CN113537626A - Neural network combined time sequence prediction method for aggregating information difference - Google Patents
Neural network combined time sequence prediction method for aggregating information difference Download PDFInfo
- Publication number
- CN113537626A CN113537626A CN202110886769.8A CN202110886769A CN113537626A CN 113537626 A CN113537626 A CN 113537626A CN 202110886769 A CN202110886769 A CN 202110886769A CN 113537626 A CN113537626 A CN 113537626A
- Authority
- CN
- China
- Prior art keywords
- matrix
- neural network
- passenger flow
- formula
- subway
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Abstract
The invention discloses a method for predicting a neural network combined with a time sequence based on aggregate information difference, belongs to the field of deep learning and intelligent transportation, and can simultaneously capture time and space characteristics of subway passenger flow data. According to the method, through a gating mechanism in a GateRecurrentUnit model, the change trend of historical information is kept as much as possible while the passenger flow information at the current moment is captured, the time dependence relationship among the passenger flow information is fully considered, and finally, the method is tested on a real traffic network Shanghai subway traffic network, so that the influence of a smoothing filter defined in a Fourier domain of a graph neural network on a prediction result can be effectively counteracted, and the problem of low prediction precision caused by the fact that the prediction result is smooth when peak passenger flow data is predicted is solved.
Description
Technical Field
The invention relates to the field of deep learning and intelligent transportation, in particular to a method for predicting a neural network combined time sequence by aggregating information difference.
Background
Nowadays, the artificial intelligence technology is rapidly developed, meanwhile, with the massive application of the cheap traffic sensor technology, the explosively increased traffic data, the human beings begin to enter the times of traffic big data and intelligent traffic, the intelligent traffic system aims to establish a set of complete traffic information service and traffic management control system, the system can relieve the traffic jam problem, such as ETC (electronic toll collection) popularized recently, the automatic vehicle identification technology is utilized to complete the automatic vehicle toll collection, the passing efficiency of a highway toll gate is accelerated, the traffic jam problem in holidays is effectively relieved, the intelligent traffic system can also monitor and predict the road traffic in real time, and the traffic state of a road network is reasonably planned, so that the passing efficiency of the vehicles is improved, the occurrence of traffic accidents is reduced, and with the rapid development of economy, the acceleration of pace of urban modern construction and the stable increase of population, the problem of urban traffic congestion is more and more serious, subways are used as important components of urban rail transit, play a great role in relieving traffic congestion, and develop rapidly in the world, in an intelligent traffic system, the prediction of subway passenger flow is an important research direction, the passenger flow of future subway stations is reasonably predicted by using historical passenger flow data, reasonable planning suggestions can be provided for subway operation planners, the operation efficiency of a subway network is improved and improved, with the increase of subway lines in cities, the subway line structure is more and more complex, the safety and efficiency of subway operation face severe challenges, the subway passenger flow is one of main factors influencing subway operation scheduling, so the accurate passenger flow prediction is an important means for coping with the challenges, and a manager is helped to make a reasonable operation plan and relieve traffic congestion, the method improves the operation efficiency of the subway, can judge the basic scale of the newly built subway station, the length of the platform, the capability of evacuating people from the platform and the like according to various data of passenger flow prediction, eliminates the dangerous factors of the subway station from the root and avoids various safety problems of the subway station. The accurate passenger flow prediction can also help travelers to plan an optimal travel route, peak passenger flow is effectively avoided, travel efficiency is improved, and travel safety is guaranteed. The accurate traffic flow prediction has important reference value for planning, designing, managing and controlling traffic facilities, for example, by predicting the traffic flow information of road traffic, a manager can effectively relieve the congestion degree of a peak road section by controlling signal lamps or adding traffic polices and the like; reference information is provided for future planned road design, for example, a road should be designed into a plurality of lanes, which part of the road can reduce the number of lanes, which part of the road needs to increase the number of lanes, which roads are one-way roads and the like, so that the newly planned road can be ensured to effectively alleviate or even solve the problem of road congestion to the greatest extent;
the traditional traffic flow prediction method is mainly divided into two major categories, namely a statistical-based method and a traditional machine learning-based method, wherein the statistical-based method comprises ARIMA and variants thereof, VAR and Kalman filtering, the methods have mathematical basis and have strong interpretability, however, the traffic data are highly nonlinear and dynamic, so that the methods do not conform to the assumptions of linearity and stability, and therefore, the method is poor in practical application performance, the traditional machine learning-based method comprises methods such as a support vector machine and K neighbor, the methods can model nonlinear relations and extract complex relations in the traffic data, but in the era background of big data, the methods need to manually extract features in the data, when a data set is huge, a large amount of manpower is consumed, and the accuracy cannot be guaranteed, so that the invention provides a method for predicting a neural network combined with time sequence by aggregating information difference so as to solve the existing problem of time sequence prediction in the neural network combined with the existing method for predicting the information difference There are problems in the art.
Disclosure of Invention
In view of the above problems, an object of the present invention is to provide a method for predicting a time series by combining a neural network with aggregated information differences, which improves the capability of capturing time and space features of traffic data, particularly extracts space features, and in an original graph convolution network model, due to a smoothing filter defined in a fourier domain, when a passenger flow prediction is performed by using the graph convolution network model, the prediction of peak data is smoother than that of real peak data and the features of the peak are missing The influence of the result, in the aspect of obtaining the time characteristic, the invention uses the gate control recursion unit model, through the gate control mechanism, the passenger flow information at the current moment can be obtained, the change trend of the historical passenger flow information can be obtained as much as possible, and the final prediction result is obtained through the fusion of various information. Compared with a recurrent neural network model, the gated recursion unit model selected by the invention has the advantages of simple structure, less parameter quantity and short training time, and finally, the method disclosed by the invention is applied to the Shanghai subway traffic network to successfully predict the passenger flow of each subway station, so that the method has great significance for relieving traffic jam and improving the subway operation efficiency, and provides reference information for travelers to improve the travel efficiency.
In order to realize the purpose of the invention, the invention is realized by the following technical scheme: a method of aggregating neural networks of information differences in conjunction with time series predictions, comprising the steps of:
the method comprises the following steps: firstly, inputting data, inputting an adjacency matrix of a subway network G ═ V, E, then inputting historical passenger flow data X of each station, regarding the passenger flow as the attribute of the station, and expressing the attribute as X ∈ RN×F；
Step two: enhancing the weight of the node per se, and enhancing the weight proportion of the node in the process of extracting the spatial features by increasing alpha times of the unit matrix on the diagonal line of the normalized Laplace matrix after the normalization of the Laplace matrix by a method based on the maximization of the difference of the aggregated information;
step three: extracting spatial features, namely extracting the spatial features through the historical passenger flow data of each station input in the step one and the preprocessed adjacency matrix of the subway network;
step four: reset Gate in Gate Recurrent Unit model is rtThen, the calculation is carried out to reset the door rtReset gate rtMainly for controlling the candidate states ctWhether the value of (d) depends on the previous state ht-1；
Step five: perform calculation update door utThe update Gate in the Gate Recurrent Unit model is mainly used to control the current state htHow much needs to be kept from the previous state ht-1And the required slave candidate state ctHow much information is received;
step six: performing calculation candidate state ctThe reset gate r is obtained through the step fivetBy resetting the gate rtCan calculate the candidate state ctThe information of (a);
step seven: proceed to update the cell state htFinally, the prediction result [ P ] is output through the full connection layert+1,Pt+2,...,Pt+T]And predicting passenger flow data of T time steps in the future by using the historical passenger flow data of the T time steps.
The further improvement lies in that: in the first step, V ═ { V ] in the subway network Gi|i∈[1,N]And E ═ E }ij＝(vi,vj)|i,j∈[1,N]I ≠ j }, ifviAnd vjWhen there is an edge in between eijThe value of E is 1, V represents the set of subway stations, N represents the number of stations, E represents the physical connecting edges between the stations, F represents the characteristic number of the node attributes and XtRepresenting the traffic values for all stations at time t.
The further improvement lies in that: in the first step, the Laplace matrix is obtained through the input calculation of the first step
In the formula
And is
Representing the adjacency matrix A plus the identity matrix INAnd is and
to represent
The degree matrix of (c).
The further improvement lies in that: in the second step, the weight proportion occupied by the nodes in the process of extracting the spatial features is enhanced by increasing alpha times of the identity matrix, and the weight proportion is expressed by using the following formula
Wherein when alpha is 0.7, the final prediction result is optimal, the degree matrix is diagonal matrix, the element value on the diagonal represents the degree of each vertex, and INIs a matrix of the units,
representing the adjacency matrix A plus the identity matrix IN，
To represent
The coefficient alpha ranges from 0 to 1.
The further improvement lies in that: in the third step, the historical passenger flow volume data of each station and the preprocessed adjacency matrix of the subway network extract the spatial characteristics, and the spatial characteristics are expressed by using the following formula
By the formula, when l is 0, HlX, and WlRepresents the l < th > layer learnable parameter, and H(l+1)Representing the output of the l-th layer, and the value is used as the input of the next layer, and the graph neural network of the two layers is represented by the following formula
In the said formula
W in the formula(0)Is a learning parameter matrix, W, of the first layer neural network(1)For the learning parameter matrix of the second layer neural network, X' is the time sequence data with spatial features output by the two-layer neural network model, HlRepresents the output of the first layer of the neural network of the graph, sigma (-) represents the nonlinear activation function,
Representing the adjacency matrix A plus the identity matrix IN、
To represent
And X represents the historical passenger flow data matrix of each station input in the step one.
The further improvement lies in that: in the fourth step, the reset gate r is calculatedtIs calculated by the following formula
rt＝σ(Wr[X′t,ht-1]+br)
In the formula WrAnd brIs the parameter matrix and deviation in the training process, and r represents the calculated rtParameters and deviations of (2), Xt' represents that time series data with a spatial characteristic at the time t is obtained from a graph neural network model.
The further improvement lies in that: in the fifth step, the door u is updated by calculationtIs calculated by the following formula
ut＝σ(Wu[X′t,ht-1]+bu)
In the formula WuAnd buIs the parameter matrix and bias in the training process, where u represents the calculated utParameters and deviations of.
The further improvement lies in that: in the sixth step, the gate r is resettComputing candidate states ctIs calculated by the following formula
ct＝tanh(Wc[X′t,(rt*ht-1)]+bc)
In the formula WcAnd bcIs the parameter matrix and bias in the training process, where c represents the calculation of ctTime of day parameter matrix and bias.
The further improvement lies in that: in the seventh step, the unit state h at the current moment is controlled by the two gantry cranes in the fifth step and the sixth steptAnd is related to the cell state at the previous time, updates the cell state, and calculates by the following formula
ht＝ut*ht-1+(1-ut)*ct
U in the formulatAnd ctIs the result calculated by the procedure of step six and step seven.
The invention has the beneficial effects that: according to the method, the time characteristics of traffic data and the spatial characteristics of the traffic data can be utilized by combining the graph convolution network model for enhancing the self-node weight and the time sequence prediction model; for the extraction of the spatial features, the method optimizes the graph neural network model by utilizing the method of maximizing the difference of the aggregated information, so that the extraction effect of the spatial features is better; meanwhile, the method is suitable for predicting the passenger flow of the subway station, and the comparison test is carried out on the real world network by using the scheme of the invention and other methods, and the result shows that the scheme of the invention is superior to other comparison methods, so that the method has higher accuracy and can effectively predict the passenger flow in a period of time in the future.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a detailed illustration of the present invention.
Fig. 3 shows the effect of the prediction performed on the data of the Shanghai subway network according to the present invention.
Fig. 4 shows the real network data set size.
FIG. 5 is a comparison of passenger flow prediction effects of multiple methods on a real traffic network
Detailed Description
In order to further understand the present invention, the following detailed description will be made with reference to the following examples, which are only used for explaining the present invention and are not to be construed as limiting the scope of the present invention.
Example one
According to fig. 1-3, the present embodiment provides a method for aggregating information-differentiated neural networks with time series prediction, which includes the following steps:
the method comprises the following steps: firstly, inputting data, inputting an adjacency matrix of a subway network G ═ V, E, then inputting historical passenger flow data X of each station, regarding the passenger flow as the attribute of the station, and expressing the attribute as X ∈ RN×F；
Step two: enhancing the weight of the node per se, and enhancing the weight proportion of the node in the process of extracting the spatial features by increasing alpha times of the unit matrix on the diagonal line of the normalized Laplace matrix after the normalization of the Laplace matrix by a method based on the maximization of the difference of the aggregated information;
step three: extracting spatial features, namely extracting the spatial features through the historical passenger flow data of each station input in the step one and the preprocessed adjacency matrix of the subway network;
step four: reset Gate in Gate Recurrent Unit model is rtThen, the calculation is carried out to reset the door rtReset gate rtMainly for controlling the candidate states ctWhether the value of (d) depends on the previous state ht-1；
Step five: perform calculation update door utThe update Gate in the Gate Recurrent Unit model is mainly used to control the current state htHow much needs to be kept from the previous state ht-1And the required slave candidate state ctHow much information is received;
step six: performing calculation candidate state ctThe reset gate r is obtained through the step fivetBy resetting the gate rtCan calculate the candidate state ctThe information of (a);
step seven: proceed to update the cell state htFinally, the prediction result [ P ] is output through the full connection layert+1,Pt+2,...,Pt+T]And predicting passenger flow data of T time steps in the future by using the historical passenger flow data of the T time steps.
In the first step, V ═ V in the subway network Gi|i∈[1,N]And E ═ E }ij＝(vi,vj)|i,j∈[1,N]I ≠ j }, where if v isiAnd vjWhen there is an edge in between eijThe value of E is 1, V represents the set of subway stations, N represents the number of stations, E represents the physical connecting edges between the stations, F represents the characteristic number of the node attributes and XtIndicating the traffic of all the stations at time tThe value is obtained.
In the first step, the Laplace matrix is obtained through the input calculation of the first step
In the formula
And is
Representing the adjacency matrix A plus the identity matrix INAnd is and
to represent
The degree matrix of (c).
In the second step, the weight proportion occupied by the nodes in the process of extracting the spatial features is enhanced by increasing alpha times of the identity matrix, and the weight proportion is expressed by using the following formula
In the third step, the historical passenger flow volume data of each station and the preprocessed adjacency matrix of the subway network are used for extracting the spatial characteristics, and the spatial characteristics are expressed by using the following formula
By the formula, when l is 0, HlX, and WlRepresents the l < th > layer learnable parameter, and H(l+1)Representing the output of the l-th layer, and the value is used as the input of the next layer, and the graph neural network of the two layers is represented by the following formula
In the formula
W in the formula(0)Is a learning parameter matrix, W, of the first layer neural network(1)For the learning parameter matrix of the second layer neural network, X' is the time sequence data with spatial features output by the two-layer neural network model, HlRepresents the output of the first layer of the neural network of the graph, sigma (-) represents the nonlinear activation function,
Representing the adjacency matrix A plus the identity matrix IN、
To represent
And X represents the historical passenger flow data matrix of each station input in the step one.
In step four, the reset gate r is calculatedtIs calculated by the following formula
rt＝σ(Wr[X′t,ht-1]+br)
In the formula WrAnd brIs the parameter matrix and deviation in the training process, and r represents the calculated rtParameters and deviations of (2), Xt' represents that time series data with a spatial characteristic at the time t is obtained from a graph neural network model.
In the fifth step, the door u is calculated and updatedtIs calculated by the following formula
ut＝σ(Wu[X′t,ht-1]+bu)
In the formula WuAnd buIs the parameter matrix and bias in the training process, where u represents the calculated utParameters and deviations of.
In step six, the gate r is resettComputing candidate states ctIs calculated by the following formula
ct＝tanh(Wc[X′t,(rt*ht-1)]+bc)
In the formula WcAnd bcIs the parameter matrix and bias in the training process, where c represents the calculation of ctTime of day parameter matrix and bias.
In the seventh step, the unit state h at the current moment is controlled by the two gantry cranes in the fifth step and the sixth steptAnd is related to the cell state at the previous time, updates the cell state, and calculates by the following formula
ht＝ut*ht-1+(1-ut)*ct
U in the formulatAnd ctIs the result calculated by the procedure of step six and step seven.
When the EST-GCN model is used, namely a graph convolution network and a gating recursion unit for enhancing self-node weight, firstly, subway network topological structures at different moments and passenger flow data of each station are extracted to serve as input data of the ES-GCN model, then based on an aggregation information difference maximization method, the model solves the influence of a smoothing filter defined in a Fourier domain in an original graph convolution network on a prediction result, the smoothing filter can cause the model to be smoother compared with real peak data when the model predicts the peak data, the prediction result lacks the characteristic of the peak, and the spatial characteristic of traffic data can be better extracted by solving the problem. The result obtained by enhancing the graph convolution network model of the self-node weight is time sequence data with space characteristic information, then in order to obtain the time characteristic, the obtained time sequence data with the space characteristic is input into a gate control recursion unit model, then the dynamic change is obtained by transferring information between units, the gate control recursion unit model resets a gate r through two gate mechanismstMainly for controlling candidate states ctWhether the value of (d) depends on the previous state ht-1Update the door utMainly used for controlling the current state htHow much needs to be kept from the previous state ht-1And need to be driven from candidate state ctHow much information is received. Through the two door mechanisms, the door control recursion unit model can not only acquire the passenger flow information at the current moment, but also keep historical passenger flow information as much as possible, and finally obtain a final prediction result from the full-connection layer.
Example two
According to the illustration in fig. 3, the embodiment provides a method for predicting a neural network by combining aggregated information differences with a time series, regions a to d in fig. 3 show comparison results of real values and predicted values of passenger flows of various stations at different time granularities from eight morning, and as can be seen from visualization results, there are a few differences between actual passenger flows and predicted passenger flows of some stations, but more than 80% of the real passenger flows of the stations at different time granularities are basically fit with the predicted passenger flows; the areas e to h in FIG. 3 show v3The site starts from different time, the comparison result of the actual value and the predicted value of the passenger flow under different time granularities shows that under the time granularity of 15min, the site v3The actual value and the predicted value of the passenger flow rate have a large difference at the beginning, but the difference is smaller and smaller along with the time, the training data is increased, the prediction capability of the model is improved, and on the whole, the influence of a smoothing filter in a Fourier domain of a graph convolution neural network on a spatial feature extraction result is counteracted by increasing the weight value of a node on a diagonal line of an adjacent matrix, so that the model has the same excellent prediction result on the peak value, the beginning and the end of the peak time can be detected, and the prediction result has a similar mode with the change trend of the passenger flow rate of an actual rail transit station.
According to fig. 4, the present embodiment provides a method for predicting a neural network by aggregating information differences in combination with a time series, which is to use each item of data of real subway network data for testing, wherein the number of subway stations is 285 stations in total, the time span of historical passenger flow data is 2015-4-month 6-2015-4-month 26-2015-20 days in total, each data interval is 15min, and 80% of the data is used as a training set, and the other 20% of the data is used as a test set to verify the performance of a model.
According to fig. 5, the present embodiment provides a method for predicting passenger flow by combining a neural network with time series, which shows how many comparison methods compare passenger flow prediction effects on multiple real networks, and evaluates the passenger flow prediction effect accurately by means of average absolute error, root mean square error, decision coefficient, interpretation of variance score, and the bold term in each row indicates that the method has the best effect on the data set of the corresponding row, so that it is seen that the EST-GCN method provided by the present invention has better effect on passenger flow prediction than other methods.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (9)
1. A method for aggregating neural networks of information differences in conjunction with time series predictions, comprising the steps of:
the method comprises the following steps: firstly, inputting data, inputting an adjacency matrix of a subway network G ═ V, E, then inputting historical passenger flow data X of each station, regarding the passenger flow as the attribute of the station, and expressing the attribute as X ∈ RN×F；
Step two: enhancing the weight of the node per se, and enhancing the weight proportion of the node in the process of extracting the spatial features by increasing alpha times of the unit matrix on the diagonal line of the normalized Laplace matrix after the normalization of the Laplace matrix by a method based on the maximization of the difference of the aggregated information;
step three: extracting spatial features, namely extracting the spatial features through the historical passenger flow data of each station input in the step one and the preprocessed adjacency matrix of the subway network;
step four: reset Gate in Gate Recurrent Unit model is rtThen, the calculation is carried out to reset the door rtReset gate rtMainly for controlling the candidate states ctWhether the value of (d) depends on the previous state ht-1；
Step five: perform calculation update door utThe update Gate in the Gate Recurrent Unit model is mainly used to control the current state htHow much needs to be kept from the previous state ht-1And the required slave candidate state ctHow much information is received;
step six: performing calculation candidate state ctThe reset gate r is obtained through the step fivetBy resetting the gate rtCan calculate the candidate state ctThe information of (a);
step seven: proceed to update the cell state htFinally, the prediction result [ P ] is output through the full connection layert+1,Pt+2,...,Pt+T]And predicting passenger flow data of T time steps in the future by using the historical passenger flow data of the T time steps.
2. The method of claim 1, wherein the neural network of aggregate information variation incorporates time series prediction, and further comprising: in the first step, V ═ { V ] in the subway network Gi|i∈[1,N]And E ═ E }ij＝(vi,vj)|i,j∈[1,N]I ≠ j }, where if v isiAnd vjWhen there is an edge in between eijThe value of E is 1, V represents the set of subway stations, N represents the number of stations, E represents the physical connecting edges between the stations, F represents the characteristic number of the node attributes and XtRepresenting the traffic values for all stations at time t.
3. The method of claim 1, wherein the neural network of aggregate information variation incorporates time series prediction, and further comprising: in the first step, the Laplace matrix is obtained through the input calculation of the first step
In the formula
And is
Representing the adjacency matrix A plus the identity matrix INAnd is and
to represent
The degree matrix of (c).
4. The method of claim 1, wherein the neural network of aggregate information variation incorporates time series prediction, and further comprising: in the second step, the weight proportion occupied by the nodes in the process of extracting the spatial features is enhanced by increasing alpha times of the identity matrix, and the weight proportion is expressed by using the following formula
Wherein when alpha is 0.7, the final prediction result is optimal, the degree matrix is diagonal matrix, the element value on the diagonal represents the degree of each vertex, and INIs a matrix of the units,
representing the adjacency matrix A plus the identity matrix IN，
To represent
The coefficient alpha ranges from 0 to 1.
5. The method of claim 1, wherein the neural network of aggregate information variation incorporates time series prediction, and further comprising: in the third step, the historical passenger flow volume data of each station and the preprocessed adjacency matrix of the subway network extract the spatial characteristics, and the spatial characteristics are expressed by using the following formula
By the formula, when l is 0, HlX, and WlRepresents the l < th > layer learnable parameter, and H(l+1)Representing the output of the l-th layer, and the value is used as the input of the next layer, and the graph neural network of the two layers is represented by the following formula
In the said formula
W in the formula(0)Is a learning parameter matrix, W, of the first layer neural network(1)For the learning parameter matrix of the second layer neural network, X' is the time sequence data with spatial features output by the two-layer neural network model, HlRepresents the output of the first layer of the neural network of the graph, sigma (-) represents the nonlinear activation function,
Representing the adjacency matrix A plus the identity matrix IN,
To represent
And X represents the historical passenger flow data matrix of each station input in the step one.
6. The method of claim 1, wherein the neural network of aggregate information variation incorporates time series prediction, and further comprising: in the fourth step, the reset gate r is calculatedtIs calculated by the following formula
rt＝σ(Wr[X′t,ht-1]+br)
In the formula WrAnd brIs the parameter matrix and deviation in the training process, and r represents the calculated rtParameters and deviations of (2), Xt' represents that time series data with a spatial characteristic at the time t is obtained from a graph neural network model.
7. The method of claim 1, wherein the neural network of aggregate information variation incorporates time series prediction, and further comprising: in the fifth step, the door u is updated by calculationtIs calculated by the following formula
ut＝σ(Wu[X′t,ht-1]+bu)
In the formula WuAnd buIs the parameter matrix and bias in the training process, where u represents the calculated utParameters and deviations of.
8. The method of claim 1, wherein the neural network of aggregate information variation incorporates time series prediction, and further comprising: in the sixth step, the gate r is resettComputing candidate states ctIs calculated by the following formula
ct＝tanh(Wc[X′t,(rt*ht-1)]+bc)
In the formula WcAnd bcIs the parameter matrix and bias in the training process, where c represents the calculation of ctTime of day parameter matrix and bias.
9. The method of claim 1, wherein the neural network of aggregate information variation incorporates time series prediction, and further comprising: in the seventh step, the unit state h at the current moment is controlled by the two gantry cranes in the fifth step and the sixth steptAnd is related to the cell state at the previous time, updates the cell state, and calculates by the following formula
ht＝ut*ht-1+(1-ut)*ct
U in the formulatAnd ctIs the result calculated by the procedure of step six and step seven.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110886769.8A CN113537626B (en) | 2021-08-03 | 2021-08-03 | Method for predicting neural network combined time sequence by aggregating information difference |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110886769.8A CN113537626B (en) | 2021-08-03 | 2021-08-03 | Method for predicting neural network combined time sequence by aggregating information difference |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113537626A true CN113537626A (en) | 2021-10-22 |
| CN113537626B CN113537626B (en) | 2023-05-05 |
Family
ID=78090332
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110886769.8A Active CN113537626B (en) | 2021-08-03 | 2021-08-03 | Method for predicting neural network combined time sequence by aggregating information difference |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113537626B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114330868A (en) * | 2021-12-27 | 2022-04-12 | 西北工业大学 | Passenger flow prediction method based on self-attention personalized enhanced graph convolution network |
| CN114722334A (en) * | 2022-04-11 | 2022-07-08 | 哈尔滨工程大学 | STFT-based online identification method for gas injection time characteristics of high-pressure natural gas direct injection engine |
| CN114973653A (en) * | 2022-04-27 | 2022-08-30 | 中国计量大学 | Traffic flow prediction method based on space-time graph convolution network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170214943A1 (en) * | 2016-01-22 | 2017-07-27 | Mitsubishi Electric Research Laboratories, Inc. | Point Cloud Compression using Prediction and Shape-Adaptive Transforms |
| CN111126704A (en) * | 2019-12-25 | 2020-05-08 | 河海大学常州校区 | Multi-region precipitation prediction model construction method based on multi-graph convolution and memory network |
| CN112350899A (en) * | 2021-01-07 | 2021-02-09 | 南京信息工程大学 | Network flow prediction method based on graph convolution network fusion multi-feature input |
| CN112418547A (en) * | 2020-12-03 | 2021-02-26 | 北京工业大学 | Bus stop passenger flow prediction method based on GCN-LSTM combined model |
| CN112819213A (en) * | 2021-01-22 | 2021-05-18 | 华南理工大学 | Expressway freight volume prediction method and system based on deep learning network |
-
2021
- 2021-08-03 CN CN202110886769.8A patent/CN113537626B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170214943A1 (en) * | 2016-01-22 | 2017-07-27 | Mitsubishi Electric Research Laboratories, Inc. | Point Cloud Compression using Prediction and Shape-Adaptive Transforms |
| CN111126704A (en) * | 2019-12-25 | 2020-05-08 | 河海大学常州校区 | Multi-region precipitation prediction model construction method based on multi-graph convolution and memory network |
| CN112418547A (en) * | 2020-12-03 | 2021-02-26 | 北京工业大学 | Bus stop passenger flow prediction method based on GCN-LSTM combined model |
| CN112350899A (en) * | 2021-01-07 | 2021-02-09 | 南京信息工程大学 | Network flow prediction method based on graph convolution network fusion multi-feature input |
| CN112819213A (en) * | 2021-01-22 | 2021-05-18 | 华南理工大学 | Expressway freight volume prediction method and system based on deep learning network |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114330868A (en) * | 2021-12-27 | 2022-04-12 | 西北工业大学 | Passenger flow prediction method based on self-attention personalized enhanced graph convolution network |
| CN114722334A (en) * | 2022-04-11 | 2022-07-08 | 哈尔滨工程大学 | STFT-based online identification method for gas injection time characteristics of high-pressure natural gas direct injection engine |
| CN114973653A (en) * | 2022-04-27 | 2022-08-30 | 中国计量大学 | Traffic flow prediction method based on space-time graph convolution network |
| CN114973653B (en) * | 2022-04-27 | 2023-12-19 | 中国计量大学 | Traffic flow prediction method based on space-time diagram convolutional network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113537626B (en) | 2023-05-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113537626A (en) | Neural network combined time sequence prediction method for aggregating information difference | |
| WO2023056696A1 (en) | Urban rail transit short-term passenger flow forecasting method based on recurrent neural network | |
| US20210407284A1 (en) | Vehicle traffic flow prediction method with missing data | |
| CN110276950B (en) | Urban traffic trip chain reconstruction method based on bayonet video data | |
| CN106228808B (en) | City expressway travel time prediction method based on Floating Car space-time grid data | |
| CN109215350B (en) | Short-term traffic state prediction method based on RFID electronic license plate data | |
| CN114330868A (en) | Passenger flow prediction method based on self-attention personalized enhanced graph convolution network | |
| CN110866649A (en) | Method and system for predicting short-term subway passenger flow and electronic equipment | |
| CN112489426B (en) | Urban traffic flow space-time prediction scheme based on graph convolution neural network | |
| CN109410577A (en) | Adaptive traffic control sub-area division method based on Spatial Data Mining | |
| CN110299011A (en) | A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data | |
| CN110390349A (en) | Bus passenger flow volume based on XGBoost model predicts modeling method | |
| CN110111576A (en) | A kind of urban transportation elastic index and its implementation based on space-time congestion group | |
| CN108597227A (en) | Road traffic flow forecasting method under freeway toll station | |
| CN107656987A (en) | A kind of subway station function method for digging based on LDA models | |
| CN110310479A (en) | A kind of Forecast of Urban Traffic Flow forecasting system and method | |
| CN107729938A (en) | Rail station classification method based on bus connection radiation zone characteristics | |
| CN105679025A (en) | Urban trunk road travel time estimation method based on variable weight mixed distribution | |
| CN106355905A (en) | Control method for overhead signal based on checkpoint data | |
| CN104318757A (en) | Operation time forecasting method of buses on road segments of bus lanes | |
| CN108876064A (en) | Urban mass transit network key station recognition methods based on the volume of the flow of passengers | |
| CN113657645B (en) | Space-time connection enhanced 3DCNN traffic prediction method based on attention mechanism | |
| CN109299825A (en) | A kind of prediction technique and forecasting system based on the real-time passenger flow of rail traffic | |
| CN113409578A (en) | Traffic network health portrait method and system based on fuzzy clustering | |
| CN109447352B (en) | Bus trip OD matrix reverse-thrust combination method |
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 |