of the Invention

[0005] In view of the above problems, the present invention provides a multi-factor fusion prediction method for airport passenger flow based on causal inference.

[0006] According to an embodiment of the present invention, a multi-factor fusion prediction method for airport passenger flow based on causal inference is provided, including the following steps:

[0007] A multi-factor fusion prediction method for airport passenger flow based on causal inference, characterized in that it includes the following steps:

[0008] Step S0: Obtain input data information, including passenger flow data, flight schedules, carbon dioxide concentration, PM2.5 concentration, temperature, and humidity information obtained through databases and sensors. All information in the input data information is time series data;

[0009] Step S1: Based on the obtained input data information, conduct a Granger causality inference test on the airport passenger flow, construct a multi-factor causal association, and obtain the input features for predicting the airport passenger flow;

[0010] Step S2: Determine whether Granger tests and analyses have been performed on the flight schedule, carbon dioxide concentration, PM2.5 concentration, temperature, and humidity information in the input data information obtained in Step S0. If so, use all the obtained input features as the multi-factor Granger causality inference results and execute the next step. Otherwise, return to Step S1 to conduct a Granger causality inference test on the remaining influencing factors in the input data information;

[0011] Step S3: Divide the time series of the input features to obtain a new time series;

[0012] Step S4: Extract features within each time period of the obtained new time series, establish an attention mechanism within the time period, and obtain local time series information;

[0013] Step S5: Extract features between time periods of the obtained new time series, establish an attention mechanism between time periods, and obtain global time series information to capture global correlation;

[0014] Step S6: Integrate the obtained local time series information and global time series information with the Granger causality inference results of Step S2 to achieve a weighted effect and obtain the predicted value of the airport passenger flow;

[0015] Step S7: Calculate the error between the obtained predicted value and the true value of the passenger flow data obtained in Step S0, and use the gradient descent method to update all weight parameters in the neural network;

[0016] Step S8: Compare the obtained error with a predetermined threshold, and compare the number of iterations with the maximum number of iterations. When either of the two conditions that the error is less than the predetermined threshold and the number of iterations reaches the maximum number of iterations is satisfied, the training of the model ends, and a trained airport passenger flow prediction model is obtained. Otherwise, return to Step S3 to re-divide the time series at different scales S;

[0017] Step S9: Input the measured real-time data into the obtained trained airport passenger flow prediction model to obtain airport passenger flow prediction information for airport resource scheduling and allocation.

[0018] Optionally, step S1 specifically includes the following steps:

[0019] Step S1.1: Select features from the obtained input data information including flight schedules, carbon dioxide concentrations, PM2.5 concentrations, temperatures, and humidity information, and construct a multivariate regression model for airport passenger flow as:

[0020]

[0021] where, Y t represents the airport passenger flow at time t, represents the value of the feature at time t-q, β p , η t , γ q represent regression parameters and are obtained by data training, p represents the variable at time p, and q represents the variable at time q;

[0022] Step S1.2: Through adding the past values of influencing factors, conduct joint regression analysis and construct a univariate regression model for airport passenger flow as:

[0023] Y t =α 0 +α 1 Y t-1 +…+α p Y t-p +ε t

[0024] where, α p , ε t represent regression parameters and are obtained by data training;

[0025] Step S1.3: Conduct an F-test on the established multivariate regression model and univariate regression model for airport passenger flow to determine whether the past values of feature X t can significantly improve the prediction ability of the airport passenger flow Y t at time t.

[0026] Optionally, step S1.3 specifically includes the following steps:

[0027] Step S1.3.1: Conduct an F-test on the established multivariate regression model and univariate regression model for airport passenger flow to obtain the test value:

[0028]

[0029] where, RSS r is the sum of squared residuals of the univariate autoregressive model, RSS ur is the sum of squared residuals of the multivariate regression model, q is the number of lag terms introduced, n is the sample size, and k is the number of explanatory variables in the model;

[0030] Step S1.3.2: Analyze the result obtained in Step S1.3.1. If the obtained test value F is greater than the preset critical value, it indicates that the current feature X t Granger causes Y t , and use the current feature as the input feature for predicting the airport passenger flow. Otherwise, it means that the feature has no effect on the prediction, and it is deleted.

[0031] Optionally, Step S3 specifically includes:

[0032] Take the time series data X ∈ R of the obtained input features H×d as the input, where H represents the length of the time series, d represents the dimension of the feature, and R H×d represents the set of time series data. Use a scale of length S to divide the data and obtain a new set of time series:

[0033] (X 1 ,..., X i ..., X P ) ∈ R P×S×d

[0034] where X i represents the input data of the i-th time period and X i ∈ R S×d , P represents the number of time periods obtained after dividing the time series data, and each time period is used as a new unit.

[0035] Optionally, Step S4 specifically includes the following steps:

[0036] Step S4.1: For the input data X 1 ,..., X i ..., X P ) of the i-th time period in the new time series (X i ∈ R S×d , encode it on the feature dimension d to obtain:

[0037]

[0038] where represents the new input data within the i-th time period after encoding on the feature dimension d, and d m represents encoding the feature dimension d to d m ;

[0039] Step S4.2: Perform a linear transformation on to obtain the variable in the attention mechanism:

[0040]

[0041] Among them, represents the key vector within a time period, V i intra represents the value vector within a time period;

[0042] Step S4.3: Use trainable variables to merge the temporal information before and after within a time period, and calculate the input data X for the i-th time period i with the attention information within:

[0043]

[0044] Step S4.4: Execute Steps S4.1 - S4.3 for the input data of each time period in the newly partitioned time series, and use the obtained attention information as local temporal information.

[0045] Optionally, Step S5 specifically includes the following steps:

[0046] Step S5.1: For the newly partitioned time series X ∈ R P×S×d , perform encoding in the feature dimension d to d m ′, and the mapped data is:

[0047]

[0048] Among them, d m ′ = S × d;

[0049] Step S5.2: Perform self-attention operations between time periods, thereby modeling the airport passenger flow data based on the correlation between time periods, and obtaining the variables between time periods through a linear mapping of X inter :

[0050]

[0051] Among them, Q inter represents the query vector between time periods, K inter represents the key vector between time periods, and V inter represents the value vector between time periods;

[0052] Step S5.3: Obtain the attention information between time periods:

[0053]

[0054] Step S5.4: Execute Steps S5.1 - 5.3 for all time periods in the newly partitioned time series, and the obtained attention information between time periods is the global temporal information.

[0055] The present invention provides a multi-factor fusion prediction method for airport passenger flow based on causal inference, including: (1) Causal coupling strength inference: The airport passenger flow is affected by multiple factors. In this patent, flight schedule information table, CO2 concentration, PM2.5 concentration, temperature, humidity, and historical airport passenger flow data are considered. All data are time series data. Among them, the flight schedule information table refers to the number of airplanes arriving at the airport per unit time. The Granger causality inference test is used to construct the causal relationship of multiple factors, and the multi-factor coupling strength information is obtained. (2) Temporal feature block extraction: The time series is segmented into sub-sequence-level patches. The patches are used as the input of the Transformer model, and the self-attention mechanism is used to extract the information within the patches to obtain local temporal information; then the self-attention mechanism is used to extract features between the patches to obtain global temporal information. (3) Information fusion: The obtained temporal information is fused with the Granger causality information, and a prediction result containing causal coupling strength inference information is output. In the case where the passenger flow is affected by many static and dynamic factors and the relationships between the factors are complex, the inference of the coupling strength of airport passenger flow under multiple factors is realized, and an accurate prediction result of airport passenger flow is obtained.

[0056] Compared with the prior art, the multi-factor fusion prediction method for airport passenger flow based on causal inference provided by the embodiments of the present invention has at least the following beneficial effects:

[0057] (1) Considering the characteristics of multi-factor coupling of airport passenger flow, through Granger causality analysis, the coupling strength of multiple factors is accurately extracted, and an accurate predicted value of airport passenger flow is obtained.

[0058] (2) By means of time series division, taking "time blocks" as new units, the local fluctuations of airport passenger flow are extracted, and the global trend of the data is captured through feature extraction between time blocks.

[0059] (3) In practical applications, it can effectively predict the passenger flow in the airport, so as to achieve the balance between airport resources and passenger transport demand.

[0060] (4) By providing an accurate predicted value of airport passenger flow, it is beneficial to the reasonable and efficient allocation of resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. By referring to the drawings, the features and advantages of the present invention can be more clearly understood. The drawings are schematic and should not be construed as any limitation to the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0062] Figure 1 Schematic diagram of the principle of the multi-factor fusion prediction method for airport passenger flow based on causal inference provided for the embodiments of the present invention.

[0063] Figure 2 Flowchart of the multi-factor fusion prediction method for airport passenger flow based on causal inference provided for the embodiments of the present invention. Specific embodiments

[0064] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0065] In the following description, many specific details are set forth in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0066] The following provides a detailed description of the multi-factor fusion prediction method for airport passenger flow based on causal inference according to the embodiments of the present invention with reference to the accompanying drawings.

[0067] As Figure 1 and Figure 2 shown, the multi-factor fusion prediction method for airport passenger flow based on causal inference according to the embodiments of the present invention includes the following steps.

[0068] Step S0: Obtain input data information, including obtaining passenger flow data, flight schedules, carbon dioxide concentration, PM2.5 concentration, temperature, humidity and other information through databases and sensors. All data in the input data information are time series data. Optionally, the input data information can be historical data information for model construction and training.

[0069] Step S1: Based on the obtained input data information, perform Granger causality inference test on the airport passenger flow to construct a multi-factor causal association. The Granger causality inference test judges the causal relationship through a regression model, including establishing a multivariate regression model and a univariate regression model. Step S1 specifically includes the following steps.

[0070] Step S1.1: Select features from the flight schedule, carbon dioxide concentration, PM2.5 concentration, temperature and humidity information in the obtained input data information to construct a multivariate regression model of airport passenger flow as:

[0071]

[0072] where Y tDenote the passenger flow of the airport at time t as Y t-p Denote the passenger flow of the airport at time t - p Denote the value of the feature at time t - q as β p , η t , γ q Denote the regression parameters obtained by training with data. p represents the variable at time p, and q represents the variable at time q

[0073] Step S1.2: By adding the past values of influencing factors, perform joint regression analysis to construct a univariate regression model for the passenger flow of the airport as follows

[0074] Y t = α 0 + α 1 Y t-1 +…+ α p Y t-p + ε t

[0075] where α p , ε t Denote the regression parameters obtained by training with data

[0076] Step S1.3: Perform an F - test on the established multivariate regression model and univariate regression model of the airport passenger flow to determine whether the past values of feature X t can significantly improve the prediction ability for Y t . This step S1.3 specifically includes the following steps

[0077] Step S1.3.1: Perform an F - test on the established multivariate regression model and univariate regression model of the airport passenger flow to obtain the test value

[0078]

[0079] where RSS r is the sum of squared residuals of the univariate autoregressive model, RSS ur is the sum of squared residuals of the multivariate regression model, q is the number of lag terms introduced, n is the sample size, and k is the number of explanatory variables in the model

[0080] Step S1.3.2: Analyze the result obtained in Step S1.3.1. If the obtained test value F is greater than the preset critical value, it means that the current feature X t Granger - causes Y t . Then use the current feature as the input feature for predicting the airport passenger flow. Otherwise, it means that the current feature has no effect on the prediction and it will be deleted

[0081] Step S2: Determine whether Granger tests and analyses have been performed on all features such as flight schedules, carbon dioxide concentrations, PM2.5 concentrations, temperatures, and humidity information in the input data information obtained in Step S0. If so, use all the input features for predicting airport passenger flow as the results of multi-factor Granger causal inference, and proceed to the next step. Otherwise, return to Step S1 to perform Granger causal inference tests on the remaining influencing factors in the input data information.

[0082] Step S3: Divide the time series to obtain a new time series. Use the time series data X ∈ R of the input features for predicting airport passenger flow obtained as the input, where H represents the length of the time series, d represents the dimension of the feature, and R H×d represents the set of time series data. Use a scale of length S for data division to obtain a new set of time series: H ×d (X

[0083] ,..., X 1 ,..., X i ..., X P ) ∈ R P×S×d

[0084] where X i represents the input data for the i-th time period, and X i ∈ R S×d , P represents the number of time periods obtained after dividing the time series data, Each time period is used as a new unit, called a Patch. The self-attention within each time period needs to be established within each Patch to obtain local temporal information.

[0085] Step S4: Extract features within each time period of the new time series after division, establish an attention mechanism within the time period, and obtain local temporal information. This Step S4 specifically includes the following steps.

[0086] Step S4.1: For the input data X 1 ,..., X i ..., X P ) of the i-th time period in the new time series after division, first encode it in the feature dimension d to obtain: i

[0087]

[0088] where represents the new input data within the i-th time period after encoding in the feature dimension d, and d m represents encoding the feature dimension d to d m .

[0089] Step S4.2. Then, perform a linear transformation on to obtain in the attention mechanism:

[0090]

[0091] where represents the key vector within the time period, and V i intra represents the value vector within the time period.

[0092] Step S4.3. Use the trainable variable to merge the temporal information before and after within the time period, and then calculate the attention information within the i-th time period X i :

[0093]

[0094] Step S4.4. Execute Steps S4.1 - 4.3 for the input data of each time period in the newly partitioned time series, and use the obtained attention information as the local temporal information.

[0095] Step S5. Extract features between the time periods in the newly partitioned time series, establish an attention mechanism between the time periods, and obtain the global temporal information to capture the global correlation. This Step S5 specifically includes the following steps.

[0096] Step S5.1. For the newly partitioned time series X ∈ R P×S×d , first encode it from the feature dimension d to d m ′, and the mapped data is:

[0097]

[0098] where d m ′ = S × d.

[0099] Step S5.2. Perform a self-attention operation between the time periods, so as to model the airport passenger flow data based on the correlation between the time periods. This includes obtaining through the linear mapping of X inter :

[0100]

[0101] Thus, three variables are obtained, Q inter , K inter , V interThey respectively represent the query vector, key vector, and value vector between time periods. The query vector is used to measure the correlation with each key vector. The key vector is used to construct the calculation basis of the attention score. The value vector is used to perform weighted summation according to the attention score to generate the final attention output.

[0102] Step S5.3: Obtain the attention information between time periods:

[0103]

[0104] Step S5.4: Execute Steps S5.1 - 5.3 for all time periods of the newly partitioned time series. The obtained attention information between time periods is the global time series information.

[0105] Step S6: After the above operations, the local time series information including the self - attention information within time periods and the global time series information including the self - attention information between time periods are obtained. They are fused with the Granger causality inference results obtained previously to achieve a weighted effect, and the predicted value of the airport passenger flow is obtained.

[0106] In this step, the model captures the local details and global correlations of the time series data, and fuses the intensity information of causal coupling to obtain the predicted value of the airport passenger flow.

[0107] Step S7: Calculate the error between the obtained predicted value and the true value of the airport passenger flow data obtained in Step S0, and use the gradient descent method to update all weight parameters in the neural network.

[0108] Step S8: Compare the obtained error with a predetermined threshold, and compare the number of iterations with the maximum number of iterations. When either of the two conditions that the error is less than the predetermined threshold and the number of iterations reaches the maximum number of iterations is satisfied, the training of the model ends, and the trained airport passenger flow prediction model is obtained. Otherwise, return to Step S3 to re - partition the time series at a different scale S.

[0109] Step S9: Input the measured real - time data into the obtained trained airport passenger flow prediction model to obtain the airport passenger flow prediction information for airport resource scheduling and allocation. This real - time data is other data and information obtained in real - time at the airport, and may include passenger flow data, flight schedules, carbon dioxide concentration, PM2.5 concentration, temperature, humidity, and other information.

[0110] In airports of various scales, the above-mentioned multi-factor fusion prediction method for airport passenger flow based on causal inference provided by the embodiments of the present invention can be used to predict the airport passenger flow. According to the historical passenger flow data, flight schedules, carbon dioxide concentration, PM2.5 concentration, temperature, and humidity information, after model prediction, the airport passenger flow value at a future time can be obtained, providing advance data support for the airport to reallocate resources and conduct traffic scheduling.

[0111] Any combination of the above all optional technical solutions can form an optional embodiment of the present application, which will not be elaborated here one by one.

[0112] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0113] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A multi-factor fusion prediction method for airport passenger flow based on causal inference, characterized by: The following steps are involved: Step S0, obtaining input data information, including obtaining passenger flow data, flight schedule, carbon dioxide concentration, PM2.5 concentration, temperature and humidity information through databases and sensors, and all information in the input data information is time series data; Step S1: Based on the input data information obtained, Granger causality inference test is performed on the airport passenger flow, causal association of multiple factors is constructed, and input features for predicting the airport passenger flow are obtained; Step S2, determining whether the flight schedule, carbon dioxide concentration, PM2.5 concentration, temperature and humidity information in the input data information obtained in step S0 have been Granger tested and analyzed. If so, all the input features obtained are used as the Granger causal inference results of multiple factors, and the next step is executed; otherwise, returning to step S1, Granger causal inference test is performed on the remaining influencing factors in the input data information; Step S3, dividing the time series of the input features to obtain a new time series; Step S4: extract features in each time period of the obtained new time series, establish an attention mechanism within the time period and obtain local time series information; Step S5, extracting features between time periods in the obtained new time series, establishing an attention mechanism between time periods and obtaining global time series information to capture global correlation; Step S6: Fusing the obtained local time series information and global time series information with the Granger causal inference result of step S2 to achieve a weighted effect, thereby obtaining a predicted value of airport passenger flow; Step S7, calculating the error between the predicted value obtained and the true value of the passenger flow data obtained in step S0, and updating all weight parameters in the neural network using the gradient descent method; Step S8, comparing the obtained error with a predetermined threshold, and comparing the number of iterations with a maximum number of iterations. When one of the two conditions, that the error is less than the predetermined threshold and the number of iterations reaches the maximum number of iterations, is met, the model training ends and a trained airport passenger flow prediction model is obtained. Otherwise, the method returns to step S3 and re-divides the time series with different scales S. Step S9: input the measured real-time data into the trained airport passenger flow prediction model to obtain airport passenger flow prediction information for airport resource scheduling and allocation.

2. The airport passenger flow multi-factor fusion prediction method based on causal inference according to claim 1 is characterized in that: The step S1 specifically includes the following steps: Step S1.1, select features from the flight schedule, carbon dioxide concentration, PM2.5 concentration, temperature and humidity information in the obtained input data information, and construct a multivariate regression model for airport passenger flow: Among them, Y t represents the airport passenger flow at time t, represents the value of the feature at time tq, β p ,η t ,γ q represents the regression parameter and is obtained by data training, p represents the variable at time p, and q represents the variable at time q; Step S1.2: By adding the past values ​​of the influencing factors, a joint regression analysis is performed to construct a univariate regression model of airport passenger flow: Y t =α0+α1Y t-1 +…+a p Y t-p +e t Among them, α p ,ε t represents the regression parameters and is obtained by data training; Step S1.3: Perform an F test on the established multivariate regression model and univariate regression model of airport passenger flow to determine whether the feature X t Can the past value of significantly improve the airport passenger flow Y at time t? t predictive ability.

3. The airport passenger flow multi-factor fusion prediction method based on causal inference according to claim 2 is characterized in that: The step S1.3 specifically includes the following steps: Step S1.3.1: Perform an F test on the established multivariate regression model and univariate regression model of airport passenger flow to obtain the test value: Among them, RSS r is the residual sum of squares of the univariate autoregression model, RSS ur is the residual sum of squares of the multivariate regression model, q is the number of lag terms introduced, n is the sample size, and k is the number of explanatory variables in the model; Step S1.3.2: Analyze the result obtained in step S1.3.

1. If the obtained test value F is greater than the preset critical value, it means that the current feature X t Granger leads to Y t , the current feature is used as the input feature for predicting airport passenger flow. Otherwise, it means that the feature has no effect on the prediction and it is deleted.

4. The airport passenger flow multi-factor fusion prediction method based on causal inference according to claim 3 is characterized in that: Step S3 specifically includes: The time series data X∈R of the input features obtained H×d As input, H represents the length of the time series, d represents the dimension of the feature, and R H×d Represents a set of time series data, and uses a scale of length S to divide the data to obtain a new set of time series: (X1,...,X i ...,X P )∈R P×S×d Among them, X i represents the input data of the i-th time period and X i ∈R S×d , P represents the number of time periods obtained after dividing the time series data, Each time period is regarded as a new unit.

5. The airport passenger flow multi-factor fusion prediction method based on causal inference according to claim 4 is characterized in that: Step S4 specifically includes the following steps: Step S4.1: For the new time series (X1,...,X i ...,X P ) in the i-th time period. o ∈R S×d , encoded on the feature dimension d, we get: in, represents the new input data in the i-th time period after encoding on the feature dimension d, d m Indicates encoding the feature dimension d to d m ; Step S4.2: Perform a linear transformation to obtain the variables in the attention mechanism: in, represents the key vector within the time period, V i intra A vector of values ​​representing a time period; Step S4.3: Using trainable variables Combine the previous and next time series information within the time period to calculate the input data X of the i-th time period i Attention information within: Step S4.4: Execute steps S4.1-S4.3 on the input data of each time period in the divided new time series, and obtain the attention information as local time series information.

6. The airport passenger flow multi-factor fusion prediction method based on causal inference according to claim 5 is characterized in that: Step S5 specifically includes the following steps: Step S5.1: The new time series X∈R after division P×S×d , encoded into d in feature dimension d m ′, the mapped data is: Among them, d m ′ = S × d; Step S5.2: Perform self-attention operation between time periods, so as to model the airport passenger flow data based on the correlation between time periods. inter The linear mapping of derives the variables between time periods: Q inter ,K inter ,V inter ∈R P×dm Among them, Q inter represents the query vector between time periods, K inter represents the key vector between time periods, V inter A vector of values ​​representing time periods; Step S5.3, obtaining attention information between time periods: Step S5.4: Execute steps S5.1-5.3 between all time periods of the divided new time series, and the attention information between the obtained time periods is the global timing information.

Images