Prediction model training method, passenger volume prediction method, device and electronic equipment

Abstract

The present invention provides a prediction model training method, a passenger volume prediction method, a device and an electronic device, and relates to the field of data processing technology. The method comprises: obtaining a target route associated with the route to be processed according to the route information of the route to be processed; inferring the inference result of the total number of historical passengers carried by the route to be processed on the first historical date according to the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date, and the time characteristic data corresponding to the first historical date, and obtaining the passenger volume prediction model by iteratively training the initial model according to the inference result of the total number of historical passengers carried by the route to be processed on the first historical date, and the time characteristic data corresponding to the first historical date and the exogenous variables corresponding to the first historical date. The present invention can improve the prediction accuracy of passenger volume.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a prediction model training method, a passenger volume prediction method, a device and an electronic device. Background Art

[0002] International routes are one of the most important modes of transportation between different countries. How to efficiently and accurately predict the passenger volume of international flights is one of the technical problems that the industry urgently needs to solve.

[0003] In the related technology, when predicting the passenger volume of a flight, a large amount of historical data and prior knowledge are usually required to build a prediction model, and long-term and continuous observation data are required to train a prediction model with good prediction performance. However, due to the limitations of data privacy protection, the passenger volume data recorded by flights between different airlines cannot be shared in real time and in large quantities, and the passenger volume of the entire market is missing. Before predicting the full market traffic of future routes, it is necessary to obtain the historical full market passenger volume in order to accurately train the model, which results in a shortage of training sample data, making it impossible to train a high-performance prediction model, and thus making it difficult to accurately predict the passenger volume of the route.

[0004] Therefore, there is an urgent need for a prediction model training method, a passenger volume prediction method, a device and an electronic device to solve the above problems. Summary of the invention

[0005] The present invention provides a prediction model training method, a passenger volume prediction method, a device and an electronic device, which are used to solve the defect in the prior art that it is difficult to accurately predict the passenger volume of each airline due to the limitation of data privacy protection, and to improve the prediction accuracy of the passenger volume.

[0006] The present invention provides a prediction model training method, which is applied to a target data storage system, and the method comprises:

[0007] According to the route information of the route to be processed, obtaining a target route associated with the route to be processed;

[0008] Inferring the inferred result of the total number of historical passengers carried by the route to be processed on the first historical date according to the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date corresponding to the target route;

[0009] Iteratively training the initial model based on the inferred result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date to obtain a passenger volume prediction model;

[0010] Among them, the first historical total number of passengers carried and the first planned total number of seats for sale are respectively the actual total number of passengers carried and the planned total number of seats for sale of the first airline on the first historical date under the target route; the second historical total number of passengers carried and the second planned total number of seats for sale are respectively the actual total number of passengers carried and the planned total number of seats for sale of the first airline on the second historical date before the first historical date under the target route; the third planned total number of seats for sale is the planned total number of seats for sale of the second airline on the first historical date under the target route; the exogenous variable corresponding to the first historical date is determined according to the proportional coefficient between the prices of the destination and the departure place of the route to be processed on the first historical date; the first airline is the airline stored in the target data storage system; the second airline is an airline in other regions except the airline stored in the target data storage system.

[0011] According to a prediction model training method provided by the present invention, obtaining a target route associated with the route to be processed according to the route information of the route to be processed includes:

[0012] According to the route information of the route to be processed, searching in the target data storage system for a route that is the same as the route to be processed;

[0013] If a route identical to the route to be processed is found, the route identical to the route to be processed is used as the target route;

[0014] If no route identical to the route to be processed is found, the itinerary route, the location coordinates of the departure airport, and the location coordinates of the destination airport of the route to be processed are obtained based on the route information of the route to be processed, and the target route is obtained in the target data storage system based on the itinerary route, the location coordinates of the departure airport, and the location coordinates of the destination airport of the route to be processed.

[0015] According to a prediction model training method provided by the present invention, obtaining the target route in the target data storage system according to the itinerary route of the route to be processed, the location coordinates of the departure airport and the location coordinates of the destination airport includes:

[0016] Determine a first relative distance between each of the historical routes and the route to be processed according to the included angle between the itinerary of each of the historical routes and the itinerary of the route to be processed in the target data storage system, and the distance between the itinerary of each of the historical routes;

[0017] Determine a second relative distance between each of the historical routes and the route to be processed according to the length of a perpendicular line segment from the location coordinates of the departure airport of the route to be processed to the itinerary of each of the historical routes, and the length of a perpendicular line segment from the location coordinates of the destination airport of the route to be processed to the itinerary of each of the historical routes;

[0018] Determine a third relative distance between each of the historical routes and the route to be processed according to the distance between the location coordinates of the departure airport of the route to be processed and the projection of the itinerary of each of the historical routes to the location coordinates of the departure airport of each of the historical routes, and the distance between the location coordinates of the destination airport of the route to be processed and the projection of the itinerary of each of the historical routes to the location coordinates of the destination airport of each of the historical routes;

[0019] The target route is determined from the plurality of historical routes according to the first relative distance, the second relative distance, and the third relative distance.

[0020] According to a prediction model training method provided by the present invention, determining the target route among the plurality of historical routes according to the first relative distance, the second relative distance, and the third relative distance includes:

[0021] Performing weighted fusion on the first relative distance, the second relative distance and the third relative distance to obtain a total relative distance between each of the historical routes and the route to be processed;

[0022] Among the multiple historical routes, the route with the smallest total relative distance is determined as the target route.

[0023] According to a prediction model training method provided by the present invention, the method further includes:

[0024] Obtaining a plurality of the first historical dates, a plurality of the to-be-processed routes, and an inference result of the total number of historical passengers carried by each of the to-be-processed routes on each of the first historical dates;

[0025] The iterative training of the initial model based on the inference result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date to obtain the passenger volume prediction model includes:

[0026] From the plurality of said first historical dates, obtaining at least one target historical date and a plurality of historical dates before each of said target historical dates;

[0027] Inputting the inference results of the total number of passengers carried by each of the to-be-processed routes on multiple historical dates before each of the target historical dates, as well as the time characteristic data corresponding to each of the target historical dates and the exogenous variables corresponding to each of the target historical dates, into the initial model to obtain the predicted value of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates;

[0028] The initial model is iteratively optimized based on the deviation between the predicted value of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates and the inferred result of the historical total number of passengers carried by each of the to-be-processed routes on each of the target historical dates to obtain the passenger volume prediction model.

[0029] According to a prediction model training method provided by the present invention, the time feature data corresponding to the first historical date is obtained based on the following steps:

[0030] Acquire date data corresponding to the first historical date; the date data includes at least one of year data, month data, day data and week data;

[0031] Determine whether the first historical date and multiple dates before and after the first historical date are holidays, and obtain holiday feature data corresponding to the first historical date;

[0032] Determine whether the first historical date is a compensatory holiday, and obtain compensatory holiday feature data corresponding to the first historical date;

[0033] The time characteristic data corresponding to the first historical date is acquired according to the date data, the holiday characteristic data and the adjusted holiday characteristic data.

[0034] The present invention also provides a passenger volume prediction method, comprising:

[0035] Obtaining the total number of passengers actually carried by the route to be predicted on the third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted;

[0036] Inputting the actual total number of passengers carried by the route to be predicted on the third historical date before the date to be predicted, the time characteristic data corresponding to the date to be predicted, and the exogenous variables corresponding to the date to be predicted into the passenger volume prediction model to obtain a predicted value of the total number of passengers carried by the route to be predicted on the date to be predicted;

[0037] Wherein, the passenger volume prediction model is trained based on any of the prediction model training methods described above.

[0038] The present invention also provides a prediction model training device, which is applied to a target data storage system, and the device comprises:

[0039] A first acquisition module, configured to acquire a target route associated with the route to be processed according to the route information of the route to be processed;

[0040] an inference module, configured to infer the inference result of the total number of historical passengers carried by the route to be processed on the first historical date according to the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date corresponding to the target route;

[0041] A training module, configured to iteratively train the initial model based on the inference result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date, so as to obtain a passenger volume prediction model;

[0042] Among them, the first historical total number of passengers carried and the first planned total number of seats for sale are respectively the actual total number of passengers carried and the planned total number of seats for sale of the first airline on the first historical date under the target route; the second historical total number of passengers carried and the second planned total number of seats for sale are respectively the actual total number of passengers carried and the planned total number of seats for sale of the first airline on the second historical date before the first historical date under the target route; the third planned total number of seats for sale is the planned total number of seats for sale of the second airline on the first historical date under the target route; the exogenous variable corresponding to the first historical date is determined according to the proportional coefficient between the prices of the destination and the departure place of the route to be processed on the first historical date; the first airline is the airline stored in the target data storage system; the second airline is an airline in other regions except the airline stored in the target data storage system.

[0043] The present invention also provides a passenger volume prediction device, comprising:

[0044] The second acquisition module is used to obtain the total number of passengers actually carried by the route to be predicted on the third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted;

[0045] A prediction module, for inputting the actual total number of passengers carried by the route to be predicted on a third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted, into a passenger volume prediction model to obtain a predicted value of the total number of passengers carried by the route to be predicted on the date to be predicted;

[0046] Wherein, the passenger volume prediction model is trained based on any of the prediction model training methods described above.

[0047] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any one of the prediction model training methods described above, or implements any one of the passenger volume prediction methods described above.

[0048] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements any of the prediction model training methods described above, or implements any of the passenger volume prediction methods described above.

[0049] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the prediction model training methods described above, or implements any of the passenger volume prediction methods described above.

[0050] The prediction model training method, passenger volume prediction method, device and electronic device provided by the present invention infer and supplement the inference result of the historical total number of passengers carried by the target route associated with the route to be processed on the first historical date through the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date, and construct a large amount of historical sample data based on the inference result of the historical total number of passengers carried by each route to be processed on each first historical date, as well as the time characteristic data corresponding to each first historical date and the exogenous variables corresponding to each first historical date, so as to train and obtain the historical actual total number of passengers carried by the route for which the passenger volume prediction is required, as well as the time characteristic data of the future date and the exogenous variables corresponding to the future date, and accurately predict the passenger volume prediction model of the predicted value of the total number of passengers carried by the route for all market flights on the future date, thereby effectively improving the accuracy of the prediction of the passenger volume in the whole market even under the restriction of data privacy protection, and expanding the application scope of the existing passenger prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1It is one of the flow charts of the prediction model training method provided by the present invention;

[0053] Figure 2 This is the second flow chart of the prediction model training method provided by the present invention;

[0054] Figure 3 It is a flow chart of the passenger volume prediction method provided by the present invention;

[0055] Figure 4 It is a structural schematic diagram of the prediction model training device provided by the present invention;

[0056] Figure 5 It is a flow chart of the passenger volume prediction method provided by the present invention;

[0057] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] As one of the most important modes of transportation between different countries, international routes play an indispensable role. As a hub of aviation network transportation, the grasp and prediction of international route passenger volume will greatly affect the allocation and decision-making of airport-related resources, and then affect the travel experience of passengers entering and leaving the country. Therefore, the prediction of international route passenger volume will have outstanding practical significance for airport operations and the development of the civil aviation industry.

[0060] As the demand for air transport continues to pay attention to passenger volume forecasting, its demand forecasting theories and methods emerge in an endless stream, including regression models, gravity models, time series models, neural networks, gray models, etc. For example, in the prior art, by obtaining the historical passenger volume feature data of the day to be predicted and the date feature data of the day to be predicted, calculating the interval day, and inputting the feature data and the normalized date feature data into the trained passenger volume forecasting model, the passenger volume forecast value with specific time characteristics is obtained.

[0061] In the above method, the passenger volume of the same origin to destination (OD) or airport in the time series is mainly based on the periodicity and trend characteristics of the passenger volume data, so as to predict the passenger volume of the same OD or airport in the future, which can meet the actual application needs to a certain extent. Although the model provides a method for predicting passenger volume in the route or airport dimension, when predicting the passenger volume of a flight, a large amount of historical data and prior knowledge are usually required to build a prediction model, and long-term and continuous observation data are required to train a prediction model with good prediction performance. However, the passenger data of different airlines, that is, airlines in different countries and regions, is different due to the different protection of their operating data, overseas passenger privacy data, departure systems, and booking systems by airlines in different countries and regions. As a result, different systems in different countries cannot timely and accurately share and obtain and record and store the actual passenger volume data of airlines in other countries and regions, resulting in a lack of training sample data. Therefore, the existing technology is limited by the lack of model input data. Therefore, the application scope of the above-mentioned public models and technologies is limited to the passenger volume forecasts of flights of some airlines stored in the model owner's system. In the flight passenger analysis and forecast of airlines outside the system, as well as the inference and forecast of international flight passenger volume data of the entire market (that is, all airlines including domestic airlines in the target country and foreign airlines in the target country), the data fitting and prediction accuracy are poor, which cannot meet the needs of flight passenger analysis and forecast of airlines outside the system, as well as the inference and forecast of international flight passenger volume data of the entire market.

[0062] In response to the above problems, the present embodiment provides a method that can supplement and infer the total historical passenger volume of flights operated by airlines across the market, and can accurately predict the future inbound and outbound international flight passenger volume for the entire market at a certain airport based on the inferred total historical passenger volume of flights and the historical passenger volume characteristics and time characteristics actually stored in the target data storage system. This is used to solve the problem of poor prediction accuracy in the prior art due to different system usage and privacy restrictions, which makes it impossible to timely and accurately obtain historical passenger volume data for the entire market, especially for foreign airlines' flights.

[0063] Figure 1 One of the flow charts of the prediction model training method provided by the present invention; Figure 1 As shown, the method is applied to a target data storage system; the target data storage system is a storage system corresponding to the prediction model holder; the method comprises the following steps:

[0064] Step 110, acquiring a target route associated with the route to be processed according to the route information of the route to be processed;

[0065] The pending routes here are routes that participate in the passenger volume prediction model training and need to infer and supplement the actual total number of passengers carried on historical dates. They can be international routes; international routes refer to routes that take off from airports in one country or region and go to other countries or regions, such as flights from a target country or region to another country or region, or flights from another country or region to a target country or region.

[0066] The flights corresponding to the routes to be processed are all international flights in the market, including international flights operated by airlines in the target country (hereinafter referred to as the first airline), that is, the flight data stored in the storage system corresponding to the prediction model holder; and international flights operated by airlines in other countries (hereinafter referred to as the second airline), that is, the flight data outside the storage system corresponding to the prediction model holder. The so-called international flights can be flights from the target country to other countries or flights from other countries to the target country.

[0067] Optionally, when it is necessary to train the passenger volume prediction model, the route information of the route to be processed is first obtained, and then the target route associated with the route to be processed stored in the target data storage system is obtained, so as to infer and supplement the historical passenger information of the route to be processed based on the passenger volume associated information of the target route stored in the target data storage system. The route information here includes but is not limited to the itinerary route, the location coordinates of the departure airport, and the location coordinates of the destination airport.

[0068] Here, the method of obtaining the target route can be to compare the itinerary route, location coordinates of the departure airport, and location coordinates of the destination airport of the route to be processed with the itinerary routes, location coordinates of the departure airport, and location coordinates of the destination airport of each historical route in the target data storage system for route itinerary identity or similarity, and based on the comparison result, select from the target data storage system a route that is identical or similar to the route to be processed as the target route associated with the route to be processed.

[0069] Step 120, based on the first historical total number of passengers carried, the first planned total number of seats for sale, the second historical total number of passengers carried, the second planned total number of seats for sale, the third planned total number of seats for sale, and the time characteristic data corresponding to the first historical date corresponding to the target route, infer the inference result of the historical total number of passengers carried for the route to be processed on the first historical date; wherein the first historical total number of passengers carried and the first planned total number of seats for sale are respectively the actual total number of passengers carried and the planned total number of seats for sale of the first airline stored in the target data storage system under the target route on the first historical date; the second historical total number of passengers carried and the second planned total number of seats for sale are respectively the actual total number of passengers carried and the planned total number of seats for sale of the first airline stored in the target data storage system under the target route on the second historical date before the first historical date; the third planned total number of seats for sale is the planned total number of seats for sale of a second airline other than the target data storage system under the target route on the first historical date; the first airline is an airline stored in the target data storage system; the second airline is an airline in other regions other than the airline stored in the target data storage system; here, the second airline is an airline in other countries and regions among all market airlines except the first airline.

[0070] Optionally, after the target route is obtained, the planned flight record data of all international flights operated by all airlines in the market, i.e., the first airline and the second airline, on the first historical date and the second historical date before the first historical date are obtained; the so-called planned flight record data includes but is not limited to the planned flight date, the planned number of seats sold, the departure airport and the destination airport, and the latitude and longitude information of the departure airport and the destination airport;

[0071] In addition, actual flight record data of each international flight operated by the first airline under the target route on the first historical date and the second historical date before the first historical date are obtained from the target data storage system; the so-called actual flight record data includes but is not limited to the actual number of passengers carried, the planned number of seats sold, the departure airport and the destination airport, and the latitude and longitude information of the departure airport and the destination airport.

[0072] The so-called second historical date is a historical date that is several days before the first historical date, for example, a historical date that is 6 days before, or it may be a historical date that is the same as the first historical date in the historical year.

[0073] Next, the actual flight record data and the planned flight record data are summarized, and the target route is used as the statistical caliber to obtain the actual total number of passengers carried and the total number of planned seats sold for all flights operated by the first airline under the target route on the first historical date, so as to obtain the first historical total number of passengers carried x1 and the first planned total number of seats sold x2; using the target route as the statistical caliber, the actual total number of passengers carried and the total number of planned seats sold for all flights operated by the first airline under the target route on the second historical date are obtained to obtain the second historical total number of passengers carried x3 and the second planned total number of seats sold x4; using the target route as the statistical caliber, other airlines stored in the non-target data storage system under the target route, that is, the total number of planned seats sold by the second airline on the first historical date x5 are obtained.

[0074] Furthermore, obtaining time characteristic data x6 corresponding to the first historical date;

[0075] The time feature data here is obtained by extracting time features through a time feature extraction algorithm; the so-called time feature extraction algorithm includes but is not limited to an extraction algorithm for extracting date data for the first historical date, and a feature extraction algorithm for determining whether the first historical date is a weekend, a day of the week, a holiday, a compensatory day, a winter or summer vacation, and whether the three days before / after the first historical date are holidays.

[0076] In some embodiments, the time characteristic data corresponding to the first historical date is obtained based on the following steps:

[0077] Acquire date data corresponding to the first historical date; the date data includes at least one of year data, month data, day data and week data;

[0078] Determine whether the first historical date and multiple dates before and after the first historical date are holidays, and obtain holiday feature data corresponding to the first historical date;

[0079] Determine whether the first historical date is a compensatory holiday, and obtain compensatory holiday feature data corresponding to the first historical date;

[0080] The time characteristic data corresponding to the first historical date is acquired according to the date data, the holiday characteristic data and the adjusted holiday characteristic data.

[0081] Optionally, date data corresponding to the first historical date is obtained, including but not limited to specific year data, month data, day data and week data, as well as determining whether the first historical date is a holiday, determining whether multiple dates before and after the first historical date are holidays, and determining whether the first historical date is a compensatory holiday, thereby obtaining time feature data corresponding to the first historical date.

[0082] Here, the multiple dates before and after the first historical date can be set according to actual needs, such as three days before and after the first historical date; the so-called holidays include but are not limited to winter and summer vacations and traditional festivals in different regions.

[0083] Next, the first historical total number of passengers carried x1, the first planned total number of seats sold x2, the second historical total number of passengers carried x3, the second planned total number of seats sold x4, the third planned total number of seats sold x5, and the time feature data x6 corresponding to the first historical date are subjected to standardized preprocessing, so as to infer the total number of actual passengers carried by all flights of the to-be-processed route on the first historical date, i.e., all flights operated by the first airline and the second airline, based on the standardized preprocessed data. The so-called standardized processing may include normalization processing, maximum and minimum scaling processing, etc.

[0084] Here, the method for obtaining the inference result of the total number of passengers carried historically for the route to be processed on the first historical date can be to input the first historical total number of passengers carried x1, the first planned total number of seats sold x2, the second historical total number of passengers carried x3, the second planned total number of seats sold x4, the third planned total number of seats sold x5, and the time characteristic data x6 corresponding to the first historical date into the route passenger volume inference model to infer the inference result of the total number of passengers carried historically for all market flights of the route to be processed on the first historical date.

[0085] Here, the route passenger volume inference model can be obtained through training based on historical samples, that is, based on the total number of passengers actually carried on a preset historical date on a small number of historical routes with the same departure airport or destination airport as the route to be processed, and the total number of passengers actually carried and the total number of seats planned to be sold by the first airline on the preset historical date, the total number of passengers actually carried and the total number of seats planned to be sold by the first airline on multiple historical dates before the preset historical date, and the total number of seats planned to be sold by the second airline on the preset historical date under the target route associated with the historical route. The total number of passengers actually carried on a small number of historical routes on the preset historical date can be manually inferred and marked, or can be marked based on analysis of industry research reports.

[0086] Step 130, iteratively train the initial model based on the inference result of the total number of passengers carried by the route to be processed on the first historical date, as well as the time characteristic data corresponding to the first historical date and the exogenous variables corresponding to the first historical date, to obtain a passenger volume prediction model; the exogenous variables corresponding to the first historical date are determined based on the proportional coefficient between the prices of the destination and the departure place of the route to be processed on the first historical date.

[0087] Optionally, steps 110-120 are iterated to obtain inferred results of the total number of passengers carried by multiple pending routes on multiple first historical dates, and multiple historical sample data are constructed based on the inferred results of the total number of passengers carried by each pending route on each first historical date, as well as the time characteristic data corresponding to each first historical date and the exogenous variables corresponding to each first historical date, and the initial model is iteratively trained based on each historical sample data, thereby obtaining a passenger volume prediction model that can accurately predict the total number of passengers carried by all market flights of the route on future dates based on the historical actual total number of passengers carried for various routes that need to be predicted, as well as the time characteristic data of future dates and the exogenous variables corresponding to the future dates.

[0088] Here, the historical actual total number of passengers carried on the route to be predicted can be inferred based on the inference steps of steps 110-120, which will not be repeated here.

[0089] The so-called initial model can be constructed based on the Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX). The core idea of ​​SARIMAX is to use the autoregressive algorithm to link the historical sequence values, current sequence values ​​and external factors through a certain model, use historical sample data for training, and then derive appropriate model parameters, and comprehensively consider endogenous and exogenous factors to improve prediction accuracy. The calculation formula of SARIMAX is as follows:

[0090]

[0091] Where Y(t) represents the observed value on date t, i.e., the predicted value of the total number of passengers, c represents a constant, {Φ1,…,Φp} represents the autoregressive coefficient, {θ1,…,θq} represents the moving average coefficient, {Φs,…,Φ Ps} represents the seasonal autoregressive coefficient, ε(t) is the white noise error term, α represents the weight coefficient of the exogenous variable, Represents the observed value of the exogenous variable on date t; p, q, P, s and each coefficient are model parameters obtained after training based on historical sample data.

[0092] The method provided in this embodiment infers and supplements the inference result of the historical total number of passengers carried of the target route associated with the target route on the first historical date through the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date, and constructs a large amount of historical sample data based on the inference result of the historical total number of passengers carried of each target route on each first historical date, as well as the time characteristic data corresponding to each first historical date and the exogenous variables corresponding to each first historical date, so as to train and obtain the historical actual total number of passengers carried by the route for which passenger volume prediction is required, as well as the time characteristic data of future dates and the exogenous variables corresponding to future dates, and accurately predict the passenger volume prediction model for the predicted value of the total number of passengers carried by the route for all market flights on the future date, thereby effectively improving the accuracy of the all-market passenger volume prediction even under the restriction of data privacy protection, and expanding the application scope of the existing passenger prediction model.

[0093] In some embodiments, in step 110, obtaining a target route associated with the route to be processed according to the route information of the route to be processed includes:

[0094] According to the route information of the route to be processed, searching in the target data storage system for a route that is the same as the route to be processed;

[0095] If a route identical to the route to be processed is found, the route identical to the route to be processed is used as the target route;

[0096] If no route identical to the route to be processed is found, the itinerary route, the location coordinates of the departure airport, and the location coordinates of the destination airport of the route to be processed are obtained based on the route information of the route to be processed, and the target route is obtained in the target data storage system based on the itinerary route, the location coordinates of the departure airport, and the location coordinates of the destination airport of the route to be processed.

[0097] Optionally, the step of obtaining a target route associated with the route to be processed specifically includes:

[0098] The route information of the route to be processed is matched and searched with the route information of each historical route in the target data storage system to determine whether the route to be processed is a newly added route in the target data storage system; if the same route as the route to be processed is found, it is determined that the route to be processed is not a newly added route; at this time, the route that is the same as the route to be processed can be used as the target route, and the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date under the same route are input into the route passenger volume inference model to infer the historical total number of passengers carried for the route to be processed on the first historical date.

[0099] If no route identical to the route to be processed is found, it is determined that the route to be processed is a newly added route; at this time, it is necessary to parse the route information of the route to be processed to obtain the itinerary route, the location coordinates of the departure airport and the location coordinates of the destination airport of the route to be processed, and calculate the itinerary similarity between the route to be processed and each historical route in the flight data storage system based on the itinerary route, the location coordinates of the departure airport and the location coordinates of the destination airport of the route to be processed, and obtain the route with the closest itinerary to the route to be processed as the target route, and pull down the similar route, and input the corresponding first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date into the route passenger volume inference model to infer the historical total number of passengers carried for the route to be processed on the first historical date.

[0100] The method provided in this embodiment can quickly and accurately match and obtain the target route associated with the route to be processed by searching and judging based on route information, and calculating the itinerary similarity based on the itinerary route, the location coordinates of the departure airport, and the location coordinates of the destination airport, so as to subsequently train the prediction model more accurately, thereby improving the prediction performance of the prediction model.

[0101] In some embodiments, acquiring the target route in the target data storage system according to the itinerary of the route to be processed, the location coordinates of the departure airport, and the location coordinates of the destination airport includes:

[0102] Determine a first relative distance between each of the historical routes and the route to be processed according to the included angle between the itinerary of each of the historical routes and the itinerary of the route to be processed in the target data storage system, and the distance between the itinerary of each of the historical routes;

[0103] Determine a second relative distance between each of the historical routes and the route to be processed according to the length of a perpendicular line segment from the location coordinates of the departure airport of the route to be processed to the itinerary of each of the historical routes, and the length of a perpendicular line segment from the location coordinates of the destination airport of the route to be processed to the itinerary of each of the historical routes;

[0104] Determine a third relative distance between each of the historical routes and the route to be processed according to the distance between the location coordinates of the departure airport of the route to be processed and the projection of the itinerary of each of the historical routes to the location coordinates of the departure airport of each of the historical routes, and the distance between the location coordinates of the destination airport of the route to be processed and the projection of the itinerary of each of the historical routes to the location coordinates of the destination airport of each of the historical routes;

[0105] The target route is determined from the plurality of historical routes according to the first relative distance, the second relative distance, and the third relative distance.

[0106] Optionally, the first relative distance d θ It is determined by the product of the angle sinθ between the itinerary of each historical route and the itinerary of the route to be processed and the distance L2 of each historical route ||L2||. The specific calculation formula is as follows:

[0107] d θ =||L2||·sinθ;

[0108] The second relative distance d ⊥ It is the length d of the vertical line segment from the location A1 coordinate of the departure airport of the route to be processed to the itinerary L2 of each historical route ⊥,a d ⊥,a The square value of the destination airport B1 coordinate of the route to be processed and the length d of the vertical line segment from the itinerary L2 of each historical route ⊥,b and d ⊥,b The specific calculation formula is as follows:

[0109]

[0110] The third relative distance d ‖ It is the distance d between the projection of the location coordinate A1 of the departure airport of the route to be processed on the itinerary line L2 of each historical route and the location coordinate A2 of the departure airport of each historical route. ‖,a The distance d between the projection of the location coordinate B1 of the destination airport of the route to be processed on the itinerary line L2 of each historical route to the location coordinate B2 of the destination airport of each historical route ‖,b The specific calculation formula is as follows:

[0111] d ‖ =min(d ‖,a ,d ‖,b );

[0112] Here, the location coordinates A1 of the departure airport and B1 of the destination airport are the latitude and longitude of the departure airport of the route to be processed and the coordinates of the latitude and longitude of the destination airport in the Lambert projection. Lambert projection is a cylindrical projection used to convert the latitude and longitude coordinates on the earth's surface into plane coordinates so as to accurately represent the earth's surface area on the map. Similarly, the location coordinates A2 of each historical route and the location coordinates B2 of the destination airport are the latitude and longitude of the departure airport of each historical route and the coordinates of the latitude and longitude of the destination airport in the Lambert projection.

[0113] By using the above formula, the first relative distance d between each historical route and the route to be processed is obtained. θ , the second relative distance d ⊥ and the third relative distance d ‖ , and combine the first relative distance d between each historical route and the route to be processed θ , the second relative distance d ⊥ and the third relative distance d ‖ The itinerary similarity between each historical route and the route to be processed is determined, so as to determine the route with the closest itinerary to the route to be processed from the multiple historical routes as the target route.

[0114] The method provided in this embodiment determines the itinerary similarity between routes in a multi-level and multi-dimensional manner, so as to accurately and efficiently obtain the target route that is closest to the itinerary between the newly added routes to be processed, so as to conduct more accurate prediction model training in the future, thereby improving the prediction performance of the prediction model.

[0115] In some embodiments, determining the target route from a plurality of the historical routes according to the first relative distance, the second relative distance, and the third relative distance includes:

[0116] Performing weighted fusion on the first relative distance, the second relative distance and the third relative distance to obtain a total relative distance between each of the historical routes and the route to be processed;

[0117] Among the multiple historical routes, the route with the smallest total relative distance is determined as the target route.

[0118] Optionally, the total relative distance between each historical route and the route to be processed This can be achieved by adjusting the first relative distance d θ , the second relative distance d ⊥ and the third relative distance d‖ The specific calculation formula is as follows:

[0119]

[0120] Among them, w1, w2 and w3 are the second relative distances d ⊥ , the third relative distance d ‖ and the first relative distance d θ Each weight coefficient can be pre-configured according to actual needs, such as w1, w2 and w3 can be configured to 0.4, 0.3 and 0.3 respectively, or it can be calculated based on a weight algorithm.

[0121] The total relative distance between each historical route and the route to be processed is obtained by traversing the above formula Due to the total relative distance between each historical route and the route to be processed The smaller the distance, the higher the similarity of the two routes. Therefore, the total relative distance between the routes to be processed can be obtained from multiple historical routes. The smallest route is used as the target route with the highest itinerary similarity to the route to be processed.

[0122] The method provided in this embodiment calculates the total relative distance between the historical routes and the routes to be processed by weighted fusion of relative distances, and determines the best matching target route based on the principle of minimum total relative distance. This can improve the accuracy and efficiency of route matching, and then quickly find the historical route that is most similar to the route to be processed, so that the prediction model can be trained more accurately in the future, thereby improving the prediction performance of the prediction model.

[0123] In some embodiments, the method further comprises:

[0124] Obtaining a plurality of the first historical dates, a plurality of the to-be-processed routes, and an inference result of the total number of historical passengers carried by each of the to-be-processed routes on each of the first historical dates;

[0125] The iterative training of the initial model based on the inference result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date to obtain the passenger volume prediction model includes:

[0126] From the plurality of said first historical dates, obtaining at least one target historical date and a plurality of historical dates before each of said target historical dates;

[0127] Inputting the inference results of the total number of passengers carried by each of the to-be-processed routes on multiple historical dates before each of the target historical dates, as well as the time characteristic data corresponding to each of the target historical dates and the exogenous variables corresponding to each of the target historical dates, into the initial model to obtain the predicted value of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates;

[0128] The initial model is iteratively optimized based on the deviation between the predicted value of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates and the inferred result of the historical total number of passengers carried by each of the to-be-processed routes on each of the target historical dates to obtain the passenger volume prediction model.

[0129] Optionally, the training steps of the passenger volume prediction model specifically include:

[0130] A plurality of first historical dates and a plurality of routes to be processed are obtained, and based on steps 110 - 120 , the inference result of the total number of historical passengers carried by each route to be processed on each first historical date is obtained through traversal.

[0131] Next, at least one target historical date and multiple historical dates before each target historical date are obtained from the multiple first historical dates, so as to divide the multiple first historical dates into multiple historical time periods;

[0132] Next, the historical total passenger volume inferred results of each pending route on multiple historical dates before each target historical date, as well as the time characteristic data corresponding to each target historical date and the exogenous variables corresponding to each target historical date are used as historical samples and input into the initial model. The initial model predicts the total passenger volume forecast value of each pending route on each target historical date, and calculates the deviation between the total passenger volume forecast value of each pending route on each target historical date and the historical total passenger volume inferred results on each target historical date. The initial model is iteratively optimized with minimizing the deviation as the optimization goal, so as to obtain a passenger volume prediction model that can accurately predict the total passenger volume forecast value of all-market flights for each route.

[0133] Figure 2 The second flowchart of the prediction model training method provided by the present invention; Figure 2 As shown, the complete process steps of this method include:

[0134] Step 210, obtaining data required by the model, including flight plan data for the entire market, actual flight data of historical flights stored in the system, characteristic data of dates, price ratio data of the destination and departure points of the route, and pre-processing the data to meet the format requirements required by the model;

[0135] Step 220, determining whether the route to be processed is a newly added route;

[0136] Step 230, selecting an inference model corresponding to a non-newly added route from the prediction model set, inputting the processed routes that are the same as the routes to be processed and the data associated with the routes to be processed, as well as the characteristic data of the date, into the inference model to obtain an estimated value of the historical passenger volume of all market flights;

[0137] Step 240, selecting an inference model corresponding to the newly added route from the prediction model set;

[0138] Step 250, according to the similarity calculation formula, the itinerary similarity between the routes is calculated, and the route with the highest itinerary similarity with the route to be processed is obtained by traversing, and the associated data of the processed route with the highest itinerary similarity and the route to be processed, as well as the characteristic data of the date, are input into the inference model to obtain the estimated value of the historical passenger volume of all market flights;

[0139] Step 260, input the inferred estimated value of the historical passenger volume of all market flights, the processed actual operation data of historical flights, the characteristic data of future dates, and exogenous variables into the trained passenger prediction model to obtain the predicted value of the total passenger volume of all market flights on future dates.

[0140] In summary, the method provided in this embodiment, under the restrictions of data privacy protection, infers and supplements the actual total number of passengers carried by the missing routes to be processed on historical dates by utilizing other data features such as passenger capacity data and planned flight data in the target data storage system, so as to enrich the training sample set, thereby training a passenger volume prediction model that can accurately predict the passenger volume of flights in the entire market, thereby improving the passenger volume prediction performance, and solving the problem in the prior art that the input parameters, i.e., long-term and continuous historical observation values, cannot be accurately obtained by the corresponding systems of different airlines due to various reasons, resulting in the shortcomings of the prior art in estimating the passenger volume of foreign flights, such as low accuracy and high subjectivity, which affects the final prediction decision, etc.

[0141] Figure 3 A flow chart of the passenger volume prediction method provided by the present invention is shown in FIG. Figure 3 As shown, the method includes:

[0142] Step 310, obtaining the total number of passengers actually carried by the route to be predicted on the third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted;

[0143] Step 320, inputting the actual total number of passengers carried by the route to be predicted on the third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted, into the passenger volume prediction model to obtain a predicted value of the total number of passengers carried by the route to be predicted on the date to be predicted;

[0144] The passenger volume prediction model is obtained by training based on the prediction model training methods provided in the above embodiments.

[0145] Optionally, in the actual prediction process, the actual total number of passengers carried by the route to be predicted on multiple historical dates before the date to be predicted can be traversed and obtained with reference to steps 110-120, which will not be described in detail here.

[0146] In addition, the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted are obtained; the actual total number of passengers carried on the route to be predicted on multiple historical dates before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted are preprocessed, so that the preprocessed data can better meet the needs of the model and improve the performance and accuracy of the model.

[0147] Next, the preprocessed data is input into the passenger volume prediction model, which predicts the total number of passengers on all market flights for the predicted route to obtain the predicted value of the total number of passengers carried on the predicted route on the predicted date.

[0148] The method provided in this embodiment forms a large amount of historical sample data by inferring and supplementing the actual total number of passengers carried by the route to be processed on the first historical date, as well as the time feature data corresponding to each first historical date and the exogenous variables corresponding to each first historical date, and trains a passenger volume prediction model to perform passenger volume prediction, thereby effectively improving the accuracy of passenger volume prediction for the entire market even under the restrictions of data privacy protection, and expanding the application scope of existing passenger prediction models.

[0149] The prediction model training device provided by the present invention is described below. The prediction model training device described below and the prediction model training method described above can be referenced to each other.

[0150] Figure 4 A schematic diagram of the structure of the prediction model training device provided by the present invention; Figure 4 As shown, the device comprises:

[0151] The first acquisition module 410 is used to acquire a target route associated with the route to be processed according to the route information of the route to be processed;

[0152] The inference module 420 is used to infer the inference result of the total number of historical passengers carried by the route to be processed on the first historical date according to the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date corresponding to the target route;

[0153] The training module 430 is used to iteratively train the initial model based on the inference result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date to obtain a passenger volume prediction model;

[0154] Among them, the first historical total number of passengers carried and the first planned total number of seats for sale are respectively the actual total number of passengers carried and the planned total number of seats for sale of the first airline on the first historical date under the target route; the second historical total number of passengers carried and the second planned total number of seats for sale are respectively the actual total number of passengers carried and the planned total number of seats for sale of the first airline on the second historical date before the first historical date under the target route; the third planned total number of seats for sale is the planned total number of seats for sale of the second airline on the first historical date under the target route; the exogenous variable corresponding to the first historical date is determined according to the proportional coefficient between the prices of the destination and the departure place of the route to be processed on the first historical date; the first airline is the airline stored in the target data storage system; the second airline is an airline in other regions except the airline stored in the target data storage system.

[0155] The prediction model training device provided in this embodiment infers and supplements the inference result of the historical total number of passengers carried by the target route associated with the route to be processed on the first historical date through the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date, and constructs a large amount of historical sample data based on the inference result of the historical total number of passengers carried of each route to be processed on each first historical date, as well as the time characteristic data corresponding to each first historical date and the exogenous variables corresponding to each first historical date, so as to train the historical actual total number of passengers carried by the route for which the passenger volume can be predicted as required, as well as the time characteristic data of the future date and the exogenous variables corresponding to the future date, and accurately predict the passenger volume prediction model of the predicted value of the total number of passengers carried by the full market flights of the route on the future date, thereby effectively improving the accuracy of the full market passenger volume prediction even under the restriction of data privacy protection, and expanding the application scope of the existing passenger prediction model.

[0156] Figure 5 A schematic diagram of the structure of the passenger volume prediction device provided by the present invention; Figure 5 As shown, the device comprises:

[0157] The second acquisition module 510 is used to obtain the total number of passengers actually carried by the route to be predicted on the third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted;

[0158] The prediction module 520 is used to input the actual total number of passengers carried by the route to be predicted on the third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted, into the passenger volume prediction model to obtain the predicted value of the total number of passengers carried by the route to be predicted on the date to be predicted;

[0159] The passenger volume prediction model is obtained by training based on the prediction model training method provided in the above embodiments.

[0160] The device provided in this embodiment forms a large amount of historical sample data by inferring and supplementing the actual total number of passengers carried on the route to be processed on the first historical date, as well as the time feature data corresponding to each first historical date and the exogenous variables corresponding to each first historical date, and trains a passenger volume prediction model to perform passenger volume prediction, thereby effectively improving the accuracy of passenger volume prediction for the entire market even under the restrictions of data privacy protection, and expanding the application scope of existing passenger prediction models.

[0161] The device provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0162] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other via the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the prediction model training method or the passenger volume prediction method provided by the above methods.

[0163] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0164] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the prediction model training method or passenger volume prediction method provided by the above methods.

[0165] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the prediction model training method or the passenger volume prediction method provided by the above-mentioned methods.

[0166] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0167] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A prediction model training method, characterized in that: Applied to a target data storage system, the method comprises: According to the route information of the route to be processed, obtaining a target route associated with the route to be processed; Inferring the inferred result of the total number of historical passengers carried by the route to be processed on the first historical date according to the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date corresponding to the target route; Iteratively training the initial model based on the inferred result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date to obtain a passenger volume prediction model; Among them, the first historical total number of passengers carried and the first planned total number of seats for sale are respectively the total number of passengers actually carried and the total number of seats planned for sale by the first airline on the first historical date under the target route; the second historical total number of passengers carried and the second planned total number of seats for sale are respectively the total number of passengers actually carried and the total number of seats planned for sale by the first airline on the second historical date before the first historical date under the target route; the third planned total number of seats for sale is the total number of seats planned for sale by the second airline on the first historical date under the target route; the exogenous variable corresponding to the first historical date is determined according to the proportional coefficient between the prices of the destination and the departure place of the route to be processed on the first historical date; the first airline is the airline stored in the target data storage system; the second airline is an airline in other regions except the airline stored in the target data storage system; The method further comprises: Obtaining a plurality of the first historical dates, a plurality of the to-be-processed routes, and an inference result of the total number of historical passengers carried by each of the to-be-processed routes on each of the first historical dates; The iterative training of the initial model based on the inference result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date to obtain the passenger volume prediction model includes: From the plurality of said first historical dates, obtaining at least one target historical date and a plurality of historical dates before each of said target historical dates; Inputting the inference results of the total number of passengers carried by each of the to-be-processed routes on multiple historical dates before each of the target historical dates, as well as the time characteristic data corresponding to each of the target historical dates and the exogenous variables corresponding to each of the target historical dates, into the initial model to obtain the predicted value of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates; Iteratively optimizing the initial model according to the deviation between the predicted value of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates and the inferred result of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates, to obtain the passenger volume prediction model; The initial model is constructed based on an autoregressive moving average model with exogenous variables and seasonal adjustment.

2. The prediction model training method according to claim 1, characterized in that: The step of acquiring a target route associated with the route to be processed according to the route information of the route to be processed includes: According to the route information of the route to be processed, searching in the target data storage system for a route that is the same as the route to be processed; If a route identical to the route to be processed is found, the route identical to the route to be processed is used as the target route; If no route identical to the route to be processed is found, the itinerary route, the location coordinates of the departure airport, and the location coordinates of the destination airport of the route to be processed are obtained based on the route information of the route to be processed, and the target route is obtained in the target data storage system based on the itinerary route, the location coordinates of the departure airport, and the location coordinates of the destination airport of the route to be processed.

3. The prediction model training method according to claim 2, characterized in that: The step of acquiring the target route in the target data storage system according to the itinerary of the route to be processed, the location coordinates of the departure airport, and the location coordinates of the destination airport comprises: Determine a first relative distance between each of the historical routes and the route to be processed according to the included angle between the itinerary of each of the historical routes and the itinerary of the route to be processed in the target data storage system, and the distance between the itinerary of each of the historical routes; Determine a second relative distance between each of the historical routes and the route to be processed according to the length of a perpendicular line segment from the location coordinates of the departure airport of the route to be processed to the itinerary of each of the historical routes, and the length of a perpendicular line segment from the location coordinates of the destination airport of the route to be processed to the itinerary of each of the historical routes; Determine a third relative distance between each of the historical routes and the route to be processed according to the distance between the location coordinates of the departure airport of the route to be processed and the projection of the itinerary of each of the historical routes to the location coordinates of the departure airport of each of the historical routes, and the distance between the location coordinates of the destination airport of the route to be processed and the projection of the itinerary of each of the historical routes to the location coordinates of the destination airport of each of the historical routes; The target route is determined from the plurality of historical routes according to the first relative distance, the second relative distance, and the third relative distance.

4. The prediction model training method according to claim 3, characterized in that: The step of determining the target route from a plurality of historical routes according to the first relative distance, the second relative distance, and the third relative distance includes: Performing weighted fusion on the first relative distance, the second relative distance and the third relative distance to obtain a total relative distance between each of the historical routes and the route to be processed; Among the multiple historical routes, the route with the smallest total relative distance is determined as the target route.

5. The prediction model training method according to any one of claims 1 to 4, characterized in that: The time characteristic data corresponding to the first historical date is obtained based on the following steps: Acquire date data corresponding to the first historical date; the date data includes at least one of year data, month data, day data and week data; Determine whether the first historical date and multiple dates before and after the first historical date are holidays, and obtain holiday feature data corresponding to the first historical date; Determine whether the first historical date is a compensatory holiday, and obtain compensatory holiday feature data corresponding to the first historical date; The time characteristic data corresponding to the first historical date is acquired according to the date data, the holiday characteristic data and the adjusted holiday characteristic data.

6. A passenger volume prediction method, characterized in that: include: Obtaining the total number of passengers actually carried by the route to be predicted on the third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted; Inputting the actual total number of passengers carried by the route to be predicted on the third historical date before the date to be predicted, the time characteristic data corresponding to the date to be predicted, and the exogenous variables corresponding to the date to be predicted into the passenger volume prediction model to obtain a predicted value of the total number of passengers carried by the route to be predicted on the date to be predicted; The passenger volume prediction model is trained based on the prediction model training method according to any one of claims 1 to 5.

7. A prediction model training device, characterized in that: Applied to a target data storage system, the device comprises: A first acquisition module, configured to acquire a target route associated with the route to be processed according to the route information of the route to be processed; an inference module, configured to infer the inference result of the total number of historical passengers carried by the route to be processed on the first historical date according to the first historical total number of passengers carried, the first planned total number of seats sold, the second historical total number of passengers carried, the second planned total number of seats sold, the third planned total number of seats sold, and the time characteristic data corresponding to the first historical date corresponding to the target route; A training module, configured to iteratively train the initial model based on the inference result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date, so as to obtain a passenger volume prediction model; Among them, the first historical total number of passengers carried and the first planned total number of seats for sale are respectively the total number of passengers actually carried and the total number of seats planned for sale by the first airline on the first historical date under the target route; the second historical total number of passengers carried and the second planned total number of seats for sale are respectively the total number of passengers actually carried and the total number of seats planned for sale by the first airline on the second historical date before the first historical date under the target route; the third planned total number of seats for sale is the total number of seats planned for sale by the second airline on the first historical date under the target route; the exogenous variable corresponding to the first historical date is determined according to the proportional coefficient between the prices of the destination and the departure place of the route to be processed on the first historical date; the first airline is the airline stored in the target data storage system; the second airline is an airline in other regions except the airline stored in the target data storage system; The device also includes: Obtaining a plurality of the first historical dates, a plurality of the to-be-processed routes, and an inference result of the total number of historical passengers carried by each of the to-be-processed routes on each of the first historical dates; The iterative training of the initial model based on the inference result of the total number of passengers carried by the route to be processed on the first historical date, the time characteristic data corresponding to the first historical date, and the exogenous variables corresponding to the first historical date to obtain the passenger volume prediction model includes: From the plurality of said first historical dates, obtaining at least one target historical date and a plurality of historical dates before each of said target historical dates; Inputting the inference results of the total number of passengers carried by each of the to-be-processed routes on multiple historical dates before each of the target historical dates, as well as the time characteristic data corresponding to each of the target historical dates and the exogenous variables corresponding to each of the target historical dates, into the initial model to obtain the predicted value of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates; Iteratively optimizing the initial model according to the deviation between the predicted value of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates and the inferred result of the total number of passengers carried by each of the to-be-processed routes on each of the target historical dates, to obtain the passenger volume prediction model; The initial model is constructed based on an autoregressive moving average model with exogenous variables and seasonal adjustment.

8. A passenger volume prediction device, characterized in that: include: The second acquisition module is used to obtain the total number of passengers actually carried by the route to be predicted on the third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted; A prediction module, for inputting the actual total number of passengers carried by the route to be predicted on a third historical date before the date to be predicted, as well as the time characteristic data corresponding to the date to be predicted and the exogenous variables corresponding to the date to be predicted, into a passenger volume prediction model to obtain a predicted value of the total number of passengers carried by the route to be predicted on the date to be predicted; The passenger volume prediction model is trained based on the prediction model training method according to any one of claims 1 to 5.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the prediction model training method as described in any one of claims 1 to 5, or implements the passenger volume prediction method as described in claim 6.

Images