CN105206040B - A kind of public transport bunching Forecasting Methodology based on IC-card data - Google Patents

Abstract

The invention discloses a bus train prediction method based on IC card data, and belongs to the technical field of public transport information processing. The prediction method includes bus IC card data collection, data processing, actual bus situation detection, training and learning data and bus collision prediction, wherein the bus collision prediction adopts the least squares support vector machine (LS-SVM) algorithm. The present invention combines bus IC card data to extract a large number of passenger information for multiple trips, without the need for a vehicle-mounted GPS system, which is convenient and fast, and reduces data processing costs; at the same time, the method of least squares support vector machine can be better, faster and more effective Realize bus string prediction accurately, enable passengers to better understand bus operation conditions, reasonably control travel time, enable bus operation departments to adjust bus departure intervals in time, and improve bus service level; the invention is simple in data processing, low in cost, and have higher prediction accuracy.

Description

A method for predicting bus strings based on IC card data

technical field

The invention relates to the technical field of public transport information processing, in particular to a bus train prediction method based on IC card data.

Background technique

In the actual bus operation process, due to factors such as traffic congestion, station stop time, and changes in the number of people getting on and off the bus, the arrival of buses is irregular. Especially during peak hours, passengers often wait at the bus station for ten minutes or more without seeing a bus coming, but once they arrive, they find that it is not just one car, but several cars arrive at the same time, and the bus arrives at the same time. Passenger loads tend to be uneven. Reduced bus service levels, causing safety hazards.

In fact, the delay of a bus at a certain station may lead to an increase in the time it takes to reach the next station, and at the same time cause an increase in the number of passengers and waiting time at the next station, further increasing the delay time of the bus. On the other hand, the number of passengers carried by the next bus will be reduced, and at the same time, the delay time at the station will be reduced, and the time interval between the bus in front and the bus in front will be shortened. This is like a snowball effect. There is a high probability that the buses will meet at a certain stop. This phenomenon is the bus phenomenon. Therefore, it is predicted that the bus train can reduce the waiting time of passengers, improve the service level of the bus, and increase the bus sharing rate.

In recent years, there have been predictions of bus arrivals at bus stops in some big cities (such as Nanjing, Jiangsu, and Hangzhou, Zhejiang, etc.), but there are very few literatures on the prediction of bus trains. However, the current bus arrival predictions are all combined with the vehicle-mounted GPS system, and only for one vehicle, the distance from the station and the estimated arrival time are given. Although this can give bus passengers a certain reference, in fact, during peak hours, road congestion is serious, leading to the phenomenon of bus stringing, causing subsequent vehicles to enter the station before the previous one, making the predicted bus arrival time different from the passenger's actual time. The waiting time does not match, and the vehicle-mounted GPS system requires a large storage space, and the positioning accuracy is low. We need to find a better way to solve the above problems.

Contents of the invention

In view of the above-mentioned problems, the present invention provides a high-precision prediction method for bus trains based on IC card data, which fully considers the relevant factors of a certain train's arrival at the downstream station and has real-time dynamic performance. Based on the bus IC card data, the present invention predicts the arrival interval of two adjacent buses and the bus train from the perspective of passengers, can better understand the bus operation situation, reasonably regulate the travel time, and improve the travel efficiency; at the same time, for the bus operation department In other words, it is also possible to adjust the bus departure interval in time to avoid the occurrence of bus jamming and better improve the bus service level.

The described method for high-precision prediction of bus strings based on IC card data extracts number identification, line identification, site identification, arrival time, and passenger flow of getting on and off the bus for multiple vehicle trips at two adjacent stations of the same bus line. quantity and other information. Firstly, the outliers where the train identifications of the two stations do not correspond are eliminated, and the data of the train identifications are completely corresponding and consistent, and the headway obtained by arranging the second station according to the order of the train numbers of the first station is calculated. By analyzing the normality of the headway Negative to detect the actual bus train situation to the second station. If it is positive, it means that there is no tandem traffic, on the contrary, if it is negative, it means that tandem traffic has occurred. Then, it is necessary to predict the situation of bus trains arriving at the second station for a certain train, and extract the small sample data of each day in the training and learning based on the above-mentioned extracted line identification, train identification, arrival time, and passenger flow, including The travel time of two stations, the headway of a certain train at the first station, the number of people getting on and off at the first station of a certain train and the adjacent previous train respectively, and the number of people getting on and off at the second station of the adjacent previous train The number of people getting on and off the bus and other information, these small sample data of each day form a large sample data, according to the large sample data to establish a prediction model, combined with the least squares support vector machine algorithm to predict a certain number of buses arriving at the second station String situation.

The advantages of the present invention are:

1. The present invention combines bus IC card data to extract a large number of passenger information for multiple trips, without the need for a vehicle-mounted GPS system, which is convenient and fast, and reduces data processing costs;

2. The present invention adopts the method of least squares support vector machine to better, faster and more effectively realize the prediction of bus stringing, so that passengers can better understand the bus operation situation, and reasonably control the travel time; at the same time, the bus operation department can also timely Adjust bus departure intervals to improve bus service levels;

3. The present invention considers multiple factors such as the number of people getting on and off the bus, the arrival time, the travel time between two stations, the headway of two adjacent trains, etc., the data processing is simple, the cost is low, and the prediction accuracy is high.

Description of drawings

Fig. 1 is the schematic diagram of the bus string prediction method based on IC card data of the present invention;

Fig. 2 is a flow chart of the method for predicting bus trains based on IC card data according to the present invention.

detailed description

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments, so that those skilled in the art can implement it with reference to the description.

The present invention provides a kind of bus stringing prediction method based on IC card data, comprising the following steps:

The first step is bus IC card data collection: real-time acquisition of bus IC card swiping information through the 3G transmission network, and establishment of a bus operation route and vehicle operation information database. The bus IC card data includes information such as bus number identification, line identification, station identification, arrival time, date, passenger flow on and off the bus, and the like. Obtain a bus line that is prone to cross-traffic from the IC card data collected above, find two adjacent target stations on the basis of this bus line, and further extract the buses that pass through the above two target stations every day The number of trains, the arrival time of each train to the two target stations, and the number of people getting on and off.

Due to the different traffic conditions every day, the number of buses issued by the bus company is not the same, and the daily bus number can be extracted according to different vehicle numbers or bus intervals.

The second step, data processing: Since the number of people getting on and off the bus at the bus station is random and uneven, and the passenger card information is also abnormal, it is necessary to match the bus number identification, and the data that does not correspond to the bus number identification of the two target stations is eliminated as the wrong data. At the same time, information such as station identification, arrival time, and passenger flow corresponding to the train number is also removed accordingly, and only the data with the same train number at the two target stations is retained.

The 3rd step, the actual bus situation detection: the present invention divides the bus situation of the bus station into a binary state, the situation of the bus is 1, and the situation of the bus is 0. According to the second step, when the two target stations have the same train identification, the second target station is arranged according to the order of the first target station, and the headway of two adjacent trains (i.e. Headway) can be obtained In the actual situation of bus tandem, if the calculated headway is positive, it means that there is no tandem situation, which is recorded as 0; is 1.

The first target site is defined as the site that arrives first among the two adjacent target sites, and the second target site is defined as the site that arrives later among the two adjacent target sites.

The headway (ie Headway) of the two adjacent trains is the difference between the arrival time of the current train at a certain target site and the arrival time of the adjacent previous train at the same target site.

The fourth step is to train and learn data: to predict in real time the bus train situation of the current train arriving at the second target site, the relevant factors include the arrival time of the adjacent previous train at the first target site and the number of people getting on and off , the arrival time and the number of people getting on and off at the second target station, and the arrival time and the number of people getting on and off at the first target station of the current train number. The input factors in the training and learning include the travel time of two target stations, the headway distance between two adjacent trains at the first target station, the up and down of the adjacent previous train and the current train at the first target station respectively. The number of people in the car, and the number of people getting on and off at the second target station of the adjacent previous trip. There is only one factor as the output variable, that is, the situation of the bus running in parallel. In the present invention, the small sample data of each day is first extracted, and then a large sample data is formed in chronological order, and the training set is selected according to the sample data ratio of 3:1 between the training set and the test set.

The travel time of the two target sites, under the corresponding situation of the train number identification, the travel time of the two target sites is exactly the difference between the arrival time of the current train number at the second target site and the arrival time of the first target site. difference. Since there is a station distance between the two target sites, according to the maximum speed of the bus stipulated by the state, the travel time between the two target sites is a positive value and greater than a certain value, so the travel time data that does not meet the regulations must be eliminated. At the same time, information such as the corresponding train number identification, station identification, arrival time, and passenger flow on and off the bus are also removed accordingly.

The 5th step, the public bus run-in prediction: the present invention adopts the least squares support vector machine algorithm to predict the bus run-in situation, according to the training set that chooses in the 4th step, establishes the prediction model to the bus run-in of the current train number that arrives at the second target site Predict the situation and get the predicted value.

The least squares support vector machine (LS-SVM) algorithm is a kernel function learning machine following the principle of structural risk minimization, and the application of LS-SVM to bus train prediction mainly uses its regression algorithm. The LS-SVM model is established by using adjacent historical data. After the model is trained, a regression function is obtained. The predicted input vector is brought into the regression function, and the output value obtained is the data to be predicted.

The application of the LS-SVM to the bus train prediction includes two processes of training modeling and prediction evaluation.

Among them, during the training process, for the training samples solve equation

In the equation, y is a 1-dimensional column vector, which is composed of the output y _i (i=1...l) of the training sample; is a 1-dimensional column vector, the number of 1 is l; γ is a determined hyperparameter, b and α are unknowns to be solved, b is a real number, α is a 1-dimensional column vector (called Lagrange multiplier), and solving The process of b and α is the modeling process, Ω is the kernel function matrix, and the input _xi with input samples is calculated by the kernel function, the formula is

Ω _ij =K(x _i ,x _j ),

The radial basis function (RBF) function is selected as the kernel function in the formula, expressed as The center of each radial basis function corresponds to a support vector, and the support vector machine obtained at this time is a radial basis function classifier;

The key of solving equation (1) is to seek the inverse matrix of A, A=Ω+γ ^- 1I, after obtaining the inverse matrix of A, both can obtain parameter b as:

The parameter α can also be obtained as:

After obtaining b and α, the training process ends, and the obtained model is as follows:

According to the model described by formula (3), the output f(X) is calculated for the new input X, and this process is called the prediction process.

In specific applications, the amount of calculation in the training process is larger, and the above calculation process is refined to obtain the following process:

The formation process of the kernel function matrix:

The formation of the kernel function matrix is mainly to calculate the kernel function of different input vectors. The kernel function adopts the RBF function, and its specific form is:

The parameter σ is a hyperparameter determined before training, which is determined by K cross-validation. The specific process is:

Step a, select the initial value of σ, σ=0.01;

In step b, the training set is divided into k equal subsets, and each time k-1 data is used as training data, and the other data is used as test data. This is repeated k times, and the expected generalization error is estimated according to the average MSE obtained after k iterations, and finally a set of optimal parameters is selected as the parameter σ of the kernel function K(x, _xi ).

The calculation of the RBF function involves the calculation of the 2-norm of the vector and the calculation of the exponential function. According to the definition of Ω _ij =K( _xi , x _j ), for l training samples, Ω _ij is an l×l-dimensional matrix, that is, l Kernel function calculations are performed on any two of the samples to obtain a kernel function matrix.

Kernel function matrix inversion process:

After Ω _ij is obtained, matrix A can be formed. From the calculation process of Ω _ij , A is a symmetric positive definite matrix. If the inverse matrix A ^-1 of the matrix A is obtained, b and α can be obtained according to formula (2). The process of finding the inverse matrix of matrix A is the key link in the training process.

Example

As shown in Figure 1, two adjacent buses V1 and V2 are divided into two situations in the process of driving on the same line: Situation 1: The ideal state without bus strings: first at 9:02, Buses V1 and V2 are at the 1st and 4th stations respectively; at 9:18, buses V1 and V2 arrive at the 5th and 8th stations respectively; then at 9:24, V1 arrives After reaching the 7th station, V2 reached the 10th station. The two buses V1 and V2 have always maintained a distance of about three stops, and the number of passengers waiting at each stop is basically average, and there is no encounter or crossing.

Situation 2: The status of the bus string: First, at 9:02, buses V1 and V2 are at the first stop and the fourth stop respectively, and the passengers waiting at each stop are basically average; at 9:10, V1 arrives At the 3rd station, V2 arrives between the 5th and 6th stations, the distance between the two buses begins to shorten, V2 travels slowly and arrives at the station later, and the number of passengers waiting at the station after the 6th station increases; 9:13 At the minute, V1 arrived at the 4th station, and V2 immediately arrived at the 6th station; then at 9:19, V1 arrived at the 6th station, and V2 just left the 7th station, the distance between the two buses was very close; finally at At 9:34, V1 and V2 met at the 10th station, and the number of passengers waiting at the next station, the 11th station, was very large, and a bus jam occurred. This shows a snowball effect, a bus V2 delay increases the number of passengers at the next stop, and also increases the stop delay time, obviously, this also increases the delay of the bus. On the other hand, the passengers of the next bus V1 will be reduced, and the parking delay time is also reduced, and there is no delay.

The prediction evaluation metrics are defined as follows:

For the classification problem with two categories, such as the bus shuffle, the samples are divided into the shuffle situation, represented by 1, and the non-shuffle situation, represented by 0. For a split car problem, if a sample is a split car 1 and is also predicted to be a split car 1, this sample is a correct number of split cars; correspondingly, a sample that is not a split car 0 is predicted as not a split car Car 0, this sample is a correct number of cars that are not linked together;

The following predictors are often used to evaluate the performance of classification algorithms:

(1) Accuracy rate: the calculation is the proportion of the predicted correct samples (including the correct sample of the train and the correct sample of the non-travel) among all the samples.

(2) Correct splitting rate: the calculation is the proportion of correct splitting samples to all splitting samples.

In this embodiment, in order to facilitate the parameter understanding and algorithm realization of the present invention, the specific basic data in the six steps are described in detail.

The bus IC card data is provided by Beijing XX Company. The bus IC card data includes train number identification, line identification, station identification, arrival time and passenger flow, record number, transaction type, transaction serial number, transaction date, transaction time, SAM Card number, city number, card issue number, card type, line number, vehicle number, boarding station, getting off station, driver number and card number and other information. Taking the XX bus of Beijing XX Company to arrive at two target stations SA and SB as an example, the basic data of the four-month bus number, station logo, arrival time and passenger flow of boarding and disembarking are shown in Table 1 and Table 2:

Table 1: Basic data of Beijing XX company XX bus arriving at the first target stop SA

Table 2: Basic data of Beijing XX company XX bus arriving at the second target stop SB

Data processing mainly includes the following steps:

1. Remove abnormal points.

First, the processed data is screened according to the data of one day. For example, the basic data of a certain day is extracted. The train number, arrival time and passenger flow of the second station should be arranged in the order of the train number of the first station. The travel time of the station, according to the station distance between the two stations, and knowing the maximum speed of the bus stipulated by the state, it can be concluded that the travel time is a positive value and greater than a certain value, so the data that does not meet the requirements, including the number of trains, is eliminated , arrival time, passenger flow and other information;

Then calculate the headway of two adjacent trains, that is, Headway. Since the data that does not meet the requirements are eliminated in the first step, some trains are not adjacent, so only the previous Headway and the next Headway can be calculated;

2. Input variables.

In order to predict in real time the situation of the current train arriving at the next target station SB, the relevant factors include the arrival time of the adjacent previous train at the station SA and the number of people getting on and off, and the arrival time and the number of people getting on and off at the next station SB. The number of people on the bus, as well as the arrival time of this train at the previous station SA and the number of people getting on and off.

There are 8 factors as input variables, including the travel time of the two target stations, the headway distance between the two adjacent trains of the current train at the first target site, and the distance between the last train and the current train at the first target site respectively. The number of people getting on and off, and the number of people getting on and off at the second target station in the last trip.

3. Output variables.

The present invention is based on the IC card data to predict the situation of bus tandem, and there is one factor as an output variable, that is, the bus tandem situation, if this bus trip occurs at the next station SB, it is recorded as 1, if there is no tandem, it is recorded as 0.

Taking the XX bus of Beijing XX Company to arrive at two target stations SA and SB as an example, the basic data after four months of processing include 8 input variables and 1 output variable, and the date is shown in Table 3 below:

table 3

The present invention predicts the situation of bus stringing vehicles based on bus IC card data, the algorithm that adopts is Least Squares Support Vector Machines (Least Squares Support Vector Machines, LS-SVM), the tool that adopts is MATLAB 2013b, selects the first three months (20120702- The data of 20120930) is used as the training data, and the data of the next month (20121001-20121029) is used as the test data. The prediction results are finally obtained through the budget as shown in Table 4 below:

Table 4 prediction results

Prediction results of bus stringing Accuracy(%) Correct shuffle rate (%) Prediction accuracy 93.77% 85.54%

The bus train prediction method based on the IC card data of the present invention has high prediction precision, and the accuracy rate reaches 93.77%. Real-time prediction of the bus traffic situation at the next target station for this trip can be made, so that bus passengers can timely understand the bus operation status, arrange travel time reasonably, and improve travel efficiency; at the same time, the bus operation department can also use this prediction result to properly dispatch buses , Adjust the departure interval, avoid the occurrence of bus strings, and improve the service level and quality of public transport.

Claims (6)

1. a method for predicting bus strings based on IC card data, is characterized in that: comprise the following steps, The first step, bus IC card data collection: real-time acquisition of bus IC card swiping information through the 3G transmission network, the bus IC card data includes bus number identification, line identification, station identification, arrival time, date and boarding and alighting Passenger flow information; select a bus line that is prone to cross-traffic from the above-mentioned collected IC card data, find two adjacent target sites on the basis of this bus line, and further extract the traffic that passes through the above two target sites every day The number of bus trips and the arrival time of each bus trip to the two target stations and the number of people getting on and off the bus; The second step, data processing: it is necessary to match the train number identification, and remove the data that does not correspond to the train number identification of the two target stations as error data, and at the same time, the station identification, arrival time and passenger flow information corresponding to the train number identification are also corresponding Eliminate, only keep the data with the same train number at the two target stations; The third step is the detection of the actual bus train situation: in the case that two target sites have the same train number, the second target site is arranged according to the order of the first target site's train numbers to obtain the headway distance of two adjacent trains, if If the obtained headway is positive, it means that there is no cross-traffic situation, which is recorded as 0; if the obtained headway is negative or 0, it indicates that there is a cross-traffic situation, which is recorded as 1; The fourth step is to train and learn data: to predict in real time the bus connection situation of the current train to the second target site, the relevant factors include the arrival time of the adjacent previous train at the first target site and the number of people getting on and off , the arrival time and the number of people getting on and off at the second target site, and the arrival time and the number of people getting on and off at the first target site of the current train number; the input factors in the training and learning include the travel time of the two target sites, The headway distance between two adjacent trains at the first target station, the number of people getting on and off at the first target station of the previous adjacent train and the current train respectively, and the number of people getting on and off at the second target station of the previous adjacent train The number of people getting on and off the bus; there is only one factor as the output variable, that is, the bus situation; first extract the small sample data of each day, and then form a large sample data in chronological order, according to the 3:1 sample of the training set and the test set The data ratio selects the training set; The fifth step, bus tandem prediction: use the least squares support vector machine algorithm to predict the bus tandem situation, and establish a prediction model based on the training set selected in the fourth step to predict the bus tandem situation of the current bus number arriving at the second target station. Forecast, get the predicted value. 2. a kind of bus string prediction method based on IC card data according to claim 1, is characterized in that, the first target site described in the 3rd step is defined as arriving earlier in two adjacent target sites site, the second target site is defined as the later-arrived site among the two adjacent target sites. 3. a kind of bus string prediction method based on IC card data according to claim 1, is characterized in that, the headway distance of two adjacent trips described in the 3rd step is exactly current trip at a certain target site The difference between the arrival time of the train and the arrival time of the previous adjacent train at the same target station. 4. a kind of bus string prediction method based on IC card data according to claim 1, is characterized in that, the travel time of two target sites described in the 4th step, under the corresponding situation of train number identification exactly, The difference between the arrival time of the current train at the second destination station and the arrival time of the first destination station. 5. a kind of bus string prediction method based on IC card data according to claim 1, is characterized in that, the prediction model described in the 5th step is expressed as Among them, the radial basis function is selected as the kernel function, expressed as α _i is the array element of Lagrange multiplier α, and Offset where matrix Kernel function matrix Ω=K( _xi ,x _j ); is a one-dimensional column vector; σ is a hyperparameter determined before training; γ is a hyperparameter that has been determined; x _i and x are the input samples. 6. a kind of bus string prediction method based on IC card data according to claim 5, is characterized in that, in described radial basis kernel function K (x, x _i ), parameter σ takes K cross-validation mode to come OK, the specific process is: Step a, select the initial value of σ, σ=0.01; Step b, establishing the LS-SVM model; In step c, the selected training set is divided into k equal subsets, and each time k-1 data is used as training data, and the other data is used as test data; this is repeated k times, according to k iterations The average value of the obtained MSE is used to estimate the expected generalization error, and finally a set of optimal parameter values is selected as the parameter σ of the kernel function K(x, _xi ).