Multi-source data fusion-based traffic distribution model parameter rapid checking method
Technical Field
The invention belongs to the field of rail transit, and particularly relates to a rapid checking method for traffic distribution model parameters based on multi-source data fusion.
Background
The passenger flow prediction is a key link in the planning of the rail transit network, directly influences the necessity and the layout form of the planning of the rail transit network, and the traffic distribution model directly influences the accuracy of a passenger flow prediction result. As a classical traffic distribution model, the gravity model has been widely applied to various traffic modes such as high-speed railways, inter-city railways, urban rail transit, roads and the like. Considering that 3 parameters in the gravity model are influenced by multi-dimensional factors such as urban heterogeneity, functional area distribution, job and live balance and the like, model parameter calibration must be carried out on a specific research object.
The current standard checking process adopted by the gravity model is basically as follows: a travel distance Distribution (OTLD) is generated using the fundamental year P-A matrix and the impedance matrix. The datcA is used as cA reference standard, and TLD generated by cA P-A matrix and an impedance matrix calculated by cA model is similar to OTLD as much as possible by adjusting model parameters.
If the od data of the local traffic area is not matched with the survey data, the od data is adjusted by using the K-factor coefficient. The K-factor is cA set of parameters that describe the region-to-region relationship and is calculated by taking the ratio between the observed value and the estimated value generated from the friction factor or resistance parameter, which results in cA calibrated gravity model that accurately replicates the P- cA matrix of the base year. This may improve the model accuracy to some extent, but the K-factor dereferencing method is questionable.
Such prior art has the following problems:
(1) at present, the sampling rate of resident trip survey is different according to different urban scales, and generally, the minimum sampling rate of cities with more than 100 ten thousand of population is not less than 1%, the minimum sampling rate of cities with 50-100 ten thousand of population is not less than 2%, the minimum sampling rate of cities with 20-50 ten thousand of population is not less than 3%, and the minimum sampling rate of cities with less than 20 ten thousand of population is not less than 5%. From the survey data of more than 20 cities completed in our hospital, such a low sampling rate results in a lot of cell data missing, and the sample expansion amplifies the error, but we still use the data for model prediction.
(2) In the actual comparison of TLD and OTLD data, the distance segment value is larger, the data fitting degree is generally higher, but once the distance is shorter and the value is smaller, the data fitting error is increased, which is mainly caused by the data loss of a part of traffic cells.
(3) At present, big data is well developed, especially mobile phone signaling data, od in the traffic region level can be obtained through mobile phone data in many cities, and the sampling rate can reach 60% or higher generally, but the accuracy of the traffic cell data is not enough due to the density of base stations and the ping-pong effect of mobile phone switching, and how to apply the data to a model is researched.
(4) The existing gravity model parameter calibration method is based on data survey, and does not integrate big data such as mobile phones, bus card swiping and the like generated in the transportation trip China of residents; in addition, the method is mostly limited to the theoretical level, and the calibration speed of the model parameters is less considered.
Disclosure of Invention
Aiming at least one of the defects or the improvement requirements in the prior art, the invention provides a traffic distribution model parameter rapid checking method based on multi-source data fusion, which realizes the combined utilization of multi-source data and survey data; the heuristic algorithm is applied, so that automatic parameter iteration and calibration can be realized; the model precision is greatly improved, the parameter value is more practical, and the simulation can be better realized.
In order to achieve the above object, according to an aspect of the present invention, there is provided a method for fast checking parameters of a traffic distribution model based on multi-source data fusion, wherein the checking and calibration of the parameters are performed in two stages, and the first stage comprises the steps of:
s1, cleaning data, and denoising;
screening out data comprising repeated data, error data and incomplete data, and removing the data to obtain relatively accurate data;
s2, circularly expanding samples and checking for multiple times;
according to administrative divisions, performing data sample expansion according to the mobile phone occupancy of each mobile phone operator in each administrative district, and checking the data after sample expansion with the population total number, age structure and gender ratio of the administrative district; if the error proportion of all indexes is larger than a preset value, circularly expanding samples according to the occupancy, population total number, age structure and sex proportion again; stopping iteration when the error proportion of all indexes is lower than a preset value;
s3, collecting counting data;
respectively collecting and counting the mobile phone data, the motor vehicle trip GPS data and the bus card swiping data after sample expansion according to the level of the middle zone in traffic, transversely comparing the collected data with other people's mouth general survey data and economic general survey data, and searching whether data with errors exceeding a preset value exist or not; if the error exceeds the preset value, returning to the step S1;
s4, model operation;
combining a shortest path algorithm to obtain a time matrix1 between traffic cells in the current year; the method comprises the steps of obtaining a new round of traffic cell simulation data matrix4 by using a traffic cell od data matrix2 obtained by status investigation and a traffic middle area od data matrix3 obtained by previous sample expansion as initial model inputs and using a three-dimensional balance model in a gravity model;
s5, calibrating parameters;
calculating the occurrence quantity, namely P quantity, and the attraction quantity, namely A quantity, of each traffic cell by using a traffic generation model as model input, introducing matrix1 in S4 as impedance, and inputting parameters a, b and c by using a gravity model to obtain a new round of traffic distribution matrix 5;
s6, regulating the parameters again;
multiplying matrix1 obtained in S4 by matrix4 and matrix5 obtained in S5 respectively to obtain OTLD and TLD respectively, comparing the OTLD and the TLD by taking the distance as an abscissa and the turnover amount as an ordinate, and if the OTLD and the TLD exceed a preset range, adjusting parameters by using a genetic algorithm, and repeating S4 and S5 for multiple times until the parameters return to the preset range, so that the latest parameters a1, b1 and c1 are obtained;
the second stage comprises the following steps:
s7, distinguishing secondary traffic;
collecting traffic cells by means of city administrative region division to obtain a secondary traffic middle area;
s8, calculating the comprehensive utility value of the traffic cell;
and (3) statistically analyzing the population and post data of each traffic cell, and calculating the comprehensive utility value according to the following formula:
Pitotal=Pi×α+Gi×β (1)
wherein, PitotalFor the traffic cell comprehensive utility value, PiFor the traffic sector population, Giα and β are weight coefficients for the posts of the traffic cell;
s9, calculating a smooth matrix;
summing the comprehensive utility values of all the traffic cells in each secondary traffic middle area, and smoothing the traffic volume according to the proportion of the comprehensive utility value of each traffic cell to the total utility value of the secondary traffic middle area to obtain a smoothed traffic cell traffic volume matrix 6;
s10, model operation;
calculating the occurrence quantity, namely P quantity, and the attraction quantity, namely A quantity, of each traffic cell by using a traffic generation model as model input, introducing matrix1 in S4 as impedance, and inputting parameters a2, b2 and c2 by using a gravity model to obtain a new round of traffic distribution matrix 7;
s11, regulating the parameters again;
multiplying matrix1 obtained in S4 by matrix6 obtained in S9 and matrix7 obtained in S10 respectively to obtain OTLD1 and TLD1, comparing OTLD1 with TLD1 by taking the distance as an abscissa and the turnover amount as an ordinate, and if the distance exceeds a preset range, adjusting parameters by using a genetic algorithm, and repeating S9 and S10 for multiple times until the parameters return to the preset range, so that the latest parameters a3, b3 and c3 are obtained;
after the first stage and the second stage are respectively completed, the method further comprises the following steps:
s12, final parameter determination;
and respectively and correspondingly averaging the model parameters a1, b1 and c1 obtained in the first stage and the model parameters a3, b3 and c3 obtained in the second stage to obtain model parameters a4, b4 and c 4.
Preferably, in step S1, the screening and elimination processing method includes any one or any combination of consistency check, evaluation, whole case elimination and variable elimination.
Preferably, in step S7, the number of secondary middle of traffic zones is greater than the number of middle of traffic zones, and each secondary middle of traffic zone has a set of calculated volumes.
Preferably, in step S8, α takes 0.4 and β takes 0.6.
The above-described preferred features may be combined with each other as long as they do not conflict with each other;
generally, compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects:
1. the method for rapidly checking the traffic distribution model parameters based on multi-source data fusion respectively adopts a bottom-up parameter checking method and a top-down parameter checking method, so that parameter calibration is also carried out in two stages, parameters obtained in the first stage are a1, b1 and c1, parameters obtained in the second stage are a3, b3 and c3, and finally the obtained model parameters adopt the average numbers a4, b4 and c4 of the two methods.
2. In the model parameter verification from bottom to top, after data cleaning, noise reduction and noise elimination, the data of different mobile phone operators are subjected to sample expansion to obtain the travel data of each traffic cell, the travel data, the motor vehicle GPS data and the bus card swiping data form the travel data in the middle traffic area after being collected, and accordingly the gravity model parameters can be calculated. In the top-down model parameter verification, traffic middle areas are divided according to urban administrative areas, and the traffic volume of each traffic cell is calculated according to the comprehensive utility of each traffic cell in the traffic middle areas, so that the gravity model parameters are calculated.
3. Multi-source data such as mobile phone operators, motor vehicle GPS, public transport GPS and the like are introduced, so that the data sample size is increased, and the workload of passenger flow investigation is reduced; changing the forward survey data into backward record data increases the authenticity and reliability of the data.
4. Comprehensive utilization of multi-source data and investigation data is achieved, and the problems of bias distribution and volatility in model prediction are solved.
5. Two model parameter calibration methods of 'top-down' and 'bottom-up' are fused, so that the defects of 'counting' and 'disaggregation' caused by unidirectional parameter calibration are overcome, and the model prediction precision is improved.
6. A heuristic solving algorithm (genetic algorithm) is adopted for 3 parameters in the gravity model, so that the convergence speed of the parameters is accelerated, and the rapid calibration of the parameters is realized.
Drawings
FIG. 1 is a flow diagram of a traffic distribution model parameter rapid checking method based on multi-source data fusion.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments; it should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention; furthermore, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other; the present invention will be described in further detail with reference to specific embodiments;
as a preferred embodiment of the present invention, as shown in fig. 1, the present invention provides a method for rapidly checking parameters of a traffic distribution model based on multi-source data fusion, and two methods are mainly used in the process of checking and calibrating parameters, so that the checking and calibrating of parameters are performed in two stages. The parameters obtained in the first stage are a1, b1 and c1, the parameters obtained in the second stage are a3, b3 and c3, and the average of the finally obtained model parameters is a4, b4 and c4 by adopting two methods.
The first stage comprises the following steps:
s1, cleaning data, and denoising;
and screening out the data comprising the repeated data, the error data and the incomplete data, and rejecting the data to obtain relatively accurate data. The common treatment methods are: consistency check, estimation, whole case deletion and variable deletion.
(1) Consistency check (consistency check) is to check whether data meets requirements according to the reasonable value range and the mutual relation of each variable, and find out data which exceeds a normal range, is logically unreasonable or is mutually contradictory;
(2) the estimation is to replace an invalid value and a missing value by a sample mean value, a median or a mode of a certain variable;
(3) the whole deletion (casewise deletion) is to eliminate the samples containing missing values.
(4) Variable deletion (variable deletion). If there are many invalid and missing values for a variable and the variable is not particularly important to the problem under study, then the variable may be considered deleted.
S2, circularly expanding samples and checking for multiple times;
according to administrative divisions, performing data sample expansion according to the mobile phone occupancy of each mobile phone operator (such as mobile, Unicom, telecom and the like) of each administrative area, and checking the data after sample expansion with the population total number, age structure and gender ratio of the administrative area; if the error proportion of all indexes is more than a preset value of 5%, circularly expanding samples according to the occupancy, the population number, the age structure and the sex proportion again; when the error proportions of all the indexes are lower than a predetermined value of 5%, the iteration is stopped.
S3, collecting counting data;
respectively carrying out centralized counting on the mobile phone data, the motor vehicle trip GPS data, the bus card swiping data and the like after sample expansion according to the level of the middle zone of traffic, transversely comparing the mobile phone data, the motor vehicle trip GPS data, the bus card swiping data and the like with other people's mouth general survey data and economic general survey data, and searching whether data with errors exceeding a preset value exist or not; if the error exceeds the predetermined value, the process returns to S1.
S4, model operation;
combining a shortest path algorithm to obtain a time matrix1 between traffic cells in the current year; and (3) obtaining a new round of traffic cell simulation data matrix4 by using the traffic cell od data matrix2 obtained by the current situation investigation and the traffic middle area od data matrix3 obtained by the previous sample expansion as initial model inputs and using a three-dimensional balance model in the gravity model.
S5, calibrating parameters;
and (3) calculating the occurrence quantity, namely P quantity, and the attraction quantity, namely A quantity of each traffic cell by using the traffic generation model as model inputs, citing matrix1 in S4 as impedance, and inputting parameters a, b and c by using a gravity model to obtain a new round of traffic distribution matrix 5.
S6, regulating the parameters again;
multiplying matrix1 obtained in S4 by matrix4 and matrix5 obtained in S5 respectively to obtain OTLD (assisted Trip Length distribution) and TLD (Trip Length distribution), comparing the OTLD and the TLD by taking the distance as an abscissa and the turnover amount as an ordinate, and if the distance exceeds a preset range, performing parameter adjustment by using a genetic algorithm, and repeating S4 and S5 for multiple times until the parameters return to the preset range, so that the latest parameters a1, b1 and c1 are obtained.
In the first stage of the parameter calibration process, during model parameter verification from bottom to top, after data cleaning, noise reduction and noise elimination, the data of different mobile phone operators are subjected to sample expansion to obtain the travel data of each traffic cell, and the travel data, the motor vehicle GPS data and the bus card swiping data are integrated to form the travel data of the central traffic area, so that the gravity model parameters can be calculated.
The second stage parameter adjustment is performed by a second method, and the second stage comprises the following steps:
s7, distinguishing secondary traffic;
and (4) collecting the traffic cells by means of city administrative region division to obtain a secondary traffic middle area. The stage should ensure that the number of the secondary traffic middle areas is larger than that of the traffic middle areas as much as possible, and each secondary traffic middle area has a set traffic volume.
S8, calculating the comprehensive utility value of the traffic cell;
and (3) statistically analyzing the population and post data of each traffic cell, and calculating the comprehensive utility value according to the following formula:
Pitotal=Pi×α+Gi×β (1)
wherein, PitotalFor the traffic cell comprehensive utility value, PiFor the traffic sector population, Giα and β are weight coefficients for the traffic cell posts, preferably, α takes 0.4 and β takes 0.6 in step S8.
S9, calculating a smooth matrix;
and summing the comprehensive utility values of all the traffic cells in each secondary traffic middle area, and smoothing the traffic volume according to the proportion of the comprehensive utility value of each traffic cell to the total utility value in the secondary traffic middle area to obtain a smoothed traffic cell traffic volume matrix 6.
S10, model operation;
and (3) calculating the occurrence quantity, namely P quantity, and the attraction quantity, namely A quantity of each traffic cell by using the traffic generation model as model inputs, citing matrix1 in S4 as impedance, and inputting parameters a2, b2 and c2 by using a gravity model to obtain a new round of traffic distribution matrix 7.
S11, regulating the parameters again;
multiplying matrix1 obtained in S4 by matrix6 obtained in S9 and matrix7 obtained in S10 respectively to obtain OTLD1 and TLD1, comparing OTLD1 with TLD1 by taking the distance as an abscissa and the turnover number as an ordinate, adjusting parameters by using a genetic algorithm if the distance exceeds a preset range, and repeating S9 and S10 for multiple times until the parameters return to the preset range, so that the latest parameters a3, b3 and c3 are obtained.
And the second stage of the parameter calibration process is up to this point, in the model parameter verification from top to bottom, the traffic middle area is divided according to the urban administrative area, the output of each traffic cell is calculated according to the comprehensive utility of each traffic cell in the traffic middle area, and therefore the gravity model parameter is calculated.
After the first and second stages are completed, respectively, the final weighted summation is performed:
s12, final parameter determination;
and respectively and correspondingly averaging the model parameters a1, b1 and c1 obtained in the first stage and the model parameters a3, b3 and c3 obtained in the second stage to obtain model parameters a4, b4 and c 4.
According to the rapid checking method for the traffic distribution model parameters based on the multi-source data fusion, multi-source data such as mobile phone operators, motor vehicle GPS (global positioning system), public transport GPS (global positioning system) and the like are introduced, so that the data sample size is increased, and the workload of passenger flow investigation is reduced; the forward investigation data is changed into backward recording data, so that the authenticity and reliability of the data are improved; comprehensive utilization of multi-source data and survey data is realized, and the problems of bias distribution and volatility in model prediction are solved; two model parameter calibration methods of 'top-down' and 'bottom-up' are fused, so that the defects of 'counting' and 'disaggregation' caused by unidirectional parameter calibration are overcome, and the model prediction precision is improved; a heuristic solving algorithm (genetic algorithm) is adopted for 3 parameters in the gravity model, so that the convergence speed of the parameters is accelerated, and the rapid calibration of the parameters is realized.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.