CN106710222B - A kind of traffic flow forecasting method and device - Google Patents
A kind of traffic flow forecasting method and device Download PDFInfo
- Publication number
- CN106710222B CN106710222B CN201710174865.3A CN201710174865A CN106710222B CN 106710222 B CN106710222 B CN 106710222B CN 201710174865 A CN201710174865 A CN 201710174865A CN 106710222 B CN106710222 B CN 106710222B
- Authority
- CN
- China
- Prior art keywords
- traffic flow
- neural network
- magnitude
- obtains
- moth
- 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
Landscapes
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of traffic flow forecasting methods, method includes the following steps: obtaining the magnitude of traffic flow time series of set period of time region of interest within before current time;WAVELET PACKET DECOMPOSITION is carried out to magnitude of traffic flow time series, resolves into multiple subsequences;Each subsequence is predicted using the flame algorithm Neural Network Optimization model that training obtains in advance is caught based on moth dynamic sensing, obtains multiple predicted values;Reconstruct is superimposed each predicted value, obtains the short-term following traffic flow magnitude in target area.Using technical solution provided by the embodiment of the present invention, the changing rule of the magnitude of traffic flow can be captured, promotes accuracy and the generalization ability of predicting traffic flow amount time series.The invention also discloses a kind of traffic flow forecasting devices, have relevant art effect.
Description
Technical field
The present invention relates to computer application technologies, more particularly to a kind of traffic flow forecasting method and device.
Background technique
With the rapid development of intelligent transportation system, the magnitude of traffic flow is carried out accurately predicting to be traffic programme and friendship in real time
The basis of logical induction.Short-term traffic flow prediction has the nonlinear characteristics such as sudden, uncertain and index of chaotic degree, is to hand at present
The hot spot of logical expert and scholar's research.
The nonlinear model progress short-term traffic flow for being normally based on neural network and support vector machines at present is pre-
It surveys.Model of traffic flux forecast and inclusiveness such as based on BP neural network technology examine the magnitude of traffic flow of SVM vector machine pre-
Survey model etc..
These methods have some disadvantages.Wherein, BP neural network is calculated using gradient descent method adjustment threshold value and weight
Method is easily precocious, so that precision of prediction is lower;The quality of vector machine SVM prediction result is closely related with the selection of parameter, and performance is not
Stablize, generalization ability is weaker.
Summary of the invention
The object of the present invention is to provide a kind of traffic flow forecasting method and devices, to promote predicting traffic flow amount time sequence
The accuracy of column and generalization ability.
In order to solve the above technical problems, the invention provides the following technical scheme:
A kind of traffic flow forecasting method, comprising:
The magnitude of traffic flow time series of set period of time region of interest within before acquisition current time;
WAVELET PACKET DECOMPOSITION is carried out to the magnitude of traffic flow time series, resolves into multiple subsequences;
The flame algorithm Neural Network Optimization model that training obtains in advance is caught to each subsequence using based on moth dynamic sensing
It is predicted, obtains multiple predicted values;
Reconstruct is superimposed each predicted value, obtains the short-term following traffic flow magnitude in the target area.
It is described that wavelet packet point is carried out to the magnitude of traffic flow time series in a kind of specific embodiment of the invention
Solution, resolves into multiple subsequences, comprising:
WAVELET PACKET DECOMPOSITION is carried out to the magnitude of traffic flow time series according to the following formula, obtains multiple subsequences:
Wherein, diFor the WAVELET PACKET DECOMPOSITION frequency band coefficient of tier I, h0And g0It is that WAVELET PACKET DECOMPOSITION conjugate filter is
Number, l are the time parameter of positioning index, and k is the frequency domain parameter of scale index.
In a kind of specific embodiment of the invention, being based on moth dynamic sensing by following steps, to catch flame algorithm preparatory
Training obtains the Neural Network Optimization model:
Elman neural network is established, determines basic parameter;
Population of the initialization package containing multiple individuals;
The training Elman neural network, the fitness value of each individual in the population is calculated according to training result;
Artificial moth carries out light source dynamic sensing and catches flame, decodes to the individual of the population, optimizing search exploitation, adaptively
Relevant weight and threshold value are adjusted, and feeds back to the Elman neural network;
The step of repeating the training Elman neural network sets until the number of iterations of the population is greater than
The value for determining the fitness function of first threshold or the Elman neural network is less than setting second threshold, obtains the nerve
Network Optimization Model.
It is described that each individual in the population is calculated according to training result in a kind of specific embodiment of the invention
Fitness value, comprising:
The fitness value of each individual in the population is calculated according to the following formula:
Wherein, fobj is the evaluation fitness function of the Elman neural network, yk(w) andIt is described respectively
The desired output and real output value of Elman neural network, m are output layer training dimensions.
In a kind of specific embodiment of the invention, the artificial moth carries out light source dynamic sensing and catches flame, comprising:
Artificial moth carries out light source dynamic sensing according to the following formula and catches flame:
S=a1·S(Mi,Fα)+a2·S(Mi,Fβ)+a3·S(Mi,Fγ);
Wherein, S is the updated position of artificial moth, a1、a2、a3For the random number on [0,1], Fα, FβAnd FγFor the overall situation
First three optimal flame, S (Mi,Fbest)=Fbest-A·DiFor artificial moth MiTowards global optimum's flames F exitingbestLinear motion updates
Position, artificial moth MiTo flames F exitingbestDistance be Di=| CFbest-Mi|, C=2r2, A=2ar1- a, a are in section
Linear decrease on [0,2], r1、r2For the random number on [0,1].
A kind of traffic flow forecasting device, comprising:
Magnitude of traffic flow time series obtains module, for obtaining the friendship of set period of time region of interest within before current time
Through-current capacity time series;
WAVELET PACKET DECOMPOSITION module resolves into multiple sons for carrying out WAVELET PACKET DECOMPOSITION to the magnitude of traffic flow time series
Sequence;
Predicted value obtains module, for catching the flame algorithm neural network that training obtains in advance using based on moth dynamic sensing
Optimized model predicts each subsequence, obtains multiple predicted values;
Traffic flow magnitude prediction module is superimposed each predicted value for reconstructing, and obtains the short-term following friendship in the target area
Through-flow magnitude.
In a kind of specific embodiment of the invention, the WAVELET PACKET DECOMPOSITION module is specifically used for:
WAVELET PACKET DECOMPOSITION is carried out to the magnitude of traffic flow time series according to the following formula, obtains multiple subsequences:
Wherein, diFor the WAVELET PACKET DECOMPOSITION frequency band coefficient of tier I, h0And g0It is that WAVELET PACKET DECOMPOSITION conjugate filter is
Number, l are the time parameter of positioning index, and k is the frequency domain parameter of scale index.
It further include that Neural Network Optimization model obtains module, for passing through in a kind of specific embodiment of the invention
Following steps be based on moth dynamic sensing catch flame algorithm in advance training obtain the Neural Network Optimization model:
Elman neural network is established, determines basic parameter;
Population of the initialization package containing multiple individuals;
The training Elman neural network, the fitness value of each individual in the population is calculated according to training result;
Artificial moth carries out light source dynamic sensing and catches flame, decodes to the individual of the population, optimizing search exploitation, adaptively
Relevant weight and threshold value are adjusted, and feeds back to the Elman neural network;
The step of repeating the training Elman neural network sets until the number of iterations of the population is greater than
The value for determining the fitness function of first threshold or the Elman neural network is less than setting second threshold, obtains the nerve
Network Optimization Model.
In a kind of specific embodiment of the invention, the Neural Network Optimization model obtains module, is specifically used for:
The fitness value of each individual in the population is calculated according to the following formula:
Wherein, fobj is the evaluation fitness function of the Elman neural network, yk(w) andIt is described respectively
The desired output and real output value of Elman neural network, m are output layer training dimensions.
In a kind of specific embodiment of the invention, the Neural Network Optimization model obtains module, is specifically used for:
Artificial moth carries out light source dynamic sensing according to the following formula and catches flame:
S=a1·S(Mi,Fα)+a2·S(Mi,Fβ)+a3·S(Mi,Fγ);
Wherein, S is the updated position of artificial moth, a1、a2、a3For the random number on [0,1], Fα, FβAnd FγFor the overall situation
First three optimal flame, S (Mi,Fbest)=Fbest-A·DiFor artificial moth MiTowards global optimum's flames F exitingbestLinear motion updates
Position, artificial moth MiTo flames F exitingbestDistance be Di=| CFbest-Mi|, C=2r2, A=2ar1- a, a are in section
Linear decrease on [0,2], r1、r2For the random number on [0,1].
Using technical solution provided by the embodiment of the present invention, set period of time region of interest within before current time is obtained
Magnitude of traffic flow time series after, to magnitude of traffic flow time series carry out WAVELET PACKET DECOMPOSITION, resolve into multiple subsequences, utilize base
The flame algorithm Neural Network Optimization model that training obtains in advance is caught in moth dynamic sensing to predict each subsequence, can be obtained
Multiple predicted values are obtained, superposition is reconstructed to each predicted value, the short-term following traffic flow magnitude in target area, capture can be obtained
To the changing rule of the magnitude of traffic flow, accuracy and the generalization ability of predicting traffic flow amount time series are promoted.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of implementation flow chart of traffic flow forecasting method in the embodiment of the present invention;
Fig. 2 is three layers of decomposition contrast schematic diagram of signal tree of wavelet decomposition and WAVELET PACKET DECOMPOSITION in the embodiment of the present invention;
Fig. 3 is a kind of schematic diagram that artificial moth light source dynamic sensing catches flame in the embodiment of the present invention;
Fig. 4 is a kind of schematic diagram of Neural Network Optimization model training in the embodiment of the present invention;
Fig. 5 is a kind of schematic diagram of traffic flow forecasting process in the embodiment of the present invention;
Fig. 6 is a kind of structural schematic diagram of traffic flow forecasting device in the embodiment of the present invention.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.Obviously, described embodiments are only a part of the embodiments of the present invention, rather than
Whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
It is shown in Figure 1, it is a kind of implementation flow chart of traffic flow forecasting method provided by the embodiment of the present invention, it should
Method may comprise steps of:
S110: the magnitude of traffic flow time series of set period of time region of interest within before current time is obtained.
Target area is the region of pending traffic flow forecasting.It can be with by installing image capture device in target area
The traffic conditions of target area are monitored.In monitoring process, the magnitude of traffic flow of target area can be obtained in real time, thus
It is easy to get the magnitude of traffic flow time series of set period of time region of interest within before current time.
Set period of time before current time can be a period close to current time, when can also be current
Carve the set period of time in several days continuous before.The period can be set and be adjusted according to the actual situation, and the present invention is real
It is without limitation to apply example.
S120: WAVELET PACKET DECOMPOSITION is carried out to the magnitude of traffic flow time series, resolves into multiple subsequences.
WAVELET PACKET DECOMPOSITION (Wavelet Packet Decomposition, WPD) is in wavelet decomposition (Wavelet
Decomposition, WD) only to the parsing of signal low frequency part on the basis of, increase a kind of letter for parsing to signal high frequency section
Number complete decomposition method.
Fig. 2 show the time series of magnitude of traffic flow size respectively by the comparative situation of wavelet decomposition and WAVELET PACKET DECOMPOSITION,
Decomposition order is 3.As can be known from Fig. 2, in decomposing at i-th layer, WAVELET PACKET DECOMPOSITION parses (2 more than wavelet decompositioni- 2) a frequency range
Signal, the signal detail of more magnitude of traffic flow time serieses can be captured, grasp its wave characteristic, compared to wavelet decomposition, more
Time Series suitable for the complicated non-linear magnitude of traffic flow.
In a kind of specific embodiment of the invention, magnitude of traffic flow time series can be carried out according to formula (1) small
Wave packet decomposes, and obtains multiple subsequences:
Wherein, diFor the WAVELET PACKET DECOMPOSITION frequency band coefficient of tier I, h0And g0It is that WAVELET PACKET DECOMPOSITION conjugate filter is
Number, l are the time parameter of positioning index, and k is the frequency domain parameter of scale index.
S130: the flame algorithm Neural Network Optimization model that training obtains in advance is caught to each son using based on moth dynamic sensing
Sequence is predicted, multiple predicted values are obtained.
In embodiments of the present invention, can based on moth dynamic sensing catch flame algorithm in advance training obtain Neural Network Optimization
Model.Each subsequence can be predicted using the Neural Network Optimization model, obtain multiple predicted values.
In a kind of specific embodiment of the invention, moth dynamic sensing can be based on by following steps and catch flame algorithm
Training obtains Neural Network Optimization model in advance:
Step 1: Elman neural network is established, determines basic parameter;
Step 2: population of the initialization package containing multiple individuals;
Step 3: the training Elman neural network calculates the fitness of each individual in population according to training result
Value;
Step 4: artificial moth carries out light source dynamic sensing and catches flame, is decoded to the individual of the population, optimizing search
Exploitation, adaptively adjusts relevant weight and threshold value, feeds back to the Elman neural network;
Step 5: repeating the operation of step 3, until population the number of iterations be greater than setting first threshold or
The value of the fitness function of Elman neural network is less than setting second threshold, obtains Neural Network Optimization model.
It is illustrated for ease of description, above-mentioned five steps are combined.
In embodiments of the present invention, Elman neural network can first be established.
Elman neural network is a kind of BP network of band feedback, has the forward direction of local memory unit, LOCAL FEEDBACK connection
Neural network and multilayered structure similar with Multilayer Feedforward Neural Networks.Compared with traditional BP neural network, Elman neural network
Prediction model with stronger dynamic behaviour and computing capability, suitable for settling time sequence.But Elman nerve net
Network is same as BP neural network to use momentum gradient descent method adjustment weight and threshold value, local optimum is easily trapped into, moreover, working as shadow
When the factor of sound and learning sample increase, the calculation amount and weight number of neural network will be sharply increased, and cause convergence rate slower.
In embodiments of the present invention, moth dynamic sensing is added in the training of Elman neural network and catches flame (Chaotic
Moth-Flame Optimization Algorithm Based On Dynamic Perceived Of Light Source,
CMFO) algorithm optimizes Elman neural network, makes the threshold value and Weight number adaptively tune of Elman neural network by CMFO algorithm
It is whole, improve precision of prediction and speed.
CMFO algorithm be caught for moth flame optimization (Moth-Flame Optimization, MFO) algorithm be easy it is precocious and
The slow defect of convergence rate proposes, introduces light source dynamic sensing mechanism for the flame behavior of catching of artificial moth, enables artificial moth
It is enough that offline mode is dynamically changed to accelerate convergence rate according to fitness weight factor.
In MFO algorithm, the searching process of artificial moth individual, which can be abstracted as, catches flame behavior and abandoning two kinds of rows of flame behavior
For.
Wherein, in catching flame behavior, artificial moth MiBased on itself phototactic characteristics, perceive in flame group from itself recently and
Optimal flames F exitingi, follow logatithmic spiral to move towards flame, shown in capture flame motion profile such as formula (2):
S(Mi,Fi)=Di·ebt·cos(2πt)+FiFormula (2)
Wherein S (Mi,Fi) indicate artificial moth MiAround flames F exitingiThe updated position of screw, Di=| Fi-Mi| it is
I-th artificial moth MiTo i-th of flames F exitingiDistance, b is the constant for moulding logatithmic spiral track.T is on [- 1,1] section
Random number, defines how far of the artificial moth apart from flame the next position, and t=-1 is closest to flame, t=1 be from
Flame is farthest.
In an iterative process, artificial moth can discard useless flame, reduce its number adaptively, finally converge on same
In a flame, accelerate late convergence, flame number flame_no can be indicated with formula (3):
Wherein, N is current iteration number, and T is total the number of iterations.
Only considered artificial moth in MFO algorithm is only influenced by itself corresponding short distance flame diverging light, and office is surrounded
The optimal flame in portion, which is done, updates the lesser helical curve movement of step-length, the phototactic characteristics of artificial moth is ignored, not to fitness
Value more preferably draw close by remote flame, under the attraction of its parallel rays, does and updates the biggish linear motion of step-length.
In order to improve convergence rate, the embodiment of the present invention introduces light source fitness weight factor ω, enables artificial moth
Enough according to ω dynamic sensing fitness value, more preferably flame automatically switches to catch flame mode under the attracting of its irradiation light, separate
Step-size in search biggish directional light linear motion is done when optimal solution, does the exploitation lesser diverging light spiral of step-length when close to optimal solution
Movement, the combination of both motor patterns accelerate the convergence rate of population.
Shown in light source fitness weight factor ω such as formula (4):
Wherein, i is flames F exitingiCorresponding index subscript, the i the big, and corresponding fitness value is more excellent, and flame_no is flame number
Mesh can be obtained by formula (3).As position iteration updates, flame_no monotone decreasing, ω monotonic increase, artificial moth
First attracted by remote directional light and accelerate search, then enhancing exploitation is attracted by short distance diverging light.
Fig. 3 illustrates artificial moth to be changed according to light source, and dynamic sensing catches the process of flame.When (δ is parallel rays to ω≤δ
Critical value between divergent rays) when, itself is smaller around the fitness value of light source, and it is remote most that artificial moth is intended to perception
The directional light of excellent flame switches lateral rectilinear flight motor pattern, captures global optimum's flame, increases step-length and accelerates search, such as
Shown in formula (5).
S(Mi,Fbest)=Fbest-A·DiFormula (5)
Wherein, S (Mi,Fbest) indicate artificial moth MiTowards global optimum's flames F exitingbestMove along a straight line the position updated, manually
Moth MiTo flames F exitingbestDistance be Di=| CFbest-Mi|, C=2r2, A=2ar1- a, a are linear on section [0,2]
Successively decrease, r1、r2For the random number on [0,1].
To avoid artificial moth from falling into local optimum in linear motion, the embodiment of the present invention introduces harmony search
Most beautiful harmony mechanism in (Harmony Search, HS) algorithm allows artificial moth towards first three global optimal flames F exitingα,
FβAnd FγMovement, to excavate potential optimal flame solution.Shown in the updated position S of artificial moth such as formula (6).Wherein, a1、
a2、a3For the random number on [0,1].
S=a1·S(Mi,Fα)+a2·S(Mi,Fβ)+a3·S(Mi,Fγ) formula (6)
As ω > δ, itself is larger around the fitness value of light source, and artificial moth is intended to perceive the hair of short distance flame
Astigmatism switches logarithmic spiral sporting flying mode to capture flame, step-size in search enhancing exploitation is reduced, as shown in formula (2).
A kind of schematic diagram that Fig. 4 show Neural Network Optimization model training determines base after establishing Elman neural network
This parameter are as follows: input layer vector is n dimension, and hidden layer and the vector for accepting layer are h dimension, and the vector of output layer is m dimension.Initialization
Population comprising multiple individuals.
Individual UVR exposure II=1 ..., N=[IW11...IWhnCW11...CWhhOW11...OWmhbH1...bHhbC1...bChbM1..
.bMm], wherein IWhn、OWmhAnd CWhhRespectively input layer to hidden layer, hidden layer to output layer and accepts layer to hidden layer
Connection weight, bHh、bCh、bMmRespectively hidden layer, undertaking layer and the corresponding threshold value of output layer.Elman neural network obtains training
Start to train after sample set, the fitness value of each individual in population can be calculated according to training result.Artificial moth carries out light
Source dynamic sensing catches flame.Specifically, artificial moth, which can carry out light source dynamic sensing according to formula (6), catches flame.To of population
Body decoding, optimizing search exploitation, adaptively adjusts relevant weight and threshold value, and feed back to Elman neural network.
Elman neural network provides mean square error MSE, the CMFO calculation of training result to CMFO algorithm in its training process
The weight W and threshold value B of the corresponding training layer of normal direction Elman neural network feedback.Elman neural network is fed back according to CMFO algorithm
It is worth adaptive adjusting training, shown in evaluation fitness function such as formula (7), yk(w) andIt is Elman neural network respectively
Desired output and real output value, m be output layer training dimension.
The fitness value of each individual in population can be calculated according to the formula (7).
The step of repeating trained Elman neural network, until reaching preset termination condition.Preset termination condition
Are as follows: the value that the number of iterations of population is greater than the fitness function of setting first threshold or Elman neural network is less than setting the
Two threshold values.First threshold and second threshold can be set and be adjusted according to the actual situation, for example, set second threshold as
0.01。
After the completion of Neural Network Optimization model training, it can use the Neural Network Optimization model and each subsequence carried out in advance
Survey processing, obtains multiple predicted values.
S140: reconstruct is superimposed each predicted value, obtains the short-term following traffic flow magnitude in target area.
Superposition is reconstructed to multiple predicted values that step S130 is obtained, the short-term following traffic in target area can be obtained
Flow value.
Specifically, wavelet package reconstruction can be carried out to multiple predicted values according to formula (8):
Wherein, fiIt is the wavelet package reconstruction frequency band coefficient of tier I, p1And q1It is that wavelet package reconstruction conjugate filter is
Number, l are the time parameter of positioning index, and k is the frequency domain parameter of scale index.
As shown in figure 5, being a kind of schematic diagram of traffic flow forecasting process, magnitude of traffic flow time series passes through wavelet packet point
Xie Hou obtains subsequence AAA3, subsequence AAD3, subsequence DDD3Deng these subsequences are input to Neural Network Optimization model
In, available corresponding forecasting sequence AAA3, forecasting sequence AAD3, forecasting sequence DDD3Deng predicted value, to each predicted value into
Row superposition reconstruct, can obtain prediction result, i.e., the short-term following traffic flow magnitude.
Combination of embodiment of the present invention WAVELET PACKET DECOMPOSITION, moth dynamic sensing catch flame algorithm and the building of Elman neural network is non-
Linear model can cope with the processing of chaotic sea now and the magnitude of traffic flow time series of complexity, realize simple, operation height
Effect.And moth dynamic sensing catches flame algorithm with preferable convergence rate and solving precision, catches flame algorithm with moth dynamic sensing
Optimize Elman neural network, solves traditional neural network using gradient descent method to adjust threshold value and weight, cause to predict
Model convergence rate is slow and the problem of easily falling into locally optimal solution, at the same can be to avoid the optimization algorithm existing defects because of use due to
The case where influencing estimated performance, occurs.
Using method provided by the embodiment of the present invention, the friendship of set period of time region of interest within before current time is obtained
After through-current capacity time series, WAVELET PACKET DECOMPOSITION is carried out to magnitude of traffic flow time series, resolves into multiple subsequences, using based on winged
Moth dynamic sensing is caught the flame algorithm Neural Network Optimization model that training obtains in advance and is predicted each subsequence, can obtain more
Superposition is reconstructed to each predicted value in a predicted value, can obtain the short-term following traffic flow magnitude in target area, capture friendship
The changing rule of through-current capacity promotes accuracy and the generalization ability of predicting traffic flow amount time series.
Corresponding to above method embodiment, the embodiment of the invention also provides a kind of traffic flow forecasting devices, hereafter
A kind of traffic flow forecasting device of description can correspond to each other reference with a kind of above-described traffic flow forecasting method.
Shown in Figure 6, which comprises the following modules:
Magnitude of traffic flow time series obtains module 210, for obtaining set period of time region of interest within before current time
Magnitude of traffic flow time series;
WAVELET PACKET DECOMPOSITION module 220 resolves into multiple for carrying out WAVELET PACKET DECOMPOSITION to the magnitude of traffic flow time series
Subsequence;
Predicted value obtains module 230, for catching the flame algorithm nerve that training obtains in advance using based on moth dynamic sensing
Network Optimization Model predicts each subsequence, obtains multiple predicted values;
Traffic flow magnitude prediction module 240 is superimposed each predicted value for reconstructing, and obtains the target area short-term future
Traffic flow magnitude.
Using device provided by the embodiment of the present invention, the friendship of set period of time region of interest within before current time is obtained
After through-current capacity time series, WAVELET PACKET DECOMPOSITION is carried out to magnitude of traffic flow time series, resolves into multiple subsequences, using based on winged
Moth dynamic sensing is caught the flame algorithm Neural Network Optimization model that training obtains in advance and is predicted each subsequence, can obtain more
Superposition is reconstructed to each predicted value in a predicted value, can obtain the short-term following traffic flow magnitude in target area, capture friendship
The changing rule of through-current capacity promotes accuracy and the generalization ability of predicting traffic flow amount time series.
In a kind of specific embodiment of the invention, the WAVELET PACKET DECOMPOSITION module 220 is specifically used for:
WAVELET PACKET DECOMPOSITION is carried out to the magnitude of traffic flow time series according to the following formula, obtains multiple subsequences:
Wherein, diFor the WAVELET PACKET DECOMPOSITION frequency band coefficient of tier I, h0And g0It is that WAVELET PACKET DECOMPOSITION conjugate filter is
Number, l are the time parameter of positioning index, and k is the frequency domain parameter of scale index.
It further include that Neural Network Optimization model obtains module, for passing through in a kind of specific embodiment of the invention
Following steps be based on moth dynamic sensing catch flame algorithm in advance training obtain the Neural Network Optimization model:
Elman neural network is established, determines basic parameter;
Population of the initialization package containing multiple individuals;
The training Elman neural network, the fitness value of each individual in the population is calculated according to training result;
Artificial moth carries out light source dynamic sensing and catches flame, decodes to the individual of the population, optimizing search exploitation, adaptively
Relevant weight and threshold value are adjusted, and feeds back to the Elman neural network;
The step of repeating the training Elman neural network sets until the number of iterations of the population is greater than
The value for determining the fitness function of first threshold or the Elman neural network is less than setting second threshold, obtains the nerve
Network Optimization Model.
In a kind of specific embodiment of the invention, the Neural Network Optimization model obtains module, is specifically used for:
The fitness value of each individual in the population is calculated according to the following formula:
Wherein, fobj is the evaluation fitness function of the Elman neural network, yk(w) andIt is described respectively
The desired output and real output value of Elman neural network, m are output layer training dimensions.
In a kind of specific embodiment of the invention, the Neural Network Optimization model obtains module, is specifically used for:
Artificial moth carries out light source dynamic sensing according to the following formula and catches flame:
S=a1·S(Mi,Fα)+a2·S(Mi,Fβ)+a3·S(Mi,Fγ);
Wherein, S is the updated position of artificial moth, a1、a2、a3For the random number on [0,1], Fα, FβAnd FγFor the overall situation
First three optimal flame, S (Mi,Fbest)=Fbest-A·DiFor artificial moth MiTowards global optimum's flames F exitingbestLinear motion updates
Position, artificial moth MiTo flames F exitingbestDistance be Di=| CFbest-Mi|, C=2r2, A=2ar1- a, a are in section
Linear decrease on [0,2], r1、r2For the random number on [0,1].
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with it is other
The difference of embodiment, same or similar part may refer to each other between each embodiment.For being filled disclosed in embodiment
For setting, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is referring to method part
Explanation.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Used herein a specific example illustrates the principle and implementation of the invention, and above embodiments are said
It is bright to be merely used to help understand technical solution of the present invention and its core concept.It should be pointed out that for the common of the art
, without departing from the principle of the present invention, can be with several improvements and modifications are made to the present invention for technical staff, these
Improvement and modification are also fallen within the protection scope of the claims of the present invention.
Claims (6)
1. a kind of traffic flow forecasting method characterized by comprising
The magnitude of traffic flow time series of set period of time region of interest within before acquisition current time;
WAVELET PACKET DECOMPOSITION is carried out to the magnitude of traffic flow time series, resolves into multiple subsequences;
Each subsequence is carried out using the flame algorithm Neural Network Optimization model that training obtains in advance is caught based on moth dynamic sensing
Prediction, obtains multiple predicted values;
Reconstruct is superimposed each predicted value, obtains the short-term following traffic flow magnitude in the target area;
Wherein, by following steps be based on moth dynamic sensing catch flame algorithm in advance training obtain the Neural Network Optimization mould
Type:
Elman neural network is established, determines basic parameter;
Population of the initialization package containing multiple individuals;
The training Elman neural network, the fitness value of each individual in the population is calculated according to training result;
Artificial moth carries out light source dynamic sensing and catches flame, decodes to the individual of the population, optimizing search exploitation, adaptive to adjust
Relevant weight and threshold value, and feed back to the Elman neural network;
The step of repeating the training Elman neural network, until the number of iterations of the population is greater than setting the
The value of the fitness function of one threshold value or the Elman neural network is less than setting second threshold, obtains the neural network
Optimized model;
The artificial moth carries out light source dynamic sensing and catches flame, comprising:
Artificial moth carries out light source dynamic sensing according to the following formula and catches flame:
S=a1·S(Mi,Fα)+a2·S(Mi,Fβ)+a3·S(Mi,Fγ);
Wherein, S is the updated position of artificial moth, a1、a2、a3For the random number on [0,1], Fα, FβAnd FγFor it is global first three
Optimal flame, S (Mi,Fbest)=Fbest-A·DiFor artificial moth MiTowards global optimum's flames F exitingbestMove along a straight line the position updated
It sets, artificial moth MiTo flames F exitingbestDistance be Di=| CFbest-Mi|, C=2r2, A=2ar1- a, a be section [0,
2] variable of linear decrease, r on1、r2For the random number on [0,1].
2. traffic flow forecasting method according to claim 1, which is characterized in that described to the magnitude of traffic flow time sequence
Column carry out WAVELET PACKET DECOMPOSITION, resolve into multiple subsequences, comprising:
WAVELET PACKET DECOMPOSITION is carried out to the magnitude of traffic flow time series according to the following formula, obtains multiple subsequences:
Wherein, diFor the WAVELET PACKET DECOMPOSITION frequency band coefficient of tier I, h0And g0It is the coefficient of WAVELET PACKET DECOMPOSITION conjugate filter, l
For the time parameter of positioning index, k is the frequency domain parameter of scale index.
3. traffic flow forecasting method according to claim 1 or 2, which is characterized in that described to be calculated according to training result
The fitness value of each individual in the population, comprising:
The fitness value of each individual in the population is calculated according to the following formula:
Wherein, fobj is the evaluation fitness function of the Elman neural network, yk(w) andIt is the Elman mind respectively
Desired output and real output value through network, m are output layer training dimensions.
4. a kind of traffic flow forecasting device characterized by comprising
Magnitude of traffic flow time series obtains module, for obtaining the traffic flow of set period of time region of interest within before current time
Measure time series;
WAVELET PACKET DECOMPOSITION module resolves into multiple subsequences for carrying out WAVELET PACKET DECOMPOSITION to the magnitude of traffic flow time series;
Predicted value obtains module, for catching the flame algorithm Neural Network Optimization that training obtains in advance using based on moth dynamic sensing
Model predicts each subsequence, obtains multiple predicted values;
Traffic flow magnitude prediction module is superimposed each predicted value for reconstructing, and obtains the short-term following traffic flow in the target area
Magnitude;
Neural Network Optimization model obtains module, trains in advance for catching flame algorithm based on moth dynamic sensing by following steps
Obtain the Neural Network Optimization model:
Elman neural network is established, determines basic parameter;
Population of the initialization package containing multiple individuals;
The training Elman neural network, the fitness value of each individual in the population is calculated according to training result;
Artificial moth carries out light source dynamic sensing and catches flame, decodes to the individual of the population, optimizing search exploitation, adaptive to adjust
Relevant weight and threshold value, and feed back to the Elman neural network;
The step of repeating the training Elman neural network, until the number of iterations of the population is greater than setting the
The value of the fitness function of one threshold value or the Elman neural network is less than setting second threshold, obtains the neural network
Optimized model;
The Neural Network Optimization model obtains module, is specifically used for:
Artificial moth carries out light source dynamic sensing according to the following formula and catches flame:
S=a1·S(Mi,Fα)+a2·S(Mi,Fβ)+a3·S(Mi,Fγ);
Wherein, S is the updated position of artificial moth, a1、a2、a3For the random number on [0,1], Fα, FβAnd FγFor it is global first three
Optimal flame, S (Mi,Fbest)=Fbest-A·DiFor artificial moth MiTowards global optimum's flames F exitingbestMove along a straight line the position updated
It sets, artificial moth MiTo flames F exitingbestDistance be Di=| CFbest-Mi|, C=2r2, A=2ar1- a, a be section [0,
2] variable of linear decrease, r on1、r2For the random number on [0,1].
5. traffic flow forecasting device according to claim 4, which is characterized in that the WAVELET PACKET DECOMPOSITION module, specifically
For:
WAVELET PACKET DECOMPOSITION is carried out to the magnitude of traffic flow time series according to the following formula, obtains multiple subsequences:
Wherein, diFor the WAVELET PACKET DECOMPOSITION frequency band coefficient of tier I, h0And g0It is the coefficient of WAVELET PACKET DECOMPOSITION conjugate filter, l
For the time parameter of positioning index, k is the frequency domain parameter of scale index.
6. traffic flow forecasting device according to claim 4 or 5, which is characterized in that the Neural Network Optimization model
Module is obtained, is specifically used for:
The fitness value of each individual in the population is calculated according to the following formula:
Wherein, fobj is the evaluation fitness function of the Elman neural network, yk(w) andIt is the Elman mind respectively
Desired output and real output value through network, m are output layer training dimensions.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710174865.3A CN106710222B (en) | 2017-03-22 | 2017-03-22 | A kind of traffic flow forecasting method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710174865.3A CN106710222B (en) | 2017-03-22 | 2017-03-22 | A kind of traffic flow forecasting method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106710222A CN106710222A (en) | 2017-05-24 |
CN106710222B true CN106710222B (en) | 2019-07-02 |
Family
ID=58887018
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710174865.3A Expired - Fee Related CN106710222B (en) | 2017-03-22 | 2017-03-22 | A kind of traffic flow forecasting method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106710222B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108399749A (en) * | 2018-03-14 | 2018-08-14 | 西南交通大学 | A kind of traffic trip needing forecasting method in short-term |
CN110109350A (en) * | 2019-03-29 | 2019-08-09 | 广东工业大学 | A kind of power capture optimization method of wave-power device that catching flame algorithm based on chaos moth |
CN110299005B (en) * | 2019-06-10 | 2020-11-17 | 浙江大学 | Urban large-scale road network traffic speed prediction method based on deep ensemble learning |
CN111161538B (en) * | 2020-01-06 | 2021-07-02 | 东南大学 | Short-term traffic flow prediction method based on time series decomposition |
CN112669606B (en) * | 2020-12-24 | 2022-07-12 | 西安电子科技大学 | Traffic flow prediction method for training convolutional neural network by utilizing dynamic space-time diagram |
CN113380026B (en) * | 2021-05-28 | 2022-07-22 | 南京理工大学 | HMM-based highway traffic prediction method |
CN113743738A (en) * | 2021-08-11 | 2021-12-03 | 湖北省食品质量安全监督检验研究院 | Method and device for predicting food safety risk grade interval |
CN114664105B (en) * | 2022-04-21 | 2023-04-07 | 合肥工业大学 | Optimal path prediction method based on improved OLF-Elman neural network |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105788249A (en) * | 2014-12-16 | 2016-07-20 | 高德软件有限公司 | Traffic flow prediction method, prediction model generation method and device |
CN106357456A (en) * | 2016-10-11 | 2017-01-25 | 广东工业大学 | Prediction method of network traffic and device thereof |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH1079096A (en) * | 1996-09-03 | 1998-03-24 | Toshiba Corp | Device for providing path selection support information |
-
2017
- 2017-03-22 CN CN201710174865.3A patent/CN106710222B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105788249A (en) * | 2014-12-16 | 2016-07-20 | 高德软件有限公司 | Traffic flow prediction method, prediction model generation method and device |
CN106357456A (en) * | 2016-10-11 | 2017-01-25 | 广东工业大学 | Prediction method of network traffic and device thereof |
Non-Patent Citations (1)
Title |
---|
基于神经网络的短时交通流预测方法的研究与应用;朱学明;《CNKI优秀硕士学位论文全文库》;20131215;正文第29、33页 |
Also Published As
Publication number | Publication date |
---|---|
CN106710222A (en) | 2017-05-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106710222B (en) | A kind of traffic flow forecasting method and device | |
CN106357456B (en) | A kind of prediction technique and device of network flow | |
Fang et al. | Densely connected search space for more flexible neural architecture search | |
Stern et al. | Blockwise parallel decoding for deep autoregressive models | |
CN108063961B (en) | Self-adaptive code rate video transmission method and system based on reinforcement learning | |
US9984683B2 (en) | Automatic speech recognition using multi-dimensional models | |
CN106570597B (en) | Content popularity prediction method based on deep learning under SDN architecture | |
CN107291232A (en) | A kind of somatic sensation television game exchange method and system based on deep learning and big data | |
CN110858973B (en) | Cell network flow prediction method and device | |
CN109359140A (en) | A kind of sequence of recommendation method and device based on adaptive attention | |
JP2021502650A5 (en) | ||
CN113361777B (en) | Runoff prediction method and system based on VMD decomposition and IHHO optimization LSTM | |
CN111079899A (en) | Neural network model compression method, system, device and medium | |
CN107688493A (en) | Train the method, apparatus and system of deep neural network | |
CN107506821A (en) | A kind of improved particle group optimizing method | |
CN115100864B (en) | Traffic signal control optimization method based on improved sparrow search algorithm | |
US20220124387A1 (en) | Method for training bit rate decision model, and electronic device | |
Kaddour | Stop wasting my time! saving days of imagenet and bert training with latest weight averaging | |
CN108766585A (en) | Generation method, device and the computer readable storage medium of influenza prediction model | |
CN112717415B (en) | Information bottleneck theory-based AI (Artificial intelligence) training method for reinforcement learning battle game | |
CN109359404A (en) | Medium-and Long-Term Runoff Forecasting method based on experience wavelet noise and neural network fusion | |
CN109740695A (en) | Image-recognizing method based on adaptive full convolution attention network | |
CN109993302A (en) | The convolutional neural networks channel of knowledge based migration is compressed from selection and accelerated method | |
TW200304096A (en) | Self-organizing feature map with improved performance by non-monotonic variation of the learning rate | |
CN113887397B (en) | Classification method and classification system of electrophysiological signals based on ocean predator algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
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: 20190702 Termination date: 20200322 |
|
CF01 | Termination of patent right due to non-payment of annual fee |