CN106781489B - A kind of road network trend prediction method based on recurrent neural network - Google Patents
A kind of road network trend prediction method based on recurrent neural network Download PDFInfo
- Publication number
- CN106781489B CN106781489B CN201611244476.5A CN201611244476A CN106781489B CN 106781489 B CN106781489 B CN 106781489B CN 201611244476 A CN201611244476 A CN 201611244476A CN 106781489 B CN106781489 B CN 106781489B
- Authority
- CN
- China
- Prior art keywords
- road network
- layer
- output
- period
- section
- 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.)
- Expired - Fee Related
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
- G08G1/0133—Traffic data processing for classifying traffic situation
Landscapes
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of road network trend prediction method based on recurrent neural network, comprising the following steps: step 1 establishes sample set;Step 2, recurrent neural network modeling.Step 3, subsequent time road network status predication.The present invention holds road network state evolution rule from the angle of macroscopic view, the timing rule of road network state change is fully considered using recurrent neural network algorithm, to preferably make prediction.
Description
Technical field
The present invention relates to public traffic information processing technology field, specifically a kind of road based on recurrent neural network
Net state prediction technique.
Background technique
As urbanization process is accelerated, traffic jam issue becomes increasingly conspicuous, especially big city, and traffic jam issue is more
Sternness has seriously affected the daily trip of people.Urban road day hastens towards saturation, and car ownership rises year by year, this confession
Unbalanced relationship is needed to aggravate traffic congestion, and traffic congestion prediction is the important channel for alleviating traffic congestion.
In existing patent, there are some methods for traffic status prediction, the method for comparing mainstream includes
Kalman filter model, time series models, neural network model, Partial Linear Models etc..Traffic behavior variation has non-thread
Property the characteristics of, can also exist because of some emergency events uncertain, Kalman filter is applicable in as a kind of linear filter
Property is limited, and there are hysteresis.Time series models not only need a large amount of historical data, and to traffic behavior variation
Chronesthesy is poor, it is difficult to which solution of emergent event, neural network is very sensitive to parameter initialization, needs repeatedly to predict to seek average
Value, it is computationally intensive, and there are locally optimal solutions, are easy to appear over-fitting, parameter transplantability is poor.Partial Linear Models are difficult to table
Up to the uncertainty, complexity and dynamic characteristic etc. of traffic behavior.
In addition, existing patent is the traffic status prediction based on section level mostly, it is difficult to hold road from macroscopic perspective
Net state development law, it is difficult to hold complexity, the uncertainty of traffic behavior.And the prediction based on section level is to traffic
Equipment requirement is high, and the shortage of data put individually will cause large effect to precision of prediction, and will based on the prediction of road network level
Effectively solve the problems, such as this because from macroscopic perspective, the missings of individual point datas on entire road network status predication influence compared with
It is small.
Summary of the invention
The present invention is to solve the above the deficiencies in the prior art, provides a kind of road network status predication based on recurrent neural network
Method, this method can fully consider the timing of road network state change.Based on the status predication of road network level, traveler is come
It says, can preferably plan trip route, greatly improve out line efficiency.For manager, road network is held from macroscopic perspective
State evolution trend can preferably analyze road grid traffic situation, plan transportation network, realize traffic optimization control.
To solve the above-mentioned problems, technical solution provided by the invention includes:
A kind of road network trend prediction method based on recurrent neural network, the described method comprises the following steps: step;One,
Establish sample set;A road network is chosen, road network is divided into k section, and each section is numbered, be denoted as (1,2,3 ...,
K), it was divided into the j the same period of each time span for 24 hours one day, calculates each section being averaged in each period
Speed;After the average speed that each section has been calculated, the road network state of the period, i.e. V are indicated with a state vectorj=
[v1,j,v2,j,…,vk,j];The road network state for considering first three period predicts the road network state of next period,
Therefore, single sample is [(Vj-2,Vj-1,Vj),(Vj+1)] its expanded form are as follows:
Entire sample set is the set of all period samples;
Step 2: recurrent neural network models;Firstly, determining input/output variable;Input variable is three one-dimensional states
Vector, difference j-2, j-1, the road network state vector of j period, output variable one is one-dimensional state vector, i.e., when next
Between section road network state, the dimension of each state vector is k, i.e., the section number in road network;Secondly, determining training set and survey
Examination collection, is divided into training set and test set by predetermined ratio according to the sample set in step 1;Finally, carrying out recurrent neural network mould
Shape parameter calibration;The recurrent neural network includes input layer, hidden layer and output layer.Learn to input by training data
Layer and weight matrix and bias vector between hidden layer, hidden layer and input layer, wherein each memory list of the hidden layer
For member all there are three input and two outputs, input content includes x (t), h (t-1), c (t-1), and output includes h (t), c (t).He
Between relationship be to be controlled by three doors, be input gate respectively, forget door and out gate, have in t moment:
it=sigmoid (whiht-1+wxixt+bi)
ft=sigmoid (whfht-1+whfxt+bf)
ct=ft·ct-1+it·tanh(whcht-1+wxcxt+bc)
ot=sigmold (whoht-1+whxxt+wcoct+bo)
ht=ot·tanh(ct)
Wherein it,ft,otRespectively for the output of input gate, forgetting door and out gate, w·,b·Be respectively coefficient matrix and partially
Vector is set, " " is point multiplication operation, and sigmoid and tanh are activation primitive, ctRepresent the defeated of the memory unit of t moment hidden layer
Out, ht-1It is the output in t-1 moment hidden layer, htIt is the output in t moment hidden layer;
Weight matrix W between hidden layer is [k, a z] matrix, and z is to hide layer unit number, and k is the quantity in section;
The weight matrix U connected between input layer and hidden layer is [z, a z] matrix;The output square between output layer is hidden in connection
Battle array V is [z, a k] matrix;The relationship of hidden layer and output layer are as follows:
Yj+1=HV+by
Wherein, H=[h1,h2,……hz], hiIt is the output valve of the hiding layer unit at j moment, the by is that connection is hidden
The bias function Y of layer and output layerj+1=[y1,j+1,y2,j+1,…,yk,j+1];
Loss function is established after obtaining output layer, loss function is for measuring the output valve of output layer and the difference of true value
It is different, using loss function to parameters derivation, calculate its gradient.Using batch ladder when the training of the recurrent neural network
Descent method learning model parameter is spent, model parameter includes all weight matrix and bias vector;
Step 3: subsequent time road network status predication;The input variable of test data is input in step 2 and has been trained
In good model, output vector is obtained, then the vector is exactly the road network state for the next period predicted.
Advantages of the present invention:
(1) biggest advantage of the present invention is to provide a kind of road network trend prediction method, holds road network shape from the angle of macroscopic view
State development law.
(2) present invention fully considers the timing rule of road network state change using recurrent neural network algorithm, thus more preferably
It makes prediction on ground.
Detailed description of the invention
Fig. 1 is flow chart of the invention;
Fig. 2 is LSTM structure chart;
Fig. 3 is recurrent neural networks model figure used in the present invention;
Fig. 4 is that recurrent neural network and BP neural network predict comparison diagram.
Specific embodiment
Below in conjunction with drawings and examples, the present invention is described in further detail, so that those skilled in the art join
Book text can be implemented accordingly as directed.
The present invention provides a kind of road network trend prediction method based on recurrent neural network, process as shown in Figure 1, include with
Lower step:
Step 1 establishes sample set.
A certain road network is chosen, road network is divided into each small section, it is assumed that is divided into k section, and each section is carried out
Number, is denoted as (1,2,3 ..., k), will be divided within 24 hours one day each time span the same period and (for example makees for every 2 minutes
For a period), each section is calculated in the average speed of each period.The calculation method of average speed is: a certain
In period, in the mean value for all average vehicle speeds that certain a road section passes through, as shown in formula (1), wherein n was represented in the time
The number of vehicles that passes through in the section of section, m represent m-th of section in road network, and m ∈ (1,2, k), j is to compile the period
Number, the length in the behalf section, Δ t represents the length of period,Vehicle i is represented in the average speed of Δ t.
If certain a road section passes through in Δ t without vehicle, current time is being substituted with the average speed of a period
The average speed of section, i.e.,
vm,j=vm,j-1。
After the average speed that each section has been calculated, the road network state of the period is indicated with a state vector, i.e.,
Vj=[v1,j,v2,j,…,vk,j]。
The road network state that first three period is considered in the specific embodiment of the invention, predicts the road network of next period
State,
Therefore single sample of the invention is shaped like [(Vj-2,Vj-1,Vj),(Vj+1)]
Its expanded form are as follows:
Entire sample set is the set of all period samples.
Step 2, recurrent neural network modeling.
Firstly, determining input/output variable.Input variable is three one-dimensional state vectors, respectively j-2, j-1, the j period
Road network state vector, output variable one is one-dimensional state vector, i.e., the road network state of next period, each state to
The dimension of amount is k, i.e., the section number in road network.
Secondly, determining training set and test set, training set and survey are divided into according to the sample set in step 1 by a certain percentage
Examination collection.
Finally, recurrent neural networks model parameter calibration.The recurrent neural network include input layer, hidden layer and
Output layer.The step needs to learn the weight matrix between input layer and hidden layer, hidden layer and input layer by training data
And bias vector, the present invention is in order to fully consider the temporal characteristics of road network state evolution, using LSTM (Long Short-Term
Memory, long short-term memory) for unit as layer unit is hidden, structure is as shown in Figure 2.
Each memory unit includes x (t), h (t-1), c (t-1), output there are three input, two outputs, input content
Including h (t), c (t).Relationship between them is controlled by three doors, is input gate (input gate) respectively, is lost
Forget door (forget gate), out gate (output gate), have in t moment:
it=sigmoid (whiht-1+wxixt+bi) (3)
ft=sigmoid (whfht-1+whfxt+bf) (4)
ct=ft·ct-1+it·tanh(whcht-1+wxcxt+bc) (5)
ot=sigmoid (whoht-1+whxxt+wcoct+bo) (6)
ht=ot·tanh(ct) (7)
Wherein it,ft,otRespectively for the output of input gate, forget gate, output gate, w·,b·It is respectively
Coefficient matrix and bias vector, " " are point multiplication operations, and sigmoid and tanh are activation primitive, ctRepresent t moment LSTM memory
The output of cell unit, ht-1It is the output in t-1 moment LSTM unit, htIt is the output in t moment LSTM unit.
Model structure of the invention is as shown in figure 3, wherein Yj+1It is prediction result, i.e. the road network state of subsequent time period.
U is the weight matrix connected between input layer and hidden layer, and W is the weight matrix connected between hidden layer, V be connection hide with
Output layer between output layer.And the weight between different sides is shared.W is [k, a h] matrix, and h is hidden layer list
First number.U is [h, a h] matrix, and V is [h, a k] matrix.Shown in the relationship of hidden layer and output layer such as formula (8):
Yj+1=HV+by (8)
Wherein,
H=[h1,h2,…,hh], hiIt is the output valve of the LSTM unit at j moment, Yj+1=[y1,j+1,y2,j+1,…,
yk,j+1]。
It obtains needing to establish loss function after output layer, using under batch gradient when the training of the recurrent neural network
Calligraphy learning model parameter drops.
Model parameter calibration of the invention is partially completed by above-mentioned.
Step 3, subsequent time road network status predication.The input variable of test data is input in step 2 and has been trained
In good model, output vector is obtained, then the vector is exactly the road network state for the next period predicted.Due in reality
In the application of border, road network state limited reliability is described with the section velocity amplitude acquired in us, therefore we turn velocity amplitude
Congestion level is turned to, specific targets are as shown in table 1 below:
1 congestion level index table of table
Velocity amplitude | Congestion index |
Less than or equal to 20km/h | Congestion |
20~40km/h | Jogging |
More than or equal to 40km/h | It is unimpeded |
Embodiment
It should be noted that data used in the present invention, by a certain road network in Beijing that certain company provides, data include 9
A field, as shown in table 2, section data be update within every 2 minutes it is primary, wherein thering is the direct relation data field to include with the present invention
Time, section number, three fields of speed, time span 3 months, section number was 278.
Table 2:
Realization route of the invention includes the following steps:
Step 1 establishes sample set.
Above-mentioned data are indicated in the road network state of certain time period with a state vector, time segment length is 2 minutes, to
Element value in amount is the velocity amplitude in each section, and element value will be by section number sorting, i.e. Vj=[v1,j,v2,j,…,vk,j],
J indicates j-th of period.As shown below, just there is a state vector within every in this way 2 minutes.
The present invention considers the road network state of first three period, predicts the road network state of next period, therefore this hair
Bright single sample is shaped like [(Vj-2,Vj-1,Vj),(Vj+1)] its expanded form are as follows:
What is indicated is with j-2, j-1, the road network state of+1 period of road network status predication jth of j period.Entire sample
Collection is the set of all period samples.
Step 2, recurrent neural network modeling.
Firstly, determining input/output variable.Input variable is three one-dimensional state vectors, respectively j-2, j-1, the j period
Road network state vector, output variable one is one-dimensional state vector, i.e., the road network state of next period, each state to
The dimension of amount is 278, i.e., the section number in road network.
Secondly, determining training set and test set, training set and test are divided into 2:1 ratio according to the sample set in step 1
Collection.
Finally, recurrent neural networks model parameter calibration.The recurrent neural network include input layer, hidden layer and
Output layer.The step needs to learn the weight matrix between input layer and hidden layer, hidden layer and input layer by training data
And bias vector, the present invention is in order to fully consider the temporal characteristics of road network state evolution, using LSTM (Long Short-Term
Memory, long short-term memory) the hiding layer unit of unit conduct.Structure is as shown in Figure 2.Input that there are three each memory units,
Two outputs, input content include x (t), h (t-1), c (t-1), and output includes h (t), c (t).Relationship between them is logical
It crosses three doors to be controlled, is input gate (input gate) respectively, forgets door (forget gate), out gate (output
Gate), have in t moment:
it=sigmoid (whiht-1+wxixt+bi) (3)
ft=sigmoid (whfht-1+whfxt+bf) (4)
ct=ft·ct-1+it·tanh(whcht-1+wxcxt+bc) (5)
ot=sigmoid (whoht-1+whxxt+wcoct+bo) (6)
ht=ot·tanh(ct) (7)
Wherein it,ft,otRespectively for the output of input gate, forget gate, output gate, w·,b·It is respectively
Coefficient matrix and bias vector, " " are point multiplication operations, and sigmoid and tanh are activation primitive, ctRepresent t moment LSTM memory
The output of cell unit, ht-1It is the output in t-1 moment LSTM unit, htIt is the output in t moment LSTM unit.
Model structure of the invention is as shown in figure 3, wherein Yj+1It is prediction result, i.e. the road network state of subsequent time period.
U is the weight matrix connected between input layer and hidden layer, and W is the weight matrix connected between hidden layer, V be connection hide with
Output layer between output layer.And the weight between different sides is shared.W is [278, a 500] matrix, and h is hiding
Layer unit number.U is [500, a 500] matrix, and V is [500, a 278] matrix.The relationship of hidden layer and output layer such as formula
(8) shown in:
Yj+1=HV+by (8)
Wherein, H=[h1,h2,…,hh], hiIt is the output valve of the LSTM unit at j moment, Yj+1=[y1,j+1,y2,j+1,…,
yk,j+1]。
Need to establish loss function after obtaining output layer, the recurrent neural network uses BP (back
Propagation) algorithm learns, so that it is determined that the weight of model and biasing, in order to accelerate training speed, the present invention is using batch
Gradient descent algorithm is iterated update to parameter.Parameter initialization of the invention is all made of the equal distribution random numbers of (0,1), defeated
Enter after variable normalizes in being input to model.
Model parameter calibration of the invention is partially completed by above-mentioned.
Step 3, subsequent time road network status predication.The input variable of test data is input in step 2 and has been trained
In good model, output vector is obtained, then the vector is exactly the road network state for the next period predicted.
Road network status predication of the invention, using recurrent neural network algorithm, tool is Python2.7, before selection
Bimestrial sample set is as training set, and trimestral sample set is as test set.By the present invention and traditional BP nerve net
Network compares, and shown in Fig. 4 is the prediction result of period a RNN and BPNN randomly selecting.
As can be seen from the figure for RNN predicted value ratio BPNN closer to true value, precision of prediction is higher, due to actually answering
In, road network state limited reliability is described with the section velocity amplitude acquired in us, therefore we convert velocity amplitude to
Congestion level, i.e., there is different congestion level in each section in road network, and specific targets are as shown in table 1, as long as true congestion journey
Degree is consistent with predicted congestion degree, i.e., it is believed that predicting correctly, therefore the present invention is efficient right with entirety effective percentage using entirety
Model is evaluated, as shown in table 3.Each state vector has 278 sections in the present invention, there is 720 state vectors (two daily
Minute update primary), one shares 21600 state vectors in test set (one month).If by each section in different time
If congestion level is considered as a sample, one shares 6004800 samples.
Positive sample: congestion level predicts correct sample
Unimpeded sample in positive sample: predicted value or true value are smooth section
Whole accuracy rate: positive sample number/total number of samples
It is whole efficient: (the unimpeded sample in positive sample number-positive sample)/(unimpeded sample in total number of samples-positive sample
This)
3 model evaluation table of table
Model | Whole accuracy rate | It is whole efficient |
Recurrent neural networks model | 86.04% | 81.48% |
BP neural network model | 75.88% | 71.23% |
It is compared with BP neural network, the whole effective percentage of model (recurrent neural network) of the present invention improves
10.16%, whole effective percentage improves 10.25%, and from the point of view of the precision of prediction, the present invention provides a kind of reliable and steady
Fixed road network trend prediction method, facilitate traveler preferably planning path can more preferable planning path, improve out line efficiency.
Claims (2)
1. a kind of road network trend prediction method based on recurrent neural network, which is characterized in that the described method comprises the following steps:
Step 1: establishing sample set
A road network is chosen, road network k section is divided into, and each section is numbered, (1,2,3 ..., k) is denoted as, by one
It is divided into the j the same period of each time span for 24 hours, calculates each section in the average speed of each period;Meter
After the average speed for having calculated each section, the road network state of the period, i.e. V are indicated with a state vectorj=[v1,j,
v2,j,…,vk,j];The road network state for considering first three period predicts the road network state of next period,
Therefore, single sample is [(Vj-2,Vj-1,Vj),(Vj+1)] its expanded form are as follows:
Entire sample set is the set of all period samples;
Step 2: recurrent neural network models
Firstly, determining input/output variable;Input variable is three one-dimensional state vectors, respectively j-2, j-1, the road of j period
Net state vector, output variable one is one-dimensional state vector, i.e., the road network state of next period, each state vector
Dimension is k, i.e., the section number in road network;
Secondly, determining training set and test set, training set and test set are divided by predetermined ratio according to the sample set in step 1;
Finally, carrying out recurrent neural networks model parameter calibration;The recurrent neural network include input layer, hidden layer and
Output layer;Learn the weight matrix and bias vector between input layer and hidden layer, hidden layer and input layer by training data,
Wherein for each memory unit of the hidden layer there are three input and two outputs, input content includes xt, ht-1, ct-1, defeated
It out include ht, ct;Relationship between them is controlled by three doors, is input gate respectively, is forgotten door and out gate, in t
Shi Keyou:
Wherein it,ft,otRespectively for the output of input gate, forgetting door and out gate, w·,b·It is coefficient matrix and to be biased towards respectively
Amount, " " is point multiplication operation, and sigmoid and tanh are activation primitive, ctThe output of the memory unit of t moment hidden layer is represented,
ht-1It is the output in t-1 moment hidden layer,It is the output in t moment hidden layer;
Weight matrix W between hidden layer is [k, a z] matrix, and z is to hide layer unit number, and k is the quantity in section;Connection
Weight matrix U between input layer and hidden layer is [z, a z] matrix;The output matrix V between output layer is hidden in connection
It is [z, a k] matrix;The relationship of hidden layer and output layer are as follows:
Yj+1=HV+by
Wherein, H=[h1,h2,……hz], hzIt is the output valve of the hiding layer unit at z moment, the byIt is connection hidden layer and defeated
The bias function Y of layer outj+1=[y1,j+1,y2,j+1,…,yk,j+1];
Obtain establishing loss function after output layer, loss function be for measuring the output valve of output layer and the difference of true value,
Using loss function to parameters derivation, its gradient is calculated;Batch gradient is used when the training of the recurrent neural network
Descent method learning model parameter, model parameter include all weight matrix and bias vector;
Step 3: subsequent time road network status predication
The input variable of test data is input in step 2 in trained model, output vector is obtained, then this is defeated
Outgoing vector is exactly the road network state for the next period predicted.
2. the method according to claim 1, wherein the calculation method of the average speed is:
In a certain period of time, the mean value of all average vehicle speeds passed through in certain a road section, as shown in formula (1), wherein n generation
The number of vehicles that table passes through in the period in the section, m represent m-th of section in road network, m ∈ (1,2 ..., k), when j is
Between segment number, the length in behalf m-th of section, Δ t represents the length of period,Vehicle i is represented in the average speed of Δ t
Degree;
If certain a road section passes through in Δ t without vehicle, in the average speed substitution current slot with a period
Average speed, i.e.,
vm,j=vm,j-1;
After the average speed that each section has been calculated, the road network state of the period, V are indicated with a state vectorj=
[v1,j,v2,j,…,vk,j]。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611244476.5A CN106781489B (en) | 2016-12-29 | 2016-12-29 | A kind of road network trend prediction method based on recurrent neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611244476.5A CN106781489B (en) | 2016-12-29 | 2016-12-29 | A kind of road network trend prediction method based on recurrent neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106781489A CN106781489A (en) | 2017-05-31 |
CN106781489B true CN106781489B (en) | 2019-07-26 |
Family
ID=58927856
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611244476.5A Expired - Fee Related CN106781489B (en) | 2016-12-29 | 2016-12-29 | A kind of road network trend prediction method based on recurrent neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106781489B (en) |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107579816B (en) * | 2017-09-06 | 2020-05-19 | 中国科学院半导体研究所 | Method for generating password dictionary based on recurrent neural network |
CN107786369B (en) * | 2017-09-26 | 2020-02-04 | 广东电网有限责任公司电力调度控制中心 | Power communication network security situation perception and prediction method based on IRT (intelligent resilient test) hierarchical analysis and LSTM (local Scale TM) |
CN108133295B (en) * | 2018-01-11 | 2020-07-07 | 安徽优思天成智能科技有限公司 | Motor vehicle exhaust concentration continuous time prediction method for target road section |
CN108335487B (en) * | 2018-03-20 | 2020-04-10 | 北方工业大学 | Road traffic state prediction system based on traffic state time sequence |
CN109118014A (en) * | 2018-08-30 | 2019-01-01 | 浙江工业大学 | A kind of traffic flow speed prediction technique based on time recurrent neural network |
CN109919387A (en) * | 2019-03-15 | 2019-06-21 | 福建工程学院 | A kind of Metro Passenger crowding prediction technique using resampling recurrent neural network |
CN111460738B (en) * | 2020-04-16 | 2023-06-16 | 中南大学 | RNN-ARX modeling method and RNN-ARX model of magnetic suspension system |
CN111524353B (en) * | 2020-04-28 | 2021-08-17 | 中国计量大学 | Method for traffic text data for speed prediction and travel planning |
CN111862595B (en) * | 2020-06-08 | 2021-12-31 | 同济大学 | Speed prediction method, system, medium and device based on road network topological relation |
CN112446537A (en) * | 2020-11-20 | 2021-03-05 | 国网浙江省电力有限公司宁波供电公司 | Short-term load prediction method based on deep long-term and short-term memory network |
CN113449402B (en) * | 2021-06-22 | 2022-08-05 | 武汉大学 | Road network efficiency gain prediction method after broken road is opened |
CN113611119B (en) * | 2021-08-10 | 2022-06-28 | 北京航空航天大学 | Vehicle induction method based on gated recursion unit |
CN113554878A (en) * | 2021-09-18 | 2021-10-26 | 深圳市城市交通规划设计研究中心股份有限公司 | Road section impedance function determination method, calculation device and storage medium |
CN115394084B (en) * | 2022-08-29 | 2023-07-25 | 郑州轻工业大学 | Urban road network short-time traffic flow prediction method based on NMF-BiLSTM |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2007224206A1 (en) * | 2006-03-03 | 2007-09-13 | Inrix, Inc. | Assessing road traffic conditions using data from mobile data sources |
CN101866143B (en) * | 2009-04-14 | 2012-05-02 | 北京宏德信智源信息技术有限公司 | Road traffic service level prediction method based on space-time characteristic aggregation |
CN102737508B (en) * | 2012-06-19 | 2014-03-26 | 银江股份有限公司 | Urban road traffic state detection method combined with support vector machine (SVM) and back propagation (BP) neural network |
CN102750824B (en) * | 2012-06-19 | 2014-04-16 | 银江股份有限公司 | Urban road traffic condition detection method based on voting of network sorter |
CN105206056B (en) * | 2015-09-25 | 2017-06-16 | 珠海高凌信息科技有限公司 | Road traffic pollution source intelligent Forecasting and system |
CN105788272B (en) * | 2016-05-16 | 2018-03-16 | 杭州智诚惠通科技有限公司 | A kind of method and system of vehicle flow jam alarming |
CN106205126B (en) * | 2016-08-12 | 2019-01-15 | 北京航空航天大学 | Large-scale Traffic Network congestion prediction technique and device based on convolutional neural networks |
-
2016
- 2016-12-29 CN CN201611244476.5A patent/CN106781489B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN106781489A (en) | 2017-05-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106781489B (en) | A kind of road network trend prediction method based on recurrent neural network | |
Li et al. | Prediction for tourism flow based on LSTM neural network | |
CN107180530B (en) | A kind of road network trend prediction method based on depth space-time convolution loop network | |
CN110223517B (en) | Short-term traffic flow prediction method based on space-time correlation | |
Zheng et al. | Traffic state estimation using stochastic Lagrangian dynamics | |
CN110164128A (en) | A kind of City-level intelligent transportation analogue system | |
CN110782093B (en) | PM fusing SSAE deep feature learning and LSTM2.5Hourly concentration prediction method and system | |
Chiappone et al. | Traffic simulation models calibration using speed–density relationship: An automated procedure based on genetic algorithm | |
Mingheng et al. | Accurate multisteps traffic flow prediction based on SVM | |
Bernard et al. | Using artificial neural networks to identify learning styles | |
CN108108808B (en) | Position prediction method and device based on deep belief network | |
CN108062561A (en) | A kind of short time data stream Forecasting Methodology based on long memory network model in short-term | |
CN110164129B (en) | Single-intersection multi-lane traffic flow prediction method based on GERNN | |
CN106504535B (en) | A kind of trip distribution modeling method of combination Gravity Models and Fratar models | |
CN111767517B (en) | BiGRU multi-step prediction method, system and storage medium applied to flood prediction | |
CN104636801A (en) | Transmission line audible noise prediction method based on BP neural network optimization | |
CN104091216A (en) | Traffic information predication method based on fruit fly optimization least-squares support vector machine | |
CN105335619A (en) | Collaborative optimization method applicable to parameter back analysis of high calculation cost numerical calculation model | |
CN113205698A (en) | Navigation reminding method based on IGWO-LSTM short-time traffic flow prediction | |
CN108416458A (en) | A kind of tunnel rich water rock mass Synthetic Geological Prediction Ahead of Construction method based on BP neural network | |
CN113449905A (en) | Traffic jam early warning method based on gated cyclic unit neural network | |
Adak et al. | An elective course suggestion system developed in computer engineering department using fuzzy logic | |
CN109146055A (en) | Modified particle swarm optimization method based on orthogonalizing experiments and artificial neural network | |
Robati et al. | Inflation rate modeling: adaptive neuro-fuzzy inference system approach and particle swarm optimization algorithm (ANFIS-PSO) | |
Zhang et al. | Direction-decision learning based pedestrian flow behavior investigation |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190726 Termination date: 20201229 |
|
CF01 | Termination of patent right due to non-payment of annual fee |