CN110443422B

CN110443422B - OD attraction degree-based urban rail transit OD passenger flow prediction method

Info

Publication number: CN110443422B
Application number: CN201910717323.5A
Authority: CN
Inventors: 陈�峰; 张金雷; 王蕊; 朱亚迪; 胡舟
Original assignee: Beijing Jiaotong University
Current assignee: Beijing Jiaotong University
Priority date: 2019-08-05
Filing date: 2019-08-05
Publication date: 2021-11-19
Anticipated expiration: 2039-08-05
Also published as: CN110443422A

Abstract

The invention provides an OD passenger flow prediction method for urban rail transit based on OD attractiveness. The method comprises the following steps: counting attraction values among OD pairs at different time periods in the rail transit network and passenger flows among the corresponding OD pairs according to historical data, wherein the attraction values are respectively expressed as an OD attraction matrix and a first OD matrix, and the attraction values reflect the degree of attraction of the OD pairs to the passenger flows; dividing the attraction degree value between the OD pairs into a plurality of attraction degree grades and selecting a reference grade; according to the OD attraction degree matrix, extracting OD pairs with attraction degree values exceeding the reference level from the first OD matrix to form a second OD matrix; and inputting the second OD matrix into a deep learning model, and obtaining the predicted passenger flow between OD pairs through training. The method can improve the accuracy, effectiveness and real-time performance of passenger flow prediction.

Description

OD attraction degree-based urban rail transit OD passenger flow prediction method

Technical Field

The invention relates to the technical field of rail transit operation management, in particular to an OD (origin-destination) passenger flow prediction method for urban rail transit based on OD attractiveness.

Background

Urban rail transit has been developed explosively in China in recent years by virtue of the advantages of large transportation capacity, high speed, high punctuality, low pollution, high safety and the like, and becomes an indispensable public transport means in large and medium-sized cities. By the end of 2017, 165 urban rail transit operation lines are opened in 34 cities in China, the total mileage is up to 5032.7 kilometers, wherein 3883.6 kilometers of subways account for 77.2 percent. In addition, 254 lines in 56 cities are under construction, the total mileage is up to 6246.3 km, and the number of cities under construction, the number of lines under construction and the number of kilometers under construction exceed the operation scale. The total number of the planned and developed urban rail transit covered cities is 62, and the total mileage of all planned lines is up to 7321.1 kilometers. Such large-scale construction activities also put more stringent requirements on the operation management of the constructed urban rail transit.

In the operation management work of urban rail transit, no matter the train operation chart and the operation management strategy are formulated and adjusted, or the real-time passenger transport organization work is carried out, the real-time space-time distribution characteristics of passenger flow need to be accurately mastered, and the space-time distribution of the passenger flow of the road network in a short period of time is predicted in time, so that how to carry out timely refined short-time passenger flow prediction becomes an important research content.

For urban rail transit, the short-time passenger flow prediction can be divided into station-entering short-time passenger flow prediction, OD (origin destination) short-time passenger flow prediction and section short-time passenger flow prediction. The OD short-time passenger flow prediction, which is an important bridge connecting the inbound short-time passenger flow prediction and the section short-time passenger flow prediction, is a predicted passenger flow destination of a known travel starting point, and can provide decision support for real-time network operation management, such as passenger flow congestion control, abnormal order detection, and the like, and also provide input information for real-time path planning and dynamic allocation.

In the prior art, in the aspect of OD short-time passenger flow prediction, there are technical solutions: 1) the short-time arrival quantity of the urban rail transit is predicted by adopting a weighted historical average self-adaptive model, then short-time OD and section short-time passenger flow prediction results are obtained by means of OD matrix prediction and multi-body simulation modeling, and verification is carried out by adopting a space-time binary verification method, so that the whole short-time passenger flow prediction system is relatively perfect. 2) The short-time passenger flow OD matrix estimation model is constructed based on a dynamic OD matrix estimation model under a moving average strategy proposed by a least square method and a state space method. 3) The short-time passenger flow OD multi-model combined estimation method based on the state space model and the multi-time granularity is provided on the basis of the least square method, and accuracy of short-time OD estimation is improved.

However, the OD short-time traffic prediction in the prior art mainly has the following problems: 1) and extremely complex space-time relations exist among OD passenger flows, a single mathematical optimization method such as a least square method, a state space model and the like cannot well depict the complex space-time relations, and the accuracy needs to be improved. 2) Most of the existing researches are applied to small-scale subway networks, and along with the expansion of the networks, the calculation complexity of the model is greatly improved, and the requirement of real-time performance cannot be met. 3) The number of the OD pairs is far larger than the number of stations and is n-1 times of the number of the stations (n is the number of the stations), but the passenger flow between a plurality of the OD pairs is very small or even has no passenger flow, so that all the OD pairs are treated equally in the existing research, the complexity and the calculated amount of a model are increased, and negative effects are generated on a prediction result. 4) The selection of the time granularity is a basic work for short-time passenger flow prediction, although refined information of the passenger flow can be described by the passenger flow sequence extracted under the smaller time granularity, the regularity is poor, the prediction accuracy is lower, and although the passenger flow sequence extracted under the larger time granularity loses the detail information of the passenger flow, the regularity is stronger, the prediction accuracy is higher, and how to grasp the selection of the time granularity is also one of the existing problems.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provides an OD passenger flow prediction method for urban rail transit based on OD attractiveness, which can accurately predict short-time OD passenger flow distribution and guide traffic operation management.

According to the first aspect of the invention, an OD passenger flow prediction method for urban rail transit based on OD attractiveness is provided. The method comprises the following steps:

step S1: counting attraction values among OD pairs at different time periods in the rail transit network and passenger flows among the corresponding OD pairs according to historical data, wherein the attraction values are respectively expressed as an OD attraction matrix and a first OD matrix, and the attraction values reflect the degree of attraction of the OD pairs to the passenger flows;

step S2: dividing the attraction degree value between the OD pairs into a plurality of attraction degree grades and selecting a reference grade;

step S3: according to the OD attraction degree matrix, extracting OD pairs with attraction degree values exceeding the reference level from the first OD matrix to form a second OD matrix;

step S4: and inputting the second OD matrix into a deep learning model, and obtaining the predicted passenger flow between OD pairs through training.

In one embodiment, the attraction value is an average passenger flow volume between pairs of ODs, and the OD attraction matrix is:

the first OD matrix is represented as:

wherein i is the number of inbound, j is the number of outbound, n is the total number of days considered in the consecutive time period, k is a specific day in the time period, t is the t-th time period in the k-th day, q is the OD passenger flow between i and j in the t-th time period of the k-th day,

is a row vector.

In one embodiment, step S3 includes: deleting OD pairs equal to and less than the reference level from the first OD matrix; and synchronously moving each OD pair in the first OD matrix to the left, and supplementing 0 to the vacant position to obtain the second OD matrix.

In one embodiment, the criteria for classifying the level of attractiveness is determined based on a temporal granularity of statistical passenger flow.

In one embodiment, the attractiveness level is divided according to the following principle:

in case the time granularity of the statistical passenger flow is 15 minutes: when the attraction value ODAD is equal to 0, the lowest level corresponds to the ODAD, the ODAD is more than or equal to 0 and less than or equal to 1, the low level corresponds to the ODAD, the ODAD is more than or equal to 1 and less than or equal to 2, the middle level corresponds to the ODAD, the ODAD is more than or equal to 2 and less than or equal to 3, the high level corresponds to the ODAD, and the ODAD is more than or equal to 3, the highest level corresponds to the ODAD;

when the time granularity is 30 minutes, the lowest level corresponds to the attraction value ODAD being equal to 0, the low level corresponds to the ODAD being more than 0 and less than or equal to 2, the medium level corresponds to the ODAD being more than 2 and less than or equal to 4, the high level corresponds to the ODAD being more than 4 and less than or equal to 6, and the highest level corresponds to the ODAD being more than 6;

in the case of a time granularity of 60 minutes, an attraction value ODAD equal to 0 corresponds to the lowest level, ODAD greater than 0 and equal to 4 or less corresponds to the low level, ODAD greater than 4 and equal to 8 or less corresponds to the medium level, ODAD greater than 8 and equal to 12 or less corresponds to the high level, and ODAD greater than 12 corresponds to the highest level.

In one embodiment, the time granularity is 30 minutes, and the attraction level is a low level as a reference level.

In one embodiment, the deep learning model is a long-and-short memory network.

In one embodiment, in step S4, the training process uses one or more of the root mean square error, the average absolute error, and the weighted average absolute percentage error to evaluate the accuracy of the prediction.

Compared with the prior art, the invention has the advantages that: the OD short-time passenger flow prediction method based on the OD attractiveness, which is disclosed by the invention, firstly puts forward the concept of the OD attractiveness, and divides all OD pairs of the whole network into a plurality of OD attractiveness levels; verifying the effectiveness of the proposed OD attraction degree on improving the prediction precision by using a deep learning method; the best combination of OD attractiveness level and time granularity was explored to guide engineering practice.

Drawings

The invention is illustrated and described only by way of example and not by way of limitation in the scope of the invention as set forth in the following drawings, in which:

FIG. 1 is a flow chart of an OD passenger flow prediction method for urban rail transit based on OD attractiveness according to one embodiment of the invention;

FIG. 2 is a schematic diagram of a memory module of the long and short term memory network LSTM;

FIG. 3 is a schematic structural diagram of the LSTM model;

FIG. 4 is a graph of OD versus number after applying different levels of OD attractiveness at different time granularity, according to one embodiment of the present invention;

FIG. 5 is a model evaluation result at different time granularities and OD attractiveness levels for different time periods according to one embodiment of the invention;

FIG. 6 is a graph of mean model estimates at different time granularity and OD attractiveness levels throughout the day for all periods of time, according to one embodiment of the invention.

Detailed Description

In order to make the objects, technical solutions, design methods, and advantages of the present invention more apparent, the present invention will be further described in detail by specific embodiments with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

In all examples shown and discussed herein, any particular value should be construed as merely illustrative, and not as a limitation. Thus, other examples of the exemplary embodiments may have different values.

Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.

According to one embodiment of the invention, an OD passenger flow prediction method for urban rail transit based on OD attractiveness is provided, which comprises the following steps: defining OD attraction degree and OD attraction degree level by taking subway card swiping data as an example; extracting an OD matrix based on the OD attraction degree; and taking the extracted OD matrix as an input of an LSTM model to obtain a short-time OD passenger flow predicted value. Specifically, referring to fig. 1, the method comprises the steps of:

step S110, counting the passenger flow attraction value between each OD pair according to the historical data.

Herein, the OD attraction degree is used to represent the degree of attraction of the passenger flow between the OD pairs, for example, the average passenger flow between the OD pairs at different periods of time over a long period of time is used as an index describing the degree of attraction of the passenger flow between the OD pairs.

In the present invention, a subway card swiping data is taken as an example for detailed description, and the used data is Beijing subway card swiping data of 2016 year 2, month 29 to 2016 year 4, month 3 for 5 weeks continuously, which comprises 17 lines and 276 stations (not including airport lines), so there are 76176 OD pairs in total, and the card swiping time is 05:00 to 23: 00.

Research shows that OD with low passenger flow volume has obvious influence on prediction accuracy, while in an urban rail transit network, a large number of OD pairs have little or no OD passenger flow, so that travel behaviors generated among the OD pairs attracting little passenger flow are mostly generated randomly, and the regularity is low, thereby influencing the prediction accuracy. In order to eliminate the influence of random factors on the prediction process and improve the prediction accuracy, the concept of OD attraction degree (ODAD) is introduced herein, for example, the OD attraction degree, i.e., the average passenger flow rate among ODs in different periods over a long time, is an index for describing the degree of attraction passenger flow between OD pairs. The matrix of attractiveness between pairs of ODs may be calculated using the following equation:

wherein i is the number of inbound, j is the number of outbound, n is the total number of days considered in the continuous time period, k is a specific certain day in the time period, t is the t time period in the k day, and q is the OD passenger flow between i and j in the t time period on the k day. As can be seen from equation (1), the OD attraction is a dynamic indicator, and for the same OD pair, the value of the OD attraction may be higher during peak hours and lower during midnight hours. Further, the OD attraction degree is an average index, and therefore the influence of random factors is excluded.

In the embodiment of the invention, the value ranges of i and j are 1 to 276, which represents 276 stations. n is 25, representing 25 workdays from 29/2016 to 3/4/2016, and k ranges from 1 to 25. When the predicted time granularity is 15 minutes, the variation range of t is 1-72 (18h), and when the predicted time granularity is 30 minutes or 60 minutes, the variation range of t is 1-36 and 1-18 respectively.

Step S120, determining an attraction degree grade division standard and determining the grade corresponding to each OD pair according to the attraction degree value.

In order to further improve the accuracy of prediction, the embodiment of the invention divides the OD attractiveness into a plurality of levels, and gives a division standard of the OD attractiveness level based on the OD attractiveness value and the time granularity. For example, when the time granularity is 15 minutes, the OD attraction level is shown in table 1 below, which can be used as a reference standard for OD matrix pre-processing. When the time granularity was 30 minutes or 60 minutes, the corresponding attraction value (ODAD) became 2-fold or 4-fold, respectively.

Table 1: OD attraction degree horizontal division standard

In table 1, the unit of the ODAD value is the number of passengers.

Step S130, selecting a reference OD attraction degree grade, extracting an OD pair higher than the reference OD attraction degree grade, and obtaining an OD matrix used for reflecting the historical passenger flow among the ODs.

In this step S130, an OD matrix extraction process for the LSTM model is given based on the OD attraction proposed by the present invention.

First, the historical OD matrix extracted from the processed card swiping data is expressed as the following formula (2), the extracted OD matrix is in one-to-one correspondence with the OD attraction degree matrix (formula (1)), then a certain OD attraction degree level is selected as a reference (for example, the OD attraction degree is "low"), and the OD pairs equal to or less than the selected OD attraction degree level in the OD matrix are deleted according to the OD attraction degree matrix, and then the OD matrix is converted into the following formula (3). Finally, as the OD pairs are deleted, the variable length sequences appear, so that the OD pairs are synchronously shifted to the left, and the vacant positions are supplemented with 0 to obtain the finally required OD matrix, as shown in formula (4).

Wherein the content of the first and second substances,

is the OD matrix for the kth time period on day k,

OD traffic going from station i to station j for the kth time period on day k,

is a row vector.

And step S140, taking the extracted OD matrix as input, and obtaining the predicted passenger flow between each OD pair by using a deep learning network.

The deep neural network adopted by the embodiment of the invention is an LSTM model, the LSTM model can receive a variable length sequence by means of a masking layer, namely, 0' of a complete vacant position can be automatically filtered in the model training process, so that the vacant position is completely output as 0, and the characteristic ensures that the introduced OD attractiveness concept can improve the prediction precision. Subjecting the product obtained in the formula (4)

It is used as input to the LSTM model.

Note that the OD pairs in the prediction result are located at the same positions as the input. Thus, after determining the output, the OD response should be relocated to its original position (i.e., from the position of equation (4) to the position of equation (2)).

Compared with the traditional recurrent neural network RNN model, the LSTM model can effectively solve the problems of gradient disappearance and gradient explosion generated in the long-term dependence process, selectively extracts historical useful information and transmits the useful information to the current moment for prediction. The memory block of the LSTM model is shown in fig. 2 and mainly consists of an input gate, an output gate, a forgetting gate and memory cells, wherein the information flow is transmitted according to equations (5) - (12).

f_t＝σ(W_fx_t+U_fh_t-1+V_fc_t-1+b_f) (5)

i_t＝σ(W_ix_t+U_ih_t-1+V_ic_t-1+b_i) (6)

o_t＝σ(W_ox_t+U_oh_t-1+V_oc_t+b_o) (9)

h_t＝o_t⊙tanh(c_t) (10)

Wherein x_t，i_t，o_t，f_t，c_tAnd h_tRepresenting the input data, input gate, output gate, forget gate, cell state and final output, respectively, and their positions in the LSTM memory block are shown in fig. 2. W, U and V denote weight matrices, and b is a bias vector. t is a time step, and a indicates a Hadamard product. c. C_tAnd h_tIs set to 0. In the embodiment of the invention, a time sequence propagation through time (BPTT) and an rmsprop optimizer are used for optimizing the model parameters.

The present invention uses the root mean square error RMSE, the mean absolute error MAE and the weighted mean absolute percentage error WMAPE to evaluate the model prediction results. Wherein three indexes at different time periods in a day are calculated according to the expressions (13) to (15), and the average value of the three indexes in a day is calculated according to the expressions (16) to (18).

Wherein

In order to predict the value of the target,

and the real value is obtained, t represents different time periods in the kth day, n represents the number of stations, and m represents the number of time periods in one day.

In order to verify the effect of the invention, an LSTM model is adopted to verify the application effect of the OD attraction degree, and on the basis of the LSTM model and the OD attraction degree, 9 cases with 3 time granularity and 3 OD attraction degree levels are selected to verify the rationality of the proposed OD attraction degree index and explore the optimal combination mode of the OD attraction degree level and the time granularity.

For the urban rail transit system, the passengers can get in the station by swiping cards, the time of getting in the station can be obtained in real time, however, the passengers need certain travel time when arriving at the destination station, and time lag exists, so that the real-time OD matrix cannot be obtained. In order to predict the OD matrix of the t-th time period on the k-th day, the OD matrix before the t-th time period cannot be used as input naturally, otherwise, errors are accumulated, and the like. Based on the method, the OD matrixes of the t time period of the k-1 day and the t-1 time period of the k-1 day and the real-time traffic volume of the station entering and exiting on the k day are used as the input of the LSTM model. In addition, the method adopts the parallel computing technology to respectively predict the OD sequences of 276 stations, finally combines the prediction results into a complete OD matrix as final output and evaluates the output. The inputs to the LSTM model are shown in equations (19) - (20).

Wherein

The row vectors in the formula (4) respectively represent OD passenger flow vectors of station arrival from the station i in the t time period on the k day, the t time period on the k-1 day and the t time period on the k-1 day,

and outputting the OD passenger flow for final prediction.

Indicating OD traffic after OD sync left shift and fill 0.

And

and (5) the traffic volume of the station entering and exiting at the t time period on the kth day.

In the embodiment of the invention, OD data of the first 22 days are used as a training set, data of the 23 th day are used as a test set to test a model, 3 time granularities (15 minutes, 30 minutes and 60 minutes) and three OD attractiveness levels (lowest, low and medium) are selected to predict 9 cases in total, and the optimal combination of the time granularity and the OD attractiveness levels is explored according to the prediction result to guide engineering practice.

The network structure adopted by the invention is shown in figure 3 and comprises a masking layer, two LSTM layers and a dense layer, wherein an input layer is represented as a sequential layer, and an output layer is represented as a Linear Activation layer. There are 276 features per time step in the LSTM input, and 4 classes of input data represent 4 time steps respectively. Due to the existence of the masking layer, the filling 0 can be filtered out completely, so that the LSTM can be trained on a variable-length sequence, and the complexity and the calculation amount of the model are reduced remarkably. The epoch parameters were set to 50, 100 and 200 for 15 minutes, 30 minutes and 60 minutes, respectively. The batch size is set to 72.

After verification, the remaining number of OD pairs is shown in fig. 4 after applying different OD attraction levels, and it can be seen from the figure that when the OD attraction level is "lowest" (reference attraction level), the OD pairs with no passenger flow volume of about 10000-. As can be seen from this figure, there is a large number of OD pairs with little or no traffic, which severely affects the prediction accuracy. It can also be seen that the level of OD attraction should be chosen carefully, both to reduce the influence of the OD pairs with lower attraction on the prediction accuracy as much as possible and to prevent the omission of too many OD pairs.

The results of model evaluation at different time granularities and OD attraction levels at different times of the day calculated according to equations (13) - (15) are shown in fig. 5. It can be seen from the figure that, as the level of the OD attractiveness is increased, all indexes are reduced at all time granularities, wherein the indexes with the OD attractiveness reduced from the lowest degree to the low degree are most obvious, and the indexes from the low degree to the medium degree are reduced to a certain extent, so that the level of the OD attractiveness adopted by the method is low, the prediction accuracy can be greatly improved, and the phenomenon that too many OD pairs are ignored in model evaluation can be avoided.

The average model evaluation results at different time granularities and OD attractiveness levels for all time periods throughout the day calculated according to equations (16) - (18) are shown in fig. 6 (for three bars for each time granularity, corresponding to the lowest, low and medium attractiveness levels in order from left to right) and tables 2-4. As can be seen from the figure, the performance of each index at all time granularities increased with increasing level of OD attraction applied. At the same time according to

Table 4 shows that when the time granularity is 30 minutes and the OD attraction level is "low", WMAPE starts to decrease to below 30%, whereas in the existing research, the index is difficult to decrease to below 30%, so according to the verification result, the recommended time granularity of the present invention is 30 minutes, which not only ensures the prediction accuracy, but also avoids the omission of the passenger flow detail information caused by the excessively large time granularity.

Table 2: average RMSE throughout the day

Table 3: average MAE throughout the day

Table 4: average WMAPE throughout the day

As can be seen from the above verification, the effect of applying the OD attraction degree on improving the model accuracy is significant, and three evaluation indexes will be described below with reference to tables 2 to 4 when the OD attraction degree level (i.e., the reference OD attraction degree level) is from "lowest" to "medium".

For the RMSE, it can be seen from Table 2 that the effect was increased by 10.14%, 5.46% and 3.56% for the 15 minute time particle size from 1.38 to 1.24, the 30 minute time particle size from 2.38 to 2.25 and the 60 minute time particle size from 4.49 to 4.33, respectively.

For the MAE, it can be seen from Table 3 that the effect increases by 42.19%, 45.83% and 46.48% for the 15 minute time particle size down to 0.64, the 30 minute time particle size down to 0.52, and the 60 minute time particle size down to 0.76, respectively, from 0.96 to 0.52 and from 1.42 to 0.76.

As can be seen from Table 4, for WMAPE, the particle size decreased from 46.91% to 31.10% in 15 minutes, from 36.45% to 22.97% in 30 minutes, and from 28.22% to 17.12% in 60 minutes, the effects were respectively improved by 15.81%, 13.48%, and 11.10%.

The invention combines the prediction result to give the best combination of time granularity and OD attraction level, namely preferably, the time granularity can be selected from 30 minutes, and the OD attraction level is 'low' to guide the engineering practice.

In summary, the invention provides the concept of OD attraction for solving the problems of low accuracy, poor real-time performance, large number of OD pairs and the like in the existing OD short-time passenger flow prediction field, and divides all OD pairs in the whole network into a plurality of attraction levels based on the OD attraction. In addition, the effectiveness of the OD attraction degree to improve the prediction accuracy is verified by using lstm (long short term memory) in deep learning. Further, a total of 9 cases of 3 passenger flow time granularities and 3 OD attractions were selected for testing to evaluate the effectiveness of the model and to explore the best combination of time granularity and OD attractions. The model framework provided by the invention can improve the accuracy of OD short-time passenger flow prediction, meet the requirement of real-time performance, has guiding significance to engineering, and can meet the requirements of the urban rail transit short-time passenger flow prediction on the accuracy and the real-time performance so as to realize the real-time and dynamic operation management.

It is understood that the skilled person can make appropriate modifications or changes to the above embodiments without departing from the spirit of the present invention, for example, the attraction value is calculated by using the relative value of the average passenger flow between the ODs with respect to a reference passenger flow, and for example, the deep learning network may use a gru (gated Recurrent unit) model or the like in addition to the LSTM.

It should be noted that, although the steps are described in a specific order, the steps are not necessarily performed in the specific order, and in fact, some of the steps may be performed concurrently or even in a changed order as long as the required functions are achieved.

The present invention may be a system, method and/or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied therewith for causing a processor to implement various aspects of the present invention.

The computer readable storage medium may be a tangible device that retains and stores instructions for use by an instruction execution device. The computer readable storage medium may include, for example, but is not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing.

Having described embodiments of the present invention, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An OD (origin-destination) attraction degree-based urban rail transit OD passenger flow prediction method comprises the following steps:

step S1: counting the attraction degree values among OD pairs at different time periods in the rail transit network and the passenger flows among the corresponding OD pairs according to historical data, and respectively representing the attraction degree values as an OD attraction degree matrix and a first OD matrix, wherein the OD attraction degree is an index for describing the degree of attraction of the passenger flows among the OD pairs, and the attraction degree values reflect the degree of attraction of the passenger flows among the OD pairs;

step S2: dividing the attraction degree value between the OD pairs into a plurality of attraction degree grades and selecting a reference grade, wherein the division standard of the attraction degree grades is determined according to the time granularity;

step S4: inputting the second OD matrix into a deep learning model, and training to obtain the predicted passenger flow between OD pairs, wherein the deep learning model is a long-time and short-time memory network;

wherein the attraction value is an average passenger flow volume between OD pairs, and the OD attraction matrix is:

the first OD matrix is represented as:

is a row vector.

2. The method according to claim 1, wherein step S3 includes:

deleting OD pairs equal to and less than the reference level from the first OD matrix;

and synchronously moving each OD pair in the first OD matrix to the left, and supplementing 0 to the vacant position to obtain the second OD matrix.

3. The method of claim 1, wherein the attractiveness level is divided according to the following principle:

when the time granularity is 30 minutes, the lowest level corresponds to the attraction value ODAD being equal to 0, the low level corresponds to the ODAD being greater than 0 and less than or equal to 2, the medium level corresponds to the ODAD being greater than 2 and less than or equal to 4, the high level corresponds to the ODAD being greater than 4 and less than or equal to 6, and the highest level corresponds to the ODAD being greater than 6;

4. The method according to claim 3, wherein the time granularity is 30 minutes, and the attraction level is a low level as a reference level.

5. The method of claim 1, wherein in step S4, the training process uses one or more of root mean square error, mean absolute error, and weighted mean absolute percentage error to evaluate the accuracy of the prediction.

6. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 5.

7. A computer device comprising a memory and a processor, on which memory a computer program is stored which is executable on the processor, characterized in that the steps of the method of any of claims 1 to 5 are implemented when the processor executes the program.