Journey time fusion forecasting and querying method based on the big data of traffic
Technical field
The present invention relates to the traffic status prediction technology of road network or bicycle, be related specifically to a kind of journey time fusion forecasting based on the big data of traffic and querying method.
Background technology
Existing road network or bicycle traffic status prediction method mainly utilize the traffic data that Floating Car or roadside device provide, and even only rely on the traffic data of roadside device offer to predict the traffic behavior of road network or bicycle. This kind of method directly obtains the transport information such as road network flow, average speed mainly through roadside device, or obtain certain class transport information by floating car data being carried out secondary operations, and based on road network topology, model recursion mode is adopted to be predicted, the transport information of prediction is concentrated mainly on traffic three elements, causes that popularization face is wideless, and promotional value is not high yet, namely there is versatility problem, this causes bigger puzzlement to operation expanding. And; existing method is generally from up time to going to consider the impact of floating car data; the time factor of floating car data would generally be neglected; and adopt linear mathematical calculation mode; such as summation, ask impartial computing; departing from the primitive character of road grid traffic, causing that the extraction accuracy of transport information and stability are not high, practicality is not strong yet. Particularly when baroque magnanimity floating car data, existing road network or bicycle traffic status prediction method generally seem at a loss what to do. Obviously, existing road network or bicycle traffic status prediction technology also exist that real-time is poor, versatility is poor and the problem such as practicality is not strong.
Summary of the invention
For the problem such as solve that the real-time that existing road network or bicycle traffic status prediction technology exist is poor, versatility is poor and practicality is not strong, the present invention proposes a kind of journey time fusion forecasting based on the big data of traffic and querying method.
The present invention is based on the journey time fusion forecasting method of the big data of traffic, and the data that all online vehicles are uploaded carry out calculated off line or training obtains all kinds of forecast model and parameter, sets up according to all kinds of forecast models and parameter and dynamically updates data dictionary;Call all kinds of forecast model and parameter, and in conjunction with the travel time data of real-time section and path, the traffic behavior of road network or bicycle is predicted; Described data dictionary includes vehicle data dictionary, section data dictionary and path data dictionary; Described online vehicle refers to login network access and automatically uploads the vehicle of location and speed data.
Further, the present invention is based on the journey time fusion forecasting method of the big data of traffic, the data that all online vehicles are uploaded carry out calculated off line or training, to obtain all kinds of forecast model and parameter, including, set every day from 0:00 time to 24:00 time as a complete cycle cycle, according to the trip mode of working day, Saturday, Sunday or festivals or holidays, travel time data is carried out trip Type division, what obtain trip type goes out line number; The travel time data in each cycle is arranged in a journey time sequence according to the sequencing of time.
Further, the present invention is based on the journey time fusion forecasting method of the big data of traffic, and described all kinds of forecast models and parameter include fusion forecasting model and intersection delay relation analysis model when section or path forms time data periods rules model, section or the statistical rules model of path forms time, section or path forms time long-term prediction model, section or path forms time short-time forecasting model, section or path forms time length; Wherein,
Adopt periodic law series approximation PLSA(PeriodicLawSeriesApproximation, PLSA) algorithm obtains section or path forms time data periods rules model, and adopt method of least square LSM(LeastSquareMethod, LSM) solve approximate model parameter;
Statistical law is adopted to extract SRE(StatisticalRuleExtraction, SRE) algorithm obtains section or path forms time statistical rules model, and adopt Density Estimator KDE(KernelDensityEstimation, KDE) obtain the probability density Changing Pattern of section or path forms time;
Correction prediction LRCF(Long-timeRollingCorrectionForecast is rolled when adopting long, LRCF) algorithm obtains section or path forms time long-term prediction model, and obtain long-term prediction model parameter by calculated off line, by quickly realizing section or path forms time prediction in line computation;
Adopt and roll matching prediction SRFF(Short-timeRollingFittingForecast in short-term, SRFF) algorithm obtains section or path forms time short-time forecasting model, and adopt time series autoregressive moving average ARMA (Auto-RegressiveandMovingAverage, ARMA) algorithm construction short-time forecasting model, and adopt method of least square LSM(LeastSquareMethod, LSM) solving model parameter;
Adopt sieve-like fusion forecasting SFF(SieveFusionForecast, SFF) algorithm obtains fusion forecasting model when section or path forms time length, and pass through off-line training, adopt Gaussian-Newton method GNIM(Guassian-NewtonIterativeMethod, GNIM) obtain fusion forecasting model parameter;
Adopt intersection delay association analysis IDCA(IntersectionDelayCorrelationAnalysis, IDCA) algorithm construction intersection delay relation analysis model, and by method of least square LSM(LeastSquareMethod, LSM) solving model parameter, model parameter is obtained, by the quick compensation in line computation realizing route journey time by off-line training;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes;Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
Further, the present invention is based on the journey time fusion forecasting method of the big data of traffic, and described section data dictionary is for storing the periodic sequence of journey time, probability sequence and various model and parameter, and the storage of its data and renewal comprise the following steps:
S101, reading historical data, read historical data from the data dictionary of section, including section numbering, the Link Travel Time data of even time interval sampling, date and hour data;
S102, choose section, choose a untreated section according to the sequencing of section numbering;
S103, Link Travel Time data are classified, theoretical based on road network tidal current, set every day from 0:00 time to 24:00 time as a complete cycle cycle, according to working day, Saturday, Sunday, the trip mode of festivals or holidays, travel time data is carried out trip Type division and numbering, namely going out line number, the travel time data in each cycle is arranged in a journey time sequence;
S104, choose Link Travel Time data, choose the untreated Link Travel Time data of class according to the sequencing going out line number;
S105, acquisition travel time data periods rules model and parameter, adopt PLSA algorithm to obtain section or path forms time data periods rules model, and adopt LSM to solve approximate model parameter; Including:
S1051, journey time sequence to any two different cycles carry out degree of association cluster analysis, extract the big travel time data of degree of association and form set and carry out asking all calculating, it is thus achieved that average travel time sequence;
S1052, based on " 4 π/hour " pi, construct a Fourier space model and approach average travel time sequence, solved by LSM and approach equation and obtain model parameter;
S1053, according to energy order from high to low, intercept gross energy >=98% model parameter, all the other parameter zero setting, thus obtaining PLSA model parameter;
S1054, by PLSA algorithm generate Link Travel Time a periodic sequence, the information such as this periodic sequence and PLSA model parameter are stored in the data dictionary of section;
S106, acquisition Link Travel Time statistical law model and parameter, adopt SRE algorithm to obtain section or the statistical rules model of path forms time, and adopt KDE to obtain section or the probability density Changing Pattern of path forms time; Including:
S1061, demarcation sometime, are selected the travel time data in this moment of all cycles and are formed data set, obtain probability density function with KDE, find the journey time that probability density maximum is corresponding, i.e. maximum probability journey time;
S1062, solve maximum probability journey time corresponding to all moment, be arranged in the probability sequence of Link Travel Time chronologically and store in the data dictionary of section;
S107, acquisition Link Travel Time long-term prediction model and parameter, adopt LRCF algorithm to obtain Link Travel Time long-term prediction model and parameter, and obtain long-term prediction model parameter by calculated off line; Including:
S1071, by the periodic sequence of journey time and probability sequence sum-average arithmetic, it is thus achieved that value sequence at the beginning of during journey time long
S1072, demarcating sometime, calculating all cycles compares with initial value time long in journey time corresponding to this moment, it is thus achieved that difference, and is arranged in a sequence of differences chronologically;
S1073, the natural law spacing considered between adjacent difference, construct a binary repeatedly multinomial model and approach this sequence of differences, and the polynomial item number of self-adaptative adjustment finds the multinomial model that minimum fitness bias is corresponding;
S1074, the process solving multinomial model are LRCF algorithm, obtain LRCF model parameter corresponding to all moment and are stored in the data dictionary of section;
S108, the short-time forecasting model obtaining Link Travel Time and parameter, adopt SRFF algorithm to obtain Link Travel Time short-time forecasting model, and adopt ARMA algorithm construction short-time forecasting model, and adopt LSM solving model parameter; Including:
S1081, chronologically rhythmic for institute journey time series arrangement is become a long sequence;
S1082, suppose that the time interval of this long sequence is impartial, with ARMA algorithm construction oneNAfter item multinomial model carrys out matchingNIndividual journey time, asks its parameter and error of fitting with LSM;
S1083, by adjustNRegulate the size of error of fitting, choose error minimum time corresponding multinomial model;
S1084, the process solving this multinomial model are SRFF algorithm, are stored in the data dictionary of section by SRFF model parameter;
Fusion forecasting model and parameter when S109, acquisition Link Travel Time length, adopt SFF algorithm to obtain fusion forecasting model during Link Travel Time length, and by off-line training, adopts GNIM acquisition fusion forecasting model parameter; Including:
S1091, demarcation initial time, obtain the item number of SRFF modelN, construct one 2 ×NSieve-like coefficient matrix, wherein, the element sum perseverance that each element is non-negative and every string is 1, and element value is unknown;
S1092, start journey time is predicted from initial time, value sequence at the beginning of when compensating long by LRCF algorithm, obtain futureNIndividual long-term prediction value, obtains future by SRFF algorithmNIndividual short-term prediction value, by both predictive value sequences form one 2 ×NPrediction matrix;
S1093, coefficient matrix is added process with row after prediction matrix dot product, it is thus achieved that one 1 ×NMerge vector, approach corresponding journey time sequence with fusion vector, it is thus achieved that corresponding correlation coefficient equation;
S1094, progressively adjust initial time backward, obtain corresponding correlation coefficient equation by same method, be made up of a correlation coefficient equation group these equations;
S1095, solve equation group with GNIM, it is thus achieved that the element value of sieve-like coefficient matrix, namely SFF model parameter, SFF model parameter is stored in the data dictionary of section;
S110, judge whether that all Link Travel Time are disposed? being then, order performs step S111, otherwise, returns and performs step S104;
Do you S111, judge that all sections are disposed? being then, order performs step S112, otherwise, returns and performs step S102;
S112, the data storage terminating this section data dictionary and renewal;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
Further, the present invention is based on the journey time fusion forecasting method of the big data of traffic, and described path data dictionary is for storing the periodic sequence of journey time, probability sequence and various model and parameter, and the storage of its data and renewal comprise the following steps:
S201, reading historical data, reading historical data from path and section data dictionary, the section sampled including road-net node, section numbering, path number, even time interval and path forms time data, date and hour data;
S202, choose combination of nodes, the permutation and combination relation according to road-net node numbering, choose one group of untreated two combination of nodes;
S203, path selection, choose a untreated path according to two combination of nodes;
S204, path forms time data are classified, theoretical based on road network tidal current, set every day from 0:00 time to 24:00 time as a complete cycle cycle, according to working day, Saturday, Sunday, the trip mode of festivals or holidays, travel time data is carried out trip Type division and numbering, namely going out line number, the travel time data in each cycle is arranged in a journey time sequence;
S205, path selection travel time data, choose a untreated path forms time data of class according to the sequencing going out line number;
S206, acquisition travel time data periods rules model and parameter, adopt PLSA algorithm to obtain section or path forms time data periods rules model, and adopt LSM to solve approximate model parameter; Including:
S2061, journey time sequence to any two different cycles carry out degree of association cluster analysis, extract the big travel time data of degree of association and form set and carry out asking all calculating, it is thus achieved that average travel time sequence;
S2062, based on " 4 π/hour " pi, construct a Fourier space model and approach average travel time sequence, solved by LSM and approach equation and obtain model parameter;
S2063, according to energy order from high to low, intercept gross energy >=98% model parameter, all the other parameter zero setting, thus obtaining PLSA model parameter;
S2064, by PLSA algorithm generate the path forms time a periodic sequence, the information such as this periodic sequence and PLSA model parameter are stored in path data dictionary;
S207, acquisition Link Travel Time statistical law model and parameter, adopt SRE algorithm to obtain section or the statistical rules model of path forms time, and adopt KDE to obtain section or the probability density Changing Pattern of path forms time; Including:
S2071, demarcation sometime, are chosen the travel time data in this moment of all cycles and are formed data set, obtain probability density function with KDE, find the journey time that probability density maximum is corresponding, i.e. maximum probability journey time;
S2072, solve maximum probability journey time corresponding to all moment, be arranged in the probability sequence of path forms time chronologically and store in path data dictionary;
S208, acquisition intersection delay relation analysis model and parameter, adopt IDCA algorithm construction intersection delay relation analysis model, and by LSM solving model parameter, and obtain model parameter by off-line training; Including:
S2081, demarcate certain cycle sometime, calculate all Link Travel Time sums in this path and path forms time and Link Travel Time and difference, this difference is crossing total delay;
S2082, travel through all cycles and all moment, calculate all crossings total delay, set up the incidence relation between path forms time with corresponding crossing total delay by LSM and solve, it is thus achieved that LSM model parameter;
S2083, all crossings total delay is arranged chronologically, it is contemplated that the natural law spacing between consecutive value, construct a many order polynomial of binary and approach this sequence, the polynomial item number of self-adaptative adjustment, finds the multinomial model that minimum fitness bias is corresponding;
S2084, LSM model of fit and multinomial model being carried out equal weight merging, pooled model is IDCA model, IDCA model parameter is stored in path data dictionary;
S209, acquisition approach journey time long-term prediction model and parameter, adopt LRCF algorithm to obtain section or path forms time long-term prediction model, and obtain long-term prediction model parameter by calculated off line; Including:
S2091, by the periodic sequence of journey time and probability sequence sum-average arithmetic, it is thus achieved that value sequence at the beginning of during journey time long
S2092, demarcating sometime, calculating all cycles compares with initial value time long in journey time corresponding to this moment, it is thus achieved that difference, and is arranged in a sequence of differences chronologically;
S2093, the natural law spacing considered between adjacent difference, construct a binary repeatedly multinomial model and approach this sequence of differences, and the polynomial item number of self-adaptative adjustment finds the multinomial model that minimum fitness bias is corresponding;
S2094, the process solving multinomial model are LRCF algorithm, obtain LRCF model parameter corresponding to all moment and are stored in path data dictionary;
S210, acquisition approach journey time short-time forecasting model and parameter, adopt SRFF algorithm acquisition approach journey time short-time forecasting model, and adopt ARMA algorithm construction short-time forecasting model, and adopt LSM solving model parameter; Including:
S2101, chronologically rhythmic for institute journey time series arrangement is become a long sequence;
S2102, suppose that the time interval of this long sequence is impartial, with ARMA algorithm construction oneNAfter item multinomial model carrys out matchingNIndividual journey time, asks its parameter and error of fitting with LSM;
S2103, by adjustNRegulate the size of error of fitting, choose error minimum time corresponding multinomial model;
S2104, the process solving this multinomial model are SRFF algorithm, are stored in path data dictionary by SRFF model parameter;
Fusion forecasting model and parameter when S211, acquisition approach journey time length, fusion forecasting model during employing SFF algorithm acquisition approach journey time length, and by off-line training, adopt GNIM to obtain fusion forecasting model parameter; Including:
S2111, demarcation initial time, obtain the item number of SRFF modelN, construct one 2 ×NSieve-like coefficient matrix, wherein each element is the element sum perseverance of non-negative and every string is 1, and element value is unknown;
S2112, start journey time is predicted from initial time, value sequence at the beginning of when compensating long by LRCF algorithm, obtain futureNIndividual long-term prediction value, obtains future by SRFF algorithmNIndividual short-term prediction value, by both predictive value sequences form one 2 ×NPrediction matrix;
S2112, coefficient matrix is added process with row after prediction matrix dot product, it is thus achieved that one 1 ×NMerge vector, approach corresponding journey time sequence with fusion vector, it is thus achieved that corresponding correlation coefficient equation;
S2114, progressively adjust initial time backward, obtain corresponding correlation coefficient equation by same method, be made up of a correlation coefficient equation group these equations;
S2115, solve equation group with GNIM, it is thus achieved that the element value of sieve-like coefficient matrix, namely SFF model parameter, SFF model parameter is stored in path data dictionary;
Do you S212, judge that all path forms time datas are disposed? being then, order performs step S213, otherwise, returns and performs step S205;
S213, judging whether all paths are disposed, be then, order performs step S214, otherwise, returns and performs step S203;
S214, judging whether all combination of nodes are disposed, be then, order performs step S215, otherwise, returns and performs step S202;
S215, the data storage terminating this path data dictionary and renewal;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
The present invention is based on the journey time fusion query method of the big data of traffic, adopt the present invention based on the big data of traffic journey time fusion forecasting method to set query path and go out line number journey time change be predicted and will predict the outcome being supplied to user, including, adopt the data that all online vehicles are uploaded obtain all kinds of forecast models and parameter by calculated off line or training and set up according to all kinds of forecast models and parameter and dynamically update data dictionary; According to user set query path and go out line number, call all kinds of forecast model and the parameter real-time travel time data in conjunction with this path and respective stretch and be predicted the journey time in this path and respective stretch and will predict the outcome being supplied to user; Described data dictionary includes vehicle data dictionary, section data dictionary and path data dictionary; Described online vehicle refers to login network access and automatically uploads the vehicle of location and speed data.
Further, the present invention is based on the journey time fusion query method of the big data of traffic, described go out line number refer to based on road network tidal current theoretical, set every day from 0:00 time to 24:00 time as a complete cycle cycle, according to working day, Saturday, Sunday, the trip mode of festivals or holidays, travel time data carrying out trip Type division and numbering, namely goes out line number, the travel time data in each cycle is arranged in a journey time sequence.
Further, the present invention is based on the journey time fusion query method of the big data of traffic, and described all kinds of forecast models and parameter include fusion forecasting model and intersection delay relation analysis model when section or path forms time data periods rules model, section or the statistical rules model of path forms time, section or path forms time long-term prediction model, section or path forms time short-time forecasting model, section or path forms time length; Wherein,
Adopt periodic law series approximation PLSA(PeriodicLawSeriesApproximation, PLSA) algorithm obtains section or path forms time data periods rules model, and adopt method of least square LSM(LeastSquareMethod, LSM) solve approximate model parameter;
Statistical law is adopted to extract SRE(StatisticalRuleExtraction, SRE) algorithm obtains section or path forms time statistical rules model, and adopt Density Estimator KDE(KernelDensityEstimation, KDE) obtain the probability density Changing Pattern of section or path forms time;
Correction prediction LRCF(Long-timeRollingCorrectionForecast is rolled when adopting long, LRCF) algorithm obtains section or path forms time long-term prediction model, and obtain long-term prediction model parameter by calculated off line, by quickly realizing section or path forms time prediction in line computation;
Adopt and roll matching prediction SRFF(Short-timeRollingFittingForecast in short-term, SRFF) algorithm obtains section or path forms time short-time forecasting model, and adopt time series autoregressive moving average ARMA (Auto-RegressiveandMovingAverage, ARMA) algorithm construction short-time forecasting model, and adopt method of least square LSM(LeastSquareMethod, LSM) solving model parameter;
Adopt sieve-like fusion forecasting SFF(SieveFusionForecast, SFF) algorithm obtains fusion forecasting model when section or path forms time length, and pass through off-line training, adopt Gaussian-Newton method GNIM(Guassian-NewtonIterativeMethod, GNIM) obtain fusion forecasting model parameter;
Adopt intersection delay association analysis IDCA(IntersectionDelayCorrelationAnalysis, IDCA) algorithm construction intersection delay relation analysis model, and by method of least square LSM(LeastSquareMethod, LSM) solving model parameter, model parameter is obtained, by the quick compensation in line computation realizing route journey time by off-line training;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
Further, the present invention, based on the journey time fusion query method of the big data of traffic, comprises the following steps:
S301, user's inquiry;
S302, collection history and real time traffic data, the section sampled including road-net node numbering, section numbering, path number, even time interval and path forms time, all kinds of Forecasting of Travel Time data, all kinds of model parameter, date and hour, data classifying and numbering;
S303, inquire about set path journey time change? it is then, continues executing with step S304, otherwise, perform step S314;
S304, querying condition are arranged, user inquires about content according to oneself requirements set, including the period of the duration of short-term prediction and long-term prediction, according to even time interval principle, demarcate corresponding in short-term with the starting point and ending point of long-term prediction step-length, and set according to calendar and go out line number accordingly; Meanwhile, user sets the starting point and ending point of road-net node, by searching route data dictionary, demarcates the path between the two road-net node and path number;
S305, path selection travel time data, choose a untreated path forms time data of class according to the sequencing going out line number;
S306, path selection, choose a untreated path according to the sequencing of path number, by the section numbering that path data dictionary search to this path comprises;
S307, choose section, choose a untreated section according to the sequencing of section numbering;
S308, Link Travel Time fusion forecasting, read corresponding model and parameter according to user's search request from the data dictionary of section, in conjunction with real-time travel time data, as follows by step process:
S3081, by LRCF algorithm obtain journey time long time at the beginning of value sequence and sequence of differences, both additions obtain the long-term prediction sequence of journey time;
S3082, the short-term prediction step-length chosen with user starting point and ending point for benchmark, obtained the short-term prediction sequence of journey time by SRFF algorithm, obtain fusion forecasting sequence in short-term again through SFF algorithm;
S3083, the long-term prediction step-length chosen with user starting point and ending point be as the criterion, expand by long-term prediction sequence pair fusion forecasting sequence in short-term, obtain the final fusion forecasting sequence of Link Travel Time;
Do you S309, judge that all sections are disposed?Being then, order performs step S310, otherwise, returns and performs step S307;
S310, path forms Fusion in Time are predicted, set according to user in predicting, as follows by step process:
S3101, obtained the forecasting sequence of crossing total delay by IDCA algorithm, according to temporal order, the forecasting sequence of crossing total delay is added with the final fusion forecasting sequence in comprised section, this path, obtains the compensation forecasting sequence of journey time;
S3102, by LRCF algorithm obtain journey time long time at the beginning of value sequence and sequence of differences, namely both additions obtain the long-term prediction sequence of journey time;
S3103, the short-term prediction step-length chosen with user starting point and ending point for benchmark, obtained the short-term prediction sequence of journey time by SRFF algorithm, obtain fusion forecasting sequence in short-term again through SFF algorithm;
S3104, the long-term prediction step-length chosen with user starting point and ending point be as the criterion, expand by long-term prediction sequence pair fusion forecasting sequence in short-term, obtain the expansion type forecasting sequence of journey time;
S3105, expansion type forecasting sequence and compensation forecasting sequence are added it are averaging, obtain the final fusion forecasting sequence of path forms time;
Do you S311, judge that all paths are disposed? being then, order performs step S312, otherwise, returns and performs step S306;
Do you S312, judge that all trips are disposed? being then, order performs step S313, otherwise, returns and performs step S305;
S313, path forms time prediction restructuring and Dynamic Announce, by final for all of path forms time fusion forecasting sequence, by path, the logical order of classifying and numbering and sequential carry out restructuring arrangement, obtain the final forecasting sequence of every paths, and on road network Dynamic Announce chronologically;
S314, terminate this path query;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
The present invention is overcome the problems such as the real-time that the traffic status prediction method of prior art road network or bicycle exists is poor, versatility is poor and practicality is not strong based on the journey time fusion forecasting of the big data of traffic and the Advantageous Effects of querying method, survey precision is greatly improved, the speed of on-line prediction is greatly improved, it is ensured that real-time in engineer applied, practicality and versatility simultaneously.
Accompanying drawing explanation
Accompanying drawing 1 is section of the present invention data dictionary data storage and updates schematic flow sheet;
Accompanying drawing 2 is path data dictionary data of the present invention storage and updates schematic flow sheet;
Accompanying drawing 3 is path forms time of the present invention change querying flow schematic diagram.
Below in conjunction with the drawings and specific embodiments, the present invention journey time fusion forecasting based on the big data of traffic and querying method are further described.
Detailed description of the invention
The present invention, based on the journey time fusion forecasting method of the big data of traffic, adopts the data that all online vehicles are uploaded to obtain all kinds of forecast models and parameter by calculated off line or training, and sets up according to all kinds of forecast models and parameter and dynamically update data dictionary;Call all kinds of forecast model and parameter in conjunction with real-time section and path forms time data, the traffic behavior of road network or bicycle to be predicted; Described data dictionary includes vehicle data dictionary, section data dictionary and path data dictionary; Described online vehicle refers to login network access and automatically uploads the vehicle of location and speed data. Compared with prior art Forecasting Methodology, the inventive method combines big data technique, and each data dictionary contains multiple index, and data storage has better systematicness, can meet the requirement of real-time of data storage and read-write very well. Of particular concern is, fully taking into account the requirement to aspects such as precision of prediction, real-time, versatility and practicality in engineer applied, most amounts of calculation of this case, the most complicated calculating task complete each through off-line. It is to say, this case obtains all kinds of prediction model parameters by off-line training, the forecast model having only to call off-line generation in line computation carries out simple computation. This have the effect that: can better be improved the precision of prediction of forecast model by calculated off line, on-line prediction speed can also be greatly improved simultaneously, reduce on-line calculation so that prediction can real-time implementation, thus the real-time that ensure that in engineer applied and practicality.
Additionally, the present invention is based on the journey time fusion forecasting method of the big data of traffic, the data that all online vehicles are uploaded are adopted to obtain all kinds of forecast models and parameter by calculated off line or training, including, set every day from 0:00 time to 24:00 time as a complete cycle cycle, according to working day, Saturday, Sunday, the trip mode of festivals or holidays, travel time data is carried out division and the numbering of trip type, namely goes out line number; The travel time data in each cycle is arranged in a journey time sequence. Advantage of this is that: the periodic characteristic in conjunction with road net traffic state divides, better met the substitutive characteristics of road grid traffic, the precision of Forecasting of Travel Time can be greatly improved, can better hold the periodic law of road net traffic state simultaneously.
The present invention is based on the journey time fusion forecasting method of the big data of traffic, and described all kinds of forecast models and parameter include fusion forecasting model and intersection delay relation analysis model when section or path forms time data periods rules model, section or the statistical rules model of path forms time, section or path forms time long-term prediction model, section or path forms time short-time forecasting model, section or path forms time length; Wherein,
Adopt periodic law series approximation PLSA(PeriodicLawSeriesApproximation, PLSA) algorithm obtains section or path forms time data periods rules model, and adopt method of least square LSM(LeastSquareMethod, LSM) solve approximate model parameter. Thus, the cyclically-varying rule of section or path forms time can be extracted preferably, simultaneously, moreover it is possible to arithmetic speed is greatly improved, there is good versatility and engineer applied is worth.
Statistical law is adopted to extract SRE(StatisticalRuleExtraction, SRE) algorithm obtains section or path forms time statistical rules model, and adopt Density Estimator KDE(KernelDensityEstimation, KDE) obtain the probability density Changing Pattern of section or path forms time. Compared to model-driven, more respect the Self-variation rule of Link Travel Time.SRE algorithm energy rapid extraction section or periods rules information the most stable in the path forms time, and computational complexity is not high yet, has good versatility and engineer applied is worth.
Correction prediction LRCF(Long-timeRollingCorrectionForecast is rolled when adopting long, LRCF) algorithm obtains section or path forms time long-term prediction model, and obtain long-term prediction model parameter by calculated off line, by quickly realizing section or path forms time prediction in line computation. Can be good at solving the real time problems in engineer applied with this, meanwhile, good precision of prediction also ensure that the engineering practicability of this algorithm.
Adopt and roll matching prediction SRFF(Short-timeRollingFittingForecast in short-term, SRFF) algorithm obtains section or path forms time short-time forecasting model, and adopt time series autoregressive moving average ARMA (Auto-RegressiveandMovingAverage, ARMA) algorithm construction short-time forecasting model, and adopt method of least square LSM(LeastSquareMethod, LSM) solving model parameter. The method has respected fully the tidal fluctuations rule of section or path forms time. Owing to this algorithm obtains short-time forecasting model parameter by off-line training, the amount of calculation of on-line prediction is very little, it is possible to well solving the real time problems in engineer applied, meanwhile, good precision of prediction also ensure that the engineering practicability of this algorithm.
Adopt sieve-like fusion forecasting SFF(SieveFusionForecast, SFF) algorithm obtains fusion forecasting model when section or path forms time length, and pass through off-line training, adopt Gaussian-Newton method GNIM(Guassian-NewtonIterativeMethod, GNIM) obtain fusion forecasting model parameter. This algorithm decreases LRCF algorithm and SRFF algorithm part weakness to a certain extent, and the advantage simultaneously strengthening again them, is LRCF algorithm and the effective integration of SRFF algorithm. This algorithm passes through off-line training, adopts GNIM to obtain fusion forecasting model parameter, precision of prediction can be effectively greatly improved, the speed of on-line prediction is greatly improved simultaneously, it is ensured that real-time in engineer applied and practicality.
Adopt intersection delay association analysis IDCA(IntersectionDelayCorrelationAnalysis, IDCA) algorithm construction intersection delay relation analysis model, and by method of least square LSM(LeastSquareMethod, LSM) solving model parameter, model parameter is obtained, by the quick compensation in line computation realizing route journey time by off-line training. The method can effectively make up the deviation that path journey time is caused by intersection delay. Owing to IDCA algorithm obtains model parameter by off-line training, by the quick compensation in line computation realizing route journey time, both ensure that computational accuracy, in turn ensure that the speed in line computation simultaneously, there is good engineering practicability.
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
Accompanying drawing 1 is section of the present invention data dictionary data storage and updates schematic flow sheet, as seen from the figure, the present invention is based on the journey time fusion forecasting method of the big data of traffic, described section data dictionary is for storing the periodic sequence of journey time, probability sequence and various model and parameter, and the storage of its data and renewal comprise the following steps:
S101, reading historical data, read historical data from the data dictionary of section, including section numbering, the Link Travel Time data of even time interval sampling, date and hour data;
S102, choose section, choose a untreated section according to the sequencing of section numbering;
S103, Link Travel Time data are classified, theoretical based on road network tidal current, set every day from 0:00 time to 24:00 time as a complete cycle cycle, according to working day, Saturday, Sunday, the trip mode of festivals or holidays, travel time data is carried out trip Type division and numbering, namely going out line number, the travel time data in each cycle is arranged in a journey time sequence;
S104, choose Link Travel Time data, choose the untreated Link Travel Time data of class according to the sequencing going out line number;
S105, acquisition travel time data periods rules model and parameter, adopt PLSA algorithm to obtain section or path forms time data periods rules model, and adopt LSM to solve approximate model parameter; Including:
S1051, journey time sequence to any two different cycles carry out degree of association cluster analysis, extract the big travel time data of degree of association and form set and carry out asking all calculating, it is thus achieved that average travel time sequence;
S1052, based on " 4 π/hour " pi, construct a Fourier space model and approach average travel time sequence, solved by LSM and approach equation and obtain model parameter;
S1053, according to energy order from high to low, intercept gross energy >=98% model parameter, all the other parameter zero setting, thus obtaining PLSA model parameter;
S1054, by PLSA algorithm generate Link Travel Time a periodic sequence, the information such as this periodic sequence and PLSA model parameter are stored in the data dictionary of section;
S106, acquisition Link Travel Time statistical law model and parameter, adopt SRE algorithm to obtain section or the statistical rules model of path forms time, and adopt KDE to obtain section or the probability density Changing Pattern of path forms time; Including:
S1061, demarcation sometime, are selected the travel time data in this moment of all cycles and are formed data set, obtain probability density function with KDE, find the journey time that probability density maximum is corresponding, i.e. maximum probability journey time;
S1062, solve maximum probability journey time corresponding to all moment, be arranged in the probability sequence of Link Travel Time chronologically and store in the data dictionary of section;
S107, acquisition Link Travel Time long-term prediction model and parameter, adopt LRCF algorithm to obtain Link Travel Time long-term prediction model and parameter, and obtain long-term prediction model parameter by calculated off line; Including:
S1071, by the periodic sequence of journey time and probability sequence sum-average arithmetic, it is thus achieved that value sequence at the beginning of during journey time long
S1072, demarcating sometime, calculating all cycles compares with initial value time long in journey time corresponding to this moment, it is thus achieved that difference, and is arranged in a sequence of differences chronologically;
S1073, the natural law spacing considered between adjacent difference, construct a binary repeatedly multinomial model and approach this sequence of differences, and the polynomial item number of self-adaptative adjustment finds the multinomial model that minimum fitness bias is corresponding;
S1074, the process solving multinomial model are LRCF algorithm, obtain LRCF model parameter corresponding to all moment and are stored in the data dictionary of section;
S108, the short-time forecasting model obtaining Link Travel Time and parameter, adopt SRFF algorithm to obtain Link Travel Time short-time forecasting model, and adopt ARMA algorithm construction short-time forecasting model, and adopt LSM solving model parameter; Including:
S1081, chronologically rhythmic for institute journey time series arrangement is become a long sequence;
S1082, suppose that the time interval of this long sequence is impartial, with ARMA algorithm construction oneNAfter item multinomial model carrys out matchingNIndividual journey time, asks its parameter and error of fitting with LSM;
S1083, by adjustNRegulate the size of error of fitting, choose error minimum time corresponding multinomial model;
S1084, the process solving this multinomial model are SRFF algorithm, are stored in the data dictionary of section by SRFF model parameter;
Fusion forecasting model and parameter when S109, acquisition Link Travel Time length, adopt SFF algorithm to obtain fusion forecasting model during Link Travel Time length, and by off-line training, adopts GNIM acquisition fusion forecasting model parameter; Including:
S1091, demarcation initial time, obtain the item number of SRFF modelN, construct one 2 ×NSieve-like coefficient matrix, wherein, the element sum perseverance that each element is non-negative and every string is 1, and element value is unknown;
S1092, start journey time is predicted from initial time, value sequence at the beginning of when compensating long by LRCF algorithm, obtain futureNIndividual long-term prediction value, obtains future by SRFF algorithmNIndividual short-term prediction value, by both predictive value sequences form one 2 ×NPrediction matrix;
S1093, coefficient matrix is added process with row after prediction matrix dot product, it is thus achieved that one 1 ×NMerge vector, approach corresponding journey time sequence with fusion vector, it is thus achieved that corresponding correlation coefficient equation;
S1094, progressively adjust initial time backward, obtain corresponding correlation coefficient equation by same method, be made up of a correlation coefficient equation group these equations;
S1095, solve equation group with GNIM, it is thus achieved that the element value of sieve-like coefficient matrix, namely SFF model parameter, SFF model parameter is stored in the data dictionary of section;
S110, judge whether that all Link Travel Time are disposed? being then, order performs step S111, otherwise, returns and performs step S104;
Do you S111, judge that all sections are disposed? being then, order performs step S112, otherwise, returns and performs step S102;
S112, the data storage terminating this section data dictionary and renewal;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
Accompanying drawing 2 is path data dictionary data of the present invention storage and updates schematic flow sheet, as seen from the figure, the present invention is based on the journey time fusion forecasting method of the big data of traffic, described path data dictionary is for storing the periodic sequence of journey time, probability sequence and various model and parameter, and the storage of its data and renewal comprise the following steps:
S201, reading historical data, reading historical data from path and section data dictionary, the section sampled including road-net node, section numbering, path number, even time interval and path forms time data, date and hour data;
S202, choose combination of nodes, the permutation and combination relation according to road-net node numbering, choose one group of untreated two combination of nodes;
S203, path selection, choose a untreated path according to two combination of nodes;
S204, path forms time data are classified, theoretical based on road network tidal current, set every day from 0:00 time to 24:00 time as a complete cycle cycle, according to working day, Saturday, Sunday, the trip mode of festivals or holidays, travel time data is carried out trip Type division and numbering, namely going out line number, the travel time data in each cycle is arranged in a journey time sequence;
S205, path selection travel time data, choose a untreated path forms time data of class according to the sequencing going out line number;
S206, acquisition travel time data periods rules model and parameter, adopt PLSA algorithm to obtain section or path forms time data periods rules model, and adopt LSM to solve approximate model parameter; Including:
S2061, journey time sequence to any two different cycles carry out degree of association cluster analysis, extract the big travel time data of degree of association and form set and carry out asking all calculating, it is thus achieved that average travel time sequence;
S2062, based on " 4 π/hour " pi, construct a Fourier space model and approach average travel time sequence, solved by LSM and approach equation and obtain model parameter;
S2063, according to energy order from high to low, intercept gross energy >=98% model parameter, all the other parameter zero setting, thus obtaining PLSA model parameter;
S2064, by PLSA algorithm generate the path forms time a periodic sequence, the information such as this periodic sequence and PLSA model parameter are stored in path data dictionary;
S207, acquisition Link Travel Time statistical law model and parameter, adopt SRE algorithm to obtain section or the statistical rules model of path forms time, and adopt KDE to obtain section or the probability density Changing Pattern of path forms time; Including:
S2071, demarcation sometime, are chosen the travel time data in this moment of all cycles and are formed data set, obtain probability density function with KDE, find the journey time that probability density maximum is corresponding, i.e. maximum probability journey time;
S2072, solve maximum probability journey time corresponding to all moment, be arranged in the probability sequence of path forms time chronologically and store in path data dictionary;
S208, acquisition intersection delay relation analysis model and parameter, adopt IDCA algorithm construction intersection delay relation analysis model, and by LSM solving model parameter, and obtain model parameter by off-line training; Including:
S2081, demarcate certain cycle sometime, calculate all Link Travel Time sums in this path and path forms time and Link Travel Time and difference, this difference is crossing total delay;
S2082, travel through all cycles and all moment, calculate all crossings total delay, set up the incidence relation between path forms time with corresponding crossing total delay by LSM and solve, it is thus achieved that LSM model parameter;
S2083, all crossings total delay is arranged chronologically, it is contemplated that the natural law spacing between consecutive value, construct a many order polynomial of binary and approach this sequence, the polynomial item number of self-adaptative adjustment, finds the multinomial model that minimum fitness bias is corresponding;
S2084, LSM model of fit and multinomial model being carried out equal weight merging, pooled model is IDCA model, IDCA model parameter is stored in path data dictionary;
S209, acquisition approach journey time long-term prediction model and parameter, adopt LRCF algorithm to obtain section or path forms time long-term prediction model, and obtain long-term prediction model parameter by calculated off line; Including:
S2091, by the periodic sequence of journey time and probability sequence sum-average arithmetic, it is thus achieved that value sequence at the beginning of during journey time long
S2092, demarcating sometime, calculating all cycles compares with initial value time long in journey time corresponding to this moment, it is thus achieved that difference, and is arranged in a sequence of differences chronologically;
S2093, the natural law spacing considered between adjacent difference, construct a binary repeatedly multinomial model and approach this sequence of differences, and the polynomial item number of self-adaptative adjustment finds the multinomial model that minimum fitness bias is corresponding;
S2094, the process solving multinomial model are LRCF algorithm, obtain LRCF model parameter corresponding to all moment and are stored in path data dictionary;
S210, acquisition approach journey time short-time forecasting model and parameter, adopt SRFF algorithm acquisition approach journey time short-time forecasting model, and adopt ARMA algorithm construction short-time forecasting model, and adopt LSM solving model parameter; Including:
S2101, chronologically rhythmic for institute journey time series arrangement is become a long sequence;
S2102, suppose that the time interval of this long sequence is impartial, with ARMA algorithm construction oneNAfter item multinomial model carrys out matchingNIndividual journey time, asks its parameter and error of fitting with LSM;
S2103, by adjustNRegulate the size of error of fitting, choose error minimum time corresponding multinomial model;
S2104, the process solving this multinomial model are SRFF algorithm, are stored in path data dictionary by SRFF model parameter;
Fusion forecasting model and parameter when S211, acquisition approach journey time length, fusion forecasting model during employing SFF algorithm acquisition approach journey time length, and by off-line training, adopt GNIM to obtain fusion forecasting model parameter; Including:
S2111, demarcation initial time, obtain the item number of SRFF modelN, construct one 2 ×NSieve-like coefficient matrix, wherein each element is the element sum perseverance of non-negative and every string is 1, and element value is unknown;
S2112, start journey time is predicted from initial time, value sequence at the beginning of when compensating long by LRCF algorithm, obtain futureNIndividual long-term prediction value, obtains future by SRFF algorithmNIndividual short-term prediction value, by both predictive value sequences form one 2 ×NPrediction matrix;
S2112, coefficient matrix is added process with row after prediction matrix dot product, it is thus achieved that one 1 ×NMerge vector, approach corresponding journey time sequence with fusion vector, it is thus achieved that corresponding correlation coefficient equation;
S2114, progressively adjust initial time backward, obtain corresponding correlation coefficient equation by same method, be made up of a correlation coefficient equation group these equations;
S2115, solve equation group with GNIM, it is thus achieved that the element value of sieve-like coefficient matrix, namely SFF model parameter, SFF model parameter is stored in path data dictionary;
Do you S212, judge that all path forms time datas are disposed? being then, order performs step S213, otherwise, returns and performs step S205;
S213, judging whether all paths are disposed, be then, order performs step S214, otherwise, returns and performs step S203;
S214, judging whether all combination of nodes are disposed, be then, order performs step S215, otherwise, returns and performs step S202;
S215, the data storage terminating this path data dictionary and renewal;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
Accompanying drawing 3 is path forms time of the present invention change querying flow schematic diagram, as seen from the figure, the present invention is based on the journey time fusion query method of the big data of traffic, adopt the present invention based on the big data of traffic journey time fusion forecasting method to set query path and go out line number journey time change be predicted and will predict the outcome being supplied to user, including, adopt the data that all online vehicles are uploaded obtain all kinds of forecast models and parameter by calculated off line or training and set up according to all kinds of forecast models and parameter and dynamically update data dictionary; According to user set query path and go out line number, call all kinds of forecast model and the parameter real-time travel time data in conjunction with this path and respective stretch and be predicted the journey time in this path and respective stretch and will predict the outcome being supplied to user; Described data dictionary includes vehicle data dictionary, section data dictionary and path data dictionary; Described online vehicle refers to login network access and automatically uploads the vehicle of location and speed data.
Equally, the present invention is based on the journey time fusion query method of the big data of traffic, described go out line number refer to based on road network tidal current theoretical, set every day from 0:00 time to 24:00 time as a complete cycle cycle, according to working day, Saturday, Sunday, the trip mode of festivals or holidays, travel time data carrying out trip Type division and numbering, namely goes out line number, the travel time data in each cycle is arranged in a journey time sequence.
Equally, the present invention is based on the journey time fusion query method of the big data of traffic, and described all kinds of forecast models and parameter include fusion forecasting model and intersection delay relation analysis model when section or path forms time data periods rules model, section or the statistical rules model of path forms time, section or path forms time long-term prediction model, section or path forms time short-time forecasting model, section or path forms time length; Wherein,
Adopt periodic law series approximation PLSA(PeriodicLawSeriesApproximation, PLSA) algorithm obtains section or path forms time data periods rules model, and adopt method of least square LSM(LeastSquareMethod, LSM) solve approximate model parameter;
Statistical law is adopted to extract SRE(StatisticalRuleExtraction, SRE) algorithm obtains section or path forms time statistical rules model, and adopt Density Estimator KDE(KernelDensityEstimation, KDE) obtain the probability density Changing Pattern of section or path forms time;
Correction prediction LRCF(Long-timeRollingCorrectionForecast is rolled when adopting long, LRCF) algorithm obtains section or path forms time long-term prediction model, and obtain long-term prediction model parameter by calculated off line, by quickly realizing section or path forms time prediction in line computation;
Adopt and roll matching prediction SRFF(Short-timeRollingFittingForecast in short-term, SRFF) algorithm obtains section or path forms time short-time forecasting model, and adopt time series autoregressive moving average ARMA (Auto-RegressiveandMovingAverage, ARMA) algorithm construction short-time forecasting model, and adopt method of least square LSM(LeastSquareMethod, LSM) solving model parameter;
Adopt sieve-like fusion forecasting SFF(SieveFusionForecast, SFF) algorithm obtains fusion forecasting model when section or path forms time length, and pass through off-line training, adopt Gaussian-Newton method GNIM(Guassian-NewtonIterativeMethod, GNIM) obtain fusion forecasting model parameter;
Adopt intersection delay association analysis IDCA(IntersectionDelayCorrelationAnalysis, IDCA) algorithm construction intersection delay relation analysis model, and by method of least square LSM(LeastSquareMethod, LSM) solving model parameter, model parameter is obtained, by the quick compensation in line computation realizing route journey time by off-line training;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
The present invention, based on the journey time fusion query method of the big data of traffic, comprises the following steps:
S301, user's inquiry;
S302, collection history and real time traffic data, the section sampled including road-net node numbering, section numbering, path number, even time interval and path forms time, all kinds of Forecasting of Travel Time data, all kinds of model parameter, date and hour, data classifying and numbering;
S303, inquire about set path journey time change? it is then, continues executing with step S304, otherwise, perform step S314;
S304, querying condition are arranged, user inquires about content according to oneself requirements set, including the period of the duration of short-term prediction and long-term prediction, according to even time interval principle, demarcate corresponding in short-term with the starting point and ending point of long-term prediction step-length, and set according to calendar and go out line number accordingly; Meanwhile, user sets the starting point and ending point of road-net node, by searching route data dictionary, demarcates the path between the two road-net node and path number;
S305, path selection travel time data, choose a untreated path forms time data of class according to the sequencing going out line number;
S306, path selection, choose a untreated path according to the sequencing of path number, by the section numbering that path data dictionary search to this path comprises;
S307, choose section, choose a untreated section according to the sequencing of section numbering;
S308, Link Travel Time fusion forecasting, read corresponding model and parameter according to user's search request from the data dictionary of section, in conjunction with real-time travel time data, as follows by step process:
S3081, by LRCF algorithm obtain journey time long time at the beginning of value sequence and sequence of differences, both additions obtain the long-term prediction sequence of journey time;
S3082, the short-term prediction step-length chosen with user starting point and ending point for benchmark, obtained the short-term prediction sequence of journey time by SRFF algorithm, obtain fusion forecasting sequence in short-term again through SFF algorithm;
S3083, the long-term prediction step-length chosen with user starting point and ending point be as the criterion, expand by long-term prediction sequence pair fusion forecasting sequence in short-term, obtain the final fusion forecasting sequence of Link Travel Time;
Do you S309, judge that all sections are disposed? being then, order performs step S310, otherwise, returns and performs step S307;
S310, path forms Fusion in Time are predicted, set according to user in predicting, as follows by step process:
S3101, obtained the forecasting sequence of crossing total delay by IDCA algorithm, according to temporal order, the forecasting sequence of crossing total delay is added with the final fusion forecasting sequence in comprised section, this path, obtains the compensation forecasting sequence of journey time;
S3102, by LRCF algorithm obtain journey time long time at the beginning of value sequence and sequence of differences, namely both additions obtain the long-term prediction sequence of journey time;
S3103, the short-term prediction step-length chosen with user starting point and ending point for benchmark, obtained the short-term prediction sequence of journey time by SRFF algorithm, obtain fusion forecasting sequence in short-term again through SFF algorithm;
S3104, the long-term prediction step-length chosen with user starting point and ending point be as the criterion, expand by long-term prediction sequence pair fusion forecasting sequence in short-term, obtain the expansion type forecasting sequence of journey time;
S3105, expansion type forecasting sequence and compensation forecasting sequence are added it are averaging, obtain the final fusion forecasting sequence of path forms time;
Do you S311, judge that all paths are disposed? being then, order performs step S312, otherwise, returns and performs step S306;
Do you S312, judge that all trips are disposed? being then, order performs step S313, otherwise, returns and performs step S305;
S313, path forms time prediction restructuring and Dynamic Announce, by final for all of path forms time fusion forecasting sequence, by path, the logical order of classifying and numbering and sequential carry out restructuring arrangement, obtain the final forecasting sequence of every paths, and on road network Dynamic Announce chronologically;
S314, terminate this path query;
Wherein, described long-term prediction is referred to and is obtained section or the long-term prediction value of path forms time by LRCF algorithm, the span of long-term prediction duration is set as 0 minute to 3 months, and concrete long-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes; Described short-term prediction is referred to and is obtained section or the short-term prediction value of path forms time by SRFF algorithm, the span of short-term prediction duration is set as 0 minute to 3 hours, and concrete short-term prediction duration can obtain by linearly converting on the prediction duration basis that user selectes.
Obviously, the present invention is overcome the problems such as the real-time that the traffic status prediction method of prior art road network or bicycle exists is poor, versatility is poor and practicality is not strong based on the journey time fusion forecasting of the big data of traffic and the Advantageous Effects of querying method, survey precision is greatly improved, the speed of on-line prediction is greatly improved, it is ensured that real-time in engineer applied, practicality and versatility simultaneously.