CN109598935A - A kind of traffic data prediction technique based on ultra-long time sequence - Google Patents
A kind of traffic data prediction technique based on ultra-long time sequence Download PDFInfo
- Publication number
- CN109598935A CN109598935A CN201811532408.8A CN201811532408A CN109598935A CN 109598935 A CN109598935 A CN 109598935A CN 201811532408 A CN201811532408 A CN 201811532408A CN 109598935 A CN109598935 A CN 109598935A
- Authority
- CN
- China
- Prior art keywords
- data
- traffic
- traffic data
- long time
- ultra
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000012545 processing Methods 0.000 claims abstract description 6
- 238000009499 grossing Methods 0.000 claims abstract description 5
- 210000004027 cell Anatomy 0.000 claims description 12
- 238000003062 neural network model Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 238000012417 linear regression Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 230000000737 periodic effect Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000000306 recurrent effect Effects 0.000 claims description 3
- 230000002441 reversible effect Effects 0.000 claims description 3
- 230000003203 everyday effect Effects 0.000 claims description 2
- 230000008447 perception Effects 0.000 abstract description 5
- 230000001537 neural effect Effects 0.000 abstract 1
- 230000007787 long-term memory Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0133—Traffic data processing for classifying traffic situation
Landscapes
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of traffic data prediction techniques based on ultra-long time sequence, first, position mark processing is carried out to historical traffic data, the method for carrying out position mark processing to historical traffic data is as follows: determining the corresponding relationship for the data space that individual data position and ultra-long time sequence are formed in traffic data ultra-long time sequence；Position mark is made to the traffic data in traffic data ultra-long time sequence according to corresponding relationship；The ultra-long time sequence refers to across multiple time serieses for dividing the period；Finally, choosing the historical traffic data with position mark as input data, traffic data prediction is carried out.Using a kind of method arbitrary in neural net model method, the method for moving average, exponential smoothing, AR modelling when traffic data is predicted.The present invention improves traffic data precision of prediction by the correlation of Data Position in perception traffic data ultra-long time sequence.
Description
Technical Field
The invention relates to the field of traffic data prediction, in particular to a traffic data prediction method based on an ultra-long time sequence.
Background
The development of real-time dynamic traffic makes real-time accurate traffic data popular. However, traffic data is greatly affected by the time factor. Mainly divided into active factors and passive factors. The active factors are: weekday and non-weekday, seasonal changes, time lapse, and the like. The passive factors are: holidays, activities that lead traffic changes, etc.
The traffic data time sequence can be divided according to time periods of minutes, hours, days, weeks, months, years and the like, and traffic data prediction mainly utilizes traffic data of historical time periods to predict traffic data of future time periods. The existing traffic prediction methods, such as a moving average method, an exponential smoothing method, an AR model method and the like, have weak perception on an ultra-long time sequence spanning multiple divided time periods; therefore, it is necessary to design a traffic data prediction method based on an ultra-long time series.
Disclosure of Invention
The invention aims to overcome the defects and provides a traffic data prediction method based on an overlong time sequence.
The invention achieves the aim through the following technical scheme: a traffic data prediction method based on an ultra-long time sequence comprises the following steps:
(1) the method for carrying out the position marking processing on the historical traffic data comprises the following steps:
(1.1) determining the corresponding relation between a single data position in the traffic data super-long time sequence and a data space formed by the super-long time sequence;
(1.2) marking the position of the traffic data in the traffic data super-long time sequence according to the corresponding relation; the ultra-long time sequence refers to a time sequence spanning multiple divided time periods;
(2) and selecting historical traffic data with position marks as input data to predict the traffic data.
Preferably, in the step (1.1), any one of a regression model method, a user setting method and a neural network model method can be adopted to determine the corresponding relation between a single data position in the traffic data super-long time sequence and a data space formed by the super-long time sequence; preferably, a regression model method is used, the formula is as follows:
Q＝∑(y-∑(wpxp+bp))＝min
wherein y is traffic data at a single data position, p represents a data space position, and w and b which minimize the Q function are solved as the corresponding relation between the single data position and the data space.
Preferably, in the step (1.2), the traffic data in the traffic data super-long time sequence is marked according to the corresponding relation, wherein w and b are obtained by adopting a regression model, and different grades are divided according to the size of w and are used as position marks.
Preferably, the divided time interval is determined according to a periodic phenomenon presented by traffic data, a plurality of groups of traffic data time sequences with a time window as a unit are selected, the average similarity is calculated, and when the average similarity meets a set threshold value, the time window can be used as the divided time interval; the similarity can be calculated by cosine similarity, Pearson correlation coefficient and Jaccard similarity coefficient.
Preferably, in the step (2), the traffic data prediction is performed by using any one of a neural network model method, a moving average method, an exponential smoothing method, and an AR model method.
Preferably, the traffic data includes, but is not limited to, data collected by traffic devices, data converted from data collected by traffic devices; the data collected by the traffic equipment includes, but is not limited to, traffic flow and traffic speed, and the data converted according to the data collected by the traffic equipment includes, but is not limited to, traffic state and traffic index.
Preferably, the neural network model method preferably adopts an LSTM long-term neural network to predict traffic data, specifically as follows:
(i) performing traffic flow prediction by setting an input driving sequence, and establishing a linear regression model for a predicted value and a remote traffic flow value;
(ii) after training, screening out the flow value of which the position mark meets the requirement, and marking the position of the remote flow value with large weight obtained by screening;
(iii) adding the traffic flow value with the position mark and the weight into an LSTM network input layer, carrying out back propagation training of LSTM +, firstly calculating the output value of each neuron in a forward direction, secondly calculating the error term value of each neuron in a reverse direction, and finally calculating the gradient of each weight according to the corresponding error term and updating the weight value;
(iv) and predicting the traffic flow in a future preset period by using the trained LSTM network model.
Preferably, in step (i), the drive sequence of the setting input is X ═ X (X)1，1，1，x1，1，2，...，xkk，nn，mm，ykk，nn+1，1，ykk，nn+1，2，...，ykk，nn+1，t-1，)，
Wherein y represents data of the prediction day, and the flow prediction problem is essentially that the flow y is built and predicted through the driving sequence Xkk，nn+1，tTo (3) is performed.
Preferably, in the step (i), when a linear regression model is established, data around the time interval t of each day is selected, and the regression model is as follows:
wherein,d is the step size of the image,and b are the weight and bias of the regression model, respectively.
Preferably, the LSTM network is a special RNN recurrent neural network, and compared with RNN, an "input gate", "forget gate", "output gate", "input gate", "forget gate", "output gate" are added, and LSTM cell states are as follows:
an input gate:
forget the door:
cell state:
an output gate:
where I represents the length of the input current sequence, C represents the number of LSTM cells, P represents the number of labeled high impact traffic values, l represents the input status,indicating the state of the forgetting gate, o indicating the state of the output gate, f and g being activation functions,represents the c-th cell state value at time t, tgpThe marked traffic flow value representing the pth weighted, the final output is:
the invention has the beneficial effects that: the method and the device improve the traffic data prediction accuracy by sensing the relevance of the data positions in the traffic data super-long time sequence, marking the importance of the data positions, and screening the traffic data of the positions to serve as the basis of traffic data prediction. Preferably, the traffic flow value of the important position is added into the LSTM, so that the LSTM model not only has a long-term and short-term memory function, but also has a certain memory function for an ultra-long term, and the traffic flow prediction is more accurate.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
Detailed Description
The invention will be further described with reference to specific examples, but the scope of the invention is not limited thereto:
example 1: as shown in fig. 1, a traffic data prediction method based on an ultra-long time series includes the following steps:
(1) the method for carrying out the position marking processing on the historical traffic data comprises the following steps:
(1.1) determining the corresponding relation between a single data position in the traffic data super-long time sequence and a data space formed by the super-long time sequence;
(1.2) marking the position of the traffic data in the traffic data super-long time sequence according to the corresponding relation; the ultra-long time sequence refers to a time sequence spanning multiple divided time periods;
the traffic data includes data collected by traffic equipment, such as traffic flow, traffic speed, and the like, and data converted according to the data collected by the traffic equipment, such as traffic state, traffic index, and the like.
The ultra-long time series refers to a time series spanning multiple divided periods, such as: the 4 divided time periods are respectively 1 minute, 1 hour, 1 day and 1 week, x [ K, M, N, L ] represents traffic data of the nth hour and the ith minute on the mth day of the kth week, K, M, N and L are integers, K is an optional range [1-K ], M is an optional range [1-M ],1< M <8, N is an optional range [1-N ],1< N <25, L is an optional range [1-L ], and 1< L < 61; due to different division time intervals, the ultra-long time sequence forms a data space [1-K,1-M,1-N,1-L ], [ K, M, N, L ] corresponding to a certain data space position. Similarly, the time series spanning 3 divided periods also form a lengthy time series.
The divided time interval is determined according to the periodic phenomenon presented by the traffic data, a plurality of groups of traffic data time sequences with the time window as a unit are selected, the average similarity is calculated, and when the average similarity meets a set threshold value, the time window can be used as the divided time interval. In this embodiment, a set of historical traffic data sets is obtained: { (x)1,t1),(x2,t2),…(xj,tj),…,(x300,t300)},t2-t11 hour, the time window TK is 12 hours, and a plurality of groups of traffic data time sequences taking TK as a unit are taken, taking 3 groups as an example, xTK1＝{x1,x2,…,x12},xTK2＝{x25,x26,…,x37},xTK3＝{x13,x14,…,x24And the selection can be carried out sequentially at 12-hour intervals or randomly.
The similarity can be calculated by cosine similarity, Pearson correlation coefficient, Jaccard similarity coefficient and the like. S<xTK1,xTK2>Is a time sequence xTK1And xTK2The similarity of (c). Average similarity S is calculatedTK＝(S<xTK1,xTK2>+S<xTK1,xTK3>+S<xTK2,xTK3>) A/3; the average similarity satisfies a set threshold, and a 12-hour time window may be used as the divided period.
Similarly, the time window TK is 24 hours, the average similarity satisfies a set threshold, and the 24 hour time window may be used as the divided period.
The method may use a regression model, user settings, a neural network model, and the like to determine a correspondence between a single data position in the traffic data super-long time sequence and a data space formed by the super-long time sequence, in this embodiment, the regression model is preferably used, and the formula is as follows:
Q＝∑(y-∑(wpxp+bp))＝min
wherein y is traffic data at a single data position, p represents a data space position, and w and b which minimize the Q function are solved as the corresponding relation between the single data position and the data space. And marking the position of the traffic data in the traffic data super-long time sequence according to the corresponding relation, acquiring w and b by adopting a regression model, and dividing different levels according to the size of w to be used as the position mark.
(2) And selecting historical traffic data with position marks as input data to predict the traffic data. The traffic data prediction may be performed by using a neural network model method, a moving average method, an exponential smoothing method, an AR model method, or the like.
In the present embodiment, a neural network model method is used below to predict traffic data; specifically, an LSTM long-time neural network is adopted for traffic data prediction, and the method specifically comprises the following steps:
(i) the traffic flow data is subjected to weaker perception analysis through an LSTM network, and the characteristic enhancement is performed on the high-influence flow value far away from the predicted value; performing traffic flow prediction by setting an input driving sequence, and establishing a linear regression model for a predicted value and a remote traffic flow value; after training, screening out the flow value of which the position mark meets the requirement, and marking the position of the remote flow value with large weight obtained by screening;
the long and short time neural network (LSTM) is widely applied to prediction of time series, and is a special Recurrent Neural Network (RNN), compared with the RNN, the LSTM is additionally provided with an input gate, an output gate and a forgetting gate, and the gates enable the LSTM to have the function of long-term memory. In traffic flow data prediction, predicted traffic flow data is closely related to flow data of the first few time stamps and also related to flow data near the same time stamp in the first few days of working days, however, due to the existence of LSTM error accumulation, LSTM has weak data perception on an ultra-long sequence, which makes LSTM have weak perception capability on prediction data of the first few days with certain influence in traffic flow, thereby leading to suboptimal analysis and prediction.
The scheme comprises two main steps, firstly, a linear regression model is established for far-distance traffic flow values and predicted values to sense the influence of the far-distance traffic flow values on the predicted values, then, the traffic flow values with position marks meeting requirements are screened out, and the screened traffic flow values with the position marks are added into an LSTM input layer.
It is assumed here that the driving sequence of the input is
X＝(x1，11，x1，1，2，...，xkk，nn，mm，ykk，nn+1，1，ykk，nn+1，2，...，ykk，nn+1，t-1，)，
For the convenience of distinguishing, data of the predicted day is represented by y, and the flow prediction problem is to establish and predict the flow y through a driving sequence Xkk，nn+1，tTo (3) is performed.
Relatively speaking, the flow value near the same time interval n days before has a higher influence on the predicted value, and the farther the flow value is from the time interval t, the weaker the influence is, so in order to reduce the redundancy of the data amount, when performing regression, data near the time interval t every day is selected, and the regression model is as follows:
whereind is the step size of the image,and b are the weight and the offset of the regression model respectively, and after the training is finished, the positions of the flow values with large weights are marked.
In this example, regression was used for weight screening to predict the kth Tuesday 12: traffic data of 00. Setting the step length d as 1, and after the training is finished, the internal corresponding weight matrix is as follows:
first week
Second week
When training is completed, each position will have a corresponding weight, which is represented by < weight, tag > to indicate whether to mark or not, and weight represents the corresponding weight, and the above example marks positions with weights greater than 0.13, and these marked position data represent high correlation positions.
(ii) Adding the traffic flow value with the position mark and the weight into an LSTM network input layer, carrying out back propagation training of LSTM +, firstly calculating the output value of each neuron in a forward direction, secondly calculating the error term value of each neuron in a reverse direction, and finally calculating the gradient of each weight according to the corresponding error term and updating the weight value;
during the training of the LSTM, these marked weighted traffic data are added to the LSTM network, and the states of the input gate, the forgetting gate, the output gate, and the LSTM cell in the LSTM network are as follows:
an input gate:
forget the door:
cell state:
an output gate:
where I represents the length of the input current sequence, C represents the number of LSTM cells, P represents the number of labeled high impact traffic values, l represents the input status,indicating the state of the forgetting gate, o indicating the state of the output gate, f and g being activation functions,represents the c-th cell state value at time t, tgpThe marked traffic flow value representing the pth weighted, the final output is:
(iii) and predicting the traffic flow in a future preset period by using the trained LSTM network model.
Example 2: a traffic data prediction method based on an ultra-long time sequence comprises the following steps:
(1) the method for carrying out the position marking processing on the historical traffic data comprises the following steps:
(1.1) determining the corresponding relation between a single data position in the traffic data super-long time sequence and a data space formed by the super-long time sequence;
(1.2) marking the position of the traffic data in the traffic data super-long time sequence according to the corresponding relation; the ultra-long time sequence refers to a time sequence spanning multiple divided time periods;
the traffic data includes data collected by traffic equipment, such as traffic flow, traffic speed, and the like, and data converted according to the data collected by the traffic equipment, such as traffic state, traffic index, and the like.
The ultra-long time series refers to a time series spanning multiple divided periods, such as: the 4 divided time periods are respectively 1 minute, 1 hour, 1 day and 1 week, x [ K, M, N, L ] represents traffic data of the nth hour and the ith minute on the mth day of the kth week, K, M, N and L are integers, K is an optional range [1-K ], M is an optional range [1-M ],1< M <8, N is an optional range [1-N ],1< N <25, L is an optional range [1-L ], and 1< L < 61; due to different division time intervals, the ultra-long time sequence forms a data space [1-K,1-M,1-N,1-L ], [ K, M, N, L ] corresponding to a certain data space position. Similarly, the time series spanning 3 divided periods also form a lengthy time series.
The divided time interval is determined according to the periodic phenomenon presented by the traffic data, a plurality of groups of traffic data time sequences with the time window as a unit are selected, the average similarity is calculated, and when the average similarity meets a set threshold value, the time window can be used as the divided time interval. In this embodiment, a set of historical traffic data sets is obtained: { (x)1,t1),(x2,t2),…(xj,tj),…,(x300,t300)},t2-t11 hour, the time window TK is 12 hours, and a plurality of groups of traffic data time sequences taking TK as a unit are taken, taking 3 groups as an example, xTK1＝{x1,x2,…,x12},xTK2＝{x25,x26,…,x37},xTK3＝{x13,x14,…,x24And the selection can be carried out sequentially at 12-hour intervals or randomly.
The similarity can be calculated by cosine similarity, Pearson correlation coefficient, Jaccard similarity coefficient and the like. S<xTK1,xTK2>Is a time sequence xTK1And xTK2The similarity of (c). Average similarity S is calculatedTK＝(S<xTK1,xTK2>+S<xTK1,xTK3>+S<xTK2,xTK3>) A/3; the average similarity satisfies a set threshold, and a 12-hour time window may be used as the divided period.
Similarly, the time window TK is 24 hours, the average similarity satisfies a set threshold, and the 24 hour time window may be used as the divided period.
The method may use a regression model, user settings, a neural network model, and the like to determine a correspondence between a single data position in the traffic data super-long time sequence and a data space formed by the super-long time sequence, in this embodiment, the regression model is preferably used, and the formula is as follows:
Q＝∑(y-∑(wpxp+bp))＝min
wherein y is traffic data at a single data position, p represents a data space position, and w and b which minimize the Q function are solved as the corresponding relation between the single data position and the data space. And marking the position of the traffic data in the traffic data super-long time sequence according to the corresponding relation, acquiring w and b by adopting a regression model, and dividing different levels according to the size of w to be used as the position mark.
(2) And selecting historical traffic data with position marks as input data to predict the traffic data.
In this embodiment, a moving average method is used to predict traffic data of 8:00 on kth +1 th week, which is specifically as follows:
ultra-long time series of historical traffic data:
week 1
Week 2
x1,1,1Traffic data at 1:00 on 1 week Monday, xkk,nn,mmTraffic data for the mth hour from 1 o integer on the nnth day from the monday on the kk week.
Historical traffic data with position markers is formed<xkk,nn,mm,tgkk,nn,mm>tgkk,nn,mmIs xkk,nn,mmLocation markers for the data.
Predict the traffic data of 1:00 every kth +1 week, select tgkk,nn,mmHistorical traffic data satisfying a threshold { xkk,7,24,xkk,7,23,xkk,7,22,xkk,7,1,xkk,6,24,xkk,6,23,xkk-1,7,24,…}，xkk,7,24,xkk,7,23,xkk,7,22First 3 hours, x, of 1:00 on kk +1 weekkk,7,1,xkk,6,24,xkk,6,23The time x around 1:00 of the first 1 daykk-1,7,24About 1:00 for the first 2 days. The importance of different time positions on data prediction is reflected.
And performing average operation on the R historical traffic data, and sequentially sliding until all the data are processed to finally obtain an average value serving as a predicted value of the traffic data, wherein R is smaller than the time sequence length of the selected historical traffic data.
While the invention has been described in connection with specific embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. A traffic data prediction method based on an ultra-long time sequence is characterized by comprising the following steps:
(1) the method for carrying out the position marking processing on the historical traffic data comprises the following steps:
(1.1) determining the corresponding relation between a single data position in the traffic data super-long time sequence and a data space formed by the super-long time sequence;
(1.2) marking the position of the traffic data in the traffic data super-long time sequence according to the corresponding relation; the ultra-long time sequence refers to a time sequence spanning multiple divided time periods;
(2) and selecting historical traffic data with position marks as input data to predict the traffic data.
2. The traffic data prediction method based on the ultra-long time series according to claim 1, characterized in that: in the step (1.1), any one of a regression model method, a user setting method and a neural network model method can be adopted to determine the corresponding relation between a single data position in the traffic data super-long time sequence and a data space formed by the super-long time sequence; preferably, a regression model method is used, the formula is as follows:
Q＝∑(y-∑(wpxp+bp))＝min
wherein y is traffic data at a single data position, p represents a data space position, and w and b which minimize the Q function are solved as the corresponding relation between the single data position and the data space.
3. The traffic data prediction method based on the ultra-long time series according to claim 1, characterized in that: and (3) marking the position of the traffic data in the traffic data super-long time sequence according to the corresponding relation in the step (1.2), wherein a regression model is adopted to obtain w and b, and different grades are divided according to the size of w and are used as position marks.
4. The traffic data prediction method based on the ultra-long time series according to claim 1, characterized in that: the divided time interval is determined according to the periodic phenomenon presented by the traffic data, a plurality of groups of traffic data time sequences with the time window as a unit are selected, the average similarity is calculated, and when the average similarity meets a set threshold value, the time window can be used as the divided time interval; the similarity can be calculated by cosine similarity, Pearson correlation coefficient and Jaccard similarity coefficient.
5. The traffic data prediction method based on the ultra-long time series according to claim 1, characterized in that: in the step (2), the traffic data is predicted by adopting any one of a neural network model method, a moving average method, an exponential smoothing method and an AR model method.
6. The traffic data prediction method based on the ultra-long time series according to claim 1, characterized in that: the traffic data comprises but is not limited to data collected by traffic equipment and data converted according to the data collected by the traffic equipment; the data collected by the traffic equipment includes, but is not limited to, traffic flow and traffic speed, and the data converted according to the data collected by the traffic equipment includes, but is not limited to, traffic state and traffic index.
7. The traffic data prediction method based on the ultra-long time series according to claim 5, characterized in that: the neural network model method preferably adopts an LSTM long-time neural network to predict traffic data, and specifically comprises the following steps:
(i) performing traffic flow prediction by setting an input driving sequence, and establishing a linear regression model for a predicted value and a remote traffic flow value;
(ii) after training, screening out the flow value of which the position mark meets the requirement, and marking the position of the remote flow value with large weight obtained by screening;
(iii) adding the traffic flow value with the position mark and the weight into an LSTM network input layer, carrying out back propagation training of LSTM +, firstly calculating the output value of each neuron in a forward direction, secondly calculating the error term value of each neuron in a reverse direction, and finally calculating the gradient of each weight according to the corresponding error term and updating the weight value;
(iv) and predicting the traffic flow in a future preset period by using the trained LSTM network model.
8. The method of claim 7, wherein the traffic data is predicted based on the very long time seriesCharacterized in that: in step (i), the input driving sequence is set to X ═ X1，1，1，x1，1，2，…，xkk，nn，mm，ykk，nn+1,1，ykk，nn+1,2，…,ykk，nn+1,t-1Y) where y represents the data of the predicted day and the flow prediction problem is essentially the creation and prediction of flow y by driving sequence Xkk，nn+1，tTo (3) is performed.
9. The traffic data prediction method based on the ultra-long time series according to claim 7, characterized in that: when a linear regression model is established, selecting data near a time interval t every day, wherein the regression model is as follows:
wherein,d is the step size of the image,and b are the weight and bias of the regression model, respectively.
10. The traffic data prediction method based on the ultra-long time series according to claim 7, characterized in that: the LSTM network is a special RNN recurrent neural network, and compared with RNNs, "input gate", "forget gate", "output gate" are added, and LSTM cell states are as follows:
an input gate:
forget the door:
cell state:
an output gate:
where I represents the length of the input current sequence, C represents the number of LSTM cells, P represents the number of labeled high impact traffic values, l represents the input status,indicating the state of the forgetting gate, o indicating the state of the output gate, f and g being activation functions,represents the c-th cell state value at time t, tgpThe marked traffic flow value representing the pth weighted, the final output is:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811532408.8A CN109598935B (en) | 2018-12-14 | 2018-12-14 | Traffic data prediction method based on ultra-long time sequence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811532408.8A CN109598935B (en) | 2018-12-14 | 2018-12-14 | Traffic data prediction method based on ultra-long time sequence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109598935A true CN109598935A (en) | 2019-04-09 |
| CN109598935B CN109598935B (en) | 2020-12-15 |
Family
ID=65962562
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811532408.8A Active CN109598935B (en) | 2018-12-14 | 2018-12-14 | Traffic data prediction method based on ultra-long time sequence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109598935B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110009429A (en) * | 2019-04-10 | 2019-07-12 | 金瓜子科技发展（北京）有限公司 | A kind of method, apparatus and computer equipment of predicted flow rate data |
| CN110837888A (en) * | 2019-11-13 | 2020-02-25 | 大连理工大学 | Traffic missing data completion method based on bidirectional cyclic neural network |
| CN112330442A (en) * | 2020-11-17 | 2021-02-05 | 深圳市欢太科技有限公司 | Modeling method and device based on ultra-long behavior sequence, terminal and storage medium |
| CN112508305A (en) * | 2019-12-29 | 2021-03-16 | 山西大学 | Public place entrance pedestrian flow prediction method based on LSTM |
| CN112766597A (en) * | 2021-01-29 | 2021-05-07 | 中国科学院自动化研究所 | Bus passenger flow prediction method and system |
| CN113112795A (en) * | 2021-04-06 | 2021-07-13 | 中移(上海)信息通信科技有限公司 | Road condition prediction method, device and equipment |
| CN114283590A (en) * | 2021-09-02 | 2022-04-05 | 青岛海信网络科技股份有限公司 | Traffic flow peak prediction method and device and electronic equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101650876A (en) * | 2009-08-26 | 2010-02-17 | 重庆大学 | Method for obtaining average speed of traffic flow of urban road sections |
| CN102034350A (en) * | 2009-09-30 | 2011-04-27 | 北京四通智能交通系统集成有限公司 | Short-time prediction method and system of traffic flow data |
| CN104134351A (en) * | 2014-08-14 | 2014-11-05 | 中国科学院自动化研究所 | Short-term traffic flow predicting method |
| CN105185106A (en) * | 2015-07-13 | 2015-12-23 | 丁宏飞 | Road traffic flow parameter prediction method based on granular computing |
| CN105389980A (en) * | 2015-11-09 | 2016-03-09 | 上海交通大学 | Short-time traffic flow prediction method based on long-time and short-time memory recurrent neural network |
-
2018
- 2018-12-14 CN CN201811532408.8A patent/CN109598935B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101650876A (en) * | 2009-08-26 | 2010-02-17 | 重庆大学 | Method for obtaining average speed of traffic flow of urban road sections |
| CN102034350A (en) * | 2009-09-30 | 2011-04-27 | 北京四通智能交通系统集成有限公司 | Short-time prediction method and system of traffic flow data |
| CN104134351A (en) * | 2014-08-14 | 2014-11-05 | 中国科学院自动化研究所 | Short-term traffic flow predicting method |
| CN105185106A (en) * | 2015-07-13 | 2015-12-23 | 丁宏飞 | Road traffic flow parameter prediction method based on granular computing |
| CN105389980A (en) * | 2015-11-09 | 2016-03-09 | 上海交通大学 | Short-time traffic flow prediction method based on long-time and short-time memory recurrent neural network |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110009429A (en) * | 2019-04-10 | 2019-07-12 | 金瓜子科技发展（北京）有限公司 | A kind of method, apparatus and computer equipment of predicted flow rate data |
| CN110009429B (en) * | 2019-04-10 | 2021-04-16 | 金瓜子科技发展（北京）有限公司 | Method and device for predicting flow data and computer equipment |
| CN110837888A (en) * | 2019-11-13 | 2020-02-25 | 大连理工大学 | Traffic missing data completion method based on bidirectional cyclic neural network |
| CN112508305A (en) * | 2019-12-29 | 2021-03-16 | 山西大学 | Public place entrance pedestrian flow prediction method based on LSTM |
| CN112330442A (en) * | 2020-11-17 | 2021-02-05 | 深圳市欢太科技有限公司 | Modeling method and device based on ultra-long behavior sequence, terminal and storage medium |
| CN112766597A (en) * | 2021-01-29 | 2021-05-07 | 中国科学院自动化研究所 | Bus passenger flow prediction method and system |
| CN112766597B (en) * | 2021-01-29 | 2023-06-27 | 中国科学院自动化研究所 | Bus passenger flow prediction method and system |
| CN113112795A (en) * | 2021-04-06 | 2021-07-13 | 中移(上海)信息通信科技有限公司 | Road condition prediction method, device and equipment |
| CN114283590A (en) * | 2021-09-02 | 2022-04-05 | 青岛海信网络科技股份有限公司 | Traffic flow peak prediction method and device and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109598935B (en) | 2020-12-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109598935B (en) | Traffic data prediction method based on ultra-long time sequence | |
| CN110298501B (en) | Electrical load prediction method based on long-time and short-time memory neural network | |
| CN109615226B (en) | Operation index abnormity monitoring method | |
| CN110555561B (en) | Medium-and-long-term runoff ensemble forecasting method | |
| CN104408203A (en) | Method for predicting path destination of moving object | |
| CN111695722A (en) | Method for predicting short-term passenger flow in holidays of urban rail transit station | |
| CN112990585A (en) | Hen laying rate prediction method based on LSTM-Kalman model | |
| CN113449905A (en) | Traffic jam early warning method based on gated cyclic unit neural network | |
| CN110414715B (en) | Community detection-based passenger flow volume early warning method | |
| CN108415884A (en) | A kind of modal parameters real-time tracing method | |
| CN110705646A (en) | Mobile equipment streaming data identification method based on model dynamic update | |
| CN109272144B (en) | BPNN-based prediction method for NDVI (normalized difference of variance) in northern grassland area of China | |
| CN107331164B (en) | A kind of prediction technique of freeway toll station entrance vehicle number | |
| CN116930609A (en) | Electric energy metering error analysis method based on ResNet-LSTM model | |
| CN109920249B (en) | Multi-model fusion-based expressway ramp running time prediction method | |
| CN116153125A (en) | Space dimension bus arrival time prediction method based on IDA-RNN model | |
| CN115526117A (en) | Step type landslide displacement rate prediction method based on LSO-RF model | |
| CN112532429B (en) | Multivariable QoS prediction method based on position information | |
| Treboux et al. | A predictive data-driven model for traffic-jams forecasting in smart santader city-scale testbed | |
| CN118114888A (en) | Energy consumption tracking analysis method and system based on energy efficiency acquisition | |
| CN107590346B (en) | Downscaling correction model based on spatial multi-correlation solution set algorithm | |
| CN110009132A (en) | A kind of short-term electric load fining prediction technique based on LSTM deep neural network | |
| CN113159224B (en) | Runoff forecasting model construction method and device, electronic equipment and medium | |
| CN116126827A (en) | Method for simultaneously realizing modeling and visual analysis of power production data | |
| CN115495661A (en) | Self-adaptive interest point recommendation method based on long-term and short-term preference of user |
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 | ||
| CP03 | Change of name, title or address |
Address after: 310012 1st floor, building 1, 223 Yile Road, Hangzhou City, Zhejiang Province Patentee after: Yinjiang Technology Co.,Ltd. Address before: 310023 floor 1, building 1, No. 223, Yile Road, Hangzhou, Zhejiang Patentee before: ENJOYOR Co.,Ltd. |
|
| CP03 | Change of name, title or address |