CN115618934A - Subway short-time passenger flow prediction method based on space-time diagram convolutional network - Google Patents
Subway short-time passenger flow prediction method based on space-time diagram convolutional network Download PDFInfo
- Publication number
- CN115618934A CN115618934A CN202211246533.9A CN202211246533A CN115618934A CN 115618934 A CN115618934 A CN 115618934A CN 202211246533 A CN202211246533 A CN 202211246533A CN 115618934 A CN115618934 A CN 115618934A
- Authority
- CN
- China
- Prior art keywords
- subway
- time
- network
- passenger flow
- space
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000010586 diagram Methods 0.000 title claims description 7
- 230000008859 change Effects 0.000 claims abstract description 11
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 23
- 125000004122 cyclic group Chemical group 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000000737 periodic effect Effects 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 description 9
- 230000002776 aggregation Effects 0.000 description 5
- 238000004220 aggregation Methods 0.000 description 5
- 239000010410 layer Substances 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 230000003595 spectral effect Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000000547 structure data Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 230000003442 weekly effect Effects 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002902 bimodal effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Optimization (AREA)
- Development Economics (AREA)
- General Engineering & Computer Science (AREA)
- General Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Marketing (AREA)
- Pure & Applied Mathematics (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Databases & Information Systems (AREA)
- Algebra (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Quality & Reliability (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
Abstract
The invention discloses a subway short-time passenger flow prediction method based on a space-time Graph Convolutional Network (GCN). The method comprises the following steps: acquiring historical data of a subway; the time dependency relationship of the subway network is obtained by learning the historical data of the subway by using the gate control circulation unit, and a hidden historical passenger flow volume change characteristic is obtainedSigned hidden state H t (ii) a And acquiring the dynamic spatial dependency relationship of the subway network by using the graph convolutional neural network so as to predict the passenger flow at the future moment. When the invention carries out space-time prediction on urban subway passenger flow, the invention considers the dynamic change characteristics of the space dependency relationship in addition to the time dependency relationship of the subway network, and obtains the passenger flow of all subway stations in the urban subway network at the time of t +1 by utilizing first-order approximate Cheb graph convolution
Description
Technical Field
The invention relates to a subway short-time passenger flow prediction method based on a space-time diagram convolutional network
Background
With the soundness of urban rail transit networks, the passenger flow intensity gradually rises, and it is particularly important to master the dynamic change trend of the passenger flow in a short period in the future. On one hand, real-time and effective information can be provided for travelers, which is beneficial to the selection of traffic modes and paths and the realization of passenger flow induction; on the other hand helps to guide the operation unit to adjust the transportation scheme in time, the transportation capacity is accurately matched, the large-passenger-flow early warning emergency work is well done, and the trip experience and the safety are improved.
The urban rail transit short-time passenger flow prediction, namely the passenger flow prediction in 5 to 15 minutes in the future, mostly takes historical AFC data as the basis and takes the subway as a research object. From the passenger flow attribute analysis, the short-time passenger flow prediction of the urban rail transit can be divided into four types, including short-time arrival passenger flow prediction, short-time departure passenger flow prediction, short-time OD passenger flow prediction and short-time passenger flow distribution prediction. In addition, short-time passenger transfer flow, in-station passenger flow, cross-section passenger flow and the like can be summarized into the category of short-time passenger flow distribution research.
The model of short-time passenger flow prediction follows up with the development of machine learning research, and especially in recent 2 years, attention mechanisms which are widely concerned and applied are obtained, and specific algorithms such as a depth residual shrinkage network and a Transformer LSTM improved model based on the attention mechanism. The most common traditional LSTM network model at present has certain defects or limitations.
A great deal of research is focused on extending convolution operators to graph structure data, and a graph neural network based on neighborhood feature aggregation is proposed. The method models the spatial dependence in the subway network by simulating the information transmission and aggregation process between the subway networks. At present, the graph convolution neural network has been widely used in the task of predicting and classifying graph structure data including urban rail transit. For example, the STGCN proposed by Yu et al is composed of two time domain convolution modules, in which the spatial dependency relationship is learned using the GLU method, and one spatial domain convolution module, in which the spatial dependency relationship is learned using the graph convolution network. Although the STGCN method has good effect on urban traffic prediction, it assumes that the spatial dependency relationship in the traffic network does not change with time. The DCRNN method proposed by Li et al is to model the spatial dependency in the directed graph traffic network by using a graph convolution network, and then input the graph data after feature aggregation into a Gated Round Unit (GRU) to learn the spatial dependency relationship. This approach still assumes that the spatial dependency relationships in the traffic network are fixed.
However, in an actual subway traffic network, the spatial dependency relationship between subway stations changes as the amount of passenger flow to and from adjacent subway stations changes. Therefore, when the space-time prediction is performed on the urban subway passenger flow, the dynamic change characteristic of the space dependency relationship must be considered in addition to the time dependency relationship of the subway network. Based on the characteristic, the invention provides a subway passenger flow prediction method considering time dependence and dynamic space dependence on the premise of keeping the topological structure of the urban subway network. The method mainly utilizes a recurrent neural network and a graph convolution network in deep learning to respectively model the time dependence and the space dependence in the urban subway network, and then predicts the passenger flow in the urban subway network.
Disclosure of Invention
In order to overcome the defects, the invention provides a subway short-time passenger flow prediction method based on a space-time diagram convolutional network.
In order to achieve the above purpose, the invention provides a method for predicting subway short-time passenger flow based on a Convolutional Network of space-time diagrams (GCN), which comprises the following steps:
acquiring historical data of a subway;
the time dependency relationship of the subway network is obtained by learning the subway historical data by using the gate control circulation unit, and a hidden state H which implies the historical passenger flow change characteristics is obtained t ;
And acquiring the dynamic space dependency relationship of the subway network by using the graph convolutional neural network so as to predict the passenger flow at the future time.
Further, the input data of the gated loop unit network includes: 1) All subway stations are at t-N time Passenger flow to time tAs a proximity timing feature; 2) All subway stations are at the front N day Passenger flow at time t +1 of dayAs a daily periodicity characteristic; 3) All subway stations are at the front N week Passenger flow at t +1 weekAs a periodic feature.
Further, the step of obtaining the dynamic spatial dependency relationship of the subway network by using the graph convolution neural network specifically comprises:
using hidden layer states H t The dynamic weight matrix of the subway network is calculated, and the calculation process is as follows:
Passenger flow of all subway stations in urban subway network at the moment of t +1 can be obtained by using first-order approximate Cheb graph convolutionComprises the following steps:
in the formulaParameters optimized for the need to train;and the passenger flow volume prediction value of each subway station in the urban subway network at the t +1 th moment.
When the invention carries out space-time prediction on urban subway passenger flow, not only the time dependency relationship of a subway network is considered, but also the dynamic change characteristic of the space dependency relationship is considered, and the t +1 moment can be obtained by utilizing the first-order approximation Cheb graph convolutionPassenger flow of all subway stations in urban subway network
Drawings
Fig. 1 is a schematic structural diagram of an embodiment of the present invention.
Fig. 2 shows the predicted results of the inventive method on the AFC dataset of the beijing subway, wherein,
(a) Predicting the result of the inbound passenger flow volume of the country trade station of line No. 1 in 10 minutes in the future;
(b) Predicting the result of the outbound passenger flow of the No. 1 wire country trade station in 10 minutes in the future;
(c) Predicting results of 10 minutes in the future of the station-entering passenger flow volume of No. 2 Lingximan;
(d) Predicting the result of 10 minutes in the future of the outbound passenger flow volume of No. 2 Linxizhmen;
(e) Predicting the result of 10 minutes in the future of the station-entering passenger flow volume of the No. 7 linear magnet device;
(f) Predicting the result of 10 minutes in the future of the outbound passenger flow volume of the No. 7 linear magnetizer port;
(g) Predicting the result of 10 minutes in the future of the number 5 line skynestem station arrival passenger flow;
(h) The result of predicting the 10 minutes future outbound passenger flow of the number 5 linear heavenly stems;
(i) Predicting results of 10 minutes in the future of the taxi-entering passenger flow volume of No. 13 line west two flags;
(j) And predicting the result of 10 minutes in the future of the taxi passenger volume when the taxi stops in west two-flag line 13.
Fig. 3 is a basic structure of a gated loop cell network.
Detailed Description
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
In the description of the present invention, it is to be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate orientations or positional relationships based on those shown in the drawings, and are merely for convenience of description and simplicity of description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed in a particular orientation, and be operated, and thus, are not to be construed as limiting the present invention.
The terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or to implicitly indicate the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless otherwise specified.
In the description of the present invention, it should be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in a specific case to those of ordinary skill in the art.
The invention provides a subway short-time passenger flow prediction method based on a space-time graph convolutional network (Seq 2 GCN). Firstly, a time dependency relationship of an urban subway network is learned by using a gate control cycle unit to obtain a hidden state H with implicit historical passenger flow change characteristics t And the dynamic spatial dependency relationship of the urban subway network is obtained by combining with the Laplace matrix of the subway network and then utilizing the graph convolution neural network, so that the passenger flow at the future moment is predicted.
1. Time-dependent modeling
Compared with the traditional cyclic neural network and the long-short term memory network, the gated cyclic unit network not only overcomes the problems of gradient disappearance and gradient explosion in the back propagation process to a certain extent, but also uses less parameters to learn the implicit relation between time sequence data on the premise of ensuring the prediction accuracy so as to reduce the calculated amount. Therefore, the time dependence of the gated cyclic unit network on the urban subway network is modeled.
The passenger flow of subway stations in an urban subway network has a time dependency relationship, the closer the time is, the stronger the dependency relationship is, and the characteristics of periodic variation in daily and weekly also exist. Therefore, the input data of the gated cyclic unit network in the Seq2GCN method provided by the present invention includes three blocks, which are respectively the adjacent timing feature, the daily cycle feature and the cycle feature, so as to more fully model the time dependency relationship.
Assuming that passenger flow of all subway stations in an urban subway network at the time of t +1 needs to be predicted, input data of a gating cycle unit network in a Seq2GCN model comprises three parts: 1) All subway stations are at t-N time Passenger flow to time tAs a proximity timing feature; 2) All subway stations are at the front N day Passenger flow at time t +1 of dayAs a daily periodicity characteristic; 3) All subway stations are at the front N week Passenger flow at t +1 weekAs a periodic feature. Wherein The passenger flow of the ith subway station at the time t.
The proximity time series signature data, the daily periodicity signature data, and the weekly periodicity signature data are then combined and input as sequence data into a gating cycle unit, as shown in FIG. 3. The input sequence data will pass through an update gate z t And combining the currently input features with the hidden state H to learn and record the current features. Then passes through a reset gate r t Insignificant history features are discarded. Finally, the gate z will be updated t And a reset gate r t And outputting and combining, and updating the state H of the hidden layer. Updating the door z in the process t Reset gate r t And the updating formulas of the hidden layer state H are shown as formula (1), formula (2) and formula (3), and the whole calculation process is shown as figure 3.
z t =σ(W xz X t +W hz H t-1 +b z ) (1)
r t =σ(W xt X t +W ht H t-1 +b r (2)
H t =z t ⊙H t-1 +(1-z t )⊙tanh(W xh X t +r t ⊙W hh H t-1 +b h ) (3)
In the formula W xz 、W hz 、W xr 、W hr 、W xh 、W hh For the weight matrix to be trained, σ is the activation function, b z 、b r 、b h For the bias coefficient, tanh (·) is a hyperbolic tangent function.
Through the training and learning of the gated cyclic unit network, H with implicit time dependency characteristics can be obtained t . The gated cyclic unit network in the Seq2GCN model provided by the invention is a single-layer network, so that the gated cyclic unit network hasWherein N is h The dimension size of the hidden layer in the network of gated loop units.
2. Dynamic spatial dependency modeling
The input data in the gate control cycle unit network is one-dimensional characteristic data after the urban subway network is expanded, and the data loses the spatial topological structure characteristics in the urban subway network. Therefore, after modeling the time dependence through the gated round-robin cell network, the Seq2GCN model in turn models the spatial dependence in the urban subway network using a spectrum Method (Spectral Method) based GCN.
The graph neural network based on spectrum method is also called GCN, and the method adopts a signal processing modeAnd information aggregation of the neighborhood subway stations is completed by performing Laplace transform on the urban subway network. For defined urban subway networksIts laplacian matrix L can be defined as:
L=D-A (4)
wherein A is an adjacency matrix and D is a degree matrix. Its symmetric normalized Laplace matrix L sym Comprises the following steps:
then the feature aggregation function of the graph neural network based on the spectral method can be defined as:
aggregate(X)=L sym X (6)
the laplacian matrix of an urban subway network has the following characteristics:
1) The Laplace matrix L is a real symmetric matrix, i.e., has L T =L;;
2) The laplacian matrix L is a semi-positive definite matrix, so it can be decomposed into the following form:
in the formulaIs the J eigenvectors of the Laplace matrixIs a matrix of column vectors, λ i Is a feature vectorI.e.: 0= λ 0 ≤λ 1 ≤…≤λ J-1 ;Λ=diag([λ 0 ,λ 1 ,…λ J-1 ])。
Based on the above characteristics, bruna et al first proposed a graph convolution neural network based on a spectral method that convolves the feature values in equation (7) with fourier transform, but the amount of computation of this method increases with the increase in the network. Deffererd et al subsequently reduced the temporal and spatial complexity of the convolution operation by optimizing the convolution kernel. The optimized convolution operation is as follows:
in the formula g θ (. Cndot.) is a convolution operation and θ is a parameter that needs to be trained by a feedback function. The value of J is far smaller than the number of subway stations in the network, so the calculation amount of convolution operation is greatly reduced.
In addition, deffererard et al also use Chebyshev polynomials instead of convolution kernels to further simplify the convolution operation. Based on the Chebyshev polynomial, there are:
whereinT j (. Cndot.) is a Chebyshev polynomial of order j, and substituting equation (9) into equation (8) yields:
kipf et al then again optimizes the graph convolution operation using a first order approximation, which sets the value of J to 1 and the maximum eigenvalue of the Laplace matrix to be approximately equal to 2, i.e., λ J-1 Is 2, and substituting equation (10) results in:
to reduce the amount of computation and prevent the over-fitting phenomenon, let θ 0 =-θ 1 = θ then:
Supposing that passenger flow of all subway stations in an urban subway network at t +1 moment needs to be predicted, when the characteristics of the subway stations at t moment are constructed, firstly, all subway stations at t-N are selected time Passenger flow to time tAs a proximity timing feature, wherein The passenger flow of the ith subway station at the time t. Then selecting the front N of all subway stations day Passenger flow at time t +1 of dayAs a characteristic of daily periodicity, where T day Representing the time periods of the day. Then selecting the front N of all subway stations week Passenger flow at t +1 weekAs a characteristic of daily periodicity, where T week Representing the time period of the week. Finally, the passenger flow capacity of the subway network at the time t is characterized in that:
the existing research results show that the spatial dependency relationship in the urban subway network changes along with the change of time, so that the hidden layer state H which is learned by the project through the gated cyclic unit network and contains the characteristics of the historical change rule t To calculate a dynamic weight matrix for the subway network. The specific calculation process is as follows:
in the formulaAndfor the weight matrix to be trained, b 1 And b 2 As a bias factor, finally haveWeight matrix of subway network at time tComprises the following steps:
in the formulaSince the dynamic spatial weight is calculated by the Hadamard product method adopted by the formula (16), the original topological structure of the metro network is completely maintained. Finally, the passenger flow of all subway stations in the urban subway network at the time of t +1 can be obtained by utilizing the first-order approximate Cheb graph convolution proposed by Kipf et alIs composed of [7] :
In the formulaTo train optimized parameters.And the passenger flow volume prediction value of each subway station in the urban subway network at the t +1 th moment.
Example of the implementation
The invention utilizes the Seq2GCN method to predict the passenger flow of the urban subway network in a large scale on the AFC data of the Beijing subway. First, 30 days of Beijing subway AFC data in 2016 were divided into a training set, a validation set and a test set, wherein the training set contained 15 days of subway AFC data, and the validation set and the test set each contained 7 days and 8 days of subway AFC data. The other parameter settings are consistent with the parameters in section 2.2.3, and then the Seq2GCN method of the present invention is used to predict subway passenger flow for 10 minutes in the future, wherein the results of the prediction of inbound and outbound passenger flow for the country trade station on line 1, the west gatekeeper station on line 2, the magnet port station on line 7, the skynut station on line 5, and the west two-flag station on line 13 are shown in fig. 2. Wherein the abscissa is time and the ordinate is passenger flow.
As can be seen from the figure, the variation pattern of inbound and outbound passenger traffic differs from site to site. For example, the inbound traffic peak of the country trade station is the off-duty peak from 5 pm to 7 pm, and the outbound traffic peak is actually the on-duty peak from 7 pm to 9 pm. The mode of the Tiantong yuan and the Xidi flag is just opposite to the mode of the national trade station, the peak of the inbound passenger flow of the Tiantong yuan and the Xidi flag is the peak of the office work from 7 to 9 am, and the peak of the outbound passenger flow is the peak of the off-work from 5 to 7 pm. The inbound and outbound passenger flow at the west-straightaway station presents a bimodal pattern, with peak hours occurring during the commuter peak hours from 7 to 9 am and the off-work peak hours from 5 to 7 pm. But the outbound traffic during peak periods is greater than the inbound traffic during peak periods. The variation mode of the incoming and outgoing passenger flow of the five stations on the working day is obviously different from that of the incoming and outgoing passenger flow on the weekend: the amount of incoming and outgoing passenger on weekends is significantly less than that on weekdays.
In addition, it can be seen that the two west flag outbound traffic volume for line 13 suddenly drops from a large value to a small value, even to 0, at some point in time. This is because no subway arrives at the station within the statistical interval (10 minutes), and therefore few passengers, even no passengers, are out of the station during this time period. This case is also a cause of poor prediction accuracy of the partial line, such as the line No. 4 and the line No. 13 in the following table.
TABLE 1 forecast result of passenger flow of each line of Beijing subway
TABLE 2 forecast results of outbound passenger flow of each line of Beijing subway
Line | Number of stations | MAE | RMSE | MAPE |
No. 1 line | 23 | 18.85 | 29.45 | 22.80 |
No. 2 line | 18 | 19.36 | 22.92 | 22.38 |
No. 4 line | 24 | 18.08 | 21.37 | 21.62 |
No. 5 line | 23 | 17.33 | 26.16 | 19.31 |
No. 6 cable | 26 | 16.33 | 26.78 | 28.54 |
No. 7 wire | 19 | 11.17 | 17.27 | 28.54 |
No. 8 line | 18 | 12.40 | 18.63 | 26.56 |
No. 9 line | 13 | 16.15 | 22.02 | 22.44 |
No. 10 wire | 22 | 19.70 | 23.57 | 22.85 |
No. 13 line | 16 | 20.56 | 21.44 | 24.99 |
14 # line | 26 | 11.22 | 18.82 | 24.78 |
No. 15 line | 19 | 10.83 | 17.29 | 25.86 |
In the description herein, particular features, structures, materials, or characteristics may be combined in any suitable manner in any one or more embodiments or examples.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily think of the changes or substitutions within the technical scope of the present invention, and shall cover the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (3)
1. A subway short-time passenger flow prediction method based on a space-time graph convolutional network is characterized by comprising the following steps:
(1) Acquiring historical data of a subway;
(2) The time dependency relationship of the subway network is obtained by learning the subway historical data by using the gate control circulation unit, and a hidden state H which implies the historical passenger flow change characteristics is obtained t ;
(3) And acquiring the dynamic space dependency relationship of the subway network by using the graph convolutional neural network so as to predict the passenger flow at the future time.
2. The method for predicting the short-time passenger flow volume of the subway based on the space-time diagram convolutional network as claimed in claim 1, wherein the input data of the gated cyclic unit network comprises: 1) All subway stations are at t-N time Passenger flow volume to time tAs a proximity timing feature; 2) All subway stations are at the front N day Passenger flow at time t +1 of dayAs a characteristic of daily periodicity, T day Represents the time period encompassed by a day; 3) All subway stations are at the front N week Passenger flow at t +1 weekAs a periodic feature, T week Indicating the time period encompassed by the week.
3. A subway short-time passenger flow predicting method based on a space-time graph convolutional network as claimed in claim 1, wherein said step of obtaining the dynamic spatial dependency relationship of the subway network by using the graph convolutional neural network specifically comprises:
using hidden layer state H t The dynamic weight matrix of the subway network is calculated, and the calculation process is as follows:
passenger flow of all subway stations in urban subway network at the moment of t +1 can be obtained by utilizing first-order approximate Cheb graph convolutionComprises the following steps:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211246533.9A CN115618934A (en) | 2022-10-12 | 2022-10-12 | Subway short-time passenger flow prediction method based on space-time diagram convolutional network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211246533.9A CN115618934A (en) | 2022-10-12 | 2022-10-12 | Subway short-time passenger flow prediction method based on space-time diagram convolutional network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115618934A true CN115618934A (en) | 2023-01-17 |
Family
ID=84862509
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211246533.9A Pending CN115618934A (en) | 2022-10-12 | 2022-10-12 | Subway short-time passenger flow prediction method based on space-time diagram convolutional network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115618934A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116205383A (en) * | 2023-05-05 | 2023-06-02 | 杭州半云科技有限公司 | Static dynamic collaborative graph convolution traffic prediction method based on meta learning |
-
2022
- 2022-10-12 CN CN202211246533.9A patent/CN115618934A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116205383A (en) * | 2023-05-05 | 2023-06-02 | 杭州半云科技有限公司 | Static dynamic collaborative graph convolution traffic prediction method based on meta learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107610464B (en) | A kind of trajectory predictions method based on Gaussian Mixture time series models | |
CN109919358B (en) | Real-time station flow prediction method based on neural network space-time attention mechanism | |
CN109886444B (en) | Short-time traffic passenger flow prediction method, device, equipment and storage medium | |
CN108280551A (en) | A kind of photovoltaic power generation power prediction method using shot and long term memory network | |
CN109754113A (en) | Load forecasting method based on dynamic time warping Yu length time memory | |
CN108985965A (en) | A kind of photovoltaic power interval prediction method of combination neural network and parameter Estimation | |
CN111695731B (en) | Load prediction method, system and equipment based on multi-source data and hybrid neural network | |
CN110599236A (en) | Short-time parking demand prediction method based on GRU model | |
CN111292525A (en) | Traffic flow prediction method based on neural network | |
CN109583565A (en) | Forecasting Flood method based on the long memory network in short-term of attention model | |
CN109492817A (en) | Following berth quantity required Forecasting Approach for Short-term in a kind of closed area | |
WO2022104696A1 (en) | Traffic flow modal fitting method based on quantum random walk | |
CN114626585B (en) | Urban rail transit short-time passenger flow prediction method based on generation countermeasure network | |
CN115618934A (en) | Subway short-time passenger flow prediction method based on space-time diagram convolutional network | |
Unterluggauer et al. | Short‐term load forecasting at electric vehicle charging sites using a multivariate multi‐step long short‐term memory: A case study from Finland | |
CN112054943A (en) | Traffic prediction method for mobile network base station | |
Liu et al. | Short-term forecasting of rail transit passenger flow based on long short-term memory neural network | |
Jia et al. | Deep learning‐based hybrid model for short‐term subway passenger flow prediction using automatic fare collection data | |
CN115081717A (en) | Rail transit passenger flow prediction method integrating attention mechanism and graph neural network | |
Zhang et al. | A hybrid spatiotemporal deep learning model for short-term metro passenger flow prediction | |
CN113205698A (en) | Navigation reminding method based on IGWO-LSTM short-time traffic flow prediction | |
CN112766600A (en) | Urban area crowd flow prediction method and system | |
CN115392554A (en) | Track passenger flow prediction method based on depth map neural network and environment fusion | |
CN113516304A (en) | Space-time joint prediction method and device for regional pollutants based on space-time graph network | |
CN115423189A (en) | Rail transit passenger flow prediction model and method of adaptive graph convolution recurrent neural network combined with attention mechanism |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |