CN104064023B - A kind of Dynamic Traffic Flow Prediction method based on space time correlation - Google Patents

A kind of Dynamic Traffic Flow Prediction method based on space time correlation Download PDF

Info

Publication number
CN104064023B
CN104064023B CN201410272800.9A CN201410272800A CN104064023B CN 104064023 B CN104064023 B CN 104064023B CN 201410272800 A CN201410272800 A CN 201410272800A CN 104064023 B CN104064023 B CN 104064023B
Authority
CN
China
Prior art keywords
data
traffic flow
neighbour
time
matrix
Prior art date
Application number
CN201410272800.9A
Other languages
Chinese (zh)
Other versions
CN104064023A (en
Inventor
李建元
李丹
陈涛
倪升华
王浩
Original Assignee
银江股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 银江股份有限公司 filed Critical 银江股份有限公司
Priority to CN201410272800.9A priority Critical patent/CN104064023B/en
Publication of CN104064023A publication Critical patent/CN104064023A/en
Application granted granted Critical
Publication of CN104064023B publication Critical patent/CN104064023B/en

Links

Abstract

The present invention relates to intelligent transportation field, particularly relate to a kind of Dynamic Traffic Flow Prediction method based on space time correlation, the method sets up space-time matrix after traffic flow data is carried out pretreatment, by neighbour's local linear reconstructing method, space-time matrix is trained, find one group of neighbour and the weights of prediction, it is predicted after non-negative correction, finally by predictive value, space-time matrix is updated.The beneficial effects of the present invention is: 1, the suitability is strong, can be suitably used for any microwave detection section;2, feasibility is strong, it is only necessary to given historical traffic flows data base, just can be trained data and predict;3, calculating speed fast, complexity is relatively low, and the calculating time is second level;4, precision of prediction is high, eliminates randomness and the undulatory property of dynamic data, improves the accuracy and reliability predicted the outcome;5, predictive efficiency is high, can realize the multistep forecasting traffic flow of multiple 5 minutes sections, can accomplish efficient the most in short-term with forecasting traffic flow time long.

Description

A kind of Dynamic Traffic Flow Prediction method based on space time correlation
Technical field
The present invention relates to intelligent transportation field, particularly relate to a kind of Dynamic Traffic Flow Prediction method based on space time correlation.
Background technology
Along with development and the economic growth of society, urban traffic jam is the most serious.For Effective Regulation traffic Flow, optimizes the service efficiency of road, and intelligent transportation system becomes emphasis of concern, and along with the most progressively intelligence of research Energyization, mobilism and informationization.As the important component part of intelligent transportation system, vehicle guidance system has become traffic administration The effective way of road traffic is dredged by department, and its key technology is the prediction to road traffic condition, the most effectively utilizes history Traffic data and real time traffic data carry out dynamic prediction to the traffic flow of future time instance road.Traffic flow forecasting mainly wraps Include two parts: set up traffic flow histories data base and build forecast model.By historical data base is carried out pretreatment also Data are trained drawing and predict the outcome by the algorithm utilizing forecast model to provide.In terms of the two, it was predicted that model algorithm is The most insoluble key component, its direct relation the quality predicted the outcome, and is emphasis and the difficulty of traffic flow forecasting Point.
At present the method to traffic flow forecasting mainly has history averaging method, neural network model, support vector regression, Little square law, time series method etc..Said method is simple to operate, convenience of calculation, is suitable for the prediction of relatively rule data, but for Road model is more complicated and the unstable irregular fluctuation of data stream caused of traffic flow, it was predicted that accuracy rate is relatively low.It addition, mostly Counting method is all to predict according to temporal associativity, have ignored traffic flow relatedness on section, space so that prediction Result is the most accurate.Therefore, in order to overcome random fluctuation and the unstable prediction effect that brings of fluctuation of dynamic data and examine Consider the impact in space, section, need to introduce new method and traffic flow is predicted on space time correlation.
Summary of the invention
The present invention is to overcome above-mentioned weak point, it is therefore intended that provide a kind of have high accuracy, reliability based on The Dynamic Traffic Flow Prediction method of space time correlation.
The present invention is to reach above-mentioned purpose by the following technical programs: a kind of Dynamic Traffic Flow Prediction based on space time correlation Method, including:
1) pass through layout data acquisition equipment on microwave section to gather historical traffic flows data, predict and work as day interval Time is the traffic flow data of T;
2) traffic flow data carries out pretreatment, cleans extraneous data according to arranging threshold value, and to missing data number According to interpolation;
3) with data acquisition time as the longitudinal axis, microwave section is transverse axis, builds traffic flow data space-time matrix;
4) carry out traffic flow data space-time matrix training based on the reconstruct of neighbour's local linear, find the weights square of prediction Battle array and test sample neighbour;
5) weight matrix is carried out non-negative correction, set up the weight matrix that neighbour is the most just weighting;
6) traffic flow forecasting is carried out according to weight matrix and test sample neighbour, it was predicted that formula is:
Wherein: XijFor test sample XiJth neighbour, wijFor sample XiThe weights of jth neighbour, K ' is long for weight matrix Degree, K ' is less than test sample neighbour's number;
7) traffic flow data of prediction is contrasted with truthful data, obtain forecast error;
8) predictive value adding traffic flow data space-time matrix, and remove the flow value in moment the earliest, combination is formed new Traffic flow data space-time matrix, repeats step 4).
As preferably, described traffic flow data pretreatment comprises the following steps:
2.1) data cleansing: T minute inside lane flow data more than 300 replace with this time point data on flows in history Meansigma methods;
2.2) missing values interpolation: for the disappearance of T minutes groove data on certain microwave section, utilize linear programming side Method:
min||x||1St. Ax=y (2)
Wherein, the list of matrix A shows that each microwave link flow in the most complete T minutes groove, vector y represent There is the non-lack part of each link flow of T time groove of disappearance, x represent in A each column vector coefficient when linear reconstruct y to Amount, solving the x obtained is sparse vector.
As preferably, described step 4) in traffic flow data space-time matrix training based on the reconstruct of neighbour's local linear, tool Body comprises the following steps:
4.1) with predicted time point traffic flow data as test sample, moment and historical time before this time point Traffic flow data is training sample, based on the data of different microwave sections, calculates training sample and closes at space-time with test sample Euclidean distance under Lian, finds out K the minimum training sample of distance K the neighbour as test sample;Europe between two vectors Family name is apart from as follows:
D = Σ i = 1 n ( x i - y i ) 2 - - - ( 3 )
Wherein n is vector length, XiFor test sample, xiAnd yiIt is respectively the element in vector X and Y;
4.2) weight matrix is calculated:
Set up error minimize function:
Wherein, XiFor test sample, N is total sample number, XijFor test sample XiJth neighbour, neighbour's sum is K, wijFor Sample XiThe weights of jth neighbour.The local covariance matrix of sample is:
Cjk(i)=(Xi-Xij)T(Xi-Xik) (5),
It is the matrix of a K x K, it is carried out Regularization and obtains
Cjk(i)=Cjk(i)+rI (6),
Wherein I is the unit matrix of K x K, and r is regularization coefficient;Can get weights according to local covariance matrix:
w i j = Σ k = 1 K ( C j k ( i ) ) - 1 Σ j = 1 K Σ k = 1 K ( C j k ( i ) ) - 1 - - - ( 7 )
As preferably, described interval time, T was 5 minutes.
The beneficial effects of the present invention is: 1, the suitability is strong, acquire the traffic flow data of multiple different microwave point, energy It is applicable to any microwave detection section;2, feasibility is strong, it is only necessary to given historical traffic flows data base, just can carry out data Training and prediction;3, calculating speed fast, the method complexity is relatively low, and for hundreds of thousands data, the calculating time is second level;4, pre- Survey precision is high, and the inventive method eliminates randomness and the undulatory property of dynamic data, reduces prediction data error, improves prediction The accuracy of result and reliability;5, predictive efficiency is high, and the inventive method can realize the multistep traffic flow of multiple 5 minutes sections Prediction, can accomplish efficient the most in short-term with forecasting traffic flow time long, and precision of prediction kept stable.
Accompanying drawing explanation
Fig. 1 is the flow chart of steps of the present invention;
Fig. 2 is that the inventive method and the history averaging method volume forecasting result at [2013-9-5] 10:45 on the same day is with true The fitting effect contrast schematic diagram of value;
Fig. 3 is that the inventive method and the history averaging method volume forecasting result at [2013-9-5] 10:50 on the same day is with true The fitting effect contrast schematic diagram of value;
Fig. 4 is that the inventive method and the history averaging method volume forecasting result at [2013-9-5] 10:55 on the same day is with true The fitting effect contrast schematic diagram of value;
Fig. 5 is that the inventive method and the history averaging method volume forecasting result at [2013-9-5] 11:00 on the same day is with true The fitting effect contrast schematic diagram of value;
Fig. 6 be the inventive method and history averaging method in 45 minutes grooves based on predictive value on the basis of predict again Error schematic diagram.
Detailed description of the invention
Below in conjunction with specific embodiment, the present invention is described further, but protection scope of the present invention is not limited in This:
Embodiment 1: as it is shown in figure 1, a kind of Dynamic Traffic Flow Prediction method based on space time correlation comprises the following steps:
Step 1: gather historical traffic flows data and multiple microwave road section traffic volume flow data of prediction Time of Day point, its Middle data are the Short-Term Traffic Flow data of every 5 minutes.
Step 2: traffic flow microwave data pretreatment.
2.1) data cleansing.The data caused due to data acquisition unit fault or other reasons are abnormal, need it Being carried out, cleaning rule is as follows:
Think that these data are abnormal when 5 minutes inside lane flows are more than 300, these data are replaced with in history this time Between put the meansigma methods of data on flows.
2.2) missing values interpolation.The shortage of data caused due to factors such as communication equipment faults, needs to carry out missing values Interpolation, interpolation rule is as follows:
For the disappearance of certain 5 minutes groove data of some section, linear programming method is utilized to train other times This disappearance section of point and the linear relationship in other sections, lack the data of section time point further according to the coefficient reconstruct trained. Linear programming is for solving following problems:
min||x||1St. Ax=y (2)
Wherein, the list of matrix A shows that each link flow in 5 the most complete minutes grooves, vector y represent existence The non-lack part of each link flow of certain time slot of disappearance, x represents each column vector coefficient vector when linear reconstruct y in A. According to formula (2), solving the x obtained is sparse vector.
Step 3: build traffic flow data space-time matrix.By pretreated data with the 5 minutes grooves of continuous 30 days For the time longitudinal axis, all microwave sections are space transverse axis, build the space-time matrix of traffic flow microwave data.
Step 4: space-time matrix training based on the reconstruct of neighbour's local linear.
4.1) neighbour is found.With certain time point traffic flow data as test sample, before this time point the moment and The data of historical time are training sample, based on the data of different microwave sections, calculate training sample and test sample time Euclidean distance under null Context, finds out K the minimum training sample of distance K the neighbour as test sample.Between two vectors Euclidean distance as follows:
D = Σ i = 1 n ( x i - y i ) 2 - - - ( 3 )
Wherein n is vector length, XiFor test sample, xiAnd yiIt is respectively the element in vector X and Y.By trying to achieve test Sample and the distance of each time point of training sample, and be ranked up by distance size, K sample of distance minimum is tested exactly The neighbour of sample.
4.2) weight matrix W is calculated.
Set up error minimize function:
Wherein, XiFor test sample, N is total sample number, XijFor test sample XiJth neighbour, neighbour's sum is K, wijFor Sample XiThe weights of jth neighbour.The local covariance matrix of sample is:
Cjk(i)=(Xi-Xij)T(Xi-Xik) (5),
It is the matrix of a K x K, it is carried out Regularization and obtains
Cjk(i)=Cjk(i)+rI (6),
Wherein I is the unit matrix of K x K, and r is regularization coefficient;Can get weights according to local covariance matrix:
w i j = Σ k = 1 K ( C j k ( i ) ) - 1 Σ j = 1 K Σ k = 1 K ( C j k ( i ) ) - 1 - - - ( 7 )
Step 5: weight matrix is carried out non-negative correction, sets up the weight matrix that neighbour is the most just weighting.
i f w i j < 0 , t h e n w i j = 0 s t . &Sigma; j = 1 K &prime; w i j - - - ( 8 )
To weight matrix W being negative item, more accurate for making to predict the outcome, W will be set to 0 for negative item, then carry out Normalization, the weight matrix just weighted of so available training set, the length of weight matrix will become K ', K '≤K.
Step 6: forecasting traffic flow.Predictive value is calculated, it was predicted that be worth for test sample time point according to weight matrix and neighbour The traffic flow of next 5 minutes, computing formula is as follows:
Y i = &Sigma; j = 1 K &prime; w i j X i j - - - ( 1 )
Wherein XijIt is the weights corresponding in the neighbour items more than 0.
Step 7: calculating predictive value and the error of actual value:
εi=Σ | Yi-Zi|/ΣYi (9)
Wherein, ZiIt it is the actual value in predictive value correspondence moment.
Step 8: space-time matrix incremental update.Predictive value is added space-time matrix, and removes the flow value in moment the earliest, group Close and form new space-time matrix, repeat said process, carry out multi-step prediction.
The a plurality of section of microwave point is set in city, Hangzhou as acquisition target, with [2013-8-7] to [2013-9-5] even In continuous 30 days, 10:00 to 11:00 is sampling periods every day, adds up every 5 minutes and passes through the vehicle number of microwave point on section, as friendship Through-current capacity source data.When training, half an hour and this time point and above half in 30 days in history before predicted time point Hour data be predicted.With the data of 10:00 to 10:40 as training sample set, 2013-9-5 10:45 to 11:00 on the same day Data be test sample collection, carry out 2013-9-5 10:45 to 11:00 on the same day according to the model of traffic flux forecast that obtains of training Volume forecasting.
Before prediction, it is necessary first to traffic flow data is carried out pretreatment:
Data cleansing.The traffic flow data collected is carried out according to above described cleaning rule, cleans threshold Value is 300, will be greater than the flow of 300 and replaces with the meansigma methods of this time point data on flows in history.
Missing values interpolation.To in the data collected, for the disappearance of certain 5 minutes groove data of some section, profit Train the linear relationship in this disappearance section of other times point and other sections with linear programming method, further according to train be The data of number reconstruct disappearance section time point.Linear programming can realize with the linprog function of matlab.
The traffic flow data that pretreatment is good is utilized to build traffic flow data space-time matrix.With [2013-9-5] 10:35 on the same day Traffic flow data be test sample, [2013-8-7] to [2013-9-4] in continuous 29 days 10:00 to 10:30 and The data of [2013-9-5] 10:00 to 10:25 on the same day are training sample set, i.e. have 209 time point data, and each time point has 330 section microwave points, i.e. have 330 data, then test sample is the array of 330*1, and training sample is the space-time of 330*209 Matrix.
Traffic flow data space-time matrix is carried out based on the reconstruct training of neighbour's local linear.
First look for neighbour.Calculate test sample and the Euclidean distance of 209 training samples in space-time matrix, find out K The training sample of distance minimum is as K neighbour of test sample, and in the present invention, K takes 20.In like manner, [2013-9-5] same day 10: When the traffic flow data of 35 and 10:40 is test sample with the testing time put front 209 time points data on flows for training Collection.
Secondly weight matrix W is calculated.Set up error minimize function:
min | X - &Sigma; j = 1 K w j X j | 2 , s t . &Sigma; j = 1 K w j - - - ( 4 )
Wherein, wherein, XiFor test sample, N is total sample number, XijFor test sample XiJth neighbour, neighbour sum For K=20, wijFor sample XiThe weights of jth neighbour.The local covariance matrix of sample is:
Cjk(i)=(Xi-Xij)T(Xi-Xik) (5),
It is the matrix of a K x K, it is carried out Regularization and obtains
Cjk(i)=Cjk(i)+rI (6),
Wherein I is the unit matrix of K x K, and r is regularization coefficient, and in the present invention, r value is that the mark of C is multiplied by 0.001, asks mark to transport The trace function calculating available Matlab is tried to achieve;Can get weights according to local covariance matrix:
w j = &Sigma; k = 1 K ( C j k ) - 1 &Sigma; m = 1 K &Sigma; n = 1 K ( C m n ) - 1 - - - ( 7 )
Step 5: weight matrix is carried out non-negative correction, sets up the weight matrix that neighbour is the most just weighting:
i f w j < 0 , w j = 0 s t . &Sigma; j = 1 K &prime; w j - - - ( 8 )
Each weight, after non-negative correction, meets wj> > 0 andWherein k ' is that non-negative is revised just Weights number.
Predictive value is calculated, it was predicted that be worth for test sample according to the revised weight matrix of non-negative and 20 test sample neighbours The time point traffic flow of next 5 minutes, i.e. with the training result of [2013-9-5] 10:30 on the same day predict [2013-9-5] when The traffic flow of it 10:35.Computing formula is as follows:
Y = &Sigma; j = 1 K &prime; w j X j - - - ( 1 )
Wherein XijIt is the weights corresponding in the neighbour items more than 0.
Calculate predictive value and the error of actual value: εi=Σ | Yi-Zi|/ΣYi (9)
Wherein, ZiIt it is the actual value in predictive value correspondence moment.
After completing above-mentioned prediction, in order to reduce data random error, increase forecasting accuracy, to traffic flow data space-time square Battle array increment is updated.By the predictive value of [2013-9-5] 10:35 on the same day and this moment traffic flow data addition in first 29 days Space-time matrix, and the flow value of 10:00 in removing 30 days, combination forms new space-time matrix, repeats said process, it was predicted that The traffic flow of [2013-9-5] 10:40 on the same day, the most repeatedly, carries out multi-step prediction.
As shown in Figures 2 to 5, the flow of prediction [2013-9-5] same day 10:45,10:50,10:55,11:00 it is respectively Predict the outcome the fitting effect contrast schematic diagram with actual value, in figure, on fitting result chart 330 section microwave points is divided into 3 Part, (a) figure is the volume forecasting fitting result chart of front 1 to No. 110 microwaves point, and (b) figure is the plan of 111 to No. 220 microwave points Closing design sketch, (c) figure is the fitting result chart of 221 to No. 330 microwave points.It will be seen that it is bigger for traffic flow fluctuation Section, the inventive method can obtain predicting the outcome more accurately.
As shown in Figure 6, the inventive method is compared with history averaging method, show that history averaging method precision of prediction is 78%, the inventive method precision of prediction reaches 85%, substantially increases the accuracy of prediction, meets the precision of traffic flow forecasting Demand.It addition, for the multi-step prediction on fundamentals of forecasting, the inventive method precision of prediction is maintained at about 83%, it was predicted that precision The highest with stability.
It is the specific embodiment of the present invention and the know-why used described in Yi Shang, if conception under this invention institute Make change, function produced by it still without departing from description and accompanying drawing contained spiritual time, must belong to the present invention's Protection domain.

Claims (4)

1. a Dynamic Traffic Flow Prediction method based on space time correlation, it is characterised in that including:
1) pass through layout data acquisition equipment on microwave section to gather historical traffic flows data, predict interval time on the same day Traffic flow data for T;
2) traffic flow data carries out pretreatment, cleans extraneous data according to arranging threshold value, and missing data is carried out data benefit Insert;
3) with data acquisition time as the longitudinal axis, microwave section is transverse axis, builds traffic flow data space-time matrix;
4) traffic flow data space-time matrix training based on the reconstruct of neighbour's local linear is carried out, by calculating the Euclidean of space-time matrix Distance finds test sample neighbour, minimizes error function by foundation and finds the weight matrix of prediction;
5) weight matrix is carried out non-negative correction, set up the weight matrix that neighbour is the most just weighting;
6) traffic flow forecasting is carried out according to weight matrix and test sample neighbour, it was predicted that formula is:
Wherein: XijFor test sample XiJth neighbour, wijFor sample XiThe weights of jth neighbour, K ' is long for weight matrix Degree, K ' is less than test sample neighbour's number;
7) traffic flow data of prediction is contrasted with truthful data, obtain forecast error;
8) predictive value adding traffic flow data space-time matrix, and remove the flow value in moment the earliest, combination forms new traffic Flow data space-time matrix, repeats step 4).
A kind of Dynamic Traffic Flow Prediction method based on space time correlation the most according to claim 1, it is characterised in that described Traffic flow data pretreatment comprises the following steps:
2.1) data cleansing: T minute inside lane flow data more than 300 replace with the flat of this time point data on flows in history Average;
2.2) missing values interpolation: for the disappearance of T minutes groove data on certain microwave section, utilizes linear programming method:
min||x||1St. Ax=y (2)
Wherein, the list of matrix A shows that each microwave link flow in the most complete T minutes groove, vector y represent existence The non-lack part of each link flow of T time groove of disappearance, x represents each column vector coefficient vector when linear reconstruct y in A, Solving the x obtained is sparse vector.
A kind of Dynamic Traffic Flow Prediction method based on space time correlation the most according to claim 1, it is characterised in that described Step 4) in traffic flow data space-time matrix training based on the reconstruct of neighbour's local linear, specifically include following steps:
4.1) with predicted time point traffic flow data as test sample, moment and the traffic of historical time before this time point Flow data is training sample, based on the data of different microwave sections, calculates training sample and test sample under space time correlation Euclidean distance, find out K the minimum training sample of distance K the neighbour as test sample;Euclidean between two vectors away from From as follows:
D = &Sigma; i = 1 n ( x i - y i ) 2 - - - ( 3 )
Wherein n is vector length, XiFor test sample, xiAnd yiIt is respectively the element in vector X and Y;
4.2) weight matrix is calculated:
Set up error minimize function:
Wherein, XiFor test sample, N is total sample number, XijFor test sample XiJth neighbour, neighbour's sum is K, wijFor Sample XiThe weights of jth neighbour;The local covariance matrix of sample is:
Cjk(i)=(Xi-Xij)T(Xi-Xik) (5),
It is the matrix of a KxK, it is carried out Regularization and obtains
Cjk(i)=Cjk(i)+rI (6),
Wherein I is the unit matrix of KxK, and r is regularization coefficient;Can get weights according to local covariance matrix:
w i j = &Sigma; k = 1 K ( C j k ( i ) ) - 1 &Sigma; j = 1 K &Sigma; k = 1 K ( C j k ( i ) ) - 1 - - - ( 7 )
Wherein, wijFor sample XiThe weights of jth neighbour, K is neighbour's sum, and k represents the kth neighbour in K neighbour.
A kind of Dynamic Traffic Flow Prediction method based on space time correlation the most according to claim 1, it is characterised in that described Interval time, T was 5 minutes.
CN201410272800.9A 2014-06-18 2014-06-18 A kind of Dynamic Traffic Flow Prediction method based on space time correlation CN104064023B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410272800.9A CN104064023B (en) 2014-06-18 2014-06-18 A kind of Dynamic Traffic Flow Prediction method based on space time correlation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410272800.9A CN104064023B (en) 2014-06-18 2014-06-18 A kind of Dynamic Traffic Flow Prediction method based on space time correlation

Publications (2)

Publication Number Publication Date
CN104064023A CN104064023A (en) 2014-09-24
CN104064023B true CN104064023B (en) 2016-12-07

Family

ID=51551709

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410272800.9A CN104064023B (en) 2014-06-18 2014-06-18 A kind of Dynamic Traffic Flow Prediction method based on space time correlation

Country Status (1)

Country Link
CN (1) CN104064023B (en)

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105006147B (en) * 2015-06-19 2017-03-15 武汉大学 A kind of Link Travel Time estimating method based on road spatial and temporal association
CN105046956B (en) * 2015-06-24 2017-04-26 银江股份有限公司 Traffic flow simulating and predicting method based on turning probability
CN105632193B (en) * 2015-12-25 2017-12-22 银江股份有限公司 A kind of shortage of data section speed calculation method based on space-time relationship
CN105679021B (en) * 2016-02-02 2018-11-06 招商局重庆交通科研设计院有限公司 Journey time fusion forecasting and querying method based on traffic big data
CN106781457B (en) * 2016-11-29 2019-04-30 东南大学 A kind of freeway traffic flow parameter correction method based on multi-source fusion data
CN107944605A (en) * 2017-11-10 2018-04-20 河海大学常州校区 A kind of dynamic traffic paths planning method based on data prediction
CN108492568B (en) * 2018-04-25 2020-06-12 南京邮电大学 Short-term traffic flow prediction method based on time-space characteristic analysis
CN110517479B (en) * 2018-05-22 2020-11-03 杭州海康威视系统技术有限公司 Urban road traffic prediction method and device and electronic equipment
CN109272169A (en) * 2018-10-10 2019-01-25 深圳市赛为智能股份有限公司 Traffic flow forecasting method, device, computer equipment and storage medium
CN109979195B (en) * 2019-03-22 2020-07-03 浙江大学城市学院 Sparse regression-based short-term traffic flow prediction method integrating space-time factors
CN110113226B (en) * 2019-04-16 2021-03-12 新华三信息安全技术有限公司 Method and device for detecting equipment abnormity
CN110474815A (en) * 2019-09-23 2019-11-19 北京达佳互联信息技术有限公司 Bandwidth prediction method, apparatus, electronic equipment and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103247177A (en) * 2013-05-21 2013-08-14 清华大学 Large-scale road network traffic flow real-time dynamic prediction system
CN103700255A (en) * 2013-12-30 2014-04-02 复旦大学 Time and space related data mining-based traffic flow prediction method

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080033630A1 (en) * 2006-07-26 2008-02-07 Eun-Mi Lee System and method of predicting traffic speed based on speed of neighboring link

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103247177A (en) * 2013-05-21 2013-08-14 清华大学 Large-scale road network traffic flow real-time dynamic prediction system
CN103700255A (en) * 2013-12-30 2014-04-02 复旦大学 Time and space related data mining-based traffic flow prediction method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于K近邻非参数回归的短时交通流预测方法;张涛等;《系统工程理论与实践》;20100228;第30卷(第2期);第376-384页 *

Also Published As

Publication number Publication date
CN104064023A (en) 2014-09-24

Similar Documents

Publication Publication Date Title
Laptev et al. Time-series extreme event forecasting with neural networks at uber
CN105389980B (en) Short-time Traffic Flow Forecasting Methods based on long short-term memory recurrent neural network
Yuan et al. Wind power prediction using hybrid autoregressive fractionally integrated moving average and least square support vector machine
Pan et al. Remaining useful life estimation using an inverse Gaussian degradation model
Dehghani et al. Uncertainty analysis of streamflow drought forecast using artificial neural networks and Monte‐Carlo simulation
Wu et al. Evaluating uncertainty estimates in distributed hydrological modeling for the Wenjing River watershed in China by GLUE, SUFI-2, and ParaSol methods
Zhao et al. Research and application based on the swarm intelligence algorithm and artificial intelligence for wind farm decision system
Bhardwaj et al. Estimation of solar radiation using a combination of Hidden Markov Model and generalized Fuzzy model
Li et al. A hybrid model based on synchronous optimisation for multi-step short-term wind speed forecasting
Mukerji et al. Flood forecasting using ANN, neuro-fuzzy, and neuro-GA models
Du et al. A novel hybrid model for short-term wind power forecasting
De Giorgi et al. Error analysis of short term wind power prediction models
CN103413443B (en) Short-term traffic flow forecasting method based on hidden Markov model
CN102496069B (en) Cable multimode safe operation evaluation method based on fuzzy analytic hierarchy process (FAHP)
Wang et al. A novel hybrid system based on a new proposed algorithm—Multi-Objective Whale Optimization Algorithm for wind speed forecasting
Li et al. Novel analysis–forecast system based on multi-objective optimization for air quality index
Ravansalar et al. Wavelet-linear genetic programming: a new approach for modeling monthly streamflow
Feng et al. A state-space-based prognostic model for hidden and age-dependent nonlinear degradation process
Giustolisi et al. Advances in data-driven analyses and modelling using EPR-MOGA
Kişi River suspended sediment concentration modeling using a neural differential evolution approach
Agrawal et al. Long term load forecasting with hourly predictions based on long-short-term-memory networks
Mellit et al. An ANFIS-based forecasting for solar radiation data from sunshine duration and ambient temperature
Guo et al. A new wind speed forecasting strategy based on the chaotic time series modelling technique and the Apriori algorithm
Jiang et al. Short-term wind speed forecasting using a hybrid model
Mirzaei et al. Comparison of interpolation methods for the estimation of groundwater contamination in Andimeshk-Shush Plain, Southwest of Iran

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant