A passenger flow multi-source data statistical method for bus doorless signal

Abstract

The application discloses a kind of for bus no door signal passenger flow multi-source data statistics method, it is related to data statistical technical field, including based on target vehicle in operation plan time window inside multiple source time series data, identify multiple candidate boundary determination events, and give each event a quantitative confidence evaluation value;For current shift, respectively based on the confidence evaluation value, for the first station operation boundary and the last station operation boundary independently generates an ordered candidate event list, the event in list is arranged according to its confidence evaluation value high-low order;With the goal of eliminating the passenger flow data mutual string between the first and last station non-operation period, an optimization objective function is constructed;This method identifies boundary events through multi-source data fusion, even in the case of no door signal, the operation boundary can also be accurately positioned, thereby finally eliminating the data mutual string problem of non-operation period.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data statistics, and specifically relates to a passenger flow multi-source data statistical method for a bus without a door signal. BACKGROUND

[0002] In the intelligent bus operation management, accurately counting the passenger flow of each bus is the core basis for line optimization, scheduling, benefit analysis and government financial subsidy accounting. At present, buses equipped with automatic passenger counters have become the industry mainstream. In an ideal situation, such a system can accurately associate and intercept the passenger flow data of the vehicle during each operation shift (from the starting station to the terminal station) by detecting the door opening and closing signal, thereby realizing the passenger flow statistics at the shift level.

[0003] However, in the reality of bus operation in cities and urban and rural areas in China, there is a long-standing technical problem that has not been properly solved: a large number of buses do not output or cannot stably output the collectable door opening and closing signal during operation (i.e., the "no door signal" phenomenon). This results in that the automatic passenger counter can continuously record the cumulative passenger flow, but cannot automatically and accurately identify the accurate starting point and ending point of an operation shift, so that it is difficult to correctly segment the continuous passenger flow data to the corresponding shift.

[0004] Therefore, under the reality condition of "no door signal", how to dynamically and accurately define the boundary of the operation shift has become a key technical bottleneck for reliable passenger flow statistics. SUMMARY

[0005] The purpose of the present application is to provide a passenger flow multi-source data statistical method for a bus without a door signal to solve the problems mentioned in the background.

[0006] A passenger flow multi-source data statistical method for a bus without a door signal, comprising:

[0007] Based on the multi-source time series data of the target vehicle within the operation plan time window, a plurality of candidate boundary determination events are identified, and each event is assigned a quantitative confidence evaluation value;

[0008] For the current shift, an ordered candidate event list is independently generated for the first station operation boundary and the last station operation boundary based on the confidence evaluation value, and the events in the list are arranged in order of their confidence evaluation value from high to low;

[0009] An optimization objective function is constructed to eliminate the mutual stringing of passenger flow data between the first and last stations.

[0010] Based on the target function, an optimal combination of a first station boundary event and a terminal station boundary event is cooperatively searched and determined from the ordered list generated for the first station operating boundary and the ordered list generated for the terminal station operating boundary;

[0011] According to the time points corresponding to the first station boundary event and the terminal station boundary event in the optimal combination, the readings of the on-board passenger flow meter are obtained and the difference is calculated, and the passenger flow statistical result belonging to the current shift is output.

[0012] The terms of the method are defined as follows: multi-source time series data refers to multi-type data with timestamps collected in real time during the operation of the target vehicle, specifically including on-board positioning data (latitude and longitude, speed), on-board power state data (ignition and extinguishing signals), operation scheduling system data (planned departure time, planned arrival time);

[0013] The boundary determination event refers to a specific state change event that can represent the operating boundary (first station departure, terminal station arrival), such as the vehicle changing from stationary to continuous driving, the ignition signal being triggered and the positioning being within the first station range, or the speed dropping to zero and the positioning being within the terminal station range for more than 5 minutes;

[0014] The confidence evaluation value is used to quantify the credibility of a candidate event being a real operating boundary, with a value range of 0 to 1, and a larger value indicating higher credibility;

[0015] The first station operating boundary is the time point of the actual departure of the current shift at the first station, and the terminal station operating boundary is the time point of the actual arrival of the current shift at the terminal station;

[0016] The non-operating period passenger flow data mutual string refers to the phenomenon that the passenger flow data during the non-operating time of the current shift (such as passengers boarding before the first station departure and passengers alighting after the terminal station arrival) is included in the statistical result of the current shift.

[0017] Preferably, the calculation of the confidence evaluation value includes the basic weight of the event type, the real-time signal quality of the data source, and the absolute time difference between the event occurrence time and the planned time.

[0018] Preferably, cooperatively searching and determining an optimal combination of a first station boundary event and a terminal station boundary event specifically includes:

[0019] The first candidate pair is formed by the first-ranked event in the first station boundary ordered list and the first-ranked event in the terminal station boundary ordered list;

[0020] The target function value of the first candidate pair is calculated;

[0021] If the value is higher than a preset threshold, the first candidate pair is determined as the optimal combination;

[0022] If the value is lower than the preset threshold, the first-ranked event in the first station list and the second-ranked event in the last station list are combined to form a second candidate pair, and / or the second-ranked event in the first station list and the first-ranked event in the last station list are combined to form a third candidate pair, and the objective function values of the candidate pairs are calculated, and the candidate pair with the highest objective function value and higher than the preset threshold is selected as the optimal combination.

[0023] Preferably, an effectiveness evaluation function for evaluating the effectiveness of the candidate event combination in isolating non-operational activities is constructed, and the function is used to calculate the auxiliary score of the candidate combination, including:

[0024] The length of the time period defined by the candidate first and last station event pair is calculated and compared with the length of the standard operational time period based on the planned time, and the smaller the length deviation is, the higher the score contribution is;

[0025] It is checked whether the time period contains a location event of the vehicle entering a non-operational area, and if yes, the score of the combination is substantially deducted.

[0026] Preferably, when identifying the boundary determination event, for the vehicle ignition event, if the occurrence location is not within the preset operational departure yard electronic fence, the confidence evaluation value is set to zero.

[0027] Preferably, when the objective function values of the first candidate pair, the second candidate pair and the third candidate pair are all lower than the preset threshold, the following steps are performed:

[0028] The ordered lists of the first station boundary and the last station boundary are merged into a joint event sequence;

[0029] In the joint event sequence, a continuous event pair capable of forming a complete operational time period and having the highest objective function value is searched as the optimal combination.

[0030] Preferably, the setting logic of the preset threshold is specifically:

[0031] Based on the historical shift data, the objective function values of all successfully used first candidate pairs are counted;

[0032] The lower quartile of the distribution of the counted objective function values is taken as the benchmark of the preset threshold.

[0033] Preferably, the specific execution steps of checking whether the time period contains a location event of the vehicle entering a non-operational area include:

[0034] The entire location sequence of the vehicle in the time period is obtained;

[0035] The location sequence is compared with the preset multiple non-operational area electronic fences;

[0036] If the position sequence intersects with any non-operation area electronic fence, it is determined as containing a non-operation area event.

[0037] Preferably, the rule of searching for a continuous event pair that can form a complete operation period and has the highest target function value is embodied as follows:

[0038] In the joint event sequence, the first station boundary event must occur before the last station boundary event.

[0039] And there is no other vehicle shutdown event between the two events.

[0040] Among all the continuous event pairs that meet the above conditions, the one with the highest target function value is selected as the optimal combination.

[0041] Preferably, when the target function values of the second candidate pair and the third candidate pair generated in sequence are both higher than a preset threshold and the same, the following steps are performed:

[0042] Calculate the first absolute time difference between the first station boundary event and the planned departure time, and the second absolute time difference between the last station boundary event and the planned return time in the two candidate pairs;

[0043] Select the candidate pair with the smaller sum of the first absolute time difference and the second absolute time difference as the optimal combination.

[0044] Compared with the prior art, the present application has the following beneficial effects:

[0045] The present application assigns a confidence evaluation value to each type of event in multi-source data, generates an ordered candidate event list according to the evaluation value, and then determines the optimal combination of the first station boundary event and the last station boundary event by using a collaborative search and target function optimization method. This method effectively distinguishes the operation state and non-operation state of the bus in the first and last station area without relying on the vehicle door signal, and solves the problem of mutual mixing of passenger flow data between adjacent trips caused by non-operation activities such as vehicle charging, shifting and cleaning. Finally, it realizes accurate attribution of passenger flow data to the corresponding operation trip, obtains accurate passenger flow statistics at the trip level, and significantly improves the accuracy and reliability of public transport passenger flow data. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The figure is a schematic diagram of the method framework structure of the present application. DETAILED DESCRIPTION

[0047] The technical solutions of the present application will be described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0048] Referring to Figure 1 The application provides a passenger flow multi-source data statistical method for bus doorless signals, comprising:

[0049] Based on the multi-source time sequence data of the target vehicle within the operation plan time window, a plurality of candidate boundary determination events are identified, and each event is assigned a quantitative confidence evaluation value;

[0050] It should be noted that the operation plan time window is set according to the operation plan of the target vehicle, specifically from 30 minutes before the planned first station departure time to 30 minutes after the planned last station arrival time. The purpose of this setting is to ensure that the time range in which the real operation boundary may appear is covered; at the same time, the collection frequency of multi-source time sequence data is unified to 1 Hz, that is, each type of data is collected once per second and time stamped synchronously, and the time stamp accuracy reaches millisecond level, thereby ensuring the time sequence consistency of the data.

[0051] The identification of candidate boundary determination events adopts a multi-source data feature fusion method, and identification rules are set for the first station and the last station operation boundary respectively. The rules related to the activation of the vehicle state at the first station will uniformly increase the electronic fence verification of the operation departure site. The electronic fence is a polygonal area outlined according to the actual site boundary of the operation departure site, which is accurately defined by latitude and longitude coordinates with a coordinate error controlled within 1 meter. The latitude and longitude coordinates adopt the CGCS2000 national geodetic coordinate system. The electronic fence data is measured and mapped once before the acceptance of the bus operation line and is fixed in the vehicle terminal. If adjustment is needed during the operation period, it must be refiled.

[0052] Specifically, the candidate event identification rules for the first station operation boundary are as follows:

[0053] Rule 1 is that the positioning data shows that the vehicle is in the preset range of the first station (a circular area with the first station platform center point as the center and a radius of 50 meters, determined based on the platform length and positioning error) and the operation departure site electronic fence, the speed increases from 0 to more than 5 kilometers per hour and maintains for 10 seconds, and the vehicle-mounted power supply state is the ignition state;

[0054] Rule 2 is that the operation dispatching system receives the departure confirmation instruction sent by the driver, and the positioning data shows that the vehicle is in the preset range of the first station and the operation departure site electronic fence;

[0055] Rule 3 is that the vehicle-mounted video monitoring data identifies that the number of people in the vehicle compartment increases from 0 or a small amount (≤2 people) to more than 5 people, and the positioning is in the first station range and the operation departure site electronic fence, and the speed starts to rise.

[0056] Candidate event identification rules for the end station operation boundary: Rule 1 is that the positioning data shows that the vehicle is in the preset range of the end station (a circular area with the center of the end station platform as the center and a radius of 50 meters, set in accordance with the first station), the speed is reduced to 0 and maintained for more than 5 minutes, and the vehicle power state is off; Rule 2 is that the operation dispatching system receives the driver's confirmation instruction to stop the vehicle, and the positioning data shows that the vehicle is in the preset range of the end station; Rule 3 is that the vehicle video monitoring data identifies that the number of people in the vehicle cabin is reduced from ≥5 to 0 or a small number (≤2) and maintained for 3 minutes, and the positioning is in the end station range and the speed is 0.

[0057] In actual operation, multiple source time series data in the operation plan time window are traversed, and as long as any of the above rules is met, it is identified as a candidate event for the corresponding boundary and the time point is recorded; it needs to be noted that if the first station involves the ignition event location not in the operation departure field electronic fence, the candidate event will be directly excluded.

[0058] The confidence evaluation value is given by using a multi-dimensional weighted scoring method, which comprehensively considers the basic weight of the event type, the real-time signal quality of the data source, and the absolute time difference between the event occurrence time and the planned time as three core bases for calculation. After normalization processing, the total score is the confidence evaluation value (value range 0 to 1, the larger the value, the higher the confidence), and the weight of each dimension (basic weight accounts for 50%, signal quality accounts for 30%, and time difference accounts for 20%) is the best weight ratio optimized through big data correlation analysis and regression model training on more than 100 historical operation data of the same line. Compared with average allocation or other experience allocation methods, this allocation can improve the accuracy of boundary determination. The specific calculation process and parameter setting are as follows:

[0059] The first dimension is the basic weight setting of the event type. Since different boundary determination events have different degrees of association with the actual operation boundary, we assign different basic weights to the event types corresponding to each rule, and the sum of these basic weights accounts for 50% of the total confidence score.

[0060] The specific allocation is as follows: in the first station operation boundary event, the basic weight of rule 1 (positioning, speed, and power linkage) is 0.2 (accounting for 40% of the total score of 50%), the basic weight of rule 2 (dispatching instruction and positioning) is 0.18 (accounting for 36% of the total score of 50%), and the basic weight of rule 3 (video personnel, positioning, and speed) is 0.12 (accounting for 24% of the total score of 50%);

[0061] The basic weight allocation of the end station operation boundary event is consistent with that of the first station, and rules 1, 2, and 3 are 0.2, 0.18, and 0.12 respectively. This allocation method can ensure that event types with high correlation have a higher proportion in scoring.

[0062] The second dimension is the real-time signal quality score of the data source, which accounts for 30% of the total confidence score. The data sources involved include vehicle positioning data, vehicle power status data, operation scheduling system data, and vehicle video monitoring data. According to the importance of each data source, weights of 0.4, 0.2, 0.3, and 0.1 are respectively assigned.

[0063] For each data source, a clear signal quality score standard threshold is set, which is determined based on 100+ shift equipment normal operating condition statistics:

[0064] Positioning data accuracy ≤ 5 meters gets 1 point, 5 to 10 meters gets 0.7 points, and > 10 meters gets 0.3 points; power signal without fluctuation gets 1 point, occasional fluctuation (≤ 2 times within 1 minute) gets 0.6 points, frequent fluctuation (> 2 times within 1 minute) gets 0.2 points; scheduling instruction signal reception complete without delay gets 1 point, delay ≤ 3 seconds gets 0.8 points, delay > 3 seconds or signal incomplete gets 0.4 points;

[0065] And the video monitoring picture is clear, the personnel identification accuracy is ≥ 95% gets 1 point, the picture is blurred but the identification accuracy is ≥ 85% gets 0.6 points, and the identification accuracy is < 85% gets 0.3 points. The final score of this dimension is the sum of the product of the quality score of each data source and the corresponding weight.

[0066] It should be noted that the video monitoring picture is authorized and does not involve the collection of personal privacy.

[0067] It should be understood that in the present scheme, the video monitoring data as a data source refers to the anonymized people count result after real-time processing by the vehicle edge computing device, which does not contain any personal biological features or portrait information. The present application itself does not involve the collection, storage, transmission or analysis of original video images, thereby ensuring privacy security at the technical implementation level.

[0068] The third dimension is the absolute time difference score of event occurrence time and planned time, which is based on the absolute difference between event occurrence time and operation planning time, and the dimension score accounts for 20% of the total confidence score.

[0069] The specific standard is:

[0070] The absolute difference between the first station event time and the planned first station departure time ≤ 5 minutes gets 1 point, 5 to 10 minutes gets 0.7 points, 10 to 15 minutes gets 0.4 points, and > 15 minutes gets 0 points;

[0071] The absolute difference between the last station event time and the planned last station arrival time is scored according to the same standard as the first station, so that the closer the event time and the planned time, the higher the score.

[0072] Further, the final calculation of the confidence evaluation value is performed according to the following steps: first, the candidate event of the first station involving ignition is subjected to the electronic fence check of the operation departure site, and if it is not in the fence, the confidence evaluation value is directly set to 0, and subsequent calculation is not needed; then, for the candidate event passing the check or not involving ignition, the basic weight score of the first dimension, the signal quality score of the second dimension and the time difference score of the third dimension are respectively calculated;

[0073] The total score is further calculated, and the calculation formula is total score = basic weight score + signal quality score multiplied by 0.3 + time difference score multiplied by 0.2; finally, the total score is constrained, if the total score > 1, 1 is taken, and if the total score < 0, 0 is taken, and the result is the final confidence evaluation value.

[0074] In combination with the present embodiment, a certain candidate event of the first station satisfies rule 1 (basic weight score 0.2), passes the electronic fence check, the signal quality score of each data source quality score is calculated to be 0.7, the event and the planned departure time difference is 2 minutes (time difference score 1 point), the total score is 0.61 by substituting into the formula, and the confidence evaluation value is 0.61. Another candidate event of the first station satisfies rule 1 but the ignition point is outside the fence, the check is not passed, and the confidence is directly set to 0. There is still a candidate event of the first station satisfying rule 2 (basic weight score 0.18), the signal quality score is 0.96 after passing the check, the time difference is 1 minute (score 1 point), the total score is 0.668, and the confidence evaluation value is 0.67.

[0075] For the current shift, an ordered candidate event list is independently generated for the first station operation boundary and the last station operation boundary based on the confidence evaluation value, and the events in the list are arranged in order from high to low according to the confidence evaluation value. In the present embodiment, the first station operation boundary of the current shift identifies 3 candidate events, and the confidence is 0.67, 0.61 and 0.45 respectively, so the first station candidate event list is sorted from high to low as event A (0.67, time 7:59), event B (0.61, time 7:58) and event C (0.45, time 8:12), and the time sequence is 7:58 < 7:59 < 8:12;

[0076] The last station operation boundary identifies 4 candidate events, and the confidence is 0.82, 0.75, 0.53 and 0.31 respectively, and the last station candidate event list is sorted as event 1 (0.82, time 9:29), event 2 (0.75, time 9:33), event 3 (0.53, time 9:37) and event 4 (0.31, time 9:41), and the time sequence is 9:29 < 9:33 < 9:37 < 9:41. If there are events with the same confidence evaluation value (difference ≤ 0.01), the events are sorted according to the time points corresponding to the events in order, and the event with earlier time is arranged in front.

[0077] An optimization objective function is constructed to eliminate the passenger flow data inter-string between the first and last stations during non-operating period.

[0078] Specifically, the objective function is realized by a composite function, the core of which is to maximize the product of the confidence degrees of the first and last station boundary events, and an effectiveness evaluation function is introduced as an important positive incentive factor;

[0079] The final expression of the optimization objective function is:

[0080] The objective function value=(first station event confidence degree x last station event confidence degree) x (1+effectiveness evaluation score);

[0081] Wherein, the effectiveness evaluation score is calculated by the following way:

[0082] Period length deviation score: calculate the actual time period length defined by the candidate first and last station events, and compare it with the standard operating period length. If the deviation rate is ≤5%, the score is 1, 5% to 10% is 0.8, 10% to 15% is 0.5, 15% to 20% is 0.3, and >20% is 0. The final score of this dimension (denoted as S1) is the above deviation score multiplied by 0.7.

[0083] Non-operating area deduction: check whether the vehicle position sequence enters the preset non-operating area electronic fence during the time period defined by the candidate combination. If not, the deduction value S2=0; if the entering time is ≤5 minutes, S2=0.3; 5 to 10 minutes, S2=0.5; >10 minutes, S2=0.7.

[0084] Effectiveness evaluation score=S1-S2. This score can be positive or negative, which is used to adjust the core confidence degree product.

[0085] The design of this function not only ensures that the selected boundary event combination itself has the highest credibility by maximizing the confidence degree product, but also encourages the system to select combinations with high consistency with the planned period and no entry into non-operating areas, thereby synergistically optimizing from the root cause and avoiding data inter-string during non-operating period.

[0086] It should be understood that here we first analyze the core performance of passenger flow data inter-string during non-operating period: if the first station boundary event time is later than the actual departure time, it will lead to the passenger flow before departure not being counted;

[0087] If the first station boundary time is earlier than the actual departure time, it will lead to the passenger flow of the previous trip at the last station being mixed in;

[0088] If the last station boundary time is earlier than the actual arrival time, it will lead to the passenger flow after arrival not being counted;

[0089] If the end station boundary time is later than the actual arrival time, it will cause the passenger flow of the next bus to mix in. In the end, the interval formed by the first station boundary time and the end station boundary time needs to accurately cover the actual operation time of the current bus, which is the core starting point of building the optimization objective function. The construction of the optimization objective function needs to consider three constraints first:

[0090] Constraint condition 1: the first station boundary event time point must be earlier than the end station boundary event time point;

[0091] Constraint condition 2: the first station boundary event time point should be within 15 minutes before and after the planned first station departure time, that is, planned first station time-15 minutes≤first station time≤planned first station time+15 minutes;

[0092] Constraint condition 3: the end station boundary event time point should be within 15 minutes before and after the planned end station arrival time, that is, planned end station time-15 minutes≤end station time≤planned end station time+15 minutes.

[0093] Based on the constructed objective function, an optimal combination of the first station boundary event and the end station boundary event is determined from the ordered list generated for the first station operation boundary and the end station operation boundary;

[0094] The specific search process can be divided into the following steps: first, preset the target function value threshold, set the logic based on statistical analysis of historical bus data, and the specific operation is to collect historical operation bus data (sample size≥100 buses) of the same line and the same vehicle type in the past 30 days, extract the target function value of all successfully used first candidate pairs (i.e. the candidate pair composed of the first ranked first station and the first ranked end station), and form a data set after excluding abnormal buses with missing positioning or scheduling records;

[0095] Then, sort the data set from small to large, calculate the lower quartile (i.e. the 25th percentile, when the data size is n, the position is (n+1) multiplied by 0.25, if it is an integer position, take the corresponding value, and if it is a decimal position, take the weighted average value of adjacent values), and take this lower quartile as the threshold reference. Finally, the threshold is controlled between 0.4 and 0.6, and if it exceeds this range, take the corresponding boundary value.

[0096] In this embodiment, the lower quartile calculated from 100 historical data is 0.52, so the threshold is set to 0.52. Secondly, the first candidate pair is formed by combining the first ranked first station and the first ranked end station, and the target function value (i.e. the first station confidence multiplied by the end station confidence) is calculated.

[0097] Then according to the calculation result, it is judged that if the value is higher than the threshold value, the combination is directly determined as the optimal combination; if it is lower than the threshold value, the combination range is expanded, respectively, the first station first and the last station second, the first station second and the last station first, the target function values of the two combinations are calculated, and the combination with the highest value and exceeding the threshold value is selected.

[0098] If the expanded combination still does not meet the requirement, it is continued to be expanded to the first station first and the last station third, the first station second and the last station second, the first station third and the last station first, until a combination exceeding the threshold value is found;

[0099] If all combinations are lower than the threshold value, the combination with the highest target function value is selected, and a parameter calibration reminder of the system is triggered. Finally, when the target function values are the same, the combination with higher first station event confidence is preferred, and if the first station confidence is also the same, the combination with earlier last station event time is selected.

[0100] In combination with the embodiment, it is specified that the current shift plan departs from the first station at 8:00 and arrives at the last station at 9:30, and the threshold value is 0.52. The first station candidate list is A (0.67, 7:59), B (0.61, 7:58), and C (0.45, 8:12), and the last station candidate list is 1 (0.82, 9:29), 2 (0.75, 9:33), etc. The first candidate pair (A, 1) is combined, the target function value is 0.67*0.82=0.5494>0.52, and it is determined as the optimal combination. If the first station A confidence is 0.58, the last station 1 confidence is 0.85, and the first candidate pair value is 0.493<0.52, the value of the highest (B, 1) combination is selected after the combination is expanded, and a calibration reminder is triggered.

[0101] According to the time points corresponding to the first and last station boundary events in the optimal combination, the readings of the vehicle-mounted passenger flow instrument are obtained and the difference is calculated, and finally the passenger flow statistical result belonging to the current shift is output. It should be noted that the vehicle-mounted passenger flow instrument adopts an infrared passenger flow statistical device with cumulative counting function, and its counting principle belongs to existing mature technology, which can record the cumulative values of the number of passengers getting on and off in real time, and store the cumulative reading data with time stamp, and the reading accuracy is 1 person.

[0102] Specifically, from the historical reading data of the vehicle-mounted passenger flow instrument, the cumulative number of passengers getting on and off at the time point corresponding to the first station boundary event, and the cumulative number of passengers getting on and off at the time point corresponding to the last station boundary event are extracted;

[0103] The passenger flow data of the current shift is calculated, wherein the number of passengers getting on is equal to the cumulative number of passengers getting on at the last station minus the cumulative number of passengers getting on at the first station, the number of passengers getting off is equal to the cumulative number of passengers getting off at the last station minus the cumulative number of passengers getting off at the first station, and the total passenger flow of the current shift (calculated by the number of passengers getting on) is the number of passengers getting on calculated;

[0104] Also need to consider the processing of abnormal situations, if the end station cumulative number of people is less than the first station cumulative number of people (such as passenger flow instrument temporary fault caused by reading jump), linear interpolation method is used for correction, the specific method is to take the first and last station boundary time interval as the basis, calculate the average number of people in unit time, and then multiply the time interval to get the corrected number of people.

[0105] As an optional embodiment, the optimal combination of the first station time 7:59 corresponds to the passenger flow instrument reading cumulative boarding 0 people, and the cumulative alighting 0 people, because there is no passenger boarding and alighting before the first station departure, and the last station time 9:29 corresponds to the cumulative boarding 46 people and the cumulative alighting 43 people, and the current trip boarding 46 people and the alighting 43 people are calculated.

[0106] We assume an abnormal situation, if the end station reading shows that the cumulative boarding number = 44 due to failure, and this value is less than the reading 45 at a certain time, at this time, the reading at 8:59 (interval 60 minutes) is taken as the intermediate cumulative boarding number, and the value is 31. By using linear interpolation to calculate the corrected cumulative boarding number of the last station = 31 + (44-31) multiplied by (90-60) / (90-60) = 44, since the jump amplitude is small and within a reasonable error range, the actual reading is still used for calculation during actual statistics, to ensure the authenticity of the statistical results.

[0107] The method identifies the boundary event in a multi-source data fusion manner, can accurately position the operation boundary even in the absence of door signals, quantifies the confidence through three dimensions of event basic weight, signal quality and time difference, and cooperatively optimizes the boundary combination, greatly improves the credibility of the boundary determination, and at the same time, through the time constraint of the target function, eliminates the data mutual string problem of the non-operation period from the root.

[0108] As an embodiment of the present application, an effectiveness evaluation function for evaluating the effectiveness of the candidate event combination in isolating non-operation activities is constructed, and the function is used to calculate the auxiliary score of the candidate combination, including:

[0109] The actual time period length defined by the candidate first and last station event pair is calculated, and then compared with the standard operation time period length determined by the operation plan time. The smaller the deviation, the higher the contribution of this dimension score.

[0110] In specific operation, the candidate first station event time is taken as the actual starting time, and the candidate last station event time is taken as the actual ending time, and the difference between the two is the actual time period length;

[0111] Then, the difference between the departure time at the first station and the arrival time at the last station in the operation plan is used as the standard operation period length; and then the deviation rate is calculated by the formula: length deviation rate = absolute value of the difference between the actual period length and the standard operation period length / the standard operation period length.

[0112] Finally, the deviation rate is scored according to the corresponding scoring standard, and the scoring standard is based on the statistical operation data of 100 overtime shifts. Specifically, if the deviation rate is less than or equal to 5%, 1 point is scored; if the deviation rate is between 5% and 10%, 0.8 points are scored; if the deviation rate is between 10% and 15%, 0.5 points are scored; if the deviation rate is between 15% and 20%, 0.3 points are scored; and if the deviation rate is greater than 20%, 0 point is scored. Then, the final score of this dimension is obtained by multiplying the deviation score by 0.7. In this embodiment, the departure time at the first station in the current shift plan is 8:00, the arrival time at the last station is 9:30, and the standard operation period length is 90 minutes.

[0113] It is checked whether there is a position event of a vehicle entering a non-operation area in the time period defined by the candidate combination. If there is, the score of the combination is substantially reduced. It should be noted that the non-operation area is defined in advance by latitude and longitude coordinates, and covers the operation departure field, the parking field outside the preset range of the first and last stations, the maintenance area, the no-entry area, etc. These areas form a non-operation area electronic fence library, and the coordinate error is controlled within 1 meter. The specific execution steps are as follows: first, all position sequences of the vehicle in the time period are obtained; then, the obtained position sequences are compared with the preset multiple non-operation area electronic fences; if the position sequences intersect with any non-operation area electronic fence, it is determined that there is a non-operation area event. In terms of deduction rules, if there is no such intersection of positioning points, the deduction score is 0; if there is an intersection, the deduction score is deducted according to the length of time of entering the non-operation area. Specifically, if the entering time is less than or equal to 5 minutes, 0.3 points are deducted; if the entering time is between 5 minutes and 10 minutes, 0.5 points are deducted; and if the entering time is greater than 10 minutes, 0.7 points are deducted.

[0114] The final score of the optimization objective function is calculated according to the following steps: first, the score of the first dimension, the period length deviation, is calculated; then, the second dimension, the non-operation area position verification, is carried out to determine the corresponding deduction score; then, the preliminary score is obtained by subtracting the deduction score from the period length deviation score; finally, the preliminary score is constrained, and if the score is greater than 1, 1 is taken, and if the score is less than 0, 0 is taken. In this way, the final evaluation result is obtained.

[0115] With the example, the candidate combination is the first station event A (time 7:59) and the last station event 1 (time 9:29), the actual time period length is 90 minutes, which is consistent with the standard operation time period length, the length deviation rate is 0%, the deviation score is 1, and the time period length deviation score = 1 x 0.7 = 0.7; further checking the positioning data in the period, no positioning point is found in the non-operation area, and 0 is deducted, and the final score is 0.7, which indicates that the effectiveness of the combination in isolating non-operation activities is higher. Looking at another candidate combination, the first station event B (time 7:50) and the last station event 2 (time 9:40), the actual time period length is 110 minutes, the deviation rate is about 22.2%, the deviation score is 0, and the time period length deviation score is 0; the positioning verification found that the vehicle entered the non-operation maintenance area from 9:35 to 9:38, and the time length was 3 minutes, which deducted 0.3 points, the preliminary score was -0.3, and after the constraint, it was 0. It can be seen that the effectiveness of this combination is very low.

[0116] In the cooperative search link of bus doorless signal passenger flow statistics, a special situation is often encountered: the first candidate pair of the first station ranking first and the last station ranking first, the second candidate pair of the first station first and the last station second, and the third candidate pair of the first station second and the last station first, and the objective function values of the three pairs of core candidates are all lower than the preset threshold.

[0117] Therefore, as an embodiment of the present application, the preset threshold is determined by still using the historical data statistics method: collecting data of 100 overtime trips of the same line and the same vehicle type in the past 30 days, extracting the objective function values of all successfully used first candidate pairs, calculating the lower quartile, and controlling the threshold between 0.4 and 0.6.

[0118] The joint event sequence is to combine the first station candidate event list and the last station candidate event list, and then sort them in ascending order according to the time stamp of each event, and clearly mark whether each event is a first station candidate or a last station candidate, and its confidence evaluation value.

[0119] The complete operation time period refers to the time interval from the beginning of a first station candidate event to the end of a last station candidate event, and meets several requirements: the beginning time is earlier than the end time, and there is no vehicle shutdown between the two, at the same time, the first station event is within 15 minutes before and after the planned departure, and the last station event is within 15 minutes before and after the planned arrival.

[0120] In the specific execution, the following process is executed, so that each link is connected, and the accuracy of screening can be guaranteed:

[0121] The candidate list of the first station and the last station is combined into a joint event sequence. Specifically, the timestamp, attribute, confidence of all events in the candidate list of the first station are extracted, and the candidate list of the last station is also extracted. Then the two parts of information are integrated into a data set, and finally the joint event sequence is obtained by sorting the timestamp in ascending order.

[0122] It should be noted that if the timestamp of two events is similar (the difference is within 1 second), the candidate event of the first station is placed in front;

[0123] Moreover, the original information of each event in the sequence should be kept, and the confidence and attribute label cannot be missed. For example, the candidate list of the first station of the current shift has A (0.58, 7:59, first station), B (0.55, 7:57, first station), and C (0.42, 8:10, first station), and the candidate list of the last station has 1 (0.85, 9:35, last station), 2 (0.72, 9:40, last station), and 3 (0.51, 9:28, last station). The joint event sequence after merging and sorting is B (0.55, 7:57, first station), A (0.58, 7:59, first station), C (0.42, 8:10, first station), 3 (0.51, 9:28, last station), 1 (0.85, 9:35, last station), and 2 (0.72, 9:40, last station).

[0124] In the joint event sequence, the event pair that meets the requirement of the complete operation period is screened. Specifically, each candidate event of the first station in the joint event sequence is taken as a starting event, and all candidate events of the last station after the starting event are matched as ending events to form a "first station-last station" event pair.

[0125] Each event pair is then checked to ensure its validity: First, a time sequence check is performed to ensure the first station event time is less than the last station event time; second, a time range check is performed to determine if the first station event time is within 15 minutes before or after the planned first station departure time, and if the last station event time is within 15 minutes before or after the planned last station arrival time; third, an intermediate state check is performed to verify the onboard power status data between the first and last station events to confirm that there are no vehicle engine shutdown events. Only when all of the above conditions are met is the event pair considered valid. To make it easier to understand, let's take the previous example again. Suppose the planned first station departure is 8:00 AM and the last station arrival time is 9:30 AM, then the time constraints are 7:45 AM to 8:15 AM from the first station and 9:15 AM to 9:45 AM from the last station. When traversing the combined sequence, starting with the first station B (7:57), the subsequent last stations 3 (9:28), 1 (9:35), and 2 (9:40) all meet the requirements, and there are no shutdown records in between, thus forming the three groups (B, 3), (B, 1), and (B, 2). The same applies when starting with the first stations A (7:59) and C (8:10). After matching the events of the subsequent last stations, the resulting event pairs all meet the above requirements and are all valid event pairs.

[0126] Calculate the objective function value of each valid event pair and select the optimal combination. Specifically, calculate the objective function value of each valid event pair according to the objective function formula, and select the event pair with the highest objective function value from all valid event pairs as the optimal combination.

[0127] If multiple events have the same objective function value (difference ≤ 0.01), the combination with the higher confidence of the first event should be selected first.

[0128] If the confidence levels of the first station are also the same, the combination with the earlier event time at the last station is selected. Since the objective function values ​​of all candidate pairs are below the preset threshold at this point, a parameter calibration reminder needs to be triggered, prompting technicians to check whether parameters such as weight allocation and signal quality scoring standards in the boundary identification rules need adjustment. Referring to the previous example, the preset threshold is 0.52. Among the calculated objective function values ​​of each valid event pair, the (A, 1) combination has the highest value of 0.493, so it is determined to be the optimal combination, and a parameter calibration reminder is triggered.

[0129] This step breaks the limitations of independent screening by merging the event sequences of the first and last stations, and mines the optimal combination from the full time series associations. Even if the high-ranking candidate pairs do not meet the threshold requirements, the relative rationality of the boundary judgment can still be guaranteed.

[0130] As one embodiment of the present invention, when the objective function values ​​of the sequentially generated second candidate pair and third candidate pair are both higher than a preset threshold and are the same, the following steps are performed:

[0131] a first absolute time difference between the first station boundary event and the planned departure time in the two candidate pairs, and a second absolute time difference between the last station boundary event and the planned return time;

[0132] The candidate pair with a smaller sum of the first absolute time difference and the second absolute time difference is selected as the optimal combination.

[0133] Specifically, the preset threshold determination method remains consistent, that is, 100 overtime historical data of the same line and the same vehicle model in the past 30 days are collected, and the fourth quartile of the target function value of the successful combination is extracted.

[0134] The second candidate pair is a combination of the first ranked event in the first station candidate list and the second ranked event in the last station candidate list, and the third candidate pair is a combination of the second ranked event in the first station candidate list and the first ranked event in the last station candidate list.

[0135] The first absolute time difference is the absolute value of the difference between the actual occurrence time of the first station boundary event in the candidate pair and the departure time of the first station in the operation plan, and the second absolute time difference is the absolute value of the difference between the actual occurrence time of the last station boundary event in the candidate pair and the return time of the last station in the operation plan, and the calculation results of both are in minutes.

[0136] In the collaborative search process, the target function values of the first candidate pair (first station first and last station first), the second candidate pair (first station first and last station second), and the third candidate pair (first station second and last station first) are calculated.

[0137] When it is detected that the target function values of the second candidate pair and the third candidate pair are both higher than the preset threshold, and the values of the two are equal (the difference is less than or equal to 0.01), the screening logic of this step is triggered.

[0138] At the same time, the key information of the two candidate pairs needs to be extracted, including the actual time of the first station boundary event in each candidate pair, the actual time of the last station boundary event, and the planned departure time and the planned return time of the current shift, in order to prepare for the subsequent time difference calculation. In this embodiment, the preset threshold is 0.52, the planned departure time of the current shift is 8:00, and the planned return time is 9:30; the first station candidate list is A (0.68, 7:58, ranked first), B (0.65, 8:02, ranked second), and the last station candidate list is 1 (0.80, 9:29, ranked first), 2 (0.78, 9:31, ranked second).

[0139] The target function value of the second candidate pair (A, 2) is calculated as 0.68*0.78=0.5304, and the target function value of the third candidate pair (B, 1) is calculated as 0.65*0.80=0.52. Both of them are higher than the threshold 0.52 and the values are equal, so this screening step is triggered.

[0140] Then, the first absolute time difference and the second absolute time difference of the two candidate pairs are calculated respectively. The specific calculation rule is that the first absolute time difference is the absolute value of the difference between the actual time of the first station boundary event and the planned departure time, and the second absolute time difference is the absolute value of the difference between the actual time of the last station boundary event and the planned return time. It should be noted that the time unit should be unified during calculation, and all should be converted into minutes for calculation, and less than 1 minute is counted as 1 minute.

[0141] For the two candidate pairs in this embodiment, in the second candidate pair (A, 2), the difference between the actual time 7:58 of the first station event A and the planned departure time 8:00 is 2 minutes, and the first absolute time difference is 2; the difference between the actual time 9:31 of the last station event 2 and the planned return time 9:30 is 1 minute, and the second absolute time difference is 1. In the third candidate pair (B, 1), the difference between the actual time 8:02 of the first station event B and the planned departure time 8:00 is 2 minutes, and the first absolute time difference is 2; the difference between the actual time 9:29 of the last station event 1 and the planned return time 9:30 is 1 minute, and the second absolute time difference is 1.

[0142] Finally, the total time difference is calculated and the optimal combination is selected. Specifically, the first absolute time difference and the second absolute time difference of each candidate pair are added to obtain the total time difference of the candidate pair;

[0143] The total time difference of the two candidate pairs is compared, and the candidate pair with smaller total time difference is selected as the optimal combination. If the total time difference of the two candidate pairs is also equal, the candidate pair with higher confidence of the first station boundary event is selected;

[0144] If the confidence of the first station is still the same, the candidate pair with higher confidence of the last station boundary event is selected. In combination with this embodiment, the total time difference of the second candidate pair (A, 2) is 2+1=3, and the total time difference of the third candidate pair (B, 1) is also 2+1=3. At this time, the confidence of the first station event A is 0.68, which is higher than that of B, 0.65, so the second candidate pair (A, 2) is determined as the optimal combination.

[0145] This step solves the screening problem that the objective function values of the second and third candidate pairs are both up to standard and the same by introducing the time difference dimension. The core logic is that the smaller the total time difference is, the more consistent the operation boundary corresponding to the candidate pair is with the planned operation time, and the closer it is to the actual operation scene.

[0146] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A passenger flow multi-source data statistical method for bus doorless signals, characterized in that, The method comprises the following steps: Based on the multi-source time series data of the target vehicle within the operation plan time window, a plurality of candidate boundary determination events are identified, and a quantitative confidence evaluation value is assigned to each event; The identification of the candidate boundary determination events adopts a multi-source data feature fusion manner, and identification rules are respectively set for the first station and the last station operation boundaries, wherein the multi-source time series data refers to the multi-type data with time stamps collected in real time during the operation process of the target vehicle, specifically including vehicle positioning data containing latitude, longitude and speed; vehicle power state data containing ignition and extinguishing signals; operation scheduling system data containing planned departure time and planned arrival time; For the current shift, an ordered candidate event list is independently generated for the first station operation boundary and the last station operation boundary based on the confidence evaluation value, and the events in the list are arranged in order of their confidence evaluation value from high to low; An optimization objective function is constructed to eliminate the passenger flow data mutual stringing between the first and last stations; wherein the passenger flow data mutual stringing refers to the phenomenon that the passenger flow data outside the current shift operation time is counted into the statistical results of the current shift; The expression of the optimization objective function is: objective function value=(first station event confidence×last station event confidence)×(1+effectiveness evaluation score), wherein the effectiveness evaluation score is calculated based on the operation period length deviation rate defined by the candidate events and whether they cover the non-operation area; Based on the target function, an optimal combination of the first station boundary event and the last station boundary event is searched and determined from the ordered list generated for the first station operation boundary and the ordered list generated for the last station operation boundary. According to the time points corresponding to the first and last station boundary events in the optimal combination, the readings of the vehicle passenger flow meter are obtained and the difference is calculated, and the passenger flow statistical results belonging to the current shift are output.

2. The passenger flow multi-source data statistical method for bus doorless signal according to claim 1, characterized in that, The calculation of the confidence evaluation value includes the basic weight of the event type, the real-time signal quality of the data source, and the absolute time difference between the event occurrence time and the planned time.

3. The passenger flow multi-source data statistical method for bus doorless signal according to claim 1, characterized in that, The specific process of searching and determining an optimal combination of the first station boundary event and the last station boundary event is as follows: The first candidate pair is formed by the first-ranked event in the first station boundary ordered list and the first-ranked event in the last station boundary ordered list; The target function value of the first candidate pair is calculated; If the value is higher than the preset threshold, the first candidate pair is determined as the optimal combination; If the value is lower than the preset threshold, the second candidate pair is formed by the first-ranked event in the first station list and the second-ranked event in the last station list, and / or the third candidate pair is formed by the second-ranked event in the first station list and the first-ranked event in the last station list, and the target function values of the two candidate pairs are calculated, and the candidate pair with the highest target function value and higher than the preset threshold is selected as the optimal combination.

4. The passenger flow multi-source data statistical method for bus doorless signal according to claim 3, characterized in that, An effectiveness evaluation function is constructed to evaluate the effectiveness of the candidate event combination in isolating non-operation activities, which is used to calculate the auxiliary score of the candidate combination, including: The length of the time period defined by the candidate first / last station event pair is calculated and compared with the length of the standard operating time period based on the scheduled time. The smaller the length deviation, the higher the score contribution. Check if the time period contains a vehicle entering a non-operating area location event. If it does, the score of the combination is substantially reduced.

5. The passenger flow multi-source data statistical method for bus doorless signal according to claim 1, characterized in that, When identifying boundary determination events, for vehicle ignition events, if the occurrence location is not within the preset operating departure yard electronic fence, the confidence evaluation value is set to zero.

6. The passenger flow multi-source data statistical method for bus doorless signal according to claim 3, characterized in that, When the objective function values of the first, second, and third candidate pairs are all lower than the preset threshold, the following steps are performed: Merge the ordered list of first station boundaries and last station boundaries into a joint event sequence. In the joint event sequence, search for a continuous event pair that can form a complete operating time period and has the highest objective function value as the optimal combination.

7. The passenger flow multi-source data statistical method for bus doorless signal according to claim 3, characterized in that, The setting logic of the preset threshold is as follows: Based on historical shift data, the objective function values of all successfully used first candidate pairs are counted. The lower quartile of the distribution of the counted objective function values is used as the benchmark for the preset threshold.

8. The passenger flow multi-source data statistical method for bus doorless signal according to claim 4, characterized in that, The specific execution steps for checking whether the time period contains a vehicle entering a non-operating area location event include: Obtain the entire location sequence of the vehicle within the time period. Compare the location sequence with the preset multiple non-operating area electronic fences. If the location sequence intersects with any non-operating area electronic fence, it is determined to contain a non-operating area event.

9. The passenger flow multi-source data statistical method for bus doorless signal according to claim 6, characterized in that, The rule for searching for a continuous event pair that can form a complete operating time period and has the highest objective function value is as follows: In the joint event sequence, the first station boundary event must occur before the last station boundary event. And there is no other vehicle ignition event between the two events. Among all continuous event pairs that meet the above conditions, the one with the highest objective function value is selected as the optimal combination.

10. The passenger flow multi-source data statistical method for bus doorless signal according to claim 3, characterized in that, When the objective function values of the sequentially generated second and third candidate pairs are both higher than the preset threshold and the same, the following steps are performed: Calculate the first absolute time difference between the first station boundary event and the scheduled departure time, and the second absolute time difference between the last station boundary event and the scheduled return yard time in the two candidate pairs. Select the candidate pair with the smaller sum of the first absolute time difference and the second absolute time difference as the optimal combination.

Images