CN105679021B - Journey time fusion forecasting and querying method based on traffic big data - Google Patents

Journey time fusion forecasting and querying method based on traffic big data Download PDF

Info

Publication number
CN105679021B
CN105679021B CN201610075689.3A CN201610075689A CN105679021B CN 105679021 B CN105679021 B CN 105679021B CN 201610075689 A CN201610075689 A CN 201610075689A CN 105679021 B CN105679021 B CN 105679021B
Authority
CN
China
Prior art keywords
time
model
sequence
data
section
Prior art date
Application number
CN201610075689.3A
Other languages
Chinese (zh)
Other versions
CN105679021A (en
Inventor
付建胜
王川久
熊正荣
谯志
王少飞
Original Assignee
招商局重庆交通科研设计院有限公司
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 招商局重庆交通科研设计院有限公司 filed Critical 招商局重庆交通科研设计院有限公司
Priority to CN201610075689.3A priority Critical patent/CN105679021B/en
Publication of CN105679021A publication Critical patent/CN105679021A/en
Application granted granted Critical
Publication of CN105679021B publication Critical patent/CN105679021B/en

Links

Abstract

The present invention is based on the journey time fusion forecasting methods and querying method of traffic big data, off-line calculation or training are carried out to the data that all online vehicles upload obtaining all kinds of prediction models and parameter, being established according to all kinds of prediction models and parameter and dynamically updating the data dictionary;All kinds of prediction models and parameter are called, and combines the travel time data of real-time section and path, the traffic behavior of road network or bicycle is predicted;The data dictionary includes vehicle data dictionary, section data dictionary and path data dictionary;The online vehicle refers to login network access and uploads the vehicle of positioning and speed data automatically.The method have the benefit that overcoming the problems such as real-time existing for the traffic status prediction method of prior art road network or bicycle is poor, versatility is poor and practicability is not strong, survey precision greatly improved, the speed that on-line prediction greatly improved simultaneously, ensure that real-time, practicability and the versatility in engineer application.

Description

Journey time fusion forecasting and querying method based on traffic big data

Technical field

The present invention relates to the traffic status prediction technologies of road network or bicycle, are related specifically to a kind of based on traffic big data Journey time fusion forecasting and querying method.

Background technology

Existing road network or bicycle traffic status prediction method mainly utilize the traffic number that Floating Car or roadside device provide According to, or even the traffic data of roadside device offer is only relied on to predict the traffic behavior of road network or bicycle.Such methods are mainly led to It crosses roadside device and directly acquires the traffic informations such as road network flow, average speed, or by carrying out secondary add to floating car data Work obtains certain class traffic information, and is based on road network topology, is predicted using model recursion mode, the traffic information master of prediction Traffic three elements are concentrated on, cause popularization face wideless, promotional value is not also high, that is, there is general sex chromosome mosaicism, this expands to business Exhibition causes larger puzzlement.Also, existing method usually from up time to go consider floating car data influence, it will usually neglect floating The time factor of motor-car data, and linear mathematical computations mode is used, it such as sums, ask impartial operation, departing from road grid traffic Primitive character, cause the extraction accuracy of traffic information and stability not high, practicability is not also strong.It is especially multiple in face of structure When miscellaneous magnanimity floating car data, existing road network or bicycle traffic status prediction method usually seem helpless.Obviously, existing Some road networks or bicycle traffic status prediction technology there is real-times it is poor, versatility is poor and practicability is not strong the problems such as.

Invention content

To solve, real-time is poor existing for existing road network or bicycle traffic status prediction technology, versatility is poor and real The problems such as not strong with property, the present invention propose a kind of journey time fusion forecasting and querying method based on traffic big data.

The present invention is based on the journey time fusion forecasting method of traffic big data, data that all online vehicles are uploaded into Row off-line calculation is trained to obtain all kinds of prediction models and parameter, is established according to all kinds of prediction models and parameter and dynamic updates Data dictionary;All kinds of prediction models and parameter are called, and combines the travel time data of real-time section and path, to road network or list The traffic behavior of vehicle is predicted;The data dictionary includes vehicle data dictionary, section data dictionary and path data dictionary; The online vehicle refers to login network access and uploads the vehicle of positioning and speed data automatically.

Further, the present invention is based on the journey time fusion forecasting methods of traffic big data, on all online vehicles The data of biography carry out off-line calculation or training, to obtain all kinds of prediction models and parameter, including, it sets daily from 0:It is arrived when 00 24:It it is a complete cycle period when 00, when according to the trip mode of working day, Saturday, Sunday or festivals or holidays to stroke Between data carry out trip Type division, obtain trip type goes out row number;The travel time data in each period is according to the time Sequencing be arranged in a journey time sequence.

Further, the present invention is based on the journey time fusion forecasting method of traffic big data, all kinds of prediction models And parameter includes section or path forms time data periods rules model, the section or statistical rules mould of path forms time Type, section or path forms time long-term prediction model, section or path forms time short-time forecasting model, section or path row The journey time grows fusion forecasting model and intersection delay relation analysis model in short-term;Wherein,

Using periodic law series approximation PLSA(Periodic Law Series Approximation, PLSA)Algorithm obtains Section or path forms time data periods rules model are taken, and using least square method LSM(Least Square Method, LSM)Solve approximate model parameter;

SRE is extracted using statistical law(Statistical Rule Extraction, SRE)Algorithm obtains section or road Diameter journey time statistical rules model, and using Density Estimator KDE(Kernel Density Estimation, KDE)It obtains The probability density changing rule of section or path forms time;

Using it is long when roll correction prediction LRCF(Long-time Rolling Correction Forecast, LRCF) Algorithm obtains section or path forms time long-term prediction model, and obtains long-term prediction model parameter by off-line calculation, leads to It crosses and fast implements section or path forms time prediction in line computation;

SRFF is predicted using fitting is rolled in short-term(Short-time Rolling Fitting Forecast, SRFF)It calculates Method obtains section or path forms time short-time forecasting model, and using time series autoregressive moving average ARMA (Auto- Regressive and Moving Average, ARMA) algorithm construction short-time forecasting model, and using least square method LSM (Least Square Method, LSM)Solving model parameter;

Using sieve-like fusion forecasting SFF(Sieve Fusion Forecast, SFF)Algorithm obtains section or path forms Time grows fusion forecasting model in short-term, and by off-line training, using Gaussian-Newton method GNIM(Guassian-Newton Iterative Method, GNIM)Obtain fusion forecasting model parameter;

Using intersection delay association analysis IDCA(Intersection Delay Correlation Analysis, IDCA)Algorithm construction intersection delay relation analysis model, and pass through least square method LSM(Least Square Method, LSM)Solving model parameter obtains model parameter, by the quick of line computation realizing route journey time by off-line training Compensation;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

Further, the present invention is based on the journey time fusion forecasting method of traffic big data, the section data dictionaries Periodic sequence, probability sequence and various models for storing journey time and parameter, data storage and update include following Step:

S101, historical data is read, historical data is read from the data dictionary of section, including section number, even time interval are adopted Link Travel Time data, the date and hour data of sample;

S102, section is chosen, a untreated section is chosen according to the sequencing of section number;

S103, the classification of Link Travel Time data, it is theoretical based on road network tidal current, it sets daily from 0:To 24 when 00:00 The Shi Weiyi complete cycle period, according to the trip mode on working day, Saturday, Sunday, festivals or holidays, to travel time data Trip Type division and number are carried out, that is, goes out row number, the travel time data in each period is arranged in a journey time sequence Row;

S104, Link Travel Time data are chosen, a kind of untreated section is chosen according to the sequencing for going out row number Travel time data;

S105, travel time data periods rules model and parameter are obtained, section or path is obtained using PLSA algorithms Travel time data periods rules model, and approximate model parameter is solved using LSM;Including:

S1051, degree of correlation clustering is carried out to the journey time sequence of any two different cycles, the extraction degree of correlation is big Travel time data form collection and be merged into row and ask and calculate, obtain average travel time sequence;

S1052, take " 4 π/hour " as basis pi, one Fourier space model of construction is come when approaching average stroke Between sequence, equation is approached by LSM solutions and obtains model parameter;

S1053, the sequence according to energy from high to low, the interception model parameter of gross energy >=98%, remaining parameter zero setting, from And obtain PLSA model parameters;

S1054, a cycle sequence that Link Travel Time is generated by PLSA algorithms, by the periodic sequence and PLSA models In the information storages such as parameter to section data dictionary;

S106, Link Travel Time statistical law model and parameter are obtained, section or path forms is obtained using SRE algorithms The statistical rules model of time, and section or the probability density changing rule of path forms time are obtained using KDE;Including:

Sometime, the travel time data for selecting all periods at the moment forms data set, uses KDE for S1061, calibration Probability density function is obtained, the corresponding journey time of probability density maximum value, i.e. maximum probability journey time are found;

S1062, corresponding maximum probability journey time of all moment is solved, is chronologically arranged in the general of Link Travel Time Rate sequence is simultaneously stored into section data dictionary;

S107, Link Travel Time long-term prediction model and parameter are obtained, Link Travel Time is obtained using LRCF algorithms Long-term prediction model and parameter, and long-term prediction model parameter is obtained by off-line calculation;Including:

S1071, by the periodic sequence of journey time and probability sequence sum-average arithmetic, obtain journey time it is long when initial value sequence Row

S1072, calibration sometime, calculate all periods the moment corresponding journey time with it is long when initial value compared Compared with, acquisition difference, and chronologically it is arranged in a sequence of differences;

S1073, in view of the number of days spacing between adjacent difference, construction one multiple multinomial model of binary approaches this Sequence of differences adaptively adjusts polynomial item number, finds the corresponding multinomial model of minimum fitness bias;

S1074, the process for solving multinomial model are LRCF algorithms, obtain corresponding LRCF model parameters of all moment And it is stored in the data dictionary of section;

S108, the short-time forecasting model and parameter for obtaining Link Travel Time, when obtaining link travel using SRFF algorithms Between short-time forecasting model, and use ARMA algorithm construction short-time forecasting models, and use LSM solving model parameters;Including:

S1081, chronologically by the rhythmic journey time series arrangement of institute at a long sequence;

S1082, assume that the time interval of the long sequence is impartial, with ARMA algorithm constructions oneNMultinomial model is intended After conjunctionNA journey time asks its parameter and error of fitting with LSM;

S1083, by adjustingNThe size of error of fitting is adjusted, chooses corresponding multinomial model when error minimum;

S1084, the process for solving the multinomial model are SRFF algorithms, and SRFF model parameters are stored in section data word In allusion quotation;

S109, Link Travel Time length fusion forecasting model and parameter in short-term are obtained, link travel is obtained using SFF algorithms Time grows fusion forecasting model in short-term, and by off-line training, and fusion forecasting model parameter is obtained using GNIM;Including:

S1091, calibration initial time, obtain the item number of SRFF modelsN, construction one 2 ×NSieve-like coefficient matrix, wherein It is 1 that each element, which is the sum of the element of non-negative and each row perseverance, and element value is unknown;

S1092, journey time is predicted since initial time, by LRCF algorithms come initial value sequence when compensating long Row obtain futureNA long-term prediction value obtains future by SRFF algorithmsNA short-term prediction value, by both prediction value sequences Composition one 2 ×NPrediction matrix;

S1093, coefficient matrix is added to processing with row after prediction matrix dot product, obtain one 1 ×NFusion vector, with melting Resultant vector approaches corresponding journey time sequence, obtains corresponding related coefficient equation;

S1094, initial time is gradually adjusted backward, corresponding related coefficient equation is obtained with same method, by these Equation constitutes a related coefficient equation group;

S1095, equation group is solved with GNIM, obtains the element value namely SFF model parameters of sieve-like coefficient matrix, it will SFF model parameters are stored in the data dictionary of section;

S110, judge whether that all Link Travel Times are disposed?It is that then, sequence executes step S111, otherwise, returns Execute step S104;

S111, judge whether all sections are disposed?It is that then, sequence executes step S112 and otherwise returns to step S102;

S112, the data storage and update for terminating this section data dictionary;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

Further, the present invention is based on the journey time fusion forecasting method of traffic big data, the path data dictionaries Periodic sequence, probability sequence and various models for storing journey time and parameter, data storage and update include following Step:

S201, historical data is read, historical data, including road-net node, road is read from path and section data dictionary Segment number, path number, the section of even time interval sampling and path forms time data, date and hour data;

S202, combination of nodes is chosen, according to the permutation and combination relationship that road-net node is numbered, chooses one group of untreated two section Point combination;

S203, path selection choose a untreated path according to two combination of nodes;

S204, the classification of path forms time data, it is theoretical based on road network tidal current, it sets daily from 0:To 24 when 00:00 The Shi Weiyi complete cycle period, according to the trip mode on working day, Saturday, Sunday, festivals or holidays, to travel time data Trip Type division and number are carried out, that is, goes out row number, the travel time data in each period is arranged in a journey time sequence Row;

S205, path selection travel time data choose a kind of untreated path according to the sequencing for going out row number Travel time data;

S206, travel time data periods rules model and parameter are obtained, section or path is obtained using PLSA algorithms Travel time data periods rules model, and approximate model parameter is solved using LSM;Including:

S2061, degree of correlation clustering is carried out to the journey time sequence of any two different cycles, the extraction degree of correlation is big Travel time data form collection and be merged into row and ask and calculate, obtain average travel time sequence;

S2062, take " 4 π/hour " as basis pi, one Fourier space model of construction is come when approaching average stroke Between sequence, equation is approached by LSM solutions and obtains model parameter;

S2063, the sequence according to energy from high to low, the interception model parameter of gross energy >=98%, remaining parameter zero setting, from And obtain PLSA model parameters;

S2064, a cycle sequence that the path forms time is generated by PLSA algorithms, by the periodic sequence and PLSA models In the information storages such as parameter to path data dictionary;

S207, Link Travel Time statistical law model and parameter are obtained, section or path forms is obtained using SRE algorithms The statistical rules model of time, and section or the probability density changing rule of path forms time are obtained using KDE;Including:

Sometime, the travel time data for choosing all periods at the moment forms data set, uses KDE for S2071, calibration Probability density function is obtained, the corresponding journey time of probability density maximum value, i.e. maximum probability journey time are found;

S2072, corresponding maximum probability journey time of all moment is solved, is chronologically arranged in the general of path forms time Rate sequence is simultaneously stored into path data dictionary;

S208, intersection delay relation analysis model and parameter are obtained, is associated with using IDCA algorithm construction intersection delays Analysis model, and by LSM solving model parameters, and model parameter is obtained by off-line training;Including:

S2081, it demarcates some period sometime, calculates the sum of all Link Travel Times in the path, Yi Jilu Diameter journey time and Link Travel Time and its difference, the difference are intersection total delay;

S2082, all periods and all moment are traversed, calculates all intersection total delays, path forms is established by LSM Incidence relation between time and corresponding intersection total delay and solution obtain LSM model parameters;

S2083, all intersection total delays are chronologically arranged, it is contemplated that the number of days spacing between consecutive value, construction one A multiple multinomial of binary approaches the sequence, adaptively adjusts polynomial item number, it is corresponding more to find minimum fitness bias Item formula model;

S2084, LSM model of fit and multinomial model are subjected to equal weight merging, pooled model is IDCA models, will In the storage to path data dictionary of IDCA model parameters;

S209, acquisition approach journey time long-term prediction model and parameter obtain section or path row using LRCF algorithms Journey time long-term prediction model, and long-term prediction model parameter is obtained by off-line calculation;Including:

S2091, by the periodic sequence of journey time and probability sequence sum-average arithmetic, obtain journey time it is long when initial value sequence Row

S2092, calibration sometime, calculate all periods the moment corresponding journey time with it is long when initial value compared Compared with, acquisition difference, and chronologically it is arranged in a sequence of differences;

S2093, in view of the number of days spacing between adjacent difference, construction one multiple multinomial model of binary approaches this Sequence of differences adaptively adjusts polynomial item number, finds the corresponding multinomial model of minimum fitness bias;

S2094, the process for solving multinomial model are LRCF algorithms, obtain corresponding LRCF model parameters of all moment And it is stored in path data dictionary;

S210, acquisition approach journey time short-time forecasting model and parameter, using SRFF algorithm acquisition approach journey times Short-time forecasting model, and ARMA algorithm construction short-time forecasting models are used, and use LSM solving model parameters;Including:

S2101, chronologically by the rhythmic journey time series arrangement of institute at a long sequence;

S2102, assume that the time interval of the long sequence is impartial, with ARMA algorithm constructions oneNMultinomial model is intended After conjunctionNA journey time asks its parameter and error of fitting with LSM;

S2103, by adjustingNThe size of error of fitting is adjusted, chooses corresponding multinomial model when error minimum;

S2104, the process for solving the multinomial model are SRFF algorithms, and SRFF model parameters are stored in path data word In allusion quotation;

S211, acquisition approach journey time grow fusion forecasting model and parameter in short-term, using SFF algorithm acquisition approach strokes Time grows fusion forecasting model in short-term, and by off-line training, and fusion forecasting model parameter is obtained using GNIM;Including:

S2111, calibration initial time, obtain the item number of SRFF modelsN, construction one 2 ×NSieve-like coefficient matrix, wherein It is 1 that each element, which is the sum of the element of non-negative and each row perseverance, and element value is unknown;

S2112, journey time is predicted since initial time, by LRCF algorithms come initial value sequence when compensating long Row obtain futureNA long-term prediction value obtains future by SRFF algorithmsNA short-term prediction value, by both prediction value sequences Composition one 2 ×NPrediction matrix;

S2112, coefficient matrix is added to processing with row after prediction matrix dot product, obtain one 1 ×NFusion vector, with melting Resultant vector approaches corresponding journey time sequence, obtains corresponding related coefficient equation;

S2114, initial time is gradually adjusted backward, corresponding related coefficient equation is obtained with same method, by these Equation constitutes a related coefficient equation group;

S2115, equation group is solved with GNIM, obtains the element value namely SFF model parameters of sieve-like coefficient matrix, it will SFF model parameters are stored in path data dictionary;

S212, judge whether all path forms time datas are disposed?It is that then, sequence executes step S213, otherwise, Return to step S205;

S213, judge whether all paths are disposed, be that then, sequence executes step S214 and otherwise returns to step S203;

S214, judge whether all combination of nodes are disposed, be that then, sequence executes step S215, otherwise, return and execute Step S202;

S215, the data storage and update for terminating this path data dictionary;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

The present invention is based on the journey time fusion query methods of traffic big data, using the present invention is based on traffic big datas Journey time fusion forecasting method is to the query path of setting and goes out the journey time variation of row number and carries out prediction and will prediction As a result it is supplied to user, including, the data uploaded using all online vehicles obtain all kinds of predictions by off-line calculation or training Model and parameter are simultaneously established according to all kinds of prediction models and parameter and dynamically update the data dictionary;According to inquiry road set by user Diameter and go out row number, all kinds of prediction models and parameter is called to combine the real-time travel time data of the path and respective stretch to this The journey time of path and respective stretch predict and prediction result is supplied to user;The data dictionary includes vehicle number According to dictionary, section data dictionary and path data dictionary;The online vehicle refers to login network access and uploads positioning and speed automatically The vehicle of degrees of data.

Further, the present invention is based on the journey time fusion query method of traffic big data, it is described go out row number refer to It is theoretical based on road network tidal current, it sets daily from 0:To 24 when 00:It it is a complete cycle period when 00, according to working day, star The trip mode of phase six, Sunday, festivals or holidays carry out trip Type division and number to travel time data, that is, go out row number, The travel time data in each period is arranged in a journey time sequence.

Further, the present invention is based on the journey time fusion query method of traffic big data, all kinds of prediction models And parameter includes section or path forms time data periods rules model, the section or statistical rules mould of path forms time Type, section or path forms time long-term prediction model, section or path forms time short-time forecasting model, section or path row The journey time grows fusion forecasting model and intersection delay relation analysis model in short-term;Wherein,

Using periodic law series approximation PLSA(Periodic Law Series Approximation, PLSA)Algorithm obtains Section or path forms time data periods rules model are taken, and using least square method LSM(Least Square Method, LSM)Solve approximate model parameter;

SRE is extracted using statistical law(Statistical Rule Extraction, SRE)Algorithm obtains section or road Diameter journey time statistical rules model, and using Density Estimator KDE(Kernel Density Estimation, KDE)It obtains The probability density changing rule of section or path forms time;

Using it is long when roll correction prediction LRCF(Long-time Rolling Correction Forecast, LRCF) Algorithm obtains section or path forms time long-term prediction model, and obtains long-term prediction model parameter by off-line calculation, leads to It crosses and fast implements section or path forms time prediction in line computation;

SRFF is predicted using fitting is rolled in short-term(Short-time Rolling Fitting Forecast, SRFF)It calculates Method obtains section or path forms time short-time forecasting model, and using time series autoregressive moving average ARMA (Auto- Regressive and Moving Average, ARMA) algorithm construction short-time forecasting model, and using least square method LSM (Least Square Method, LSM)Solving model parameter;

Using sieve-like fusion forecasting SFF(Sieve Fusion Forecast, SFF)Algorithm obtains section or path forms Time grows fusion forecasting model in short-term, and by off-line training, using Gaussian-Newton method GNIM(Guassian-Newton Iterative Method, GNIM)Obtain fusion forecasting model parameter;

Using intersection delay association analysis IDCA(Intersection Delay Correlation Analysis, IDCA)Algorithm construction intersection delay relation analysis model, and pass through least square method LSM(Least Square Method, LSM)Solving model parameter obtains model parameter, by the quick of line computation realizing route journey time by off-line training Compensation;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

Further, the present invention is based on the journey time fusion query method of traffic big data, include the following steps:

S301, user's inquiry;

S302, acquisition history and real time traffic data, including road-net node number, section number, path number, even time interval The section of sampling and path forms time, all kinds of Forecasting of Travel Time data, all kinds of model parameters, date and hour, data point Class is numbered;

S303, setting path forms time change whether is inquired?It is then, to continue to execute step S304, otherwise, executes step S314;

S304, querying condition setting, user according to the demand of oneself set inquiry content, including the duration of short-term prediction and The period of long-term prediction, according to even time interval principle, the corresponding starting point and ending point with long-term prediction step-length in short-term of calibration, and Go out row number accordingly according to calendar setting;Meanwhile user sets the starting point and ending point of road-net node, passes through searching route Data dictionary demarcates the path between the two road-net nodes and path number;

S305, path selection travel time data choose a kind of untreated path according to the sequencing for going out row number Travel time data;

S306, path selection choose a untreated path, passage path data according to the sequencing of path number It numbers in the section that dictionary search is included to the path;

S307, section is chosen, a untreated section is chosen according to the sequencing of section number;

S308, Link Travel Time fusion forecasting are read accordingly according to user's search request from the data dictionary of section Model and parameter are as follows by step process in conjunction with real-time travel time data:

S3081, by LRCF algorithms obtain journey time it is long when at the beginning of value sequence and sequence of differences, the two, which is added, is gone The long-term prediction sequence of journey time;

S3082, by user choose short-term prediction step-length starting point and ending point on the basis of, obtained by SRFF algorithms The short-term prediction sequence of journey time, then fusion forecasting sequence in short-term is obtained by SFF algorithms;

The starting point and ending point for the long-term prediction step-length that S3083, user of being subject to choose, it is short with long-term prediction sequence pair When fusion forecasting sequence expanded, obtain the final fusion forecasting sequence of Link Travel Time;

S309, judge whether all sections are disposed?It is that then, sequence executes step S310 and otherwise returns to step S307;

S310, the prediction of path forms Fusion in Time, set according to user in predicting, as follows by step process:

S3101, the forecasting sequence that intersection total delay is obtained by IDCA algorithms, it is according to temporal order, intersection is total The forecasting sequence of delay is added with the final fusion forecasting sequence in the included section in the path, obtains the compensation pre- of journey time Sequencing row;

S3102, by LRCF algorithms obtain journey time it is long when at the beginning of value sequence and sequence of differences, the two be added i.e. obtains The long-term prediction sequence of journey time;

S3103, by user choose short-term prediction step-length starting point and ending point on the basis of, obtained by SRFF algorithms The short-term prediction sequence of journey time, then fusion forecasting sequence in short-term is obtained by SFF algorithms;

The starting point and ending point for the long-term prediction step-length that S3104, user of being subject to choose, it is short with long-term prediction sequence pair When fusion forecasting sequence expanded, obtain the expansion type forecasting sequence of journey time;

S3105, expansion type forecasting sequence is added averaging with compensation forecasting sequence, obtains the path forms time most Whole fusion forecasting sequence;

S311, judge whether all paths are disposed?It is that then, sequence executes step S312 and otherwise returns to step S306;

S312, judge whether all trips are disposed?It is that then, sequence executes step S313 and otherwise returns to step S305;

S313, path forms time prediction recombination and Dynamic Announce, by all path forms time final fusion forecastings Sequence, by path, the logical order of classifying and numbering and sequential carry out recombination arrangement, obtain the final forecasting sequence of each path, And the chronologically Dynamic Announce on road network;

S314, terminate this path query;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

Advantageous effects the present invention is based on the journey time fusion forecasting of traffic big data and querying method are to overcome Real-time is poor, versatility is poor and practicability is not strong existing for the traffic status prediction method of prior art road network or bicycle The problems such as, survey precision greatly improved, while the speed of on-line prediction greatly improved, ensure that real-time in engineer application Property, practicability and versatility.

Description of the drawings

Attached drawing 1 is that data dictionary data in section of the present invention store and update flow diagram;

Attached drawing 2 is that path data dictionary data of the present invention stores and update flow diagram;

Attached drawing 3 is path forms time change querying flow schematic diagram of the present invention.

In the following with reference to the drawings and specific embodiments to the present invention is based on the journey time fusion forecasting of traffic big data and looking into Inquiry method is further described.

Specific implementation mode

The present invention is based on the journey time fusion forecasting method of traffic big data, the data uploaded using all online vehicles Obtain all kinds of prediction models and parameter by off-line calculation or training, and establish according to all kinds of prediction models and parameter and dynamic more New data dictionary;Call all kinds of prediction models and parameter combination real-time section and path forms time data to road network or bicycle Traffic behavior is predicted;The data dictionary includes vehicle data dictionary, section data dictionary and path data dictionary;It is described Online vehicle refers to login network access and uploads the vehicle of positioning and speed data automatically.Compared with prior art prediction technique, this Inventive method combines big data technology, and each data dictionary contains multiple indexes, and data storage has preferable systematicness, can be very Meet the requirement of real-time of data storage and read-write well.It of particular concern is, fully take into account in engineer application to prediction The requirement of precision, real-time, versatility and practicability etc., most calculation amounts, most complicated calculating task of this case By completing offline.That is, this case obtains all kinds of prediction model parameters by off-line training, only needed in line computation The prediction model generated offline is called to carry out simple computation.This have the effect that:It can preferably be improved by off-line calculation The precision of prediction of prediction model, while on-line prediction speed can also be greatly improved, reduce on-line calculation so that prediction can be real Shi Shixian, to ensure that the real-time and practicability in engineer application.

In addition, the present invention is based on the journey time fusion forecasting method of traffic big data, uploaded using all online vehicles Data all kinds of prediction models and parameter are obtained by off-line calculation or training, including, set daily from 0:To 24 when 00:When 00 For a complete cycle period, according to working day, Saturday, Sunday, festivals or holidays trip mode to travel time data into The division of row trip type and number, that is, go out row number;The travel time data in each period is arranged in a journey time sequence Row.The advantage of doing so is that:It is divided in conjunction with the periodic characteristic of road net traffic state, has preferably met the essence of road grid traffic Feature can greatly improve the precision of Forecasting of Travel Time, while can preferably hold the periodic law of road net traffic state.

The present invention is based on the journey time fusion forecasting method of traffic big data, all kinds of prediction models and parameter include Section or path forms time data periods rules model, section or statistical rules model, section or the road of path forms time Diameter journey time long-term prediction model, section or path forms time short-time forecasting model, section or path forms time length When fusion forecasting model and intersection delay relation analysis model;Wherein,

Using periodic law series approximation PLSA(Periodic Law Series Approximation, PLSA)Algorithm obtains Section or path forms time data periods rules model are taken, and using least square method LSM(Least Square Method, LSM)Solve approximate model parameter.The cyclically-varying rule of section or path forms time can be preferably extracted as a result, together When, moreover it is possible to arithmetic speed is greatly improved, there is preferable versatility and engineering application value.

SRE is extracted using statistical law(Statistical Rule Extraction, SRE)Algorithm obtains section or road Diameter journey time statistical rules model, and using Density Estimator KDE(Kernel Density Estimation, KDE)It obtains The probability density changing rule of section or path forms time.For model-driven, Link Travel Time has more been respected Self-variation rule.SRE algorithms energy rapid extraction section or periods rules information the most stable in the path forms time, and And computational complexity is not also high, has preferable versatility and engineering application value.

Using it is long when roll correction prediction LRCF(Long-time Rolling Correction Forecast, LRCF) Algorithm obtains section or path forms time long-term prediction model, and obtains long-term prediction model parameter by off-line calculation, leads to It crosses and fast implements section or path forms time prediction in line computation.It can be good at solving the real-time in engineer application with this Problem, meanwhile, preferable precision of prediction also ensures the engineering practicability of the algorithm.

SRFF is predicted using fitting is rolled in short-term(Short-time Rolling Fitting Forecast, SRFF)It calculates Method obtains section or path forms time short-time forecasting model, and using time series autoregressive moving average ARMA (Auto- Regressive and Moving Average, ARMA) algorithm construction short-time forecasting model, and using least square method LSM (Least Square Method, LSM)Solving model parameter.The method has respected fully the tide of section or path forms time Nighttide changing rule.Since the algorithm obtains short-time forecasting model parameter by off-line training, the calculation amount of on-line prediction is very small, It can be good at solving the real time problems in engineer application, meanwhile, preferable precision of prediction also ensures the engineering of the algorithm Practicability.

Using sieve-like fusion forecasting SFF(Sieve Fusion Forecast, SFF)Algorithm obtains section or path forms Time grows fusion forecasting model in short-term, and by off-line training, using Gaussian-Newton method GNIM(Guassian-Newton Iterative Method, GNIM)Obtain fusion forecasting model parameter.The algorithm reduce to a certain extent LRCF algorithms and SRFF algorithms part weakness, while being the effective integration of LRCF algorithms and SRFF algorithms the advantages of strengthen them again.The algorithm By off-line training, fusion forecasting model parameter is obtained using GNIM, precision of prediction can be effectively greatly improved, greatly improve simultaneously The speed of on-line prediction ensure that real-time and practicability in engineer application.

Using intersection delay association analysis IDCA(Intersection Delay Correlation Analysis, IDCA)Algorithm construction intersection delay relation analysis model, and pass through least square method LSM(Least Square Method, LSM)Solving model parameter obtains model parameter, by the quick of line computation realizing route journey time by off-line training Compensation.This method can effectively make up intersection delay deviation caused by the journey time of path.Due to IDCA algorithms pass through it is offline Training obtains model parameter and both ensure that computational accuracy, simultaneously by the quick compensation in line computation realizing route journey time It in turn ensures the speed in line computation, there is preferable engineering practicability.

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

Attached drawing 1 is that data dictionary data in section of the present invention store and update flow diagram, as seen from the figure, the present invention is based on The journey time fusion forecasting method of traffic big data, the section data dictionary be used for store journey time periodic sequence, Probability sequence and various models and parameter, data storage and update include the following steps:

S101, historical data is read, historical data is read from the data dictionary of section, including section number, even time interval are adopted Link Travel Time data, the date and hour data of sample;

S102, section is chosen, a untreated section is chosen according to the sequencing of section number;

S103, the classification of Link Travel Time data, it is theoretical based on road network tidal current, it sets daily from 0:To 24 when 00:00 The Shi Weiyi complete cycle period, according to the trip mode on working day, Saturday, Sunday, festivals or holidays, to travel time data Trip Type division and number are carried out, that is, goes out row number, the travel time data in each period is arranged in a journey time sequence Row;

S104, Link Travel Time data are chosen, a kind of untreated section is chosen according to the sequencing for going out row number Travel time data;

S105, travel time data periods rules model and parameter are obtained, section or path is obtained using PLSA algorithms Travel time data periods rules model, and approximate model parameter is solved using LSM;Including:

S1051, degree of correlation clustering is carried out to the journey time sequence of any two different cycles, the extraction degree of correlation is big Travel time data form collection and be merged into row and ask and calculate, obtain average travel time sequence;

S1052, take " 4 π/hour " as basis pi, one Fourier space model of construction is come when approaching average stroke Between sequence, equation is approached by LSM solutions and obtains model parameter;

S1053, the sequence according to energy from high to low, the interception model parameter of gross energy >=98%, remaining parameter zero setting, from And obtain PLSA model parameters;

S1054, a cycle sequence that Link Travel Time is generated by PLSA algorithms, by the periodic sequence and PLSA models In the information storages such as parameter to section data dictionary;

S106, Link Travel Time statistical law model and parameter are obtained, section or path forms is obtained using SRE algorithms The statistical rules model of time, and section or the probability density changing rule of path forms time are obtained using KDE;Including:

Sometime, the travel time data for selecting all periods at the moment forms data set, uses KDE for S1061, calibration Probability density function is obtained, the corresponding journey time of probability density maximum value, i.e. maximum probability journey time are found;

S1062, corresponding maximum probability journey time of all moment is solved, is chronologically arranged in the general of Link Travel Time Rate sequence is simultaneously stored into section data dictionary;

S107, Link Travel Time long-term prediction model and parameter are obtained, Link Travel Time is obtained using LRCF algorithms Long-term prediction model and parameter, and long-term prediction model parameter is obtained by off-line calculation;Including:

S1071, by the periodic sequence of journey time and probability sequence sum-average arithmetic, obtain journey time it is long when initial value sequence Row

S1072, calibration sometime, calculate all periods the moment corresponding journey time with it is long when initial value compared Compared with, acquisition difference, and chronologically it is arranged in a sequence of differences;

S1073, in view of the number of days spacing between adjacent difference, construction one multiple multinomial model of binary approaches this Sequence of differences adaptively adjusts polynomial item number, finds the corresponding multinomial model of minimum fitness bias;

S1074, the process for solving multinomial model are LRCF algorithms, obtain corresponding LRCF model parameters of all moment And it is stored in the data dictionary of section;

S108, the short-time forecasting model and parameter for obtaining Link Travel Time, when obtaining link travel using SRFF algorithms Between short-time forecasting model, and use ARMA algorithm construction short-time forecasting models, and use LSM solving model parameters;Including:

S1081, chronologically by the rhythmic journey time series arrangement of institute at a long sequence;

S1082, assume that the time interval of the long sequence is impartial, with ARMA algorithm constructions oneNMultinomial model is intended After conjunctionNA journey time asks its parameter and error of fitting with LSM;

S1083, by adjustingNThe size of error of fitting is adjusted, chooses corresponding multinomial model when error minimum;

S1084, the process for solving the multinomial model are SRFF algorithms, and SRFF model parameters are stored in section data word In allusion quotation;

S109, Link Travel Time length fusion forecasting model and parameter in short-term are obtained, link travel is obtained using SFF algorithms Time grows fusion forecasting model in short-term, and by off-line training, and fusion forecasting model parameter is obtained using GNIM;Including:

S1091, calibration initial time, obtain the item number of SRFF modelsN, construction one 2 ×NSieve-like coefficient matrix, wherein It is 1 that each element, which is the sum of the element of non-negative and each row perseverance, and element value is unknown;

S1092, journey time is predicted since initial time, by LRCF algorithms come initial value sequence when compensating long Row obtain futureNA long-term prediction value obtains future by SRFF algorithmsNA short-term prediction value, by both prediction value sequences Composition one 2 ×NPrediction matrix;

S1093, coefficient matrix is added to processing with row after prediction matrix dot product, obtain one 1 ×NFusion vector, with melting Resultant vector approaches corresponding journey time sequence, obtains corresponding related coefficient equation;

S1094, initial time is gradually adjusted backward, corresponding related coefficient equation is obtained with same method, by these Equation constitutes a related coefficient equation group;

S1095, equation group is solved with GNIM, obtains the element value namely SFF model parameters of sieve-like coefficient matrix, it will SFF model parameters are stored in the data dictionary of section;

S110, judge whether that all Link Travel Times are disposed?It is that then, sequence executes step S111, otherwise, returns Execute step S104;

S111, judge whether all sections are disposed?It is that then, sequence executes step S112 and otherwise returns to step S102;

S112, the data storage and update for terminating this section data dictionary;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

Attached drawing 2 is that path data dictionary data of the present invention stores and updates flow diagram, as seen from the figure, the present invention is based on The journey time fusion forecasting method of traffic big data, the path data dictionary be used for store journey time periodic sequence, Probability sequence and various models and parameter, data storage and update include the following steps:

S201, historical data is read, historical data, including road-net node, road is read from path and section data dictionary Segment number, path number, the section of even time interval sampling and path forms time data, date and hour data;

S202, combination of nodes is chosen, according to the permutation and combination relationship that road-net node is numbered, chooses one group of untreated two section Point combination;

S203, path selection choose a untreated path according to two combination of nodes;

S204, the classification of path forms time data, it is theoretical based on road network tidal current, it sets daily from 0:To 24 when 00:00 The Shi Weiyi complete cycle period, according to the trip mode on working day, Saturday, Sunday, festivals or holidays, to travel time data Trip Type division and number are carried out, that is, goes out row number, the travel time data in each period is arranged in a journey time sequence Row;

S205, path selection travel time data choose a kind of untreated path according to the sequencing for going out row number Travel time data;

S206, travel time data periods rules model and parameter are obtained, section or path is obtained using PLSA algorithms Travel time data periods rules model, and approximate model parameter is solved using LSM;Including:

S2061, degree of correlation clustering is carried out to the journey time sequence of any two different cycles, the extraction degree of correlation is big Travel time data form collection and be merged into row and ask and calculate, obtain average travel time sequence;

S2062, take " 4 π/hour " as basis pi, one Fourier space model of construction is come when approaching average stroke Between sequence, equation is approached by LSM solutions and obtains model parameter;

S2063, the sequence according to energy from high to low, the interception model parameter of gross energy >=98%, remaining parameter zero setting, from And obtain PLSA model parameters;

S2064, a cycle sequence that the path forms time is generated by PLSA algorithms, by the periodic sequence and PLSA models In the information storages such as parameter to path data dictionary;

S207, Link Travel Time statistical law model and parameter are obtained, section or path forms is obtained using SRE algorithms The statistical rules model of time, and section or the probability density changing rule of path forms time are obtained using KDE;Including:

Sometime, the travel time data for choosing all periods at the moment forms data set, uses KDE for S2071, calibration Probability density function is obtained, the corresponding journey time of probability density maximum value, i.e. maximum probability journey time are found;

S2072, corresponding maximum probability journey time of all moment is solved, is chronologically arranged in the general of path forms time Rate sequence is simultaneously stored into path data dictionary;

S208, intersection delay relation analysis model and parameter are obtained, is associated with using IDCA algorithm construction intersection delays Analysis model, and by LSM solving model parameters, and model parameter is obtained by off-line training;Including:

S2081, it demarcates some period sometime, calculates the sum of all Link Travel Times in the path, Yi Jilu Diameter journey time and Link Travel Time and its difference, the difference are intersection total delay;

S2082, all periods and all moment are traversed, calculates all intersection total delays, path forms is established by LSM Incidence relation between time and corresponding intersection total delay and solution obtain LSM model parameters;

S2083, all intersection total delays are chronologically arranged, it is contemplated that the number of days spacing between consecutive value, construction one A multiple multinomial of binary approaches the sequence, adaptively adjusts polynomial item number, it is corresponding more to find minimum fitness bias Item formula model;

S2084, LSM model of fit and multinomial model are subjected to equal weight merging, pooled model is IDCA models, will In the storage to path data dictionary of IDCA model parameters;

S209, acquisition approach journey time long-term prediction model and parameter obtain section or path row using LRCF algorithms Journey time long-term prediction model, and long-term prediction model parameter is obtained by off-line calculation;Including:

S2091, by the periodic sequence of journey time and probability sequence sum-average arithmetic, obtain journey time it is long when initial value sequence Row

S2092, calibration sometime, calculate all periods the moment corresponding journey time with it is long when initial value compared Compared with, acquisition difference, and chronologically it is arranged in a sequence of differences;

S2093, in view of the number of days spacing between adjacent difference, construction one multiple multinomial model of binary approaches this Sequence of differences adaptively adjusts polynomial item number, finds the corresponding multinomial model of minimum fitness bias;

S2094, the process for solving multinomial model are LRCF algorithms, obtain corresponding LRCF model parameters of all moment And it is stored in path data dictionary;

S210, acquisition approach journey time short-time forecasting model and parameter, using SRFF algorithm acquisition approach journey times Short-time forecasting model, and ARMA algorithm construction short-time forecasting models are used, and use LSM solving model parameters;Including:

S2101, chronologically by the rhythmic journey time series arrangement of institute at a long sequence;

S2102, assume that the time interval of the long sequence is impartial, with ARMA algorithm constructions oneNMultinomial model is intended After conjunctionNA journey time asks its parameter and error of fitting with LSM;

S2103, by adjustingNThe size of error of fitting is adjusted, chooses corresponding multinomial model when error minimum;

S2104, the process for solving the multinomial model are SRFF algorithms, and SRFF model parameters are stored in path data word In allusion quotation;

S211, acquisition approach journey time grow fusion forecasting model and parameter in short-term, using SFF algorithm acquisition approach strokes Time grows fusion forecasting model in short-term, and by off-line training, and fusion forecasting model parameter is obtained using GNIM;Including:

S2111, calibration initial time, obtain the item number of SRFF modelsN, construction one 2 ×NSieve-like coefficient matrix, wherein It is 1 that each element, which is the sum of the element of non-negative and each row perseverance, and element value is unknown;

S2112, journey time is predicted since initial time, by LRCF algorithms come initial value sequence when compensating long Row obtain futureNA long-term prediction value obtains future by SRFF algorithmsNA short-term prediction value, by both prediction value sequences Composition one 2 ×NPrediction matrix;

S2112, coefficient matrix is added to processing with row after prediction matrix dot product, obtain one 1 ×NFusion vector, with melting Resultant vector approaches corresponding journey time sequence, obtains corresponding related coefficient equation;

S2114, initial time is gradually adjusted backward, corresponding related coefficient equation is obtained with same method, by these Equation constitutes a related coefficient equation group;

S2115, equation group is solved with GNIM, obtains the element value namely SFF model parameters of sieve-like coefficient matrix, it will SFF model parameters are stored in path data dictionary;

S212, judge whether all path forms time datas are disposed?It is that then, sequence executes step S213, otherwise, Return to step S205;

S213, judge whether all paths are disposed, be that then, sequence executes step S214 and otherwise returns to step S203;

S214, judge whether all combination of nodes are disposed, be that then, sequence executes step S215, otherwise, return and execute Step S202;

S215, the data storage and update for terminating this path data dictionary;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

Attached drawing 3 is path forms time change querying flow schematic diagram of the present invention, and as seen from the figure, it is big that the present invention is based on traffic The journey time fusion query method of data is set using the journey time fusion forecasting method pair the present invention is based on traffic big data Fixed query path and the journey time variation for going out row number predict and prediction result are supplied to user, including, it uses The data that all online vehicles upload are by off-line calculation or all kinds of prediction models of training acquisition and parameter and according to all kinds of predictions Model and parameter are established and dynamically update the data dictionary;According to query path set by user and go out row number, calls all kinds of pre- When survey model and parameter combine stroke of the real-time travel time data of the path and respective stretch to the path and respective stretch Between carry out predict and prediction result is supplied to user;The data dictionary include vehicle data dictionary, section data dictionary and Path data dictionary;The online vehicle refers to login network access and uploads the vehicle of positioning and speed data automatically.

Equally, the present invention is based on the journey time fusion query method of traffic big data, it is described go out row number refer to being based on Road network tidal current is theoretical, sets daily from 0:To 24 when 00:When 00 be a complete cycle period, according to working day, Saturday, The trip mode on Sunday, festivals or holidays carries out trip Type division and number to travel time data, that is, goes out row number, each The travel time data in period is arranged in a journey time sequence.

Equally, the present invention is based on the journey time fusion query method of traffic big data, all kinds of prediction models and ginsengs Number includes section or path forms time data periods rules model, section or the statistical rules model of path forms time, road When section or path forms time long-term prediction model, section or path forms time short-time forecasting model, section or path forms Between the long model of fusion forecasting in short-term and intersection delay relation analysis model;Wherein,

Using periodic law series approximation PLSA(Periodic Law Series Approximation, PLSA)Algorithm obtains Section or path forms time data periods rules model are taken, and using least square method LSM(Least Square Method, LSM)Solve approximate model parameter;

SRE is extracted using statistical law(Statistical Rule Extraction, SRE)Algorithm obtains section or road Diameter journey time statistical rules model, and using Density Estimator KDE(Kernel Density Estimation, KDE)It obtains The probability density changing rule of section or path forms time;

Using it is long when roll correction prediction LRCF(Long-time Rolling Correction Forecast, LRCF) Algorithm obtains section or path forms time long-term prediction model, and obtains long-term prediction model parameter by off-line calculation, leads to It crosses and fast implements section or path forms time prediction in line computation;

SRFF is predicted using fitting is rolled in short-term(Short-time Rolling Fitting Forecast, SRFF)It calculates Method obtains section or path forms time short-time forecasting model, and using time series autoregressive moving average ARMA (Auto- Regressive and Moving Average, ARMA) algorithm construction short-time forecasting model, and using least square method LSM (Least Square Method, LSM)Solving model parameter;

Using sieve-like fusion forecasting SFF(Sieve Fusion Forecast, SFF)Algorithm obtains section or path forms Time grows fusion forecasting model in short-term, and by off-line training, using Gaussian-Newton method GNIM(Guassian-Newton Iterative Method, GNIM)Obtain fusion forecasting model parameter;

Using intersection delay association analysis IDCA(Intersection Delay Correlation Analysis, IDCA)Algorithm construction intersection delay relation analysis model, and pass through least square method LSM(Least Square Method, LSM)Solving model parameter obtains model parameter, by the quick of line computation realizing route journey time by off-line training Compensation;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

The present invention is based on the journey time fusion query methods of traffic big data, include the following steps:

S301, user's inquiry;

S302, acquisition history and real time traffic data, including road-net node number, section number, path number, even time interval The section of sampling and path forms time, all kinds of Forecasting of Travel Time data, all kinds of model parameters, date and hour, data point Class is numbered;

S303, setting path forms time change whether is inquired?It is then, to continue to execute step S304, otherwise, executes step S314;

S304, querying condition setting, user according to the demand of oneself set inquiry content, including the duration of short-term prediction and The period of long-term prediction, according to even time interval principle, the corresponding starting point and ending point with long-term prediction step-length in short-term of calibration, and Go out row number accordingly according to calendar setting;Meanwhile user sets the starting point and ending point of road-net node, passes through searching route Data dictionary demarcates the path between the two road-net nodes and path number;

S305, path selection travel time data choose a kind of untreated path according to the sequencing for going out row number Travel time data;

S306, path selection choose a untreated path, passage path data according to the sequencing of path number It numbers in the section that dictionary search is included to the path;

S307, section is chosen, a untreated section is chosen according to the sequencing of section number;

S308, Link Travel Time fusion forecasting are read accordingly according to user's search request from the data dictionary of section Model and parameter are as follows by step process in conjunction with real-time travel time data:

S3081, by LRCF algorithms obtain journey time it is long when at the beginning of value sequence and sequence of differences, the two, which is added, is gone The long-term prediction sequence of journey time;

S3082, by user choose short-term prediction step-length starting point and ending point on the basis of, obtained by SRFF algorithms The short-term prediction sequence of journey time, then fusion forecasting sequence in short-term is obtained by SFF algorithms;

The starting point and ending point for the long-term prediction step-length that S3083, user of being subject to choose, it is short with long-term prediction sequence pair When fusion forecasting sequence expanded, obtain the final fusion forecasting sequence of Link Travel Time;

S309, judge whether all sections are disposed?It is that then, sequence executes step S310 and otherwise returns to step S307;

S310, the prediction of path forms Fusion in Time, set according to user in predicting, as follows by step process:

S3101, the forecasting sequence that intersection total delay is obtained by IDCA algorithms, it is according to temporal order, intersection is total The forecasting sequence of delay is added with the final fusion forecasting sequence in the included section in the path, obtains the compensation pre- of journey time Sequencing row;

S3102, by LRCF algorithms obtain journey time it is long when at the beginning of value sequence and sequence of differences, the two be added i.e. obtains The long-term prediction sequence of journey time;

S3103, by user choose short-term prediction step-length starting point and ending point on the basis of, obtained by SRFF algorithms The short-term prediction sequence of journey time, then fusion forecasting sequence in short-term is obtained by SFF algorithms;

The starting point and ending point for the long-term prediction step-length that S3104, user of being subject to choose, it is short with long-term prediction sequence pair When fusion forecasting sequence expanded, obtain the expansion type forecasting sequence of journey time;

S3105, expansion type forecasting sequence is added averaging with compensation forecasting sequence, obtains the path forms time most Whole fusion forecasting sequence;

S311, judge whether all paths are disposed?It is that then, sequence executes step S312 and otherwise returns to step S306;

S312, judge whether all trips are disposed?It is that then, sequence executes step S313 and otherwise returns to step S305;

S313, path forms time prediction recombination and Dynamic Announce, by all path forms time final fusion forecastings Sequence, by path, the logical order of classifying and numbering and sequential carry out recombination arrangement, obtain the final forecasting sequence of each path, And the chronologically Dynamic Announce on road network;

S314, terminate this path query;

Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, The value range of long-term prediction duration is set as 0 minute to 3 months, and specific long-term prediction duration can be in the prediction that user selectes It is obtained on the basis of duration by linearly converting;The short-term prediction refers to when obtaining section or path forms by SRFF algorithms Between short-term prediction value, the value range of short-term prediction duration is set as 0 minute to 3 hour, specific short-term prediction duration It can be obtained on the basis of the prediction duration that user selectes by linearly converting.

Obviously, the present invention is based on the advantageous effects of the journey time fusion forecasting of traffic big data and querying method is Overcome existing for the traffic status prediction method of prior art road network or bicycle that real-time is poor, versatility is poor and practicability Survey precision greatly improved in the problems such as not strong, while the speed of on-line prediction greatly improved, and ensure that the reality in engineer application Shi Xing, practicability and versatility.

Claims (8)

1. a kind of journey time fusion forecasting method based on traffic big data, which is characterized in that uploaded to all online vehicles Data carry out off-line calculation or training obtains all kinds of prediction models and parameter, established simultaneously according to all kinds of prediction models and parameter Dynamic updates the data dictionary;All kinds of prediction models and parameter are called, and combines the travel time data of real-time section and path, it is right The traffic behavior of road network or bicycle is predicted;The data dictionary includes vehicle data dictionary, section data dictionary and path Data dictionary;The online vehicle refers to login network access and uploads the vehicle of positioning and speed data automatically;Wherein,
All kinds of prediction models and parameter include section or periods rules model, section or the path of path forms time data Statistical rules model, section or the long-term prediction model of path forms time of journey time, section or path forms time The length of short-time forecasting model, section or path forms time fusion forecasting model and intersection delay relation analysis model in short-term; Wherein,
Using periodic law series approximation PLSA algorithms, Periodic Law Series Approximation, PLSA obtain road Section or path forms time data periods rules model, and least square method LSM, Least Square Method, LSM are used, Solve approximate model parameter;
SRE algorithms are extracted using statistical law, Statistical Rule Extraction, SRE obtain section or path row Journey time statistical rules model, and Density Estimator KDE algorithms, Kernel Density Estimation, KDE is used to obtain The probability density changing rule of section or path forms time;
Using it is long when roll correction prediction LRCF algorithms, Long-time Rolling Correction Forecast, LRCF, Obtain section or path forms time long-term prediction model, and long-term prediction model parameter obtained by off-line calculation, by Line computation fast implements section or path forms time prediction;
Using fitting prediction SRFF algorithms are rolled in short-term, Short-time Rolling Fitting Forecast, SRFF are obtained Section or path forms time short-time forecasting model are taken, and uses time series autoregressive moving average ARMA algorithms, Auto- Regressive and Moving Average, ARMA construct short-time forecasting model, and using least square method LSM, Least Square Method, LSM, solving model parameter;
Using sieve-like fusion forecasting SFF algorithms, Sieve Fusion Forecast, SFF obtain section or path forms time Long fusion forecasting model in short-term, and by off-line training, using Gaussian-Newton method GNIM, Guassian-Newton Iterative Method, GNIM obtain fusion forecasting model parameter;
Using intersection delay association analysis IDCA algorithms, Intersection Delay Correlation Analysis, IDCA constructs intersection delay relation analysis model, and by least square method LSM, Least Square Method, LSM, Solving model parameter obtains model parameter by off-line training, passes through the quick compensation in line computation realizing route journey time;
Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, when long The value range of prediction duration is set as 0 minute to 3 months, the prediction duration base that specific long-term prediction duration is selected in user It is obtained on plinth by linearly converting;The short-term prediction refers to obtaining the short of section or path forms time by SRFF algorithms When predicted value, the value range of short-term prediction duration is set as 0 minute to 3 hour, and specific short-term prediction duration is in user It is obtained on the basis of selected prediction duration by linearly converting.
2. the journey time fusion forecasting method based on traffic big data according to claim 1, which is characterized in that all The data that online vehicle uploads carry out off-line calculation or training, to obtain all kinds of prediction models and parameter, including, setting daily from 0:To 24 when 00:It it is a complete cycle period when 00, according to the trip mode of working day, Saturday, Sunday or festivals or holidays Trip Type division is carried out to travel time data, obtain trip type goes out row number;The travel time data in each period It is arranged in a journey time sequence according to the sequencing of time.
3. the journey time fusion forecasting method based on traffic big data according to claim 1, which is characterized in that the road Segment data dictionary is used to store periodic sequence, probability sequence and the various models and parameter of all Link Travel Times, data Storage and update include the following steps:
S101, historical data is read, historical data is read from the data dictionary of section, including section number, even time interval sample Link Travel Time data, date and hour data;
S102, section is chosen, a untreated section is chosen according to the sequencing of section number;
S103, the classification of Link Travel Time data, it is theoretical based on road network tidal current, it sets daily from 0:To 24 when 00:It is when 00 One complete cycle period carries out travel time data according to the trip mode on working day, Saturday, Sunday, festivals or holidays Trip Type division and number, that is, go out row number, the travel time data in each period is arranged in a journey time sequence;
S104, Link Travel Time data are chosen, a kind of untreated link travel is chosen according to the sequencing for going out row number Time data;
S105, travel time data periods rules model and parameter are obtained, section or path forms is obtained using PLSA algorithms Time data periods rules model, and approximate model parameter is solved using LSM;Including:
S1051, degree of correlation clustering, the big row of the extraction degree of correlation are carried out to the journey time sequence of any two different cycles Journey time data is formed to collect to be merged into go to ask and be calculated, and obtains average travel time sequence;
S1052, take " 4 π/hour " as basic pi, one Fourier space model of construction approaches average travel time sequence Row approach equation to obtain model parameter by LSM solutions;
S1053, the sequence according to energy from high to low, the interception model parameter of gross energy >=98%, remaining parameter zero setting, to obtain Obtain PLSA model parameters;
S1054, a cycle sequence that Link Travel Time is generated by PLSA algorithms, by the periodic sequence and PLSA model parameters In information storage to section data dictionary;
S106, Link Travel Time statistical law model and parameter are obtained, section or path forms time is obtained using SRE algorithms Statistical rules model, and section or the probability density changing rule of path forms time are obtained using KDE;Including:
Sometime, the travel time data for selecting all periods at the moment forms data set, is obtained with KDE for S1061, calibration Probability density function finds the corresponding journey time of probability density maximum value, i.e. maximum probability journey time;
S1062, corresponding maximum probability journey time of all moment is solved, is chronologically arranged in the probability sequence of Link Travel Time It arranges and stores into section data dictionary;
S107, Link Travel Time long-term prediction model and parameter are obtained, when long using LRCF algorithms acquisition Link Travel Time Prediction model and parameter, and long-term prediction model parameter is obtained by off-line calculation;Including:
S1071, by the periodic sequence of journey time and probability sequence sum-average arithmetic, obtain journey time it is long when at the beginning of value sequence;
S1072, calibration sometime, calculate all periods the moment corresponding journey time with it is long when initial value be compared, Difference is obtained, and is chronologically arranged in a sequence of differences;
S1073, in view of the number of days spacing between adjacent difference, construction one multiple multinomial model of binary approaches the difference Sequence adaptively adjusts polynomial item number, finds the corresponding multinomial model of minimum fitness bias;
S1074, the process for solving multinomial model are LRCF algorithms, obtain corresponding LRCF model parameters of all moment and deposit In approach segment data dictionary;
It is short to obtain Link Travel Time using SRFF algorithms for S108, the short-time forecasting model and parameter for obtaining Link Travel Time When prediction model, and use ARMA algorithm construction short-time forecasting models, and use LSM solving model parameters;Including:
S1081, chronologically by the rhythmic journey time series arrangement of institute at a long sequence;
S1082, assume that the time interval of the long sequence is impartial, after being fitted with one N multinomial model of ARMA algorithm constructions N number of journey time asks its parameter and error of fitting with LSM;
S1083, the size of error of fitting is adjusted by adjusting N, chooses corresponding multinomial model when error minimum;
S1084, the process for solving the multinomial model are SRFF algorithms, and SRFF model parameters are stored in section data dictionary In;
S109, Link Travel Time length fusion forecasting model and parameter in short-term are obtained, Link Travel Time is obtained using SFF algorithms Long fusion forecasting model in short-term, and by off-line training, fusion forecasting model parameter is obtained using GNIM;Including:
S1091, calibration initial time, obtain the item number N of SRFF models, construct 2 × N sieve-like coefficient matrix, wherein is each Element is that the sum of the element of non-negative and each row perseverance is 1, and element value is unknown;
S1092, journey time is predicted since initial time, by LRCF algorithms come value sequence at the beginning of when compensating long, is obtained Following N number of long-term prediction value is taken, following N number of short-term prediction value is obtained by SRFF algorithms, is made of both prediction value sequences One 2 × N prediction matrix;
S1093, coefficient matrix with row after prediction matrix dot product is added to processing, obtains 1 × N fusion vector, with fusion to Amount approaches corresponding journey time sequence, obtains corresponding related coefficient equation;
S1094, initial time is gradually adjusted backward, corresponding related coefficient equation is obtained with same method, by these equations Constitute a related coefficient equation group;
S1095, equation group is solved with GNIM, the element value namely SFF model parameters of sieve-like coefficient matrix is obtained, by SFF moulds Shape parameter is stored in the data dictionary of section;
S110, judge whether that all Link Travel Times are disposed, be that then, sequence executes step S111, otherwise, return and execute Step S104;
S111, judge whether all sections are disposed, be that then, sequence executes step S112 and otherwise returns to step S102;
S112, the data storage and update for terminating this section data dictionary;
Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, when long The value range of prediction duration is set as 0 minute to 3 months, the prediction duration base that specific long-term prediction duration is selected in user It is obtained on plinth by linearly converting;The short-term prediction refers to obtaining the short of section or path forms time by SRFF algorithms When predicted value, the value range of short-term prediction duration is set as 0 minute to 3 hour, and specific short-term prediction duration is in user It is obtained on the basis of selected prediction duration by linearly converting.
4. the journey time fusion forecasting method based on traffic big data according to claim 1, which is characterized in that the road Diameter data dictionary is used to store periodic sequence, probability sequence and the various models and parameter of journey time, data storage and more Newly include the following steps:
S201, historical data is read, historical data is read from path and section data dictionary, including road-net node, section are compiled Number, path number, even time interval sampling section and path forms time data, date and hour data;
S202, combination of nodes is chosen, according to the permutation and combination relationship that road-net node is numbered, chooses one group of untreated two nodes group It closes;
S203, path selection choose a untreated path according to two combination of nodes;
S204, the classification of path forms time data, it is theoretical based on road network tidal current, it sets daily from 0:To 24 when 00:It is when 00 One complete cycle period carries out travel time data according to the trip mode on working day, Saturday, Sunday, festivals or holidays Trip Type division and number, that is, go out row number, the travel time data in each period is arranged in a journey time sequence;
S205, path selection travel time data choose a kind of untreated path forms according to the sequencing for going out row number Time data;
S206, travel time data periods rules model and parameter are obtained, section or path forms is obtained using PLSA algorithms Time data periods rules model, and approximate model parameter is solved using LSM;Including:
S2061, degree of correlation clustering, the big row of the extraction degree of correlation are carried out to the journey time sequence of any two different cycles Journey time data is formed to collect to be merged into go to ask and be calculated, and obtains average travel time sequence;
S2062, take " 4 π/hour " as basic pi, one Fourier space model of construction approaches average travel time sequence Row approach equation to obtain model parameter by LSM solutions;
S2063, the sequence according to energy from high to low, the interception model parameter of gross energy >=98%, remaining parameter zero setting, to obtain Obtain PLSA model parameters;
S2064, a cycle sequence that the path forms time is generated by PLSA algorithms, by the periodic sequence and PLSA model parameters In information storage to path data dictionary;
S207, Link Travel Time statistical law model and parameter are obtained, section or path forms time is obtained using SRE algorithms Statistical rules model, and section or the probability density changing rule of path forms time are obtained using KDE;Including:
Sometime, the travel time data for choosing all periods at the moment forms data set, is obtained with KDE for S2071, calibration Probability density function finds the corresponding journey time of probability density maximum value, i.e. maximum probability journey time;
S2072, corresponding maximum probability journey time of all moment is solved, is chronologically arranged in the probability sequence of path forms time It arranges and stores in path data dictionary;
S208, intersection delay relation analysis model and parameter are obtained, using intersection delay association analysis IDCA algorithm constructions Intersection delay relation analysis model, and by LSM solving model parameters, and model parameter is obtained by off-line training;Including:
S2081, it demarcates some period sometime, calculates the sum of all Link Travel Times in the path and path row Journey time and Link Travel Time and its difference, the difference are intersection total delay;
S2082, all periods and all moment are traversed, calculates all intersection total delays, the path forms time is established by LSM Incidence relation between corresponding intersection total delay and solution obtain LSM model parameters;
S2083, all intersection total delays are chronologically arranged, it is contemplated that the number of days spacing between consecutive value constructs one two First repeatedly multinomial approaches the sequence, adaptively adjusts polynomial item number, finds the corresponding multinomial of minimum fitness bias Model;
S2084, LSM model of fit and multinomial model are subjected to equal weight merging, pooled model is IDCA models, by IDCA In model parameter storage to path data dictionary;
S209, acquisition approach journey time long-term prediction model and parameter, when obtaining section or path forms using LRCF algorithms Between long-term prediction model, and pass through off-line calculation obtain long-term prediction model parameter;Including:
S2091, by the periodic sequence of journey time and probability sequence sum-average arithmetic, obtain journey time it is long when at the beginning of value sequence;
S2092, calibration sometime, calculate all periods the moment corresponding journey time with it is long when initial value be compared, Difference is obtained, and is chronologically arranged in a sequence of differences;
S2093, in view of the number of days spacing between adjacent difference, construction one multiple multinomial model of binary approaches the difference Sequence adaptively adjusts polynomial item number, finds the corresponding multinomial model of minimum fitness bias;
S2094, the process for solving multinomial model are LRCF algorithms, obtain corresponding LRCF model parameters of all moment and deposit Enter in the data dictionary of path;
S210, acquisition approach journey time short-time forecasting model and parameter, in short-term using SRFF algorithm acquisition approachs journey time Prediction model, and ARMA algorithm construction short-time forecasting models are used, and use LSM solving model parameters;Including:
S2101, chronologically by the rhythmic journey time series arrangement of institute at a long sequence;
S2102, assume that the time interval of the long sequence is impartial, after being fitted with one N multinomial model of ARMA algorithm constructions N number of journey time asks its parameter and error of fitting with LSM;
S2103, the size of error of fitting is adjusted by adjusting N, chooses corresponding multinomial model when error minimum;
S2104, the process for solving the multinomial model are SRFF algorithms, and SRFF model parameters are stored in path data dictionary In;
S211, acquisition approach journey time grow fusion forecasting model and parameter in short-term, using SFF algorithm acquisition approach journey times Long fusion forecasting model in short-term, and by off-line training, fusion forecasting model parameter is obtained using GNIM;Including:
S2111, calibration initial time, obtain the item number N of SRFF models, 2 × N sieve-like coefficient matrix are constructed, wherein each Element is that the sum of the element of non-negative and each row perseverance is 1, and element value is unknown;
S2112, journey time is predicted since initial time, by LRCF algorithms come value sequence at the beginning of when compensating long, is obtained Following N number of long-term prediction value is taken, following N number of short-term prediction value is obtained by SRFF algorithms, is made of both prediction value sequences One 2 × N prediction matrix;
S2112, coefficient matrix with row after prediction matrix dot product is added to processing, obtains 1 × N fusion vector, with fusion to Amount approaches corresponding journey time sequence, obtains corresponding related coefficient equation;
S2114, initial time is gradually adjusted backward, corresponding related coefficient equation is obtained with same method, by these equations Constitute a related coefficient equation group;
S2115, solve equation group with GNIM, the element value namely SFF model parameters of sieve-like coefficient matrix are obtained, by SFF Model parameter is stored in path data dictionary;
S212, judge whether all path forms time datas are disposed, be that then, sequence executes step S213, otherwise, return Execute step S205;
S213, judge whether all paths are disposed, be that then, sequence executes step S214 and otherwise returns to step S203;
S214, judge whether all combination of nodes are disposed, be that then, sequence executes step S215 and otherwise returns to step S202;
S215, the data storage and update for terminating this path data dictionary;
Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, when long The value range of prediction duration is set as 0 minute to 3 months, the prediction duration base that specific long-term prediction duration is selected in user It is obtained on plinth by linearly converting;The short-term prediction refers to obtaining the short of section or path forms time by SRFF algorithms When predicted value, the value range of short-term prediction duration is set as 0 minute to 3 hour, and specific short-term prediction duration is in user It is obtained on the basis of selected prediction duration by linearly converting.
5. a kind of journey time fusion query method based on traffic big data, which is characterized in that appointed using Claims 1-4 The stroke of row number to the query path of setting and is gone out based on the journey time fusion forecasting method of traffic big data described in one Time change predict and prediction result is supplied to user, including, the data uploaded using all online vehicles by from Line computation or training obtain all kinds of prediction models and parameter and establish according to all kinds of prediction models and parameter and dynamically update the data Dictionary;According to query path set by user and go out row number, all kinds of prediction models and parameter is called to combine the path and corresponding The real-time travel time data in section to the journey time in the path and respective stretch predict and is supplied to prediction result User;The data dictionary includes vehicle data dictionary, section data dictionary and path data dictionary;The online vehicle refers to Login network access and the vehicle for uploading positioning and speed data automatically.
6. the journey time fusion query method based on traffic big data according to claim 5, which is characterized in that it is described go out Row number refers to theoretical based on road network tidal current, is set daily from 0:To 24 when 00:It is a complete cycle period when 00, according to The trip mode on working day, Saturday, Sunday, festivals or holidays carries out trip Type division and number, i.e., to travel time data Go out row number, the travel time data in each period is arranged in a journey time sequence.
7. the journey time fusion query method based on traffic big data according to claim 5, which is characterized in that described each Class prediction model and parameter include section or path forms time data periods rules model, the section or system of path forms time Count rule model, section or path forms time long-term prediction model, section or path forms time short-time forecasting model, section Or the path forms time grows fusion forecasting model and intersection delay relation analysis model in short-term;Wherein,
Using periodic law series approximation PLSA algorithms, Periodic Law Series Approximation, PLSA obtain road Section or path forms time data periods rules model, and least square method LSM, Least Square Method, LSM are used, Solve approximate model parameter;
SRE algorithms are extracted using statistical law, Statistical Rule Extraction, SRE obtain section or path row Journey time statistical rules model, and Density Estimator KDE algorithms, Kernel Density Estimation, KDE is used to obtain The probability density changing rule of section or path forms time;
Using it is long when roll correction prediction LRCF algorithms, Long-time Rolling Correction Forecast, LRCF, Obtain section or path forms time long-term prediction model, and long-term prediction model parameter obtained by off-line calculation, by Line computation fast implements section or path forms time prediction;
Using fitting prediction SRFF algorithms are rolled in short-term, Short-time Rolling Fitting Forecast, SRFF are obtained Section or path forms time short-time forecasting model are taken, and uses time series autoregressive moving average ARMA algorithms, Auto- Regressive and Moving Average, ARMA construct short-time forecasting model, and using least square method LSM, Least Square Method, LSM, solving model parameter;
Using sieve-like fusion forecasting SFF algorithms, Sieve Fusion Forecast, SFF obtain section or path forms time Long fusion forecasting model in short-term, and by off-line training, using Gaussian-Newton method GNIM, Guassian-Newton Iterative Method, GNIM obtain fusion forecasting model parameter;
Using intersection delay association analysis IDCA algorithms, Intersection Delay Correlation Analysis, IDCA,)Intersection delay relation analysis model is constructed, and by least square method LSM, Least Square Method, LSM, Solving model parameter obtains model parameter by off-line training, passes through the quick compensation in line computation realizing route journey time;
Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, when long The value range of prediction duration is set as 0 minute to 3 months, the prediction duration base that specific long-term prediction duration is selected in user It is obtained on plinth by linearly converting;The short-term prediction refers to obtaining the short of section or path forms time by SRFF algorithms When predicted value, the value range of short-term prediction duration is set as 0 minute to 3 hour, and specific short-term prediction duration is in user It is obtained on the basis of selected prediction duration by linearly converting.
8. according to the journey time fusion query method based on traffic big data described in any one of claim 5 to 7, feature exists In including the following steps:
S301, user's inquiry;
S302, acquisition history and real time traffic data, including road-net node number, the sampling of section number, path number, even time interval Section and the path forms time, all kinds of Forecasting of Travel Time data, all kinds of model parameters, date and hour, data classification compile Number;
S303, setting path forms time change whether is inquired, is then, to continue to execute step S304, otherwise, execute step S314;
S304, querying condition setting, user according to the demand of oneself set inquiry content, including the duration of short-term prediction and it is long when The period of prediction, according to even time interval principle, the corresponding starting point and ending point with long-term prediction step-length in short-term of calibration, and according to Calendar setting goes out row number accordingly;Meanwhile user sets the starting point and ending point of road-net node, passes through searching route data Dictionary demarcates the path between the two road-net nodes and path number;
S305, path selection travel time data choose a kind of untreated path forms according to the sequencing for going out row number Time data;
S306, path selection choose a untreated path, passage path data dictionary according to the sequencing of path number Search the section number that the path is included;
S307, section is chosen, a untreated section is chosen according to the sequencing of section number;
S308, Link Travel Time fusion forecasting read corresponding model according to user's search request from the data dictionary of section And parameter, it is as follows by step process in conjunction with real-time travel time data:
S3081, by LRCF algorithms obtain journey time it is long when at the beginning of value sequence and sequence of differences, when the two is added to obtain stroke Between long-term prediction sequence;
S3082, by user choose short-term prediction step-length starting point and ending point on the basis of, pass through SRFF algorithms obtain stroke The short-term prediction sequence of time, then fusion forecasting sequence in short-term is obtained by SFF algorithms;
The starting point and ending point for the long-term prediction step-length that S3083, user of being subject to choose, is melted in short-term with long-term prediction sequence pair It closes forecasting sequence to be expanded, obtains the final fusion forecasting sequence of Link Travel Time;
S309, judge whether all sections are disposed, be that then, sequence executes step S310 and otherwise returns to step S307;
S310, the prediction of path forms Fusion in Time, set according to user in predicting, as follows by step process:
S3101, the forecasting sequence that intersection total delay is obtained by IDCA algorithms, according to temporal order, by intersection total delay Forecasting sequence be added with the final fusion forecasting sequence in the included section in the path, obtain the compensation pre- sequencing of journey time Row;
S3102, by LRCF algorithms obtain journey time it is long when at the beginning of value sequence and sequence of differences, the two be added i.e. obtains stroke The long-term prediction sequence of time;
S3103, by user choose short-term prediction step-length starting point and ending point on the basis of, pass through SRFF algorithms obtain stroke The short-term prediction sequence of time, then fusion forecasting sequence in short-term is obtained by SFF algorithms;
The starting point and ending point for the long-term prediction step-length that S3104, user of being subject to choose, is melted in short-term with long-term prediction sequence pair It closes forecasting sequence to be expanded, obtains the expansion type forecasting sequence of journey time;
S3105, expansion type forecasting sequence is added averaging with compensation forecasting sequence, obtains the final of path forms time and melts Close forecasting sequence;
S311, judge whether all paths are disposed, be that then, sequence executes step S312 and otherwise returns to step S306;
S312, judge whether all trips are disposed, be then, sequence executes step S313 and otherwise returns to step S305;
S313, path forms time prediction recombination and Dynamic Announce, by all path forms time final fusion forecasting sequences, By path, the logical order of classifying and numbering and sequential carries out recombination arrangement, obtains the final forecasting sequence of each path, and on road Online chronologically Dynamic Announce;
S314, terminate this path query;
Wherein, the long-term prediction refers to obtaining section or the long-term prediction value of path forms time by LRCF algorithms, when long The value range of prediction duration is set as 0 minute to 3 months, the prediction duration base that specific long-term prediction duration is selected in user It is obtained on plinth by linearly converting;The short-term prediction refers to obtaining the short of section or path forms time by SRFF algorithms When predicted value, the value range of short-term prediction duration is set as 0 minute to 3 hour, and specific short-term prediction duration is in user It is obtained on the basis of selected prediction duration by linearly converting.
CN201610075689.3A 2016-02-02 2016-02-02 Journey time fusion forecasting and querying method based on traffic big data CN105679021B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610075689.3A CN105679021B (en) 2016-02-02 2016-02-02 Journey time fusion forecasting and querying method based on traffic big data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610075689.3A CN105679021B (en) 2016-02-02 2016-02-02 Journey time fusion forecasting and querying method based on traffic big data

Publications (2)

Publication Number Publication Date
CN105679021A CN105679021A (en) 2016-06-15
CN105679021B true CN105679021B (en) 2018-11-06

Family

ID=56303439

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610075689.3A CN105679021B (en) 2016-02-02 2016-02-02 Journey time fusion forecasting and querying method based on traffic big data

Country Status (1)

Country Link
CN (1) CN105679021B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106251625B (en) * 2016-08-18 2019-10-01 上海交通大学 Three-dimensional urban road network global state prediction technique under big data environment
CN106781468B (en) * 2016-12-09 2018-06-15 大连理工大学 Link Travel Time Estimation method based on built environment and low frequency floating car data
CN106988588B (en) * 2016-12-12 2019-07-26 蔚来汽车有限公司 Parking stall floor recognition methods based on two dimensional motion track
CN107862865A (en) * 2017-10-27 2018-03-30 沈阳世纪高通科技有限公司 The Forecasting Methodology and device of a kind of trip information
CN108052979A (en) * 2017-12-15 2018-05-18 阿里巴巴集团控股有限公司 The method, apparatus and equipment merged to model predication value

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102081859B (en) * 2009-11-26 2014-10-01 上海遥薇(集团)有限公司 One kind of bus arrival time prediction model control method
CN102646332B (en) * 2011-02-21 2014-03-12 日电(中国)有限公司 Traffic state estimation device and method based on data fusion
CN102903240B (en) * 2012-10-09 2016-03-16 潮州市创佳电子有限公司 Based on real-time traffic car Beidou terminal perceptual system
CN103106793B (en) * 2013-01-11 2015-02-04 福州大学 Traffic state discriminated method based on real-time driving direction and transit time quantum information
CN103678917B (en) * 2013-12-13 2016-11-23 杭州易和网络有限公司 One kind of real-time bus arrival time prediction method based on simulated annealing algorithm
WO2015134311A1 (en) * 2014-03-03 2015-09-11 Inrix Inc Traffic obstruction detection
CN104064023B (en) * 2014-06-18 2016-12-07 银江股份有限公司 Dynamic traffic forecasting method based on spatio-temporal association
CN104217593B (en) * 2014-08-27 2016-05-25 北京航空航天大学 Traffic information oriented mobile phone speed real-time access method
TWI522974B (en) * 2014-10-06 2016-02-21 Chunghwa Telecom Co Ltd Arrival time prediction system and method

Also Published As

Publication number Publication date
CN105679021A (en) 2016-06-15

Similar Documents

Publication Publication Date Title
Lawrance et al. A new autoregressive time series model in exponential variables (NEAR (1))
Mukerji et al. Flood forecasting using ANN, neuro-fuzzy, and neuro-GA models
Moretti et al. Urban traffic flow forecasting through statistical and neural network bagging ensemble hybrid modeling
Vernieuwe et al. Comparison of data-driven Takagi–Sugeno models of rainfall–discharge dynamics
Dharia et al. Neural network model for rapid forecasting of freeway link travel time
Sfetsos A comparison of various forecasting techniques applied to mean hourly wind speed time series
Chau A split-step particle swarm optimization algorithm in river stage forecasting
Chang et al. An improved neural network-based approach for short-term wind speed and power forecast
Bliemer Dynamic queuing and spillback in analytical multiclass dynamic network loading model
Yalcinoz et al. Short term and medium term power distribution load forecasting by neural networks
Baratti et al. River flow forecast for reservoir management through neural networks
Wu et al. Evolving RBF neural networks for rainfall prediction using hybrid particle swarm optimization and genetic algorithm
Xie et al. Work travel mode choice modeling with data mining: decision trees and neural networks
Roy et al. Spatial interaction modelling
Wey Model formulation and solution algorithm of traffic signal control in an urban network
EP2375364A1 (en) Method and system for time-dependent routing
Krishnankutty Long-range monsoon rainfall prediction of 2005 for the districts and sub-division Kerala with artificial neural network
Pan et al. State space neural networks for short term rainfall-runoff forecasting
Archibald et al. Nested Benders decomposition and dynamic programming for reservoir optimisation
Sözen et al. Solar potential in Turkey
Afan et al. ANN based sediment prediction model utilizing different input scenarios
Haghani et al. A system dynamics approach to land use/transportation system performance modeling part I: Methodology
Azadeh et al. A hybrid simulation-adaptive network based fuzzy inference system for improvement of electricity consumption estimation
Agarwal et al. Runoff modelling through back propagation artificial neural network with variable rainfall-runoff data
Ibrahim et al. A novel hybrid model for hourly global solar radiation prediction using random forests technique and firefly algorithm

Legal Events

Date Code Title Description
PB01 Publication
C06 Publication
SE01 Entry into force of request for substantive examination
C10 Entry into substantive examination
TA01 Transfer of patent application right

Effective date of registration: 20171113

Address after: 400060 Chongqing Nan'an District University Avenue, No. 33

Applicant after: China Merchants Chongqing Communications Research & Design Institute Co., Ltd.

Address before: 401336, Chongqing City, South Bank Road, No. 8 Jade Road, science and technology innovation center, building No. 7, British building (via the expansion zone)

Applicant before: Chongqing Yun Tu transport science and techonologies company limited

TA01 Transfer of patent application right
GR01 Patent grant
GR01 Patent grant