CN106935034A - Towards the regional traffic flow forecasting system and method for car networking - Google Patents

Abstract

The invention discloses a kind of regional traffic flow forecasting system and method towards car networking.The system includes：External action data module (1), car networking data module (2), data processing module (3) and support vector regression module (4).Data processing module (3) includes weather, festivals or holidays, the row vector data of date and time using the data genaration of external action data module (1) and car networking data module (2)；Support vector regression module (4) learns forecast model using the training of these data, and the row vector data at next moment in cycle are combined using forecast model, completes the prediction to the regional traffic flow at next moment in cycle.The present invention considers the influence of weather, festivals or holidays, date and time to regional traffic flow, can effectively predict regional traffic flow.Can be used to dredge traffic and car networking resource is allocated, improve the ability of traffic control and the utilization ratio of car networking resource.

Description

Towards the regional traffic flow forecasting system and method for car networking

Technical field

The invention belongs to traffic forecasting technique field, more particularly to a kind of regional traffic flow Forecasting Methodology can be used to hand over Siphunculus control and car networking resource allocation.

Background technology

Car networking IoV be vehicle Intranet, vehicle-mounted mobile internet and car border net based on, car and car, Che Yulu, Between car and pedestrian, car and internet and car and high in the clouds, entered with Data Exchange Standard by the communication protocol of unified agreement Row wireless telecommunications and the big grid of information exchange, are management, the information clothes of Intelligent Dynamic that can carry out intelligent traffic The integrated network of business and the control of Vehicular intelligent etc..The network is by GPS, RFID, sensor, camera image treatment etc. Device, completes the collection of itself environment and status information；By Internet technology, all of vehicle passes the various information of itself It is defeated to converge to central processing unit；By computer technology, the information of these a large amount of vehicles is analyzed and processed, so as to calculate not With the best route of vehicle, report without delay road conditions and arrangement signal lamp cycle.The magnitude of traffic flow refers to pass through in seclected time section The traffic entity number in a certain region, a certain road section or a certain track.The information of forecasting of the magnitude of traffic flow is to carry out intelligence in ITS Traffic control, dynamic traffic state identification and prediction and the key of real-time traffic flowable state induction.

Current traffic flow forecasting technical method mainly has two classes：One is statistical forecast algorithm model, such as rolling average, from Regressive is average, Kalman filtering and linear regression etc.；Two is based on the artificial intelligence i.e. model of machine learning algorithm.But The technical method of the existing magnitude of traffic flow is concentrated mainly on the traffic flow forecasting in a certain road section or a certain track, few Technical method is predicted for regional traffic flow.But in the traffic environment of reality, due to the movement of vehicle, in same period Carve, some regions occur the magnitude of traffic flow higher, and the magnitude of traffic flow in some regions is then relatively low, these regional traffic flows It is unbalanced to bring serious influence to traffic control and the car networking level of resources utilization.

The content of the invention

It is an object of the invention to be directed to above-mentioned the deficiencies in the prior art, a kind of Regional Traffic Flow towards car networking is proposed Amount forecasting system and method, to improve the ability of traffic control and the utilization ratio of car networking resource.

Technical thought of the invention is：The gps data information of each vehicle user collected by car networking, is considered Weather, festivals or holidays, date, time, learn forecast model using support vector regression training, for regional traffic flow is provided Accurately prediction.

According to above-mentioned thinking, regional traffic flow forecasting system of the present invention towards car networking, it is characterised in that including：

External action data module, weather conditions and every day of its record every day whether be festivals or holidays data letter Breath, for the external action data source as data processing module；

Car networking data module, the gps data information of the vehicle user of all travelings in its record car networking, for conduct The internal influence data source of data processing module；

Data processing module, for external action data and car networking data by being input into external action data module The internal influence data of module input carry out numerical quantization treatment, produce the row vector of multidimensional and are input to support vector regression Module；

Support vector regression module, it is pre- for being trained using the multidimensional row vector being input into by data processing module Survey, learn forecast model, be predicted with the magnitude of traffic flow to the future period moment.

According to above-mentioned thinking, the present invention carries out the Forecasting Methodology of regional traffic flow using said system, it is characterised in that Comprise the following steps：

1) initialize：Determine benchmark time, the cycle T of prediction and number of training m；

2) data processing module is provided according to the result and external action data module of initialization with car networking data module The data genaration current period moment and its preceding at moment in m-1 cycle at common moment in m cycle data；

3) support vector regression module learns forecast model using the data training that data processing module is generated, and utilizes The magnitude of traffic flow at forecast model prediction output m+1 moment in cycle；

4) when the historical juncture is turned at the m+1 moment in cycle of prediction, when the m+1 moment in cycle is updated into current period Carve；

5) circulation performs step 2) -4), the regional traffic flow at next moment in cycle is uninterruptedly predicted in completion.

The invention has the advantages that：

First, the present invention combines the gps data of vehicle in external action data and car networking, special by data processing module Different quantification treatment, produce multidimensional row vector, and these row vectors are analyzed using support vector regression, train and Study obtains regional traffic flow and weather, festivals or holidays, the internal relation of date and time, can construct forecast model；

Second, the multiple regions of forecast model energy built using the present invention carry out traffic flow forecasting, can analyze multi-region The traffic flow conditions at domain future period moment, and obtain predicting the outcome for multizone future period moment；

3rd, using it is of the invention predict the outcome may indicate that traffic dispersion and and car networking resource is allocated so that Improve the ability of traffic control and the utilization ratio of car networking resource.

Brief description of the drawings

System diagram of Fig. 1 present invention towards the prediction of car networking regional traffic flow；

Flow chart of Fig. 2 present invention towards car networking regional traffic flow Forecasting Methodology.

Specific embodiment

The present invention is described in detail with reference to the accompanying drawings and examples.

Reference picture 1, it is of the invention towards car networking regional traffic flow forecasting system, including：External action data module 1, Car networking data module 2, data processing module 3 and support vector regression module 4.Wherein：

External action data module 1, weather conditions and every day of its record every day whether be festivals or holidays data letter Breath, and the data message is input to data processing module 3, as the external action data source of data processing module 3；

Car networking data module 2, the gps data information of the vehicle user of all travelings in its record car networking, and should Data message is input to data processing module 3, used as the internal influence data source of data processing module 3；

Data processing module 3, for external action data and car networking number by being input into external action data module 1 The internal influence data being input into according to module 2 carry out numerical quantization treatment, produce the row vector of multidimensional and are input to supporting vector and return Return machine module 4, as the training data of support vector regression module 4；

Support vector regression module 4, it is pre- for being trained using the multidimensional row vector being input into by data processing module 3 Survey, learn forecast model, be predicted with the magnitude of traffic flow to the future period moment, and output predicts the outcome.

The data of the record of external action data module 1, its content at least includes date, time, weather conditions, and will Festivals or holidays are labeled as 1, and non-festivals or holidays are labeled as 0.

The gps data of the vehicle user in the record of car networking data module 2 car networking in each traveling, its data Form at least includes date, time and longitude and latitude.

The multidimensional row vector that the data processing module 3 is produced, is expressed as：

Wherein：

x_weatherRepresent the quantized value of weather conditions, its value according to weather conditions set, when weather conditions be thunderstorm gale, When hail, cyclone, local heavy showers, the bad weather of severe snow these one of them, 1 is set to；When weather conditions are non-severe During weather, 0 is set to；

x_hdayRepresent whether be festivals or holidays quantized value, its value according to whether be festivals or holidays setting, when the date be festivals or holidays When, it is set to 1；It is non-festivals or holidays when the date, is set to 0, wherein festivals or holidays includes weekend and official holiday；

x_yearThe quantized value in the time on date is represented, its value sets according to the selected benchmark time, the selected benchmark time 1 is set to, the quantized value in other times is equal to its time with the difference in benchmark time plus 1；

x_week1 year quantized value of inner circumferential number is represented, its value sets according to which week in 1 year, is set to 1 within first week, its The quantized value of Later Zhou Dynasty, one of the Five Dynasties's number is incremented by successively；

x_dayQuantized value one week seven days is represented, its value is set to 1 according to all several settings in a week, Monday, quantization thereafter Value is incremented by successively；

x_timeThe prediction time quantized value of one day is represented, its value sets according to predetermined period T, then had within one dayIt is individual pre- The moment is surveyed, wherein first quantized value of prediction time is 1, the quantized value of prediction time is incremented by successively thereafter, and last is pre- Survey the moment quantized value beWherein, the unit of T is minute；

K-th quantized value of the vehicle user number of prediction time region i in representing one day, its quantification manner is：

Wherein,I represents some region of mark,Represent that (k-1) in one day individual and kth Individual prediction time is all in region i vehicle numbers；(k-1) individual prediction time is represented in one day not in region i, but pre- k-th Survey vehicle number of the moment in region i；(k-1) individual prediction time is represented in one day in region i, but k-th prediction time Again not in the vehicle number of region i；

Specific calculating formula is as follows：

Wherein, g represents a certain gps data of vehicle user；G represents the gps data of all vehicle users；Region_i Represent statistical regions i；g_kRepresent k-th gps data of prediction time a certain vehicle user in a day.

The support vector regression module 4, is trained pre- using the multidimensional row vector being input into by data processing module 3 Survey, the forecast model for learning is expressed as follows：

Wherein, m is the sample number of training data；Its function expression is：Gaussian kernel function is selected in this example, its expression Formula is：κ(x_i, x)=exp (- g | | x_i-x||²), | | x_i-x||²Representation vector x_iEuclidean distance between vector x, setting Penalty constant C=80, Gauss nuclear parameter g=20 and interval ε=0.1 during training；To train the prediction for learning The weight parameter of model, b is the forecast model offset parameter that training learns.

With reference to Fig. 2, the present invention is comprised the following steps towards car networking regional traffic flow Forecasting Methodology：

Step 1：Initialization.

Determine benchmark time, predetermined period T and number of training m.The benchmark time of this example is set to, prediction week in 2015 Phase T=15 minutes, number of training

Step 2：Result and external action data module 1 and car networking data mould of the data processing module 3 according to initialization Data genaration current period moment and its preceding 959 moment in cycle totally 960 data at moment in cycle that block 2 is provided.

This step is implemented as follows：

2a) each cycle time data is quantified as follows：

The quantized value x of weather conditions 2a1) is set according to weather conditions_weather：If when weather conditions are thunderstorm gale, ice When hail, cyclone, local heavy showers, the bad weather of severe snow these one of them, if x_weatherIt is 1；If when weather conditions are During non-bad weather, if x_weatherIt is 0；

2a2) according to festivals or holidays setting whether be festivals or holidays quantized value x_hday：When the date being festivals or holidays, then x is set_hdayFor 1；It it is non-festivals or holidays when the date, if x_hdayIt is 0；

The quantized value x in time on date 2a3) is set according to the time_year：The quantized value x in selected benchmark time_yearIt is set to 1, The quantized value x in other times_yearDifference equal to its time and benchmark time adds 1；

1 year quantized value x of inner circumferential number 2a4) is set according to which week in 1 year_week：By the amount of first week in a year Change value x_week1 is set to, the quantized value x of its Later Zhou Dynasty, one of the Five Dynasties's number_weekIt is incremented by successively；

2a5) setting quantized value x one week seven days_day：By the quantized value x of Monday_day1 is set to, quantized value x thereafter_dayAccording to It is secondary to be incremented by；

The prediction time quantized value of one day 2a6) is set according to intraday which prediction time and predetermined period T x_time：By the quantized value x of intraday first prediction time_time1 is set to, thereafter the quantized value x of prediction time_timePass successively Increase, the quantized value x of last prediction time in a day_timeIt is set to 96；

2a7) calculate k-th vehicle user number of prediction time region i in a day

If k ∈ { 1,2,3 ..., 96 }, i represent some region of mark, g represents a certain gps data of vehicle user； G represents the gps data of all vehicle users；Region_i represents statistical regions i；g_kRepresent k-th prediction time in one day The gps data of one vehicle user.

(k-1) is individual and k-th prediction time is all designated as in region i vehicle numbers in one dayIts calculating formula is：

By (k-1) individual prediction time in one day not in region i, but k-th prediction time is designated as in the vehicle number of region iIts calculating formula is：

By (k-1) individual prediction time in 7 one days in region i, but k-th prediction time be not again in the vehicle number of region i It is designated asIts calculating formula is：

Vehicle user number scale by k-th prediction time region i in a day isAnd according toMeter CalculateNumerical value:

2b) according to step 2a) quantized result, the data to each moment in cycle carry out format specification treatment, i.e., By step 2a) in each moment in cycle corresponding x_weather、x_hday、x_year、x_week、x_day、x_timeWithTreatment forms multidimensional The form of row vector, is expressed as：

Step 3：Support vector regression module 4 learns prediction mould using the data training that data processing module 3 is generated Type.

3a) will in step 2 generate 960 data inputs at moment in cycle to support vector regression module 4, the module These data are carried out with [0,1] normalized；

3b) support vector regression module 4 utilizes step 3a) in data training after normalized learn prediction mould Type, the forecast model is：

Wherein, κ (x_i, it is x) gaussian kernel function, expression formula is：κ(x_i, x)=exp (- g | | x_i-x||²)；G is Gauss Nuclear parameter, g=20；||x_i-x||²Representation vector x_iEuclidean distance between vector x；For training learn it is pre- Survey the weight parameter of model；B is the forecast model offset parameter that training learns.

Step 4：The forecast model learnt using support vector regression module 4, during to current period in training data The next moment in cycle carved carries out regional traffic flow and is predicted, and completes the renewal at current period moment.

4a) learn the forecast model based on support vector regression for using training in step 3, with reference to the 961st cycle The vector data at moment, the result of the regional traffic flow at prediction the 961st moment in cycle of output, completed to the 961st moment in cycle Regional traffic flow prediction；

4b) when the historical juncture is turned at the 961st moment in cycle of prediction, the 961st moment in cycle was updated to current period Moment.

Step 5：Circulation performs step 2- steps 4, and the regional traffic flow at next moment in cycle is uninterruptedly predicted in completion.

Above description is only example of the present invention, does not constitute any limitation of the invention, it is clear that for For one of skill in the art, various modifications, equivalent, change etc. can also be made to the present invention, as long as these are converted Without departing from spirit of the invention, all should be within protection scope of the present invention.

Claims (10)

1. towards the regional traffic flow forecasting system of car networking, it is characterised in that including：

External action data module (1), weather conditions and every day of its record every day whether be festivals or holidays data message, For the external action data source as data processing module (3)；

Car networking data module (2), the gps data information of the vehicle user of all travelings in its record car networking, for conduct The internal influence data source of data processing module (3)；

Data processing module (3), for external action data and car networking number by being input into external action data module (1) The internal influence data being input into according to module (2) carry out numerical quantization treatment, produce the row vector of multidimensional and are input to supporting vector Regression machine module (4)；

Support vector regression module (4), it is pre- for being trained using the multidimensional row vector being input into by data processing module (3) Survey, learn forecast model, be predicted with the magnitude of traffic flow to the future period moment.

2. system according to claim 1, it is characterised in that the data of external action data module (1) record, its content At least include date, time, weather conditions, and will festivals or holidays be labeled as 1, non-festivals or holidays be labeled as 0.

3. system according to claim 1, it is characterised in that each row in car networking data module (2) record car networking The gps data of the vehicle user in sailing, its data form at least includes date, time and longitude and latitude.

4. system according to claim 1, it is characterised in that the multidimensional row vector that data processing module (3) is produced, represents For：Wherein：

x_weatherRepresent the quantized value of weather conditions, its value according to weather conditions set, when weather conditions be thunderstorm gale, hail, When cyclone, local heavy showers, the bad weather of severe snow these one of them, 1 is set to；When weather conditions are non-bad weather When, it is set to 0；

x_hdayRepresent whether be festivals or holidays quantized value, its value according to whether be festivals or holidays setting, when the date be festivals or holidays when, if It is 1；It is non-festivals or holidays when the date, is set to 0, wherein festivals or holidays includes weekend and official holiday；

x_yearThe quantized value in the time on date is represented, its value sets according to the selected benchmark time, the selected benchmark time is set to 1, The quantized value in other times is equal to its time with the difference in benchmark time plus 1；

x_week1 year quantized value of inner circumferential number is represented, its value sets according to which week in 1 year, is set to 1 within first week, its Later Zhou Dynasty, one of the Five Dynasties's number Quantized value it is incremented by successively；

x_dayRepresent quantized value one week seven days, its value is set to 1 according to all several settings in a week, Monday, quantized value thereafter according to It is secondary to be incremented by；

x_timeThe prediction time quantized value of one day is represented, its value sets according to predetermined period T, then had within one dayDuring individual prediction Carve, wherein first quantized value of prediction time is 1, the quantized value of prediction time is incremented by successively thereafter, during last prediction The quantized value at quarter isWherein, the unit of T is minute；

K-th quantized value of the vehicle user number of prediction time region i in representing one day, its quantification manner is：

$N_i^k=N_{i\_0}^k+N_{i\_in}^k-N_{i\_out}^k$

Wherein,I represents some region of mark,Represent (k-1) in one day it is individual and k-th it is pre- The moment is surveyed all in region i vehicle numbers；(k-1) individual prediction time is represented in one day not in region i, but during k-th prediction It is engraved in the vehicle number of region i；(k-1) individual prediction time is represented in one day in region i, but k-th prediction time is again not The vehicle number of i in region；

Specific calculating formula is as follows：

$N_{i\_0}^k=\underset{g\∈G}{\Σ}|(g_k\∈\mathrm{Re}gion\_i)\∩(g_{k-1}\∈\mathrm{Re}gion\_i)|$

$N_{i\_in}^k=\underset{g\∈G}{\Σ}|(g_k\∈\mathrm{Re}gion\_i)\∩(g_{k-1}\∉\mathrm{Re}gion\_i)|$

$N_{i\_out}^k=\underset{g\∈G}{\Σ}|(g_k\∉\mathrm{Re}gion\_i)\∩(g_{k-1}\∈\mathrm{Re}gion\_i)|$

Wherein, g represents a certain gps data of vehicle user；G represents the gps data of all vehicle users；Region_i is represented Statistical regions i；g_kRepresent k-th gps data of prediction time a certain vehicle user in a day.

5. system according to claim 1, it is characterised in that support vector regression module (4) is using by data processing module (3) the multidimensional row vector of input is trained prediction, and the forecast model for learning is expressed as follows：

$f\left(x\right)=\underset{i=1}{\overset{m}{\Σ}}({\mathrm{\α}}_i^{*}-{\mathrm{\α}}_i)\mathrm{\κ}(x_i,x)+b,$

Wherein, m is the sample number of training data；κ(x_i, x)=φ (x_i)^Tφ (x) is kernel function；For training learns Forecast model weight parameter, b is that training learns the forecast model offset parameter that.

6. regional traffic flow Forecasting Methodology is carried out using system described in claim 1, it is characterised in that comprise the following steps：

1) initialize：Determine benchmark time, the cycle T of prediction and number of training m；

2) result and external action data module (1) and car networking data module (2) of the data processing module (3) according to initialization The data at the data genaration current period moment of offer and its preceding at moment in m-1 cycle at common moment in m cycle；

3) support vector regression module (4) learns forecast model using the data training of data processing module (3) generation, profit The magnitude of traffic flow at output m+1 moment in cycle is predicted with the forecast model；

4) when the historical juncture is turned at the m+1 moment in cycle of prediction, the m+1 moment in cycle is updated to the current period moment；

5) circulation performs step 2) -4), the regional traffic flow at next moment in cycle is uninterruptedly predicted in completion.

7. method according to claim 6, wherein step 2) in the m data at moment in cycle of generation, enter as follows OK:

2a) each cycle time data is quantified as follows：

The quantized value x of weather conditions is set according to weather conditions_weather：When weather conditions are thunderstorm gale, hail, cyclone, office When portion's heavy showers, the bad weather of severe snow these one of them, if x_weatherIt is 1；When weather conditions are non-bad weather, if x_weatherIt is 0；

According to festivals or holidays setting whether be festivals or holidays quantized value x_hday：When the date being festivals or holidays, if x_hdayIt is 1；It is when the date Non- festivals or holidays, if x_hdayIt is 0；

The quantized value x in time on date is set according to the time_year：The quantized value x in selected benchmark time_year1 is set to, other times Quantized value x_yearDifference equal to its time and benchmark time adds 1；

1 year quantized value x of inner circumferential number is set according to which week in 1 year_week：By the quantized value x of first week in a year_week 1 is set to, the quantized value x of its Later Zhou Dynasty, one of the Five Dynasties's number_weekIt is incremented by successively；

Setting quantized value x one week seven days_day：By the quantized value x of Monday_day1 is set to, quantized value x thereafter_dayIt is incremented by successively；

The prediction time quantized value x of one day is set according to intraday which prediction time and predetermined period T_time：By in one day First prediction time quantized value x_time1 is set to, thereafter the quantized value x of prediction time_timeIt is incremented by successively, it is last in one day One quantized value x of prediction time_timeIt is set to

IfI represents some region of mark,

(k-1) is individual and k-th prediction time is all designated as in region i vehicle numbers in one day

By (k-1) individual prediction time in one day not in region i, but k-th prediction time is designated as in the vehicle number of region i

By (k-1) individual prediction time in one day in region i, but k-th prediction time is not designated as in the vehicle number of region i again

Vehicle user number scale by k-th prediction time region i in a day isAnd according toCalculateNumerical value:

2b) according to step 2a) quantized result, the data to each moment in cycle carry out format specification treatment, Ji Jiangbu Rapid 2a) in each moment in cycle corresponding x_weather、x_hday、x_year、x_week、x_day、x_timeWithTreatment formed multidimensional row to The form of amount, is expressed as：

8. (k-1) is individual and k-th prediction time is all in region i vehicle numbers in a day according to claim 7 Its calculating formula is：

$N_{i\_0}^k=\underset{g\∈G}{\Σ}|(g_k\∈\mathrm{Re}gion\_i)\∩(g_{k-1}\∈\mathrm{Re}gion\_i)|$

Wherein, g represents a certain gps data of vehicle user；G represents the gps data of all vehicle users；Region_i is represented Statistical regions i；g_kRepresent k-th gps data of prediction time a certain vehicle user in a day.

9. but k-th prediction time exists or not region i (k-1) individual prediction time in a day according to claim 7 The vehicle number of region iIts calculating formula is：

$N_{i\_in}^k=\underset{g\∈G}{\Σ}|(g_k\∈\mathrm{Re}gion\_i)\∩(g_{k-1}\∉\mathrm{Re}gion\_i)|$

Wherein, g represents a certain gps data of vehicle user；G represents the gps data of all vehicle users；Region_i is represented Statistical regions i；g_kRepresent k-th gps data of prediction time a certain vehicle user in a day.

10. but k-th prediction time is again in region i (k-1) individual prediction time in a day according to claim 7 Not in the vehicle number of region iIts calculating formula is：

$N_{i\_out}^k=\underset{g\∈G}{\Σ}|(g_k\∉\mathrm{Re}gion\_i)\∩(g_{k-1}\∈\mathrm{Re}gion\_i)|$

Wherein, g represents a certain gps data of vehicle user；G represents the gps data of all vehicle users；Region_i is represented Statistical regions i；g_kRepresent k-th gps data of prediction time a certain vehicle user in a day.