Passenger flow prediction method and device, computing device, and storage medium

Abstract

The application discloses a passenger flow prediction method and device, a computing device and a storage medium. The method comprises the following steps: obtaining historical passenger flow data of ordinary days in the same period in the past of a target space in a to-be-predicted period and a first coefficient, wherein the to-be-predicted period is located in a first holiday, and the first coefficient represents the difference between the first holiday and the ordinary day in the historical passenger flow data in the same year before the to-be-predicted period; and predicting the passenger flow prediction result of the target space in the to-be-predicted period according to the historical passenger flow data of the ordinary days in the same period in the past of the target space in the to-be-predicted period and the first coefficient. The method distinguishes different types of holidays, analyzes different characteristics of different holidays, and can realize accurate passenger flow prediction under various types of holidays.

Description

Technical Field

[0001] This application relates to the field of transportation technology, and in particular to a passenger flow prediction method, apparatus, computing device, and storage medium. Background Technology

[0002] Currently, urban rail transit lines operate independently, and trains are not rationally scheduled according to passenger flow fluctuations, often resulting in operational problems such as peak-hour congestion and off-peak waste. Especially during holidays or large events, the surge in passenger flow poses a significant challenge to the operation and management of urban rail transit. Using traditional, passive emergency response methods can easily lead to passenger congestion and delays, severely impacting the passenger travel experience. Therefore, urban rail transit passenger flow forecasting has become a necessary tool. By predicting passenger flow over a certain period, data support can be provided for the design and adjustment of traffic operation plans, thereby achieving objectives such as providing emergency command suggestions, coordinating line capacity, and comprehensive energy conservation.

[0003] However, some existing passenger flow forecasting schemes do not differentiate between specific types of holidays in their holiday passenger flow forecasting models, resulting in insufficient accuracy in passenger flow forecasting for various types of holidays and making it difficult to provide accurate passenger flow forecasting data for urban rail transit operators. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this application provides a passenger flow prediction method, apparatus, computing device, and storage medium, which distinguishes different types of holidays, analyzes the different characteristics of different types of holidays, and can achieve accurate passenger flow prediction under various types of holidays.

[0005] In a first aspect, this application provides a passenger flow forecasting method, which includes: acquiring historical passenger flow data and a first coefficient for ordinary days in the same past period of the target space during the forecast period, wherein the forecast period is located on a first holiday, and the first coefficient represents the difference between the first holiday and ordinary days in the same year before the forecast period in historical passenger flow data; and forecasting the passenger flow forecast result of the target space during the forecast period based on the historical passenger flow data and the first coefficient for ordinary days in the same past period of the target space during the forecast period.

[0006] It can be seen that the period to be predicted falls within the first holiday. The first holiday can be any type of holiday, such as the Spring Festival holiday, the May Day holiday, the National Day holiday, etc. This application embodiment does not specifically limit it. It should be understood that these holidays only occur once a year. Therefore, the historical passenger flow data of a certain type of holiday refers to the historical passenger flow data of that type of holiday in previous years during the period to be predicted. It can also be seen that the first coefficient is related to the specific type of holiday. Different types of first holidays have different first coefficients, which can distinguish various types of holidays, realize the characteristic analysis of different types of holidays, and improve the accuracy of passenger flow prediction under various types of holidays.

[0007] It should be noted that the first coefficient represents the difference between historical passenger flow data for the first holiday and historical passenger flow data for ordinary days in a past year within the period to be predicted. In other words, it is necessary to compare historical passenger flow data for holidays and ordinary days within the same year preceding the period to be predicted to determine the first coefficient. Ordinary days refer to dates other than holidays and event dates; "the same year preceding the period to be predicted" can be the previous year or the year before, without specific limitation; the "difference" can be a multiple of the historical passenger flow data for the first holiday relative to the historical passenger flow data for ordinary days, or it can be the difference between the historical passenger flow data for the first holiday and the historical passenger flow data for ordinary days, without specific limitation. This application embodiment, through the first coefficient, can correlate the passenger flow patterns of the first holiday with the passenger flow patterns of ordinary days, thereby enabling the prediction of passenger flow during the first holiday using historical passenger flow data for ordinary days, which can improve the accuracy of passenger flow prediction for the first holiday.

[0008] Based on the first aspect, in a possible embodiment, the first coefficient represents the difference between historical passenger flow data of the target space during the first holiday of the same period in the past of the period to be predicted and historical passenger flow data of the target space during the same period in the past of the period to be predicted, in the year preceding the period to be predicted.

[0009] Based on the first aspect, in a possible embodiment, the method further includes: obtaining a second coefficient, the second coefficient representing the difference in annual growth rate between historical passenger flow data of the first holiday and historical passenger flow data of ordinary days; predicting the passenger flow forecast result of the target space in the period to be predicted based on historical passenger flow data of ordinary days in the same past period of the target space in the period to be predicted and the first coefficient, including: predicting the passenger flow forecast result of the target space in the period to be predicted based on historical passenger flow data of ordinary days in the same past period of the target space in the period to be predicted, the first coefficient and the second coefficient.

[0010] It should be noted that the first coefficient is determined based on historical passenger flow data for the first holiday and ordinary days in the same year preceding the forecast period, and does not involve passenger flow growth rates between different years. The second coefficient represents the difference in annual growth rates between historical passenger flow data for the first holiday and historical passenger flow data for ordinary days. Therefore, combining the first and second coefficients allows us to take into account the difference in annual growth rates between historical passenger flow data for the first holiday and historical passenger flow data for ordinary days, which can further improve the accuracy of passenger flow forecasting.

[0011] Based on the first aspect, in a possible embodiment, the second coefficient represents the difference in annual growth rate between the historical passenger flow data of the target space during the same past period of the predicted time and the historical passenger flow data of the target space during the same past period of the predicted time on ordinary days.

[0012] Based on the first aspect, in possible embodiments, the target space includes any one or more types of stations, lines, networks, and cross-sections; ordinary daily historical passenger flow data includes one or more types of historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume; passenger flow prediction results include one or more types of predicted inbound volume, predicted outbound volume, predicted transfer volume, and predicted passenger volume.

[0013] In other words, the target space can be at the spatial granularity of stations, lines, networks, and cross-sections, without specific limitations; the embodiments of this application can provide predictions of various passenger flow indicators such as the number of passengers entering the station, the number of passengers exiting the station, the number of transfers, and the passenger volume.

[0014] Based on the first aspect, in possible embodiments, the method further includes: acquiring weather characteristics of the target space during the forecast period, the weather characteristics including one or more of temperature, rainfall, wind force, humidity, and air quality; predicting the passenger flow forecast result of the target space during the forecast period based on historical passenger flow data of ordinary days in the same past period of the target space and a first coefficient, including: predicting the passenger flow forecast result of the target space during the forecast period based on historical passenger flow data of ordinary days in the same past period of the target space, the first coefficient, and weather characteristics.

[0015] In other words, it is also possible to obtain the weather characteristics of the target space during the forecast period, and to achieve passenger flow forecasting for the target space during the forecast period based on historical passenger flow data of ordinary days, the first coefficient, and the weather characteristics. It is understood that weather conditions are an important factor affecting passenger flow; therefore, by adding weather characteristics, the embodiments of this application can further improve the accuracy of passenger flow forecasting.

[0016] Based on the first aspect, in a possible embodiment, the period to be predicted is located on the first holiday and the day of the event; the method further includes: obtaining the activity characteristics of the target space in the period to be predicted; predicting the passenger flow prediction result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same past period in the period to be predicted and a first coefficient, including: predicting the passenger flow prediction result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same past period in the period to be predicted, the first coefficient and the activity characteristics.

[0017] It can be seen that the period to be predicted not only falls within a certain type of holiday, but may also simultaneously fall within the date of an event (i.e., the event day). Therefore, in this scenario where holidays overlap with event days, it is also necessary to consider the impact of the event on passenger flow. This application embodiment improves the accuracy of passenger flow prediction by obtaining the activity characteristics of the target space during the period to be predicted and using these activity characteristics to reflect the impact of the event on passenger flow in the target space during the period to be predicted.

[0018] Based on the first aspect, in possible embodiments, the activity features include one or more of the following: date, time period, activity start time, activity end time, weekday, activity type, activity scale, time influence coefficient, and distance influence coefficient, wherein the time influence coefficient represents the degree to which the time period to be predicted is affected by the activity event in time, and the distance influence coefficient represents the degree to which the target space is affected by the activity event in distance.

[0019] As can be seen, activity characteristics can include one or more dimensions of information, which can comprehensively reflect the impact of activity events on passenger flow in the target space during the predicted period, thereby improving the accuracy of passenger flow prediction and obtaining more accurate passenger flow prediction results.

[0020] Based on the first aspect, in a possible embodiment, obtaining historical passenger flow data of the target space on ordinary days during the period to be predicted includes: obtaining origin-destination OD data; and performing path matching on the OD data based on the K-shortest path algorithm and in combination with one or more of driving time, walking time, and waiting time, thereby obtaining historical passenger flow data of the target space on ordinary days during the period to be predicted.

[0021] In other words, the embodiments of this application are based on the K-shortest path algorithm for path matching. The K-shortest path algorithm is a general term for algorithms used to solve the K shortest path problem, and its specific type is not limited. By combining one or more factors such as travel time, walking time, and waiting time to perform path matching on OD data, the embodiments of this application can improve the accuracy of path matching, thereby obtaining more accurate historical passenger flow data. It is understandable that with more accurate data, the performance of passenger flow prediction models or algorithms can be improved, helping to increase the accuracy of passenger flow prediction.

[0022] Based on the first aspect, in a possible embodiment, the path matching of OD data based on the K-shortest path algorithm and in combination with one or more of driving time, walking time, and waiting time includes: obtaining basic road network information; determining K corresponding shortest paths for OD pairs in the basic road network information based on the K-shortest path algorithm and driving time, where K is a positive integer; determining a recommended path for the OD pair from the K shortest paths of the OD pair in combination with one or more of walking time and waiting time; and matching the OD data to the recommended path of the OD pair corresponding to the OD data.

[0023] In other words, this embodiment first uses the K-shortest path algorithm and travel time to determine K shortest paths for each OD pair in the road network. Each OD pair has corresponding K shortest paths, and these K shortest paths have relatively short travel times, so passengers are more likely to choose these paths. These K shortest paths are used as the candidate path set for that OD pair. It is understood that waiting time and walking time are also important factors influencing passenger route selection. Therefore, this embodiment further combines one or more factors such as waiting time and walking time to determine a recommended path for that OD pair from its K shortest paths. Then, the OD data related to that OD pair is matched to the recommended path, thereby obtaining more accurate historical passenger flow data, which helps improve the accuracy of subsequent passenger flow prediction.

[0024] Based on the first aspect, in possible embodiments, walking time includes one or more of the following: walking time to enter the station, walking time to exit the station, and walking time for transfers; waiting time includes one or more of the following: waiting time to enter the station, waiting time for transfers, and waiting time to exit the station.

[0025] Secondly, embodiments of this application provide a passenger flow prediction device, which includes: an acquisition module, configured to acquire historical passenger flow data and a first coefficient for ordinary days in the same past period of the time to be predicted for a target space, wherein the time to be predicted is located on a first holiday, and the first coefficient represents the difference in historical passenger flow data between the first holiday and ordinary days in the same year prior to the time to be predicted; and a prediction module, configured to predict the passenger flow prediction result of the target space in the time to be predicted based on the historical passenger flow data and the first coefficient for ordinary days in the same past period of the time to be predicted for the target space.

[0026] Each module in the above-mentioned passenger flow prediction device is specifically used to implement the method of any embodiment in the first aspect.

[0027] Thirdly, embodiments of this application provide a computing device including a processor and a memory; the processor and memory can be interconnected via a bus or integrated together. The processor is used to read program code stored in the memory, so that the computing device executes the method of any of the embodiments in the first aspect described above.

[0028] Fourthly, embodiments of this application provide a computer-readable storage medium for storing implementation code of the method of any of the embodiments in the first aspect described above.

[0029] Fifthly, embodiments of this application provide a computer program (product) including program instructions that, when executed, perform the method of any of the embodiments in the first aspect described above.

[0030] In summary, this application embodiment, by acquiring historical passenger flow data and a first coefficient for ordinary days in the same past period of the target space during the period to be predicted, can predict the passenger flow of the target space during the period to be predicted and obtain the corresponding passenger flow prediction results. The first coefficient represents the difference between historical passenger flow data for the first holiday and historical passenger flow data for ordinary days in the same year prior to the period to be predicted. Therefore, the passenger flow patterns of the first holiday can be correlated with the passenger flow patterns of ordinary days, thus enabling accurate prediction of passenger flow during the first holiday using historical data for ordinary days. Furthermore, the first coefficient is related to the type of holiday, thus enabling the differentiation of various types of holidays, analysis of the different characteristics of different holidays, and thereby achieving accurate passenger flow prediction for various types of holidays.

[0031] This application embodiment can also obtain a second coefficient, which represents the difference in annual growth rate between historical passenger flow data of the first holiday and historical passenger flow data of ordinary days. Passenger flow prediction is performed based on the first coefficient and combined with the second coefficient, taking into account the difference in annual growth rate between historical passenger flow data of the first holiday and historical passenger flow data of ordinary days, which can further improve the accuracy of passenger flow prediction. This application embodiment can also obtain the weather characteristics of the target space during the prediction period, and combine the weather characteristics with passenger flow prediction to improve the accuracy of passenger flow prediction. If the prediction period not only falls on a certain type of holiday but also on the date of an event, then this application embodiment can also obtain the event characteristics of the target space during the prediction period, and combine the event characteristics with passenger flow prediction, taking into account the impact of the event on passenger flow in the target space during the prediction period, thus realizing passenger flow prediction in scenarios where holidays overlap with event days, which can improve the accuracy of passenger flow prediction.

[0032] The embodiments of this application can also be optimized in the data analysis. Based on the K-shortest path algorithm, and combined with one or more of driving time, walking time, and waiting time, path matching is performed on OD data to obtain more accurate historical passenger flow data. This improvement is made at the source of data acquisition and can enhance the effect of subsequent passenger flow prediction. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0034] Figure 1 This is a schematic diagram of the architecture of a passenger flow prediction system provided in an embodiment of this application;

[0035] Figure 2 This is a flowchart illustrating a passenger flow prediction method provided in an embodiment of this application;

[0036] Figure 3 This is a schematic diagram of an urban rail transit network provided in an embodiment of this application;

[0037] Figure 4 This is a schematic diagram of a network topology constructed based on road network infrastructure information, provided in an embodiment of this application;

[0038] Figure 5 This is a flowchart illustrating another passenger flow prediction method provided in the embodiments of this application;

[0039] Figure 6 This is a flowchart illustrating another passenger flow prediction method provided in the embodiments of this application;

[0040] Figure 7This is a flowchart illustrating another passenger flow prediction method provided in the embodiments of this application;

[0041] Figure 8 This is a schematic diagram of the structure of a passenger flow prediction device provided in an embodiment of this application;

[0042] Figure 9 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application;

[0043] Figure 10 This is a flowchart illustrating a training method for a passenger flow prediction model provided in an embodiment of this application. Detailed Implementation

[0044] To facilitate understanding of the technical solutions in the embodiments of this application, some terms and concepts involved in the embodiments of this application will be briefly introduced below.

[0045] 1. Automated fare collection (AFC) system and origin-destination (OD) data:

[0046] AFC system, or automatic fare collection system for urban rail transit, is an automated network system centrally controlled by a computer. Based on technologies such as computing, communication, networking, and automatic control, AFC system can realize functions such as automatic ticketing, ticket checking, fare calculation, collection, statistics, clearing, and management of rail transit.

[0047] The AFC (Automatic Fare Collection) system collects passenger card swipe data (either upon entry or exit) and card swipe data upon exit. It then uses two consecutive entry and exit card swipe records of the same passenger as the origin and destination of a single trip to obtain OD (Original Location) data. It should be understood that OD data is a common type of data in fields such as transportation, urban planning, and geographic information systems. Its characteristic is that each OD data record includes information such as the location and time of the origin and destination of a single trip.

[0048] 2. LightGBM (light gradient boosting machine):

[0049] LightGBM is an open-source gradient boosting framework, a machine learning algorithm based on decision trees. LightGBM primarily employs gradient-based one-side sampling (GOSS) and exclusive feature bundling (EFB) algorithms: GOSS aims to reduce the number of samples by excluding most samples with small gradients, using only the remaining samples to calculate information gain, thus balancing data reduction with accuracy; EFB fuses and binds some features, further reducing the number of features.

[0050] The passenger flow prediction system 1000 involved in the embodiments of this application is described below.

[0051] Please see Figure 1 , Figure 1 This is a schematic diagram of the architecture of a passenger flow prediction system 1000 provided in an embodiment of this application. The passenger flow prediction system 1000 includes an input system 100, a passenger flow analysis system 200, and a prediction system 300.

[0052] (1) Input system 100 to obtain the raw data information required for passenger flow forecasting. The raw data information may include OD data, road network basic information, weather data, holiday data, event data, etc.

[0053] The OD data can be provided by the AFC system. It is understood that the AFC system collects passenger card swipe data at the gate, processes it simply to obtain individual OD data, and then provides the required OD data to the input system 100. In possible embodiments, the AFC system can be located outside the input system 100, inputting corresponding OD data to the input system 100 according to its needs; the AFC system can also be directly integrated into the input system 100, which is not specifically limited in this application.

[0054] Basic road network information refers to the fundamental information of an urban rail transit network, including the number, names, and spatial relationships of stations, lines, and networks. This basic information can be provided by rail transit construction departments, operation and management departments, etc., and is not specifically limited to any particular entity.

[0055] Weather data can be provided by meteorological monitoring departments, meteorological bureaus, etc., and is not specifically limited. In possible embodiments, weather data may include information from one or more dimensions, such as temperature, rainfall, wind speed, humidity, air quality, etc.

[0056] Event data refers to the basic information of an event. For example, an event can be a concert, a music festival, a sporting event, a dance party, a celebrity meet-and-greeting, an autograph session, a themed light show, an exhibition, etc., and this application does not impose specific limitations. In possible embodiments, event data may include one or more dimensions of information, such as date, time period, event type, event scale, event venue, and event name. Event data can be obtained from event organizers, event promotional websites, etc., and is not specifically limited.

[0057] (2) Passenger flow analysis system 200 is used to analyze the raw data information obtained by input system 100 to obtain historical passenger flow data. Historical passenger flow data may include types such as historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume. It can be statistically analyzed according to spatial granularity such as station, cross section, line, and network. Taking a line as an example, there may be historical inbound volume, historical outbound volume, historical transfer volume, historical passenger volume, etc.

[0058] It's important to note that a cross-section is a relatively special concept. Simply put, it refers to two adjacent stations forming a cross-section, and it has a directional aspect. For example, stations A and B are two adjacent stations on the same transportation line. The journey from station A to station B constitutes one cross-section, and vice versa. Cross-sectional flow (or cross-sectional passenger flow / passenger volume) is a more specific type of passenger flow data. It refers to the passenger flow passing through a specific cross-section of a rail transit line per unit of time, encompassing multiple dimensions such as time, space (with directionality), and quantity (including absolute volume and occupancy rate).

[0059] (3) A prediction system 300 is used to perform spatiotemporal characteristic analysis of passenger flow (analyze the influencing factors of passenger flow), train a passenger flow prediction model, predict passenger flow based on the trained passenger flow prediction model, analyze and evaluate the passenger flow prediction results, and persistently store them, etc. In possible embodiments, the prediction system 300 can provide passenger flow predictions for scenarios such as ordinary days, holidays, and event days, at spatial granularities such as stations, lines, networks, and cross-sections, and at configurable minute-level (e.g., 5min, 60min, etc.) time granularities, including types such as inbound volume, outbound volume, transfer volume, and passenger volume.

[0060] It should be noted that the division and connection relationships of the various subsystems in the above-mentioned passenger flow prediction system 1000 are merely examples. The passenger flow prediction system 1000 may also include more or fewer subsystems / modules. For example, the same subsystem may be divided into multiple functional modules, and multiple subsystems may be merged into one subsystem. This application does not impose any specific limitations.

[0061] Based on the above-mentioned passenger flow prediction system 1000, the following describes an embodiment of the passenger flow prediction method in this application.

[0062] Please see Figure 2 , Figure 2 This is a flowchart illustrating a passenger flow prediction method provided in an embodiment of this application. The method is used for passenger flow prediction on ordinary days (i.e., days excluding holidays and event days) and includes the following steps:

[0063] S201. Obtain OD data. Based on the K-shortest path (KSP) algorithm and combined with one or more of driving time, walking time, and waiting time, perform path matching on the OD data to obtain historical passenger flow data.

[0064] The KSP algorithm is a general term for algorithms used to solve the problem of K shortest paths. The KSP algorithm can be divided into labeling algorithms, path deletion algorithms, intelligent improvement algorithms, deviation path algorithms, improved Dijkstra algorithms, etc. This application does not make specific limitations.

[0065] Waiting time can be one or more of the following: waiting time to enter the station, waiting time to transfer, and waiting time to exit the station; walking time can be one or more of the following: walking time to enter the station, walking time to exit the station, and walking time to transfer.

[0066] The aforementioned OD data can be multiple OD data entries. It should be understood that the AFC system reads the daily passenger gate card swipe data, sorts the card swipe data by card number and time, and uses two consecutive entry and exit card swipes by the same passenger as one OD data entry, thus generating a series of OD data entries.

[0067] The aforementioned historical passenger flow data can be of various types / indicators such as historical inbound volume, historical outbound volume, historical transfer volume, or historical passenger volume. It can be historical passenger flow data at spatial granularity such as station, line, network, or section. This application does not impose any specific limitations.

[0068] It should be understood that a single OD (Original Departure / Origin) data point is essentially two consecutive card swipes by a passenger upon entry and exit from the station. This is equivalent to a simple travel record, containing only basic information such as the location of the origin and destination (starting and ending stations) and the time (card swipe time upon entry and exit). The actual route taken by the passenger during this trip is not directly reflected. Therefore, path matching is needed to reconstruct the actual travel path corresponding to the OD data, thereby obtaining various historical passenger flow data. In particular, passenger flow data such as transfer volume and cross-sectional flow cannot be directly obtained from passenger card swipe data and requires precise path matching.

[0069] In a possible embodiment, step S201 may include: first, obtaining basic path information, constructing a network topology using stations in the basic road network information as nodes, intervals formed by two adjacent stations as edges, and travel time as the edge weight; then, based on the KSP algorithm and combined with factors such as walking time and waiting time, determining a recommended path for each OD pair in the network topology, matching OD data to the recommended path of the corresponding OD pair, thereby obtaining various passenger flow data. Here, an OD pair consists of any two different nodes (i.e., stations) in the network topology graph, with one node serving as the origin (O) and the other as the destination (D), exhibiting directionality.

[0070] For example, please see Figure 3 , Figure 3 This is a schematic diagram of an urban rail transit network provided in an embodiment of this application. The network includes three subway lines, denoted as Line 1, Line 2, and Line 3. Line 1 includes five stations: A, B, C, D, and E; Line 2 includes four stations: F, C, G, and H; and Line 3 includes five stations: I, J, G, D, and K. It should be noted that... Figure 3 The number, names, shapes, and location relationships of the stations and lines in the examples are merely illustrative and do not constitute a limitation on this application.

[0071] It is understandable that information can be obtained from rail transit construction departments, operation and management departments, etc. Figure 3 The basic road network information allows for abstraction of it. Figure 3 By using each station as a node (still represented by the original station names), the intervals formed by two adjacent stations as edges, and the travel time corresponding to the interval as the edge weight, a system can be constructed. Figure 4 Network topology. For example... Figure 4 As shown, there is an edge between every pair of adjacent stations, and the weight of the edge represents the travel time between the two stations. For example, station A and station B are two adjacent stations on Line 1, and the weight W of the edge between stations A and B... AB This represents the travel time required between the two stations.

[0072] The paths between any two stations in an urban rail transit network are diverse, so according to the KSP algorithm, they are respectively... Figure 4 For each OD pair, K shortest paths are calculated, where K is a positive integer and can be selected appropriately based on the actual application scenario. In other words, Figure 4For each OD pair, K shortest paths are calculated. Then, for each OD pair, based on information such as waiting time for entering / exiting / transferring and walking time for entering / exiting / transferring, a recommended path is determined from the K shortest paths corresponding to that OD pair. Thus, each OD pair has a recommended path.

[0073] It should be noted that this application does not specifically limit the method of "determining a recommended path for an OD pair from K shortest paths". For example, it can first determine whether there is a direct path (without transfers) among the K shortest paths of an OD pair. If there is a direct path, it is directly used as the recommended path. If there is no direct path, it is then judged based on the transfer waiting time required for each path, and the path with the shortest transfer waiting time among the K shortest paths is used as the recommended path for the OD pair. If there are multiple paths with the shortest transfer waiting time, it can be further judged based on the transfer walking time required for each path, and the path with the shortest transfer walking time is used as the recommended path for the OD pair. It should be understood that the walking time, waiting time, etc., corresponding to each path can be obtained based on data statistics. For example, subway staff can repeatedly count the transfer waiting time required for a certain path, and the average of the multiple statistics can be used as the transfer waiting time corresponding to this path.

[0074] Finally, the OD data provided by the AFC system is matched to the recommended paths for the corresponding OD pairs. For example, the origin and destination of a certain OD data point are respectively... Figure 3 For stations B and G in the data, this OD data can be matched to the recommended path of the corresponding OD pair (station B-station C, i.e., station B as the origin and station G as the destination), thus reconstructing the passenger's actual travel path as station B→station C→station G. Path matching for other OD data is similar, thereby enabling the collection of various historical passenger flow data.

[0075] In a possible implementation, path matching can be performed proportionally. For example, after calculating the K shortest paths for a certain OD pair, the allocation ratio of each path is determined based on factors such as the waiting time and walking time required for each path. Then, the OD data related to this OD pair is allocated to these K shortest paths according to the allocation ratio, thereby statistically analyzing various historical passenger flow data.

[0076] In a possible implementation, only the OD data related to the target space can be obtained according to the needs of passenger flow prediction, and then path matching can be performed on these OD data according to the method described above to obtain historical passenger flow data related to the target space.

[0077] It should be noted that step S201 can statistically analyze historical passenger flow data at spatial granularities such as station, line, network, and cross-section, including historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume. Based on the scenarios corresponding to the historical passenger flow data, the data can be segmented to obtain historical passenger flow data for ordinary days, holidays, event days, etc., without specific limitations. The historical passenger flow data under each scenario can be further subdivided into types such as historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume.

[0078] S202. Obtain historical data of the target space during the same period of the predicted time from the above historical passenger flow data.

[0079] The target space refers to a specific spatial location where passenger flow forecasting is needed. The target space can be at the spatial granularity level, such as a station, cross-section, line, or network, and can be selected according to the passenger flow forecasting requirements. For example, the target space can be the entire rail transit network of a city, a specific line, a specific station, a specific cross-section, etc., without specific limitations.

[0080] The period to be predicted refers to a specific time period during which passenger flow forecasting needs to be performed. For example, assuming a time window (or time period) is 60 minutes long, a day can be divided into 24 time windows, and the period to be predicted can be a specific time window within that day; assuming a time window is 5 minutes long, a day can be divided into 288 time windows, and the period to be predicted can be a specific time window within that day. Of course, time windows can also be divided only within the subway operating hours, and the length of the time window can be set to other values; this application does not impose specific limitations. It should be noted that... Figure 2 The passenger flow prediction method is based on a normal day scenario, therefore the date of the period to be predicted here is a normal day.

[0081] Historical data refers to the historical passenger flow data of the target space during the same period in the past on ordinary days. There can be multiple historical data sets. It should be understood that the historical passenger flow data for ordinary days can include one or more of the following types: historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume on ordinary days.

[0082] For example, suppose the target interval is Figure 3For station A, the predicted time period is 10-11 AM on October 15th. This predicted time period corresponds to the 11th time window of the day. The goal is to predict the inbound traffic of station A during this predicted time period. Assuming we use historical data from the past 5 days (all ordinary days) for the same period, the historical inbound traffic of station A during the predicted time period is the historical inbound traffic of station A in the 11th time window of the 5 days prior to October 15th. Specifically, this includes the historical inbound traffic of station A during 10-11 AM on October 10th, October 11th, October 12th, October 13th, and October 14th. It should be noted that the above example selects historical data from the past 5 days of the period to be predicted. This is just an example and does not constitute a limitation. For example, you can select historical data from each day of the N days before the period to be predicted, or you can select only historical data from the M days of the N days before the period to be predicted, where N is a positive integer and M is a positive integer less than N.

[0083] S203. Construct basic features of the target space in the period to be predicted based on historical data from the same period. The basic features include one or more of the following: historical data from the same period, scaled-down historical data from the same period, and moving averages of historical data from the same period.

[0084] In other words, the aforementioned basic features may include features of one or more dimensions. Historical data from the same period, scaled-down versions of historical data from the same period, and moving averages of historical data from the same period can all be used as part of the basic features.

[0085] Here, "scaled historical data" refers to the result obtained after scaling the historical data from the same period. This application does not limit the specific method of data scaling. For example, it could be an interval scaling method, which maps the historical data from the same period to the interval [0,1] through linear transformation, thus avoiding obtaining a weight value that is significantly different from other features.

[0086] The moving average of historical data refers to the data obtained by applying a moving average (MA) processing to historical data from the same period. It should be understood that a moving average is a technical analysis tool. If a set of data is obtained, and a certain number of data points are taken sequentially and their arithmetic mean is calculated, the resulting data is called the moving average. Moving averages include categories such as simple moving average (SMA), weighted moving average (WMA), and exponential moving average (EMA), which are not specifically limited in this application.

[0087] Following the example in step S202, assume that the historical inbound traffic of station A during the 11th time window from October 10th to October 14th was 500, 620, 800, 710, and 880 respectively. Assuming a simple moving average is used, with 3 terms T, by taking the average of 3 numbers sequentially from these historical data points for the same period, we can obtain... These moving averages, namely 640, 710, and 800 respectively, can then be used to construct the underlying features.

[0088] It should be noted that, in addition to using methods such as data scaling and calculating moving averages to construct basic features, other feature engineering methods can also be used to process historical data from the same period to construct basic features. This application does not specifically limit the methods.

[0089] In a possible embodiment, the basic characteristics of the target space during the period to be predicted may further include: historical data of related spaces of the target space during the same period. For example, if the target space is station B, the related spaces of the target space may be adjacent stations of station B, the line where station B is located, etc., and then the historical data of related spaces during the same period of the period to be predicted are used as part of the basic characteristics of the target space during the period to be predicted.

[0090] S204. Obtain the weather characteristics of the target space during the forecast period.

[0091] Weather characteristics refer to weather conditions and may include information from one or more dimensions. For example, weather characteristics may include one or more of temperature, rainfall, wind speed, humidity, and air quality; this application does not impose specific limitations. It is understood that by obtaining weather data for the target space during the forecast period from meteorological monitoring departments, meteorological bureaus, etc., the weather characteristics of the target space during the forecast period can be determined.

[0092] It should be noted that the execution order of step S204 is not specifically limited in the embodiments of this application. Step S204 can be executed before steps S201, S202 or S203, or it can be executed simultaneously with one of these steps.

[0093] S205. Input the basic characteristics and weather characteristics of the target space during the forecast period into the passenger flow forecast model to obtain the passenger flow forecast results of the target space during the forecast period.

[0094] The passenger flow prediction model here is a passenger flow prediction model for a normal day, which can be a LightGBM model, a gradient boosting decision tree (GBDT) model, a neural network-based model, etc., and this application does not make specific limitations. The passenger flow prediction results can include one or more types of predicted inbound volume, predicted outbound volume, predicted transfer volume, and predicted passenger volume. It should be understood that the type of passenger flow prediction results can correspond to the type of historical data obtained in step S202.

[0095] It should be noted that after obtaining the passenger flow forecast results, they can be analyzed, evaluated, and persistently stored. For example, the passenger flow forecast results can be stored on a hard drive or in a database, without specific limitations.

[0096] In summary, this application first uses the KSP algorithm combined with factors such as travel time, waiting time, and walking time to perform accurate path matching on OD data. It analyzes and processes the OD data both spatially and temporally to obtain accurate historical passenger flow data. It is understood that good data and features are prerequisites for models and algorithms to function effectively. The path matching method used in this application can obtain relatively accurate historical passenger flow data, thereby improving the accuracy of subsequent passenger flow prediction. Next, historical data for the same period in the target space during the prediction period is obtained from the aforementioned historical passenger flow data, and basic features are constructed based on this historical data. This application also considers the impact of weather factors on passenger flow. By obtaining the weather characteristics of the target space during the prediction period, and inputting the basic features and weather characteristics together into the passenger flow prediction model, the passenger flow prediction result for the target space during the prediction period can be obtained. Accurate passenger flow prediction results allow for the prediction of passenger flow trends, enabling reasonable train scheduling based on passenger flow fluctuations, promoting a balanced distribution of passenger flow across lines, meeting the travel needs of citizens, and avoiding problems such as peak-hour congestion and off-peak waste.

[0097] Please see Figure 5 , Figure 5 This is a flowchart illustrating another passenger flow prediction method provided in this application embodiment. This method is used for passenger flow prediction during holidays (i.e., statutory holidays) and includes the following steps:

[0098] S501. Obtain OD data. Based on the KSP algorithm and combined with one or more of driving time, walking time, and waiting time, perform path matching on the OD data to obtain historical passenger flow data.

[0099] For details of step S501, please refer to the description in step S201 above, which will not be repeated here.

[0100] S502. Obtain historical data of the target space during the same period of the predicted time from the above historical passenger flow data.

[0101] The target space refers to a specific spatial location where passenger flow forecasting is required. The target space can be at the spatial granularity level, such as a station, cross-section, line, or network, and can be selected based on passenger flow forecasting needs without specific limitations.

[0102] The period to be predicted refers to a specific time period during which passenger flow forecasting needs to be performed. This period falls within the first holiday. The first holiday can be any type of holiday, such as the Spring Festival, Qingming Festival, May Day, Dragon Boat Festival, Mid-Autumn Festival, or National Day, without specific limitations. For example, if the period to be predicted is 10:00-11:00 AM on May 1, 2021, then the date falls on the first day of the May Day holiday, which fits the holiday scenario.

[0103] Historical data refers to the historical passenger flow data of the target space during the same period in the past on ordinary days. There can be multiple historical data sets. It should be noted that the historical passenger flow data for ordinary days can include one or more of the following types: historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume on ordinary days.

[0104] For example, suppose the target interval is Figure 3 For station C, the predicted time period is 10-11 AM on May 1, 2021. The time window corresponding to this predicted time period is the 11th time window of the day. Now, we need to predict the inbound traffic of station A during the above-mentioned predicted time period. Obviously, the above-mentioned predicted time period falls within the May Day holiday, which is consistent with the holiday scenario.

[0105] Assuming we take the historical data for the same period of the past 3 days (all ordinary days) from May 1st, then the historical data for the same period of the above-mentioned period to be predicted for site A are the historical inbound traffic of site A in the 11th time window of the 3 days before October 15th, that is, the historical inbound traffic of site A from 10:00 AM to 11:00 AM on April 28, 2021, the historical inbound traffic of site A from 10:00 AM to 11:00 AM on April 29, 2021, and the historical inbound traffic of site A from 10:00 AM to 11:00 AM on April 30, 2021.

[0106] For further details regarding step S502, please refer to the relevant description in the aforementioned step S202; it will not be elaborated upon here.

[0107] It should be noted that, in possible embodiments, the above steps S501 and S502 can also be directly combined into one step: obtain OD data, perform path matching on the OD data based on the KSP algorithm and in combination with one or more of driving time, walking time, and waiting time, so as to obtain the historical data of the target space in the same period of the period to be predicted, that is, the historical passenger flow data of the target space on ordinary days in the same period of the past in the period to be predicted.

[0108] S503. Construct basic features of the target space in the period to be predicted based on historical data from the same period. The basic features include one or more of the following: historical data from the same period, scaled-down historical data from the same period, and moving averages of historical data from the same period.

[0109] For details of step S503, please refer to the description in step S203 above, which will not be repeated here.

[0110] S504. Obtain the first coefficient and the second coefficient. The first coefficient represents the difference in historical passenger flow data between the first holiday and ordinary days in the same year before the period to be predicted. The second coefficient represents the difference in annual growth rate between the historical passenger flow data of the first holiday and the historical passenger flow data of ordinary days.

[0111] It should be noted that both the first and second coefficients are related to the specific type of holiday. Different types of holidays have different first and second coefficients, which can distinguish various types of holidays and enable characteristic analysis of different types of holidays. Even for the same holiday, the first (or second) coefficients corresponding to different types of historical passenger flow data (historical inbound volume, historical outbound volume, historical transfer volume, or historical passenger volume) may differ. In other words, the first and second coefficients are also related to the specific type of historical passenger flow data.

[0112] Specifically, the first coefficient represents the difference between historical passenger flow data for the first holiday and historical passenger flow data for ordinary days in a past year within the period to be predicted. In other words, it requires comparing historical passenger flow data for holidays and ordinary days within the same year preceding the period to be predicted to obtain the first coefficient. The "same year preceding the period to be predicted" can be the year before the year in which the period to be predicted is located, or it can be a year earlier; this application does not specifically limit this. The "difference" can be a multiple of the historical passenger flow data for the first holiday relative to the historical passenger flow data for ordinary days, or it can be the difference between the historical passenger flow data for the first holiday and the historical passenger flow data for ordinary days; this is also not specifically limited. The second coefficient represents the difference in annual growth rate between passenger flow on the first holiday and passenger flow on ordinary days. Therefore, it requires historical passenger flow data for holidays and ordinary days from several years preceding the period to be predicted to determine the second coefficient. It can use historical passenger flow data from two years preceding the period to be predicted, or it can use historical passenger flow data from more past years; this is not specifically limited.

[0113] For example, assuming the predicted period is 10-11 AM on May 1, 2021, and the first coefficient is determined using historical passenger flow data from the year preceding the predicted period, then the historical passenger flow data for the 2020 May Day holiday can be compared with the historical passenger flow data for ordinary days in 2020 to obtain the first coefficient. For instance, the historical passenger flow data for each day of the 2020 May Day holiday can be averaged to obtain the average daily historical passenger flow for that route during the 2020 May Day holiday; then, the historical passenger flow data for some or all ordinary days of 2020 can be averaged to obtain the average daily historical passenger flow for that route during ordinary days in 2020; finally, by comparing the average daily historical passenger flow for the 2020 May Day holiday with the average daily historical passenger flow for ordinary days in 2020, the first coefficient corresponding to the May Day holiday can be determined.

[0114] Assuming the second coefficient is determined using historical passenger flow data from the two years preceding the period to be predicted, the average historical passenger flow data for each day of the 2019 May Day holiday can be calculated to obtain the average daily historical passenger flow for that holiday. Then, the average historical passenger flow data for some or all of the regular days in 2019 can be calculated to obtain the average daily historical passenger flow for that regular day in 2019. Comparing the average daily historical passenger flow for the 2020 May Day holiday with that of 2019 yields the annual growth rate of historical passenger flow data for the May Day holiday. Similarly, comparing the average daily historical passenger flow for regular days in 2020 with that of 2019 yields the annual growth rate of historical passenger flow data for regular days. Finally, comparing the annual growth rate of historical passenger flow data for the May Day holiday with that of regular days determines the second coefficient corresponding to the May Day holiday.

[0115] It should be understood that the first coefficient can be either a multiple of the historical passenger flow data for the first holiday relative to the historical passenger flow data for ordinary days, or the difference between the historical passenger flow data for the first holiday and the historical passenger flow data for ordinary days; this application does not specifically limit this. The second coefficient can be either a multiple or the difference between the annual growth rate of the historical passenger flow data for the first holiday and the annual growth rate of the historical passenger flow data for ordinary days; this is also not specifically limited.

[0116] In one possible embodiment, the first coefficient represents the difference between historical passenger flow data of the target space during the first holiday of the same period in the past year of the period to be predicted and historical passenger flow data of the target space during the same period on ordinary days in the past year of the period to be predicted.

[0117] In one possible embodiment, the two coefficients represent the difference in annual growth rate between historical passenger flow data of the target space during the same past holiday period and historical passenger flow data of the target space during the same past ordinary day period.

[0118] The following section uses the May Day holiday as an example to further introduce possible schemes for extracting the first and second coefficients.

[0119] Following the example in step S502, we now need to predict the number of passengers entering station A between 10:00 AM and 11:00 AM on May 1, 2021. First, we obtain the historical passenger volume of station A during the same period last year (May Day), specifically the historical passenger volume of station A on May 1, 2020 (Friday) between 10:00 AM and 11:00 AM, let's assume it's 1000. Next, we obtain the historical passenger volumes of station A during the same period three Fridays (assuming they are all ordinary days) before May 1, 2020, which are 600, 680, and 610 respectively. We then average these figures to obtain the first average value of 630. Finally, we compare this first average value of 630 with the historical passenger volume of 1000 during the same period last year (May Day), to obtain the first coefficient a1_rate. a1_rate represents the difference in historical passenger flow data between the May Day holiday and ordinary days in the year preceding the period to be predicted.

[0120] We obtain the historical visitor volume for site A during the same period of the previous year's May Day holiday, specifically the visitor volume for site A between 10:00 AM and 11:00 AM on May 1st, 2019 (Wednesday), let's assume it's 800. Then we obtain the historical visitor volume for site A during the same period of the previous three Wednesdays (assuming they are all ordinary days) between 10:00 AM and 11:00 AM, which are 500, 600, and 700 respectively. We calculate the average of these values ​​to obtain a second average of 600. Comparing the first average of 630 and the second average of 600, we can obtain the annual growth rate of historical visitor volume on ordinary days. By comparing the historical passenger volume of station A between 10:00 AM and 11:00 AM on May 1, 2019, with the historical passenger volume of station A between 10:00 AM and 11:00 AM on May 1, 2020, we can obtain the annual growth rate of historical passenger volume during the May Day holiday. Comparing the annual growth rate of historical passenger volume on ordinary days with the annual growth rate of historical passenger volume during the May Day holiday, we can obtain the second coefficient a2_rate as 20% (i.e., 25% - 5%). a2_rate represents the difference in the annual growth rate between the historical passenger volume during the May Day holiday and the historical passenger volume on ordinary days.

[0121] It should be noted that the above example is merely an illustration, and this application does not specifically limit the method for extracting the first and second coefficients corresponding to different types of holidays. It should also be noted that the order of steps S504 is not limited; step S504 can be performed before or simultaneously with steps S501, S502, or S503.

[0122] S505. Input the basic characteristics, first coefficient and second coefficient of the target space in the period to be predicted into the passenger flow prediction model to obtain the passenger flow prediction result of the target space in the period to be predicted.

[0123] The passenger flow prediction model here refers to a holiday passenger flow prediction model, which can be a LightGBM model, a GBDT model, a neural network model, etc., and this application does not impose specific limitations. The passenger flow prediction results can include one or more types of predicted inbound volume, predicted outbound volume, predicted transfer volume, and predicted passenger volume, which can correspond to the types of historical data obtained in step S502. It is understood that after obtaining the passenger flow prediction results, they can also be analyzed, evaluated, and persistently stored. For example, the passenger flow prediction results can be stored on a hard drive or in a database, without specific limitations. Depending on the predicted holiday period, the passenger flow prediction model can be divided into different models. For example, the first / last day passenger flow prediction model is used to predict passenger flow data on the first / last day of a holiday, and the mid-holiday passenger flow prediction model is used to predict passenger flow data excluding the first and last days of a holiday. The days before and after the holiday can also be considered as holidays for passenger flow prediction, establishing a passenger flow prediction model for predicting the days before / after the holiday.

[0124] It should be noted that before step S505, Figure 5 The passenger flow forecasting method may further include: obtaining the weather characteristics of the target space during the forecast period; therefore, step S505 may also be: inputting the basic characteristics, weather characteristics, first coefficient, and second coefficient of the target space during the forecast period into the passenger flow forecasting model to obtain the passenger flow forecasting result of the target space during the forecast period. Regarding weather characteristics, please refer to the relevant content in the aforementioned step S204, which will not be repeated here.

[0125] In summary, relative to Figure 2 A method for predicting passenger flow in a typical daily scenario, Figure 5In the passenger flow forecasting method for holiday scenarios, the basic features input into the passenger flow forecasting model can also include one or more of the following: historical data for the same period, scaled-down historical data for the same period, and moving averages of historical data for the same period, approximating the forecasting of holidays as ordinary days. However, this method also obtains the first and second coefficients corresponding to the first holiday, using these two coefficients to distinguish different types of holidays and analyze their different characteristics. Therefore, this passenger flow forecasting model can adapt to passenger flow forecasting under various types of holidays. Furthermore, by using the first and second coefficients, the passenger flow patterns of the first holiday can be correlated with those of ordinary days, allowing the prediction of passenger flow under the first holiday to be made using historical data from ordinary days. This method can also consider the impact of weather factors on passenger flow. By obtaining the weather characteristics of the target space during the forecast period, and then inputting the basic features, the first coefficient, the second coefficient, and the weather characteristics together into the passenger flow forecasting model, the accuracy of passenger flow forecasting can be improved, obtaining accurate passenger flow forecast results. This provides accurate data support for urban rail transit operators, enabling them to formulate reasonable train scheduling plans and avoid passenger congestion and delays caused by sudden surges in passenger flow during holidays.

[0126] Understandably, with the first and second coefficients, the passenger flow forecasting model for ordinary days can be directly used as the base model for the passenger flow forecasting model for holidays. Specifically, the holiday passenger flow forecasting model first calls the ordinary day passenger flow forecasting model, inputting the basic features of the target space during the forecast period (located on the first holiday) into the ordinary day passenger flow forecasting model. This approximates treating the holiday as an ordinary day for forecasting, and an output result can be obtained from the ordinary day passenger flow forecasting model. Then, the holiday passenger flow forecasting model corrects this output result using the first and second coefficients, thus obtaining the passenger flow forecast result for the target space during the forecast period. Of course, in addition to basic features, other features such as weather characteristics can also be input into the ordinary day passenger flow forecasting model; there are no specific limitations.

[0127] Please see Figure 6 , Figure 6 This is a flowchart illustrating another passenger flow prediction method provided in this application embodiment. This method is used for passenger flow prediction in an event day (i.e., the day the event occurs) scenario and includes the following steps:

[0128] S601. Obtain OD data. Based on the KSP algorithm and combined with one or more of driving time, walking time, and waiting time, perform path matching on the OD data to obtain historical passenger flow data.

[0129] For details of step S601, please refer to the description in step S201 above, which will not be repeated here.

[0130] S602. Obtain historical data of the target space during the same period of the predicted time from the above historical passenger flow data.

[0131] The target space refers to a specific spatial location where passenger flow forecasting is required. The target space can be at the spatial granularity level, such as a station, cross section, line, or network, and can be selected based on passenger flow forecasting needs.

[0132] The period to be predicted refers to a specific time period during which passenger flow forecasting is required. It should be noted that... Figure 5 The visitor flow prediction method is based on the event day scenario; therefore, the date of the period to be predicted should be the date of a specific event. Events can include concerts, music festivals, sporting events, dance parties, celebrity meet-and-greets, autograph sessions, themed light shows, exhibitions, etc., and this application does not impose any specific limitations.

[0133] For example, suppose a concert will be held in city X from 7 to 9 pm on December 31, 2021. Now we need to predict the outbound traffic of station Y in that city from 6 to 7 pm on December 31, 2021. Obviously, the date of the predicted time period is the same as the concert, which is consistent with the prediction under the event day scenario.

[0134] Historical data refers to the passenger flow data of the target space during the same past period of the period to be predicted. This historical passenger flow data can be historical passenger flow data of ordinary days or historical passenger flow data of event days, and can be of various types such as historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume, without specific limitations.

[0135] For further details regarding step S602, please refer to the relevant description in step S202 above, which will not be repeated here.

[0136] S603. Construct basic features of the target space in the period to be predicted based on historical data from the same period. The basic features include one or more of the following: historical data from the same period, scaled-down historical data from the same period, and moving average of historical data from the same period.

[0137] For details of step S603, please refer to the description in step S203 above, which will not be repeated here.

[0138] S604. Obtain the activity characteristics of the target space during the period to be predicted.

[0139] The above-mentioned activity characteristics may include one or more dimensions of information, such as date, time period, activity start time, activity end time, number of weeks, activity type, activity scale / number of people, time influence coefficient, distance influence coefficient, or one or more of these.

[0140] The time period refers to the time period (or time window) in which the event to be predicted occurs, or the time period in which the event occurs.

[0141] The week number refers to which day of the week the period to be predicted falls within.

[0142] The time impact coefficient represents the degree to which the predicted time period is affected by the event. Understandably, the degree of impact varies depending on the proximity to the event's start / end time. For example, if the predicted time period is before (or close to) the event's start time, the closer the predicted time period is to the start time, the greater the impact on the exit volume of subway stations near the event venue. Conversely, if the predicted time period is after (or close to) the event's end time, and is within one hour of the event's end time, the impact on the entry volume of stations near the event venue will be significant; if the predicted time period is more than one hour before the event's end time, the impact on the entry volume of stations near the event venue will be relatively small.

[0143] The distance impact coefficient represents the degree to which a target space is affected by an event in terms of distance. Understandably, the degree of impact varies depending on the distance of a station from the event venue. For example, the closer a station is to the event venue, the greater the impact; the farther away, the less impact. Of course, the scale of the event also affects passenger flow to some extent; the larger the event, the greater the impact on passenger flow at nearby stations and along routes.

[0144] It is understandable that the activity characteristics of the target space during the period to be predicted can be determined through activity data.

[0145] For example, Table 1 contains event data for a certain concert obtained from the event promotion webpage, including event name, event type, date, number of days of the week, event venue, event scale, event start time, and event end time.

[0146] Table 1 Concert Event Data

[0147]

[0148] Suppose that the stations affected by this event are station G, station B, and station C. Station G is closest to the event venue, station B is next, and station C is farthest from the event venue. Therefore, based on the distance between the stations and the event venue, the distance influence coefficients of these three stations can be determined as 1, 2, and 3 respectively (these values ​​are only examples).

[0149] Suppose we want to predict the outbound traffic of station G from 6 PM to 7 PM on the day of the event. Since the predicted time period is close to the event start time, within one hour, the time influence coefficient can be set to 1. Suppose we want to predict the outbound traffic of station G from 5 PM to 6 PM on the day of the event. Since the predicted time period is far from the event start time, the time influence coefficient can be set to 2.

[0150] If we want to predict the number of passengers leaving Line 1 between 6 PM and 7 PM on the day of the event, we can use the sum of the time influence coefficients and the sum of the distance influence coefficients of stations B and C on Line 1 as the corresponding time influence coefficients and distance influence coefficients for Line 1, respectively. This application does not specifically limit this.

[0151] It can be understood that, based on the activity data in Table 1, it is easy to determine the characteristics of the basic features, such as date, time period, activity start time, activity end time, number of weeks, activity type, and activity scale.

[0152] S605. Input the basic features and activity features of the target space during the period to be predicted into the passenger flow prediction model to obtain the passenger flow prediction results of the target space during the period to be predicted.

[0153] The passenger flow prediction model here refers to the passenger flow prediction model for the event day, which can be a LightGBM model, a GBDT model, a neural network model, etc., and this application does not impose specific limitations. The passenger flow prediction results can include one or more types of predicted inbound volume, predicted outbound volume, predicted transfer volume, and predicted passenger volume, which can correspond to the type of historical data obtained in step S602. It is understood that after obtaining the passenger flow prediction results, they can also be analyzed, evaluated, and persistently stored. For example, the passenger flow prediction results can be stored on a hard drive or in a database, without specific limitations.

[0154] In a possible embodiment, prior to step S605, Figure 6 The passenger flow forecasting method may further include: obtaining the weather characteristics of the target space during the forecast period; therefore, step S605 may also be: inputting the basic characteristics, activity characteristics, and weather characteristics of the target space during the forecast period into the passenger flow forecasting model to obtain the passenger flow forecasting results of the target space during the forecast period. Regarding weather characteristics, please refer to the relevant content in the aforementioned step S204, which will not be repeated here.

[0155] In a possible embodiment, the period to be predicted may not only fall on an activity day, but also on a holiday; that is, there may be a scenario where a holiday overlaps with an activity day. In this scenario, before step S605, the method may further include: obtaining a first coefficient and a second coefficient; thus, step S605 may also be: inputting the basic characteristics, activity characteristics, first coefficient, and second coefficient of the target space in the period to be predicted, into the passenger flow prediction model, to obtain the passenger flow prediction result of the target space in the period to be predicted. Regarding the first coefficient and the second coefficient, please refer to the relevant content in the aforementioned step S504, which will not be repeated here.

[0156] In summary, relative to Figure 2 A method for predicting passenger flow in a typical daily scenario, Figure 6 In passenger flow prediction methods for event days, the basic features input to the passenger flow prediction model can include one or more of the following: historical data from the same period, scaled-down historical data from the same period, and moving averages of historical data from the same period. However, this method also incorporates the activity characteristics of the target space during the prediction period, considering the impact of event activities on passenger flow, which can improve the accuracy of passenger flow prediction. This method can also consider the impact of weather factors on passenger flow. By acquiring the weather characteristics of the target space during the prediction period and then inputting the basic features, activity features, and weather features together into the passenger flow prediction model, the accuracy of passenger flow prediction can be improved, resulting in precise passenger flow predictions. This allows for the development of reasonable train scheduling plans, avoiding passenger congestion and delays caused by sudden surges in passenger flow during event days. In particular, in scenarios where holidays overlap with event days, passenger flow prediction can be performed by combining the first and second coefficients.

[0157] Please see Figure 7 , Figure 7 This is a flowchart illustrating another passenger flow prediction method provided in an embodiment of this application, which may include the following steps:

[0158] S701. Obtain historical passenger flow data and a first coefficient for ordinary days in the same past period of the target space within the period to be predicted. The period to be predicted is located on the first holiday. The first coefficient represents the difference in historical passenger flow data between the first holiday and ordinary days in the same year before the period to be predicted.

[0159] For information on the target space, the period to be predicted, and the first holiday, please refer to the relevant content in step S502 above, which will not be elaborated on here.

[0160] In one possible embodiment, the first coefficient represents the difference between historical passenger flow data of the target space during the first holiday of the same period in the past year of the period to be predicted and historical passenger flow data of the target space during ordinary days of the same period in the past year of the period to be predicted. For details regarding the first coefficient, please refer to the relevant content of step S504, which will not be elaborated upon here.

[0161] In one possible embodiment, the target space includes any one or more types of stations, lines, networks, and cross-sections; historical passenger flow data includes one or more types of historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume.

[0162] In one possible embodiment, obtaining historical passenger flow data of the target space on ordinary days in the same past period of the time to be predicted may include: obtaining origin-destination OD data; and performing path matching on the OD data based on the K-shortest path algorithm and in combination with one or more of driving time, walking time, and waiting time, thereby obtaining historical passenger flow data of the target space on ordinary days in the same past period of the time to be predicted.

[0163] In one possible embodiment, the path matching of the OD data based on the K-shortest path algorithm combined with one or more of driving time, walking time, and waiting time includes: obtaining basic road network information; determining K corresponding shortest paths for OD pairs in the basic road network information based on the K-shortest path algorithm and driving time, where K is a positive integer; determining a recommended path for the OD pair from the K shortest paths of the OD pair by combining one or more of walking time and waiting time; and matching the OD data to the recommended path of the OD pair corresponding to the OD data.

[0164] In one possible embodiment, walking time includes one or more of the following: walking time to the station, walking time to the station, and walking time for transfers; waiting time includes one or more of the following: waiting time to the station, waiting time for transfers, and waiting time to the station.

[0165] It should be noted that for details regarding the three embodiments described above, please refer to the relevant parts of steps S201 and S202 above, which will not be elaborated upon here.

[0166] S702. Based on the historical passenger flow data of the target space during the same period of the past ordinary days and the first coefficient, predict the passenger flow forecast result of the target space during the period to be predicted.

[0167] In possible embodiments, passenger flow forecast results include one or more of the following: predicted arrival volume, predicted departure volume, predicted transfer volume, and predicted passenger volume.

[0168] In a possible embodiment, basic features of the target space during the predicted period are constructed based on historical passenger flow data of ordinary days in the same past period. These basic features and a first coefficient are then input into the passenger flow prediction model to obtain the passenger flow prediction result for the target space during the predicted period. The passenger flow prediction model can be a LightGBM model, a gradient boosting decision tree (GBDT) model, a neural network-based model, etc., and this application does not impose specific limitations. For details on the construction of basic features, please refer to the relevant content in step S203 above; further details are omitted here.

[0169] In a possible embodiment, step S701 may further include: obtaining a second coefficient, the second coefficient representing the difference in annual growth rate between historical passenger flow data of the first holiday and historical passenger flow data of ordinary days; thus step S702 may also be: predicting the passenger flow forecast result of the target space in the period to be predicted based on the historical passenger flow data of ordinary days in the same period in the past of the period to be predicted, the first coefficient and the second coefficient.

[0170] In a possible embodiment, the second coefficient represents the difference in annual growth rate between the historical passenger flow data of the target space during the same past period of the forecast period (on a first holiday) and the historical passenger flow data of the target space during the same past period of the forecast period (on a regular day). For details regarding the second coefficient, please refer to the relevant content of step S504; it will not be elaborated upon here.

[0171] In a possible embodiment, step S701 may further include: acquiring weather characteristics of the target space during the forecast period, whereby the weather characteristics include one or more of temperature, rainfall, wind force, humidity, and air quality; thus, step S702 may also be: predicting the passenger flow forecast result of the target space during the forecast period based on historical passenger flow data of ordinary days in the same past period of the target space, the first coefficient, and the weather characteristics. Regarding weather characteristics, please refer to the relevant content in step S204 above, which will not be elaborated upon here.

[0172] In a possible embodiment, step S702 may also be: predicting the passenger flow forecast result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same period in the past, the first coefficient, the second coefficient and weather characteristics.

[0173] In a possible embodiment, the period to be predicted is located on the first public holiday and the day of the event; therefore, step S701 may further include: obtaining the activity characteristics of the target space during the period to be predicted; step S702 may further include: predicting the passenger flow forecast result of the target space during the period to be predicted based on the historical passenger flow data of the target space on ordinary days of the same past period during the period to be predicted, the first coefficient, and the activity characteristics. Regarding the activity characteristics, please refer to the relevant part of step S604, which will not be elaborated upon here.

[0174] In possible embodiments, the activity features include one or more of the following: date, time period, activity start time, activity end time, weekday, activity type, activity scale, time influence coefficient, and distance influence coefficient, wherein the time influence coefficient represents the degree to which the time period to be predicted is affected by the activity event in time, and the distance influence coefficient represents the degree to which the target space is affected by the activity event in distance.

[0175] In a possible embodiment, step S702 may also be: predicting the passenger flow forecast result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same period in the past during the period to be predicted, the first coefficient, the second coefficient, and the activity characteristics.

[0176] In a possible embodiment, step S702 may also be: predicting the passenger flow forecast result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, weather characteristics and activity characteristics.

[0177] In a possible embodiment, step S702 may also be: predicting the passenger flow forecast result of the target space in the period to be predicted based on the historical passenger flow data of the target space in the same period of the past ordinary days, the first coefficient, the second coefficient, weather characteristics and activity characteristics.

[0178] In summary, this application embodiment, by acquiring historical passenger flow data and a first coefficient for ordinary days in the same past period of the target space during the period to be predicted, can predict the passenger flow of the target space during the period to be predicted and obtain the corresponding passenger flow prediction results. The first coefficient represents the difference between historical passenger flow data for the first holiday and historical passenger flow data for ordinary days in the same year prior to the period to be predicted. Therefore, the passenger flow patterns of the first holiday can be correlated with the passenger flow patterns of ordinary days, thus enabling accurate prediction of passenger flow during the first holiday using historical data of ordinary days. Furthermore, the first coefficient is related to the type of holiday, thus enabling the differentiation of various types of holidays, analysis of the different characteristics of different holidays, and thereby achieving accurate passenger flow prediction for various types of holidays.

[0179] This application embodiment can also obtain a second coefficient, which represents the difference in annual growth rate between historical passenger flow data of the first holiday and historical passenger flow data of ordinary days. Passenger flow prediction is performed based on the first coefficient and combined with the second coefficient, taking into account the difference in annual growth rate between historical passenger flow data of the first holiday and historical passenger flow data of ordinary days, which can further improve the accuracy of passenger flow prediction. This application embodiment can also obtain the weather characteristics of the target space during the prediction period, and combine the weather characteristics with passenger flow prediction to improve the accuracy of passenger flow prediction. If the prediction period not only falls on a certain type of holiday but also on the date of an event, then this application embodiment can also obtain the event characteristics of the target space during the prediction period, and combine the event characteristics with passenger flow prediction, taking into account the impact of the event on passenger flow in the target space during the prediction period, thus realizing passenger flow prediction in scenarios where holidays overlap with event days, which can improve the accuracy of passenger flow prediction.

[0180] The embodiments of this application can also be optimized in the data analysis. Based on the K-shortest path algorithm, and combined with one or more of driving time, walking time, and waiting time, path matching is performed on OD data to obtain more accurate historical passenger flow data. This improvement is made at the source of historical passenger flow data acquisition, which can improve the effect of subsequent passenger flow prediction.

[0181] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of a passenger flow prediction device 800 provided in an embodiment of this application. The passenger flow prediction device 800 includes an acquisition module 801 and a prediction module 802.

[0182] The acquisition module 801 is used to acquire historical passenger flow data and a first coefficient for ordinary days in the same past period of the target space during the period to be predicted. The period to be predicted is located during the first holiday, and the first coefficient represents the difference in historical passenger flow data between the first holiday and ordinary days in the same year before the period to be predicted.

[0183] The prediction module 802 is used to predict the passenger flow forecast result of the target space during the period to be predicted based on the historical passenger flow data of ordinary days in the same period of the past and the first coefficient.

[0184] In one possible embodiment, the first coefficient mentioned above represents the difference between the historical passenger flow data of the target space during the first holiday of the same period in the past year of the period to be predicted and the historical passenger flow data of the target space during the same period on ordinary days in the past year of the period to be predicted.

[0185] In one possible embodiment, the acquisition module 801 is further configured to: acquire a second coefficient, the second coefficient representing the difference in annual growth rate between historical passenger flow data of the first holiday and historical passenger flow data of ordinary days; the prediction module 802 is further configured to: predict the passenger flow prediction result of the target space in the period to be predicted based on the historical passenger flow data of ordinary days in the same period in the past of the period to be predicted, the first coefficient and the second coefficient.

[0186] In one possible embodiment, the second coefficient represents the difference in annual growth rate between historical passenger flow data of the target space during the same past holiday period in the period to be predicted and historical passenger flow data of the target space during the same past ordinary day in the same past period in the prediction period.

[0187] In one possible embodiment, the target space includes any one or more types of stations, lines, networks, and cross-sections; historical passenger flow data includes one or more types of historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume; and passenger flow prediction results include one or more types of predicted inbound volume, predicted outbound volume, predicted transfer volume, and predicted passenger volume.

[0188] In one possible embodiment, the acquisition module 801 is further configured to: acquire the weather characteristics of the target space during the forecast period, the weather characteristics including one or more of temperature, rainfall, wind force, humidity, and air quality; the prediction module 802 is further configured to: predict the passenger flow forecast result of the target space during the forecast period based on the historical passenger flow data of ordinary days in the same past period of the target space during the forecast period, the first coefficient, and the weather characteristics.

[0189] In one possible embodiment, the period to be predicted is located on the first holiday and the day of the event; the acquisition module 801 is further configured to: acquire the activity characteristics of the target space in the period to be predicted; the prediction module 802 is further configured to: predict the passenger flow prediction result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient and the activity characteristics.

[0190] In one possible embodiment, the activity features include one or more of the following: date, time period, activity start time, activity end time, weekday, activity type, activity scale, time influence coefficient, and distance influence coefficient. The time influence coefficient represents the degree to which the time period to be predicted is affected by the activity event in time, and the distance influence coefficient represents the degree to which the target space is affected by the activity event in distance.

[0191] In one possible embodiment, obtaining historical passenger flow data of the target space on ordinary days in the same past period of the time to be predicted includes: obtaining origin-destination OD data; and performing path matching on the OD data based on the K-shortest path algorithm and in combination with one or more of driving time, walking time, and waiting time, thereby obtaining historical passenger flow data of the target space on ordinary days in the same past period of the time to be predicted.

[0192] In one possible embodiment, the path matching of OD data based on the K-shortest path algorithm combined with one or more of driving time, walking time, and waiting time includes: obtaining basic road network information; determining K corresponding shortest paths for OD pairs in the basic road network information based on the K-shortest path algorithm and driving time, where K is a positive integer; determining a recommended path for the OD pair from the K shortest paths of the OD pair by combining one or more of walking time and waiting time; and matching the OD data to the recommended path of the OD pair corresponding to the OD data.

[0193] In one possible embodiment, walking time includes one or more of the following: walking time to enter the station, walking time to exit the station, and walking time for transfers; waiting time includes one or more of the following: waiting time to enter the station, waiting time for transfers, and waiting time to exit the station.

[0194] It should be noted that the acquisition module 801 and the prediction module 802 in the aforementioned passenger flow prediction device 800 are specifically used to implement... Figure 7 Any embodiment of the passenger flow prediction method in [the context].

[0195] Figure 9 This is a schematic diagram of the structure of a computing device 900 provided in an embodiment of this application. The computing device 900 may be a laptop, tablet computer, cloud server, or other computing device, and this application does not specifically limit it.

[0196] The computing device 900 includes a processor 901, a memory 902, and a communication interface 903. The computing device 900 is specifically used to implement... Figure 7 This is an embodiment of a passenger flow prediction method. The processor 901, memory 902, and communication interface 903 can be interconnected via an internal bus 904, or they can communicate via wireless transmission or other means. This embodiment uses a connection via bus 904 as an example. Bus 904 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. Bus 904 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0197] Processor 901 may consist of at least one general-purpose processor, such as a central processing unit (CPU), or a combination of a CPU and hardware chips. The hardware chips may be application-specific integrated circuits (ASICs), programmable logic devices (PLDs), or combinations thereof. The PLDs may be complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), generic array logic (GALs), or any combination thereof. Processor 901 executes various types of digital storage instructions, such as software or firmware programs stored in memory 902, enabling computing device 900 to provide a variety of services.

[0198] The memory 902 is used to store program code, and its execution is controlled by the processor 901 to perform the above-mentioned tasks. Figure 7 Any embodiment of the passenger flow prediction method.

[0199] It should be noted that this embodiment can be implemented using a general-purpose physical server, such as an ARM server or an x86 server, or it can be implemented using a virtual machine based on a general-purpose physical server combined with NFV technology. A virtual machine refers to a complete computer system simulated by software, possessing full hardware system functionality and running in a completely isolated environment. This application does not impose specific limitations on this. It should be understood that... Figure 9 The computing device 900 shown can also be a server cluster consisting of at least one server, which is not specifically limited in this application.

[0200] Memory 902 may include volatile memory, such as random access memory (RAM); memory 902 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD); memory 902 may also include combinations of the above types. Memory 902 may store program code, specifically including code for execution. Figure 7 The program code for any embodiment of the passenger flow prediction method is not described in detail here.

[0201] The communication interface 903 can be a wired interface (e.g., an Ethernet interface), an internal interface (e.g., a high-speed serial computer expansion bus (peripheral component interconnect express, PCIe) bus interface), a wired interface (e.g., an Ethernet interface), or a wireless interface (e.g., a cellular network interface or a wireless LAN interface), for communicating with other devices or modules.

[0202] It needs to be explained that, Figure 9 This is merely one possible implementation of an embodiment of this application. In practical applications, the computing device 900 may include more or fewer components, and this is not a limitation. For any content not shown or described in the embodiments of this application, please refer to the foregoing. Figure 7 The relevant descriptions in the embodiments of the passenger flow prediction method are not repeated here.

[0203] This application also provides a computer-readable storage medium storing instructions that, when executed on a processor. Figure 7 The method of any of the embodiments can be implemented.

[0204] This application also provides a computer program product that, when run on a processor, provides a solution for... Figure 7 The method of any embodiment can be implemented.

[0205] The foregoing content details the process of obtaining corresponding passenger flow prediction results through trained passenger flow prediction models for various scenarios. It should be understood that before this, it is necessary to train the untrained passenger flow prediction model to obtain a trained passenger flow prediction model.

[0206] Please participate Figure 10 , Figure 10This is a flowchart illustrating a training method for a passenger flow prediction model provided in an embodiment of this application. The method may include the following steps:

[0207] S1001. Obtain training data, wherein the training data is the historical passenger flow data of the first space in the first time period, and the first time period is during the first holiday.

[0208] The first space can be at the spatial granularity of a station, cross section, line, or network. For example, the first space can be the entire rail transit network of a city, a line, a station, a cross section, etc., without specific limitations.

[0209] The first time period can be a specific time period / window within a past day, without specific limitations. For example, assuming a time window (or time period) is 60 minutes in length, a day can be divided into 24 time windows, and the time period to be predicted can be a specific time window within a past day.

[0210] The first public holiday can be any type of holiday, such as the Spring Festival holiday, the May Day holiday, the National Day holiday, etc., without specific restrictions.

[0211] Historical passenger flow data can be any type of historical entry volume, historical exit volume, historical transfer volume, or historical passenger volume, without specific limitations. It should be understood that since the first time period here falls within the first holiday, this historical passenger flow data is specific to the first holiday scenario and can be referred to as historical passenger flow data for the first holiday.

[0212] For example, suppose the training data obtained is the historical passenger flow of station A between 10:00 and 11:00 AM on October 1, 2021. Then, the first space is station A, the first time period is the time period between 10:00 and 11:00 AM on October 1, 2021, which is during the National Day holiday. The specific type of the training data is historical passenger flow.

[0213] S1002. Obtain historical concurrent data and a first coefficient of the training data, wherein the historical concurrent data is the historical passenger flow data of the first space on ordinary days in the same past time period in the first time period.

[0214] Here, the first coefficient represents the difference in historical passenger flow data between the first holiday and ordinary days in the same year prior to the first time period. For details regarding the first coefficient, please refer to the relevant content in step S504 above; it will not be elaborated upon here.

[0215] Continuing the previous example, we now need to obtain the historical passenger flow data for station A during the same period (10-11 AM) on October 1, 2021, on a typical day. Here, we assume we take the historical passenger flow data for the five days prior to the first period. Therefore, the historical passenger flow data for station A during the same period (10-11 AM) on the first period could be the historical passenger flow data for station A on September 26, 2021, September 27, 2021, September 28, 2021, September 29, 2021, and September 30, 2021. All of these data are historical data from the same period in the training data.

[0216] It should be noted that steps S1001 and S1002 can also be combined into one step and executed simultaneously.

[0217] S1003. Construct a feature vector for the training data based on the historical data from the same period. The feature vector includes one or more of the following: historical data from the same period, scaled-down historical data from the same period, and moving average of historical data from the same period.

[0218] For information on historical data from the same period, scaled-down historical data from the same period, and moving averages of historical data from the same period, please refer to the relevant content in step S203 above, which will not be repeated here.

[0219] In possible embodiments, the feature vector of the training data may also include weather features, activity features, etc. For details on weather features and activity features, please refer to the relevant content of steps S204 and S604 above, which will not be repeated here.

[0220] Continuing with the previous example, the weather features of this training data represent the weather conditions near station A between 10:00 AM and 11:00 AM on October 1, 2021, which may include one or more of the following: temperature, rainfall, wind speed, humidity, and air quality.

[0221] S1004. Train the untrained passenger flow prediction model using the feature vector, the first coefficient, and the training data.

[0222] The untrained passenger flow prediction model here can be the LightGBM model, the gradient boosting decision tree (GBDT) model, a neural network-based model, etc., and this application does not make any specific restrictions. The LightGBM model will be used as an example below.

[0223] It should be understood that the feature vector is the input feature vector of the passenger flow prediction model, and the training data serves as the output target value of the passenger flow prediction model. Specifically, the feature vector is input into the LightGBM model to obtain a first result; then, the first result is adjusted using a first coefficient to obtain a second result; the second result is compared with the output target value (i.e., the value of the training data), and the parameters of the LightGBM model are adjusted based on the comparison result.

[0224] In a possible embodiment, step S1002 may further include: obtaining a second coefficient, which represents the difference in annual growth rate between historical passenger flow data of the first holiday and historical passenger flow data of ordinary days; thus, step S1004 may also be: training the untrained passenger flow prediction model using the feature vector, the first coefficient, the second coefficient, and the training data. Specifically, the feature vector is input into the LightGBM model to obtain a first result; then the first result is adjusted using the first and second coefficients to obtain a second result; the second result is compared with the output target value (i.e., the value of the training data), and the parameters of the LightGBM model are adjusted according to the comparison result. Regarding the second coefficient, please refer to the relevant content in step S504 above, which will not be elaborated upon here.

[0225] Understandable. Figure 10 The execution process of steps S1001 to S1004 in the middle is similar to Figure 7 The execution process of steps S701 to S702 in the passenger flow prediction method is similar; please refer to [link / reference] for details. Figure 7 And related descriptions, which will not be elaborated here.

[0226] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0227] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Those skilled in the art will understand that implementing all or part of the processes of the above embodiments and making equivalent changes in accordance with the claims of this application are still within the scope of the invention.

Claims

1. A passenger flow prediction method, characterized in that, The method includes: Obtain historical passenger flow data and a first coefficient for ordinary days in the same past period of the period to be predicted for the target space, wherein the period to be predicted is located on a first holiday, and the first coefficient represents the difference between the historical passenger flow data of the target space on the first holiday in the same past period of the period to be predicted and the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted in the same year before the period to be predicted. Obtain a second coefficient, which represents the difference in annual growth rate between the historical passenger flow data of the first holiday and the historical passenger flow data of ordinary days; Based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, and the second coefficient, the passenger flow prediction result of the target space in the period to be predicted is predicted.

2. The method according to claim 1, characterized in that, The second coefficient represents the difference in annual growth rate between the historical passenger flow data of the target space during the same past period of the predicted period on the first holiday and the historical passenger flow data of the target space during the same past period on a regular day.

3. The method according to claim 1 or 2, characterized in that, The target space includes any one or more types of stations, lines, networks, and cross-sections; the historical passenger flow data includes one or more types of historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume; the passenger flow prediction results include one or more types of predicted inbound volume, predicted outbound volume, predicted transfer volume, and predicted passenger volume.

4. The method according to claim 1 or 2, characterized in that, The method further includes: acquiring the weather characteristics of the target space during the forecast period, wherein the weather characteristics include one or more of temperature, rainfall, wind force, humidity, and air quality; The step of predicting the passenger flow forecast result for the target space during the period to be predicted, based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, and the second coefficient, includes: Based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, the second coefficient, and the weather characteristics, the passenger flow prediction result of the target space in the period to be predicted is predicted.

5. The method according to claim 1 or 2, characterized in that, The time period to be predicted is located on the dates of the first holiday and the event. The method further includes: acquiring the activity characteristics of the target space during the time period to be predicted; The step of predicting the passenger flow forecast result for the target space during the period to be predicted, based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, and the second coefficient, includes: Based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, the second coefficient, and the activity characteristics, the passenger flow prediction result of the target space in the period to be predicted is predicted.

6. The method according to claim 5, characterized in that, The activity characteristics include one or more of the following: date, time period, activity start time, activity end time, week number, activity type, activity scale, time influence coefficient, and distance influence coefficient. The time influence coefficient represents the degree to which the time period to be predicted is affected by the activity event in terms of time, and the distance influence coefficient represents the degree to which the target space is affected by the activity event in terms of distance.

7. The method according to claim 1 or 2, characterized in that, The acquisition of historical passenger flow data for the target space during the same past period of the time to be predicted includes: Obtain origin-destination OD data; Based on the K-shortest path algorithm and combined with one or more of driving time, walking time, and waiting time, path matching is performed on the OD data to obtain historical passenger flow data of the target space on ordinary days in the same past time period during the time period to be predicted.

8. The method according to claim 7, characterized in that, The path matching based on the K-law shortest path algorithm, combined with one or more of driving time, walking time, and waiting time, performs path matching on the OD data, including: Obtain basic road network information; Based on the K-shortest path algorithm and the travel time, K shortest paths are determined for the OD pairs in the road network basic information, where K is a positive integer; By combining one or more of the walking time and the waiting time, a recommended path is determined for the OD pair from the K shortest paths of the OD pair; The OD data is matched to the recommended path of the OD pair corresponding to the OD data.

9. The method according to claim 7, characterized in that, The walking time includes one or more of the following: walking time to enter the station, walking time to exit the station, and walking time for transfers. The waiting time includes one or more of the following: waiting time to enter the station, waiting time for transfers, and waiting time to exit the station.

10. A passenger flow prediction device, characterized in that, The device includes: The acquisition module is used to acquire historical passenger flow data of the target space on ordinary days in the same past period of the predicted period and a first coefficient, wherein the predicted period is located on a first holiday, and the first coefficient represents the difference between the historical passenger flow data of the target space on the first holiday in the same past period of the predicted period and the historical passenger flow data of the target space on ordinary days in the same past period of the predicted period in the same year before the predicted period. The acquisition module is further configured to acquire a second coefficient, which represents the difference in annual growth rate between historical passenger flow data on the first holiday and historical passenger flow data on ordinary days. The prediction module is used to predict the passenger flow forecast result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, and the second coefficient.

11. The apparatus according to claim 10, characterized in that, The second coefficient represents the difference in annual growth rate between the historical passenger flow data of the target space during the same past period of the predicted period on the first holiday and the historical passenger flow data of the target space during the same past period on a regular day.

12. The apparatus according to claim 10 or 11, characterized in that, The target space includes any one or more types of stations, lines, networks, and cross-sections; the historical passenger flow data includes one or more types of historical inbound volume, historical outbound volume, historical transfer volume, and historical passenger volume; the passenger flow prediction results include one or more types of predicted inbound volume, predicted outbound volume, predicted transfer volume, and predicted passenger volume.

13. The apparatus according to claim 10 or 11, characterized in that, The acquisition module is further configured to: acquire the weather characteristics of the target space during the forecast period, wherein the weather characteristics include one or more of temperature, rainfall, wind force, humidity, and air quality; The prediction module is further configured to: predict the passenger flow forecast result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, the second coefficient, and the weather characteristics.

14. The apparatus according to claim 10 or 11, characterized in that, The time period to be predicted is located on the dates of the first holiday and the event. The acquisition module is also used to: acquire the activity characteristics of the target space during the time period to be predicted; The prediction module is further configured to: predict the passenger flow prediction result of the target space in the period to be predicted based on the historical passenger flow data of the target space on ordinary days in the same past period of the period to be predicted, the first coefficient, the first coefficient, and the activity characteristics.

15. The apparatus according to claim 14, characterized in that, The activity characteristics include one or more of the following: date, time period, activity start time, activity end time, week number, activity type, activity scale, time influence coefficient, and distance influence coefficient. The time influence coefficient represents the degree to which the time period to be predicted is affected by the activity event in terms of time, and the distance influence coefficient represents the degree to which the target space is affected by the activity event in terms of distance.

16. The apparatus according to claim 10 or 11, characterized in that, The acquisition of historical passenger flow data for the target space during the same past period of the time to be predicted includes: Obtain origin-destination OD data; Based on the K-shortest path algorithm and combined with one or more of driving time, walking time, and waiting time, path matching is performed on the OD data to obtain historical passenger flow data of the target space on ordinary days in the same past time period during the time period to be predicted.

17. The apparatus according to claim 16, characterized in that, The path matching based on the K-law shortest path algorithm, combined with one or more of driving time, walking time, and waiting time, performs path matching on the OD data, including: Obtain basic road network information; Based on the K-shortest path algorithm and the travel time, K shortest paths are determined for the OD pairs in the road network basic information, where K is a positive integer; By combining one or more of the walking time and the waiting time, a recommended path is determined for the OD pair from the K shortest paths of the OD pair; The OD data is matched to the recommended path of the OD pair corresponding to the OD data.

18. The apparatus according to claim 16, characterized in that, The walking time includes one or more of the following: walking time to enter the station, walking time to exit the station, and walking time for transfers. The waiting time includes one or more of the following: waiting time to enter the station, waiting time for transfers, and waiting time to exit the station.

19. A computing device, characterized in that, Including processor and memory; The memory is used to store computer programs; The processor is configured to execute a computer program stored in the memory, so that the device performs the method as described in any one of claims 1-9.

20. A computer-readable storage medium, characterized in that, Includes a program or instructions that, when run on a computer, execute the method as described in any one of claims 1-9.

Images