CN115394084B - Urban road network short-time traffic flow prediction method based on NMF-BiLSTM - Google Patents

Urban road network short-time traffic flow prediction method based on NMF-BiLSTM Download PDF

Info

Publication number
CN115394084B
CN115394084B CN202211042155.2A CN202211042155A CN115394084B CN 115394084 B CN115394084 B CN 115394084B CN 202211042155 A CN202211042155 A CN 202211042155A CN 115394084 B CN115394084 B CN 115394084B
Authority
CN
China
Prior art keywords
coefficient matrix
bilstm
road network
traffic flow
time
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.)
Active
Application number
CN202211042155.2A
Other languages
Chinese (zh)
Other versions
CN115394084A (en
Inventor
王永东
曹祥红
石晓艳
袁凯鑫
张征宇
武东辉
吴艳敏
白振鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhengzhou University of Light Industry
Original Assignee
Zhengzhou University of Light Industry
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhengzhou University of Light Industry filed Critical Zhengzhou University of Light Industry
Priority to CN202211042155.2A priority Critical patent/CN115394084B/en
Publication of CN115394084A publication Critical patent/CN115394084A/en
Application granted granted Critical
Publication of CN115394084B publication Critical patent/CN115394084B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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
    • G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
    • 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
    • G06Q50/40—Business processes related to the transportation industry
    • 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
    • G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Chemical & Material Sciences (AREA)
  • Computational Linguistics (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Operations Research (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Development Economics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention provides an NMF-BiLSTM-based urban road network short-time traffic flow prediction method, which comprises the following steps: extracting historical state data of urban road network traffic flow, and acquiring a base matrix and a coefficient matrix of the historical state data based on NMF decomposition; constructing a BiLSTM coefficient matrix prediction model, taking a coefficient matrix of historical state data as input of the BiLSTM coefficient matrix prediction model, and training the BiLSTM coefficient matrix prediction model to obtain a trained BiLSTM model; NMF decomposition is carried out on the real-time urban road network traffic flow sequence data, and a base matrix and a coefficient matrix of the real-time road network traffic flow sequence data are obtained; and taking the real-time coefficient matrix as the input of the BiLSTM model after training, and predicting through the basis matrix of the historical state data to obtain predicted road network traffic flow data. The invention combines the base matrix of the historical data, realizes the traffic flow prediction of the real-time data of the road network, reduces the data processing amount and improves the prediction precision.

Description

Urban road network short-time traffic flow prediction method based on NMF-BiLSTM
Technical Field
The invention belongs to the technical field of urban road network traffic flow data prediction, relates to a method for processing and mathematical modeling of urban road network traffic data, and particularly relates to an NMF-BiLSTM-based urban road network short-time traffic flow prediction method.
Background
Urban road network congestion seriously affects urban operation and sustainable development. The essential problem of traffic jam is imbalance between traffic supply capacity and traffic running demand, so that traffic flow in partial areas is too concentrated and traffic resources are wasted. The prediction of the traffic flow of the urban road network is a precondition for traffic management and control, and is a key for realizing traffic flow system induction and making traffic control strategies. The prediction of the urban road network traffic flow is an important component for realizing intelligent traffic and intelligent cities, can predict the traffic state of the whole road network in the future, and has important effects of relieving traffic jam, effectively utilizing traffic resources and reasonably optimizing traffic infrastructure configuration.
In the existing traffic flow prediction method, the prediction method based on mathematical statistics is convenient to construct, but the nonlinear capturing capacity is not strong, the adaptability is poor, and the prediction precision is not high; the prediction model based on machine learning has better nonlinear capturing capability, but is not suitable for big data; the deep learning prediction model has better data feature capturing capability, but has a certain model overfitting problem and is complex to construct. In addition, huge urban road networks contain massive traffic state data, and most traffic flow prediction researches stay on the road level without considering the overall running state of the road network.
The invention patent with application number 201911219004.8 discloses a method for estimating and predicting short-term traffic running states of an urban road network, which comprises the following steps: (1) Heterogeneous data are obtained and preprocessed, a road section between two signal intersections of a city is taken as a research unit, and a speed field of the research unit is reconstructed by using a GASM algorithm; (2) Constructing a space weight matrix of the urban road network, calculating the space-time correlation among all road segments, and identifying and quantifying the vulnerable road segments by adopting TOPSIS; (3) Taking an average value of speeds according to the reconstructed research unit speed field, and selecting a reasonable fragile road section to construct a space-time feature matrix of the urban road network; (4) And estimating and predicting the traffic state of the whole road network according to the Bi-ConvLSTM. According to the invention, the speed field of the research unit is reconstructed by fusing heterogeneous data, the prediction limitation caused by a single data source is solved, meanwhile, the influence of the traffic speed at the upstream and downstream of the research unit is considered by adopting Bi-ConvLSTM, the space-time characteristics of traffic flow are fully excavated, and the prediction accuracy is further improved. However, the method is difficult to quickly and effectively process and analyze massive urban road network traffic data.
Disclosure of Invention
Aiming at the technical problems of low prediction precision and low operation efficiency of the existing traffic flow prediction method, the invention provides the urban road network short-time traffic flow prediction method based on NMF-BiLSTM, which decomposes mass data in an NMF decomposition mode, realizes the prediction of road network traffic flow by performing BiLSTM modeling on the decomposed data, greatly reduces time complexity and calculation cost, and has the characteristics of high prediction precision and good real-time performance.
In order to achieve the above purpose, the technical scheme of the invention is realized as follows: an NMF-BiLSTM-based urban road network short-time traffic flow prediction method comprises the following steps:
step 1): extracting historical state data of urban road network traffic flow, and acquiring a base matrix and a coefficient matrix of the historical state data based on NMF decomposition;
step 2): constructing a BiLSTM coefficient matrix prediction model, taking a coefficient matrix of historical state data as input of the BiLSTM coefficient matrix prediction model, defining a loss function of the BiLSTM coefficient matrix prediction model, realizing training of the BiLSTM coefficient matrix prediction model to obtain optimal parameters, and taking the optimal parameters as a trained BiLSTM model;
step 3): extracting real-time urban road network traffic flow sequence data, and performing NMF decomposition according to the decomposition structure of the historical state data in the step 1) to obtain a base matrix and a coefficient matrix of the real-time road network traffic flow sequence data;
step 4): taking the coefficient matrix of the real-time road network traffic flow sequence data as the input of the BiLSTM model after training in the step 2), and completing the prediction of the coefficient matrix through the base matrix of the historical state data, namely the predicted road network traffic flow data.
Preferably, the implementation method of NMF decomposition in step 1) is as follows:
A m×n ＝W m×r H r×n (1)
wherein A is m×n Representing historical state data of traffic flow in a road network, wherein m represents the number of road segments in the road network, n represents the first n time periods, n is more than or equal to 1 and less than or equal to T, and T represents the total time period number; w (W) m×r Historical state data A for representing traffic flow in road network m×n The matrix after NMF decomposition, r represents the matrix W m×r The number of the medium base vectors, and r is less than or equal to min (m, n); h r×n Representing a coefficient matrix; w (W) m×r And H r×n Are all non-negative matrices.
Preferably, the BiLSTM coefficient matrix prediction model in the step 2) is expressed as:
i t ＝σ(W i X t +U i S t-1 ) (2)
f t ＝σ(W f X t +U f S t-1 ) (3)
o t ＝σ(W o X t +U o S t-1 ) (4)
wherein X is t Coefficient matrix H representing an input road network at time t after NMF decomposition r×n The Hadamard product is indicated by "; sigma and tanh represent sigmoid and tanh excitation functions, respectively; i.e t 、f t And o t Respectively representing the output of the input gate, the forget gate and the output gate of BiLSTM at the time t, W i And U i A coefficient matrix representing the BiLSTM input gates; w (W) o And U o A coefficient matrix representing the BiLSTM output gates; w (W) f And U f A coefficient matrix representing a forgetting gate;and->Respectively representing the states of the time sequence hidden layers in the forward direction and the reverse direction at the time t; w (W) c And U c Representation->And->Coefficient matrix of (a);/>representing the state of a new hidden layer, S t-1 、S t 、S t+1 And respectively representing the final output of the model at the time t-1, t and t+1.
Preferably, the training method of the BiLSTM coefficient matrix prediction model comprises the following steps:
coefficient matrix H of history state data after NMF decomposition r×n As input to the BiLSTM coefficient matrix prediction model:
H' r×(n+1) ,...,H' r×(n+p) ＝f BiLSTM (H r×(n-(q-1)) ,...,H r×n ) (9)
wherein H' r×(n+1) ,...,H' r×(n+p) A prediction coefficient matrix representing p time periods after predicting n time periods; h r×(n-(q-1)) ,...,H r×n A true coefficient matrix representing q periods before the history n period; f (f) BiLSTM Processing functions representing BiLSTM coefficient matrix prediction models, namely formulas (2) to (8); RMSE represents the root mean square error of the prediction coefficient matrix and the true coefficient matrix as a loss function of the BiLSTM coefficient matrix prediction model.
Preferably, the method for decomposing NMF in the step 3) is as follows:
wherein B is m×v Representing real-time urban road network traffic flow sequence data, wherein m represents the number of road segments in the road network, and v represents the first v time periods;matrix representing NMF decomposition, +.>Representing NMF decompositionA coefficient matrix.
Preferably, the method for inputting the coefficient matrix of the real-time road network traffic flow sequence data in the step 4) into the trained BiLSTM model includes:
wherein f BiLSTM Representing the processing functions of the BiLSTM coefficient matrix prediction model,real-time coefficient matrix representing q periods before v period,/v>A coefficient matrix representing p periods after the predicted v period.
Preferably, the method for predicting the coefficient matrix in the step 4) is as follows: combining the matrix of the history state data after NMF decomposition, the traffic flow prediction of the road network real-time data is realized as follows:
wherein B' m×(v+1) ,...,B' m×(v+p) And the road network traffic flow data of p time periods after the predicted v time period is represented, so that the short-time prediction of the real-time data is realized.
The invention has the beneficial effects that: according to the invention, the existing urban road network historical data is decomposed by an NMF decomposition method to obtain a non-negative base matrix and a coefficient matrix, and the base matrix well reflects the overall characteristics of road network traffic states; the BiLSTM model is trained through the coefficient matrix of the historical road network data, so that the dimension of the original road network data is greatly reduced; and predicting a coefficient matrix of the real-time road network data decomposed by NMF by using the trained BiLSTM model, and finally realizing short-time prediction of the urban road network data by combining a base matrix of the historical data and the predicted real-time coefficient matrix. The invention combines the base matrix of the historical data, realizes the traffic flow prediction of the real-time data of the road network, reduces the data processing amount and improves the prediction precision.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without any inventive effort, are intended to be within the scope of the invention.
As shown in FIG. 1, the urban road network short-time traffic flow prediction method based on NMF-BiLSTM comprises the following steps:
step 1): and extracting historical state data of the urban road network traffic flow, and acquiring a base matrix and a coefficient matrix of the road network traffic flow data based on NMF decomposition.
The historical traffic flow state data of each road section is obtained through the coils laid on the roadside, and the traffic state data of a plurality of road sections in the urban road network jointly form the historical traffic flow state data of the urban road network.
The general expression for NMF decomposition is as follows:
A m×n ＝W m×r H r×n (1)
wherein A is m×n Representing historical state data of traffic flow in a road network, wherein m represents the number of road segments in the road network, n represents the first n time periods, n is more than or equal to 1 and less than or equal to T, and T represents the total time period number; w (W) m×r Representing traffic flow history in road networkState data a m×n The matrix after NMF decomposition, r represents the matrix W m×r The number of the medium base vectors, and r is less than or equal to min (m, n); h r×n Representing a coefficient matrix; w (W) m×r And H r×n Are all non-negative matrices.
Step 2): constructing a BiLSTM coefficient matrix prediction model, and decomposing the road network traffic flow historical state data into a coefficient matrix H r×n As the input of the BiLSTM coefficient matrix prediction model, a loss function of the BiLSTM coefficient matrix prediction model is defined, training of the coefficient matrix is achieved, and the trained BiLSTM model is obtained.
The expression for the BiLSTM coefficient matrix prediction model is generally expressed as follows:
i t ＝σ(W i X t +U i S t-1 ) (2)
f t ＝σ(W f X t +U f S t-1 ) (3)
o t ＝σ(W o X t +U o S t-1 ) (4)
wherein X is t An input road network coefficient matrix H representing t moment after being decomposed by NFM r×n The Hadamard product is indicated by "; sigma and tanh represent sigmoid and tanh excitation functions, respectively; i.e t 、f t And o t Input gates and forget gates respectively representing BiLSTMAnd the output of the output gate at the time t, W i And U i A coefficient matrix representing the BiLSTM input gates; w (W) o And U o A coefficient matrix representing the BiLSTM output gates; w (W) f And U f A coefficient matrix representing a forgetting gate;and->Respectively representing the states of the time sequence hidden layers in the forward direction and the reverse direction at the time t; w (W) c And U c Representation->And->Coefficient matrix of (a); />Representing the state of a new hidden layer, S t-1 、S t 、S t+1 And respectively representing the final output of the model at the time t-1, t and t+1. The excitation function and hadamard product are also known, the other parameters being training results.
Coefficient matrix H for decomposing historical state data of road network traffic flow r×n As the input of BiLSTM coefficient matrix prediction model, coefficient matrix H is realized r×n The general expression is as follows:
H' r×(n+1) ,...,H' r×(n+p) ＝f BiLSTM (H r×(n-(q-1)) ,...,H r×n ) (9)
wherein H' r×(n+1) ,...,H' r×(n+p) A prediction coefficient matrix representing p time periods after predicting n time periods; h r×(n-(q-1)) ,...,H r×n A true coefficient matrix representing q periods before the history n period; f (f) BiLSTM Representing BiLSTM coefficient momentsThe general expression of the processing function of the matrix prediction model is shown in formulas (2) to (8). The training is performed by using historical data, and the input and output are known, so that the parameters can be obtained through training, and the trained parameters are applied to the subsequent real-time data. RMSE represents the root mean square error of the prediction coefficient matrix and the true coefficient matrix as a loss function of the BiLSTM coefficient matrix prediction model. Equation (10) is to calculate the real and predicted gap in the training process, and the RMSE is made smaller and smaller by continuously training the BiLSTM model of the water supply (9).
Step 3): extracting real-time urban road network traffic flow sequence data, performing NMF decomposition according to the historical state data decomposition structure in the step 1), and obtaining a base matrix and a coefficient matrix of the real-time road network traffic flow sequence data, wherein the general steps are as follows:
wherein B is m×v Representing real-time urban road network traffic flow sequence data, wherein m represents the number of road segments in the road network, and v represents the first v time periods;matrix representing NMF decomposition, +.>A coefficient matrix representing NMF decomposition.
Step 4): taking the NMF decomposed coefficient matrix of the real-time road network traffic flow sequence data as the input of the BiLSTM model trained in the step 2), and completing the prediction of the coefficient matrix through the NMF decomposed base matrix of the historical traffic flow state data, namely the predicted road network traffic flow data.
The general procedure is as follows:
wherein f BiLSTM Representing the processing functions of the BiLSTM coefficient matrix prediction model,real-time coefficient matrix representing q periods before v period,/v>A coefficient matrix representing p periods after the predicted v period.
The method combines the base matrix decomposed by the historical data NMF to realize the traffic flow prediction of the real-time data of the road network, and basically comprises the following steps:
wherein B' m×(v+1) ,...,B' m×(v+p) And the road network traffic flow data of p time periods after the predicted v time period is represented, so that the short-time prediction of the real-time data is realized.
Specific examples: an NMF-BiLSTM-based urban road network short-time traffic flow prediction method comprises the following steps:
1) Extracting historical state data of urban road network traffic flow, and obtaining base matrix and coefficient matrix of road network traffic flow data based on NMF decomposition
In the test, 6 road segments in the Beijing two-ring are used as a research road network, road network speed data (sampling interval is 2 minutes) actually measured at the same test point in 1 to 14 days (total 14 days) of 6 th month in 2011 are used as a sample sequence, and the road segment information is shown in table 1.
TABLE 1 road segment information
Extracting road traffic history data of 6 road sections from the road traffic characteristic reference sequence, wherein the acquired data of each road section is 720 each day, and converting the acquired data into an m multiplied by n matrix, namely, m is 6, n is 720, and marking as: a is that mⅹn Then
A m×n ＝W m×r H r×n (1)
Wherein A is m×n Representing road networkM represents the number of road segments in the road network, n (n is more than or equal to 1 and less than or equal to T) represents the first n time periods, and T represents the total time period number; w (W) m×r Data A representing historical traffic state of road network m×n NMF decomposed basis matrix, wherein r represents W m×r The number of the medium base vectors, and r is less than or equal to min (m, n); h r×n Representing a coefficient matrix; w (W) m×r And H r×n No element of (2) is less than 0.r is selected in the training process, and after r is obtained, the structures of the base matrix and the coefficient matrix can also be obtained.
2) Constructing a BiLSTM coefficient matrix prediction model, and decomposing historical road network traffic flow data into a coefficient matrix H r×n As an input of a BiLSTM coefficient matrix prediction model, a loss function of the prediction model is defined, training of the coefficient matrix is achieved, and a BiLSTM expression is generally expressed as follows:
i t ＝σ(W i X t +U i S t-1 ) (2)
f t ＝σ(W f X t +U f S t-1 ) (3)
o t ＝σ(W o X t +U o S t-1 ) (4)
wherein, as indicated by Hadamard product, i t ，f t And o t Respectively represent the input gate and the input gate of BiLSTMForgetting gate and output gate output at t time, W i ，W f And W is o Coefficient matrixes respectively representing the corresponding input gate, the forget gate and the output gate;and->Respectively representing the state of the forward and reverse time series hidden layer at time t,/and>representing a new hidden layer state, S t Representing the output of the final BiLSTM.
Coefficient matrix H for decomposing historical road network traffic flow data r×n As an input to the BiLSTM model, training of the coefficient matrix is achieved, with the general expression as follows:
H' r×(n+1) ,...,H' r×(n+p) ＝f BiLSTM (H r×(n-(q-1)) ,...,H r×n ) (9)
wherein H' r×(n+1) ,...,H' r×(n+p) A coefficient matrix representing p periods after the prediction of n periods; h r×(n-(q-1)) ,...,H r×n A coefficient matrix representing q periods preceding the history n period; f (f) BiLSTM The BiLSTM model is represented, and general expressions are shown in formulas (2) to (8). RMSE represents the root mean square error of the prediction coefficient matrix and the true coefficient matrix.
3) Extracting real-time urban road network traffic flow sequence data, performing NMF decomposition according to a historical data decomposition structure, and obtaining a base matrix and a coefficient matrix of the real-time data, wherein the general steps are as follows:
wherein B is m×v Representing real-time road network trafficStream sequence data, m represents the number of road segments in the road network, v represents the first v time periods;matrix representing NMF decomposition, +.>A coefficient matrix representing NMF decomposition.
4) Taking a coefficient matrix of real-time road network data decomposition as input of a BiLSTM model to complete prediction of the coefficient matrix, wherein the general steps are as follows:
wherein f BiLSTM Representing the BiLSTM model,real-time coefficient matrix representing q periods before v period,/v>A coefficient matrix representing p periods after the predicted v period.
The traffic flow prediction of the road network real-time data is realized by combining the base matrix of the historical data decomposition, and the basic steps are as follows:
wherein B' m×(v+1) ,...,B' m×(v+p) And the road network traffic flow data of p time periods after the prediction v time period is represented, so that the short-time prediction of the real-time data is realized.
The prediction effect of the invention is mainly determined by m, n, r, p, q and other parameters, namely, for different parameters (m, n, r, p and q), the influence of each parameter on the algorithm precision is independently analyzed, and the optimization of the algorithm cannot be ensured, so that the influence of all the parameters on the prediction result of the road network traffic flow should be considered simultaneously when the analysis is performed.
The average absolute percentage error (MAPE) of the real-time data prediction is introduced as a measurement index of the road network traffic flow prediction precision, and the calculation formulas are respectively shown as follows:
wherein B is t As a measurement value, B' t Is a predicted value. MAPE is the mean absolute percentage error of the predicted data.
That is, for different parameters (m, n, r, p, q), there is a mean absolute percentage error MAPE corresponding thereto, so the following equation exists:
MAPE＝θ(m,n,r,p,q) (15)
that is, the parameters (m, n, r, p, q) and MAPE have a certain distribution relation θ, and the parameter (m, n, r, p, q) corresponding to the minimum MAPE is found, namely the optimal parameter setting process. The following model can be obtained:
Minθ(m,n,r,p,q)
the final value of (m, n, r, p, q) can be determined through training of road traffic history data.
Experimental results: the data of 1 to 12 days are used as history data, and the data of 13 and 14 days are used as real-time data. Acquiring optimal parameters (m, n, r, p, q) based on road network traffic history data; the method is used for extracting road network traffic real-time data and realizing the real-time prediction of the road network traffic data based on the NMF-BiLSTM road network short-time traffic flow prediction method.
And selecting Root Mean Square Error (RMSE), mean Absolute Error (MAE) and Mean Absolute Percent Error (MAPE) as indexes of road network traffic flow prediction precision. The calculation formulas are as follows:
wherein B is t And B' t And respectively representing the real value and the predicted value of the real-time road network data in the t period.
Statistics of road network speed prediction error results at 13 and 14 days of 6 months of 2011 are shown in table 2.
TABLE 2 prediction error result statistics
Date of day RMSE MAE MAPE(％)
6 month 13 day 11.46 7.88 9.95
6 months and 14 days 9.47 8.45 9.74
As can be seen from the results in Table 2, the average absolute error MAE is controlled below 9, and the average absolute percentage error MAPE is controlled below 10%, which indicates that the error of the invention can be controlled below 10%, and the method is an effective urban road network short-time traffic flow prediction method.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.

Claims (5)

1. An NMF-BiLSTM-based urban road network short-time traffic flow prediction method is characterized by comprising the following steps:
step 1): extracting historical state data of urban road network traffic flow, and acquiring a base matrix and a coefficient matrix of the historical state data based on NMF decomposition;
step 2): constructing a BiLSTM coefficient matrix prediction model, taking a coefficient matrix of historical state data as input of the BiLSTM coefficient matrix prediction model, defining a loss function of the BiLSTM coefficient matrix prediction model, realizing training of the BiLSTM coefficient matrix prediction model to obtain optimal parameters, and taking the optimal parameters as a trained BiLSTM model;
the BiLSTM coefficient matrix prediction model in the step 2) is expressed as follows:
i t ＝σ(W i X t +U i S t-1 ) (1)
f t ＝σ(W f X t +U f S t-1 ) (2)
o t ＝σ(W o X t +U o S t-1 ) (3)
wherein X is t Coefficient matrix H representing an input road network at time t after NMF decomposition r×n The Hadamard product is indicated by "; sigma and tanh represent sigmoid and tanh excitation functions, respectively; i.e t 、f t And o t Respectively representing the output of the input gate, the forget gate and the output gate of BiLSTM at the time t, W i And U i A coefficient matrix representing the BiLSTM input gates; w (W) o And U o A coefficient matrix representing the BiLSTM output gates; w (W) f And U f A coefficient matrix representing a forgetting gate;and->Respectively representing the states of the time sequence hidden layers in the forward direction and the reverse direction at the time t; w (W) c And U c Representation->And->Coefficient matrix of (a); />Representing the state of a new hidden layer, S t-1 、S t 、S t+1 Respectively representing the final output of the model at the times t-1, t and t+1;
the training method of the BiLSTM coefficient matrix prediction model comprises the following steps:
coefficient matrix H of history state data after NMF decomposition r×n As BiLInput of STM coefficient matrix prediction model:
H' r×(n+1) ,...,H' r×(n+p) ＝f BiLSTM (H r×(n-(q-1)) ,...,H r×n ) (8)
wherein H' r×(n+1) ,...,H' r×(n+p) A prediction coefficient matrix representing p time periods after predicting n time periods; h r×(n-(q-1)) ,...,H r×n A true coefficient matrix representing q periods before the history n period; f (f) BiLSTM Processing functions representing BiLSTM coefficient matrix prediction models, namely formulas (1) - (7); RMSE represents the root mean square error of the prediction coefficient matrix and the true coefficient matrix as a loss function of the BiLSTM coefficient matrix prediction model;
step 3): extracting real-time urban road network traffic flow sequence data, and performing NMF decomposition according to the decomposition structure of the historical state data in the step 1) to obtain a base matrix and a coefficient matrix of the real-time road network traffic flow sequence data;
step 4): taking the coefficient matrix of the real-time road network traffic flow sequence data as the input of the BiLSTM model after training in the step 2), and completing the prediction of the coefficient matrix through the base matrix of the historical state data, namely the predicted road network traffic flow data.
2. The method for predicting short-term traffic flow of urban road network based on NMF-BiLSTM according to claim 1, wherein the implementation method of NMF decomposition in step 1) is as follows:
A m×n ＝W m×r H r×n (10)
wherein A is m×n Representing historical state data of traffic flow in a road network, wherein m represents the number of road segments in the road network, n represents the first n time periods, n is more than or equal to 1 and less than or equal to T, and T represents the total time period number; w (W) m×r Historical state data A for representing traffic flow in road network m×n NMF decomposed base matrix, r representsBase matrix W m×r The number of the medium base vectors, and r is less than or equal to min (m, n); h r×n Representing a coefficient matrix; w (W) m×r And H r×n Are all non-negative matrices.
3. The method for predicting short-term traffic flow of urban road network based on NMF-BiLSTM according to claim 1 or 2, wherein the method for NMF decomposition in step 3) is as follows:
wherein B is m×v Representing real-time urban road network traffic flow sequence data, wherein m represents the number of road segments in the road network, and v represents the first v time periods;matrix representing NMF decomposition, +.>A coefficient matrix representing NMF decomposition.
4. The method for predicting short-term traffic flow of urban road network based on NMF-BiLSTM according to claim 3, wherein the method for inputting the coefficient matrix of the real-time road network traffic flow sequence data in step 4) into the trained BiLSTM model is as follows:
wherein f BiLSTM Representing the processing functions of the BiLSTM coefficient matrix prediction model,real-time coefficient matrix representing q periods before v period,/v>A coefficient matrix representing p periods after the predicted v period.
5. The method for predicting short-term traffic flow of urban road network based on NMF-BiLSTM according to claim 4, wherein the method for predicting the coefficient matrix in step 4) is as follows: combining the matrix of the history state data after NMF decomposition, the traffic flow prediction of the road network real-time data is realized as follows:
wherein B' m×(v+1) ,...,B' m×(v+p) And the road network traffic flow data of p time periods after the predicted v time period is represented, so that the short-time prediction of the real-time data is realized.
CN202211042155.2A 2022-08-29 2022-08-29 Urban road network short-time traffic flow prediction method based on NMF-BiLSTM Active CN115394084B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211042155.2A CN115394084B (en) 2022-08-29 2022-08-29 Urban road network short-time traffic flow prediction method based on NMF-BiLSTM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211042155.2A CN115394084B (en) 2022-08-29 2022-08-29 Urban road network short-time traffic flow prediction method based on NMF-BiLSTM

Publications (2)

Publication Number Publication Date
CN115394084A CN115394084A (en) 2022-11-25
CN115394084B true CN115394084B (en) 2023-07-25

Family

ID=84121951

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211042155.2A Active CN115394084B (en) 2022-08-29 2022-08-29 Urban road network short-time traffic flow prediction method based on NMF-BiLSTM

Country Status (1)

Country Link
CN (1) CN115394084B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106781489A (en) * 2016-12-29 2017-05-31 北京航空航天大学 A kind of road network trend prediction method based on recurrent neural network
CN110070713A (en) * 2019-04-15 2019-07-30 浙江工业大学 A kind of traffic flow forecasting method based on two-way nested-grid ocean LSTM neural network
CN110472045A (en) * 2019-07-11 2019-11-19 中山大学 A kind of short text falseness Question Classification prediction technique and device based on document insertion
CN110490365A (en) * 2019-07-12 2019-11-22 四川大学 A method of based on the pre- survey grid of multisource data fusion about vehicle order volume
CN111865932A (en) * 2020-06-30 2020-10-30 哈尔滨工程大学 Intrusion detection method based on context correlation attention mechanism and simplified LSTM network
CN112035745A (en) * 2020-09-01 2020-12-04 重庆大学 Recommendation algorithm based on counterstudy and bidirectional long-short term memory network

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11423775B2 (en) * 2019-07-18 2022-08-23 International Business Machines Corporation Predictive route congestion management

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106781489A (en) * 2016-12-29 2017-05-31 北京航空航天大学 A kind of road network trend prediction method based on recurrent neural network
CN110070713A (en) * 2019-04-15 2019-07-30 浙江工业大学 A kind of traffic flow forecasting method based on two-way nested-grid ocean LSTM neural network
CN110472045A (en) * 2019-07-11 2019-11-19 中山大学 A kind of short text falseness Question Classification prediction technique and device based on document insertion
CN110490365A (en) * 2019-07-12 2019-11-22 四川大学 A method of based on the pre- survey grid of multisource data fusion about vehicle order volume
CN111865932A (en) * 2020-06-30 2020-10-30 哈尔滨工程大学 Intrusion detection method based on context correlation attention mechanism and simplified LSTM network
CN112035745A (en) * 2020-09-01 2020-12-04 重庆大学 Recommendation algorithm based on counterstudy and bidirectional long-short term memory network

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
Learning heterogeneous traffic patterns for travel time prediction of bus journeys;Peilan He;《Information Sciences》;第512卷;1394-1406 *
Real-time spatiotemporal prediction and imputation of traffic status based on LSTM and Graph Laplacian regularized matrix factorization;Jin-Ming Yang;《Transportation Research Part C: Emerging Technologies》;第129卷;1-16 *
Single-channel speech separation based on deep clustering with local optimization;Taotao Fu;《2017 3rd International Conference on Frontiers of Signal Processing (ICFSP)》;44-49 *
基于深度学习的城市时空序列预测模型及应用研究;杜圣东;《中国博士学位论文全文数据库工程科技Ⅰ辑》(第6期);B027-90 *
基于融合神经网络的短期交通流预测研究;李杰;《兰州交通大学学报》;第41卷(第3期);60-67 *
基于非负矩阵分解和长短时记忆网络的单通道语音分离;崔建峰;《科学技术与工程》;第19卷(第12期);206-210 *

Also Published As

Publication number Publication date
CN115394084A (en) 2022-11-25

Similar Documents

Publication Publication Date Title
CN109285346B (en) Urban road network traffic state prediction method based on key road sections
CN111161535B (en) Attention mechanism-based graph neural network traffic flow prediction method and system
CN111210633B (en) Short-term traffic flow prediction method based on deep learning
CN109034448B (en) Trajectory prediction method based on vehicle trajectory semantic analysis and deep belief network
CN108009674A (en) Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks
CN114330671A (en) Traffic flow prediction method based on Transformer space-time diagram convolution network
CN113554466B (en) Short-term electricity consumption prediction model construction method, prediction method and device
CN112910711A (en) Wireless service flow prediction method, device and medium based on self-attention convolutional network
Zhao et al. Short term traffic flow prediction of expressway service area based on STL-OMS
CN107704970A (en) A kind of Demand-side load forecasting method based on Spark
CN114596700B (en) Real-time traffic estimation method for expressway section based on portal data
CN112396234A (en) User side load probability prediction method based on time domain convolutional neural network
CN115204477A (en) Bicycle flow prediction method of context awareness graph recursive network
CN101834753A (en) Network flow prediction method based on covariant orthogonality and neural network
CN115206092A (en) Traffic prediction method of BiLSTM and LightGBM model based on attention mechanism
CN116681176A (en) Traffic flow prediction method based on clustering and heterogeneous graph neural network
CN110807508A (en) Bus peak load prediction method considering complex meteorological influence
CN116913098B (en) Short-time traffic flow prediction method integrating air quality and vehicle flow data
Khadr et al. GA-based implicit stochastic optimization and RNN-based simulation for deriving multi-objective reservoir hedging rules
CN115394084B (en) Urban road network short-time traffic flow prediction method based on NMF-BiLSTM
Zhang et al. Irregular traffic time series forecasting based on asynchronous spatio-temporal graph convolutional network
Jiao et al. Multi-step traffic flow prediction method based on the Conv1D+ LSTM
CN116993391A (en) Site type shared bicycle system use demand prediction method
CN116578858A (en) Air compressor fault prediction and health degree evaluation method and system based on graphic neural network
CN112581177B (en) Marketing prediction method combining automatic feature engineering and residual neural network

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
GR01 Patent grant
GR01 Patent grant
CB03 Change of inventor or designer information
CB03 Change of inventor or designer information

Inventor after: Wang Yongdong

Inventor after: Wu Donghui

Inventor after: Cao Xianghong

Inventor after: Shi Xiaoyan

Inventor after: Yuan Kaixin

Inventor after: Zhang Zhengyu

Inventor after: Wu Yanmin

Inventor after: Bai Zhenpeng

Inventor before: Wang Yongdong

Inventor before: Cao Xianghong

Inventor before: Shi Xiaoyan

Inventor before: Yuan Kaixin

Inventor before: Zhang Zhengyu

Inventor before: Wu Donghui

Inventor before: Wu Yanmin

Inventor before: Bai Zhenpeng