WO2021072959A1 - 一种地铁大客流预测方法、系统及电子设备 - Google Patents

一种地铁大客流预测方法、系统及电子设备 Download PDF

Info

Publication number
WO2021072959A1
WO2021072959A1 PCT/CN2019/125437 CN2019125437W WO2021072959A1 WO 2021072959 A1 WO2021072959 A1 WO 2021072959A1 CN 2019125437 W CN2019125437 W CN 2019125437W WO 2021072959 A1 WO2021072959 A1 WO 2021072959A1
Authority
WO
WIPO (PCT)
Prior art keywords
passenger flow
station
site
period
large passenger
Prior art date
Application number
PCT/CN2019/125437
Other languages
English (en)
French (fr)
Inventor
赵娟娟
须成忠
张帆
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2021072959A1 publication Critical patent/WO2021072959A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Definitions

  • This application belongs to the technical field of intelligent public transportation, and particularly relates to a method, system and electronic equipment for predicting a large passenger flow in subway.
  • Urban rail transit has gradually become the main mode of public transportation for citizens due to its advantages such as fast speed, large volume, accurate time, low pollution, and low energy consumption.
  • urban rail transit has become one of the best solutions for large cities at home and abroad to develop public transportation and relieve road traffic pressure.
  • the real-time prediction of passenger flow is the basis of passenger flow evacuation, dynamic train dispatching, and shuttle bus dispatching.
  • This application provides a method, system, and electronic equipment for predicting a large passenger flow in a subway, which aims to solve one of the above-mentioned technical problems in the prior art at least to a certain extent.
  • a method for predicting large passenger flow in subway includes the following steps:
  • Step a Extract static and dynamic characteristics of the subway based on historical passenger travel data
  • Step b Calculate the passenger flow aggregation index of each station based on the deviation between the real-time online passenger flow of the subway and the historical average passenger flow, and combine the passenger flow aggregation index, static characteristics and dynamic characteristics of each station to determine whether there will be a large passenger flow event in the subway network in the future. And the specific sites where a large passenger flow event will occur;
  • Step c Predict the accurate passenger flow of the site in the future period according to the static characteristics and dynamic characteristics corresponding to the specific site where a large passenger flow event will occur.
  • the technical solution adopted in the embodiment of the application further includes: in the step a, the static characteristics include online passenger flow Ai ,k,c , site cumulative passenger flow Ac j,c , historical average passenger flow at each time period, and historical large passenger flow
  • the number of events F j and the average time spent on the site Cd j ; among them, the historical average passenger flow in each period includes the online historical average passenger flow and the cumulative passenger flow, respectively.
  • the technical solution adopted in the embodiment of the present application further includes: in the step b, the real-time estimation of the passenger flow aggregation index of each station based on the offset between the real-time online passenger flow and the historical average passenger flow is specifically as follows: assuming that in the past period T k from The number of passengers entering the station si , the number of passengers who are still online after the time period T c, is the offset from the historical average Large, it means that R i, k, c will gather at other stations in the future period of time T c+m ; if the passenger flow of many stations in the entire subway network has a tendency to gather at station s j during the period of T c+m , It is considered that there will be a large passenger flow event at station s j in the future time period T c+m; the stations that Ri, k, c may go to and the time period affected are related to the time spent between the two stations; offset R i,k,c obey the Poisson distribution Ri ,k,c ⁇ P
  • the passenger flow aggregation index GS c,j, m at station s j in the future period T c+m is defined as: the key passenger flow pair arriving from other stations in the range of T c+mM ⁇ T c in the future period T c+m
  • the sum of the contribution rate of the large passenger flow of the station s j is calculated as:
  • the technical solution adopted in the embodiment of the application further includes: in the step b, the combination of the passenger flow aggregation index, static characteristics and dynamic characteristics of each station is used to determine whether a large passenger flow event will occur in the subway network in the future, and whether a large passenger flow will occur.
  • the specific sites of the incident include:
  • Step b1 According to the passenger flow aggregation index, screen out the set of potential gathering stations that may have a large passenger flow; in the current time period T c , in order to determine whether a large passenger flow event occurs in the future time period T c+m , first, the passenger flow aggregation index GS c,j,m
  • the top N B sites greater than the threshold G max are regarded as sites where a large passenger flow event may occur, and are added to the set of large passenger flow aggregation sites S B ;
  • Step b2 Establish a Logit model based on the static and dynamic characteristics of each site in the set of potential gathering sites to determine the specific site where a large passenger flow event will occur; firstly, determine that each site in the set of large passenger flow gathering sites S B is the nearest Whether there has been passenger flow gathering during the time period; if so, the site will be regarded as the site where the large passenger flow event occurred; if each station in the large passenger flow gathering site set S B does not have passenger flow gathering in the latest period, then each station in S B will be calculated For the probability of a large passenger flow event, the station with the highest probability is regarded as the station where the large passenger flow occurs; for each station in s j ⁇ S B , the passenger flow aggregation index GS c,j,m can reflect its dynamic characteristics, combined with the site s j The historical number of large passenger flow events F j and the average time spent by the stations Cd j , and the probability of large passenger flow events at each station is calculated based on the multi-probability selection
  • the technical solution adopted in the embodiment of the application further includes: in the step c, predicting the accurate passenger flow of the site in the future period according to the static characteristics and dynamic characteristics corresponding to the specific site where the large passenger flow event occurs specifically includes: During the time period T c , it is determined that a large passenger flow event will occur at the station s j in the future time period T c+m , the passenger flow aggregation index is GS c,j,m , and the passenger flow D j to the station s j in the future time period T c+m is predicted, c+m ; D j,c+m is expressed as the average passenger flow And the sum of the offset ⁇ , namely
  • ⁇ is expressed as the contribution amount ⁇ p during the period of I c+mN ⁇ I c and the contribution amount ⁇ f during the period of I c+1 ⁇ I c+m . It is assumed that in each event of a large passenger flow, the passengers who participate in the gathering of large passenger flow Arrival time obeys a uniform distribution, then the ratio of ⁇ f / ⁇ can be calculated as:
  • the large passenger flow contribution rate is the historical average large passenger flow contribution rate; if a large passenger flow event has never occurred at the station s j in the past, a linear regression model is constructed Contribution rate of average time spent in use And average passenger flow contribution rate Estimate the large passenger flow contribution rate of the station s i; where the time contribution rate
  • the calculation method is:
  • a metro passenger flow prediction system including:
  • Feature extraction module used to extract static and dynamic features of the subway based on historical passenger travel data
  • Large passenger flow aggregation station judgment module used to calculate the passenger flow aggregation index of each station based on the deviation between the real-time online passenger flow of the subway and the historical average passenger flow, and combine the passenger flow aggregation index, static characteristics and dynamic characteristics of each station to determine whether the subway network will be in the future Large passenger flow incidents will occur, and specific sites where large passenger flow incidents will occur;
  • Large passenger flow prediction module used to predict the precise passenger flow of the site in the future period according to the static characteristics and dynamic characteristics corresponding to the specific site where the large passenger flow event will occur.
  • the technical solution adopted in the embodiment of this application also includes: the static features include online passenger flow Ai ,k,c , site cumulative passenger flow Ac j,c , historical average passenger flow in each period, historical number of large passenger flow events F j , and site Average time spent Cd j ; among them, the historical average passenger flow in each period includes the online historical average passenger flow and the cumulative passenger flow, respectively.
  • the static features include online passenger flow Ai ,k,c , site cumulative passenger flow Ac j,c , historical average passenger flow in each period, historical number of large passenger flow events F j , and site Average time spent Cd j ; among them, the historical average passenger flow in each period includes the online historical average passenger flow and the cumulative passenger flow, respectively.
  • the dynamic characteristics include the time spent between two stations cst i,j and the average contribution rate of passenger flow
  • the large passenger flow aggregation site judgment module includes:
  • Passenger flow aggregation index calculation unit used to estimate the passenger flow aggregation index of each station in real time based on the offset between the real-time online passenger flow and the historical average passenger flow; assuming that the passengers who entered the station from the station si in the past time period T k , after the time period T c Offset of the number of passengers who are still online from the historical average Large, it means that R i, k, c will gather at other stations in the future period of time T c+m ; if the passenger flow of many stations in the entire subway network has a tendency to gather at station s j during the period of T c+m , It is considered that there will be a large passenger flow event at station s j in the future time period T c+m; the stations that Ri, k, c may go to and the time period affected are related to the time spent between the two stations; offset R i,k,c obey the Poisson distribution Ri ,k,c ⁇ P( ⁇ ), use the confidence interval to test whether the offset
  • the passenger flow aggregation index GS c,j, m at station s j in the future period of time T c+m is defined as: the key passenger flow pairs arriving from other stations in the range of T c+mM ⁇ T c are in the future period of time T c+m
  • the sum of the contribution rate of the large passenger flow of the station s j is calculated as:
  • the large passenger flow aggregation site judgment module further includes:
  • Large passenger flow aggregation site prediction unit used to screen out the potential aggregation site set that may have a large passenger flow according to the passenger flow aggregation index, and then establish a Logit model based on the static and dynamic characteristics of each station in the potential aggregation site set to determine that a large passenger flow event will occur
  • the specific site specifically:
  • the first N B stations where the passenger flow aggregation index GS c,j,m is greater than the threshold G max are taken as the possible large passenger flow events.
  • traffic aggregation index GS c, j, m may reflect the dynamics of binding sites s j history of a major traffic event frequency F j and site average time spent Cd j, and based on multiple selection of the probability
  • the model logit calculates the probability of a large passenger flow event at each station.
  • the technical solution adopted in the embodiment of the present application further includes: the large passenger flow prediction module predicts the accurate passenger flow of the site in the future period according to the static and dynamic characteristics of the specific site where the large passenger flow event occurs. Specifically, it includes: suppose that it is in the time period T At c , it is determined that a large passenger flow event will occur at station s j in the future time period T c+m , the passenger flow aggregation index is GS c,j,m , and the passenger flow rate D j,c+ into station s j in the future time period T c+m is predicted m ; D j,c+m is expressed as the average passenger flow And the sum of the offset ⁇ , namely
  • ⁇ is expressed as the contribution amount ⁇ p during the period of I c+mN ⁇ I c and the contribution amount ⁇ f during the period of I c+1 ⁇ I c+m . It is assumed that in each event of a large passenger flow, the passengers who participate in the gathering of large passenger flow Arrival time obeys a uniform distribution, then the ratio of ⁇ f / ⁇ can be calculated as:
  • the large passenger flow contribution rate is the historical average large passenger flow contribution rate; if a large passenger flow event has never occurred at the station s j in the past, a linear regression model is constructed Contribution rate of average time spent in use And average passenger flow contribution rate Estimate the large passenger flow contribution rate of the station s i; where the time contribution rate
  • the calculation method is:
  • an electronic device including:
  • At least one processor At least one processor
  • a memory communicatively connected with the at least one processor; wherein,
  • the memory stores instructions that can be executed by the one processor, and the instructions are executed by the at least one processor, so that the at least one processor can perform the following operations of the above-mentioned method for predicting a large passenger flow in subway:
  • Step a Extract static and dynamic characteristics of the subway based on historical passenger travel data
  • Step b Calculate the passenger flow aggregation index of each station based on the deviation between the real-time online passenger flow of the subway and the historical average passenger flow, and combine the passenger flow aggregation index, static characteristics and dynamic characteristics of each station to determine whether there will be a large passenger flow event in the subway network in the future. And the specific sites where a large passenger flow event will occur;
  • Step c Predict the precise passenger flow of the site in the future based on the static characteristics and dynamic characteristics of the specific site where a large passenger flow event will occur.
  • the beneficial effects produced by the embodiments of the present application are: the method, system and electronic equipment for predicting the large passenger flow of subway in the embodiments of the present application perform in-depth analysis of historical long-term passenger travel data, based on the occurrence of historical large passenger flow at each site The number of times, the amount of change in the passenger flow in the recent period, and the relevant characteristics of the site. Determine the specific site where a large passenger flow event will occur in the short term in the future, and combine the history of whether a large passenger flow event has occurred, and other dynamic and static characteristics of the site to predict the precise short-term passenger flow in the future the amount. In a large passenger flow scenario, this application has higher prediction accuracy than traditional methods.
  • Fig. 1 is a flowchart of a method for predicting a large subway passenger flow according to an embodiment of the present application
  • Figure 2 is an example diagram of large passenger flow gathering
  • Fig. 3 is a schematic structural diagram of a metro passenger flow prediction system according to an embodiment of the present application.
  • FIG. 4 is a schematic diagram of the hardware device structure of the method for predicting a large passenger flow in subway provided by an embodiment of the present application.
  • this application conducts in-depth analysis of historical long-term passenger travel data to predict in real time the stations with a large number of passenger flows (outbound passenger flows) and inbound (outbound) passenger flows in the short term in the future.
  • ⁇ and multiple stations S ⁇ s 1 ,s 2 ,...,s
  • a passenger's trip tr is associated with the four attributes S o , S d , t o , and t d , which respectively represent inbound station, outbound station, inbound time, and outbound time.
  • Definition 3 (period set): Divide a day into multiple periods T 1 , T 2 ,...T
  • Definition 4 Use O i,k to represent the number of passengers arriving from the station si during the time period T k .
  • tr ⁇ Tr,tr.s o s i ,tr.t o ⁇ T k ⁇
  • tr ⁇ Tr,tr.s d s i ,tr.t d ⁇ T k ⁇
  • Definition 7 For a certain station s j , if the difference between the outbound passenger flow during the time period T b and the historical average is greater than a certain set threshold ⁇ max , it means that the station s j occurs during the time period T b A large passenger flow event.
  • each transaction data includes the time and station of passengers entering (exiting) the station.
  • Figure 1 shows the outbound passenger flow and historical average passenger flow at the Shenzhen Convention and Exhibition Center site on September 28, 2014. If the threshold ⁇ max is set to 3000 and the time interval ⁇ is set to half an hour, it is determined that a large passenger flow event occurred at the exhibition center site from 8:00 to 8:30.
  • FIG. 2 is a flowchart of a method for predicting a large subway passenger flow according to an embodiment of the present application.
  • the method for predicting a large subway passenger flow in the embodiment of the present application includes the following steps:
  • Step 100 Extract static and dynamic characteristics of the subway based on historical long-term passenger travel data
  • step 100 the static characteristics and dynamic characteristics of the subway are the basis for the determination of large passenger flow gathering stations and the prediction of large passenger flow.
  • Static characteristics are the characteristics related to the site, including online passenger flow, cumulative passenger flow of the site, historical average passenger flow in each period, the number of historical large passenger flow events, and the average time spent on the site. details as follows:
  • Online passenger flow refers to passengers who have swiped their card to enter the station but have not yet left the station.
  • a i, k, c s i represents the site from the passenger card stop time period T k until T c then also the number of passengers in subway systems;
  • a i, k, c is calculated as:
  • the cumulative passenger flow of the station refers to the cumulative sum of the difference between the passenger flow in and out of the station in various periods in the past.
  • the following embodiment uses Ac j, c to identify the cumulative passenger flow of site s j , and the calculation method is:
  • Historical average passenger flow in each period includes online historical average passenger flow and cumulative passenger flow. This application is used separately Represents the historical average corresponding to A i,k,c and Ac j,k.
  • the number of large passenger flow incidents in history the number of historical large passenger flow occurrences at a certain site reflects the possibility of large passenger flow at this site to a certain extent.
  • F j is used to represent the number of times that a large passenger flow event occurs at the s j site. It should be noted that the number of large passenger flow incidents is counted according to continuous time periods. For example, a large passenger flow occurred at the site s j between 9:00 and 13:00 on a certain day, although it spanned multiple time periods, because the same event occurred, it was only counted once.
  • Dynamic characteristics involve multiple stations, including the time spent between two stations, the average contribution rate of passenger flow, and so on. specific:
  • Time spent between two stations the time spent between two stations is an important factor that affects the time for passengers arriving from each station to arrive at other stations.
  • Passenger travel data records the complete entry and exit time of each passenger, which provides sufficient data support for calculating the distribution of time spent between the two stations.
  • two types of time spent features are extracted respectively.
  • the first category is the average time spent between two stations, which represents a general description of the time spent between two stations.
  • the following uses cst i,j to identify the average time spent between site s i and site s j.
  • the calculation method is:
  • the second category is the distribution of time spent in different time periods of the day, describing the local characteristics of time. This is due to the influence of factors such as train scheduling time, such as the departure interval, and the time spent by passengers between two stations. Time period makes separate statistics on the time spent between the two stations.
  • train scheduling time such as the departure interval
  • time period makes separate statistics on the time spent between the two stations.
  • M is the number of time periods that spend the most time between any two stations in the subway system.
  • the average contribution rate of passenger flow refers to the proportion of passengers leaving a station from other stations.
  • Use matrix Indicates the average contribution rate of passengers entering the station from station s j, The calculation method is:
  • Tr all represents all historical travel records of passengers.
  • Step 200 Calculate the passenger flow aggregation index of each station based on the offset between the real-time online passenger flow and the historical average passenger flow, and combine the passenger flow aggregation index, static characteristics and dynamic characteristics of each station to determine whether a large passenger flow event will occur in the subway network in the future, and The specific site where a large passenger flow event occurred;
  • step 200 in order to clearly describe the prediction method of a large passenger flow event, the following detailed description is divided into two steps:
  • Step 201 Estimate the passenger flow aggregation index of each station in real time based on the deviation between the real-time online passenger flow and the historical average passenger flow;
  • step 201 since the departure time of passengers is later than the arrival time, the number of passengers leaving from a certain station s j in the future period of time T c+m is included in the previous period of time T c+mM ⁇ T c arriving from other stations Of passengers. If a certain station will have a large number of passenger inflows in the future period of time T c+m , then compared to the normal situation, the passengers entering the station in the previous period of time T c+mM ⁇ T c will have a tendency to gather at the station s j. This application In the embodiment, this trend is referred to as a passenger flow aggregation index.
  • the future traffic aggregation period T c + m index in the calculation of each site comprises: at any of these sites s j describes the whole network in the passenger line period T c + m internal site in the next period T c + m aggregate index for The tendency of abnormal aggregation of s j.
  • the abnormal clustering trend includes the following two rules: (1) Passenger travel is usually regular, and the online passenger flow is relatively close to the average in time and space distribution. However, in the case of strong random gathering of large passenger flow, it deviates from it. The average value is larger. (2) The travel time of passengers is inversely proportional to the number of passengers. That is, the shorter the travel time, the greater the number of passengers, indicating that passengers are more inclined to gather in nearby areas.
  • the passenger flow R i, k, c is likely to gather to other stations in the future period T c+m.
  • the stations that R i, k, c may go to and the time period affected are related to the time spent between the two stations. If the passenger metro-wide network of sites has a tendency to aggregate to the site j s in T c + m periods, then it is quite large passenger flow event may appear in future periods T c + m of s j station.
  • the offset Ri ,k,c obeys the Poisson distribution Ri ,k,c ⁇ P( ⁇ ), and the parameter ⁇ can be estimated by using maximum likelihood.
  • This application uses 95% as the confidence interval to test whether the offset R i,k,c > 0 is abnormal, and uses N(R i,k,c ) to identify whether R i,k,c passes the abnormality test, and if it passes the test The value is 0, otherwise it is 1. If R i,k,c is abnormal, then R i,k,c >0 is called a key passenger flow.
  • R i, k, c is the critical flow, assuming R i, k, c s j destined site, then R i, k, c s j reaches the site of the future traffic period T c + m, or R i,
  • the contribution rate of k, c to the occurrence of a large passenger flow event at station s j in the future period of time T c+m can be calculated as:
  • the passenger flow aggregation index GS c,j, m at station s j in the future time period T c+m is defined as: the key passenger flow pair coming in from other stations during the time period T c+mM ⁇ T c is in the future time period T c+
  • the sum of the contribution rate of the large passenger flow of m at the station s j can be calculated as:
  • Step 202 Screen out a set of potential gathering sites that may have a large passenger flow according to the passenger flow aggregation index, and then establish a Logit model based on the static and dynamic characteristics of each site in the potential gathering site set to determine the specific site where a large passenger flow event will occur;
  • step 202 according to the definition of the passenger flow aggregation index above, it can be seen that if the passengers of the entire subway network tend to gather to the station s j , it is also possible to gather to the neighboring stations of the station s j , that is, the passenger flow of s j and its neighboring stations.
  • the difference in the aggregation index may be small, so its neighboring sites may also be judged as sites where a large passenger flow gathers.
  • this application divides the following two steps to determine the site where a large passenger flow event occurs:
  • Step 2021 Selection of potential large passenger flow gathering sites; in the current time period T c , in order to determine whether a large passenger flow event occurs in the future time period T c+m , the first N B where the passenger flow aggregation index GS c,j,m is greater than the threshold G max
  • the site is regarded as a site where a large passenger flow event may occur, and is added to the set of large passenger flow gathering sites S B ;
  • the threshold value G max mode selection specifically is: the history data according to whether the first event into large large passenger traffic data D B and D N data types usually occur. Then calculate D N D B and the corresponding traffic aggregation index distribution f B and f n, and select a distribution density D B is much larger than the area in the aggregate index is D N. Under normal circumstances, the region is in the part where the aggregation index value is larger, and the threshold G max is set as the maximum value that satisfies the condition f B (x>G max )>95%.
  • the selection method of N B is specifically as follows: for each large passenger flow event, find all stations with a large passenger flow aggregation index greater than the threshold G max , and number these stations according to the passenger flow aggregation index from large to small, and the largest value in the number is taken as N The value of B.
  • Step 2022 determining large passenger aggregation site; sites object of large passenger aggregation is determined from a large collection site traffic aggregation S B s b in selected sites most likely to occur in large passenger.
  • a large passenger flow event is caused by a large-scale event, and the large passenger flow event will continue for a period of time, that is, if there is a gathering of passenger flow at a certain site in the current period, there may also be a gathering of passenger flow in the next period.
  • whether each station occurs and the probability of a large passenger flow event are related to the relevant characteristics of each station, such as the number of occurrences, regional characteristics, etc.
  • the method for determining a large passenger flow gathering site is specifically as follows:
  • s j ⁇ S B for each site, traffic aggregation index GS c, j, m which may reflect the dynamic characteristics, s j binding site of a major traffic event history number F j, the average time spent sites like Cd j , And based on the multi-probability selection model logit to calculate the probability of a large passenger flow event at each station.
  • the formula for calculating the probability of passengers going to the stop s j is:
  • the parameters ⁇ 1 , ⁇ 2 , and ⁇ 3 can be obtained by fitting historical passenger flow events.
  • Step 300 Predict the accurate passenger flow of the site in the future period according to the static and dynamic characteristics corresponding to the specific site where the large passenger flow event occurs;
  • Step 300 it is assumed it is determined at the time period T c future period T c + m in the s j sites large passenger event occurs, traffic aggregation index GS c, j, m, purpose of this step is to predict T c in the next period + m
  • D j,c+m can be expressed as average passenger flow And the sum of the offset ⁇ , namely In the following, the problem of passenger flow prediction is reduced to the prediction of ⁇ .
  • the time period for the arrival of passengers leaving the station at station s j in the future period T c+m will also be different.
  • Some passengers have entered the station in the past period of time I c+mN ⁇ I c , and another part of passengers will enter the station in the future period of time I c+1 ⁇ I c+m , so ⁇ can be expressed as two parts of ⁇ p and ⁇ f, respectively
  • the amount of contribution in the period of I c+mN to I c and the amount of contribution in the period of I c+1 to I c+m are the passenger flow aggregation index.
  • the large passenger flow contribution rate is estimated to be the historical average large passenger flow contribution rate. If there has never been a large passenger flow incident at this site in the past, construct a linear regression model Contribution rate of average time spent in use And average passenger flow contribution rate To estimate the large passenger flow contribution ratio of the station si. Of which time contribution rate The calculation method is:
  • FIG. 3 is a schematic structural diagram of a metro passenger flow prediction system according to an embodiment of the present application.
  • the metro passenger flow prediction system of the embodiment of the present application includes a feature extraction module, a large passenger flow aggregation site judgment module, and a large passenger flow prediction module.
  • Feature extraction module used to extract static and dynamic characteristics of the subway based on historical long-term passenger travel data; specifically, the feature extraction module includes:
  • Static feature extraction unit for extracting static features of a site: Static features are features related to the site, including online passenger flow, accumulated passenger flow at the site, historical average passenger flow in each period, the number of historical large passenger flow events, and average time spent on the site. details as follows:
  • Online passenger flow refers to passengers who have swiped their card to enter the station but have not yet left the station.
  • a i, k, c s i represents the site from the passenger card stop time period T k until T c then also the number of passengers in subway systems;
  • a i, k, c is calculated as:
  • the cumulative passenger flow of the station refers to the cumulative sum of the difference between the passenger flow in and out of the station in various periods in the past.
  • the following embodiment uses Ac j, c to identify the cumulative passenger flow of site s j , and the calculation method is:
  • Historical average passenger flow in each period includes online historical average passenger flow and cumulative passenger flow. This application is used separately Represents the historical average corresponding to A i,k,c and Ac j,k.
  • the number of large passenger flow incidents in history the number of historical large passenger flow occurrences at a certain site reflects the possibility of large passenger flow at this site to a certain extent.
  • F j is used to represent the number of times that a large passenger flow event occurs at the s j site. It should be noted that the number of large passenger flow incidents is counted according to continuous time periods. For example, a large passenger flow occurred at the site s j between 9:00 and 13:00 on a certain day, although it spanned multiple time periods, because the same event occurred, it was only counted once.
  • Dynamic feature extraction unit for extracting subway network features subway network features involve multiple stations, including the time spent between two stations, the average contribution rate of passenger flow, and so on. Specifically:
  • Time spent between two stations the time spent between two stations is an important factor that affects the time for passengers arriving from each station to arrive at other stations.
  • Passenger travel data records the complete entry and exit time of each passenger, which provides sufficient data support for calculating the distribution of time spent between the two stations.
  • two types of time spent features are extracted respectively.
  • the first category is the average time spent between two stations, which represents a general description of the time spent between two stations.
  • the following uses cst i,j to identify the average time spent between site s i and site s j.
  • the calculation method is:
  • the second category is the distribution of time spent in different time periods of the day, describing the local characteristics of time. This is due to the influence of factors such as train scheduling time, such as the departure interval, and the time spent by passengers between two stations. Time period makes separate statistics on the time spent between the two stations.
  • train scheduling time such as the departure interval
  • time period makes separate statistics on the time spent between the two stations.
  • M is the number of time periods that spend the most time between any two stations in the subway system.
  • the average contribution rate of passenger flow refers to the proportion of passengers leaving a station from other stations.
  • Use matrix Indicates the average contribution rate of passengers entering the station from station s j, The calculation method is:
  • Tr all represents all historical travel records of passengers.
  • Large passenger flow aggregation site judgment module used to calculate the passenger flow aggregation index of each station based on the offset between the real-time online passenger flow and the historical average passenger flow, and judge whether the subway network is in the future based on the passenger flow aggregation index of each station, the static characteristics of the station and the dynamic characteristics A large passenger flow event will occur, and the specific site where a large passenger flow event will occur; specifically, the large passenger flow gathering site judgment module includes:
  • Passenger flow aggregation index calculation unit used to estimate the passenger flow aggregation index of each station in real time based on the offset between the real-time online passenger flow and the historical average passenger flow; since the departure time of passengers is later than the arrival time, in the future period T c+m Passengers leaving the station at a certain station s j include passengers entering the station from other stations in the previous period T c+mM ⁇ T c. If a certain station will have a large number of passenger inflows in the future period of time T c+m , then compared to the normal situation, the passengers entering the station in the previous period of time T c+mM ⁇ T c will have a tendency to gather at the station s j. This application In the embodiment, this trend is referred to as a passenger flow aggregation index.
  • the future traffic aggregation period T c + m index in the calculation of each site comprises: at any of these sites s j describes the whole network in the passenger line period T c + m internal site in the next period T c + m aggregate index for The tendency of abnormal aggregation of s j.
  • the abnormal clustering trend includes the following two rules: (1) Passenger travel is usually regular, and the online passenger flow is relatively close to the average in time and space distribution. However, in the case of strong random gathering of large passenger flow, it deviates from it. The average value is larger. (2) The travel time of passengers is inversely proportional to the number of passengers. That is, the shorter the travel time, the greater the number of passengers, indicating that passengers are more inclined to gather in nearby areas.
  • the passenger flow R i, k, c is likely to gather to other stations in the future period T c+m.
  • the stations that R i, k, c may go to and the time period affected are related to the time spent between the two stations. If the passenger metro-wide network of sites has a tendency to aggregate to the site j s in T c + m periods, then it is quite large passenger flow event may appear in future periods T c + m of s j station.
  • the offset Ri ,k,c obeys the Poisson distribution Ri ,k,c ⁇ P( ⁇ ), and the parameter ⁇ can be estimated by using maximum likelihood.
  • This application uses 95% as the confidence interval to test whether the offset R i,k,c > 0 is abnormal, and uses N(R i,k,c ) to identify whether R i,k,c passes the abnormality test, and if it passes the test The value is 0, otherwise it is 1. If R i,k,c is abnormal, then R i,k,c >0 is called a key passenger flow.
  • R i, k, c is the critical flow, assuming R i, k, c s j destined site, then R i, k, c s j reaches the site of the future traffic period T c + m, or R i,
  • the contribution rate of k, c to the occurrence of a large passenger flow event at station s j in the future period of time T c+m can be calculated as:
  • the passenger flow aggregation index GS c,j, m at station s j in the future time period T c+m is defined as: the key passenger flow pair coming in from other stations during the time period T c+mM ⁇ T c is in the future time period T c+
  • the sum of the contribution rate of the large passenger flow of m at the station s j can be calculated as:
  • Large passenger flow aggregation station prediction unit used to screen out the potential aggregation site set that may have large passenger flow according to the passenger flow aggregation index, and then establish a Logit model based on the static and dynamic characteristics of each station in the potential aggregation site set to determine that a large passenger flow event will occur
  • the passenger flow aggregation index it can be seen that if the passengers of the entire subway network tend to gather to the station s j , then it is also possible to gather to the neighboring stations of the station s j , that is, the passenger flow between s j and its neighboring stations
  • the difference in the aggregation index may be small, so its neighboring sites may also be judged as sites where a large passenger flow gathers.
  • this application divides the following two steps to determine the site where a large passenger flow event occurs:
  • the threshold value G max mode selection specifically is: the history data according to whether the first event into large large passenger traffic data D B, and usually occurs two data D N. Then calculate D N D B and the corresponding traffic aggregation index distribution f B and f n, and select a distribution density D B is much larger than the area in the aggregate index is D N. Under normal circumstances, the region is in the part where the aggregation index value is larger, and the threshold G max is set as the maximum value that satisfies the condition f B (x>G max )>95%.
  • the selection method of N B is specifically as follows: for each large passenger flow event, find all stations with a large passenger flow aggregation index greater than the threshold G max , and number these stations according to the passenger flow aggregation index from large to small, and the largest value in the number is taken as N The value of B.
  • large passenger aggregation site is determined; large passenger aggregation site object is determined from a large collection site traffic aggregation S B s b in selected sites most likely to occur in large passenger.
  • a large passenger flow event is caused by a large-scale event, and the large passenger flow event will continue for a period of time, that is, if there is a gathering of passenger flow at a certain site in the current period, there may also be a gathering of passenger flow in the next period.
  • whether each station occurs and the probability of a large passenger flow event are related to the relevant characteristics of each station, such as the number of occurrences, regional characteristics, and so on.
  • the method of determining large passenger flow gathering stations is as follows: First, determine whether each station in the large passenger flow gathering site set S B has already had passenger flow gathering in the most recent period; the cumulative passenger flow Ac j,c of each station s j ⁇ S B can reflect the passenger flow Therefore, calculate the cumulative passenger flow of each station in S B and determine whether there is a large passenger flow event. If it is, this site is regarded as a large passenger flow event site. If each station in the large passenger flow aggregation site set S B has no passenger flow aggregation in the recent period , the probability of a large passenger flow event at each station in S B is calculated, and the station with the highest probability is regarded as the site where the large passenger flow occurs.
  • s j ⁇ S B for each site, traffic aggregation index GS c, j, m which may reflect the dynamic characteristics, s j binding site of a major traffic event history number F j, the average time spent sites like Cd j , And based on the multi-probability selection model logit to calculate the probability of a large passenger flow event at each station.
  • the formula for calculating the probability of passengers going to the stop s j is:
  • the parameters ⁇ 1 , ⁇ 2 , and ⁇ 3 can be obtained by fitting historical passenger flow events.
  • High traffic prediction module the site for accurately predict traffic according to a specific period in the future site corresponding static characteristic and a dynamic characteristic event occurring large passenger; tentative decisions in the future in the time period T c + m s j at time period T c large passenger site event occurs, traffic aggregation index GS c, j, m, purpose of this step is to predict the future traffic period D j T c + m s j of the inflow site, c + m.
  • D j,c+m can be expressed as average passenger flow And the sum of the offset ⁇ , namely In the following, the problem of passenger flow prediction is attributed to the prediction of ⁇ .
  • the time period for the arrival of passengers leaving the station at station s j in the future period of time T c+m will also be different.
  • Some passengers have entered the station in the past period of time I c+mN ⁇ I c , and another part of passengers will enter the station in the future period of time I c+1 ⁇ I c+m , so ⁇ can be expressed as two parts of ⁇ p and ⁇ f, respectively
  • the amount of contribution in the period of I c+mN to I c and the amount of contribution in the period of I c+1 to I c+m are the passenger flow aggregation index.
  • the large passenger flow contribution rate is estimated to be the historical average large passenger flow contribution rate. If there has never been a large passenger flow incident at this site in the past, construct a linear regression model Contribution rate of average time spent in use And average passenger flow contribution rate To estimate the large passenger flow contribution ratio of the station si. Of which time contribution rate The calculation method is:
  • FIG. 4 is a schematic diagram of the hardware device structure of the method for predicting a large passenger flow in subway provided by an embodiment of the present application.
  • the device includes one or more processors and memory. Taking a processor as an example, the device may also include: an input system and an output system.
  • the processor, the memory, the input system, and the output system may be connected by a bus or in other ways.
  • the connection by a bus is taken as an example.
  • the memory can be used to store non-transitory software programs, non-transitory computer executable programs, and modules.
  • the processor executes various functional applications and data processing of the electronic device by running non-transitory software programs, instructions, and modules stored in the memory, that is, realizing the processing methods of the foregoing method embodiments.
  • the memory may include a program storage area and a data storage area, where the program storage area can store an operating system and an application program required by at least one function; the data storage area can store data and the like.
  • the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices.
  • the memory may optionally include a memory remotely provided with respect to the processor, and these remote memories may be connected to the processing system through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • the input system can receive input digital or character information, and generate signal input.
  • the output system may include display devices such as a display screen.
  • the one or more modules are stored in the memory, and when executed by the one or more processors, the following operations of any of the foregoing method embodiments are performed:
  • Step a Extract static and dynamic characteristics of the subway based on historical passenger travel data
  • Step b Calculate the passenger flow aggregation index of each station based on the deviation between the real-time online passenger flow of the subway and the historical average passenger flow, and combine the passenger flow aggregation index, static characteristics and dynamic characteristics of each station to determine whether there will be a large passenger flow event in the subway network in the future. And the specific sites where a large passenger flow event will occur;
  • Step c Predict the accurate passenger flow of the site in the future period according to the static characteristics and dynamic characteristics corresponding to the specific site where a large passenger flow event will occur.
  • the embodiment of the present application provides a non-transitory (non-volatile) computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can perform the following operations:
  • Step a Extract static and dynamic characteristics of the subway based on historical passenger travel data
  • Step b Calculate the passenger flow aggregation index of each station based on the deviation between the real-time online passenger flow of the subway and the historical average passenger flow, and combine the passenger flow aggregation index, static characteristics and dynamic characteristics of each station to determine whether there will be a large passenger flow event in the subway network in the future. And the specific sites where a large passenger flow event will occur;
  • Step c Predict the accurate passenger flow of the site in the future period according to the static characteristics and dynamic characteristics corresponding to the specific site where a large passenger flow event will occur.
  • the embodiment of the present application provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions are executed by a computer To make the computer do the following:
  • Step a Extract static and dynamic characteristics of the subway based on historical passenger travel data
  • Step b Calculate the passenger flow aggregation index of each station based on the deviation between the real-time online passenger flow of the subway and the historical average passenger flow, and combine the passenger flow aggregation index, static characteristics and dynamic characteristics of each station to determine whether there will be a large passenger flow event in the subway network in the future. And the specific sites where a large passenger flow event will occur;
  • Step c Predict the accurate passenger flow of the site in the future period according to the static characteristics and dynamic characteristics corresponding to the specific site where a large passenger flow event will occur.
  • the metro passenger flow prediction method, system and electronic equipment of the embodiments of the present application make in-depth analysis of historical long-term passenger travel data, and judge the short-term future based on the number of historical passenger flow occurrences at each station, the amount of change in passenger flow in recent periods, and related characteristics of the station.
  • the specific site where a large passenger flow event will occur combined with whether there has been a large passenger flow event in history, and other dynamic and static characteristics of the site, predict the short-term accurate passenger flow in the future.
  • this application has higher prediction accuracy than traditional methods.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Tourism & Hospitality (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • General Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Quality & Reliability (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Educational Administration (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种地铁大客流预测方法、系统及电子设备，属于智能公共交通技术领域。该方法包括：步骤a：根据历史乘客出行数据提取地铁静态特征及动态特征；步骤b：基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；步骤c：根据所述会发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量。在大客流场景下，具有较高的预测精度。

Description

一种地铁大客流预测方法、系统及电子设备 技术领域
本申请属于智能公共交通技术领域，特别涉及一种地铁大客流预测方法、系统及电子设备。
背景技术
城市轨道交通以速度快、运量大、时间准、污染小、能耗低等优点，逐渐成为市民公共交通出行的主要方式。目前城市轨道交通已成为国内外大型城市发展公共交通、缓解道路交通压力的最佳解决方案之一。而对客流的实时预测尤其是对大客流的预测是客流疏散、动态列车调度、区间车调度等的基础。
目前，国内外研究学者已经在客流预测方面展开了大量的研究，比如轨道交通站点客流量、区间断面流量等的预测。但现有的预测方法主要针对通常情况下的客流做分析与预测，或者对已经发生大客流的站点做监测，无法适用于预测未来在短时间出现大客流的场景。相对于通常情况，对由大型活动等原因造成的在较短时间内在某站点汇聚大量客流的情况的预测具有更加重要的意义，例如维持公共交通安全等。
发明内容
本申请提供了一种地铁大客流预测方法、系统及电子设备，旨在至少在一定程度上解决现有技术中的上述技术问题之一。
为了解决上述问题，本申请提供了如下技术方案：
一种地铁大客流预测方法，包括以下步骤：
步骤a：根据历史乘客出行数据提取地铁静态特征及动态特征；
步骤b：基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
步骤c：根据所述会发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量。
本申请实施例采取的技术方案还包括：在所述步骤a中，所述静态特征包括在线客流A i,k,c、站点累计客流Ac j,c、各时段历史平均客流、历史发生大客流事件的次数F j、站点平均花费时间Cd j；其中，各时段历史平均客流包括在线历史平均客流量、累计客流量，分别使用
Figure PCTCN2019125437-appb-000001
表示A i,k,c,Ac j,k对应的历史平均值；所述动态特征包括两站之间花费时间cst i,j、客流平均贡献率
Figure PCTCN2019125437-appb-000002
本申请实施例采取的技术方案还包括：在所述步骤b中，所述基于实时在线客流与历史平均客流的偏移量实时估算各站点的客流聚集指数具体为：假设在过去时段T k从站点s i进站的乘客，在时间段T c之后仍然在线的乘客数量相比历史平均值的偏移量
Figure PCTCN2019125437-appb-000003
大，则表示R i,k,c这部分乘客会在未来时段T c+m向别的站点聚集；如果地铁全网很多站点的客流都有在T c+m时段向站点s j聚集的趋势，则认为在未来时段T c+m的s j站会出现大客流事件；R i,k,c可能去往的站点以及所影响的时段与两站之间花费时间有关系；偏移量R i,k,c服从泊松分布R i,k,c～P(λ)，使用置信区间检验偏移量R i,k,c＞0是否异常，如果R i,k,c异常，则将R i,k,c＞0称为关键客流；如果R i,k,c是关键客流，假设R i,k,c去往站点s j，那么R i,k,c在未来时段T c+m到达站点s j的客流量，或R i,k,c对在未来时段T c+m在s j站点的发生大客流事件的贡献率计算为：
Figure PCTCN2019125437-appb-000004
未来时段T c+m在s j站的客流聚集指数GS c,j,m定义为：在T c+m-M～T c时段范围从其它站点进站的关键客流对在未来时段T c+m在站点s j的大客流贡献率之和，计算为：
Figure PCTCN2019125437-appb-000005
本申请实施例采取的技术方案还包括：在所述步骤b中，所述结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点具体包括：
步骤b1：根据客流聚集指数筛选出可能发生大客流的潜在聚集站点集合；在当前时段T c，为了判断未来时段T c+m是否发生大客流事件，首先将客流聚集指数GS c,j,m大于阈值G max的前N B个站点作为可能发生大客流事件的站点，并添加到大客流聚集站点集合S B中；
步骤b2：根据所述潜在聚集站点集合中各个站点的静态特征及动态特征建立Logit模型，判断将要发生大客流事件的具体站点；首先，判断所述大客流聚集站点集合S B中各个站点在最近时段是否已经出现了客流聚集情况；如果是，则将该站点作为大客流事件发生站点；如果大客流聚集站点集合S B中各个站点在最近时段没有出现客流聚集情况，则计算S B中各站点发生大客流事件的概率，将概率最大的站点作为大客流发生的站点；对于s j∈S B中的每一个站点，客流聚集指数GS c,j,m可以反映其动态特征，结合站点s j的历史发生大客流事件的次数F j以及站点平均花费时间Cd j，并基于多概率选择模型logit计算各站点发生大客流事件的概率。
本申请实施例采取的技术方案还包括：在所述步骤c中，所述根据发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量具体包括：假设在时段T c时判定未来时段T c+m在s j站点将发生大客流 事件，客流聚集指数为GS c,j,m，预测在未来时段T c+m流入站点s j的客流量D j,c+m；D j,c+m表示为平均客流量
Figure PCTCN2019125437-appb-000006
与偏移量Δ之和，即
Figure PCTCN2019125437-appb-000007
Δ表示为I c+m-N～I c时段的贡献量Δ p和I c+1～I c+m时段的贡献量Δ f两部分，假设在每一次大客流事件中，参与大客流聚集的乘客进站时间服从均匀分布，则Δ f/Δ的比值可以计算为：
Figure PCTCN2019125437-appb-000008
上述公式中，Pr Δ(t o∈[I c+m-N I c])表示在Δ中，在过去时段进站的乘客比例；如果已知大客流贡献率，Δ计算为Δ＝Δ p×θ；
如果过去在站点s j发生过大客流事件，则所述大客流贡献率为历史平均大客流贡献率；如果过去在站点s j从来没有发生过大客流事件，则构建线性回归模型
Figure PCTCN2019125437-appb-000009
使用平均花费时间贡献率
Figure PCTCN2019125437-appb-000010
和客流平均贡献率
Figure PCTCN2019125437-appb-000011
估计站点s i的大客流贡献率；其中所述时间贡献率
Figure PCTCN2019125437-appb-000012
计算方法为：
Figure PCTCN2019125437-appb-000013
本申请实施例采取的另一技术方案为：一种地铁大客流预测系统，包括：
特征提取模块：用于根据历史乘客出行数据提取地铁静态特征及动态特征；
大客流聚集站点判断模块：用于基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
大客流量预测模块：用于根据所述会发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量。
本申请实施例采取的技术方案还包括：所述静态特征包括在线客流A i,k,c、站点累计客流Ac j,c、各时段历史平均客流、历史发生大客流事件的次数F j、站点平均花费时间Cd j；其中，各时段历史平均客流包括在线历史平均客流量、累计客流量，分别使用
Figure PCTCN2019125437-appb-000014
表示A i,k,c,Ac j,k对应的历史平均值；所述动态特征包括两站之间花费时间cst i,j、客流平均贡献率
Figure PCTCN2019125437-appb-000015
本申请实施例采取的技术方案还包括：所述大客流聚集站点判断模块包括：
客流聚集指数计算单元：用于基于实时在线客流与历史平均客流的偏移量实时估算各站点的客流聚集指数；假设在过去时段T k从站点s i进站的乘客，在时间段T c之后仍然在线的乘客数量相比历史平均值的偏移量
Figure PCTCN2019125437-appb-000016
大，则表示R i,k,c这部分乘客会在未来时段T c+m向别的站点聚集；如果地铁全网很多站点的客流都有在T c+m时段向站点s j聚集的趋势，则认为在未来时段T c+m的s j站会出现大客流事件；R i,k,c可能去往的站点以及所影响的时段与两站之间花费时间有关系；偏移量R i,k,c服从泊松分布R i,k,c～P(λ)，使用置信区间检验偏移量R i,k,c＞0是否异常，如果R i,k,c异常，则将R i,k,c＞0称为关键客流；如果R i,k,c是关键客流，假设R i,k,c去往站点s j，那么R i,k,c在未来时段T c+m到达站点s j的客流量，或R i,k,c对在未来时段T c+m在s j站点的发生大客流事件的贡献率计算为：
Figure PCTCN2019125437-appb-000017
未来时段T c+m在s j站的客流聚集指数GS c,j,m定义为：在T c+m-M～T c时段范围从其它站点进站的关键客流对在未来时段T c+m在站点s j的大客流贡献率之和，计算为：
Figure PCTCN2019125437-appb-000018
本申请实施例采取的技术方案还包括：所述大客流聚集站点判断模块还包括：
大客流聚集站点预测单元：用于根据客流聚集指数筛选出可能发生大客流的潜在聚集站点集合，然后根据潜在聚集站点集合中各个站点的静态特征以及动态特征建立Logit模型，判断将要发生大客流事件的具体站点；具体为：
在当前时段T c，为了判断未来时段T c+m是否发生大客流事件，首先将客流聚集指数GS c,j,m大于阈值G max的前N B个站点作为可能发生大客流事件的站点，并添加到大客流聚集站点集合S B中；判断所述大客流聚集站点集合S B中各个站点在最近时段是否已经出现了客流聚集情况；如果是，则将该站点作为大客流事件发生站点；如果大客流聚集站点集合S B中各个站点在最近时段没有出现客流聚集情况，则计算S B中各站点发生大客流事件的概率，将概率最大的站点作为大客流发生的站点；对于s j∈S B中的每一个站点，客流聚集指数GS c,j,m可以反映其动态特征，结合站点s j的历史发生大客流事件的次数F j以及站点平均花费时间Cd j，并基于多概率选择模型logit计算各站点发生大客流事件的概率。
本申请实施例采取的技术方案还包括：所述大客流量预测模块根据发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量具体包括：假设在时段T c时判定未来时段T c+m在s j站点将发生大客流事件，客流聚集指数为GS c,j,m，预测在未来时段T c+m流入站点s j的客流量D j,c+m；D j,c+m表示为平均客流量
Figure PCTCN2019125437-appb-000019
与偏移量Δ之和，即
Figure PCTCN2019125437-appb-000020
Δ表示为I c+m-N～I c时段的贡献量Δ p和I c+1～I c+m时段的贡献量Δ f两部分，假设在每一次大客流事件中，参与大客流聚集的乘客进站时间服从均匀分布，则Δ f/Δ的比值可以计算为：
Figure PCTCN2019125437-appb-000021
上述公式中，Pr Δ(t o∈[I c+m-N I c])表示在Δ中，在过去时段进站的乘客比例；如果已知大客流贡献率，Δ计算为Δ＝Δ p×θ；
如果过去在站点s j发生过大客流事件，则所述大客流贡献率为历史平均大客流贡献率；如果过去在站点s j从来没有发生过大客流事件，则构建线性回归模型
Figure PCTCN2019125437-appb-000022
使用平均花费时间贡献率
Figure PCTCN2019125437-appb-000023
和客流平均贡献率
Figure PCTCN2019125437-appb-000024
估计站点s i的大客流贡献率；其中所述时间贡献率
Figure PCTCN2019125437-appb-000025
计算方法为：
Figure PCTCN2019125437-appb-000026
本申请实施例采取的又一技术方案为：一种电子设备，包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述的地铁大客流预测方法的以下操作：
步骤a：根据历史乘客出行数据提取地铁静态特征及动态特征；
步骤b：基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
步骤c：根据所述会发生大客流事件的具体站点对应的静态特征及动态特 征预测该站点在未来时段的精确客流量。
相对于现有技术，本申请实施例产生的有益效果在于：本申请实施例的地铁大客流预测方法、系统及电子设备通过对历史长期的乘客出行数据做深入分析，基于各站点历史大客流发生次数、最近时段客流的变化量、以及站点相关特征判断未来短期将要发生大客流事件的具体站点，并结合历史是否发生过大客流事件，以及该站点其它的动态和静态特征预测未来短期的精确客流量。在大客流场景下，相比传统方法本申请具有较高的预测精度。
附图说明
图1是本申请实施例的地铁大客流预测方法的流程图；
图2为大客流聚集实例图；
图3是本申请实施例的地铁大客流预测系统的结构示意图；
图4是本申请实施例提供的地铁大客流预测方法的硬件设备结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
为了解决现有技术的不足，本申请通过对历史长期的乘客出行数据做深入分析，对未来短期内流入大量客流(出站客流)的站点和流入(出站)客流量进行实时预测。为了清楚描述及解释本申请的技术方案，首先给出本申请使用到的如下定义，然后对本申请的具体技术内容进行详细说明：
定义1(地铁系统)：一个地铁系统由多条线路L＝{l 1,l 2,…,l |L|}和多个站点S＝{s 1,s 2,…,s |S|}组成，其中|L|和|S|分别表示线路条数和站点个数；本申请所指的线路是带有方向性的逻辑线路，例如深圳1号线从罗湖到机场东与从机场东到罗湖为两条不同的线路。
定义2(乘客出行)：乘客的一次出行tr关联S o,S d,t o,t d四个属性，分别表示进站站点、出站站点、进站时间、出站时间。
定义3(时段集合)：将一天按固定的间隔τ划分为多个时段T 1,T 2,...T |T|，第k时段T k所包含的时间范围为{(k-1)τ,kτ}。
定义4(进站客流)：使用O i,k表示在T k时间段从站点s i进站的乘客数量，计算方法为：O i,k＝|{tr|tr∈Tr,tr.s o＝s i,tr.t o∈T k}|，其中Tr表示预测当天的所有已产生的出行；|.|操作用于统计符合条件的出行数量。
定义5(出站客流)：使用D i,k表示在在T k时间段从站点s i出站的乘客数量，计算方法为D i,k＝|{tr|tr∈Tr,tr.s d＝s i,tr.t d∈T k}|。
定义6(OD客流)：使用OD i,j, k,n表示在T k时间段从s i站点进站，在T k+n时段从s j出站的乘客数量，计算方法为
Figure PCTCN2019125437-appb-000027
定义7(大客流事件):对于某站点s j，如果在时间段T b出站的客流量与历史平均值的差大于某设定阈值Δ max，则表示站点s j在时间段T b发生了大客流事件。
问题定义:给出历史长期和实时的地铁智能卡交易数据，每一条交易数据包含乘客进(出)站的刷卡时间和站点，本申请的目的是预测未来时段T c+m(T c为当前时段，m＝1,2,3,…)地铁系统是否发生大客流事件，以及发生大客流事件的站点和具体客流量。例如图1展示了2014年9月28日深圳会展中心站 点的出站客流和历史平均客流。如果将阈值Δ max设定为3000，将时间间隔τ设置为半小时，则判定会展中心站点在8:00～8:30发生了大客流事件。
请参阅图2，是本申请实施例的地铁大客流预测方法的流程图。本申请实施例的地铁大客流预测方法包括以下步骤：
步骤100：根据历史长期的乘客出行数据提取地铁静态特征及动态特征；
步骤100中，地铁静态特征及动态特征是大客流聚集站点判定和大客流量预测的基础。静态特征即与站点相关的特征，包括在线客流、站点累计客流、各时段历史平均客流、历史发生大客流事件的次数、站点平均花费时间等。具体如下：
(1)在线客流；在线客流表示已经刷卡进站还没有出站的乘客。以下实施例使用A i,k,c表示在时间段T k从站点s i刷卡进站的乘客中，直到T c之后还处于地铁系统的乘客数量；A i,k,c计算方法为：
Figure PCTCN2019125437-appb-000028
(2)站点累计客流；给定站点s i，以及某天的当前时段T c,累计客流指的是在过去各个时段进出站客流之间差值的累计和。以下实施例使用Ac j,c标识站点s j的累计客流，计算方法为：
Figure PCTCN2019125437-appb-000029
(3)各时段历史平均客流；各时段历史平均客流包括在线历史平均客流量、累计客流量。本申请分别使用
Figure PCTCN2019125437-appb-000030
表示A i,k,c,Ac j,k对应的历史平均值。
(4)历史发生大客流事件的次数；某站点的历史大客流发生次数在一定程度上反应了此站点大客流发生的可能性。以下实施例使用F j表示大客流事件在s j站点发生的次数。需要注明的是，大客流事件发生的次数按连续时间段统计。例如在某天的9:00～13:00之间站点s j发生了大客流，虽然中间跨越多个时间段，由于发生的是同一事件，所以只统计一次。
(5)站点平均花费时间；一般而言，一个城市的中心区域比郊区更容易发生大客流聚集事件。给定某站点s j，来自全网其它站点的乘客前往此站点的平均花费时间Cd j在一定程度上反应了此站点的区域特征。从其它各站点到此站点出行的平均时间越长在某种程度上说明此站点可能越偏远。平均花费时间小的站点可能处于中心区域，反之为郊区。以下使用Cd j表示站点s j的区域特征，用从站点s j出站的乘客的平均花费时间表示，Cd j计算公式为：
Figure PCTCN2019125437-appb-000031
动态特征涉及多个站点，包括两站之间花费时间、客流平均贡献率等。具体的：
(1)两站之间时间花费；两站之间的花费时间是影响从各站点进站的客流到达其它站点时间的重要因素。乘客出行数据记录了每一位乘客完整的进出站时间，这为计算两站之间花费时间分布提供了充足的数据支撑。本申请实施例中，分别提取两类时间花费特征。第一类是两站之间的平均花费时间，表示两站之间花费时间的总体描述。以下使用cst i,j标识站点s i和站点s j之间的平均花费时间。计算方法为：
Figure PCTCN2019125437-appb-000032
第二类是一天不同时间段的花费时间分布，描述时间的局部特征，这是由于列车调度时间，例如发车间隔等因素的影响，两站之间乘客所花费时间也会不同，所以针对不同的时间段对两站之间的时间花费分别做统计。将在Tk时间段从s i进站前往s j的乘客中，在T k,T k+1,…T k+M时间段从s j出站的比例记做：
Figure PCTCN2019125437-appb-000033
计算方法为：
Figure PCTCN2019125437-appb-000034
公式(3)中，M为地铁系统任意两站之间花费时间最多的时间段的数量。
(2)客流平均贡献率；客流平均贡献率指的是从某站点出站的乘客从其它各站点进站的比例。使用矩阵
Figure PCTCN2019125437-appb-000035
表示从站点s j出站的乘客进站站点的平均贡献率，
Figure PCTCN2019125437-appb-000036
的计算方法为：
Figure PCTCN2019125437-appb-000037
公式(4)中，Tr all表示乘客的所有历史出行记录。
本申请实施例中，由于居民的出行受工作生活等因素影响呈现7天周期性，在工作日、周末、节假日三类情况下呈现出不同的特征，因此本申请针工作日、周末、节假日三类情况分别进行静态特征及动态特征的提取。
步骤200：基于实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及发生大客流事件的具体站点；
步骤200中，为了清楚描述大客流事件的预测方式，以下分两个步骤进行具体描述：
步骤201：基于实时在线客流与历史平均客流的偏移量实时估算各站点的客流聚集指数；
步骤201中，由于乘客的出站时间要晚于进站时间，在未来时段T c+m从某站点s j出站的乘客中包含在前期时段T c+m-M～T c从其它站点进站的乘客。如果某站点在未来时段T c+m内会发生大量客流流入，那么相比平常情况下，在前期时段T c+m-M～T c进站的乘客会存在往站点s j聚集的趋势，本申请实施例中将该趋势称之 为客流聚集指数。
进一步地，未来时段T c+m内各个站点的客流聚集指数计算方式具体包括：任意站点s j在未来时段T c+m的聚集指数用于描述全网在线乘客在T c+m时段内在站点s j异常聚集的趋势。根据分析，该异常聚集趋势包括以下两点规律：(1)通常情况下乘客出行呈现规律性，在线客流在时空分布上比较接近平均值，而在随机性强的大客流聚集情况下，其偏离平均值较大。(2)乘客的出行时间与乘客数量成反比。即出行时间越短，乘客数量越多，说明乘客更倾向于向附近区域聚集。
基于以上规律，如果在过去时段T k从站点s i进站的乘客，在时间段T c之后仍然在线的乘客数量相比历史平均值的偏移量
Figure PCTCN2019125437-appb-000038
较大，那么R i,k,c这部分客流很可能在未来时段T c+m向别的站点聚集。R i,k,c可能去往的站点以及所影响的时段与两站之间时间花费有关系。如果地铁全网很多站点的客流都有在T c+m时段向站点s j聚集的趋势，那么在未来时段T c+m的s j站很有可能出现大客流事件。
通过对历史数据分析，发现偏移量R i,k,c服从泊松分布R i,k,c～P(λ)，其中参数λ可以通过使用极大似然估计得到。本申请使用95％作为置信区间来检验偏移量R i,k,c＞0是否异常，并使用N(R i,k,c)标识R i,k,c是否通过异常检验，如果通过检验值为0，否则为1。如果R i,k,c异常，则将R i,k,c＞0称为关键客流。
如果R i,k,c是关键客流，假设R i,k,c去往站点s j，那么R i,k,c在未来时段T c+m到达站点s j的客流量，或R i,k,c对在未来时段T c+m在s j站点的发生大客流事件的贡献率可以计算为：
Figure PCTCN2019125437-appb-000039
其中，未来时段T c+m在s j站的客流聚集指数GS c,j,m定义为：在T c+m-M～T c时段范围从其它站点进站的关键客流对在未来时段T c+m在站点s j的大客流贡献率之和，可以计算为：
Figure PCTCN2019125437-appb-000040
步骤202：根据客流聚集指数筛选出可能发生大客流的潜在聚集站点集合，然后根据潜在聚集站点集合中各个站点的静态特征及动态特征建立Logit模型，判断将要发生大客流事件的具体站点；
步骤202中，根据以上客流聚集指数的定义，可以看到如果整个地铁网络的乘客有向站点s j聚集的趋势，那么也有可能向站点s j的邻居站点聚集，即s j与其邻居站点的客流聚集指数相差可能较小，因此其相邻站点也有可能被判断为大客流聚集的站点，为了防止误判断，本申请分以下两个步骤进行大客流事件发生站点的判断：
步骤2021：潜在大客流聚集站点选择；在当前时段T c，为了判断未来时段T c+m是否发生大客流事件，首先将客流聚集指数GS c,j,m大于阈值G max的前N B个站点作为可能发生大客流事件的站点，并添加到大客流聚集站点集合S B中；
步骤2021中，阈值G max的选择方式具体为：首先将历史数据按照是否发生大客流事件分为发生大客流数据D B和通常情况下数据D N两类。然后分别计算D B和D N中对应的客流聚集指数的分布f B和f n，并选择在D B中的分布密度远远大于在D N中的聚集指数的区域。正常情况下，该区域处于聚集指数值较大的部分，并设定阈值G max为满足条件f B(x＞G max)＞95％的最大值。
N B的选择方式具体为：针对每一次大客流事件，找到大客流聚集指数大于阈值G max的所有站点，并对这些站点按照客流聚集指数从大到小进行编号，编号中最大的值作为N B的值。
步骤2022：大客流聚集站点判定；大客流聚集站点判定的目的是从大客流聚集站点集合S B中选择出最可能发生大客流的站点s b。一般情况下，大客流事件是由于大型活动引起的，且大客流事件会延续一段时间，即某站点如果在当前时段出现了客流聚集，在下一个时段也有可能出现客流聚集。另外，各站点是否发生以及发生大客流事件的概率与各站点相关特征有关系，例如发生次数，区域特征等。
本申请实施例中，大客流聚集站点判定方式具体为：
首先，判断大客流聚集站点集合S B中各个站点在最近时段是否已经出现了客流聚集情况；各个站点s j∈S B的累计客流Ac j,c可以反映客流的聚集趋势，因此计算S B中每个站点的累计客流，并判断是否出现了大客流事件，如果是，则将此站点作为大客流事件发生站点。如果大客流聚集站点集合S B中各个站点在最近时段没有出现客流聚集情况，则计算S B中各站点发生大客流事件的概率，将概率最大的站点作为大客流发生的站点。对于s j∈S B中的每一个站点，客流聚集指数GS c,j,m可以反映其动态特征，结合站点s j的历史发生大客流事件的次数F j、站点平均花费时间Cd j等特征，并基于多概率选择模型logit计算各站点发生大客流事件的概率。Logit模型使用效用理论为：站点s j发生大客流事件的可能性与其效用值U j＝θ 1×GS c,j,m+θ 2×F j+θ 3×Cd j有关系，效用值越大，其发生的可能性越大。乘客前往站点s j的概率计算公式为：
Figure PCTCN2019125437-appb-000041
公式(7)中，参数θ 1,θ 2,θ 3可以通过历史大客流事件拟合得到。
步骤300：根据发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量；
步骤300中，假设在时段T c时判定未来时段T c+m在s j站点将发生大客流事件，客流聚集指数为GS c,j,m，本步骤的目的是预测在未来时段T c+m流入站点s j的客流量D j,c+m。D j,c+m可以表示为平均客流量
Figure PCTCN2019125437-appb-000042
与偏移量Δ之和，即
Figure PCTCN2019125437-appb-000043
以下将客流量预测问题归结为对Δ的预测。
由于站点间所花费时间的不同，在未来时段T c+m在站点s j出站的乘客进站时间段也会不同。一部分乘客在过去时段I c+m-N～I c已经进站，另一部分乘客将在未来时段I c+1～I c+m进站，所以Δ可以表示为Δ p和Δ f两部分，分别为I c+m-N～I c时段的贡献量和I c+1～I c+m时段的贡献量。其中客流聚集指数是在I c+m-N～I c时段内进站的关键客流对Δ的贡献，即Δ p的值。
为了计算Δ f的值，首先定义大客流贡献率。给定某次大客流事件发生的时段T k和站点s j，使用R表示各个站点的客流对Δ的贡献比例。此次大客流量D j,k与平均客流量
Figure PCTCN2019125437-appb-000044
的差值记为
Figure PCTCN2019125437-appb-000045
从任意其它站点s i进站的乘客数量
Figure PCTCN2019125437-appb-000046
与平均客流量的差值记为
Figure PCTCN2019125437-appb-000047
各个站点对Δ的贡献比例称作大客流贡献率，用向量R＝{r 1,r 2,...,r |S|}表示，其中
Figure PCTCN2019125437-appb-000048
表示站点s i对Δ的贡献比。
假设在每一次大客流事件中，参与大客流聚集的乘客进站时间服从均匀分布(在交通领域经常使用的假设，例如公交乘客到达站点的时间等)，那么Δ f/Δ的比值可以计算为：
Figure PCTCN2019125437-appb-000049
公式(8)中，Pr Δ(t o∈[I c+m-N I c])表示在Δ中，在过去时段进站的乘客比例。如果已知大客流贡献率，Δ可以计算为Δ＝Δ p×θ。
为了估计大客流贡献率，首先介绍两点发现：
(1)通过对同一站点发生过两次以上大客流事件，或者持续多个时间段的同一件大客流事件进行分析，发现相同站点在不同大客流事件中对此站点的贡献率比例基本稳定。这是因为由于区域等因素的影响，往此站点聚集的客流的源站点分布比较规律。
(2)任意一个其它站点s i对Δ的贡献率与两站之间的平均花费时间
Figure PCTCN2019125437-appb-000050
以及客流平均贡献率β j有比较强的相关性，这是因为乘客更加关注距离自己近的聚集事件，即乘客更倾向于在附近区域聚集。
基于上述两点，如果过去在站点s j发生过大客流事件，那么大客流贡献率估计为历史平均大客流贡献率。如果过去在此站点从来没有发生过大客流事件，构建线性回归模型
Figure PCTCN2019125437-appb-000051
使用平均花费时间贡献率
Figure PCTCN2019125437-appb-000052
和客流平均贡献率
Figure PCTCN2019125437-appb-000053
去估计站点s i的大客流贡献比。其中时间贡献率
Figure PCTCN2019125437-appb-000054
计算方法为：
Figure PCTCN2019125437-appb-000055
请参阅图3，是本申请实施例的地铁大客流预测系统的结构示意图。本申请实施例的地铁大客流预测系统包括特征提取模块、大客流聚集站点判断模块和大客流量预测模块。
特征提取模块：用于根据历史长期的乘客出行数据提取地铁静态特征和动态特征；具体的，特征提取模块包括：
用于提取站点静态特征的静态特征提取单元：静态特征即与站点相关的特征，包括在线客流、站点累计客流、各时段历史平均客流、历史发生大客流事件的次数、站点平均花费时间等。具体如下：
(1)在线客流；在线客流表示已经刷卡进站还没有出站的乘客。以下实施例使用A i,k,c表示在时间段T k从站点s i刷卡进站的乘客中，直到T c之后还处于地 铁系统的乘客数量；A i,k,c计算方法为：
Figure PCTCN2019125437-appb-000056
(2)站点累计客流；给定站点s i，以及某天的当前时段T c,累计客流指的是在过去各个时段进出站客流之间差值的累计和。以下实施例使用Ac j,c标识站点s j的累计客流，计算方法为：
Figure PCTCN2019125437-appb-000057
(3)各时段历史平均客流；各时段历史平均客流包括在线历史平均客流量、累计客流量。本申请分别使用
Figure PCTCN2019125437-appb-000058
表示A i,k,c,Ac j,k对应的历史平均值。
(4)历史发生大客流事件的次数；某站点的历史大客流发生次数在一定程度上反应了此站点大客流发生的可能性。以下实施例使用F j表示大客流事件在s j站点发生的次数。需要注明的是，大客流事件发生的次数按连续时间段统计。例如在某天的9:00～13:00之间站点s j发生了大客流，虽然中间跨越多个时间段，由于发生的是同一事件，所以只统计一次。
(5)站点平均花费时间；一般而言，一个城市的中心区域比郊区更容易发生大客流聚集事件。给定某站点s j，来自全网其它站点的乘客前往此站点的平均花费时间Cd j在一定程度上反应了此站点的区域特征。从其它各站点到此站点出行的平均时间越长在某种程度上说明此站点可能越偏远。平均花费时间小的站点可能处于中心区域，反之为郊区。以下使用Cd j表示站点s j的区域特征，用从站点s j出站的乘客的平均花费时间表示，Cd j计算公式为：
Figure PCTCN2019125437-appb-000059
用于提取地铁网络特征的动态特征提取单元：地铁网络特征涉及多个站点，包括两站之间花费时间、客流平均贡献率等。具体为：
(1)两站之间时间花费；两站之间的花费时间是影响从各站点进站的客 流到达其它站点时间的重要因素。乘客出行数据记录了每一位乘客完整的进出站时间，这为计算两站之间花费时间分布提供了充足的数据支撑。本申请实施例中，分别提取两类时间花费特征。第一类是两站之间的平均花费时间，表示两站之间花费时间的总体描述。以下使用cst i,j标识站点s i和站点s j之间的平均花费时间。计算方法为：
Figure PCTCN2019125437-appb-000060
第二类是一天不同时间段的花费时间分布，描述时间的局部特征，这是由于列车调度时间，例如发车间隔等因素的影响，两站之间乘客所花费时间也会不同，所以针对不同的时间段对两站之间的时间花费分别做统计。将在Tk时间段从s i进站前往s j的乘客中，在T k,T k+1,…T k+M时间段从s j出站的比例记做：
Figure PCTCN2019125437-appb-000061
计算方法为：
Figure PCTCN2019125437-appb-000062
公式(3)中，M为地铁系统任意两站之间花费时间最多的时间段的数量。
(2)客流平均贡献率；客流平均贡献率指的是从某站点出站的乘客从其它各站点进站的比例。使用矩阵
Figure PCTCN2019125437-appb-000063
表示从站点s j出站的乘客进站站点的平均贡献率，
Figure PCTCN2019125437-appb-000064
的计算方法为：
Figure PCTCN2019125437-appb-000065
公式(4)中，Tr all表示乘客的所有历史出行记录。
本申请实施例中，由于居民的出行受工作生活等因素影响呈现7天周期性，在工作日、周末、节假日三类情况下呈现出不同的特征，因此本申请针工作日、周末、节假日三类情况分别进行静态特征及动态特征的提取。
大客流聚集站点判断模块：用于基于实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并根据各站点的客流聚集指数、站点静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及发生大客流事件的具体站点；具体的，大客流聚集站点判断模块包括：
客流聚集指数计算单元：用于基于实时在线客流与历史平均客流的偏移量实时估算各站点的客流聚集指数；由于乘客的出站时间要晚于进站时间，在未来时段T c+m从某站点s j出站的乘客中包含在前期时段T c+m-M～T c从其它站点进站的乘客。如果某站点在未来时段T c+m内会发生大量客流流入，那么相比平常情况下，在前期时段T c+m-M～T c进站的乘客会存在往站点s j聚集的趋势，本申请实施例中将该趋势称之为客流聚集指数。
进一步地，未来时段T c+m内各个站点的客流聚集指数计算方式具体包括：任意站点s j在未来时段T c+m的聚集指数用于描述全网在线乘客在T c+m时段内在站点s j异常聚集的趋势。根据分析，该异常聚集趋势包括以下两点规律：(1)通常情况下乘客出行呈现规律性，在线客流在时空分布上比较接近平均值，而在随机性强的大客流聚集情况下，其偏离平均值较大。(2)乘客的出行时间与乘客数量成反比。即出行时间越短，乘客数量越多，说明乘客更倾向于向附近区域聚集。
基于以上规律，如果在过去时段T k从站点s i进站的乘客，在时间段T c之后仍然在线的乘客数量相比历史平均值的偏移量
Figure PCTCN2019125437-appb-000066
较大，那么R i,k,c这部分客流很可能在未来时段T c+m向别的站点聚集。R i,k,c可能去往的站点以及所影响的时段与两站之间时间花费有关系。如果地铁全网很多站点的客流都有在T c+m时段向站点s j聚集的趋势，那么在未来时段T c+m的s j站很有可能出现大客流事件。
通过对历史数据分析，发现偏移量R i,k,c服从泊松分布R i,k,c～P(λ)，其中参数λ可以通过使用极大似然估计得到。本申请使用95％作为置信区间来检验偏移量R i,k,c＞0是否异常，并使用N(R i,k,c)标识R i,k,c是否通过异常检验，如果通过检验值为0，否则为1。如果R i,k,c异常，则将R i,k,c＞0称为关键客流。
如果R i,k,c是关键客流，假设R i,k,c去往站点s j，那么R i,k,c在未来时段T c+m到达站点s j的客流量，或R i,k,c对在未来时段T c+m在s j站点的发生大客流事件的贡献率可以计算为：
Figure PCTCN2019125437-appb-000067
其中，未来时段T c+m在s j站的客流聚集指数GS c,j,m定义为：在T c+m-M～T c时段范围从其它站点进站的关键客流对在未来时段T c+m在站点s j的大客流贡献率之和，可以计算为：
Figure PCTCN2019125437-appb-000068
大客流聚集站点预测单元：用于根据客流聚集指数筛选出可能发生大客流的潜在聚集站点集合，然后根据潜在聚集站点集合中各个站点的静态特征以及动态特征建立Logit模型，判断将要发生大客流事件的具体站点；根据以上客流聚集指数的定义，可以看到如果整个地铁网络的乘客有向站点s j聚集的趋势，那么也有可能向站点s j的邻居站点聚集，即s j与其邻居站点的客流聚集指数相差可能较小，因此其相邻站点也有可能被判断为大客流聚集的站点，为了防止误判断，本申请分以下两个步骤进行大客流事件发生站点的判断：
一、潜在大客流聚集站点选择；在当前时段T c，为了判断未来时段T c+m是否发生大客流事件，首先将客流聚集指数GS c,j,m大于阈值G max的前N B个站点作为可能发生大客流事件的站点，并添加到大客流聚集站点集合S B中。
上述中，阈值G max的选择方式具体为：首先将历史数据按照是否发生大客流事件分为发生大客流数据D B和通常情况下数据D N两类。然后分别计算D B和D N中对应的客流聚集指数的分布f B和f n，并选择在D B中的分布密度远远大于在D N中的聚集指数的区域。正常情况下，该区域处于聚集指数值较大的部分，并设定阈值G max为满足条件f B(x＞G max)＞95％的最大值。
N B的选择方式具体为：针对每一次大客流事件，找到大客流聚集指数大于阈值G max的所有站点，并对这些站点按照客流聚集指数从大到小进行编号，编号中最大的值作为N B的值。
二、大客流聚集站点判定；大客流聚集站点判定的目的是从大客流聚集站点集合S B中选择出最可能发生大客流的站点s b。一般情况下，大客流事件是由于大型活动引起的，且大客流事件会延续一段时间，即某站点如果在当前时段出现了客流聚集，在下一个时段也有可能出现客流聚集。另外，各站点是否发生以及发生大客流事件的概率与各站点相关特征有关系，例如发生次数，区域特征等。
大客流聚集站点判定方式具体为：首先，判断大客流聚集站点集合S B中各个站点在最近时段是否已经出现了客流聚集情况；各个站点s j∈S B的累计客流Ac j,c可以反映客流的聚集趋势，因此计算S B中每个站点的累计客流，并判断是否出现了大客流事件，如果是，则将此站点作为大客流事件发生站点。如果大客流聚集站点集合S B中各个站点在最近时段没有出现客流聚集情况，则计算S B中各站点发生大客流事件的概率，将概率最大的站点作为大客流发生的站点。对于s j∈S B中的每一个站点，客流聚集指数GS c,j,m可以反映其动态特征，结合站点s j的历史发生大客流事件的次数F j、站点平均花费时间Cd j等特征，并基于多概率选择模型logit计算各站点发生大客流事件的概率。Logit模 型使用效用理论为：站点s j发生大客流事件的可能性与其效用值U j＝θ 1×GS c,j,m+θ 2×F j+θ 3×Cd j有关系，效用值越大，其发生的可能性越大。乘客前往站点s j的概率计算公式为：
Figure PCTCN2019125437-appb-000069
公式(7)中，参数θ 1,θ 2,θ 3可以通过历史大客流事件拟合得到。
大客流量预测模块：用于根据发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量；假设在时段T c时判定未来时段T c+m在s j站点将发生大客流事件，客流聚集指数为GS c,j,m，本步骤的目的是预测在未来时段T c+m流入站点s j的客流量D j,c+m。D j,c+m可以表示为平均客流量
Figure PCTCN2019125437-appb-000070
与偏移量Δ之和，即
Figure PCTCN2019125437-appb-000071
以下将客流量预测问题归结为对Δ的预测。
由于站点间所花费时间的不同，在未来时段T c+m在站点s j出站的乘客进站时间段也会不同。一部分乘客在过去时段I c+m-N～I c已经进站，另一部分乘客将在未来时段I c+1～I c+m进站，所以Δ可以表示为Δ p和Δ f两部分，分别为I c+m-N～I c时段的贡献量和I c+1～I c+m时段的贡献量。其中客流聚集指数是在I c+m-N～I c时段内进站的关键客流对Δ的贡献，即Δ p的值。
为了计算Δ f的值，首先定义大客流贡献率。给定某次大客流事件发生的时段T k和站点s j，使用R表示各个站点的客流对Δ的贡献比例。此次大客流量D j,k与平均客流量
Figure PCTCN2019125437-appb-000072
的差值记为
Figure PCTCN2019125437-appb-000073
从任意其它站点s i进站的乘客数量
Figure PCTCN2019125437-appb-000074
与平均客流量的差值记为
Figure PCTCN2019125437-appb-000075
各个站点对Δ的贡献比例称作大客流贡献率，用向量R＝{r 1,r 2,...,r |S|}表示，其中
Figure PCTCN2019125437-appb-000076
表示站点s i对Δ的贡献比。
假设在每一次大客流事件中，参与大客流聚集的乘客进站时间服从均匀分布(在交通领域经常使用的假设，例如公交乘客到达站点的时间等)，那么Δ f/Δ 的比值可以计算为：
Figure PCTCN2019125437-appb-000077
公式(8)中，Pr Δ(t o∈[I c+m-N I c])表示在Δ中，在过去时段进站的乘客比例。如果已知大客流贡献率，Δ可以计算为Δ＝Δ p×θ。
为了估计大客流贡献率，首先介绍两点发现：
(1)通过对同一站点发生过两次以上大客流事件，或者持续多个时间段的同一件大客流事件进行分析，发现相同站点在不同大客流事件中对此站点的贡献率比例基本稳定。这是因为由于区域等因素的影响，往此站点聚集的客流的源站点分布比较规律。
(2)任意一个其它站点s i对Δ的贡献率与两站之间的平均花费时间
Figure PCTCN2019125437-appb-000078
以及客流平均贡献率β j有比较强的相关性，这是因为乘客更加关注距离自己近的聚集事件，即乘客更倾向于在附近区域聚集。
基于上述两点，如果过去在站点s j发生过大客流事件，那么大客流贡献率估计为历史平均大客流贡献率。如果过去在此站点从来没有发生过大客流事件，构建线性回归模型
Figure PCTCN2019125437-appb-000079
使用平均花费时间贡献率
Figure PCTCN2019125437-appb-000080
和客流平均贡献率
Figure PCTCN2019125437-appb-000081
去估计站点s i的大客流贡献比。其中时间贡献率
Figure PCTCN2019125437-appb-000082
计算方法为：
Figure PCTCN2019125437-appb-000083
图4是本申请实施例提供的地铁大客流预测方法的硬件设备结构示意图。如图4所示，该设备包括一个或多个处理器以及存储器。以一个处理器为例，该设备还可以包括：输入系统和输出系统。
处理器、存储器、输入系统和输出系统可以通过总线或者其他方式连接，图4中以通过总线连接为例。
存储器作为一种非暂态计算机可读存储介质，可用于存储非暂态软件程序、非暂态计算机可执行程序以及模块。处理器通过运行存储在存储器中的非暂态软件程序、指令以及模块，从而执行电子设备的各种功能应用以及数据处理，即实现上述方法实施例的处理方法。
存储器可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储数据等。此外，存储器可以包括高速随机存取存储器，还可以包括非暂态存储器，例如至少一个磁盘存储器件、闪存器件、或其他非暂态固态存储器件。在一些实施例中，存储器可选包括相对于处理器远程设置的存储器，这些远程存储器可以通过网络连接至处理系统。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入系统可接收输入的数字或字符信息，以及产生信号输入。输出系统可包括显示屏等显示设备。
所述一个或者多个模块存储在所述存储器中，当被所述一个或者多个处理器执行时，执行上述任一方法实施例的以下操作：
步骤a：根据历史乘客出行数据提取地铁静态特征及动态特征；
步骤b：基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
步骤c：根据所述会发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量。
上述产品可执行本申请实施例所提供的方法，具备执行方法相应的功能模块和有益效果。未在本实施例中详尽描述的技术细节，可参见本申请实施例提 供的方法。
本申请实施例提供了一种非暂态(非易失性)计算机存储介质，所述计算机存储介质存储有计算机可执行指令，该计算机可执行指令可执行以下操作：
步骤a：根据历史乘客出行数据提取地铁静态特征及动态特征；
步骤b：基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
步骤c：根据所述会发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量。
本申请实施例提供了一种计算机程序产品，所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算机程序，所述计算机程序包括程序指令，当所述程序指令被计算机执行时，使所述计算机执行以下操作：
步骤a：根据历史乘客出行数据提取地铁静态特征及动态特征；
步骤b：基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
步骤c：根据所述会发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量。
本申请实施例的地铁大客流预测方法、系统及电子设备通过对历史长期的乘客出行数据做深入分析，基于各站点历史大客流发生次数、最近时段客流的变化量、以及站点相关特征判断未来短期将要发生大客流事件的具体站点，并结合历史是否发生过大客流事件，以及该站点其它的动态和静态特征预测未来短期的精确客流量。在大客流场景下，相比传统方法本申请具有较高的预测精 度。
对所公开的实施例的上述说明，使本领域专业技术人员能够实现或使用本申请。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的，本申请中所定义的一般原理可以在不脱离本申请的精神或范围的情况下，在其它实施例中实现。因此，本申请将不会被限制于本申请所示的这些实施例，而是要符合与本申请所公开的原理和新颖特点相一致的最宽的范围。

Claims (11)

  1. 一种地铁大客流预测方法，其特征在于，包括以下步骤：
    步骤a：根据历史乘客出行数据提取地铁静态特征及动态特征；
    步骤b：基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
    步骤c：根据所述会发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量。
  2. 根据权利要求1所述的地铁大客流预测方法，其特征在于，在所述步骤a中，所述静态特征包括在线客流A i,k,c、站点累计客流Ac j,c、各时段历史平均客流、历史发生大客流事件的次数F j、站点平均花费时间Cd j；其中，各时段历史平均客流包括在线历史平均客流量、累计客流量，分别使用
    Figure PCTCN2019125437-appb-100001
    表示A i,k,c,Ac j,k对应的历史平均值；所述动态特征包括两站之间花费时间cst i,j、客流平均贡献率
    Figure PCTCN2019125437-appb-100002
  3. 根据权利要求2所述的地铁大客流预测方法，其特征在于，在所述步骤b中，所述基于实时在线客流与历史平均客流的偏移量实时估算各站点的客流聚集指数具体为：假设在过去时段T k从站点s i进站的乘客，在时间段T c之后仍然在线的乘客数量相比历史平均值的偏移量
    Figure PCTCN2019125437-appb-100003
    大，则表示R i,k,c这部分乘客会在未来时段T c+m向别的站点聚集；如果地铁全网很多站点的客流都有在T c+m时段向站点s j聚集的趋势，则认为在未来时段T c+m的s j站会出现大客流事件；R i,k,c可能去往的站点以及所影响的时段与两站之间花费时间有关系；偏移量R i,k,c服从泊松分布R i,k,c～P(λ)，使用置信区间检验偏移量R i,k,c＞0是否异常，如果R i,k,c异常，则将R i,k,c＞0称 为关键客流；如果R i,k,c是关键客流，假设R i,k,c去往站点s j，那么R i,k,c在未来时段T c+m到达站点s j的客流量，或R i,k,c对在未来时段T c+m在s j站点的发生大客流事件的贡献率计算为：
    Figure PCTCN2019125437-appb-100004
    未来时段T c+m在s j站的客流聚集指数GS c,j,m定义为：在T c+m-M～T c时段范围从其它站点进站的关键客流对在未来时段T c+m在站点s j的大客流贡献率之和，计算为：
    Figure PCTCN2019125437-appb-100005
  4. 根据权利要求3所述的地铁大客流预测方法，其特征在于，在所述步骤b中，所述结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点具体包括：
    步骤b1：根据客流聚集指数筛选出可能发生大客流的潜在聚集站点集合；在当前时段T c，为了判断未来时段T c+m是否发生大客流事件，首先将客流聚集指数GS c,j,m大于阈值G max的前N B个站点作为可能发生大客流事件的站点，并添加到大客流聚集站点集合S B中；
    步骤b2：根据所述潜在聚集站点集合中各个站点的静态特征及动态特征建立Logit模型，判断将要发生大客流事件的具体站点；首先，判断所述大客流聚集站点集合S B中各个站点在最近时段是否已经出现了客流聚集情况；如果是，则将该站点作为大客流事件发生站点；如果大客流聚集站点集合S B中各个站点在最近时段没有出现客流聚集情况，则计算S B中各站点发生大客流事件的概率，将概率最大的站点作为大客流发生的站点；对于s j∈S B中的每一个站点，客流聚集指数GS c,j,m可以反映其动态特征，结合站点s j的历史发生大客流事件的次 数F j以及站点平均花费时间Cd j，并基于多概率选择模型logit计算各站点发生大客流事件的概率。
  5. 根据权利要求4所述的地铁大客流预测方法，其特征在于，在所述步骤c中，所述根据发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量具体包括：假设在时段T c时判定未来时段T c+m在s j站点将发生大客流事件，客流聚集指数为GS c,j,m，预测在未来时段T c+m流入站点s j的客流量D j,c+m；D j,c+m表示为平均客流量
    Figure PCTCN2019125437-appb-100006
    与偏移量Δ之和，即
    Figure PCTCN2019125437-appb-100007
    Δ表示为I c+m-N～I c时段的贡献量Δ p和I c+1～I c+m时段的贡献量Δ f两部分，假设在每一次大客流事件中，参与大客流聚集的乘客进站时间服从均匀分布，则Δ f/Δ的比值可以计算为：
    Figure PCTCN2019125437-appb-100008
    上述公式中，Pr Δ(t o∈[I c+m-N I c])表示在Δ中，在过去时段进站的乘客比例；如果已知大客流贡献率，Δ计算为Δ＝Δ p×θ；
    如果过去在站点s j发生过大客流事件，则所述大客流贡献率为历史平均大客流贡献率；如果过去在站点s j从来没有发生过大客流事件，则构建线性回归模型
    Figure PCTCN2019125437-appb-100009
    使用平均花费时间贡献率
    Figure PCTCN2019125437-appb-100010
    和客流平均贡献率
    Figure PCTCN2019125437-appb-100011
    估计站点s i的大客流贡献率；其中所述时间贡献率
    Figure PCTCN2019125437-appb-100012
    计算方法为：
    Figure PCTCN2019125437-appb-100013
  6. 一种地铁大客流预测系统，其特征在于，包括：
    特征提取模块：用于根据历史乘客出行数据提取地铁静态特征及动态特征；
    大客流聚集站点判断模块：用于基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及 动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
    大客流量预测模块：用于根据所述会发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量。
  7. 根据权利要求6所述的地铁大客流预测系统，其特征在于，所述静态特征包括在线客流A i,k,c、站点累计客流Ac j,c、各时段历史平均客流、历史发生大客流事件的次数F j、站点平均花费时间Cd j；其中，各时段历史平均客流包括在线历史平均客流量、累计客流量，分别使用
    Figure PCTCN2019125437-appb-100014
    表示A i,k,c,Ac j,k对应的历史平均值；所述动态特征包括两站之间花费时间cst i,j、客流平均贡献率
    Figure PCTCN2019125437-appb-100015
  8. 根据权利要求7所述的地铁大客流预测系统，其特征在于，所述大客流聚集站点判断模块包括：
    客流聚集指数计算单元：用于基于实时在线客流与历史平均客流的偏移量实时估算各站点的客流聚集指数；假设在过去时段T k从站点s i进站的乘客，在时间段T c之后仍然在线的乘客数量相比历史平均值的偏移量
    Figure PCTCN2019125437-appb-100016
    大，则表示R i,k,c这部分乘客会在未来时段T c+m向别的站点聚集；如果地铁全网很多站点的客流都有在T c+m时段向站点s j聚集的趋势，则认为在未来时段T c+m的s j站会出现大客流事件；R i,k,c可能去往的站点以及所影响的时段与两站之间花费时间有关系；偏移量R i,k,c服从泊松分布R i,k,c～P(λ)，使用置信区间检验偏移量R i,k,c＞0是否异常，如果R i,k,c异常，则将R i,k,c＞0称为关键客流；如果R i,k,c是关键客流，假设R i,k,c去往站点s j，那么R i,k,c在未来时段T c+m到达站点s j的客流量，或R i,k,c对在未来时段T c+m在s j站点的发生大客流事件的贡献率计算为：
    Figure PCTCN2019125437-appb-100017
    未来时段T c+m在s j站的客流聚集指数GS c,j,m定义为：在T c+m-M～T c时段范围从其它站点进站的关键客流对在未来时段T c+m在站点s j的大客流贡献率之和，计算为：
    Figure PCTCN2019125437-appb-100018
  9. 根据权利要求8所述的地铁大客流预测系统，其特征在于，所述大客流聚集站点判断模块还包括：
    大客流聚集站点预测单元：用于根据客流聚集指数筛选出可能发生大客流的潜在聚集站点集合，然后根据潜在聚集站点集合中各个站点的静态特征以及动态特征建立Logit模型，判断将要发生大客流事件的具体站点；具体为：
    在当前时段T c，为了判断未来时段T c+m是否发生大客流事件，首先将客流聚集指数GS c,j,m大于阈值G max的前N B个站点作为可能发生大客流事件的站点，并添加到大客流聚集站点集合S B中；判断所述大客流聚集站点集合S B中各个站点在最近时段是否已经出现了客流聚集情况；如果是，则将该站点作为大客流事件发生站点；如果大客流聚集站点集合S B中各个站点在最近时段没有出现客流聚集情况，则计算S B中各站点发生大客流事件的概率，将概率最大的站点作为大客流发生的站点；对于s j∈S B中的每一个站点，客流聚集指数GS c,j,m可以反映其动态特征，结合站点s j的历史发生大客流事件的次数F j以及站点平均花费时间Cd j，并基于多概率选择模型logit计算各站点发生大客流事件的概率。
  10. 根据权利要求9所述的地铁大客流预测系统，其特征在于，所述大客流量预测模块根据发生大客流事件的具体站点对应的静态特征及动态特征预测该站点在未来时段的精确客流量具体包括：假设在时段T c时判定未来时段T c+m在s j站点将发生大客流事件，客流聚集指数为GS c,j,m，预测在未来时段T c+m流入站点s j的客流量D j,c+m；D j,c+m表示为平均客流量
    Figure PCTCN2019125437-appb-100019
    与偏移量Δ之和，即
    Figure PCTCN2019125437-appb-100020
    Δ表示为I c+m-N～I c时段的贡献量Δ p和I c+1～I c+m时段的贡献量Δ f两部分，假设在每一次大客流事件中，参与大客流聚集的乘客进站时间服从均匀分布，则Δ f/Δ的比值可以计算为：
    Figure PCTCN2019125437-appb-100021
    上述公式中，Pr Δ(t o∈[I c+m-N I c])表示在Δ中，在过去时段进站的乘客比例；如果已知大客流贡献率，Δ计算为Δ＝Δ p×θ；
    如果过去在站点s j发生过大客流事件，则所述大客流贡献率为历史平均大客流贡献率；如果过去在站点s j从来没有发生过大客流事件，则构建线性回归模型
    Figure PCTCN2019125437-appb-100022
    使用平均花费时间贡献率
    Figure PCTCN2019125437-appb-100023
    和客流平均贡献率
    Figure PCTCN2019125437-appb-100024
    估计站点s i的大客流贡献率；其中所述时间贡献率
    Figure PCTCN2019125437-appb-100025
    计算方法为：
    Figure PCTCN2019125437-appb-100026
  11. 一种电子设备，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述1至5任一项所述的地铁大客流预测方法的以下操作：
    步骤a：根据历史乘客出行数据提取地铁静态特征及动态特征；
    步骤b：基于地铁实时在线客流与历史平均客流的偏移量计算各站点的客流聚集指数，并结合各站点的客流聚集指数、静态特征以及动态特征判断未来时段地铁网络是否会发生大客流事件，以及会发生大客流事件的具体站点；
    步骤c：根据所述会发生大客流事件的具体站点对应的静态特征及动态特征 预测该站点在未来时段的精确客流量。
PCT/CN2019/125437 2019-10-18 2019-12-14 一种地铁大客流预测方法、系统及电子设备 WO2021072959A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910993773.7 2019-10-18
CN201910993773.7A CN112686417B (zh) 2019-10-18 2019-10-18 一种地铁大客流预测方法、系统及电子设备

Publications (1)

Publication Number Publication Date
WO2021072959A1 true WO2021072959A1 (zh) 2021-04-22

Family

ID=75445108

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/125437 WO2021072959A1 (zh) 2019-10-18 2019-12-14 一种地铁大客流预测方法、系统及电子设备

Country Status (2)

Country Link
CN (1) CN112686417B (zh)
WO (1) WO2021072959A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113536932A (zh) * 2021-06-16 2021-10-22 中科曙光国际信息产业有限公司 人群聚集预测方法、装置、计算机设备和存储介质
CN118036868A (zh) * 2024-01-17 2024-05-14 西藏沐风文化有限公司 智能化会展客流分析与调控方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113393029B (zh) * 2021-06-11 2023-11-17 上海逸迅信息科技有限公司 一种轨交客流量的预测方法和设备
CN115564151A (zh) * 2022-12-06 2023-01-03 成都智元汇信息技术股份有限公司 一种基于形态识别的突发大客流形态识别方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105427594A (zh) * 2015-11-23 2016-03-23 青岛海信网络科技股份有限公司 一种基于双向上车客流的公交断面客流量获取方法及系统
CN108846514A (zh) * 2018-06-07 2018-11-20 同济大学 一种突发事件下轨道交通客流疏散需求预测方法
CN109308543A (zh) * 2018-08-20 2019-02-05 华南理工大学 基于ls-svm和实时大数据的地铁短期客流预测方法
CN109858681A (zh) * 2019-01-04 2019-06-07 深圳北斗应用技术研究院有限公司 一种基于ic卡的交通短时客流预测方法和相关装置

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6872331B2 (ja) * 2016-09-09 2021-05-19 株式会社日立製作所 評価システム及び評価方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105427594A (zh) * 2015-11-23 2016-03-23 青岛海信网络科技股份有限公司 一种基于双向上车客流的公交断面客流量获取方法及系统
CN108846514A (zh) * 2018-06-07 2018-11-20 同济大学 一种突发事件下轨道交通客流疏散需求预测方法
CN109308543A (zh) * 2018-08-20 2019-02-05 华南理工大学 基于ls-svm和实时大数据的地铁短期客流预测方法
CN109858681A (zh) * 2019-01-04 2019-06-07 深圳北斗应用技术研究院有限公司 一种基于ic卡的交通短时客流预测方法和相关装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113536932A (zh) * 2021-06-16 2021-10-22 中科曙光国际信息产业有限公司 人群聚集预测方法、装置、计算机设备和存储介质
CN118036868A (zh) * 2024-01-17 2024-05-14 西藏沐风文化有限公司 智能化会展客流分析与调控方法

Also Published As

Publication number Publication date
CN112686417A (zh) 2021-04-20
CN112686417B (zh) 2024-03-19

Similar Documents

Publication Publication Date Title
WO2021072959A1 (zh) 一种地铁大客流预测方法、系统及电子设备
Toqué et al. Short & long term forecasting of multimodal transport passenger flows with machine learning methods
CN104064028B (zh) 基于多元信息数据的公交到站时间预测方法及系统
CN103247177B (zh) 大规模路网交通流实时动态预测系统
CN111079875A (zh) 基于多源数据的公共交通客流监测方法、装置和存储介质
CN105405293B (zh) 一种道路旅行时间短期预测方法和系统
WO2018023331A1 (zh) 一种常规公交服务指数实时评价系统及评价方法
CN110942411A (zh) 基于区域联动的地铁车站客流分级报警方法
CN106448132A (zh) 一种常规公交服务指数实时评价系统及评价方法
CN109544908A (zh) 一种实时预测公交到站时间的方法、电子设备及存储介质
CN106777703A (zh) 一种公交乘客实时分析系统及其构建方法
CN102324128A (zh) 基于ic卡记录的公交站间od客流预测方法及装置
CN112884325A (zh) 一种客站设备应用分析及健康状况评价方法及系统
CN106327864A (zh) 一种基于高速公路联网收费数据的交通流估计方法
CN107330540A (zh) 一种考虑电压质量的配电网台区缺供电量预测方法
CN109816978B (zh) 考虑驾驶员动态响应行为的区域群体交通诱导系统及方法
CN108717583B (zh) 一种城市轨道交通区间中断条件下车站滞留旅客量实时预测方法
CN113468243A (zh) 地铁客流分析与预测方法及分析与预测系统
CN106650209A (zh) 一种车辆应用实时信息确定可靠性增长趋势及参数的方法
CN111367900B (zh) 基于afc数据的城市轨道交通网络常态限流强度的计算方法
TW201608518A (zh) 一種雲端智慧型公共自行車泊車調配系統
CN108197078A (zh) 一种基于公交ic卡数据计算公交断面客流量的方法
CN114419894B (zh) 一种路内停车泊位设置与使用监测的方法与系统
CN109523788B (zh) 一种路段运行效果评价方法
CN116384807A (zh) 基于模糊综合评价模型的航路网络拥挤态势评估方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19949459

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19949459

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19949459

Country of ref document: EP

Kind code of ref document: A1