CN107610464B - A kind of trajectory predictions method based on Gaussian Mixture time series models - Google Patents
A kind of trajectory predictions method based on Gaussian Mixture time series models Download PDFInfo
- Publication number
- CN107610464B CN107610464B CN201710683275.3A CN201710683275A CN107610464B CN 107610464 B CN107610464 B CN 107610464B CN 201710683275 A CN201710683275 A CN 201710683275A CN 107610464 B CN107610464 B CN 107610464B
- Authority
- CN
- China
- Prior art keywords
- model
- data
- prediction
- track
- models
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 239000000203 mixture Substances 0.000 title claims abstract description 46
- 238000004458 analytical method Methods 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000000926 separation method Methods 0.000 claims description 4
- 230000006641 stabilisation Effects 0.000 claims description 4
- 238000011105 stabilization Methods 0.000 claims description 4
- 238000012417 linear regression Methods 0.000 claims description 3
- 238000007476 Maximum Likelihood Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 claims description 2
- 238000003064 k means clustering Methods 0.000 claims description 2
- 238000013178 mathematical model Methods 0.000 claims description 2
- 238000001583 randomness test Methods 0.000 claims description 2
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 claims 2
- 238000005259 measurement Methods 0.000 claims 1
- 238000005311 autocorrelation function Methods 0.000 description 13
- 230000008859 change Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 230000007774 longterm Effects 0.000 description 5
- 238000005314 correlation function Methods 0.000 description 4
- 238000009499 grossing Methods 0.000 description 4
- 238000005315 distribution function Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 241001123248 Arma Species 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000012958 reprocessing Methods 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 230000000087 stabilizing effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Landscapes
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses one kind to be based on Gaussian Mixture time series models (Gauss Mixture Time Series Model, abbreviation GMTSM) trajectory predictions method, the analysis of model recurrence and section vehicle flowrate is carried out to magnanimity vehicle historical track, realizes track of vehicle prediction.Key step includes: that (1) carries out unsupervised learning cluster using k-means algorithm to vehicle historical track.(2) historical track probability Distribution Model is constructed using Gaussian Mixture time series models.(3) mobile object track is predicted by mixed model regression process.The experimental results showed that Gaussian Mixture time series models under the vehicle flowrate catastrophe of section, by the weight of model adjust automatically sub-model, select the prediction locus of maximum probability.
Description
Technical Field
The invention relates to a vehicle track prediction method based on a Gaussian mixture time series model, which belongs to the technical application field of large data value mining, and mainly applies an intelligent traffic network system.
Background
The intelligent traffic system is an intelligent system formed in the continuous advancing process of urbanization, relieves the urban traffic jam condition, but with the further expansion of urbanization, the traffic jam is still an urgent problem to be solved, the traffic condition is monitored and predicted in advance, a reasonable route is recommended, the intelligent traffic system is a reasonable scheme for relieving the traffic jam, is also a hotspot of mobile object database research in recent years, and has extremely high application value for real-time, accurate and reliable prediction of the uncertain track of the mobile object. Through research on the moving track of the vehicle, a track model is obtained, and the model can provide traffic conditions for a driver in a short time in the future and provide reasonable route instructions according to the current route.
The existing vehicle track prediction methods can be roughly divided into two types, the first method is to perform supervised learning by taking a designated vehicle as a target and predict the track through the change of vehicle curvature, speed and maneuvering state, and the method is only suitable for predicting the track of the vehicle in a short time in the future and is not suitable for predicting the traffic jam condition in a long time period in the future. The other method is to establish an unsupervised learning model according to a large amount of historical track data to realize track prediction, the method can realize prediction of future traffic conditions, but only predicts according to probability distribution of vehicle historical tracks, and does not consider the situation that due to different time period selections, such as special time periods including rush hour periods on work and off work, holiday time periods and the like, the latest traffic flow of road sections is suddenly increased or decreased, so that the referential performance of track data is reduced, and the track prediction is inaccurate and the real-time performance is not strong.
In the trajectory prediction model, modeling is required to be carried out on a large number of vehicle historical trajectories to realize model regression type trajectory prediction, and an unsupervised learning model trajectory prediction algorithm is required to ensure extraction of reliable data and improve estimation and prediction accuracy under the condition that the traffic flow of a road is large in short-time variation range, so that the method is an important problem in providing an optimal route scheme for prediction of a driver aiming at future traffic conditions.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the defects of the traditional Gaussian mixture model and other traditional models in the track prediction under the condition of road traffic flow mutation during non-working dates, the invention optimizes the traditional Gaussian mixture model from two aspects of a dynamic historical track model and a traffic flow prediction model of a sample sequence, provides a Gaussian mixture time sequence model, and judges whether the historical track is in a traffic flow sensitive period or not by predicting the traffic flow of a road, thereby more accurately predicting the track of a moving object.
Define 1 track chain { t (t) | t ═ 1,2,3, · n } time, { (x)1,y1),(x2,y2),...,(xn,yn) The method comprises the steps of obtaining two-dimensional coordinate data of a moving object, forming discrete track points of the moving object by the discrete points, and linearizing the discrete points through a certain algorithm to obtain the linearity of the moving objectChange the track to be denoted as Si＝{xi,yi,ti},1≤i≤n。
Defining 2 a historical track vector set prediction model to cluster massive historical tracks, modeling a track chain set formed by each cluster in X-axis and Y-axis directions, and representing data of the tracks by using track vectors in two directions as follows: representing a set of projection vectors of the ith track chain in the x-direction,representing a set of projection vectors of the ith trajectory in the y-direction,called track chain SiIs called the set of trajectory vectors on the trajectory data set D.
And 3, defining a time period of the road traffic flow change of sudden increase or decrease in the traffic flow sensitive period, such as a peak time of commuting and a holiday. Fig. 2 (see fig. 2) shows that the direct traffic flow of the Kyushu intersection at the Zhuhai welcome major road is manually investigated in 21 months 3 and 2003, the traffic flow is read every hour within one day, the phenomenon of relatively large increase of the traffic flow occurs in 8:00-9:00, 12:00-14:00 and 18:00-21:00, and the time periods are traffic flow sensitive periods.
Definition 4 Long-term prediction the current time is denoted as tnPredicted time thThen the predicted time scale is Δ t ═ tn-thWhen the value of delta t is smaller, the method is small-scale prediction, also called short-time prediction; when the value of delta t is larger, large-scale prediction is carried out; when the value of delta t is medium, the prediction is in a mesoscale mode. In the GMTSM model prediction, the prediction in the future moment is regarded as long-term prediction, namely delta t is more than or equal to 180 and less than or equal to 600 (seconds). When delta t is less than or equal to 30 seconds, the ruler is a small rulerDegree prediction, also known as short-term prediction; when 30 (seconds)<Δt<If the value of 180 (seconds) is medium, the prediction is the mesoscale prediction; when the delta t is more than or equal to 180 seconds, the long-term prediction is carried out.
Definition 5 (trajectory prediction error density): the trajectory prediction model takes the vehicle historical trajectory data set and the traffic flow historical data set as input and takes the vehicle future trajectory as output, see fig. 3, wherein the deviation degree existing between the predicted trajectory chain (dotted line) and the actual trajectory chain (solid line) is a prediction error. The magnitude function of the root mean square error of the deviation degree in unit time is recorded as RMSER (the unit is M/s, and the deviation distance of the track in each second is represented), as shown in formula (1), the root mean square error is the geometric space error between the calculated predicted track and the actual track, M is the clustering number of track points, T is the time consumed by the model to start regression learning to finish prediction at the next moment, T cannot be infinitely large or infinitely small within the acceptable range of long-term prediction, and X and Y are respectively coordinate information obtained by modeling in the directions of an X axis and a Y axis according to the longitude and latitude.
And 6, defining the capability of the model reliability track prediction model to complete accurate prediction under specified conditions and within specified time, wherein the capability is in inverse relation with prediction errors. Expressed by R, the formula is shown as (2):
wherein, F (t)future) Is the error of the future prediction, F (t)true) Is the error of the actual prediction.
The technical scheme is as follows: a trajectory prediction method based on a Gaussian mixture time series model comprises the following three aspects:
(1) preprocessing of trajectory data and traffic flow data
(2) Reliability analysis of partial models
(3) Weight training of partial models
(1) Preprocessing track data and traffic flow data:
the data preprocessing comprises three parts, namely ETL separation processing of the data, clustering of historical track data and smoothing processing of historical traffic flow data.
① separating ETL data, marking that the empty data and the missing data can not be processed, replacing the invalid data, normalizing the format, converting the data format extracted from the source data into a target data format convenient for warehouse processing, and replacing the illegal data or exporting the illegal data to an error file for reprocessing by establishing the main foreign key constraint.
② trajectory data clustering, wherein the proposed k-means algorithm is based on the idea of determining k value through unsupervised learning, randomly giving a k for clustering, carrying out k +1 (or k-1) on the k after obtaining a result, then respectively calculating the distance between the edge point of the trajectory and the central point in the attributed cluster under k and k +1 (or k-1) clustering, selecting the final k value with the minimum distance, then carrying out k +1 (or k-1) on the new k value, and finally obtaining a proper k value.
③ smoothing traffic data, selecting Auto-Regressive Moving Average model (ARIMA), wherein the default parameters are ARIMA (p, d, q), p is Auto-Regressive item, q is Moving Average item number, and d is difference times when the time sequence becomes stable, eliminating the complex process of uncertainty identification in the process of determining parameters by the time sequence model, reducing the time overhead of the model, only needing to solve the model order and the model parameters, reducing the realization consumption of the algorithm, ensuring the dynamic property and real-time property of the algorithm prediction, and when the first difference is not enough to make the non-stable time sequence data (t), the idea is generalized.
(2) Reliability analysis of partial models
The Gaussian mixture time series model provided by the invention comprises two subsections: the method comprises the following steps of analyzing historical track space data of the vehicle, and analyzing traffic time data of the vehicle. And comparing the reliability of the models through the prediction error rates of the sub-models, and performing model weight training by using the model reliability to realize the establishment of the Gaussian mixture time sequence prediction model.
The Gaussian mixture model and the time sequence model are obtained by training according to historical data, along with the change of time, the longer the time of the historical data selected by the model is, the larger the model error of the model is due to the assumption of parameters, and in order to reduce the error caused by the time of the historical data selection, the historical data of the two models needs to be dynamically updated; meanwhile, in order to reduce the influence of the sub-model on the final mixed model due to the fact that the prediction error of the sub-model is increased, the weights of the two sub-models (the Gaussian mixed model and the time series model) of the mixed model need to be dynamically updated, the support degree between the two sub-models can be improved through dynamic weight weighting, the data of the two sub-models have the trend characteristic, and the mixed model can obtain a more accurate prediction effect finally. The RMSER defined by the formula (1) represents the track prediction error density, the deviation distance between the predicted track and the real track of the model in unit time is obtained, the larger the value of the RMSER is, the larger the model prediction error is, the smaller the value of the RMSER is, the smaller the model prediction error is, therefore, the RMSER reflects the accuracy of the model, and the weight of the model is calculated through the RMSER.
The model reliability measurement method is a basis for determining the reliability of the model, the requirement for determining the evaluation index is a main step of the reliability analysis of the model, and if the system is not evaluated by the reliability index or is not verified by the index evaluation, the reliability of the model cannot be credible. The main purpose of the prediction model is to accurately predict the position of the moving object at the next time, the prediction error is a measure of the prediction accuracy, the prediction error rate reflects the change of the prediction error in unit time, and the stability of the model can be considered, so the reliability of the model is obtained by the prediction error rate.
The reliability analysis of the partial models is as follows:
① regression learning of Gaussian mixture model
Clustering the historical track data to form a historical track vector set DMObtaining the probability ratio of the clusters by a Gaussian mixture recursive model to form a probability matrix PMWhen the track prediction is carried out, three-dimensional space-time data of the vehicle at the occurring moment are required to be input, the maximum value is taken out as the position coordinate of the next moment by comparing the output values of the future coordinates in the clustering probability matrix, and the track prediction on the historical track is realized by analogy in sequence.
The space-time data of the track is three-dimensional data comprising longitude, latitude and time stamp, the model predicts future coordinate points of the vehicle by a method for realizing linear regression, and track points formed by the three-dimensional coordinate data are marked as Si＝(xi,yi,ti) The historical track chain is represented as: sh＝{(x1,y1,t1),(x2,y2,t2),...,(xn,yn,tn)}＝{S1,S2,...,Sn}. Obtaining a Gaussian mixture prediction model through learning of the historical track chain, and obtaining position information (x 'in the directions of x and y at the time n +1 by predicting the next time through the Gaussian mixture model'n+1,y'n+1) Then, the predicted trajectory chain obtained by the gaussian mixture model is represented as: sn+1＝(x'n+1,y'n+1,tn+1)＝(xn+Δxn+1,yn+Δyn+1,tn+1) Whereinis an increment in x and y directions, phii(x) And phii(y) weights in the x and y directions, respectively, μyiAnd muyiThe distance mean values in the x and y directions are respectively, and the actual position information at the time n +1 is (x)n+1,yn+1) Then, the prediction error of the gaussian mixture model can be expressed as:the simultaneous prediction error density (RMSER) can be expressed asWherein M is the number of trace point clusters and T is the total prediction time.
② regression learning of time series models
Aperiodic congestion is not considered in this study because generally this type of congestion is less likely to occur, and for short-term traffic flow prediction studies, it is easy to eliminate these sample points containing aperiodic congestion from the samples. Therefore, road traffic conditions are divided into two states: the former corresponds to a free-flow condition defined by VanArem, and the latter corresponds to a periodic congestion condition. And (4) checking the variance, the trend and the seasonal change rule of the dispersion, the autocorrelation function and the partial autocorrelation function graph of the time sequence by using an ADF unit root, and identifying the stationarity of the sequence. Generally speaking, neither time series of road traffic is a smooth sequence. And carrying out smoothing treatment on the non-stationary sequence. If the data sequence is non-stationary and has a certain increasing or decreasing trend, the data needs to be processed differentially, and if the data has an variance, the data needs to be processed technically until the autocorrelation function value and the partial correlation function value of the processed data are not significantly different from zero.
And establishing a corresponding model according to the identification rule of the time series model. If the partial correlation function of the stationary sequence is truncated and the autocorrelation function is trailing, it can be concluded that the sequence fits the AR model; if the partial correlation function of the stationary sequence is tail-biting and the autocorrelation function is tail-biting, it can be concluded that the sequence fits the MA model; if both the partial correlation function and the autocorrelation function of the stationary sequence are tail-shifted, the sequence fits the ARMA model. (truncation refers to the property that the autocorrelation function (ACF) or partial autocorrelation function (PACF) of the time sequence is 0 after a certain order (such as the PACF of AR), and the property that the ACF or PACF is not 0 after a certain order (such as the ACF of AR) is followed by parameter estimation, whether the parameter estimation has statistical significance is checked, hypothesis test is carried out, whether the residual sequence is white noise is diagnosed, and prediction analysis is carried out by using a model which passes the test.
Short-term traffic flow is more affected by various random factors than long-term traffic flow, and thus shows strong randomness characteristics. The traffic flow at a certain moment is related to the traffic flow at the previous moments because in many large and medium cities drivers are usually able to reasonably select and adjust driving routes in advance by obtaining traffic information through some kind of channel (traffic broadcasting station, etc.). In addition, the traffic conditions of urban roads can be divided into two states — smooth traffic and traffic congestion. In different states, the motion characteristics of traffic flow data are likely to be different. And the transition between these two traffic states is accomplished by a random process.
The root mean square error of the traffic flow data can reflect the error between the actual traffic flow and the predicted traffic flow and describe the discrete condition of the prediction error, and the larger the root mean square error value is, the larger the RMSER value is, and the more discrete the sequence of the prediction error is, the worse the prediction effect is. Predicting the traffic flow of N roads by the ARIMA model, wherein the predicted traffic flow is NfActual vehicle flow is NhThen the prediction error density of the model is:m is the number of clusters obtained by the Gaussian mixture time sequence model, and T is the predicted total time.
(3) Weight training of partial models
The Gaussian mixture model predicts the track by obtaining probability functions of different clusters through spatial data learning of historical tracks, the probability of track chains in different clusters is compared in size to determine the future track, the time sequence model predicts the traffic flow by obtaining the value of the traffic flow of the future road section, and the value ratio of the traffic flow of different road sections is the probability ratio of traffic flow conditions of different road sections.
The hybrid model provided by the invention is used for improving the accuracy of track prediction under the road emergency. In an emergency, the reliability of the historical track data of the Gaussian mixture model is reduced, so that the reliability of probability ratios of different future track chains is reduced, the prediction error is increased, the mixture model realizes the prediction of traffic flow through the time sequence model, the probability ratios of the different future track chains obtained by the Gaussian mixture model depending on the historical data are adjusted, and the accuracy of the track prediction is improved; under normal conditions, the time series model has small variation range of the predicted value of the traffic flow and the value of the historical traffic flow, but can only predict the trend of the traffic flow and can not realize the prediction of the vehicle track, and the Gaussian mixture model can realize the prediction of the future track of the vehicle according to the training of the historical track. Therefore, in order to realize the supporting degree between the two sub-models under different conditions and ensure the accuracy of the mixed model trajectory prediction, different weights are set for the Gaussian mixed model and the time series model.
The premise of the effectiveness of the trajectory prediction model is the real-time performance and the accuracy of the model, so the formula (1) of the invention provides a concept of the reliability of the model. The model reliability specifies the probability of the prediction model achieving accurate prediction within a certain time, and is in a negative correlation with the prediction error. The prediction model has smaller prediction error and larger reliability within a specified time, the model has larger prediction error and smaller reliability, and the prediction error density (RMSER) can reflect the change rate of the prediction error, so the prediction model calculates the reliability of the model through the RMSER. The prediction error of the trajectory prediction model in the prediction time is abbreviated as a trajectory prediction error distribution function and is represented by F (t):
where t is the total time predicted by the model. The error distribution function F (t) of the model, F (t), can present the prediction error distribution of the model, realize the calculation of the prediction error of the future model, and obtain the probability ratio of the prediction error of the model by calculating the future prediction error and the real prediction errorThrough the analysis of the prediction errors of the two sub-models, the reliability of the sub-models is calculated, and the similarity of the model is defined by S(R1,R2)Is represented by R1、R2Reliability of the two models, d (R), respectively1,R2)＝|R1-R2I represents the degree of difference in reliability between different models, and S (R)1,R2) And d (R)1,R2) The greater the distance between the reliability analysis results, the smaller the similarity of the two models, and the model similarity is:
according to the invention, the weight of the component model is calculated through the similarity among a Gaussian mixture time series model, a Gaussian mixture model and a time series model, and the mutual support degree of different models can be expressed as follows:wherein N ismodelIs the number of models, A (M)i) Representing the sum of the similarity of one model to the other. Finally, the mutual support degree between the two model results is respectively used as the dynamic weight of each modelHeavy, expressed as:
the Gaussian mixture model and the time sequence model are obtained by training according to historical data, along with the change of time, the longer the time of the historical data selected by the model is, the larger the model error is caused by the assumed parameters of the model, and in order to reduce the error caused by the time of the historical data selection, the historical data of the two models needs to be dynamically updated; meanwhile, in order to reduce the influence of the sub-models on the final mixed model due to the fact that the prediction errors of the sub-models are increased, the weights of the two sub-models of the mixed model need to be dynamically updated, the dynamic weight weighting can improve the supporting degree between the two sub-models, the data of the two sub-models have the trend characteristic, and the mixed model is enabled to obtain a more accurate prediction effect finally.
Drawings
FIG. 1 is a diagram of a comparative example of trajectory prediction according to an embodiment of the present invention;
FIG. 2 is a traffic flow chart defining a traffic flow sensitive area according to the present invention;
FIG. 3 is a diagram illustrating an example of defining a trajectory prediction error according to the present invention;
FIG. 4 is a frame diagram of trajectory prediction of moving objects based on Gaussian mixture time series model according to the present invention;
FIG. 5 is a diagram of a conventional k-means cluster according to the present invention;
FIG. 6 is a graph of a non-conventional k-means cluster according to the present invention;
FIG. 7 is a flow chart of the model reliability analysis of the present invention;
FIG. 8 is a flowchart of the present invention for calculating the weights of the partial models.
Detailed Description
The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, as various equivalent modifications of the invention will occur to those skilled in the art upon reading the present disclosure and fall within the scope of the appended claims.
As shown in fig. 4, the principle of the method for predicting the trajectory by the gaussian mixture time series model is divided into four steps:
(1) and preprocessing vehicle related data acquired by a GPS through an ETL technology to realize the separation of traffic flow data and vehicle historical track data. The traffic data is two-dimensional data comprising traffic and a timestamp; the historical track data is three-dimensional data including longitude, latitude, and time stamp. And converting the two types of separated data into vector data and storing the vector data in a specified database.
(2) Performing stabilization analysis processing on the two-dimensional traffic flow data to obtain stabilization data; and performing clustering analysis processing on the three-dimensional historical track data by using a k-means algorithm, and obtaining clustering model parameters by using a maximum likelihood Estimation (EM) algorithm so as to maximize probability distribution based on the historical track data and finally obtaining M clusters.
(3) Modeling is carried out on a training data set according to the thought of model regression learning, two sub-models of track prediction are obtained through training, the reliability of the sub-models in the mixed model is analyzed and evaluated, the weight of the sub-models is dynamically adjusted according to the reliability of the sub-models, and finally the Gaussian mixed time sequence model is obtained.
(4) The coordinate information of the vehicle is used as the input of the hybrid model, the obtained output is track point information predicted by the model, the track prediction of the moving object is realized, meanwhile, the actual track point coordinate information of the moving object is stored in a database, the calculation of track prediction errors and the updating of training data are carried out, the effectiveness of the data is guaranteed, and the dynamic prediction of the hybrid model is realized.
Specific description of each module is given below:
(1) data pre-processing
The unsupervised k-means clustering algorithm provided by the invention is based on a regression model, M clusters are automatically obtained according to the characteristics of track data, and the condition that the selected data points meet the requirements in the clustering process is as follows:
(1) the selection of historical track data points is random in different regions.
(2) Each data point is present in a cluster, and each cluster has a weight ωiAnd the weight satisfiesThe M value is not obtained through experimental summary, is not artificially and subjectively determined, and is obtained according to an unsupervised k-means algorithm, so that the M value can objectively reflect the discrete state of data.
The autocorrelation analysis of the time series prediction module comprises the analysis of autocorrelation coefficients and partial correlation coefficients, and the identification of time series characteristics is realized through comparative analysis. And estimating parameters of the ARIMA model through repeated tests according to the characteristics of the autocorrelation function and the partial autocorrelation function of the transformed sequence. According to the method, the model is selected according to the characteristic that the sampling frequency is required to be small (the sampling frequency is a period of five years) in the selection of the traffic flow data, so that the complex process of uncertainty identification in the parameter determining process of the time series model is eliminated, the time overhead of the model is reduced, only the model order and the model parameter need to be solved, the implementation consumption of the algorithm is reduced, and the dynamic property and the real-time property of algorithm prediction are ensured.
(2) Regression learning of partial models
The track prediction model is reasonable and efficient in predicting the future track coordinate point of the vehicle through the idea of a linear regression method, and historical track data and traffic flow data are subjected to predictionWhen predicting the trace point of the next moment by learning, the space-time trace data processing needs to respectively model the x direction and the y direction, wherein the x direction is the direction of the modelt＝f(x1,x2,...,xt-1),yt＝f(y1,y2,...,yt-1) Are regression functions of the respective trajectory predictions obtained by individual modeling. Will train the data setIs divided intoAndthe processing in two directions is carried out,vector sets representing trace points in the x and y directions, respectively. Sequence set { x) using historical tracks1,x2,...,xn}、{y1,y2,...,ynCalculate the next time position (x)n+1,yn+1,tn+1) Has a value of f (x)n+1)＝f(xn)+Δxn+1、f(yn+1)＝f(yn)+Δyn+1Wherein, f (x)n)、f(yn) Both are regression functions, Δ x, obtained by a trajectory modeln+1＝f(Δxn)+ε、Δyn+1＝f(Δyn) + ε, ε is the data noise, ε -N (0, σ)2) The probability distribution function is:
wherein the training data set DtrainX is input data, and y is output data. Test data set DtestX is input test data, y is output predicted value, M is the number of clusters,weight representing the clustering, GP (y, f)i(y),Σiy) Representing probability functions conforming to the Gaussian process, y being input data, μiyMean, sigma, representing the ith clusteriyRepresents the variance of the ith cluster; f (y, mu)i,Σi) Is a model regression function; omegaiIs the ith Gaussian weight, and the mean value is represented by the average value,representing the variance, the predicted value of y is:
the trace point formed by the three-dimensional space-time data at a certain moment is marked as (x)i,yi,ti) Then the historical track chain is represented as: s { (x)1,y1,t1),(x2,y2,t2),...,(xn,yn,tn)}＝{S1,S2,...,Sn}. Obtaining a regression prediction model through learning of the historical track chain, and predicting the next time through the regression model to obtain the position information (x ') in the x and y directions at the n +1 time'n+1,y'n+1) Then the future trajectory is represented as: sn+1＝(x'n+1,y'n+1,tn+1)＝(xn+Δxn+1,yn+Δyn+1,tn+1) Wherein, Δ xn+1,Δyn+1The actual position information at the time n +1 is (x) in the increment of x and yn+1,yn+1) Then the prediction error is:the prediction error density is then:
the time series model will be at a series of times t for the road1,t2,...,tnAre arranged in time sequence, when t is an independent variable1＜t2＜...＜tnThe resulting discrete numbers form a set of sequences { x (t) }1),x(t2),...,x(tn) And (6) performing curve fitting and parameter estimation methods through the obtained time sequence data to establish a mathematical model so as to obtain a prediction result. It generally employs curve fitting or parameter estimation methods (e.g., non-linear least squares).
The prediction of the road section traffic flow is divided into the following four steps:
(1) and selecting short-time dynamic traffic flow data from the database as a sample sequence.
(2) And (3) carrying out stationarity test on the original flow sequence through the correlation diagram and the partial correlation diagram, and stabilizing the original sequence into a stable sequence by adopting a difference method for the non-stationarity time sequence.
(3) And performing parameter estimation on the preliminarily selected ARIMA (p, d,0) model, wherein the parameter estimation comprises a significance test and a randomness test on the parameters. The least squares method is used for parameter estimation and the BIC criterion is used for determining the model order, and the criterion BIC is defined asBayesian Information Criterion (abbreviated BIC), where N represents the total amount of training data;representing the variance of the model; p represents the upper limit of the model order.
(4) In order to realize dynamic prediction of traffic flow, real-time traffic flow data needs to be collected, a database needs to be updated, and finally the steps (1) to (3) are repeated.
The root mean square error of the traffic flow data can reflect the error between the actual traffic flow and the predicted traffic flow and describe the discrete condition of the prediction error, and the larger the root mean square error value is, the larger the RMSER value is, and the more discrete the sequence of the prediction error is, the worse the prediction effect is. Predicting the traffic flow of M clusters by the ARIMA model, wherein the predicted traffic flow is NfActual vehicle flow is NhThen the prediction error density of the model is:
FIG. 7 is a flowchart of the reliability analysis of the model in the method of the present invention, the process is as follows:
vectorizing the original data, and calling a data preprocessing module. Firstly, whether data are subjected to separation processing or not is judged, and if the data are processed, model parameters are directly determined; if the data is not preprocessed, clustering or smoothing the data. Then, determining model parameters according to the dynamic data, and if the model parameters are determined, performing model training; otherwise, estimating the model parameters and determining the model parameters. And then, carrying out prediction error analysis on the model, calculating the prediction error rate, and finally, carrying out model reliability calculation.
(3) Weight training of partial models
In the reliability analysis process of the two sub-models of the Gaussian mixture model and the time series model, because the error generated by noise data is inevitable, the ratio of the weight sum of each sub-model and all sub-models is directly used as the weight, the weight error is larger, and therefore, in order to ensure the accuracy of weight calculation through the reliability of the two sub-models, the model similarity S is provided(R1,R2)Definition of (1)，R1、R2Reliability of the two models, d(R1,R2)＝|R1-R2I represents the degree of difference in reliability between different models, and S(R1,R2)And d(R1,R2)The greater the distance between the reliability analysis results, the smaller the similarity of the two models, and the model similarity is as follows:
the degree of mutual support of different models can be expressed as:
the mutual support degree between the two model results is respectively used as the dynamic weight of each model, and can be expressed as:
when the two sub-models are probability distribution models, when the sub-models compare the probabilities of different lines, if the reliability of the sub-models is not large in the working day period, the line ratio obtained by a Gaussian mixture model for predicting according to the historical track of the vehicle is almost the same as the line ratio obtained by a traffic flow prediction model, and the probability of summing the sub-models is consistent with the probabilities predicted by the two sub-models; during the non-working day, the reliability of the time series model is higher than that of the Gaussian mixture model, so that the product result of the probability ratio of the model and the weight is biased to the time series model, and the accurate prediction effect of the mixture model is achieved, therefore, the Gaussian mixture time series model is defined as follows:
the idea of GMTSM is: performing data modeling of a probability density function, separating training track data, and realizing parameter estimation of the submodel; obtaining regression functions and stable time sequences of M Gaussian components according to normal distribution; obtaining the prediction error density of the partial model and the weight of the partial model; and finally, mixing the weighted regression functions to complete regression prediction of the track. Wherein, ω is1+ω2≠1，ω1、ω2The value of (d) is based on the similarity between the GMTSM, GMM and time series model, so the sum of the weights of the two partial models is not equal to 1.Is a Gaussian mixture model, GP denotes the Gaussian process in the x, y directions, GP (x, y | μ |)i,Σi) Representing the characteristic vectors which are in line with the directions x and y, training the historical track by the model to obtain different routes at the next moment, wherein ARIMA (p, d,0) is a time sequence model, the mixed model is trained and learned according to historical data to obtain the probability of different future tracks of the vehicle, and a future track chain with high probability is the predicted future track of the mixed model.
FIG. 8 is a flow chart of the training of the weights of the hierarchal model in the method of the present invention, the process is as follows:
firstly, judging whether the sub-models are subjected to reliability analysis, if so, directly calculating the similarity of the models, and otherwise, performing the reliability analysis of the models; then, model support degree calculation is carried out, and finally, model weight calculation is carried out.
According to the embodiments, the invention provides a hybrid prediction model of a historical track learning model and a traffic flow prediction model based on probability distribution, starting from two aspects of vehicle historical track regression learning and road traffic flow prediction, aiming at the defects that the prediction accuracy is not high and the prediction instantaneity is not good when the road traffic flow is suddenly changed in the current vehicle track prediction algorithm. The hybrid model proposed herein has the advantages of: the hybrid model automatically adjusts parameter information in the model through learning historical data, realizes dynamic trajectory prediction of the vehicle, and improves the accuracy of the trajectory prediction model.
Claims (8)
1. A trajectory prediction method based on a Gaussian mixture time series model is used for vehicle trajectory prediction, provides an optimal route for a user and plans an intelligent transportation network, and is characterized by comprising the following steps:
(1) preprocessing vehicle related data acquired by a GPS through an ETL technology to realize the separation of traffic flow data and vehicle historical track data; the traffic data is two-dimensional data comprising traffic and a timestamp; the historical track data is three-dimensional data comprising longitude, latitude and time stamp; converting the two types of separated data into vector data and storing the vector data in a specified database;
(2) performing stabilization analysis processing on the two-dimensional traffic flow data to obtain stabilization data; performing clustering analysis processing on the three-dimensional historical track data by using a k-means algorithm, wherein clustering model parameters are obtained by using a maximum likelihood Estimation (EM) algorithm, so that probability distribution based on the historical track data is maximized, and finally obtaining M clusters;
(3) modeling a training data set according to model regression learning, training to obtain two sub-models of track prediction, analyzing and evaluating the reliability of the sub-models in a mixed model, dynamically adjusting the weight of the sub-models according to the reliability of the sub-models, and finally obtaining a Gaussian mixed time sequence model;
(4) the coordinate information of the vehicle is used as the input of the hybrid model, the obtained output is track point information predicted by the model, the track prediction of the moving object is realized, meanwhile, the actual track point coordinate information of the moving object is stored in a database, the calculation of track prediction errors and the updating of training data are carried out, the effectiveness of the data is guaranteed, and the dynamic prediction of the hybrid model is realized.
2. The trajectory prediction method based on the gaussian mixture time series model according to claim 1, wherein the unsupervised k-means clustering algorithm automatically obtains M clusters according to the characteristics of trajectory data based on a regression model, and in the clustering process, the condition that the selected data points meet the requirements is as follows:
(1) the selection of the historical track data points is random in different areas;
(2) each data point is present in a cluster, and each cluster has a weight ωiAnd the weight satisfiesThe M value is obtained according to an unsupervised k-means algorithm.
3. The trajectory prediction method based on the Gaussian mixture time series model as claimed in claim 1, wherein the back of the partial modelIn the regression learning, the track prediction model is reasonable and efficient in predicting the future track coordinate points of the vehicle through the idea of a linear regression method, when historical track data and traffic flow data are learned to predict track points of the next moment, the space-time track data are processed to be respectively modeled in the directions x and y, and the directions x and y are required to be processedt＝f(x1,x2,...,xt-1)、yt＝f(y1,y2,...,yt-1) Is a regression function of the respective trajectory predictions obtained by individual modeling; training setFor the x and y directionsAndseparate treatment; historical track sequence set using x-direction { x1,x2,...,xnCalculating the position x of the next timen+1Value of (a), f (x)n+1)＝f(xn)+Δxn+1，Δxn+1＝f(Δxn)+ε，f(Δxn) Is obtained by a regression function through a track prediction model, wherein epsilon is data noise and epsilon-N (0, sigma)2) (ii) a Historical track sequence set using y direction1,y2,...,ynCalculating the position y at the next momentn+1Value of f (y)n+1)＝f(yn)+Δyn+1，Δyn+1＝f(Δyn)+ε，f(Δyn) Is obtained by a regression function through a track prediction model, wherein epsilon is data noise and epsilon-N (0, sigma)2)；
The tracing point formed by the three-dimensional space-time data is marked as (x)i,yi,ti) The historical track chain is represented as: s { (x)1,y1,t1),(x2,y2,t2),...,(xn,yn,tn)}＝{S1,S2,...,Sn}; through learning historical track chainObtaining a regression prediction model, and predicting the next time by the regression model to obtain the position information (x ') in the x and y directions at the n +1 time'n+1,y'n+1) Then the future trajectory is represented as: sn+1＝(x'n+1,y'n+1,tn+1)＝(xn+Δxn+1,yn+Δyn+1,tn+1) Wherein, Δ xn+1、Δyn+1The actual position information at the time n +1 is (x) in increments of x and y directionsn+1,yn+1) Then the prediction error is:the prediction error density is then:m is the number of trace point clusters, and T is the time consumed by the model to start regression learning until the prediction of the next moment is finished;
the time series model will be at a series of times t for the road1,t2,...,tnArranged according to time sequence, when t1＜t2＜...＜tnThe resulting discrete numbers form a set of sequences { x (t) }1),x(t2),...,x(tn) And (6) performing curve fitting and parameter estimation methods through the obtained time sequence data to establish a mathematical model so as to obtain a prediction result.
4. The trajectory prediction method based on the Gaussian mixture time series model according to claim 3, characterized in that the prediction of the road section traffic flow is divided into the following four steps:
(1) selecting short-time dynamic traffic flow data from a database as a sample sequence;
(2) carrying out stationarity test on an original flow sequence through a correlation diagram and a partial correlation diagram, and carrying out stationarity treatment on a non-stationarity time sequence by adopting a difference method to convert the original sequence into a stationarity sequence;
(3) performing parameter estimation on the preliminarily selected ARIMA (p, d,0) model, includingCarrying out significance test and randomness test on the parameters; using least squares for parameter estimation and a BIC criterion for model order determination, the criterion BIC being defined as
(4) In order to realize dynamic prediction of traffic flow, real-time traffic flow data needs to be collected, a database needs to be updated, and finally the steps (1) to (3) are repeated.
5. The trajectory prediction method based on the Gaussian mixture time series model as claimed in claim 1, wherein the ARIMA model predicts the traffic flow of N roads, the predicted traffic flow is Nf, and the actual traffic flow is NhThen the prediction error rate of the model is:m is the number of trace point clusters, and T is the time consumed by the model to start regression learning until the prediction of the next moment is finished.
6. The trajectory prediction method based on the Gaussian mixture time series model as claimed in claim 1, wherein in weight training of the partial models, two partial models M1、M2Obtaining two reliability measurement results R by calculating the reliability of different models1、R2Calculating the distance between reliabilities respectivelyBy the formulaCalculating the similarity of reliability estimation results of different models; if the estimated results R of the two models1＝R2Indicate that their similarity is S(R1,R2)The larger the distance between the estimation results is, the smaller the similarity of the two models is;
the degree of mutual support of the two models can be expressed as:
the mutual support degree between the two model results is respectively used as the dynamic weight of each model, and can be expressed as:
7. the trajectory prediction method based on the Gaussian mixture time series model as claimed in claim 1, wherein after the weights of the partial models are obtained, the hybrid model is constructed;
the gaussian mixture time series model is defined as:
8. the trajectory prediction method based on the gaussian mixture time series model according to claim 1, wherein the regression equation in x and y directions is:wherein,xf、yfto output data, xh、yhAs historical data, ωiIs the ith weight of the gaussian to be weighted,m is the number of trace point clusters; the ith Gaussian density function is GP (x)h,xf|μi,Σi) Wherein, muiMeans, Σ, representing data in x, y directionsiDenotes the variance of the data in the x and y directions, f (x)i)、f(xi) Are regression functions in the x and y directions, respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710683275.3A CN107610464B (en) | 2017-08-11 | 2017-08-11 | A kind of trajectory predictions method based on Gaussian Mixture time series models |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710683275.3A CN107610464B (en) | 2017-08-11 | 2017-08-11 | A kind of trajectory predictions method based on Gaussian Mixture time series models |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107610464A CN107610464A (en) | 2018-01-19 |
| CN107610464B true CN107610464B (en) | 2019-02-12 |
Family
ID=61065268
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710683275.3A Active CN107610464B (en) | 2017-08-11 | 2017-08-11 | A kind of trajectory predictions method based on Gaussian Mixture time series models |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107610464B (en) |
Families Citing this family (73)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108334897B (en) * | 2018-01-22 | 2023-04-07 | 上海海事大学 | Offshore floater track prediction method based on self-adaptive Gaussian mixture model |
| DE102019202090A1 (en) * | 2018-03-14 | 2019-09-19 | Robert Bosch Gmbh | A method of generating a training data set for training an artificial intelligence module for a controller of a robot |
| CN108446490A (en) * | 2018-03-19 | 2018-08-24 | 辽宁石油化工大学 | A kind of open cell formula road optimum design method |
| CN108515862A (en) * | 2018-03-26 | 2018-09-11 | 唐天才 | A kind of data intelligence for electric vehicle manages system |
| CN110162379B (en) * | 2018-04-24 | 2023-04-21 | 腾讯云计算（北京）有限责任公司 | Virtual machine migration method and device and computer equipment |
| CN108763538B (en) * | 2018-05-31 | 2019-07-23 | 北京嘀嘀无限科技发展有限公司 | A kind of method and device in the geographical location determining point of interest POI |
| CN108955855B (en) * | 2018-06-08 | 2020-05-08 | 山东大学 | Vibration signal feature extraction method, monitoring method and device for rotary machine |
| CN109034448B (en) * | 2018-06-14 | 2022-02-11 | 重庆邮电大学 | Trajectory prediction method based on vehicle trajectory semantic analysis and deep belief network |
| CN109255315B (en) * | 2018-08-30 | 2021-04-06 | 跨越速运集团有限公司 | People and vehicle separation judgment method and device during vehicle leaving |
| CN109147325B (en) * | 2018-09-04 | 2022-01-28 | 广州视源电子科技股份有限公司 | Road condition prediction method and device, storage medium and processor |
| CN109165245B (en) * | 2018-09-19 | 2020-06-26 | 北京航空航天大学 | Time-space mode mining method of mobile trajectory generation model for multi-source data fusion |
| CN109543715A (en) * | 2018-10-23 | 2019-03-29 | 武汉理工大学 | A kind of ship air route is extracted and the method for track deviation detection |
| CN109598070A (en) * | 2018-12-06 | 2019-04-09 | 北京搜狐新动力信息技术有限公司 | A kind of Time Series Forecasting Methods and platform |
| CN109584557B (en) * | 2018-12-14 | 2021-02-26 | 北京工业大学 | Traffic flow prediction method based on dynamic decomposition mode and matrix filling |
| CN109783877B (en) * | 2018-12-19 | 2024-03-01 | 平安科技（深圳）有限公司 | Time sequence model establishment method, device, computer equipment and storage medium |
| CN109782754B (en) * | 2018-12-25 | 2022-05-06 | 东软睿驰汽车技术（沈阳）有限公司 | Vehicle control method and device |
| CN109800801B (en) * | 2019-01-10 | 2020-12-01 | 浙江工业大学 | K-Means cluster analysis lane flow method based on Gaussian regression algorithm |
| CN109800916B (en) * | 2019-01-23 | 2021-04-23 | 山东省交通规划设计院有限公司 | Modeling method for traffic flow of driving-in traffic in expressway service area |
| CN109829936B (en) * | 2019-01-29 | 2021-12-24 | 青岛海信网络科技股份有限公司 | Target tracking method and device |
| CN109871599B (en) * | 2019-01-30 | 2023-05-23 | 湖南湘江智车出行科技有限公司 | Unmanned vehicle operation safety index evaluation method |
| CN109859467B (en) * | 2019-01-30 | 2020-11-10 | 银江股份有限公司 | Mining analysis method for environmental impact factors in traffic model |
| CN109859477B (en) * | 2019-03-15 | 2021-03-19 | 同盾控股有限公司 | Congestion data determination method and device |
| CN111915875B (en) * | 2019-05-08 | 2024-07-19 | 阿里巴巴集团控股有限公司 | Method and device for processing traffic flow path distribution information and electronic equipment |
| CN110263811B (en) * | 2019-05-21 | 2021-03-26 | 上海应势信息科技有限公司 | Equipment running state monitoring method and system based on data fusion |
| CN110264586A (en) * | 2019-05-28 | 2019-09-20 | 浙江零跑科技有限公司 | L3 grades of automated driving system driving path data acquisitions, analysis and method for uploading |
| CN110363983B (en) * | 2019-06-24 | 2021-08-31 | 长沙理工大学 | Non-signalized intersection unmanned and non-manned real-time trajectory prediction and conflict recognition method |
| CN110264735B (en) * | 2019-06-28 | 2021-11-30 | 佛山科学技术学院 | Traffic jam prediction system and method based on big data and storage medium |
| CN110475206B (en) * | 2019-06-28 | 2020-11-13 | 安徽四创电子股份有限公司 | Monitoring and early warning method based on wifi log |
| CN110264721B (en) * | 2019-07-01 | 2021-04-20 | 北京理工大学 | Urban intersection surrounding vehicle track prediction method |
| CN110634285B (en) * | 2019-08-05 | 2021-10-08 | 江苏大学 | Road section travel time prediction method based on Gaussian mixture model |
| CN110909592B (en) * | 2019-10-11 | 2020-12-18 | 重庆特斯联智慧科技股份有限公司 | Target tracking method and system based on multi-scale characteristic quantity |
| CN112652161B (en) * | 2019-10-12 | 2023-01-06 | 阿里巴巴集团控股有限公司 | Method and device for processing traffic flow path distribution information and electronic equipment |
| CN110796858A (en) * | 2019-10-24 | 2020-02-14 | 山东科技大学 | Vehicle track prediction method and system based on video vehicle passing data |
| CN112862137A (en) * | 2019-11-27 | 2021-05-28 | 顺丰科技有限公司 | Method and device for predicting quantity, computer equipment and computer readable storage medium |
| CN111079804A (en) * | 2019-12-02 | 2020-04-28 | 中南大学 | Vehicle motion track prediction method and system of improved Gaussian mixture model |
| CN111027202B (en) * | 2019-12-04 | 2023-12-15 | 北京软通绿城科技有限公司 | Digital city prediction method, device, equipment and storage medium |
| CN111091708B (en) * | 2019-12-13 | 2020-11-03 | 中国科学院深圳先进技术研究院 | Vehicle track prediction method and device |
| CN111209475B (en) * | 2019-12-27 | 2022-03-15 | 武汉大学 | Interest point recommendation method and device based on space-time sequence and social embedded ranking |
| CN111241066B (en) * | 2020-01-10 | 2024-06-25 | 平安科技（深圳）有限公司 | Platform database automation operation and maintenance method, device and computer readable storage medium |
| CN111309770B (en) * | 2020-02-24 | 2023-03-28 | 电子科技大学 | Automatic rule generating system and method based on unsupervised machine learning |
| JP7327688B2 (en) * | 2020-03-12 | 2023-08-16 | 日本電気株式会社 | Event detection device, method and program |
| CN111445461A (en) * | 2020-03-30 | 2020-07-24 | 上海眼控科技股份有限公司 | Radar cloud picture prediction method and device, computer equipment and storage medium |
| CN111724592B (en) * | 2020-06-12 | 2022-11-25 | 重庆大学 | Highway traffic jam detection method based on charging data and checkpoint data |
| CN112070262B (en) * | 2020-07-09 | 2022-06-14 | 深圳市计量质量检测研究院（国家高新技术计量站、国家数字电子产品质量监督检验中心） | Air conditioner load prediction method based on support vector machine |
| CN112116806B (en) * | 2020-08-12 | 2021-08-24 | 深圳技术大学 | Traffic flow characteristic extraction method and system |
| CN112036082B (en) * | 2020-08-27 | 2022-03-08 | 东北大学秦皇岛分校 | Time series data prediction method based on attention mechanism |
| CN112070808A (en) * | 2020-09-01 | 2020-12-11 | 三一专用汽车有限责任公司 | Trajectory prediction method, apparatus and computer-readable storage medium |
| CN112287599B (en) * | 2020-10-09 | 2023-08-04 | 暨南大学 | Multi-step novelty-based temporal sequence iterative prediction algorithm, medium and device |
| CN112215429A (en) * | 2020-10-20 | 2021-01-12 | 清华大学 | Historical trajectory-based trajectory prediction method and device |
| CN112699536B (en) * | 2020-12-14 | 2022-09-13 | 武汉科技大学 | Self-adaptive target ship track prediction and locking method |
| CN112785863B (en) * | 2020-12-26 | 2022-05-03 | 浙江天行健智能科技有限公司 | Merging decision classification early warning method based on K-Means and entropy weighting |
| CN112734176B (en) * | 2020-12-28 | 2022-04-08 | 中创三优（北京）科技有限公司 | Charging station building method and device, terminal equipment and readable storage medium |
| CN112833903B (en) * | 2020-12-31 | 2024-04-23 | 广州文远知行科技有限公司 | Track prediction method, device, equipment and computer readable storage medium |
| CN112833884B (en) * | 2020-12-31 | 2021-09-28 | 郑州轻工业大学 | Navigation method and navigation system based on big data |
| CN112733452B (en) * | 2021-01-13 | 2024-03-29 | 北京嘀嘀无限科技发展有限公司 | Track prediction method, track prediction device, electronic equipment and readable storage medium |
| CN114792148A (en) * | 2021-01-25 | 2022-07-26 | 魔门塔(苏州)科技有限公司 | Method and device for predicting motion trail |
| CN112783950B (en) * | 2021-02-01 | 2024-04-23 | 西北工业大学 | Human movement predictability quantization method based on information entropy |
| CN112836749A (en) * | 2021-02-03 | 2021-05-25 | 中国工商银行股份有限公司 | System resource adjusting method, device and equipment |
| CN112829744B (en) * | 2021-02-09 | 2022-04-15 | 清华大学 | Vehicle long time domain track prediction method based on longitudinal and transverse coupling |
| CN113468751B (en) * | 2021-07-05 | 2022-12-27 | 河南中烟工业有限责任公司 | Recursion Lasso-based flowmeter anomaly online monitoring method and system and storage medium |
| CN113469075A (en) * | 2021-07-07 | 2021-10-01 | 上海商汤智能科技有限公司 | Method, device and equipment for determining traffic flow index and storage medium |
| CN114119648A (en) * | 2021-11-12 | 2022-03-01 | 史缔纳农业科技(广东)有限公司 | Pig counting method for fixed channel |
| CN114676176B (en) * | 2022-03-24 | 2024-07-26 | 腾讯科技（深圳）有限公司 | Method, device, equipment and program product for predicting time sequence |
| CN114664094B (en) * | 2022-05-26 | 2022-08-26 | 青岛慧拓智能机器有限公司 | Vehicle track prediction method, device, chip, terminal and computer equipment |
| CN114863685B (en) * | 2022-07-06 | 2022-09-27 | 北京理工大学 | Traffic participant trajectory prediction method and system based on risk acceptance degree |
| CN115392407B (en) * | 2022-10-28 | 2023-03-24 | 中建五局第三建设有限公司 | Non-supervised learning-based danger source early warning method, device, equipment and medium |
| CN115685792B (en) * | 2022-12-29 | 2023-03-31 | 北京万维盈创科技发展有限公司 | Wastewater intermittent drain outlet flow triggering method and device based on flow threshold |
| CN116662111B (en) * | 2023-05-05 | 2023-11-17 | 浙江锐明智能控制技术有限公司 | Intelligent network management control system for train |
| CN116401570B (en) * | 2023-05-26 | 2023-08-11 | 佛山市力天包装印刷有限公司 | Intelligent processing system for printing quality monitoring big data |
| CN117134504B (en) * | 2023-10-25 | 2024-01-26 | 陕西禄远电子科技有限公司 | Intelligent energy monitoring method and system based on safety protection |
| CN117725764B (en) * | 2024-02-07 | 2024-05-03 | 中汽研汽车检验中心（天津）有限公司 | Regression model-based vehicle chassis multi-objective optimization method, equipment and medium |
| CN117775078B (en) * | 2024-02-28 | 2024-05-07 | 山西阳光三极科技股份有限公司 | Method for judging running direction of freight train in mine based on deep learning |
| CN118400715A (en) * | 2024-06-27 | 2024-07-26 | 南京邮电大学 | Method for detecting node failure in Internet of vehicles based on Gaussian mixture model |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201307343Y (en) * | 2008-11-14 | 2009-09-09 | 上海经达实业发展有限公司 | Navigation device of vehicle dynamic route |
| CN103914985B (en) * | 2014-04-25 | 2015-10-28 | 大连理工大学 | A kind of hybrid power passenger car following speed of a motor vehicle trajectory predictions method |
| CN105185115B (en) * | 2015-09-07 | 2018-04-06 | 华迪计算机集团有限公司 | A kind of traffic forecast method and forecasting system |
| CN206378893U (en) * | 2016-12-30 | 2017-08-04 | 安徽理工大学 | A kind of condition of road surface based on Internet of Things and big data in real time and predicted query system |
-
2017
- 2017-08-11 CN CN201710683275.3A patent/CN107610464B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN107610464A (en) | 2018-01-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107610464B (en) | A kind of trajectory predictions method based on Gaussian Mixture time series models | |
| CN109658695B (en) | Multi-factor short-term traffic flow prediction method | |
| Xu et al. | Accurate and interpretable bayesian mars for traffic flow prediction | |
| CN110570651A (en) | Road network traffic situation prediction method and system based on deep learning | |
| CN111985710A (en) | Bus passenger trip station prediction method, storage medium and server | |
| CN108108831B (en) | Destination prediction method and device | |
| CN109191845B (en) | Bus arrival time prediction method | |
| CN107346460A (en) | Following operating mode Forecasting Methodology based on the lower front truck operation information of intelligent network contact system | |
| CN112884222B (en) | Time-period-oriented LSTM traffic flow density prediction method | |
| CN112101132B (en) | Traffic condition prediction method based on graph embedding model and metric learning | |
| KR101703972B1 (en) | System and method for predicting groundwater potential area using spatial information | |
| CN115269758A (en) | Passenger-guidance-oriented road network passenger flow state deduction method and system | |
| CN113888867B (en) | Parking space recommendation method and system based on LSTM (least squares) position prediction | |
| Picano et al. | Nonlinear dynamic chaos theory framework for passenger demand forecasting in smart city | |
| Toplak et al. | Novel road classifications for large scale traffic networks | |
| CN108053646B (en) | Traffic characteristic obtaining method, traffic characteristic prediction method and traffic characteristic prediction system based on time sensitive characteristics | |
| Tabibiazar et al. | Kernel-based modeling and optimization for density estimation in transportation systems using floating car data | |
| Fahs et al. | Traffic congestion prediction based on multivariate modelling and neural networks regressions | |
| CN117057848B (en) | Takeout order demand prediction method based on GAT-GPR | |
| CN112287503A (en) | Dynamic space network construction method for traffic demand prediction | |
| CN108055639A (en) | A kind of density of stream of people Forecasting Methodology and system based on signaling data analysis | |
| Vu et al. | Bus running time prediction using a statistical pattern recognition technique | |
| CN115691140B (en) | Analysis and prediction method for space-time distribution of automobile charging demand | |
| CN111524358A (en) | Regional radiation sexual communication flow prediction method | |
| CN116989801A (en) | Map matching method and device for low-frequency long tracks of complex road network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |