CN110119884A - A kind of high-speed railway passenger flow Time segments division method based on neighbour's propagation clustering - Google Patents

Abstract

The invention provides a high-speed railway passenger flow time division method based on neighbor propagation clustering. The invention divides the statistical time into several time points, then counts the passenger flow data at each time point, and constructs a time point from the preprocessed sample variables. sample sequence; then, the sample sequence is divided by the nearest neighbor propagation clustering algorithm; finally, the optimal clustering result is determined by using clustering effectiveness indicators such as CH, Hartigan, and IGP, and the division result of the annual operation period is further formed. The nearest neighbor propagation algorithm clusters and merges the time points with similar passenger flow in the year, and determines the optimal number of clusters according to CH, Hart and IGP indicators to improve the classification accuracy; at the same time, it can more objectively and accurately reflect the passenger flow demand in different periods of the year. , which overcomes the subjectivity, low efficiency and low precision of the manual division method, thus laying a foundation for the adaptive adjustment of the train running plan.

Description

A time division method of high-speed railway passenger flow based on neighbor propagation clustering

technical field

The invention relates to the technical field of high-speed railways, in particular to a method for dividing passenger flow periods of high-speed railways based on neighbor propagation clustering.

Background technique

The gradual improvement of the "eight horizontal and eight vertical" high-speed railway network and the intercity railway network has made more and more medium and long-distance passengers choose high-speed railway as their preferred travel mode. The number of trains, sections, stops, etc. In order to improve the service quality of the passenger flow of each destination on the road network, and at the same time reduce the cost of train operation as much as possible, the high-speed railway passenger transportation management department needs to adjust the train operation plan according to the fluctuation of the annual passenger flow. Make it meet the needs of passenger flow at different times of the year.

The distribution of passenger flow on trains in the railway network is an important basis for evaluating the implementation efficiency of passenger train operation plans. Usually, the actual passenger flow distribution is used to evaluate the passenger train operation scheme being implemented. However, for the passenger train operation scheme to be optimally designed, the passenger flow distribution on the train can only be evaluated by means of the passenger flow distribution method. Since the efficiency of passenger flow distribution and the rationality of the results directly affect the optimization level of the passenger train operation plan, the train passenger flow distribution method has always been one of the important basic research topics for the optimization of the railway passenger train operation plan.

However, the adjustment of the train running plan involves many factors, which is a complex and huge system engineering, and the number of adjustments to it every year is limited. It is a feasible strategy to divide the operation year of high-speed railway according to the fluctuation characteristics of passenger flow, and then adjust the train operation plan according to the passenger flow of each period. Therefore, the scientific and reasonable division of operating periods is the basic premise and important basis for the adjustment of the train operation plan, and an important guarantee for adjusting the train operation plan to adapt to the dynamic passenger flow demand. The current high-speed railway operating time division method is to divide the annual operating time period into the spring travel period, summer travel period, holiday and peak season according to the changes in the total passenger flow of the target line throughout the year. Although this approach reflects the difference in passenger flow between different time periods, the quality of time division results depends to a large extent on the experience of on-site engineers and technicians, which is highly subjective and easily leads to unreasonable time division results. , it is difficult to accurately reflect the passenger flow demand with seasonal changes in the year.

There is no relevant research at home and abroad on the division of high-speed railway operation periods. This problem is similar in nature to the division of traffic periods in the design of road intersection signals based on multi-period control (TOD). For the multi-time control (TOD) problem of road intersections, scholars at home and abroad have conducted some related research, mainly by drawing the cumulative traffic volume curve of a representative intersection for one day, and determining the time when the traffic volume curve changes significantly through manual experience. Nodes are time-segment points, so as to achieve a reasonable division of traffic time periods. There is almost no independent research on the earlier train passenger flow allocation methods, and they are usually used in the optimization of train operation plans. These studies construct passenger transfers based on a given train operation plan. network, design the generalized travel cost of passengers, including fare expenditure, travel time and congestion effect, etc., establish a static user equilibrium allocation model or a random user equilibrium allocation model, and distribute the traffic evenly to the train running sections (see Shi Feng, Deng Lianbo, Li Xinhua, Fang Qigen. Research on the operation plan of passenger trains related to passenger dedicated lines [J]. Journal of Railways, 2004, 26(2): 16-20.). Urban public transport passenger flow distribution is very similar to high-speed rail passenger flow distribution, and there is a lot of research work in this area. Hamdouch et al. constructed a travel alternative set for passengers traveling to any station on the space-time network, and then proposed a timetable-based balanced distribution model for public transport strategies under the principle of comprehensive consideration of departure time requirements, strict capacity constraints and space-time priority. Considering the capacity constraints and space-time priority characteristics in the process of passenger flow distribution, no research and analysis on the characteristics of passenger ticket purchases are given. However, this characteristic has an important impact on the travel choices of high-speed railway passengers. Therefore, it is necessary to design a suitable Passenger flow distribution methods for high-speed rail transport networks.

SUMMARY OF THE INVENTION

In view of the deficiencies of the prior art, the present invention provides a method for dividing the passenger flow period of high-speed railway based on neighbor propagation clustering. The adaptability of the opening plan and the passenger flow demand.

The technical scheme of the present invention is as follows: a method for dividing the passenger flow period of high-speed railway based on neighbor propagation clustering, comprising the following steps:

S1), divide a year into T time points, and set the high-speed railway line X to include n important sites, and count the passenger flow of each site in each time interval t _k in the upward or downward direction of the high-speed railway line X, which is

Among them, X _k represents the passenger flow of the high-speed rail line X within the time interval t _k , Represents the passenger flow sending volume of the nth station of the high-speed rail line _X within the time interval tk;

S2), adopt the threshold value δ to determine whether the passenger flow transmission volume at each time point is abnormal, specifically counting the passenger flow transmission vectors of the m adjacent time points at the time point _t1 , and calculate the mean value thereof if

Then it is considered that the passenger flow data at this time point _tl is non-abnormal data, otherwise it is abnormal data;

Wherein, X _l is the amount of passenger flow sent at time point t _l ;

If the data is abnormal, delete the abnormal data, and use the passenger flow data of m adjacent time points at this time point to fit and repair the passenger flow transmission volume. The calculation formula is as follows:

In the formula, l _k (t) is the k-1 degree fitting polynomial, L _n (t) is the Lagrangian interpolation polynomial, t is the time point to be fitted, t _j is the jth time point, t _i is the i-th time point, X _i is the passenger flow matrix of the i-th time point, and l _i (t) is the i-1 degree fitting polynomial;

Combine the fitted and repaired data with the original normal data, and then use the standard deviation standardization to eliminate the difference in scale between variables. The calculation formula is as follows:

In the formula, Z-score is the value after the standard deviation standardization, x is the passenger sending volume of the station at a certain time point, is the mean of the passenger volume of all stations in the year, σ is the standard deviation of the passenger volume of all stations in the year;

S3), divide the time period based on the neighbor propagation clustering algorithm, take the passenger flow data set at T time points in the passenger flow data set as the candidate class representative, and judge the similarity s of the passenger flow in any two time intervals ( i,k), that is, s(i,k) represents the similarity between the passenger flow transmission volume sample X _k of the time interval k and the passenger flow transmission volume sample X _i of the interval i, that is, the sample X _k is suitable as the class representative of the sample X _i When the algorithm is initialized, it is assumed that all samples have the same probability of being represented by the class, that is, all s(k,k) are assumed to be the median value p of the same attraction, where the similarity calculation formula of any two samples is:

s(i,k)=-||x _i -x _k || ² ;

Among them, x _i represents the sample of the passenger flow sending volume of i, and x _k represents the passenger flow sending volume at time k;

Define the reliability matrix r and the availability matrix a, where the reliability matrix r(i,k) points from the sample x _i to the sample x _k , indicating the representative degree that the sample x _k is suitable as the class representative of x _i ; a (i,k) is from the sample x _k to the sample _xi , indicating the appropriateness of _xi choosing x _k as the class representative; for any sample _xi, calculate the reliability r(i,k) of the amount of passengers sent in other time intervals and the sum of the availability a(i, k), if the sample x _k with the largest sum of the two is the class representative, output the classification results of all time points;

Specifically include the following steps:

S301), set the initial value of the reliability matrix r(i,k) and the availability matrix a(i,k) to 0;

S302), calculate the similarity matrix s(i,k) of the passenger flow samples at any time interval, and the matrix value adopts the Euclidean distance as the measure, that is, s(i,k)=-||x _i -x _k || ² ;

Set the diagonal elements s(k,k) to the same median attraction, that is

In the formula, N is the number of samples;

S303), update the credibility matrix r(i,k) and the availability matrix a(i,k), wherein, the update calculation formula of the credibility matrix r(i,k) is:

The update calculation formula of the availability matrix a(i,k) is:

S304), set the damping factor λ to eliminate the digital oscillation in the iteration, that is,

In the formula, r _new (i, k) and r _old (i, k) are the credibility matrices obtained from the current and previous updates, respectively; a _new (i, k) and a _old (i, k) are respectively Availability matrices obtained from the current and previous updates; λ∈(0,1) is the damping factor;

S305), obtain the sum of the reliability and availability of any passenger flow sending volume data sample and all passenger flow sending volume samples, according to Find the class center sample for each sample;

S306), the current number of iterations is updated n←n+1, and it is judged whether the information iteration process reaches the set maximum number of iterations, that is, n≤N _max , if yes, the algorithm is terminated, and the classification results of all time points are output, otherwise, return to step S302);

S4), calculate the Calinski-Harabasz, Hartigan and In-Group Proportion indicators of the classification results at different time points respectively, and select the optimal number of time points and their corresponding classification results;

S5), check and correct the division results of the operation period, traverse all the division categories in the cycle, and perform a pairwise comparative analysis of each sample. If the time points corresponding to the two samples are adjacent, they are combined into one operation period, otherwise, regarded as another operating period;

S6), passenger flow demand adaptability assessment, after the operation period is divided, the train operation plan is prepared according to the average passenger flow demand in each period and the passenger flow distribution simulation is carried out. Three indicators of passenger flow demand satisfaction rate, average train occupancy rate and passenger flow direct rate are introduced. Quantitative assessment and summary of the adaptability of the passenger flow demand and the train operation plan during the time period.

Further, the Calinski-Harabasz index is based on the measurement of the intra-class dispersion matrix and the inter-class dispersion matrix of all samples, and the number of classes corresponding to the maximum value is used as the optimal number of clusters, that is,

where k is the number of clusters, trB(k) is the trace of the inter-class dispersion matrix, trW(k) is the trace of the intra-class dispersion matrix, and n is the number of samples at time points.

Further, the Hartigan index is used when the number of clusters is 1, and the minimum number of clusters satisfying Ha≤10 is used as the optimal number of clusters, that is,

In the formula, k is the total number of categories at the time point of the sample clustering result, trW(k) is the trace of the intra-class dispersion matrix, and n is the number of samples at the time point.

Further, the In-Group Proportion index is used to measure whether the samples closest to each sample in a certain class are in the same class. The larger the average IGP index of all clusters, the better the quality of the clusters. The number of classes corresponding to the maximum value is the optimal number of clusters, that is,

In the formula, u is the class label of a certain cluster, Class(j) is the class label of sample j, j ^N is the sample closest to sample j, and # is the number of satisfied conditions.

Further, the passenger flow demand satisfaction rate is used to reflect the passenger transport capacity and passenger flow demand satisfaction degree provided by the train operation plan between the various passenger flow destinations of the high-speed railway and related road networks. Under the constraints, it is expressed by the ratio of the passenger traffic volume that can obtain effective transportation services under the condition of the established train operation plan to the total passenger flow demand. The calculation formula is as follows:

In the formula, q' _w is the total passenger flow transported by the passenger flow OD to the high-speed railway between w, and w is the number of destination of the passenger flow on the road network.

Further, the average train occupancy rate refers to the average occupancy rate of all trains within the evaluation range, and the average train occupancy rate refers to the weighted ratio of the passenger flow carried by the train in its running section to the total number of seats provided by the train, This indicator is used to reflect the selection results of various types of high-speed trains by passengers between different passenger flow OD pairs. The calculation formula of the average train occupancy rate is as follows:

In the formula, is the passenger flow carried by the train h in the section (i, j), A _h is the number of passengers of the train h, and E _h is the number of sections that the train h runs.

Further, the passenger flow demand structure is composed of different demand directions, and each demand direction will have a direct or transfer train plan to the destination. The ratio of the passenger flow that directly reaches the destination without transfer between demand points and the total passenger flow in this direction, and its calculation formula is:

In the formula, w is the number of destinations of the road network passenger flow, |e| is the number of transfers of a certain destination passenger flow, is the number of passengers who reach the destination directly without transferring between passenger flow OD and w, is the number of passengers who reach the destination through |e| transfers between passenger flow OD pair w.

The beneficial effects of the present invention are:

1. The present invention combines the passenger flow data at each time point along the station, adopts the neighbor propagation algorithm to cluster and merge the time points with similar passenger flow in the year, and determines the optimal number of clusters according to CH, Hart and IGP indicators, and improves the classification. accuracy;

2. The high-speed railway operation period division method provided by the present invention combined with the result test of the cluster validity index can more objectively and accurately reflect the passenger flow demand in different periods of the year, and overcome the subjectivity and low efficiency of the manual division method. And the shortcomings of low precision, so as to lay the foundation for the adaptive adjustment of the train running plan;

3. After the best clustering result is determined by the present invention, the clustering result is manually analyzed, and the accuracy of the planning is ensured by checking and correcting the results of the division of the operation period, and at the same time, the passenger flow demand satisfaction rate, the average train occupancy rate and the passenger flow direct rate are determined. Evaluate the adaptability of the passenger flow demand to further improve the adaptability of the passenger flow demand at each time period and the train operation plan.

4. The present invention also preprocesses the collected data, deletes abnormal data, and uses Lagrangian interpolation to fit and repair the abnormal data, thereby ensuring data availability and reliability of planning results.

Description of drawings

Fig. 1 is a flow chart of high-speed railway operation time division based on neighbor propagation clustering;

Figure 2 is a schematic diagram of the effectiveness index values of different cluster numbers in the operation period in 2014;

Figure 3 is a schematic diagram of the effectiveness index values of different cluster numbers in the operation period in 2015;

Figure 4 is a schematic diagram of the clustering results of the operation period in 2014;

Figure 5 is a schematic diagram of the clustering results of the operation period in 2015.

Detailed ways

The specific embodiments of the present invention will be further described below in conjunction with the accompanying drawings:

As shown in FIG. 1 , this embodiment provides a method for dividing high-speed railway passenger flow periods based on neighbor propagation clustering. For ease of understanding, the From January 1st to December 31st, 2015, the passenger flow of a high-speed railway that has been operating normally is used as data to illustrate the present invention. The railway has 9 stations, which specifically includes the following steps:

S1), divide January 1, 2014 to December 31, and January 1, 2015 to December 31, 2015 into 365 time points, that is, each day is a time point, and each time point is counted The passenger flow down the railway is Describe the passenger flow OD matrix in the downlink direction of the line respectively, where X _k is the passenger flow at time point k, is the passenger flow of site i at time point k;

S2), use the threshold value δ to determine whether the passenger flow transmission volume at each time point is abnormal, specifically counting the passenger flow transmission vectors of the m adjacent time points at the time point _t1 , and calculating the mean value thereof if

Then it is considered that the passenger flow data at this time point _tl is non-abnormal data, otherwise it is abnormal data;

Wherein, X _l is the amount of passenger flow sent at time point t _l ;

If the data is abnormal, delete the abnormal data, and use the passenger flow data of m adjacent time points at this time point to fit and repair the passenger flow transmission volume. The calculation formula is as follows:

In the formula, l _k (t) is the k-1 degree fitting polynomial, L _n (t) is the Lagrangian interpolation polynomial, t is the time point to be fitted, t _j is the jth time point, t _i is the i-th time point, X _i is the passenger flow matrix of the i-th time point, and l _i (t) is the i-1 degree fitting polynomial;

Combine the fitted and repaired data with the original normal data, and then use the standard deviation standardization to eliminate the difference in scale between variables. The calculation formula is as follows:

In the formula, Z-score is the value after the standard deviation standardization, x is the passenger sending volume of the station at a certain time point, is the mean of the passenger volume of all stations in the year, σ is the standard deviation of the passenger volume of all stations in the year;

S3), the time period is divided based on the neighbor propagation clustering algorithm. The neighbor propagation clustering algorithm (Affinity Propagation, AP) is to perform clustering on the similarity matrix S composed of sample data points, which is the same as other clustering algorithms. It is to minimize the distance between each data point in each category and the representative point of the category, so as to realize category division, which includes the following steps:

S301), initialize the reliability matrix r(i,k) and the value of the availability matrix a(i,k) to 0;

S302), calculate the sample similarity matrix s(i,k), the matrix value adopts the Euclidean distance as the measure, that is, s(i,k)=-||x _i -x _k || ² , set the diagonal element s( k,k) is the same median value of attractiveness, i.e.

In the formula, N is the number of samples, and in this embodiment, the value is 265;

S303), update the reliability matrix r(i,k) and the availability matrix a(i,k), the calculation formulas are respectively:

The update calculation formula of the credibility matrix r(i,k) is:

The update calculation formula of the availability matrix a(i,k) is:

S304), set the damping factor to eliminate the digital oscillation in the iteration

In the formula, r _new (i, k) and r _old (i, k) are the credibility matrices obtained from the current and previous updates, respectively; a _new (i, k) and a _old (i, k) are respectively Availability matrices obtained from this and the last update; damping factor λ=0.9 is set in this embodiment;

S305), calculate the sum of the reliability and availability for all samples for the passenger flow data samples at any time point, according to Find the class center sample of each sample, and then output the class division results of all time points;

S4), since the AP algorithm in step S3) will output a series of clustering results when clustering the samples, Calinski-Harabasz, Hartigan and In-Group Proportion indicators are used to evaluate the effectiveness of the various clustering results obtained by the algorithm The results are shown in Figure 2 and Figure 3. It can be seen from the figures that the optimal sample clustering number based on the high-speed railway passenger flow data from 2014 to 2015 is 5, which is used as the final clustering result, as shown in Figure 4 and Figure 4 5 shown;

S5), traverse all the divided categories in the cycle, conduct a pairwise comparative analysis of the passenger flow data samples at any time point, and split the discontinuous time points in the same category to form the division results of the high-speed railway operation period from 2014 to 2015. Its structure As shown in Table 1;

Table 1 Division results of operating periods

It can be seen from Table 1 that from 2014 to 2015, the high-speed railway railway time division results based on the change of passenger flow are all 5 categories, and the 365 days a year can be divided into 13 operating time periods. Among them, the time spans of operation period 3, operation period 6, operation period 7, operation period 8, and operation period 12 from 2014 to 2015 are the same, while the time spans of other operation periods are different. The reason is due to the differences in the Spring Festival travel period over the years. The Spring Festival in 2014 is on January 31, which is the 31st day; the Spring Festival in 2015 is on February 19, which is the 50th day. It can be found that every year, the operation period2 is 7 days before the Spring Festival. The annual seasons corresponding to other operating periods are more obvious, which can be summarized as follows:

The time span of operation period 1 is the period of smooth passenger flow after the New Year and before the Spring Festival; the time span of operation period 2 to operation period 4 is the peak period of passenger flow during the Spring Festival travel period; the time span of operation period 5 is the period of smooth passenger flow between the Spring Festival travel period and the Qingming Festival; The time span of time period 6 is the peak passenger flow period of Qingming Festival; the time span of operation period 7 is the smooth passenger flow period between Qingming Festival and May 1st Labor Day; the time span of operation period 8 is the peak passenger flow period of May 1st Labor Day; the time span of operation period 9 It is the passenger flow flat period between the May 1st Labor Day and the Summer Festival; the time span of the operation period 10 is the peak passenger flow period of the summer transport; the time span of the operation period 11 is the passenger flow flat period between the summer transport period and the National Day of the 11th National Day; The time span of time period 12 is the peak passenger flow period of “National Day”; the time span of operation period 13 is the period of smooth passenger flow before the National Day and the New Year.

S6), the correction of the division result of the operation period, since the time span of operation period 3, 6, and 8 is only one day or a few days, for the high-speed railway passenger transportation management department, the large-scale adjustment of the train running plan to meet the needs of passenger flow in these periods will affect the Existing transportation plans cause too much disruption and consume too much human and material resources. Therefore, according to the field work experience, in this embodiment, three operation periods with days less than 7 days are merged with adjacent operation periods, and the corrected high-speed railway operation period division results are shown in Table 2.

Table 2 The revised results of the division of operating periods

The corrected high-speed railway operating time division results can be summarized as follows: operating time period 1 is the smooth passenger flow period after the New Year and before the Spring Festival; operating time period 2 is the peak passenger flow period of the Spring Festival; operation period 3 is the time span between the Spring Festival travel period and the summer travel period The time span of operation period 4 is the peak period of passenger flow during the summer transportation period; the time span of operation period 5 is the period of smooth passenger flow between the summer transportation period and the National Day holiday; the time span of operation period 6 is the passenger flow of "National Day on the 11th National Day" Peak period; operating period 7 time spans the National Day holiday and the flat period of passenger flow before the New Year. The above time division conclusion can be used as the premise for the evaluation and adjustment of the train operation plan, and evaluate the adaptability of the passenger flow demand and the train operation plan obtained according to the prediction in each operation period. If the evaluation result is not satisfactory, the current train operation needs to be evaluated adjust the plan;

S7), passenger flow demand adaptability assessment, in order to illustrate that the train operation plan formulated based on the operating period division results has better adaptability to passenger flow demand, based on the division of high-speed railway operating periods, according to the average passenger flow of each period. Compile the train operation plan, and simulate and calculate the train operation plan and various adaptability evaluation indicators of passenger flow demand at each time period. At the same time, according to the time division of the actual operation of the high-speed railway in 2014 and 2015, the train operation plan and the adaptability of passenger flow demand were compared. The results are shown in Table 3.

Table 3 compares the results with the actual operating conditions

It can be seen from Table 3 that under the premise that the number of large-scale adjustment of the train operation plan remains unchanged, the train operation plan prepared according to the operation period division results of the neighbor propagation clustering algorithm has better adaptability to the passenger flow demand. Among them, in 2014, the passenger flow demand satisfaction rate, the average train occupancy rate and the passenger flow direct rate increased by 7.6%, 16.7% and 14.1% respectively. In 2015, the above three indicators increased by 5.7%, 18.4% and 14.4% respectively.

In this embodiment, combined with the survey data of the daily passenger traffic at the stations along the line, the neighbor propagation algorithm is used to cluster and merge the time points with similar passenger flow in the year, and the optimal number of clusters is determined according to the CH, Hart and IGP indicators. The division method of the annual operation period of high-speed railway is designed, and the main conclusions are as follows

(1) The high-speed railway operating time division method based on cluster analysis, combined with the result test of the cluster validity index, can more objectively and accurately reflect the passenger flow demand in different periods of the year, and overcome the subjectivity, low efficiency and low efficiency of manual division methods. The disadvantage of low precision is to lay the foundation for the adaptive adjustment of the train running plan.

(2) The case study using the statistical data of passenger traffic at a station along a high-speed railway as a sample shows that by determining the best clustering results of the annual operation period and manually analyzing the clustering results, the entire The year is divided into reasonable operating periods.

What is described in the above-mentioned embodiments and specification is only to illustrate the principle and best embodiment of the present invention. Without departing from the spirit and scope of the present invention, the present invention will also have various changes and improvements, and these changes and improvements all fall within the scope of the present invention. within the scope of the claimed invention.

Claims (7)

1. a kind of high-speed railway passenger flow Time segments division method based on neighbour's propagation clustering, comprising the following steps:

S1), a year is divided into T time point, and setting any one high-speed rail route X includes n important websites, statistics should Each time interval t on high-speed railway route X upstream or downstream direction_kInterior each website passenger flow traffic volume constructs passenger flow square Battle array, i.e.,

Wherein, X_kIndicate time interval t_kThe passenger flow traffic volume of interior high-speed rail route X,Indicate time interval t_kInterior high-speed rail route X's The passenger flow traffic volume of n-th of website；

S2), judge whether the passenger flow traffic volume of each time point is abnormal using threshold value δ, specially statistical time point t_lIt is adjacent The passenger flow at m time point send vector, and calculate its mean valueIf

Then think time point t_lPassenger flow traffic volume data be non-abnormal data, be otherwise abnormal data；

Wherein, X_lFor time point t_lPassenger flow traffic volume；

If data exception, suppressing exception data, and Lagrange's interpolation is utilized, according to the adjacent m time at the time point The passenger flow data of point is fitted reparation to its passenger flow traffic volume, and calculating formula is as follows:

In formula, l_kIt (t) is k-1 polynomial fitting, L_n(t) be Lagrange interpolation polynomial, t be time point to be fitted, t_j For j-th of time point, t_iFor i-th of time point, X_iFor passenger flow matrix, the l at i-th of time point_i(t) multinomial for i-1 fitting Formula；

Data after fitting is repaired are merged with former normal data, and scale between eliminating variable is then standardized using standard deviation Difference, calculating formula is as follows:

In formula, Z-score is the value carried out after standard deviation standardization, and x is website passenger's traffic volume at some time point,For year The mean value of all website passenger traffic volumes in spending, σ are the standard deviation of all website passenger traffic volumes in year；

S3), Time segments division is carried out based on neighbour's propagation clustering algorithm, by the passenger flow at T time point in passenger flow traffic volume data set Traffic volume data set is represented as candidate class, judges the similarity s (i, k) of the passenger flow traffic volume in any 2 time intervals, phase The passenger flow traffic volume sample X of time interval k is indicated like degree s (i, k)_kWith a passenger flow traffic volume sample X of interval i_iSimilarity, That is sample X_kIt is suitable as sample X_iThe quantization degree that represents of class, when algorithm initialization, it is assumed that all samples were represented for class Possibility is identical, that is, assumes that all s (k, k) are identical Attraction Degree intermediate value p, wherein the calculating formula of similarity of any two sample Are as follows:

S (i, k)=- | | x_i-x_k||²；

Wherein, x_iIndicate the passenger flow traffic volume sample of i, x_kIndicate the passenger flow traffic volume at k moment；

Define reliability matrix r and availability matrix a, wherein reliability matrix r (i, k) is from sample x_iIt is directed toward sample x_k, table This x of sample_kIt is suitable as x_iClass represent representative degree；A (i, k) is from sample x_kIt is directed toward sample x_i, indicate x_iSelect x_kMake The appropriate level represented for class；For arbitrary sample x_iCalculate other times interval passenger flow traffic volume confidence level r (i, k) and can The sum of expenditure a (i, k), if the maximum sample x of sum of the two_kFor class representative, all time point category division results are exported；

Specifically comprise the following steps:

S301), the initial value that reliability matrix r (i, k) and availability matrix a (i, k) is arranged is 0；

S302), calculate arbitrary time span passenger flow traffic volume sample similarity matrix s (i, k), matrix value using it is European away from From to estimate, i.e. s (i, k)=- | | x_i-x_k||²；

It is identical Attraction Degree intermediate value that diagonal entry s (k, k), which is arranged, i.e.,

It is N sample size in formula；

S303), reliability matrix r (i, k) and availability matrix a (i, k) are updated, wherein reliability matrix r (i, k) updates meter Calculate formula are as follows:

Availability matrix a (i, k) updates calculation formula are as follows:

S304), setting damping factor λ eliminates the number concussion in iteration, that is,

In formula, r_new(i, k) and r_old(i, k) is respectively the reliability matrix that this is obtained with last update；a_new(i, k) and a_old(i, k) is respectively the availability matrix that this is obtained with last update；λ ∈ (0,1) is damping factor；

S305), the confidence level and availability of itself and all passenger flow traffic volume samples are sought to any passenger flow traffic volume data sample The sum of, according toFind the class central sample of each sample；

S306), current iteration number updates n ← n+1, judges whether information iteration process reaches the maximum number of iterations of setting, That is n≤N_max, it is then algorithm termination, exports all time point category divisions as a result, otherwise return step S302)；

S4), Calinski-Harabasz, Hartigan and In- of different time points category division result are calculated separately Group Proportion index selects optimal time point classification number and its corresponding category division result；

S5), Time segments division product test and correction are runed, all division classifications of traversal loop carry out two-by-two each sample Comparative analysis, if time point corresponding to two samples be it is adjacent, merge into an operation period, otherwise, be considered as another One operation period；

S6), passenger flow demand adaptability teaching after runing Time segments division, is worked out train according to day part passenger flow average demand and is started Scheme simultaneously carries out bus traveler assignment simulation, introduces that passenger flow satisfactory rate of information demand, train be averaged attendance and passenger flow is gone directly rate three and referred to Mark carries out quantitative evaluation to the adaptedness of day part passenger flow demand and train running scheme and summarizes.

2. a kind of high-speed railway passenger flow Time segments division method based on neighbour's propagation clustering according to claim 1, special Sign is: in step S4), the Calinski-Harabasz index is mean dispersion error matrix and class in the class based on whole samples Between mean dispersion error matrix estimate, the corresponding class number of maximum value is as preferable clustering number, i.e.,

In formula, k is cluster numbers, trB (k) be between mean dispersion error matrix mark, trW (k) is the mark of mean dispersion error matrix in class, and n is time point Sample size.

3. a kind of high-speed railway passenger flow Time segments division method based on neighbour's propagation clustering according to claim 1, special Sign is: in step S4), the Hartigan index is used for the case where cluster numbers are 1, meets the infima species number of Ha≤10 As preferable clustering number, i.e.,

In formula, k is cluster numbers, and trW (k) is the mark of mean dispersion error matrix in class, and n is time point sample size.

4. a kind of high-speed railway passenger flow Time segments division method based on neighbour's propagation clustering according to claim 1, special Sign is: in step S4), the In-Group Proportion index is used to measure each sample of distance in certain one kind Whether nearest sample is in same class, and the average IGP index of all clusters is bigger, and the quality for indicating cluster is better, maximum value Corresponding class number is preferable clustering number, i.e.,

In formula, u is the category of certain cluster, and Class (j) is the category of sample j, j^NFor the nearest sample of distance sample j, # is to meet The number of condition.

5. a kind of high-speed railway passenger flow Time segments division method based on neighbour's propagation clustering according to claim 1, special Sign is: in step S6), the passenger flow satisfactory rate of information demand is used to embody each passenger flow whereabouts of high-speed railway and related road network Between, the passenger's conveying capacity and passenger flow need satisfaction degree that train running scheme provides especially are arranged in transport capacity resource condition Under the constraint of vehicle staffing condition, by the Passenger Traffic that can obtain effective transportation service under the conditions of set train running scheme It is indicated with the ratio of passenger flow total demand, calculation formula is as follows:

In formula, q'_wFor the passenger flow total amount that passenger flow OD conveys high-speed railway w, w is road network passenger flow whereabouts number.

6. a kind of high-speed railway passenger flow Time segments division method based on neighbour's propagation clustering according to claim 1, special Sign is: in step S6), the train attendance that is averaged refers to the average value of all train attendances in range of value, column Vehicle is averaged the volume of the flow of passengers that attendance refers to that train is carried in its running section and the train provides the weighting ratio of seat seat sum, this Index is used to reflect passenger between OD pairs of different passenger flows to various types bullet train selection result, and train is averaged attendance Calculation formula is as follows:

In formula,For the volume of the flow of passengers that train h is carried at section (i, j), A_hFor the staffing number of train h, E_hFor the area of train h operation Number of segment.

7. a kind of high-speed railway passenger flow Time segments division method based on neighbour's propagation clustering according to claim 1, special Sign is: in step S6), passenger flow demand structure is made of different demand directions, and each demand direction can have arrival mesh Ground through or transfer riding scheme, passenger flow rate of going directly refers in set train running scheme and passenger flow demand structure Under, the volume of the flow of passengers to go directly to destination between each passenger flow demand point pair without transfer removes upwards the always ratio of the volume of the flow of passengers with this, Its calculation formula is:

In formula, w is road network passenger flow whereabouts number, | e | it is the number of transfer of some whereabouts passenger flow,It is straight between not changed to w for passenger flow OD The passenger number arrived at the destination is connect,Pass through for passenger flow OD between w | e | it is secondary to change to the passenger number arrived at the destination.