CN115984077B - Traffic abnormal flow causal detection method and equipment - Google Patents
Traffic abnormal flow causal detection method and equipment Download PDFInfo
- Publication number
- CN115984077B CN115984077B CN202310160350.3A CN202310160350A CN115984077B CN 115984077 B CN115984077 B CN 115984077B CN 202310160350 A CN202310160350 A CN 202310160350A CN 115984077 B CN115984077 B CN 115984077B
- Authority
- CN
- China
- Prior art keywords
- flow
- traffic
- abnormal
- values
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A30/00—Adapting or protecting infrastructure or their operation
- Y02A30/60—Planning or developing urban green infrastructure
Landscapes
- Traffic Control Systems (AREA)
Abstract
The invention discloses a traffic abnormal flow causal detection method and equipment, which are characterized in that after an urban area is segmented to obtain an urban area node diagram, links are created between nodes according to acquired urban flow data, the urban area flow change is monitored on flow changes of other areas of the city through the links, space-time abnormal values are obtained through calculating the distance values of the links, so that the abnormal causal diagram is obtained through calculating the space-time abnormal values through an abnormal fruit tree algorithm, the normal state is distinguished from the abnormal state flow under the abnormal condition by combining the normal state causal diagram under the normal flow state, the abnormal change characteristics of the traffic flow caused by the space-time abnormal values can be captured, the causal inspection of the abnormal values and the transmissibility of the change with time are realized, the potential interaction between different urban areas and roads is revealed, the effective perception of the urban traffic state is promoted, and more accurate decision is facilitated.
Description
Technical Field
The invention relates to the technical field of intelligent traffic, in particular to a traffic abnormal flow causal detection method and equipment.
Background
The travel is a core part in the development and construction of smart cities. The intelligent travel is used as an innovative travel mode, advanced technologies such as artificial intelligence, the Internet of things, space perception, big data and cloud computing are utilized to be fused, and the whole process of comprehensive and traffic construction management is controlled and supported in the traffic fields such as urban traffic management, traffic transportation and people travel. The system can enable the transportation travel system to have excellent capabilities of Internet of things, perception, interconnection, monitoring, early warning, prevention and control and the like in a region, a city and even a wider space range. The method plays a great role in guaranteeing the safety of the transportation trip and improving the high efficiency and the comprehensive management capability of the transportation trip. Traffic is considered to be one of the most promising applications in all application scenarios of the internet of things.
With the development of urban traffic, traffic problems are more and more serious, so that the traditional solution cannot meet new traffic problems, and intelligent traffic is generated. Intelligent transportation refers to the effective integration of advanced information technology, data transmission technology, computer processing technology and the like into a transportation management system, so that people, vehicles and roads can be closely matched, the transportation environment is improved, the resource utilization rate is improved, and the like.
The anomaly detection analysis can solve some urban scale problems such as traffic, public safety, crime prevention, efficient resource utilization and the like, and improve the user experience of large public and private spaces. Outlier detection, also known as anomaly detection, refers to detecting data that deviates significantly from most data instances. It is increasingly being used in areas such as risk management, compliance, security, financial supervision, health and medical risks, and artificial intelligence, where outlier detection plays an increasingly important role.
The increasing popularity of positioning technologies, including GPS and WIFI, generates a large amount of spatio-temporal data, one of the main forms of which is spatio-temporal data in the form of trajectory data. The abnormal pattern of trajectory data generated by a large number of moving objects can generally reflect the abnormal traffic flow pattern on the traffic road network, which may be caused by non-periodic events such as celebration events, large commercial promotions, temporary traffic control, special large sporting events, etc., and may be periodic traffic congestion caused by a somewhat unreasonable planning of traffic roads. Thus, detecting outliers from the trajectory data helps the manager to perceive the outliers in time and reduce their anomalous impact on traffic flow.
For the current outlier detection technology, the detection method includes Principal Component Analysis (PCA) and variants thereof, DPMM (DirichletProcess Mixture Model, dirichlet process hybrid model) and deep neural network model. Principal component analysis and variants thereof are widely used to detect outliers from spatial and temporal data, while DPMM is more focused on studies of outlier finger detection in traffic data. Meanwhile, due to the strong learning capacity of the deep neural network model, researchers also use the deep neural network model in the field of outlier detection.
Principal Component Analysis (PCA) and variants thereof are the more popular data dimension reduction and anomaly detection methods. The main idea of principal component analysis is to use orthogonal transformation to convert linearly related data into linearly uncorrelated data. PCA is applied to a link time matrix that displays traffic on different roads in a time window to detect the root cause of abnormal traffic behavior.
DPMM may be used to derive outliers in urban traffic flow data. First, a set of all flow values is projected into an n-dimensional space. The dimensionality of the data is then reduced to a two-dimensional space by Principal Component Analysis (PCA). The traffic values are then clustered into classes according to a chinese restaurant process (Chinese restaurant process). Each traffic value is assigned to a new cluster with a probability proportional to the number parameter a, otherwise it will be assigned to the previously created cluster. Thereafter, all flow values belonging to the cluster having the largest number of elements are regarded as internal values, and the remaining flow values are regarded as outliers, i.e. outliers.
The method for detecting the abnormal value by the deep neural network comprises the following steps: researchers have trained three separate automatic encoder networks to learn three different features of video anomaly detection: appearance, motion, and appearance-motion joint characteristics. A classifier integrated by three support vector machines is trained independently on each learned feature representation to perform anomaly scoring. First, outlier detection is performed on a low-dimensional feature representation of Gao Weiyuan data generated by a deep neural network using a Linear support vector machine (Linear SVM). After further optimization, the method of unsupervised classification was used instead of SVM (SupportVector Machine) to achieve anomaly scoring in the projection space. In this technique, they first cluster the low-dimensional features of the video frames produced by the convolution-based auto-encoder and then classify the clustered tags as pseudo tags. The classification probability is used to calculate an anomaly score for each frame.
For the causal relationship verification technique, the traditional analytical verification causal relationship method is a controlled variable experiment, such as a/B detection, is widely used in industry. The A/B test is a conventional causal analysis verification method that randomly assigns groups of variables to specific univariate process levels, compares the performance of one or more test groups to control groups, and finally performs the test.
The following disadvantages exist when the above-described method:
disadvantage 1: in outlier detection, the results of Principal Component Analysis (PCA) are very sensitive to parameter settings of highly dependent data. In some cases, extreme outliers can in turn affect the calculation of PCA, leading to false positives and false negatives of outliers. At the same time, the technology lacks causal verification of outliers and transitivity over time.
Disadvantage 2: the existing space-time causal detection method is focused on the causal relation under the abnormal condition, and some attempts are made to mine the causal relation among the overall traffic roads from the data, but road causal diagrams of different conditions under the normal condition and the abnormal condition are not distinguished.
Disadvantage 3: in terms of causality verification, conventional a/B testing requires obtaining different test feedback for different variables from the same user or sample to determine causality between the variables. In practice, it is very difficult to obtain data of the same sample under different conditions, for example, traffic data is usually only obtained at a certain moment, and the influence of the flow of the outflow of a plurality of areas at the same moment on the flow of other areas cannot be obtained. Meanwhile, the causal check relation is simply related, but is a stronger association relation, so that the reliability reapplication of the A/B detection is often limited.
Disclosure of Invention
The technical problems to be solved by the invention are as follows: the method and the equipment for detecting the traffic abnormal flow cause and effect can better understand the current road traffic mode and make a more timely coping strategy by establishing a more accurate cause and effect relation graph;
in order to solve the technical problems, the invention adopts the following technical scheme:
a traffic abnormal flow causal detection method comprises the following steps:
dividing the urban area into areas to obtain an urban area node map;
acquiring flow data, and constructing links between nodes according to the urban area node diagram and the flow data;
calculating all the linked distance values, and according to the distance values, calculating space-time abnormal values;
judging the relation between the space-time abnormal values according to an abnormal fruit tree algorithm to obtain an abnormal causal graph;
and judging the relation between nodes according to the traffic city area node diagram and the traffic data to obtain a normal state causal diagram.
In order to solve the technical problems, the invention adopts another technical scheme that:
a traffic anomaly traffic cause and effect detection device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of a traffic anomaly traffic cause and effect detection method as described above when the computer program is executed.
The invention has the beneficial effects that: after the urban area is subjected to area segmentation to obtain an urban area node diagram, links are created between nodes according to the acquired urban flow data, the urban area flow change is monitored on flow changes of other areas of the city through the links, space-time anomaly values are obtained through calculation of linked distance values, so that an anomaly causal diagram is obtained through calculation of the space anomaly values through an anomaly cause and fruit tree algorithm, and flow under normal flow conditions is distinguished from flow under anomaly conditions by combining the causal diagram of the normal flow conditions, the anomaly change characteristics of traffic flow caused by the space-time anomaly values can be captured, causal inspection of the anomaly values and transmissibility of time change are achieved, potential interaction between different urban areas and roads is revealed, effective perception of urban traffic conditions is promoted, and more accurate decision is facilitated.
Drawings
FIG. 1 is a flow chart of steps of a method for causal detection of abnormal traffic flow in an embodiment of the invention;
FIG. 2 is a schematic diagram of a logic structure of a traffic anomaly traffic cause and effect detection method according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a traffic abnormal flow cause and effect detection device according to an embodiment of the present invention.
Detailed Description
In order to describe the technical contents, the achieved objects and effects of the present invention in detail, the following description will be made with reference to the embodiments in conjunction with the accompanying drawings.
Referring to fig. 1, a traffic abnormal flow causal detection method includes the steps of:
dividing the urban area into areas to obtain an urban area node map;
acquiring flow data, and constructing links between nodes according to the urban area node diagram and the flow data;
calculating all the linked distance values, and according to the distance values, calculating space-time abnormal values;
judging the relation between the space-time abnormal values according to an abnormal fruit tree algorithm to obtain an abnormal causal graph;
and judging the relation between nodes according to the traffic city area node diagram and the traffic data to obtain a normal state causal diagram.
From the above description, the beneficial effects of the present invention are as follows: after the urban area is subjected to area segmentation to obtain an urban area node diagram, links are created between nodes according to the acquired urban flow data, the urban area flow change is monitored on flow changes of other areas of the city through the links, space-time anomaly values are obtained through calculation of linked distance values, so that an anomaly causal diagram is obtained through calculation of the space anomaly values through an anomaly cause and fruit tree algorithm, and flow under normal flow conditions is distinguished from flow under anomaly conditions by combining the causal diagram of the normal flow conditions, the anomaly change characteristics of traffic flow caused by the space-time anomaly values can be captured, causal inspection of the anomaly values and transmissibility of time change are achieved, potential interaction between different urban areas and roads is revealed, effective perception of urban traffic conditions is promoted, and more accurate decision is facilitated.
Further, the performing region segmentation on the urban region to obtain a urban region node map includes:
dividing the urban area by adopting a regular hexagon to obtain a node diagram of the urban area;
each of the nodes represents an area;
the links between the nodes represent the traffic from region to region.
From the above description, it can be seen that the urban area is divided into a plurality of areas by using the honeycomb hexagon to construct an area graph, the corresponding divided areas are represented by the nodes in the graph, and the flow between the two areas is represented by the edges of the hexagon, so that the flow change of each node in the urban area and the flow change between the nodes can be accurately described.
Further, the constructing a link between nodes according to the urban area node map and the traffic data includes:
judging whether traffic interaction exists between the nodes according to the traffic data, and if yes, establishing a link between two groups of nodes with traffic interaction.
As can be seen from the above description, when there is traffic interaction between two nodes, it is indicated that there is an interaction relationship between two nodes, and by establishing a link between the nodes where the interaction relationship exists, the association between the nodes is improved.
Further, the calculating distance values of all the links, and according to the distance values, the space-time outliers include:
calculating all the linked distance characteristic values, and obtaining a time abnormal value according to the distance characteristic values;
calculating the distances between all different links, and obtaining a space abnormal value according to the distances between the links;
the spatiotemporal outliers are derived from intersections of the temporal outliers and the spatial outliers.
From the above description, it can be seen that, by obtaining the time outliers according to the distance values of the links, and obtaining the space outliers according to the distances between the links, and finally screening out the data satisfying both the time outliers and the space outliers, the space-time outliers are obtained, so that the outliers in the medium-flow data can be accurately screened out.
Further, the calculating the distance eigenvalues of all the links, and obtaining the time outliers according to the distance eigenvalues includes:
calculating the average value of the flow distance between all time frames in each link and any other time frame to obtain the average value of the distance between each time frame and other time frames;
and screening out extreme values from all the distance average values, and taking the extreme values as the time abnormal values.
As can be seen from the above description, by calculating the average value of the distances between the flows of each link on different time frames, the time frames with abnormal flow data in each link can be screened out, and the time abnormal value corresponding to each link can be obtained.
Further, the calculating the distances between all the different links, and obtaining the spatial outliers according to the distances between the links includes:
calculating flow distance values between any two groups of links under the same time frame to obtain flow distance values between each link and all other links under each time frame;
and screening out extreme values from all the flow distance values, and taking the extreme values as the space abnormal values.
As can be seen from the above description, by calculating the distance corresponding to each link and other links in the same time frame of the same day, links with a longer distance from other links can be screened out, that is, the links have spatial anomalies relative to other links, so as to obtain accurate spatial anomaly values.
Further, the filtering the extremum from all the flow distance values, and taking the extremum as the space outlier includes:
normalizing all the flow distance values;
and screening out an extremum from the normalized flow distance value, and taking the extremum as the space abnormal value.
From the above description, the influence of different values in different areas can be eliminated by normalizing the flow distance value and then screening, so that the screening precision is improved.
Further, the determining the relationship between the nodes according to the traffic city area node map and the traffic data, and obtaining the normal cause and effect map includes:
aggregating the input flow and the output flow of each node to obtain a flow time sequence;
the flow time sequence is processed through feature extraction and difference to obtain a stable flow sequence;
and carrying out the Granger causal inspection on the stable flow sequence to obtain the relation between the nodes and obtain the normal causal graph.
From the above description, it can be seen that, by aggregating the input traffic and the output traffic of each node and performing feature extraction and differential processing on the traffic time sequence obtained by aggregation, the periodicity of the traffic time sequence can be eliminated, and a stable traffic sequence can be further obtained, which is more beneficial to performing the glaring causal inspection, and the validity of the inspection result is improved.
Further, the obtaining the smooth flow sequence through feature extraction and differential processing of the flow time sequence includes:
extracting features of the flow time sequence by adopting a non-negative matrix decomposition method;
and carrying out the differential processing on the fitting combination of the non-negative matrix factorization by the flow data to obtain the stable flow sequence.
As can be seen from the above description, by performing feature extraction on the traffic time series by using the non-negative matrix decomposition method, only additive combination of the traffic time series in the traffic time series can be allowed, thereby achieving the effect of hiding features, being more suitable for processing traffic flow data, and improving the data processing effect.
Referring to fig. 3, another embodiment of the present invention provides a traffic abnormal flow cause and effect detection apparatus, which includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the steps in a traffic abnormal flow cause and effect detection method as described above when executing the computer program. And visualizing a result graph obtained by the traffic abnormal flow causal detection method.
The traffic abnormal flow causal detection method and the traffic abnormal flow causal detection equipment can be suitable for urban intelligent traffic scenes, can better understand the current road traffic mode and make more timely coping strategies by establishing an accurate causal relation graph, and are described by the following specific embodiments:
example 1
Referring to fig. 1 and 2, a traffic abnormal flow causal detection method includes the steps of:
s1, carrying out region segmentation on a city region to obtain a city region node diagram;
the grid system is important for analyzing a large-scale space data set and dividing a city area into identifiable grid units, and in the embodiment, a honeycomb hexagon is adopted to divide the city area so as to obtain a node diagram of the city area; constructing a region graph according to the track data, wherein each node in the region graph represents a region; the connection line between the nodes represents the flow between the areas;
s2, acquiring flow data, and constructing links between nodes according to the urban area node diagram and the flow data; specifically, judging whether traffic interaction exists between the nodes according to the traffic data, if so, establishing a link between two groups of nodes with traffic interaction; if the flow data is taxi order track data; the track data comprises four characteristics of taxi order numbers, time stamps, longitudes and latitudes; sequencing the same order numbers of the same taxis according to the sequence of the time stamps to obtain taxi driving tracks; for two pieces of track data of the taxi, if the two pieces of track data are continuous in time (namely continuous in time stamp) and continuous in space (namely continuous in longitude and latitude), namely the taxi moves from one area to an adjacent area, the two areas are considered to generate flow interaction; for two pieces of continuous track data, if they respectively belong to two different areas, it is considered that there is a link between the two areas at this time; namely, a link is from one divided area to the next area, and the link has directivity;
converting each of said trajectory data into a series of said links between pairs of regions by scanning the entire trajectory data set; if traffic interactions are generated between two regions, then connect the two regions through the link, where link= (total traffic, traffic out, traffic in); i.e. one of said links is a vector comprising three dimensional information; wherein the total flow is defined as the total flow on the link, i.e. the total number of vehicles passing through both areas in a certain period of time; the flow rate, i.e. the ratio of the flow rate flowing out through the link in a certain period of time to the total flow rate in the area; the flow rate, i.e. the ratio of the flow rate flowing in through the link in a certain period of time to the total flow rate in the area;
s3, calculating all the linked distance values, and according to the distance values, obtaining space-time abnormal values; specifically, defining an outlier from the distance values as the spatiotemporal outlier, wherein only outliers having anomalies in both time and space are considered to be spatiotemporal outliers;
s31, calculating all the linked distance characteristic values, and obtaining a time abnormal value according to the distance characteristic values;
s311, calculating the average value of the flow distances between all time frames in each link and any other time frame to obtain the average value of the distances between each time frame and other time frames, namely obtaining the distance characteristic value of the link; in an alternative embodiment, the euclidean distance is used to perform the above calculation, that is, the data related to time, such as the same time on different dates or the same day on different weeks, can be obtained through the above calculation, and a special traffic data mode with similar behavior can be observed; as for link L1, it will have different flow properties { (flow 1, in1, out 1), (flow 2, in2, out 2), (flow 3, in3, out 3), (flow 4, in4, out 4) } over different time frames { t1, t2, t3, t4.. }, where flow1, in1 and out1 represent the total flow, the flow out and the flow in the link definition, respectively; the distance between any two time frames is calculated by the following formula:
i.e., t1t2, t1t3, t1t4., t2t1, t2t3, t2t 4., t3t1, t3t2, t3t 4..; for a t1 time frame, its distance relative to other time frames is t1t2+t1t3+t1t4./ n; and so on, the distances from all time frames to other time frames can be obtained to obtain the average value of the distances as the characteristic value of the distances;
s312, screening out extreme values from all the distance average values, and taking the extreme values as the time abnormal values; the distance feature value calculated in step S311 is subjected to normalization processing by subtracting a minimum value from the distance feature value and dividing the minimum value by a maximum value: (x-min)/max, so that the distance characteristic value is in the range of [0,1], and the influence of the numerical values of different areas is eliminated; screening the distance characteristics of different time frames of each link according to a given threshold value, wherein the maximum value, namely the time frame far away from all other links, is regarded as a time abnormal value; as in an alternative embodiment, for the normalized distance feature values, a threshold = 0.95 is set to filter out the first 5 percent distance feature value where the distance is greatest and consider it as the maximum value of the distance feature;
s32, calculating the distances among all different links, and obtaining a space outlier according to the distances among the links; for the detection of the spatial outliers, the distance between two links is calculated in a time frame range to judge, namely, the maximum difference value between different links in the same time frame is searched to represent, and the detection is specific:
s321, calculating flow distance values between any two groups of links under the same time frame to obtain flow distance values between each link and all other links under each time frame; calculating the distance between two link traffic in the same time frame of the same day, wherein the distance is expressed by a Markov distance; that is, for the links on the same time frame, for example, any two links may have different traffic attributes such as including links { L1, L2, L3, L4. }, where L1 is { (flow 1, in1, out 1), L2 is (flow 2, in2, out 2), L3 is (flow 3, in3, out 3), and L4 is (flow 4, in4, out 4) }, the traffic distance of any two links is calculated by the following formula:
x and y are two different links, i.e. the distance between the link and all other links at time t1 can be obtained for L1, and so on, the distance between any two links at time t1 can be obtained;
s322, screening out extreme values from all the flow distance values, and taking the extreme values as the space abnormal values; normalizing all the flow distance values; screening out an extremum from the normalized flow distance values, and taking the extremum as the space abnormal value; if the extremum in all links in the t1 time frame range is detected, the links with the largest difference between the characteristics of the links and the spatial adjacent links are the spatial outliers;
s33, obtaining the space-time outlier according to the intersection of the time outlier and the space outlier; that is, the intersection obtained by intersecting the time outlier result obtained in step S31 with the space outlier result obtained in step S32 is the space-time outlier;
s4, judging the relation between the space-time abnormal values according to an abnormal fruit tree algorithm to obtain an abnormal causal graph; looking up outlier cause and effect relationships based on an outlier cause and effect tree algorithm looking up relationships between outliers in the earliest time range to the last time range; when one of said spatiotemporal outliers STO1 occurs temporally before another of said spatiotemporal outliers STO2 and they also have a spatially end-to-end relationship, then outlier STO1 is considered to be the cause of another outlier STO 2; the dependency relationship is found according to the abnormal link value obtained in the previous step, specifically:
for each of said spatiotemporal outliers it comprises a temporal attribute (i.e. the time at which the anomaly occurred), a spatial attribute (i.e. the two regions at which the anomaly link occurred); then, for the abnormal link L1 (t 1, start grid1, end grid 1) and the abnormal link L2 (t 2, start grid2, end grid 2), if t1 and t2 are continuous in time (such as 10:30 and 11:00) and end grid1 of L1 and start grid2 of L2 are continuous (i.e. end grid1 and start grid2 are the same, are the same region), then the abnormal link L1 and the abnormal link L2 are considered to be continuous in time and space, and then the two space-time abnormal values are considered to have a causal relationship;
s5, judging the relation between nodes according to the traffic city area node diagram and the traffic data to obtain a normal state causal diagram;
s51, aggregating the input flow and the output flow of each node to obtain a flow time sequence; to obtain flow information in a normal state, all flows input and output from each area are aggregated; wherein the start region of the link is denoted e.g. by Rgno, and the arrival region of the link is denoted Rgnd; then for any region Rgnk, the flow is input: rnk.info=Σi= 1 to nlinki.flow where linki.Rgno =rnnk; output flow rate: rdnk. Output flow=Σi= 1 to nlinki.flow wherelinki.Rgnd =rdnk;
s52, carrying out feature extraction and differential processing on the flow time sequence to obtain a stable flow sequence, wherein the flow time sequence is specifically:
s521, carrying out feature extraction on the flow time sequence by adopting a non-negative matrix decomposition method; thereby removing the periodicity of the flow time series in step S51;
s522, performing the difference processing on the fitting combination of the non-negative matrix factorization by the flow data to obtain the stable flow sequence; if the number of columns of W and the number of rows of H are set to be the data volume of one day, the non-negative matrix factorization can extract characteristics according to the day and carry out differential processing on the fitting result of the non-negative matrix factorization by utilizing the original data to obtain differential fluctuation of the original data, so that the periodic influence is removed, and the stable flow sequence is obtained;
s53, carrying out the Granger causal inspection on the stable flow sequence to obtain the relation between the nodes and obtain the normal causal graph;
if the glaring causal check is carried out on all the stable flow sequences, taking the result of the glaring causal check as the edge between the area directed graphs, namely, if causal relation exists between the two areas, the edge exists, otherwise, the causal relation does not exist; meanwhile, in order to further extract causality rather than relativity, the Grandis causal inspection is carried out on different time lengths, and finally, a normal regional causal graph is obtained; and the traffic flow is improved through the obtained abnormal causal graph and the obtained normal causal graph.
Example two
Referring to fig. 3, a traffic abnormal flow cause and effect detection apparatus includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a traffic abnormal flow cause and effect detection method according to the first embodiment when executing the computer program.
In summary, according to the method and the device for detecting the traffic abnormal flow cause and effect provided by the invention, after the urban area is segmented to obtain the urban area node diagram, links are created between nodes according to the acquired urban flow data, the urban area flow change is monitored on the flow change of other areas of the city through the links, the space-time abnormal value is obtained by calculating the linked distance value, the abnormal cause and effect diagram is obtained by calculating the time-space abnormal value through the abnormal cause and effect diagram algorithm, the normal cause and effect diagram in the normal state of the flow is combined, the flow is distinguished from the normal state in the abnormal state, the abnormal change characteristic of the traffic flow caused by the space-time abnormal value can be captured, the cause and effect detection of the abnormal value and the transmissibility of the change along with time are realized, the potential interaction between different urban areas and roads is disclosed, the effective perception of the urban traffic state is promoted, and more accurate decision is facilitated.
The foregoing description is only illustrative of the present invention and is not intended to limit the scope of the invention, and all equivalent changes made by the specification and drawings of the present invention, or direct or indirect application in the relevant art, are included in the scope of the present invention.
Claims (8)
1. The traffic abnormal flow causal detection method is characterized by comprising the following steps:
dividing the urban area into areas to obtain an urban area node map;
acquiring flow data, and constructing links between nodes according to the urban area node diagram and the flow data;
calculating all the linked distance values, and according to the distance values, calculating space-time abnormal values;
judging the relation between the space-time abnormal values according to an abnormal fruit tree algorithm to obtain an abnormal causal graph;
judging the relation between nodes according to the urban area node diagram and the flow data to obtain a normal causal diagram;
the calculating of the distance values of all the links and the space-time abnormal values according to the distance values comprise the following steps:
calculating all the linked distance characteristic values, and obtaining a time abnormal value according to the distance characteristic values;
calculating the distances between all different links, and obtaining a space abnormal value according to the distances between the links;
obtaining the space-time outlier according to the intersection of the time outlier and the space outlier;
the traffic flow is improved through the abnormal causal graph and the normal causal graph;
judging the relation between nodes according to the urban area node diagram and the flow data, wherein the step of obtaining a normal causal diagram comprises the following steps:
aggregating the input flow and the output flow of each node to obtain a flow time sequence;
the flow time sequence is processed through feature extraction and difference to obtain a stable flow sequence;
and carrying out the Granger causal inspection on the stable flow sequence to obtain the relation between the nodes and obtain the normal causal graph.
2. The method for detecting the cause and effect of abnormal traffic flow according to claim 1, wherein the step of dividing the urban area into areas to obtain a node map of the urban area comprises the steps of:
dividing the urban area by adopting a regular hexagon to obtain a node diagram of the urban area;
each of the nodes represents an area;
the links between the nodes represent the traffic from region to region.
3. The method for detecting the cause and effect of abnormal traffic flow according to claim 1, wherein the constructing the link between nodes according to the urban area node map and the traffic data comprises:
judging whether traffic interaction exists between the nodes according to the traffic data, and if yes, establishing a link between two groups of nodes with traffic interaction.
4. The method according to claim 1, wherein calculating the distance eigenvalues of all the links and obtaining the time outliers according to the distance eigenvalues comprises:
calculating the average value of the flow distance between all time frames in each link and any other time frame to obtain the average value of the distance between each time frame and other time frames;
and screening out extreme values from all the distance average values, and taking the extreme values as the time abnormal values.
5. The method according to claim 1, wherein calculating distances between all of the links, and obtaining a spatial outlier based on the distances between the links comprises:
calculating flow distance values between any two groups of links under the same time frame to obtain flow distance values between each link and all other links under each time frame;
and screening out extreme values from all the flow distance values, and taking the extreme values as the space abnormal values.
6. The method of claim 5, wherein the filtering the extremum from all the traffic distance values, and taking the extremum as the spatial outlier comprises:
normalizing all the flow distance values;
and screening out an extremum from the normalized flow distance value, and taking the extremum as the space abnormal value.
7. The method for causal detection of abnormal traffic flow according to claim 6, wherein said extracting and differencing said time series of traffic flows to obtain a steady traffic sequence comprises:
extracting features of the flow time sequence by adopting a non-negative matrix decomposition method;
and carrying out the differential processing on the fitting combination of the non-negative matrix factorization by the flow data to obtain the stable flow sequence.
8. A traffic anomaly traffic cause and effect detection device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of a traffic anomaly traffic cause and effect detection method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310160350.3A CN115984077B (en) | 2023-02-24 | 2023-02-24 | Traffic abnormal flow causal detection method and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310160350.3A CN115984077B (en) | 2023-02-24 | 2023-02-24 | Traffic abnormal flow causal detection method and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115984077A CN115984077A (en) | 2023-04-18 |
| CN115984077B true CN115984077B (en) | 2023-06-13 |
Family
ID=85970759
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310160350.3A Active CN115984077B (en) | 2023-02-24 | 2023-02-24 | Traffic abnormal flow causal detection method and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115984077B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111179592A (en) * | 2019-12-31 | 2020-05-19 | 合肥工业大学 | Urban traffic prediction method and system based on spatio-temporal data flow fusion analysis |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102110365B (en) * | 2009-12-28 | 2013-11-06 | 日电（中国）有限公司 | Road condition prediction method and road condition prediction system based on space-time relationship |
| EP2590151A1 (en) * | 2011-11-03 | 2013-05-08 | Deutsche Telekom AG | A framework for the systematic study of vehicular mobility and the analysis of city dynamics using public web cameras |
| US20130166188A1 (en) * | 2011-12-21 | 2013-06-27 | Microsoft Corporation | Determine Spatiotemporal Causal Interactions In Data |
| CN103258427B (en) * | 2013-04-24 | 2015-03-11 | 北京工业大学 | Urban expressway traffic real-time monitoring system and method based on information physical network |
| CN105023044B (en) * | 2015-07-21 | 2017-10-24 | 清华大学 | Traffic flow causality method for digging based on plenty of time sequence |
| GB201711408D0 (en) * | 2016-12-30 | 2017-08-30 | Maxu Tech Inc | Early entry |
| CN110111575B (en) * | 2019-05-16 | 2020-10-27 | 北京航空航天大学 | Urban traffic flow network analysis method based on complex network theory |
| CN111242521B (en) * | 2020-04-24 | 2020-07-31 | 中国人民解放军国防科技大学 | Track anomaly detection method and system |
| CN115204478A (en) * | 2022-06-30 | 2022-10-18 | 北京航空航天大学 | Public traffic flow prediction method combining urban interest points and space-time causal relationship |
-
2023
- 2023-02-24 CN CN202310160350.3A patent/CN115984077B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111179592A (en) * | 2019-12-31 | 2020-05-19 | 合肥工业大学 | Urban traffic prediction method and system based on spatio-temporal data flow fusion analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115984077A (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Dodge et al. | Revealing the physics of movement: Comparing the similarity of movement characteristics of different types of moving objects | |
| Mancas et al. | Abnormal motion selection in crowds using bottom-up saliency | |
| US20130286198A1 (en) | Method and system for automatically detecting anomalies at a traffic intersection | |
| CN105160313A (en) | Method and apparatus for crowd behavior analysis in video monitoring | |
| Piciarelli et al. | Surveillance-oriented event detection in video streams | |
| Wang et al. | An improved robust principal component analysis model for anomalies detection of subway passenger flow | |
| CN113011322B (en) | Detection model training method and detection method for monitoring specific abnormal behavior of video | |
| CN111008337A (en) | Deep attention rumor identification method and device based on ternary characteristics | |
| CN110084201A (en) | A kind of human motion recognition method of convolutional neural networks based on specific objective tracking under monitoring scene | |
| CN110032952A (en) | A kind of road boundary point detecting method based on deep learning | |
| Hu et al. | Parallel spatial-temporal convolutional neural networks for anomaly detection and location in crowded scenes | |
| Jingying | A survey on crowd counting methods and datasets | |
| CN114202803A (en) | Multi-stage human body abnormal action detection method based on residual error network | |
| Shen et al. | An attention-based digraph convolution network enabled framework for congestion recognition in three-dimensional road networks | |
| Jin et al. | A visual analytics system for improving attention-based traffic forecasting models | |
| Huang et al. | Enabling versatile analysis of large scale traffic video data with deep learning and HiveQL | |
| CN112101132B (en) | Traffic condition prediction method based on graph embedding model and metric learning | |
| CN117456730A (en) | Traffic flow prediction method based on self-adaptive partial attention spread graph convolutional neural network | |
| CN115984077B (en) | Traffic abnormal flow causal detection method and equipment | |
| Hu et al. | Detecting socially abnormal highway driving behaviors via recurrent graph attention networks | |
| Thakare et al. | Object interaction-based localization and description of road accident events using deep learning | |
| Tan et al. | Vehicle detection using YOLO and mobility tracking during COVID-19 pandemic lockdowns | |
| Priyadharshini et al. | Vehicle data aggregation from highway video of madurai city using convolution neural network | |
| CN113516850A (en) | Pipeline traffic flow data acquisition method based on space syntactic analysis | |
| CN110738692A (en) | spark cluster-based intelligent video identification 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 |