CN111339978A - Method for recognizing traffic index time series mode by using convolutional neural network model - Google Patents
Method for recognizing traffic index time series mode by using convolutional neural network model Download PDFInfo
- Publication number
- CN111339978A CN111339978A CN202010140374.9A CN202010140374A CN111339978A CN 111339978 A CN111339978 A CN 111339978A CN 202010140374 A CN202010140374 A CN 202010140374A CN 111339978 A CN111339978 A CN 111339978A
- Authority
- CN
- China
- Prior art keywords
- time series
- traffic index
- index time
- neural network
- convolutional neural
- 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.)
- Pending
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 91
- 238000000034 method Methods 0.000 title claims abstract description 54
- 239000013598 vector Substances 0.000 claims abstract description 26
- 238000000605 extraction Methods 0.000 claims abstract description 23
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000011176 pooling Methods 0.000 claims description 32
- 230000006870 function Effects 0.000 claims description 30
- 239000011159 matrix material Substances 0.000 claims description 23
- 230000015654 memory Effects 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 4
- 238000011160 research Methods 0.000 abstract description 13
- 230000008859 change Effects 0.000 abstract description 5
- 238000004422 calculation algorithm Methods 0.000 description 36
- 238000003909 pattern recognition Methods 0.000 description 28
- 238000010586 diagram Methods 0.000 description 16
- 238000012549 training Methods 0.000 description 15
- 230000008569 process Effects 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000012706 support-vector machine Methods 0.000 description 5
- 238000003064 k means clustering Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 230000003203 everyday effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000003442 weekly effect Effects 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 2
- 230000001747 exhibiting effect Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000012567 pattern recognition method Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007621 cluster analysis Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 230000002354 daily effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G06Q50/40—
Abstract
The invention discloses a method, a device, electronic equipment and a storage medium for identifying a traffic index time series mode by using a convolutional neural network model. The method comprises the following steps: acquiring traffic index time series data; performing feature extraction on the traffic index time series data by using a convolutional neural network model to obtain a feature vector; and processing the feature vectors by using a classification module to obtain a mode classification result of the traffic index time series data. The invention discovers the time sequence characteristics of the traffic index time sequence data through the convolutional neural network model, classifies the time sequence characteristics by utilizing the classification module, unifies the complex abstract time sequences into the modes, and is more accurate and effective in identifying the time sequence data modes, thereby being suitable for the research on the correlation between data presenting time stage change and the mode discovery.
Description
Technical Field
The invention relates to the field of software, in particular to a method, a device, electronic equipment and a storage medium for recognizing a traffic index time series pattern by using a convolutional neural network model.
Background
With the rapid development of urbanization, the problem of urban traffic congestion is increasingly prominent. In order to solve the problem of complex and variable urban traffic jam, a traffic management department sets out a series of policies and regulations, so that the jam condition of urban traffic is relieved to a certain extent, but the problem still exists. Based on the above, scientific research personnel develop a great deal of research in the fields of sociology, statistics, internet, geographic information and the like so as to accurately find out laws and characteristics of residents in the trip and provide reference and guidance for traffic management departments. The traffic index is an important index for researching the running state of urban traffic, reflects the quantitative result of urban congestion conditions, and has a certain rule in the time dimension. Therefore, if the potential characteristics of the travel of the residents are obtained from the historical traffic indexes through time series pattern recognition, the categories of the residents are distinguished, basic data can be provided for the research and prediction of the traffic running state, and the method has important value for the relief of urban traffic jam.
The pattern recognition is that a computer automatically recognizes the category of an object according to a certain algorithm, and the objects with certain similarity are regarded as a whole through the quantity, the size or the logical relationship of the object representations, so that the rules of the objects are mined and extracted. The pattern recognition method is classified into supervised learning and unsupervised learning, and the basis for generally distinguishing the two methods is to judge whether the classification result is known.
A common algorithm for unsupervised classification is a clustering algorithm, clustering analysis is taken as a common algorithm in the traditional machine learning algorithm, and the high efficiency and simplicity of the algorithm enable the algorithm to be widely applied to pattern recognition. In cluster analysis, there are generally three steps: feature extraction, object feature similarity calculation and grouping according to similarity. Macqueen proposed a K-Means algorithm for the first time in 1967, and the core idea of the algorithm is that Euclidean distance is used as an index for measuring similarity of objects, and the Euclidean distance is inversely proportional to the similarity. Loss function of algorithm:
in the formula, data { xiThe clustering is performed to k classes, and each class after the clustering is tiCluster center is μi. K-Means is to find the best K and tiSo that the loss function is minimized. However, due to the limitation of the K-Means algorithm, the K-Means algorithm hardly embodies the advantages of the K-Means algorithm on part of clustering problems, and particularly hardly embodies the time sequence of time sequence data.
With the increase of the complexity of the practical problem, the traditional unsupervised learning method cannot meet the requirements of the practical problem at present. While the advent of neural networks has raised machine learning to another height. The neural network is formed by connecting a large number of nodes (neurons), each node is an output function, and the nodes are connected by a weight value. Neural networks can be divided into an input layer and an output layer. In deep learning, a Support Vector Machine (SVM) algorithm is a common algorithm for pattern recognition. The SVM algorithm finds a hyperplane in object samples and divides the object samples according to similarity difference, and the core of the algorithm is how to find the hyperplane to maximize the object difference. However, in practical application, the SVM algorithm has some disadvantages, for example, the multi-classification problem is difficult, the performance of the SVM algorithm depends on the selection of kernel functions, the time sequence of time series data is difficult to be embodied, and the artificial construction features are complicated.
At present, time series pattern recognition research has become the research focus of scholars at home and abroad in the field of data mining. Research shows that the time series pattern recognition machine learning method mainly comprises the following steps: distance-based time series pattern recognition and feature-based time series pattern recognition. Distance-based time series pattern recognition, similarity between time series data is generally measured using euclidean distances, such as the k-nearest neighbor (kNN) algorithm. And the feature-based pattern recognition is to study the differential subsections of the time series data according to the differential features of the time series data, and distinguish the categories of the sequences by comparing the subsections with the most significant difference of the time series, for example, the Shapelet algorithm is to find the continuous subsequences which are most representative in the time series.
Through verification, the time series pattern recognition based on the distance and the characteristics can obtain better recognition effect only under certain conditions. Meanwhile, due to the influence of uncertainty factors, certain deformation and distortion phenomena exist in the time sequence data. Feature-based pattern recognition has limited ability to distinguish single Shapelet subsequences, and multi-time-segment Shapelet subsequence extraction is difficult to achieve. Therefore, it is desirable to provide a more accurate and efficient pattern recognition method that is more suitable for traffic index time series data.
Disclosure of Invention
An object of the present invention is to solve at least the above problems and/or disadvantages and to provide at least the advantages described hereinafter.
An object of the present invention is to provide a method for recognizing a traffic index time series pattern using a convolutional neural network model, which is more accurate and efficient for pattern recognition of traffic index time series data, and is suitable for correlation research and pattern discovery between data exhibiting time-phased variation.
It is still another object of the present invention to provide an apparatus for recognizing a traffic index time series pattern using a convolutional neural network model, which is more accurate and efficient for pattern recognition of traffic index time series data, and is suitable for correlation research and pattern discovery between data exhibiting time-phased variation.
To achieve these objects and other advantages in accordance with the purpose of the invention, there is provided a method for recognizing a traffic index time series pattern using a convolutional neural network model, including:
acquiring traffic index time series data;
performing feature extraction on the traffic index time series data by using a convolutional neural network model to obtain a feature vector;
and processing the feature vectors by using a classification module to obtain a mode classification result of the traffic index time series data.
Preferably, in the method for identifying a traffic index time series pattern by using a convolutional neural network model, the extracting features of the traffic index time series data by using the convolutional neural network model to obtain a feature vector includes:
converting the traffic index time series data into an N-dimensional matrix;
and performing feature extraction on the N-dimensional matrix by using the convolutional neural network model to obtain a feature vector.
Preferably, in the method for identifying a traffic index time series pattern by using a convolutional neural network model, the classification module is a SOFTMAX classifier; the convolutional neural network model is formed by combining 2 groups of structures of 1 convolutional layer 5 by 5 and 1 pooling layer in each group in a laminated structure; the activation function of the convolutional neural network model is a ReLU function; the pooling layer adopts maximum pooling.
Preferably, in the method for identifying a traffic index time series pattern by using a convolutional neural network model, the pattern classification result of the traffic index time series data includes a monday pattern, a middle of week pattern, a friday pattern, a saturday pattern and a sunday pattern; the traffic index time series data is traffic index time series data between 5 and 23 points in a natural day.
The invention also provides a device for identifying the traffic index time series mode by utilizing the convolutional neural network model, which comprises the following steps:
the acquisition module is used for acquiring traffic index time series data;
the extraction module is used for extracting the characteristics of the traffic index time series data by using a convolutional neural network model to obtain a characteristic vector;
and the classification module is used for processing the characteristic vector to obtain a mode classification result of the traffic index time series data.
Preferably, in the apparatus for identifying a traffic index time series pattern by using a convolutional neural network model, the extraction module includes:
the conversion sub-module is used for converting the traffic index time sequence data into an N-dimensional matrix;
and the extraction submodule is used for extracting the characteristics of the N-dimensional matrix by using the convolutional neural network model to obtain a characteristic vector.
Preferably, in the apparatus for identifying a traffic index time series pattern by using a convolutional neural network model, the classification module is a SOFTMAX classifier; the convolutional neural network model is formed by combining 2 groups of structures of 1 convolutional layer 5 by 5 and 1 pooling layer in each group in a laminated structure; the activation function of the convolutional neural network model is a ReLU function; the pooling layer adopts maximum pooling.
Preferably, in the apparatus for identifying a traffic index time series pattern using a convolutional neural network model, the pattern classification result of the traffic index time series data includes a monday pattern, a middle of the week pattern, a friday pattern, a saturday pattern, and a sunday pattern; the traffic index time series data is traffic index time series data between 5 and 23 points in a natural day.
The present invention also provides an electronic device comprising: at least one processor, and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the method.
The invention also provides a storage medium on which a computer program is stored which, when executed by a processor, implements the method.
The invention at least comprises the following beneficial effects:
the invention discovers the time sequence characteristics of the traffic index time sequence data through the convolutional neural network model, classifies the time sequence characteristics by utilizing the classification module, unifies the complex abstract time sequences into the modes, and is more accurate and effective in identifying the time sequence data modes, thereby being suitable for the research on the correlation between data presenting time stage change and the mode discovery. Based on the mode recognition result of the traffic index time series data, the traffic travel rules of the traffic index time series data can be extracted according to different modes, so that reference is provided for relevant management departments to make traffic plans.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Drawings
FIG. 1 is a flow chart of one embodiment of a method for identifying traffic index time series patterns using a convolutional neural network model provided in the present invention;
FIG. 2 is a schematic diagram of a convolutional neural network model according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of convolutional layer operation of the method for identifying traffic index time series pattern using convolutional neural network model according to the present invention;
FIG. 4 is a schematic diagram of a convolutional neural network model according to another embodiment of the method for identifying a traffic index time series pattern using the convolutional neural network model;
fig. 5(a) is a schematic diagram of traffic index time series data of a monday pattern when pattern recognition is performed on a training set in an embodiment of the method for recognizing a traffic index time series pattern by using a convolutional neural network model provided in the present invention;
FIG. 5(b) is a schematic diagram of traffic index time series data of a pattern in a week of pattern recognition for a training set in an embodiment of the method for recognizing traffic index time series patterns by using a convolutional neural network model provided in the present invention;
FIG. 5(c) is a schematic diagram of traffic index time series data of Friday pattern when pattern recognition is performed on a training set in an embodiment of the method for recognizing traffic index time series patterns by using a convolutional neural network model provided in the present invention;
FIG. 5(d) is a schematic diagram of traffic index time series data of Saturday pattern when pattern recognition is performed on a training set in an embodiment of the method for recognizing traffic index time series patterns by using a convolutional neural network model provided by the present invention;
FIG. 5(e) is a schematic diagram of traffic index time series data of a weekday pattern when pattern recognition is performed on a training set in an embodiment of the method for recognizing a traffic index time series pattern by using a convolutional neural network model provided by the present invention;
FIG. 6 is a diagram illustrating the recognition result of pattern recognition on the traffic index time series data according to an embodiment of the method for recognizing the traffic index time series pattern by using the convolutional neural network model provided in the present invention;
FIG. 7 is a diagram showing the recognition result of pattern recognition of traffic index time-series data by using the conventional K-means clustering algorithm;
FIG. 8 is a diagram illustrating an embodiment of an apparatus for identifying a traffic index time series pattern using a convolutional neural network model according to the present invention;
fig. 9 is a schematic structural diagram of an embodiment of an electronic device provided in the present invention.
Detailed Description
The present invention is further described in detail below with reference to the attached drawings so that those skilled in the art can implement the invention by referring to the description text.
A flow diagram of one embodiment of a method of the present invention for identifying traffic index time series patterns using a convolutional neural network model is shown in fig. 1. The invention provides a method for identifying a traffic index time series mode by using a convolutional neural network model, which comprises the following steps:
The traffic index is an important index for researching the running state of urban traffic, reflects the quantitative result of urban congestion conditions, and has a certain rule in the time dimension. The traffic index is a value between 0 and 10, and the higher the value is, the more congested the traffic is, and the worse the traffic condition is. The traffic index time series refers to a section of series obtained by arranging traffic indexes along with time change. The time in the time series may be year, quarter, month or any other form of time depending on the time of observation. Assume a set of random variables X ═ X1，X2，……，XnDefine time as T ═ T }1，t2，……，tnThen define Xt＝{X1，X2，……，XtI T ∈ T is a time series within time T.
And 102, performing feature extraction on the traffic index time sequence data by using a convolutional neural network model (CNN) to obtain a feature vector.
And 103, processing the feature vectors by using a classification module to obtain a mode classification result of the traffic index time series data.
The invention utilizes the convolutional neural network model to carry out mode recognition on the traffic index time sequence data, can reduce the influence of human characteristic intervention, and excavates deeper time sequence rules and characteristics in the traffic index time sequence data.
The convolutional neural network algorithm is a neural network aiming at image processing and is realized by back propagation, and the main process comprises the following steps: input layer, convolutional layer, pooling layer, fully-connected layer, and output layer (as shown in fig. 2). In the convolution layer, the convolution core performs traversal calculation on the input layer, and performs feature extraction on a local area on the original input layer. As shown in fig. 3, the convolutional layer designs the algorithm and size of the convolutional kernel according to actual requirements. Equation (2) is an expression of the convolution operation of the input layer and the convolution kernel. The convolutional layer also needs to satisfy the "weight sharing" principle, and scan the entire input layer with one convolutional layer.
In the formula (I), the compound is shown in the specification,
the output result after the convolution operation is obtained;
is the jth convolution of the ith layer;
is a bias parameter; f (x) is an activation function.
The nonlinear transformation of the function is reactivated by the result of convolutional layer operation to solve the linear irreparable problem. Common activation functions include: sigmoid function (equation 3), Tanh function (equation 4), and ReLU function (equation 5). The Sigmoid function can increase the calculated amount, and simultaneously, the conditions of gradient disappearance and gradient explosion can be caused when the function is reversely propagated, so that information loss is caused, and deep network training cannot be completed. The Tanh function is built on top of the Sigmoid function, except for the scale and range. The ReLU function can more effectively reduce the influence of gradient descent and back propagation on training and avoid the problems of gradient explosion and gradient disappearance, and the function not only optimizes the calculation process, but also reduces the calculation amount, and simultaneously reduces the dependency relationship of parameters and relieves the phenomenon of overfitting of the network. Therefore, the present invention preferably uses the ReLU function to perform the non-linear transformation after the convolution process.
The pooling layer is mainly used for further compressing and extracting the feature map after the convolutional layer is processed. The method is divided into average pooling and maximum pooling. The average pooling is to take the average value of the feature points in the neighborhood as the feature expression of the neighborhood; the maximum pooling is the feature expression that the feature point in the domain takes the maximum value as the neighborhood. The error in the convolution process comes from the selection of the neighborhood size and the adjustment of the convolution parameters. Average pooling can reduce errors caused by neighborhood size selection, but cannot reduce errors caused by mean shift due to parameter adjustment, so that more feature information of the input layer cannot be reserved. Therefore, the present invention preferably employs maximal pooling for further feature extraction.
After convolution processing and pooling processing, the feature maps enter a full-connection layer for classification. The full-connection layer integrates the feature maps and outputs a one-dimensional vector, so that the local features are integrated with global information in a higher dimension. After the integration of the full connection layer, the integrated result is classified by utilizing the SOFTMAX function (formula (6)), so that the distribution probability of each class is obtained, and the class with the maximum probability is used as the classification result according to the probability.
In the formula, ajA j-th value representing a vector in the fully-connected layer; a iskRepresenting each bit in the full link layerA value; t represents the number of classification categories set in advance.
The invention discovers the time sequence characteristics of the traffic index time sequence data through the convolutional neural network model, classifies the time sequence characteristics by utilizing the classification module, unifies the complex abstract time sequences into the modes, and is more accurate and effective in identifying the time sequence data modes, thereby being suitable for the research on the correlation between data presenting time stage change and the mode discovery. Furthermore, based on the mode recognition result of the traffic index time series data, the traffic travel rules of the traffic index time series data can be extracted according to different modes, so that reference is provided for making a traffic plan for related management departments.
In one embodiment, in the method for identifying a traffic index time series pattern by using a convolutional neural network model, the extracting features of the traffic index time series data by using the convolutional neural network model to obtain a feature vector includes: converting the traffic index time series data into an N-dimensional matrix; and performing feature extraction on the N-dimensional matrix by using the convolutional neural network model to obtain a feature vector.
Specifically, when converting the traffic index time-series data into an N-dimensional matrix, assuming that the number of numeric values of the traffic index time-series data is M, L numeric values "0" may be supplemented before the first digit of the time-series data (or after the last digit of the time-series data), and M + L ═ N × N may be supplemented, thereby converting the time-series data supplemented with a sufficient number of digits into an N-dimensional matrix. Each row of the matrix is arranged from the first column of the row during the conversion process. For example, when there are 73 time-series data, in the 10 × 10 matrix, the first 27 bits are the value 0, and the 1 st bit of the time-series data is from the 8 th column of the 3 rd row, then the time-series data is arranged in the 3 rd row in sequence, and the 1 st column of the 4 th row corresponds to the 4 th bit of the time-series data, then the time-series data is arranged in the 4 th row in sequence.
In the embodiment, the traffic index time series data are converted into an N-dimensional matrix, the preprocessed time series matrix characteristics are found through a convolutional neural network, and then the preprocessed time series matrix characteristics are classified by using a SOFTMAX classifier, so that the complex and abstract time series are unified into the affiliated mode.
FIG. 4 provides a schematic diagram of a convolutional neural network model of yet another embodiment. In the method for identifying the traffic index time series mode by using the convolutional neural network model, the classification module is a SOFTMAX classifier; the convolutional neural network model is formed by combining 2 groups of structures of 1 convolutional layer 5 by 5 and 1 pooling layer in each group in a laminated structure; the activation function of the convolutional neural network model is a ReLU function; the pooling layer adopts maximum pooling. When the convolutional neural network model is applied to pattern recognition, firstly, the N-dimensional matrix of an input layer is subjected to feature extraction, 5 × 5 convolutional kernels are adopted to perform convolutional sampling processing, and scanning output results are subjected to maximum pooling through 2 × 2 pooling layers. The low-level features are extracted through the first convolution, but because the features need to be extracted and compressed continuously, the second convolution and pooling processing is carried out, and by analogy, the features are compressed one by one, so that more reliable feature information is obtained. The result of the first pooling was subjected to a second convolution using a second 5 x 5 convolution kernel, and then the pooled result was output using a 2 x 2 maximum pooling level. Extracting features through two convolution processes, outputting one-dimensional vectors through a full connection layer, obtaining the occurrence probability of each mode by using an SOFTMAX classifier, and taking the mode corresponding to the maximum probability value as the mode of the time-of-day sequence data.
In one embodiment, in the method for identifying a traffic index time series pattern by using a convolutional neural network model, pattern classification results of the traffic index time series data include a monday pattern, a middle of week pattern, a friday pattern, a saturday pattern and a sunday pattern; the traffic index time series data is traffic index time series data between 5 and 23 points in a natural day.
The pattern classification of the traffic index time series data can be determined according to the characteristics of the peak hours and the distribution of the people flow. Fig. 5(a) to 5(e) provide schematic diagrams of five modes of traffic index time-series data, and it can be seen that there is a distinct difference in the distribution of the five modes of traffic index time-series data, that is, the traffic index time-series data of different modes can be classified and identified by time-series characteristics.
In the process of training the convolutional neural network model, a predetermined pattern is labeled in a label mode on a training set formed by daily traffic index time series data.
The time span of the traffic index time-series data preferably takes data between 5 and 23 points. In the time periods before 5 o 'clock and after 23 o' clock, the flow rate of people is less, the difference between the traffic index time series data in different modes is not large, and in order to simplify the calculation process and reduce the data amount, the time periods before 5 o 'clock and after 23 o' clock are omitted. And between 5 and 23, the human flow is relatively large, the time sequence characteristics of more traffic index time sequence data are included, and the data in the time span are analyzed, so that the accuracy of pattern recognition on the traffic index time sequence data is improved.
In one embodiment, beijing city full market traffic index data for three months of 2016 (11 months), 2017 (11 months, 2018 (30 days of data without 2018 (11 months, 2018)) was used as the subject. The time frequency was 15 minutes and traffic indices were recorded at 05:00-23:00 a day in the morning.
Specific data are shown in table 1. According to the characteristics of the difference of the peak time and the distribution of the flow of people, the traffic index categories are divided into five categories, namely a monday mode, a week middle mode, a friday mode, a saturday mode and a sunday mode.
TABLE 1 traffic instruction list in Beijing City
The data of these three 11 months in year were recorded, 2016, 2017, 11, and 2018, 1, 15, as training sets, and labeled according to their actual conditions, and are represented by 0 and 1, for example, (1,0,0,0,0) represents a monday pattern, (0,1,0,0,0) represents a monday pattern, (0,0,1,0,0) represents a friday pattern, (0,0,0,1,0) represents a saturday pattern, and (0,0,0, 1) represents a sunday pattern.
Every 15 minutes, 73 values are counted every 05:00-23:00 every day, the data of every day is supplemented with a value of '0' at the head of time sequence data to form a 10 x 10 dimensional matrix, a label is marked at the tail of the matrix to distinguish the mode of every day, and finally the experimental data form an input layer of the convolutional neural network model.
After the above pre-processing, the data has been divided into a training set and a test set. The training set data includes: 2016 month 11, 2017 month 11 and 2018 month 11 month 1 to 2018 month 11 month 15 for a total of 74 days; the test set data includes: 16/11/2018 to 29/11/2018 for 14 days. Fig. 5(a) to 5(e) show the results of five pattern classification on the training set data.
In this embodiment, a convolutional neural network model is used as a deep learning model, and is implemented in the tensrflow by using a CPU version in Python language. The traffic index time series data is identified using the convolutional neural network model shown in fig. 4.
The result obtained by the time series pattern recognition based on the convolutional neural network in the present embodiment is shown in fig. 6, and the classification result is shown in table 2.
TABLE 2 convolution operation Pattern recognition results
| Categories | Results of the experiment | |
| 1 | Day 16, day 23 | Day 16, day 23 |
| 2 | 17 days and 24 days | 17 days and 24 days |
| 3 | 18 days and 25 days | 18 days and 25 |
| 4 | 19 days, 22 days, 26 days, and 27 days | 19 days and 26 days |
| 5 | 20 days, 21 days, 28 days, and 29 days | 20 days, 21 days, 22 days, 27 days, 28 days, and 29 days |
As shown in table 2, the convolution algorithm distinguishes between the monday mode (11-month-20 days, 11-month-21 days, 11-month-28 days, 11-month-29 days), the weekday mode (11-month-19 days, 11-month-22 days, 11-month-26 days, 11-month-27 days), the friday mode (11-month-16 days, 11-month-23 days), the saturday mode (11-month-17 days, 11-month-24 days), and the sunday mode (11-month-18 days, 11-month-25 days). However, the patterns of 11 months 22 and 11 months 27 are mistakenly divided into weekly patterns, which is not in accordance with the actual situation.
In order to verify the accuracy of pattern recognition of the convolutional neural network, the embodiment adopts a K-Means clustering algorithm to realize a comparison experiment in Tensflow to verify the classification accuracy of the convolutional neural network. The data from 11 months, 16 days to 29 days in 2018 were also classified into 5 categories. The clustering results are shown in Table 3, and the recognition results are shown in FIG. 7.
TABLE 3K-Means clustering results
According to table 3, the K-Means clustering algorithm can accurately distinguish the friday pattern (11 months 16 days, 11 months 23 days), the saturday pattern (11 months 17 days, 11 months 24 days), and the sunday pattern (11 months 18 days, 11 months 25 days), but there is a mistake in distinguishing the monday pattern from the intra-week pattern. The algorithm fuses 11-month 19 days, 11-month 20 days, 11-month 22 days, 11-month 26 days, 11-month 27 days and 11-month 29 days into a mode, correspondingly fuses three conditions of Monday, Tuesday and Thursday into a unified mode, and fuses results of 11-month 21 days and 11-month 28 days into a Wednesday mode, which is contradictory to the actual condition.
TABLE 4 identification results of two algorithms
As shown in fig. 6 and 7, the method of the present embodiment and the K-Means algorithm of the prior art can accurately identify the friday pattern, the saturday pattern and the weekend pattern. However, the two methods have different recognition capabilities for the monday mode and the monday mode in which the characteristic information is relatively hidden. The K-Means algorithm is based on the distance as the category difference of the acquaintance measuring sequence, and deep characteristic information cannot be mined, so that the clustering of the algorithm is different from the real situation. The convolutional neural network can fully extract traffic index characteristic information through repeated training, the most representative subsegment sequence is searched according to the characteristic information, the traffic index difference characteristics under different modes are determined, and the mode recognition task is completed. As can be seen from the experimental identification accuracy in table 4, the traffic index time series pattern identification accuracy of the convolution algorithm is higher than that of the conventional K-Means clustering algorithm, and the feasibility and superiority of the method provided by the invention on time series pattern identification are verified.
Due to the particularity of the time series data, the traditional clustering algorithm is difficult to accurately distinguish time series patterns, so the invention provides a method for carrying out pattern recognition on the time series data based on a convolutional neural network. In the experiment, 2016-plus-2018 Beijing city wholemarket traffic index data is taken as a research object, traffic index time sequence data form a 10-dimensional matrix, a neural network is used for repeatedly training and optimizing to obtain an optimal network model, convolution operation is performed on an input layer for two times, characteristic information of the time sequence data is trained and extracted, a SOFTMAX classifier is used for carrying out mode classification, and finally data from 11 months and 15 days to 11 months and 29 days in 2018 are divided into five modes: monday mode, weekly middle mode, friday mode, saturday mode, and sunday mode. Compared with a machine learning algorithm based on distance calculation, the convolutional neural network based on feature extraction has better recognition capability on time series data pattern recognition.
The method has important significance for the traffic management department to carry out overall control and make traffic plans, and is expected to provide service for relieving urban traffic pressure.
FIG. 8 is a diagram illustrating an embodiment of an apparatus for identifying a traffic index time series pattern using a convolutional neural network model according to the present invention. The device comprises: an obtaining module 201, configured to obtain traffic index time-series data; the extraction module 202 is configured to perform feature extraction on the traffic index time series data by using a convolutional neural network model to obtain a feature vector; and the classification module 203 is used for processing the feature vectors to obtain a mode classification result of the traffic index time series data.
The invention discovers the time sequence characteristics of the traffic index time sequence data through the convolutional neural network model, classifies the time sequence characteristics by utilizing the classification module, unifies the complex abstract time sequences into the modes, and is more accurate and effective in identifying the time sequence data modes, thereby being suitable for the research on the correlation between data presenting time stage change and the mode discovery. Furthermore, based on the mode recognition result of the traffic index time series data, the traffic travel rules of the traffic index time series data can be extracted according to different modes, so that reference is provided for making a traffic plan for related management departments.
In one embodiment, in the apparatus for identifying a traffic index time series pattern using a convolutional neural network model, the extraction module includes: the conversion sub-module is used for converting the traffic index time sequence data into an N-dimensional matrix; and the extraction submodule is used for extracting the characteristics of the N-dimensional matrix by using the convolutional neural network model to obtain a characteristic vector.
In one embodiment, in the apparatus for identifying a traffic index time series pattern by using a convolutional neural network model, the classification module is a SOFTMAX classifier; the convolutional neural network model is formed by combining 2 groups of structures of 1 convolutional layer of 5 x 5 and 1 pooling layer of 2 x 2 in each group in a laminated structure; the activation function of the convolutional neural network model is a ReLU function; the pooling layer adopts maximum pooling.
In one embodiment, the pattern classification result of the traffic index time series data includes a monday pattern, a weekly middle pattern, a friday pattern, a saturday pattern, and a sunday pattern; the traffic index time series data is traffic index time series data between 5 and 23 points in a natural day.
Corresponding to the method for identifying the traffic index time series pattern by using the convolutional neural network model in fig. 1, the present invention also provides an electronic device, as shown in fig. 9, including: at least one processor, and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the method.
Specifically, the memory 1000 and the processor 2000 can be general-purpose memories and processors, and are not particularly limited herein, and when the processor 2000 executes a computer program stored in the memory 1000, the above-described method of recognizing a traffic index time-series pattern using a convolutional neural network model can be performed, so that the traffic index time-series pattern can be recognized more accurately and efficiently.
The invention also provides a storage medium on which a computer program is stored which, when executed by a processor, carries out the method. For specific implementation, reference may be made to the method embodiment, which is not described herein again.
While embodiments of the invention have been disclosed above, it is not intended to be limited to the uses set forth in the specification and examples. It can be applied to all kinds of fields suitable for the present invention. Additional modifications will readily occur to those skilled in the art. It is therefore intended that the invention not be limited to the exact details and illustrations described and illustrated herein, but fall within the scope of the appended claims and equivalents thereof.
Claims (10)
1. A method for identifying a traffic index time series pattern by using a convolutional neural network model is characterized by comprising the following steps:
acquiring traffic index time series data;
performing feature extraction on the traffic index time series data by using a convolutional neural network model to obtain a feature vector;
and processing the feature vectors by using a classification module to obtain a mode classification result of the traffic index time series data.
2. The method for identifying a traffic index time series pattern by using a convolutional neural network model as claimed in claim 1, wherein the extracting features from the traffic index time series data by using the convolutional neural network model to obtain a feature vector comprises:
converting the traffic index time series data into an N-dimensional matrix;
and performing feature extraction on the N-dimensional matrix by using the convolutional neural network model to obtain a feature vector.
3. The method for identifying traffic index time series patterns using convolutional neural network model of claim 2, wherein the classification module is a SOFTMAX classifier; the convolutional neural network model is formed by combining 2 groups of structures of 1 convolutional layer 5 by 5 and 1 pooling layer in each group in a laminated structure; the activation function of the convolutional neural network model is a ReLU function; the pooling layer adopts maximum pooling.
4. The method for identifying a traffic index time series pattern using a convolutional neural network model as claimed in claim 2 or 3, wherein the pattern classification results of the traffic index time series data include a monday pattern, a middle of the week pattern, a friday pattern, a saturday pattern, and a sunday pattern; the traffic index time series data is traffic index time series data between 5 and 23 points in a natural day.
5. An apparatus for identifying a traffic index time series pattern using a convolutional neural network model, comprising:
the acquisition module is used for acquiring traffic index time series data;
the extraction module is used for extracting the characteristics of the traffic index time series data by using a convolutional neural network model to obtain a characteristic vector;
and the classification module is used for processing the characteristic vector to obtain a mode classification result of the traffic index time series data.
6. The apparatus for identifying traffic index time series patterns using convolutional neural network model as set forth in claim 5, wherein the extraction module comprises:
the conversion sub-module is used for converting the traffic index time sequence data into an N-dimensional matrix;
and the extraction submodule is used for extracting the characteristics of the N-dimensional matrix by using the convolutional neural network model to obtain a characteristic vector.
7. The apparatus for identifying traffic index time series patterns using convolutional neural network model of claim 6, wherein the classification module is a SOFTMAX classifier; the convolutional neural network model is formed by combining 2 groups of structures of 1 convolutional layer 5 by 5 and 1 pooling layer in each group in a laminated structure; the activation function of the convolutional neural network model is a ReLU function; the pooling layer adopts maximum pooling.
8. The apparatus of claim 6 or 7, wherein the pattern classification result of the traffic index time series data comprises a monday pattern, a middle of week pattern, a friday pattern, a saturday pattern and a sunday pattern; the traffic index time series data is traffic index time series data between 5 and 23 points in a natural day.
9. An electronic device, comprising: at least one processor, and a memory communicatively coupled to the at least one processor, wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform the method of any of claims 1-4.
10. A storage medium on which a computer program is stored, which program, when being executed by a processor, is adapted to carry out the method of any one of claims 1-4.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010135586 | 2020-03-02 | ||
| CN2020101355868 | 2020-03-02 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111339978A true CN111339978A (en) | 2020-06-26 |
Family
ID=71184077
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010140374.9A Pending CN111339978A (en) | 2020-03-02 | 2020-03-03 | Method for recognizing traffic index time series mode by using convolutional neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111339978A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112330158A (en) * | 2020-11-06 | 2021-02-05 | 北京建筑大学 | Method for identifying traffic index time sequence based on autoregressive differential moving average-convolution neural network |
| CN113095422A (en) * | 2021-04-21 | 2021-07-09 | 广东电网有限责任公司 | Method and device for realizing automatic programming order issuing |
| CN113553350A (en) * | 2021-05-27 | 2021-10-26 | 四川大学 | Traffic flow partition model for similar evolution mode clustering and dynamic time zone partitioning |
| CN116051173A (en) * | 2023-03-30 | 2023-05-02 | 安徽交欣科技股份有限公司 | Passenger flow prediction method, passenger flow prediction system and bus dispatching method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862331A (en) * | 2017-10-31 | 2018-03-30 | 华中科技大学 | It is a kind of based on time series and CNN unsafe acts recognition methods and system |
| CN109034264A (en) * | 2018-08-15 | 2018-12-18 | 云南大学 | Traffic accident seriousness predicts CSP-CNN model and its modeling method |
| CN109165698A (en) * | 2018-10-16 | 2019-01-08 | 中国电子科技集团公司信息科学研究院 | A kind of image classification recognition methods and its storage medium towards wisdom traffic |
| CN110033021A (en) * | 2019-03-07 | 2019-07-19 | 华中科技大学 | A kind of Fault Classification based on one-dimensional multichannel convolutional neural networks |
| CN110503205A (en) * | 2019-07-17 | 2019-11-26 | 中国科学院计算技术研究所 | A kind of multi-modal data processing method and system |
| US20190379589A1 (en) * | 2018-06-12 | 2019-12-12 | Ciena Corporation | Pattern detection in time-series data |
-
2020
- 2020-03-03 CN CN202010140374.9A patent/CN111339978A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862331A (en) * | 2017-10-31 | 2018-03-30 | 华中科技大学 | It is a kind of based on time series and CNN unsafe acts recognition methods and system |
| US20190379589A1 (en) * | 2018-06-12 | 2019-12-12 | Ciena Corporation | Pattern detection in time-series data |
| CN109034264A (en) * | 2018-08-15 | 2018-12-18 | 云南大学 | Traffic accident seriousness predicts CSP-CNN model and its modeling method |
| CN109165698A (en) * | 2018-10-16 | 2019-01-08 | 中国电子科技集团公司信息科学研究院 | A kind of image classification recognition methods and its storage medium towards wisdom traffic |
| CN110033021A (en) * | 2019-03-07 | 2019-07-19 | 华中科技大学 | A kind of Fault Classification based on one-dimensional multichannel convolutional neural networks |
| CN110503205A (en) * | 2019-07-17 | 2019-11-26 | 中国科学院计算技术研究所 | A kind of multi-modal data processing method and system |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112330158A (en) * | 2020-11-06 | 2021-02-05 | 北京建筑大学 | Method for identifying traffic index time sequence based on autoregressive differential moving average-convolution neural network |
| CN112330158B (en) * | 2020-11-06 | 2023-10-13 | 北京建筑大学 | Method for identifying traffic index time sequence based on autoregressive differential moving average-convolution neural network |
| CN113095422A (en) * | 2021-04-21 | 2021-07-09 | 广东电网有限责任公司 | Method and device for realizing automatic programming order issuing |
| CN113553350A (en) * | 2021-05-27 | 2021-10-26 | 四川大学 | Traffic flow partition model for similar evolution mode clustering and dynamic time zone partitioning |
| CN113553350B (en) * | 2021-05-27 | 2023-07-18 | 四川大学 | Traffic flow partition model for similar evolution mode clustering and dynamic time zone division |
| CN116051173A (en) * | 2023-03-30 | 2023-05-02 | 安徽交欣科技股份有限公司 | Passenger flow prediction method, passenger flow prediction system and bus dispatching method |
| CN116051173B (en) * | 2023-03-30 | 2023-06-30 | 安徽交欣科技股份有限公司 | Passenger flow prediction method, passenger flow prediction system and bus dispatching method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111339978A (en) | Method for recognizing traffic index time series mode by using convolutional neural network model | |
| CN111695467B (en) | Spatial spectrum full convolution hyperspectral image classification method based on super-pixel sample expansion | |
| CN114067160B (en) | Small sample remote sensing image scene classification method based on embedded smooth graph neural network | |
| CN112132006B (en) | Intelligent forest land and building extraction method for cultivated land protection | |
| KR20180120056A (en) | Method and system for pre-processing machine learning data | |
| CN105354273A (en) | Method for fast retrieving high-similarity image of highway fee evasion vehicle | |
| CN111696101A (en) | Light-weight solanaceae disease identification method based on SE-Inception | |
| CN108982377A (en) | Corn growth stage spectrum picture and chlorophyll content correlation and period division methods | |
| Huang et al. | Research on optimization methods of ELM classification algorithm for hyperspectral remote sensing images | |
| CN115439754A (en) | Large-range trans-climatic region crop mapping method based on time sequence remote sensing image | |
| CN112330158A (en) | Method for identifying traffic index time sequence based on autoregressive differential moving average-convolution neural network | |
| CN112434662B (en) | Tea leaf scab automatic identification algorithm based on multi-scale convolutional neural network | |
| CN111242028A (en) | Remote sensing image ground object segmentation method based on U-Net | |
| CN113032613B (en) | Three-dimensional model retrieval method based on interactive attention convolution neural network | |
| Pal | Factors influencing the accuracy of remote sensing classifications: a comparative study | |
| CN112633169B (en) | Pedestrian recognition algorithm based on improved LeNet-5 network | |
| CN116612307A (en) | Solanaceae disease grade identification method based on transfer learning | |
| CN114417938B (en) | Electromagnetic target classification method embedded by using knowledge vector | |
| Song et al. | Study of Sugarcane Buds Classification Based on Convolutional Neural Networks. | |
| LU503005B1 (en) | A layout-unconstrained method based on graph reasoning network for reading text block | |
| Kumar et al. | Image classification of land use land cover of bengaluru city using convolutional neural network | |
| CN114757776A (en) | Quantitative stock selection technical method combining LSTM and hypergraph neural network | |
| Kumar et al. | C-SPIN: Classification of Satellite Images for Prediction Network | |
| Zhou | Comparison of Random Forests, Support Vector Machine and Artificial Neural Network Methods for Agriculture Land Cover Classification | |
| CN116778205A (en) | Citrus disease grade identification method, equipment, storage medium and device |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200626 |