CN108734614A - Traffic congestion prediction technique and device, storage medium - Google Patents
Traffic congestion prediction technique and device, storage medium Download PDFInfo
- Publication number
- CN108734614A CN108734614A CN201710241528.1A CN201710241528A CN108734614A CN 108734614 A CN108734614 A CN 108734614A CN 201710241528 A CN201710241528 A CN 201710241528A CN 108734614 A CN108734614 A CN 108734614A
- Authority
- CN
- China
- Prior art keywords
- processing
- layer module
- data
- neural network
- recognition
- 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
Classifications
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
A kind of traffic congestion prediction technique, device and storage medium, the method for one embodiment include：Embedded processing is carried out to the input data of acquisition, obtains data after embedded processing, data are the data for the format that convolutional neural networks can identify after embedded processing, and input data includes time cycle mark and time segment identification；Data after embedded handle are handled using convolutional neural networks, goes out data characteristics from extracting data after embedded handle, obtains convolutional neural networks handling result；Convolutional neural networks handling result is handled using Recognition with Recurrent Neural Network, obtains Recognition with Recurrent Neural Network handling result；Own coding processing is carried out to Recognition with Recurrent Neural Network handling result, obtained from coded treatment result；Full connection processing is carried out to own coding handling result, obtains full connection processing result；Classification processing is carried out to full connection processing result, obtains traffic congestion prediction result.This embodiment scheme improves the accuracy of traffic congestion prediction.
Description
Technical field
The present invention relates to traffic information technical fields, are gathered around more particularly to a kind of traffic congestion prediction technique, a kind of traffic
Stifled prediction meanss and a kind of storage medium.
Background technology
With the continuous growth of city automobile ownership, and the construction of urban traffic road receive many restrictions development compared with
Slowly, can not be fast-developing with the growth of car ownership, therefore, traffic congestion has become one of urban transportation more
Universal phenomenon has seriously affected the go off daily of resident.For this phenomenon, to real-time traffic in occurring at present very much
Stream is analyzed, and analysis and identification goes out the scheme of the congested link in traffic route, to provide reference to the user of trip.However,
Although this mode can so that the daily of resident can be facilitated significantly to a certain extent to avoid congested link in trip
Trip, however due to being to predict in real time traffic congestion, and do not have centainly perspective.Accordingly, to traffic congestion
Carrying out prediction becomes a more important developing direction.By predicting traffic congestion situation, to traffic guidance, most preferably
Path planning, traffic administration and control have very important significance.However, the scheme of current traffic congestion prediction, mostly
Number is predicted for traffic flow, can not be predicted the traffic congestion degree of certain specific a road section, and generally existing
The low problem of accuracy rate.
Invention content
Based on this, the embodiment of the present invention to be designed to provide a kind of traffic congestion prediction technique, a kind of traffic congestion pre-
Device and a kind of storage medium are surveyed, to improve the accuracy of traffic congestion prediction.
In order to achieve the above objectives, following technical scheme is used in one embodiment：
A kind of traffic congestion prediction technique, including step：
Embedded processing is carried out to the input data of acquisition, obtains data after embedded processing, data are after the embedded processing
The data for the format that convolutional neural networks can identify, the input data include time cycle mark and time segment identification；
Data after the embedded processing are handled using convolutional neural networks, from being carried in data after the embedded processing
Data characteristics is taken out, convolutional neural networks handling result is obtained；
The convolutional neural networks handling result is handled using Recognition with Recurrent Neural Network, is obtained at Recognition with Recurrent Neural Network
Manage result；
Own coding processing is carried out to the Recognition with Recurrent Neural Network handling result, obtained from coded treatment result；
Full connection processing is carried out to the own coding handling result, obtains full connection processing result；
Classification processing is carried out to the full connection processing result, obtains traffic congestion prediction result.
A kind of traffic congestion prediction meanss, including：Input layer module, neural net layer module and output layer module, institute
It includes that sequentially connected convolutional neural networks layer module, Recognition with Recurrent Neural Network layer module, own coding are hidden to state neural net layer module
Hide layer module and full articulamentum module；
The input layer module obtains input data, and carries out embedded processing to the input data, obtains embedded processing
Data afterwards, data are the data for the format that convolutional neural networks layer module can identify, the input number after the embedded processing
According to including time cycle mark and time segment identification；
The convolutional neural networks layer module, at using convolutional neural networks to data after the embedded processing
Reason goes out data characteristics from extracting data after the embedded processing, obtains convolutional neural networks handling result；
The Recognition with Recurrent Neural Network layer module, for using Recognition with Recurrent Neural Network to the convolutional neural networks layer module
It exports result and carries out operation processing；
The own coding hidden layer module carries out own coding for the output result to the Recognition with Recurrent Neural Network layer module
Processing, obtained from coded treatment result；
The full articulamentum module obtains full junction for carrying out full connection processing to the own coding handling result
Manage result；
The output layer module obtains traffic congestion prediction for carrying out classification processing to the full connection processing result
As a result.
A kind of storage medium, is stored thereon with computer program, is realized when which is executed by processor as described above
Traffic congestion prediction technique.
According to the scheme in embodiment as described above, using time cycle mark and time segment identification as inputting number
According to the dimension of required input data is low, and uses multiple neural networks comprising convolutional neural networks, Recognition with Recurrent Neural Network
Input data is handled, obtains traffic congestion prediction result on this basis.To which this embodiment scheme only needs very low
The data of dimension are obtained with preferable traffic congestion degree prediction result, and which raises the accuracys of traffic congestion prediction.
Description of the drawings
Fig. 1 is the schematic diagram of the working environment of a this embodiment scheme；
Fig. 2 is the schematic diagram of the composed structure of the terminal of one embodiment；
Fig. 3 is the schematic diagram of the composed structure of the server of one embodiment；
Fig. 4 is the flow diagram of the traffic congestion prediction technique in one embodiment；
Fig. 5 is the schematic diagram of the universal architecture frame in a specific example；
Fig. 6 is the structural schematic diagram of the neural network in a specific example；
Fig. 7 is with the schematic diagram of the accuracy comparison based on this embodiment scheme in application example；
Fig. 8 is the structural schematic diagram of the traffic congestion prediction meanss in one embodiment.
Specific implementation mode
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, with reference to the accompanying drawings and embodiments, to this
Invention is described in further detail.It should be appreciated that the specific embodiments described herein are only used to explain the present invention,
Do not limit protection scope of the present invention.
Unless otherwise defined, all of technologies and scientific terms used here by the article and belong to the technical field of the present invention
The normally understood meaning of technical staff is identical.Used term is intended merely to description tool in the description of the invention herein
The purpose of the embodiment of body, it is not intended that in the limitation present invention.Term as used herein " and/or " include one or more phases
Any and all combinations of the Listed Items of pass.
Fig. 1 shows the working environment schematic diagram in one embodiment of the invention, as shown in Figure 1, its working environment is related to
Terminal 101 and server 102, can be communicated between terminal 101 and server 102 by network.Terminal 101 can be from
Server 102 obtains various traffic datas, such as each section mark, each period, speed, congestion degree etc..Terminal 101, service
Device 102 can carry out the training of neural network, the nerve net that training obtains based on the various traffic datas obtained from server 102
Network can be used for terminal 101 or server 102 carries out the prediction of traffic congestion.
The structural schematic diagram of terminal 101 in one embodiment is as shown in Figure 2.The terminal 101 includes passing through system bus
Processor, non-volatile memory medium, communication interface, power interface and the memory of connection.Wherein, the processor of terminal 101 is used
In providing calculating and control ability, the operation of entire terminal 101 is supported.The non-volatile memory medium of terminal 101 is stored with behaviour
Make system and a kind of computer program of traffic congestion prediction meanss, the computer program of the traffic congestion prediction meanss is handled
Device realizes a kind of traffic congestion prediction technique when executing.The memory of terminal 101 is that the traffic in non-volatile memory medium is gathered around
The operation of stifled prediction meanss provides environment.The network interface of terminal 101 is used for server 102 through network connection and communication,
The power interface of terminal 101 for being connect with external power supply, powered by the power interface to terminal 101 by external power supply.Terminal
101 can be the equipment, such as mobile terminal, such as mobile phone, tablet computer etc. that any type can realize intelligent input output；
Can also be other equipment with above structure.
As shown in figure 3, in one embodiment server 102 include the processor connected by system bus, power supply module,
Storage medium, memory and communication interface.Wherein, which supports entire server for providing calculating and control ability
Operation.The storage medium of server is stored with operating system, database and a kind of computer application of traffic congestion prediction meanss
Program, the traffic congestion prediction meanss realize a kind of traffic congestion prediction technique when being executed by processor.Memory in server
Environment is provided for the operation of the traffic congestion prediction meanss in storage medium, communication interface is used to carry out network communication with terminal.
It will be understood by those skilled in the art that structure shown in Fig. 3, only with the frame of the relevant part-structure of example scheme
Figure, does not constitute the restriction for the server being applied thereon to this embodiment scheme, and specific server may include than figure
Shown in more or fewer components, either combine certain components or arranged with different components.
Fig. 4 shows the flow diagram of the traffic congestion prediction technique in one embodiment, as shown in figure 4, the implementation
Example in method include：
Step S401：Embedded processing is carried out to the input data of acquisition, obtains data after embedded processing, the embedded processing
Data are the data for the format that convolutional neural networks can identify afterwards, and the input data includes time cycle mark and time
Segment identification；
Step S402：Data after the embedded processing are handled using convolutional neural networks, from the embedded processing
Extracting data goes out data characteristics afterwards, obtains convolutional neural networks handling result；
Step S403：The convolutional neural networks handling result is handled using Recognition with Recurrent Neural Network, is recycled
Processing with Neural Network result；
Step S404：Own coding processing is carried out to the Recognition with Recurrent Neural Network handling result, obtained from coded treatment result；
Step S405：Full connection processing is carried out to the own coding handling result, obtains full connection processing result；
Step S406：Classification processing is carried out to the full connection processing result, obtains traffic congestion prediction result.
Wherein, in an application example of the embodiment, the input data of acquisition is carried out in above-mentioned steps S401 embedding
Before entering formula processing, it can also include the following steps：
The input data of acquisition is normalized.
To be normalized by the input data to acquisition, the facility for carrying out embedded processing can be improved
Property, improve the efficiency for carrying out embedded processing.
Wherein, when being handled data after the embedded processing using convolutional neural networks in above-mentioned steps S402,
32 characteristic patterns may be used in one example and carry out convolutional neural networks processing, and read 3 members from input data every time
Element carries out convolution operation, wherein the length of the pond layer of convolutional neural networks could be provided as 2.
In a specific example of the present embodiment, above-mentioned steps S402 obtain convolutional neural networks handling result it
Afterwards, before step S403 is handled the convolutional neural networks handling result using Recognition with Recurrent Neural Network, can also include
Step：
Ratio is exited using first the convolutional neural networks handling result is carried out exiting processing.
To, it exits ratio by using first and convolutional neural networks handling result is carried out to exit processing, it can be effective
Ground avoids over-fitting.
Similarly, after obtaining Recognition with Recurrent Neural Network handling result in above-mentioned steps S403, god is recycled described in step S40
Can also include step before carrying out own coding processing through network processes result：
Ratio is exited using second the Recognition with Recurrent Neural Network handling result is carried out exiting processing.
To exit ratio by using second and carry out exiting processing to Recognition with Recurrent Neural Network handling result, it is possibility to have
Avoid over-fitting to effect.
The convolutional neural networks handling result is carried out exiting processing it is understood that exiting ratio using first
Process exits the process that ratio to the Recognition with Recurrent Neural Network handling result exit processing with using second, can select
Any one is handled.In order to improve the effect for preventing over-fitting, the two processes for exiting processing can be carried out at the same time.
In addition, in a specific example, between above-mentioned convolutional neural networks and Recognition with Recurrent Neural Network, Recognition with Recurrent Neural Network
Own coding processing network between, own coding processing neural network and the network of full connection processing between be full connection,
To further increase the performance of traffic congestion prediction.
The traffic congestion prediction technique of the present embodiment as described above, substantially construct one rationally effectively towards
The general framework of city traffic road condition congestion degree prediction, to improve the accuracy rate predicted the traffic congestion in city.At this
In the scheme of embodiment, it is only necessary to which the data of opposite low dimensional can reach preferable precision and accuracy rate.Fig. 5 shows base
In the schematic diagram of the universal architecture frame of a specific example of this embodiment scheme.Refering to what is shown in Fig. 5, this embodiment scheme exists
After obtaining history road condition data, by the processing of Feature Engineering, then builds and be with convolutional neural networks and Recognition with Recurrent Neural Network
The neural network module of core, and be applied in the prediction of specific traffic congestion degree.
Before specifically predicting traffic congestion, first neural network can be trained, the nerve net that training obtains
Network is the neural network of structure shown in fig. 5, and the neural network module then obtained with training is applied to actual traffic and gathers around
In stifled prediction, the traffic congestion degree in following section is predicted, at this time above-mentioned convolutional neural networks, Recognition with Recurrent Neural Network,
Own coding processing, full connection processing etc. collectively constitute the neural network module by obtaining after training.It, can when being trained
It is trained with the history road condition data to input.
When being trained, the training data of input being trained includes：Section mark, time cycle mark are with timely
Between segment identification, can also include road conditions congestion degree label, these training datas can be determined based on history road condition data.It can
To understand, the data for the input data that the training data inputted when being trained obtains when being predicted with actual traffic congestion itself
Type is identical, and the input data when training data inputted when only training is predicted relative to actual traffic congestion includes road
Condition congestion degree label, to show what road conditions this data is.Wherein, above-mentioned time cycle mark, can be with one month
It is used as a time cycle daily, can also be using one week as one time cycle.In the present embodiment, can be with heaven-made
For time cycle, one week time cycle as a whole.At this point, above-mentioned time cycle mark can be：Monday, Tuesday ...
Sunday, time segment identification can be specific time segment identifications, the specific length of period can be needed in conjunction with actual techniques into
Row setting, such as each different period is divided for 0 to 24 point, the length of each period can be identical, can not also phase
Together.Each training data is trained again after being normalized, and exemplary training data of concrete application shows
Such as it is lower shown：
status times Mon Tue Wed Thu Fri Sat Sun
1 0.0231 0 0 1 0 0 0 0
Above-mentioned training example represents morning Wednesday 12：(data of the time segment identification in above-mentioned example are 30 data
Data after normalization standardization in other words), road conditions (status) are 1, i.e., unimpeded.It is appreciated that if being applied to reality
Traffic congestion prediction during, then the parameter of status then be not necessarily to input.
As it can be seen that in this example, input data can be only the data of 8 dimensions, to using neural network
When module is handled, only Processing with Neural Network can also be carried out with 8 neurons.
Refering to what is shown in Fig. 5, after inputting training data, by Feature Engineering module to the training data of the low-dimensional of input
Embedded processing, by training data processing to support the data of Processing with Neural Network, to support the processing of neural network.
After obtaining neural network after training by training, you can the neural network obtained based on training is handed over
The actual prediction of logical congestion.
When carrying out traffic congestion prediction, input data is obtained first, as described above, the input data obtained is with the time
It is main, including time cycle and time segment identification can be first to obtaining in conjunction with shown in the structural schematic diagram of neural network shown in fig. 6
The input data taken is normalized, and then carries out embedded processing to the input data after normalized, is located
Reason is that neural network can be with the data of the format of identifying processing.Can be to the embedded place of input data in conjunction with shown in Fig. 6
Reason is the data of the manageable format of convolutional neural networks, and the mode of specific embedded processing may be used any possible
Mode carries out, and in some applications, the process of embedded processing can also be integrated in the module of Processing with Neural Network.In this reality
It applies in example, a characteristic pattern is generated after being handled the input data after normalization by embedded processing, this feature figure is
Matrix based on 8*1 wherein 8 features are respectively the time cycle on Monday to Sunday and specific period, therefore inputs length
Degree is 8, and " uniform " can be used to initialize.
For the input data after embedded processing, the neural network module that above-mentioned construction may be used carries out depth
It practises.Refering to what is shown in Fig. 6, first using one-dimensional convolutional neural networks (Convolutional to the input data after embedded processing
Neural Networks, CNN) it is handled as hidden layer, further to extract data characteristics.In the present embodiment, it is adopting
When being handled with one-dimensional convolutional neural networks, a convolutional layer is only used and a pond layer carries out at convolutional neural networks
Reason.The input of pond layer derives from a upper convolutional layer, to provide very strong robustness, and reduces the quantity of parameter, prevents
The generation of over-fitting.In other examples, the number of plies of the convolutional layer, pond layer that carry out convolutional neural networks processing can be with
It is other numbers.Wherein, in a concrete application example, convolutional layer uses 32 characteristic patterns, and every time in input data
Read 3 elements and carry out convolution operations, other boundary schemes and activation primitive can use general " same " pattern and
" relu " activation primitive.And the length of pond layer could be provided as 2, the length of pond layer may be set to be it in other embodiments
His length.
After carrying out processing by convolutional neural networks and obtaining convolutional neural networks handling result, Recognition with Recurrent Neural Network is used
(Recurrent Neural Networks, RNN) is a new hidden layer and an and upper hidden layer (i.e. above-mentioned convolution god
Through network) it is connected entirely, to ensure to information remember and applied in the calculating of output.It is every in full connection i.e. one layer
A neuron is all connected with another layer of each neuron, then is referred to as this two layers full connection, in this example i.e. convolutional Neural
Each neuron of network is connected with each neuron of Recognition with Recurrent Neural Network.It is handled using Recognition with Recurrent Neural Network
When, it can be operated using 100 mnemons, other are not provided with any parameter.
In the scheme of the present embodiment, CNN and RNN are as the core for constituting the neural network module in this embodiment scheme
Neural network, and the combination of CNN and RNN can be effectively treated may input different types of data.In the present embodiment,
With the neural network of the core for being combined as neural network module of CNN and RNN, the different god of more layers can be added on demand
Through network.
After carrying out processing using Recognition with Recurrent Neural Network and obtaining Recognition with Recurrent Neural Network handling result, refering to what is shown in Fig. 6, using
Self-editing code layer handles convolutional neural networks handling result, to be processed to input information.Carrying out own coding processing
When, Sigmoid functions may be used as variation kernel function.Sigmoid functions are also referred to as S sigmoid growth curves, Sigmoid functions
It is defined as：It is formulated as s'(x to the derivative of x)=s (x) (1-s (x)).
Own coding processing is being carried out after coded treatment result, full junction is carried out to own coding handling result successively
Reason obtains full connection processing as a result, and carrying out classification processing acquisition traffic congestion prediction result to full connection processing result.Having
In the application example of body, the full connection hidden layer of band rectifier (relu) activation of 256 units can be used to be connected entirely
Processing, and classification is practiced using the output layer using Softmax functions and is handled, to be combined simultaneously to all nonlinear features
Export traffic congestion prediction result.In order to improve the performance of neural network, max-norm regularization parameters can be set for 3 with about
Beam connects the weight of hidden layer entirely, and extension of the Softmax functions as Sigmoid functions, can preferably promote more classification
The performance of handling result.In a specific example of this embodiment scheme, the traffic congestion prediction result of acquisition may be packet
It includes：Unobstructed (such as 60 kilometers or more/hour), slowly (such as 40 kilometers or more -60 kilometers or less/hour), congestion (such as 15
Kilometer or more -40 kilometers or less/hour) and heavy congestion (such as 15 kilometers or less/hour).
The structure of neural network shown in fig. 6 needs according to actual techniques application and inputs number based on CNN+RNN
According to difference, different types of neural network can be added and using more different parameters.In this embodiment scheme, it is
Over-fitting is prevented, as shown in fig. 6, between convolutional neural networks CNN and the Recognition with Recurrent Neural Network RNN, and cycle nerve net
Between network RNN and own coding hidden layer, all employs one and exit layer.The two exit layer and may be used one, in order into one
Step, which is promoted, prevents the performance of over-fitting, can two retain two and exit layer.Two are exited layer and exit ratio and identical also can may be used
With difference.At one in application example, for convenience's sake, the ratio that exits that can exit two on layer is both configured to 0.2, i.e.,
It is not included in the update cycle there are one can be random in every five inputs.
In the example depicted in fig. 6, it is the processing that convolutional neural networks are first carried out to the input data after embedded processing
It carries out illustrating for the processing of Recognition with Recurrent Neural Network again, in actual techniques application, convolutional neural networks can also be exchanged
With the sequence of Recognition with Recurrent Neural Network, it can also be and handled using other neural networks, can also be and add different god
Through network.In actual techniques in application, each different neural network can be arranged in general-purpose library, in general-purpose library it is each not
Same neural network can be the neural network done after package processing, and each neural network after package processing can be direct
It is connected, when building neural network, can directly adds different neural networks, and each neural network is set in general-purpose library
The sequence of front and back processing, then carries out the training of neural network module, and is applied to after training in traffic congestion prediction.In skill
In art application, as long as not changing the number of input neuron, the number finally exported is the same.Such as middle pond shown in Fig. 6
It is directly to connect Recognition with Recurrent Neural Network layer after layer, can also be directly to connect cycle nerve after input layer in other embodiments
Network layer.As to how add each neural network, due to having done package processing in general-purpose library, package treated each nerve net
Network itself is just supported and is connected directly, therefore can more advantageously carry out the structure of neural network.Such as needing addition 100
When mnemon uses the Recognition with Recurrent Neural Network of default setting, it is only necessary to which following code can be realized：
model.add(LSTM(100))
Wherein, model refers to entire neural network, and a variety of different layers can be added to this neural network.Such as it can be with
Pass through one new empty neural network of following code establishing：
Model=Sequential ()
Wherein, it when being added to different neural networks, can be accessed between two neural networks that arbitrary neighborhood connects
One is exited layer, to prevent over-fitting as far as possible.
In order to verify performance, sufficient experiment is carried out to the frame of example scheme as described above, has been adopted in an experiment
Use 30, the cities Liao Mou main road section (3 kilometers one section) one month totally 44580 data samples as history road condition data collection and do
10FOLD cross validations obtain the neural network module of structure to be trained.In concrete application example, pass through Keras softwares
Packet is developed, and operation is carried out on display card chip GTX 1080, will using the result that this embodiment scheme obtains with it is current general
The result that method obtains is compared, including decision tree, SVM and multi-layered perception neural networks (wherein Multilayer Perception god
Through all being set in network and this embodiment scheme using 200 epoch (epoch)), the reality of the accuracy comparison finally obtained
Test that the results are shown in Figure 7.As can be seen from Figure 7：The predictablity rate of decision tree is 0.5297, and the prediction of SVM is accurate
True rate is 0.635, and the relatively high predictablity rate of multi-layered perception neural networks is 0.726, and the prediction of this embodiment scheme
To 0.795, accuracy rate is obviously improved rate of accuracy reached.It follows that under the conditions of low latitudes data experiment, this implementation
Predictablity rate is greatly improved in the prediction of urban traffic situation congestion degree in example scheme, refering to what is shown in Fig. 7, opposite decision tree carries
Height has been more than 50%, and opposing layers layered perception neural networks improve about 10%.
The scheme of the present embodiment as described above can be one and rationally effectively be predicted towards city traffic road condition congestion degree
General framework, and can support polymorphic multi-source data, that is, the input data obtained in addition to comprising the above-mentioned time cycle identify with
Can also include other kinds of data and except time segment identification, the feature after embedded processing above-mentioned in the case
The dimension of figure also adjusts accordingly.
Based on thought same as mentioned above, Fig. 8 shows the knot of the traffic congestion prediction meanss in one embodiment
Structure schematic diagram.As shown in figure 8, the traffic congestion prediction meanss in the embodiment include：Input layer module 801, neural net layer
Module 802 and output layer module 803, wherein neural net layer module 802 includes sequentially connected convolutional neural networks layer mould
Block 8021, Recognition with Recurrent Neural Network layer module 8022, own coding hidden layer module 8023 and full articulamentum module 8024.
Input layer module 801 obtains input data, and carries out embedded processing to the input data, after obtaining embedded processing
Data, data are the data for the format that convolutional neural networks layer module can identify, the input data after the embedded processing
Including time cycle mark and time segment identification.
Convolutional neural networks layer module 8021 is handled data after the embedded processing using convolutional neural networks, from
Extracting data goes out data characteristics after the embedded processing, obtains convolutional neural networks handling result.
Output of the Recognition with Recurrent Neural Network layer module 8023 using Recognition with Recurrent Neural Network to the convolutional neural networks layer module
As a result operation processing is carried out, Recognition with Recurrent Neural Network handling result is obtained.
Own coding hidden layer module 8025 carries out own coding processing to the output result of the Recognition with Recurrent Neural Network layer module,
Obtained from coded treatment result.
Full articulamentum module 8026 carries out full connection processing to the own coding handling result, obtains full connection processing knot
Fruit.
Output layer module 803 carries out classification processing to the full connection processing result, obtains traffic congestion prediction result.
According to the scheme in embodiment as described above, using time cycle mark and time segment identification as inputting number
According to the dimension of required input data is low, and uses multiple neural networks comprising convolutional neural networks, Recognition with Recurrent Neural Network
Input data is handled, obtains traffic congestion prediction result on this basis.To which this embodiment scheme only needs very low
The data of dimension are obtained with preferable traffic congestion degree prediction result, and which raises the accuracys of traffic congestion prediction.
As shown in figure 8, the traffic congestion prediction meanss in the embodiment can also include model training module 8001, the mould
Type training module 8001 calls the neural net layer module 802 to be trained the training data of input, the training data
Including：Section mark, time cycle mark, time segment identification and road conditions congestion degree label.
At this point, neural net layer module 802 is to be trained and then be applied to through model training module 8001 specifically
Traffic congestion prediction in.
In a specific example, as shown in figure 8, the traffic congestion prediction meanss of the embodiment can also include normalization
Module 8002, the normalization module 8002 is for being normalized input data.
At this point, above-mentioned input layer module 801 is to carry out embedded processing to the input data after normalized.
To be normalized by the input data to acquisition, the facility for carrying out embedded processing can be improved
Property, improve the efficiency for carrying out embedded processing.
In one example, above-mentioned convolutional neural networks layer module 8021 may be used 32 characteristic patterns and carry out convolutional Neural
Network processes, and read 3 elements from input data every time and carry out convolution operation, wherein the pond layer of convolutional neural networks
Length could be provided as 2.
In one example, as shown in figure 8, the neural net layer module 802 in the example can also include：It is connected to volume
First between product neural net layer module 8021 and Recognition with Recurrent Neural Network layer module 8023 exits layer module 8022.This first is moved back
Go out layer module 8022 to exit ratio using first the convolutional neural networks handling result is carried out exiting processing.To pass through
Ratio is exited using first convolutional neural networks handling result is carried out to exit processing, can be effectively prevented from over-fitting.
In one example, as shown in figure 8, the neural net layer module 802 in the example can also include：It is connected to and follows
Second between ring neural net layer module 8023 and own coding hidden layer module 8025 exits layer module 8024.This second is exited
Layer module 8024 exits ratio using second and carries out exiting processing to the Recognition with Recurrent Neural Network handling result.To by adopting
Ratio is exited with second Recognition with Recurrent Neural Network handling result is carried out to exit processing, it is possibility to have avoid over-fitting to effect.
It is understood that first exits layer module 8022 and second and exits layer module 8024 and can select any one,
In order to improve the effect for preventing over-fitting, it can also be while first exit layer module 8022 and second and exit layer module 8024.
Wherein, between above-mentioned convolutional neural networks layer module 8021 and Recognition with Recurrent Neural Network layer module 8023, cycle nerve
Between network layer module 8023 and own coding hidden layer module 8025, own coding hidden layer module 8025 and full articulamentum module
It is full connection between 8026, to further increase the performance of traffic congestion prediction.
Based on example as described above, a kind of storage medium is also provided in one embodiment, is stored thereon with computer journey
Sequence, the computer program realize traffic congestion prediction technique as described above when being executed by processor.
One of ordinary skill in the art will appreciate that realizing all or part of flow in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, it is non-volatile computer-readable that the program can be stored in one
It takes in storage medium, in the embodiment of the present invention, which can be stored in the storage medium of computer system, and by the calculating
At least one of machine system processor executes, and includes the flow such as the embodiment of above-mentioned each method with realization.Wherein, described
Storage medium can be magnetic disc, CD, read-only memory (Read-Only Memory, ROM) or random access memory
(Random Access Memory, RAM) etc..
Each technical characteristic of embodiment described above can be combined arbitrarily, to keep description succinct, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, it is all considered to be the range of this specification record.
Several embodiments of the invention above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to the protection of the present invention
Range.Therefore, the protection domain of patent of the present invention should be determined by the appended claims.
Claims (11)
1. a kind of traffic congestion prediction technique, which is characterized in that including step：
Embedded processing is carried out to the input data of acquisition, obtains data after embedded processing, data are convolution after the embedded processing
The data for the format that neural network can identify, the input data include time cycle mark and time segment identification；
Data after the embedded processing are handled using convolutional neural networks, are gone out from extracting data after the embedded processing
Data characteristics obtains convolutional neural networks handling result；
The convolutional neural networks handling result is handled using Recognition with Recurrent Neural Network, obtains Recognition with Recurrent Neural Network processing knot
Fruit；
Own coding processing is carried out to the Recognition with Recurrent Neural Network handling result, obtained from coded treatment result；
Full connection processing is carried out to the own coding handling result, obtains full connection processing result；
Classification processing is carried out to the full connection processing result, obtains traffic congestion prediction result.
2. traffic congestion prediction technique according to claim 1, which is characterized in that be embedded in the input data of acquisition
Further include step before formula processing：
The input data of acquisition is normalized.
3. traffic congestion prediction technique according to claim 1, which is characterized in that obtaining convolutional neural networks processing knot
Further include step before being handled the convolutional neural networks handling result using Recognition with Recurrent Neural Network after fruit：
Ratio is exited using first the convolutional neural networks handling result is carried out exiting processing.
4. traffic congestion prediction technique according to claim 1, which is characterized in that obtaining Recognition with Recurrent Neural Network processing knot
Further include step before carrying out own coding processing to the Recognition with Recurrent Neural Network handling result after fruit：
Ratio is exited using second the Recognition with Recurrent Neural Network handling result is carried out exiting processing.
5. traffic congestion prediction technique according to claim 1, which is characterized in that convolutional neural networks and cycle nerve net
Between network, between the network of Recognition with Recurrent Neural Network and own coding processing, neural network and the full connection processing of own coding processing
It is full connection between network.
6. a kind of traffic congestion prediction meanss, which is characterized in that including：Input layer module, neural net layer module and output
Layer module, the neural net layer module include sequentially connected convolutional neural networks layer module, Recognition with Recurrent Neural Network layer module,
Own coding hidden layer module and full articulamentum module；
The input layer module obtains input data, and carries out embedded processing to the input data, obtains number after embedded processing
According to data are the data for the format that convolutional neural networks layer module can identify, the input data packet after the embedded processing
Include time cycle mark and time segment identification；
The convolutional neural networks layer module, for being handled data after the embedded processing using convolutional neural networks,
Go out data characteristics from extracting data after the embedded processing, obtains convolutional neural networks handling result；
The Recognition with Recurrent Neural Network layer module, for using output of the Recognition with Recurrent Neural Network to the convolutional neural networks layer module
As a result operation processing is carried out, Recognition with Recurrent Neural Network handling result is obtained；
The own coding hidden layer module is carried out for the output result to the Recognition with Recurrent Neural Network layer module at own coding
Reason, obtained from coded treatment result；
The full articulamentum module obtains full connection processing knot for carrying out full connection processing to the own coding handling result
Fruit；
The output layer module obtains traffic congestion prediction result for carrying out classification processing to the full connection processing result.
7. traffic congestion prediction meanss according to claim 6, which is characterized in that further include model training module, it is described
Model training module calls the neural net layer module to be trained the training data of input, and the training data includes：
Section mark, time cycle mark, time segment identification and road conditions congestion degree label.
8. traffic congestion prediction meanss according to claim 6, it is characterised in that：
Further include normalization module, for input data to be normalized；
The input layer module carries out embedded processing to the input data after the normalized.
9. traffic congestion prediction meanss according to claim 6, which is characterized in that including at least one in following two
?：
The neural net layer module further includes：It is connected to the convolutional neural networks layer module and the Recognition with Recurrent Neural Network layer
First between module exits a layer module；
The neural net layer module further includes：It is connected to the Recognition with Recurrent Neural Network layer module and the own coding hidden layer mould
Second between block exits a layer module.
10. traffic congestion prediction technique according to claim 6, which is characterized in that convolutional neural networks layer module with follow
Between ring neural net layer module, between Recognition with Recurrent Neural Network layer module and own coding hidden layer module, own coding hidden layer mould
It is full connection between block and full articulamentum module.
11. a kind of storage medium, is stored thereon with computer program, which is characterized in that the program is realized when being executed by processor
Traffic congestion prediction technique as described in claim 1 to 5 any one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710241528.1A CN108734614A (en) | 2017-04-13 | 2017-04-13 | Traffic congestion prediction technique and device, storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710241528.1A CN108734614A (en) | 2017-04-13 | 2017-04-13 | Traffic congestion prediction technique and device, storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108734614A true CN108734614A (en) | 2018-11-02 |
Family
ID=63925058
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710241528.1A Pending CN108734614A (en) | 2017-04-13 | 2017-04-13 | Traffic congestion prediction technique and device, storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108734614A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458223A (en) * | 2019-08-06 | 2019-11-15 | 湖南省华芯医疗器械有限公司 | Tumor of bronchus automatic testing method and detection system under a kind of scope |
| CN110674887A (en) * | 2019-10-08 | 2020-01-10 | 中兴飞流信息科技有限公司 | End-to-end road congestion detection algorithm based on video classification |
| CN111145541A (en) * | 2019-12-18 | 2020-05-12 | 深圳先进技术研究院 | Traffic flow data prediction method, storage medium, and computer device |
| CN111862605A (en) * | 2020-07-20 | 2020-10-30 | 腾讯科技（深圳）有限公司 | Road condition detection method and device, electronic equipment and readable storage medium |
| WO2020232732A1 (en) * | 2019-05-23 | 2020-11-26 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for predicting traffic information |
| CN112882683A (en) * | 2019-11-29 | 2021-06-01 | 阿里巴巴集团控股有限公司 | Random number generator judging method, random number processing method, device and equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105046277A (en) * | 2015-07-15 | 2015-11-11 | 华南农业大学 | Robust mechanism research method of characteristic significance in image quality evaluation |
| CN105279495A (en) * | 2015-10-23 | 2016-01-27 | 天津大学 | Video description method based on deep learning and text summarization |
| CN106205126A (en) * | 2016-08-12 | 2016-12-07 | 北京航空航天大学 | Large-scale Traffic Network based on convolutional neural networks is blocked up Forecasting Methodology and device |
| CN106250707A (en) * | 2016-08-12 | 2016-12-21 | 王双坤 | A kind of based on degree of depth learning algorithm process head construction as the method for data |
| CN106355879A (en) * | 2016-09-30 | 2017-01-25 | 西安翔迅科技有限责任公司 | Time-space correlation-based urban traffic flow prediction method |
| CN106446782A (en) * | 2016-08-29 | 2017-02-22 | 北京小米移动软件有限公司 | Image identification method and device |
-
2017
- 2017-04-13 CN CN201710241528.1A patent/CN108734614A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105046277A (en) * | 2015-07-15 | 2015-11-11 | 华南农业大学 | Robust mechanism research method of characteristic significance in image quality evaluation |
| CN105279495A (en) * | 2015-10-23 | 2016-01-27 | 天津大学 | Video description method based on deep learning and text summarization |
| CN106205126A (en) * | 2016-08-12 | 2016-12-07 | 北京航空航天大学 | Large-scale Traffic Network based on convolutional neural networks is blocked up Forecasting Methodology and device |
| CN106250707A (en) * | 2016-08-12 | 2016-12-21 | 王双坤 | A kind of based on degree of depth learning algorithm process head construction as the method for data |
| CN106446782A (en) * | 2016-08-29 | 2017-02-22 | 北京小米移动软件有限公司 | Image identification method and device |
| CN106355879A (en) * | 2016-09-30 | 2017-01-25 | 西安翔迅科技有限责任公司 | Time-space correlation-based urban traffic flow prediction method |
Non-Patent Citations (1)
| Title |
|---|
| 彭玉青 等: "基于微调优化的深度学习在语音识别中的应用", 《郑州大学学报（理学版）》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020232732A1 (en) * | 2019-05-23 | 2020-11-26 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for predicting traffic information |
| CN110458223A (en) * | 2019-08-06 | 2019-11-15 | 湖南省华芯医疗器械有限公司 | Tumor of bronchus automatic testing method and detection system under a kind of scope |
| CN110674887A (en) * | 2019-10-08 | 2020-01-10 | 中兴飞流信息科技有限公司 | End-to-end road congestion detection algorithm based on video classification |
| CN112882683A (en) * | 2019-11-29 | 2021-06-01 | 阿里巴巴集团控股有限公司 | Random number generator judging method, random number processing method, device and equipment |
| CN111145541A (en) * | 2019-12-18 | 2020-05-12 | 深圳先进技术研究院 | Traffic flow data prediction method, storage medium, and computer device |
| CN111862605A (en) * | 2020-07-20 | 2020-10-30 | 腾讯科技（深圳）有限公司 | Road condition detection method and device, electronic equipment and readable storage medium |
| CN111862605B (en) * | 2020-07-20 | 2022-03-08 | 腾讯科技（深圳）有限公司 | Road condition detection method and device, electronic equipment and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108734614A (en) | Traffic congestion prediction technique and device, storage medium | |
| Yin et al. | Deep learning on traffic prediction: Methods, analysis, and future directions | |
| Perea et al. | Prediction of irrigation event occurrence at farm level using optimal decision trees | |
| CN107038478A (en) | Road condition predicting method and device, computer equipment and computer-readable recording medium | |
| CN110555568B (en) | Road traffic running state real-time perception method based on social network information | |
| CN107045788A (en) | Traffic Forecasting Methodology and device | |
| CN109360097A (en) | Prediction of Stock Index method, apparatus, equipment and storage medium based on deep learning | |
| CN110363449A (en) | A kind of Risk Identification Method, apparatus and system | |
| CN109840628B (en) | Short-time multi-zone vehicle speed prediction method and system | |
| Song et al. | A match‐then‐predict method for daily traffic flow forecasting based on group method of data handling | |
| Sarkheyli et al. | Using an effective tabu search in interactive resources scheduling problem for LEO satellites missions | |
| CN110956807B (en) | Highway flow prediction method based on combination of multi-source data and sliding window | |
| CN110910659A (en) | Traffic flow prediction method, device, equipment and storage medium | |
| CN108275759A (en) | Method for treating water based on neural network and system | |
| CN115049152A (en) | Airport transfer area arrival passenger flow prediction method and device and electronic equipment | |
| Lin et al. | Path travel time estimation using attribute-related hybrid trajectories network | |
| CN112036598A (en) | Charging pile use information prediction method based on multi-information coupling | |
| CN113838303B (en) | Parking lot recommendation method and device, electronic equipment and storage medium | |
| Liu et al. | Dynamic scheduling for semiconductor manufacturing systems with uncertainties using convolutional neural networks and reinforcement learning | |
| CN110263136A (en) | The method and apparatus for pushing object to user based on intensified learning model | |
| CN107704942A (en) | Distribution path determines method, apparatus and equipment | |
| Pescaru et al. | Ensemble based traffic light control for city zones using a reduced number of sensors | |
| CN107729078A (en) | Background application management-control method, device, storage medium and electronic equipment | |
| Dowsland | Classical techniques | |
| Chen et al. | GDCRN: global diffusion convolutional residual network for traffic flow prediction |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181102 |
|
| RJ01 | Rejection of invention patent application after publication |