Urban rail transit passenger flow prediction method under rapid road network change condition

Abstract

The invention relates to an urban rail transit passenger flow prediction method under the condition of rapid road network change, which is oriented to a small-scale urban rail transit network with a rapid road network structure change, establishes a historical daily passenger flow mode, and extracts key information of a prediction day. By calculating the similarity of indexes such as road network conditions, date attributes, large-scale activity conditions, week types, seasons, weather, temperatures, wind power and interval durations of a prediction day and historical days, the comprehensive similarity of all the historical days is synthesized, and a plurality of historical days with the highest comprehensive similarity value are selected as feature days; introducing a road network change coefficient by considering a road network change condition on the basis of the passenger flow of the characteristic day, introducing a natural increase coefficient by considering a short-term increase condition of the passenger flow, and finally synthesizing the passenger flow of the required prediction day.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban rail transit operation management, and in particular relates to a method for predicting urban rail transit passenger flow under conditions of rapid changes in road networks. Background Art

[0002] Understanding the passenger flow characteristics of urban rail transit is the basis for compiling train operation diagrams and optimizing station passenger flow organization. Large-scale urban rail transit networks are relatively stable, and existing passenger flow prediction methods can achieve good results. However, for small-scale urban rail transit networks, due to the rapid changes in their network structure and the rapid increase in passenger flow, existing passenger flow prediction methods cannot complete the task of passenger flow prediction well. Summary of the invention

[0003] Based on this, the purpose of the present invention is to provide a method for predicting passenger flow of urban rail transit under the condition of rapid changes in the road network, and to use rich historical automatic ticket vending and checking system (Auto Fare Collection, referred to as AFC) data and historical operation data of urban rail transit to complete the passenger flow prediction work on important operating dates in the near and short term, so as to guide the work of station passenger flow organization, operation diagram compilation, etc., and improve the level of refined and intelligent operation of urban rail transit.

[0004] The above technical objectives of the present invention are achieved through the following technical solutions:

[0005] A method for predicting passenger flow of urban rail transit under the condition of rapid changes in road network, comprising:

[0006] Get the forecast day feature information;

[0007] Establish the passenger flow model of historical days, and synthesize the comprehensive similarity of each historical day by calculating the similarity of various characteristic indicators between historical dates and predicted days;

[0008] Several historical days with the highest comprehensive similarity values ​​are selected as feature days. Based on the passenger flow on the feature days, the road network change coefficient is introduced considering the road network changes, and the natural growth coefficient is introduced considering the short-term growth of passenger flow, and finally the passenger flow of the required forecast day is synthesized.

[0009] Optionally, the predicted day characteristic information includes: 0-1 type indicators, qualitative indicators and quantitative indicators.

[0010] The 0-1 type indicators are divided into two types: special days and ordinary days. Special days refer to holidays and the day before holidays. Their passenger flow patterns are quite different from those of ordinary days nearby, but their changing trends are similar to those of the same period or similar holidays in previous years. Ordinary days refer to days other than special days with little difference in passenger flow changes from adjacent days.

[0011] Qualitative indicators include what kind of large-scale events are held, the type of week, season, and weather;

[0012] Quantitative indicators include the maximum temperature during the operating period, the minimum temperature during the operating period, the wind speed during the operating period, and the interval duration.

[0013] Optionally, when calculating the similarity of a 0-1 type indicator, the similarity is 1 if the historical date and the predicted date are both special days or both ordinary days, otherwise it is 0.

[0014] Optionally, when calculating the similarity of qualitative indicators, the historical passenger flow data is classified and aggregated according to different values ​​of the qualitative indicators and dimensionally normalized, and the similarity between any two values ​​is calculated by the following formula, and finally all similarities are sorted into a similarity matrix;

[0015]

[0016] Among them, r we is the similarity of the qualitative indicator classification type,

[0017] k is the number of days of historical passenger flow data,

[0018] x m is the dimensionally normalized value of passenger flow of classification type p,

[0019] x n is the dimensionally normalized value of passenger flow for classification type q.

[0020] Optionally, for the quantitative indicator classification types of maximum temperature, minimum temperature during operation, and wind speed during operation, the similarity value is calculated by the following formula, and the quantitative indicator has a stronger impact on the travel choice of pedestrians, and its similarity calculation formula is designed as a piecewise function form;

[0021] r Th =1-a k |T i，h -T j，h |

[0022] Among them, r Th is the similarity of the quantitative indicator classification type,

[0023] T i，h is the dimensionally normalized value of the classification type of the operation period on the i-th day,

[0024] T j，h is the dimensionally normalized value of the classification type on the predicted day j,

[0025] a k is the influence coefficient, k is 1, 2; the classification type of the operation period is compared with the threshold. When the passenger flow is sensitive to the classification type, k is 2, ak Take 0.020; when the passenger flow is not sensitive to the classification type, k is 1, a k Take 0.001;

[0026] For quantitative indicators, the interval between the historical day and the predicted day is a very important indicator for calculating similarity. The closer to the predicted day, the more similar the passenger flow characteristics are. The similarity slowly decays as the distance from the predicted day increases. It can be calculated by the following formula:

[0027] r Di =w1 int(d / 7) w2 mod(d / 7)

[0028] Among them, r Di is the similarity of the interval duration indicator,

[0029] d is the number of days from the i-th historical day to the predicted day,

[0030] w1 is the weekly similarity attenuation ratio,

[0031] w2 is the daily similarity attenuation ratio.

[0032] Optionally, for each historical day, the similarities of all the calculated indicators are synthesized by the following formula to obtain a comprehensive similarity, and the n similar days with the highest values ​​are selected as historical characteristic days. According to actual application experience, when making passenger flow predictions, 4 to 6 similar days can be selected for ordinary days, and 1 to 2 similar days can be selected for special days;

[0033]

[0034] Among them, R i is the comprehensive similarity of historical days,

[0035] r i，m is the similarity between the historical day i and the predicted day in terms of the mth passenger flow influencing factor,

[0036] k m is the weight coefficient of the mth passenger flow influencing factor;

[0037] Based on the passenger flow of the selected n historical characteristic days, the quadratic exponential smoothing method is used to predict the passenger flow of the forecast day. If the road network structure of the similar day is different from that of the historical characteristic day, the road network change coefficient must be introduced to correct the passenger flow value; if the year or month of the forecast day is different from that of the historical characteristic day, the natural growth coefficient must be introduced to correct the passenger flow value.

[0038] Optionally, when the similar day and the predicted day are in different months, a monthly growth coefficient must be introduced to calculate the monthly growth coefficient between months based on the average passenger flow of each month in previous years under the same road network conditions:

[0039]

[0040] Among them, r 月 is the monthly growth coefficient,

[0041] is the average passenger flow in the i-th month of the year;

[0042] When the similar day and the predicted day belong to different years, the annual growth coefficient must be introduced and the annual growth coefficient of each year shall be calculated based on the average passenger flow of the actual date of the first month of each year under the same road network conditions;

[0043]

[0044] Among them, r 年 is the annual growth coefficient,

[0045] F i is the passenger flow on the ith day of the first month of the year,

[0046] m is the total number of days that actually generated passenger flow in the first month of year a,

[0047] n is the total number of days that actually generated passenger traffic in the first month of year b;

[0048] When the road network structure of the similar day is different from that of the predicted day, the road network change coefficient is introduced. For the starting and ending stations that exist before and after the road network change, there are OD pairs. The road network change coefficient is determined by the average daily passenger flow of the week before the current day and the average daily passenger flow of the week before the current day in the same period of previous years. The predicted passenger flow is corrected based on the passenger flow of the similar day calculated in the previous step.

[0049]

[0050] Q 修正 =Q 相似日 ·θ 既有

[0051] Among them, θ 既有 is the road network variation coefficient of the existing OD pairs,

[0052] Q t is the passenger flow of the ith adjacent day before the current day,

[0053] q t is the passenger flow of the i-th adjacent day in the same period of the previous year before the current day,

[0054] Q 修正 is the revised forecast daily passenger flow,

[0055] Q 相似日is the passenger flow on similar days calculated in the previous step.

[0056] For the newly added OD pairs after the road network changes, the road network change coefficient is determined by the daily passenger flow of the same period in previous years on the forecast day and the average daily passenger flow of the week before the current day. The forecast passenger flow is corrected based on the average daily passenger flow of the week before the current day.

[0057]

[0058] Q 修正 =Q 邻近日 `θ 新增

[0059] Among them, θ 新增 is the network change coefficient of the newly added OD pairs,

[0060] q 预测日 is the passenger flow on the same day in previous years during the same period of the forecast day.

[0061] q i is the passenger flow of the i-th adjacent day from the current day in the same period of previous years,

[0062] Q 修正 is the revised forecast daily passenger flow,

[0063] Q 邻近日 It is the average daily passenger flow in the week before the current day.

[0064] Due to the adoption of the above technical scheme, the beneficial effects of the present invention include: the present invention can well solve the characteristics of rapid changes in the network structure of small-scale urban rail transit, large fluctuations in passenger flow, and rapid increase in passenger flow attraction. The introduction of the road network change coefficient and the natural growth coefficient has significantly improved the accuracy of passenger flow prediction on holidays, and the error of the overall passenger flow prediction result is within the acceptable range for daily operation and management of urban rail transit. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a route framework diagram for predicting passenger flow of urban rail transit in a small-scale road network based on historical characteristic days.

[0066] Figure 2 It is a table of historical daily passenger flow patterns.

[0067] Figure 3 is the week type indicator similarity matrix.

[0068] Figure 4 It is a block diagram of the technical solution for correcting the passenger flow on historical characteristic days.

[0069] Figure 5 This is a schematic diagram of the existing operating lines of Nanchang Metro.

[0070] Figure 6 is a fitting curve diagram of the time-sharing entry and exit volume of Weidong Station. (a) is the fitting curve diagram of the time-sharing entry volume of Weidong Station; (b) is the fitting curve diagram of the time-sharing exit volume of Weidong Station. DETAILED DESCRIPTION

[0071] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0072] First, the various passenger flow pattern indicators for the forecast date and the historical dates within a certain period of time before are sorted into a table (see Figure 2 ), including historical day number (No), date (Date), whether it is a holiday (Holiday), week type (Week_Type), season (Season), weather (Weather), what kind of large-scale activities are held (Activity), the highest temperature during the operating period (High_Tem), the lowest temperature during the operating period (Low_Tem), and the wind power during the operating period (Wind_Power).

[0073] Secondly, the above indicators are divided into three categories according to data characteristics: 0-1 type indicators (Holiday), qualitative indicators (Activity, Week_Type, Season, Weather), and quantitative indicators (High_Tem, Low_Tem, Wind_Power, and interval_length).

[0074] (a) For the 0-1 type indicator, whether it belongs to a holiday or not is divided into two types: special days and ordinary days. Special days refer to holidays and the day before holidays. Their passenger flow patterns are usually quite different from those of ordinary days nearby, but their changing trends are similar to those of the same period or similar holidays in previous years; ordinary days refer to days with smaller differences in passenger flow changes from adjacent days except for special days. When calculating similarity, the similarity is 1 for both special days and both ordinary days, and 0 otherwise.

[0075] (b) For the qualitative indicator Week_Type, the historical passenger flow data is classified and aggregated according to the different values ​​of the weekly indicator (Sunday, Monday, Tuesday, Wednesday, Thursday, Friday, Saturday) and dimensionally normalized. The similarity between any two values ​​is calculated using formula (1). Finally, all similarities are organized into a similarity matrix (taking the similarity of week type as an example, see Figure 3 ).

[0076]

[0077] Among them, r we is the similarity of the week type indicator,

[0078] k is the number of days of historical passenger flow data,

[0079] x m is the dimensionally normalized value of the passenger flow on the day of week type p,

[0080] x n is the dimensionally normalized value of the passenger flow on the day of week type q,

[0081] m, n are the values ​​of the week type (Monday is 1,..., Sunday is 7).

[0082] (1) p and q are not dates, but classification types of qualitative indicators. p can be regarded as the prediction day and q as the historical day. For example, if the qualitative indicator is the day of the week, then p and q in the calculation formula are the days of the week, so as to calculate the indicator similarity between the days of the week.

[0083] (2) There are many methods for dimensional normalization. The most common one is min-max normalization, which scales all data to the range of 0-1 to facilitate comparison between data of different orders of magnitude.

[0084] For the qualitative indicator Season, the historical passenger flow data is classified and aggregated according to the different values ​​of the seasonal indicator (spring, summer, autumn, and winter) and dimensionally normalized. The similarity between any two values ​​is calculated using formula (2), and finally all similarities are organized into a similarity matrix.

[0085]

[0086] Among them, r se is the similarity of seasonal indicators,

[0087] k is the number of days of historical passenger flow data,

[0088] x m is the dimensionally normalized value of the passenger flow on the date with season type p,

[0089] s n is the dimensionally normalized value of the passenger flow on the date with the seasonal type q,

[0090] m, n are the values ​​of the season type (1,... for spring, 4 for winter).

[0091] For the qualitative indicator Weather, the historical passenger flow data is classified and aggregated according to different values ​​of the weather indicator (sunny, cloudy, rainy, foggy, and hazy) and dimensionally normalized. The similarity between any two values ​​is calculated using formula (3), and finally all similarities are organized into a similarity matrix.

[0092]

[0093] Among them, r Wea is the similarity of weather indicators,

[0094] k is the number of days of historical passenger flow data,

[0095] x m is the dimensionally normalized value of passenger flow on a date with weather type p,

[0096] x n is the dimensionally normalized value of passenger flow on a date with weather type q,

[0097] m, n are the values ​​of weather type (1 for clear, ..., 5 for haze).

[0098] For the qualitative indicator of what kind of large-scale activities are held, the historical passenger flow data is classified and aggregated according to different values ​​of the large-scale activity indicators (no activities, general activities, large-scale activities (non-holiday festivals / concerts, etc.)) and dimensionally normalized. The similarity between any two value conditions is calculated using formula (4), and finally all similarities are organized into a similarity matrix.

[0099]

[0100] Among them, r ac is the similarity of the indicators of what kind of large-scale events are held,

[0101] k is the number of days of historical passenger flow data,

[0102] x m is the dimensionally normalized value of the passenger flow on the date of large-scale event type p,

[0103] x n is the dimensionally normalized value of the passenger flow on the date of large-scale event type q,

[0104] m, n are the values ​​of the large-scale activity type (1 for no activity, 2 for general activity, and 3 for large-scale activity).

[0105] (c) For the quantitative indicator, the maximum temperature during the operating period (High_Tem), the similarity value can be calculated using formula (5). The maximum daily temperature has a stronger impact on pedestrians’ travel choices, so the similarity calculation formula is designed as a piecewise function.

[0106] r Th =1-a k |T i，h -T j，h | (5)

[0107] Among them, r This the similarity of the highest temperature index during the operation period,

[0108] T i，h is the dimensionally normalized value of the maximum temperature during the operation period of the i-th day,

[0109] T j，h is the dimensionally normalized value of the predicted daily maximum temperature,

[0110] a k is the influence coefficient (k is 1, 2). When the highest temperature during the operation period is higher than the high temperature threshold, the passenger flow is sensitive to the temperature. k is 2, a k Take 0.020; when the highest temperature during the operation period is lower than the high temperature threshold, the passenger flow is not sensitive to the temperature, k is 1, a k Take 0.001.

[0111] For the quantitative indicator, the lowest temperature during the operation period (Low_Tem), the similarity calculation formula is designed as a piecewise function form. The similarity value can be calculated by formula (6):

[0112] r Ti =1-a k |T i，i -T j，i | (6)

[0113] Among them, r Ti is the similarity of the lowest temperature index during the operation period,

[0114] T i，i is the dimensionally normalized value of the lowest temperature during the operation period of the i-th day,

[0115] T j，i is the dimensionally normalized value of the predicted daily minimum temperature,

[0116] a k is the influence coefficient (k is 1, 2). When the lowest temperature during the operation period is lower than the low temperature threshold, the passenger flow is sensitive to the temperature. k is 2, a k Take 0.020; when the lowest temperature during the operation period is higher than the low temperature threshold, the passenger flow is not sensitive to the temperature, k is 1, a k Take 0.001.

[0117] For the quantitative indicator wind power (Wind_Power) during the operation period, the similarity value can be calculated by formula (7), and its similarity calculation formula is designed as a piecewise function form.

[0118] r Wi =1-a k |W i，wi -W j，wi | (7)

[0119] Among them, r Wi is the similarity of wind index during the operation period,

[0120] W i，wi is the dimensionally normalized value of the wind force during the operation period of the i-th day,

[0121] W j，wi is the dimensionally normalized value of the predicted daily wind force,

[0122] a k is the influence coefficient (k is 1 or 2). When the wind speed during the operation period is higher than the wind speed threshold, the passenger flow is sensitive to the wind speed. k is 2, and a k Take 0.020; when the wind speed is lower than the wind speed threshold during the operation period, the passenger flow is not sensitive to the wind speed, k is 1, a k Take 0.001.

[0123] For quantitative indicators, the interval length (Interval_length) between the historical day and the predicted day is a very important indicator for calculating similarity. The closer to the predicted day, the more similar the passenger flow characteristics are. The similarity slowly decays as the distance from the predicted day increases. It can be calculated using formula (8).

[0124] r Di =w1 int(d / 7) w2 mod(d / 7) (8)

[0125] Among them, r Di is the similarity of the interval duration indicator,

[0126] d is the number of days from the i-th historical day to the predicted day,

[0127] w1 is the weekly similarity attenuation ratio,

[0128] w2 is the daily similarity attenuation ratio.

[0129] In summary, for each historical day, the similarities of all the calculated indicators are synthesized through formula (9) to obtain the comprehensive similarity, and the n similar days with the highest values ​​are selected as the historical characteristic days. According to actual application experience, when making passenger flow predictions, 4 to 6 similar days can be selected for ordinary days, and 1 to 2 similar days can be selected for special days.

[0130]

[0131] Among them, R i is the comprehensive similarity of historical days,

[0132] r i，m is the similarity between the historical day i and the predicted day in terms of the mth passenger flow influencing factor,

[0133] k m is the weight coefficient of the mth passenger flow influencing factor.

[0134] Finally, based on the passenger flow of the selected n historical characteristic days, the quadratic exponential smoothing method is used to predict the passenger flow of the forecast day. If the road network structure of the similar day is different from that of the historical characteristic day, the road network change coefficient must be introduced to correct the passenger flow value; if the year or month of the forecast day is different from that of the historical characteristic day, the natural growth coefficient must be introduced to correct the passenger flow value (see Figure 4 ).

[0135] When the similar day and the predicted day are in different months, the monthly growth coefficient must be introduced and the monthly growth coefficient between months must be calculated based on the average passenger flow in previous months under the same road network conditions (Formula 10).

[0136]

[0137] Among them, r 月 is the monthly growth coefficient,

[0138] is the average passenger flow in month i of the year.

[0139] When the similar day and the predicted day belong to different years, the annual growth coefficient must be introduced. The annual growth coefficient for each year is calculated based on the average passenger flow on the actual date of the first month of each year under the same road network conditions (Formula 11).

[0140]

[0141] Among them, r 年 is the annual growth coefficient,

[0142] F i is the passenger flow on the ith day of the first month of the year,

[0143] m is the total number of days that actually generated passenger flow in the first month of year a,

[0144] n is the total number of days that actually generate passenger traffic in the first month of year b.

[0145] When the road network structure of the similar day is different from that of the predicted day, the road network change coefficient is introduced. For the origin-destination (OD) stations that exist before and after the road network change, that is, the existing OD pairs, the road network change coefficient is determined by the average daily passenger flow in the week before the current day and the average daily passenger flow in the week before the current day in the same period of previous years (Formula 12). The predicted passenger flow is corrected based on the passenger flow on the similar day calculated in the previous step (Formula 13).

[0146]

[0147] Q 修正 =Q 相似日 ·θ 既有 (13)

[0148] Among them, θ 既有 is the road network variation coefficient of the existing OD pairs,

[0149] Q i is the passenger flow of the ith adjacent day before the current day,

[0150] q i is the passenger flow of the i-th adjacent day in the same period of the previous year before the current day,

[0151] Q 修正 is the revised forecast daily passenger flow,

[0152] Q 相似日 is the passenger flow on similar days calculated in the previous step.

[0153] For the newly added OD pairs after the road network changes, the road network change coefficient is determined by the daily passenger flow on the forecast day in the same period of previous years and the average daily passenger flow in the week before the current day in the same period of previous years (Formula 14). The forecast passenger flow is corrected based on the average daily passenger flow in the week before the current day (Formula 15).

[0154]

[0155] Q 修正 =Q 邻近日 ·θ 新增 (15)

[0156] Among them, θ 新增 is the network change coefficient of the newly added OD pairs,

[0157] q 预测日 is the passenger flow on the same day in previous years during the same period of the forecast day.

[0158] q i is the passenger flow of the i-th adjacent day from the current day in the same period of previous years,

[0159] Q 修正 is the revised forecast daily passenger flow,

[0160] Q 邻近日 It is the average daily passenger flow in the week before the current day.

[0161] Example 1

[0162] Nanchang Metro is an urban rail transit system serving Nanchang City, Jiangxi Province, China. Its first line was opened for trial operation on December 26, 2015, making Nanchang the 25th city in mainland China and the first city in Jiangxi Province to open rail transit. As of December 2022, Nanchang Metro has opened 4 lines (see Figure 5 ), namely Nanchang Metro Line 1, Line 2, Line 3 and Line 4, with a total of 94 stations (including 9 transfer stations, not counted repeatedly), and a total operating line length of 128.45km. Among them, Metro Line 4 has an operating mileage far exceeding the first three metro lines, with a total length of 40km, and is an important line for Nanchang Metro to achieve network operation. At present, the average daily passenger volume of the line network has reached 712,300, and the highest daily passenger volume of the line network has reached 1.815 million.

[0163] November 22, 2021 (Monday) was selected as the prediction day, and the ten stations with the highest OD ratio in the Nanchang Metro network were selected as the prediction objects. The AFC data from March 2021 to June 2022 were used as the model training sample data to obtain the similarity matrix of each qualitative indicator (see Table 1-Table 5)

[0164] Table 1 - Weekly type indicator similarity matrix

[0165]

[0166] Table 2 - Seasonal indicator similarity matrix

[0167]

[0168] Table 3 - Weather indicator similarity matrix

[0169]

[0170] Table 4-Holiday indicator similarity matrix

[0171]

[0172] Table 5 - Large-scale event indicator similarity matrix

[0173]

[0174] The present invention calculates the similarity between each historical day and the predicted day, and takes the top five with the highest similarity as historical similar days, which are November 15, 2021, November 8, 2021, November 1, 2021, November 18, 2021, and November 17, 2021; the predicted daily passenger flow of each station is calculated based on the passenger flow of each station on the historical similar day, and the error is calculated by comparing with the actual passenger flow of the station on this basis; at the same time, the traditional prediction method is used to calculate the average passenger flow of the stations in the same period of history for comparison (see Table 6-Table 7). Among them, Metro Building Station, Bayi Pavilion Station, and Bayi Square Station are transfer stations, and the passenger flow statistics of the station in and out include data of two lines.

[0175] Table 6- Forecast results of station arrivals and comparison with historical averages for the same period

[0176]

[0177] Table 7- Forecast results of the number of outbound trips at each station and comparison with the historical average values ​​for the same period

[0178]

[0179] The passenger flow of the station during the morning and evening peaks plays an important role in the operation and management of urban rail transit. It is one of the important reference indicators for the organization of train operation and passenger transport management. By searching for the historical similar days corresponding to the predicted day, the passenger flow of each station can be predicted. The time-sharing (15 minutes as the time granularity) in and out of the station is calculated for the TOP10 stations with the highest OD share, and a trend curve is drawn with the time-sharing in and out of the station predicted by the historical similar days to observe the consistency of the passenger flow prediction model for each period of the day. Taking Weidong Station as an example (see Figure 6), the predicted curve is relatively consistent with the actual curve change trend.

[0180] The present invention can avoid the limitations of insufficient and discontinuous historical card swiping data sources in urban rail transit under a certain road network to a certain extent. After calculating the similarity based on the characteristic factors, it is possible to find the historical day with the highest similarity, thereby obtaining a refined basic passenger flow forecast value for the entire network, which can avoid the shortcomings of the traditional passenger flow forecast method that relies on the historical passenger flow of the same period, and has a higher accuracy rate than the traditional historical average method of the same period; it also has a good effect on the prediction of the time-sharing passenger flow throughout the day. Through observation, it can be found that the low predicted passenger flow is mainly reflected in the decrease in the peak value of the morning and evening peaks, but the predicted curve during the morning and evening peaks is more in line with the actual curve change trend, and the accuracy calculated by the average error value test model is concentrated within 8%.

[0181] The above-mentioned related explanations and descriptions of the embodiments are for the convenience of those skilled in the art to understand and apply the present invention. It is obvious that those familiar with the art can easily make various modifications to these contents and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above-mentioned related explanations and descriptions of the embodiments. Improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the protection scope of the present invention.

Claims

1. A method for predicting passenger flow of urban rail transit under conditions of rapid changes in the road network, characterized in that: include: Get the forecast day feature information; Establish the passenger flow model of historical days, and synthesize the comprehensive similarity of each historical day by calculating the similarity of various characteristic indicators between historical dates and predicted days; Several historical days with the highest comprehensive similarity values ​​are selected as feature days. Based on the passenger flow on the feature days, the road network change coefficient is introduced considering the road network changes, and the natural growth coefficient is introduced considering the short-term growth of passenger flow, and finally the passenger flow of the required forecast day is synthesized.

2. The method for predicting passenger flow of urban rail transit under the condition of rapid changes in road network according to claim 1 is characterized in that: The forecast day characteristic information includes: 0-1 type indicators, qualitative indicators and quantitative indicators. The 0-1 type indicators are divided into two types: special days and ordinary days. Special days refer to holidays and the day before holidays. Their passenger flow patterns are quite different from those of ordinary days nearby, but their changing trends are similar to those of the same period or similar holidays in previous years. Ordinary days refer to days other than special days with little difference in passenger flow changes from adjacent days. Qualitative indicators include what kind of large-scale events are held, the type of week, season, and weather; Quantitative indicators include the maximum temperature during the operating period, the minimum temperature during the operating period, the wind speed during the operating period, and the interval duration.

3. The method for predicting passenger flow of urban rail transit under the condition of rapid changes in road network according to claim 2 is characterized in that: When calculating the similarity of 0-1 type indicators, the similarity is 1 if the historical date and the predicted date are both special days or both ordinary days, otherwise it is 0.

4. The method for predicting passenger flow of urban rail transit under the condition of rapid changes in road network according to claim 2 is characterized in that: When calculating the similarity of qualitative indicators, the historical passenger flow data is classified and aggregated according to different values ​​of qualitative indicators and dimensionally normalized. The similarity between any two values ​​is calculated using the following formula, and finally all similarities are organized into a similarity matrix; Among them, r we is the similarity of the qualitative indicator classification type, k is the number of days of historical passenger flow data, x m is the dimensionally normalized value of passenger flow of classification type p, x n is the dimensionally normalized value of passenger flow for classification type q.

5. The method for predicting passenger flow of urban rail transit under the condition of rapid changes in road network according to claim 2 is characterized in that: For the quantitative indicator classification types of maximum temperature, minimum temperature during operation, and wind speed during operation, the similarity values ​​are calculated using the following formula. Since quantitative indicators have a stronger impact on pedestrians' travel choices, the similarity calculation formula is designed as a piecewise function form. r Th =1-a k |T i，h -T j，h | Among them, r Th is the similarity of the quantitative indicator classification type, T i，h is the dimensionally normalized value of the classification type of the operation period on the i-th day, T j，h is the dimensionally normalized value of the classification type on the predicted day j, a k is the influence coefficient, k is 1, 2; the classification type of the operation period is compared with the threshold. When the passenger flow is sensitive to the classification type, k is 2, a k Take 0.020; when the passenger flow is not sensitive to the classification type, k is 1, a k Take 0.001; For quantitative indicators, the interval between the historical day and the predicted day is a very important indicator for calculating similarity. The closer to the predicted day, the more similar the passenger flow characteristics are. The similarity slowly decays as the distance from the predicted day increases. It can be calculated by the following formula: r Di =w1 int(d / 7) w2 mod(d / 7) Among them, r Di is the similarity of the interval duration indicator, d is the number of days from the i-th historical day to the predicted day, w1 is the weekly similarity attenuation ratio, w2 is the daily similarity attenuation ratio.

6. The method for predicting passenger flow of urban rail transit under the condition of rapid changes in road network according to claim 2 is characterized in that: For each historical day, the similarities of all the calculated indicators are synthesized by the following formula to obtain the comprehensive similarity, and the n similar days with the highest values ​​are selected as the historical characteristic days. According to actual application experience, when making passenger flow predictions, 4 to 6 similar days can be selected for ordinary days, and 1 to 2 similar days can be selected for special days; Among them, R i is the comprehensive similarity of historical days, r i,m is the similarity between the historical day i and the predicted day in terms of the mth passenger flow influencing factor, k m is the weight coefficient of the mth passenger flow influencing factor; Based on the passenger flow of the selected n historical characteristic days, the quadratic exponential smoothing method is used to predict the passenger flow of the forecast day. If the road network structure of the similar day is different from that of the historical characteristic day, the road network change coefficient must be introduced to correct the passenger flow value; if the year or month of the forecast day is different from that of the historical characteristic day, the natural growth coefficient must be introduced to correct the passenger flow value.

7. The method for predicting passenger flow of urban rail transit under the condition of rapid changes in the road network according to claim 6, characterized in that: When the similar day and the predicted day are in different months, the monthly growth coefficient must be introduced. The monthly growth coefficient between months is calculated based on the average passenger flow of each month in previous years under the same road network conditions: Among them, r 月 is the monthly growth coefficient, is the average passenger flow in the i-th month of the year; When the similar day and the predicted day belong to different years, the annual growth coefficient must be introduced and the annual growth coefficient of each year shall be calculated based on the average passenger flow of the actual date of the first month of each year under the same road network conditions; Among them, r 年 is the annual growth coefficient, F i is the passenger flow on the ith day of the first month of the year, m is the total number of days that actually generated passenger flow in the first month of year a, n is the total number of days that actually generated passenger traffic in the first month of year b; When the road network structure of the similar day is different from that of the predicted day, the road network change coefficient is introduced. For the starting and ending stations that exist before and after the road network change, there are OD pairs. The road network change coefficient is determined by the average daily passenger flow of the week before the current day and the average daily passenger flow of the week before the current day in the same period of previous years. The predicted passenger flow is corrected based on the passenger flow of the similar day calculated in the previous step. Q 修正 =Q 相似日 ·θ 既有 Among them, θ 既有 is the road network variation coefficient of the existing OD pairs, Q i is the passenger flow of the ith adjacent day before the current day, q i is the passenger flow of the i-th adjacent day in the same period of the previous year before the current day, Q 修正 is the revised forecast daily passenger flow, Q 相似日 is the passenger flow on similar days calculated in the previous step. For the newly added OD pairs after the road network changes, the road network change coefficient is determined by the daily passenger flow of the same period in previous years on the forecast day and the average daily passenger flow of the week before the current day. The forecast passenger flow is corrected based on the average daily passenger flow of the week before the current day. Q 修正 =Q 邻近日 ·θ 新增 Among them, θ 新增 is the network change coefficient of the newly added OD pairs, q 预测日 is the passenger flow on the same day in previous years during the same period of the forecast day. q i is the passenger flow of the i-th adjacent day from the current day in the same period of previous years, Q 修正 is the revised forecast daily passenger flow, Q 领近日 It is the average daily passenger flow in the week before the current day.

Images