1. A Bayesian network-based traffic health diagnosis method is characterized by comprising the following steps:

equally dividing the measurement day into a plurality of time segments;

acquiring traffic flow speeds of all road sections in each time period;

constructing a traffic operation network at each moment in a time period according to the traffic flow speed of each road section in each time period to obtain a first road network;

acquiring a first threshold value of traffic jam;

screening congested roads in the first road network according to the traffic congestion first threshold value to obtain a second road network;

calculating the probability that two roads in the second road network belong to a congestion sub-group, and recording the probability as a first probability, wherein the congestion sub-group is a set formed by congestion roads which can be mutually communicated;

constructing a congestion association weight matrix among roads according to the first probability, wherein the rows and the columns of the weight matrix represent the roads, and the matrix elements are values of the first probability;

performing dimensionality reduction and clustering on the weight matrix to obtain multiple types of congestion modes;

acquiring the characteristics of each type of congestion mode;

constructing a Bayesian network with the characteristics of the congestion mode as nodes;

acquiring a second threshold value of the normal congestion mode and a third threshold value of the abnormal congestion mode;

screening a normal congestion mode and an abnormal congestion mode in the plurality of congestion modes according to the second threshold value and the third threshold value;

respectively training the Bayesian network by using data in a normal congestion mode and data in an abnormal congestion mode to obtain a normal congestion Bayesian network and an abnormal congestion Bayesian network;

calculating the joint probability of the characteristics of a plurality of congestion modes when a certain congestion mode is a normal congestion mode according to the normal congestion Bayesian network, and calculating the joint probability of the characteristics of a plurality of congestion modes when the certain congestion mode is an abnormal congestion mode according to the abnormal congestion Bayesian network;

calculating the normal probability and the abnormal probability of the congestion mode according to the combined probability of the characteristics of the congestion modes in the normal congestion mode and the combined probability of the characteristics of the congestion modes in the abnormal congestion mode;

and judging whether the congestion mode is abnormal or not according to the normal probability and the abnormal probability.

2. The Bayesian network-based traffic health diagnosis method according to claim 1, wherein the first road network is a directed network, and the directed network is weighted by taking intersections as nodes, roads as edges, and road-to-vehicle flow velocities as edges.

3. The bayesian network based traffic health diagnostic method according to claim 2, wherein the traffic speed is a relative speed.

4. The Bayesian network-based traffic health diagnosis method according to claim 3, wherein a formula r is adopted_ij＝v_ij/v_i ^95％Calculating a relative velocity, wherein r_ijIndicating the relative speed, v, of the road flow i at time j_ijRepresenting the real speed, v, observed by the road flow i at time j_i ^95％And the corresponding speed of 95% of quantiles of the all-day speed distribution of the road traffic flow i is shown.

5. The bayesian network-based traffic health diagnosis method according to claim 1, wherein after the step of "obtaining the traffic flow speed of each road segment in each time period", before the step of "constructing the traffic operation network at each time in each time period according to the traffic flow speed of each road segment in each time period" to obtain the first road network ", the method further comprises: and carrying out data compensation on the traffic flow speed of the missing road section.

6. The bayesian network-based traffic health diagnosis method according to claim 1, wherein the step of screening congested roads in the first road network according to the traffic congestion first threshold obtains the second road network specifically as: and obtaining a second road network by removing the roads with the traffic flow speed larger than the traffic jam first threshold value in the first road network.

7. The Bayesian network-based traffic health diagnosis method according to claim 1, wherein a formula is adoptedCalculating a first probability, wherein CC (t)_k) Represents t_kCongestion sub-cluster at time, t_k|i，j∈CC(t_k) Indicating i, j road at t_kThe time belongs to a certain congestion sub-cluster, T represents the time length of the observation period, N (T)_k|i，j∈CC(t_k) I, j links belong to a certain congestion subgroup.

8. The Bayesian network-based traffic health diagnosis method as recited in claim 1, wherein the reducing and clustering the weight matrix to obtain a plurality of congestion patterns comprises:

performing column dimensionality reduction on the weight matrix by adopting a principal component analysis method, wherein each row of the matrix after the dimensionality reduction represents the characteristics of a congested road;

and clustering the features of the congested roads by adopting a K-means clustering algorithm to obtain a plurality of congestion modes.

9. The bayesian network-based traffic health diagnosis method according to claim 1, wherein said obtaining the characteristics of each type of congestion mode includes:

constructing a sub-network by taking roads in the same congestion mode as edges and taking intersections as nodes;

and calculating the node degree mean value, the node degree variance, the betweenness and the clustering coefficient of the sub-network to obtain the characteristics of the congestion mode.

10. The bayesian network-based traffic health diagnosis method according to claim 1, wherein the normal congestion mode and the abnormal congestion mode are divided according to a range of traffic congestion.

Background

With the rapid development of economy in China, the reserved quantity of private automobiles and the annual driving mileage are remarkably increased; however, along with the deepening of the urbanization process, the urban traffic system faces the serious challenges that traffic infrastructure is seriously congested and cannot fully exert the service capability of the traffic infrastructure, and whether the traffic system is healthy or not increasingly affects the urban operation efficiency and the quality of life of residents; the urban traffic system is increasingly complex due to the expansion of urban scale and the diversity of travel demands in cities, so that the search for the source of the congestion problem becomes increasingly troublesome, and a lot of difficulties are brought to managers in cities and users of road networks. Therefore, how to diagnose the health problem of the traffic system, locking the cause of the traffic fault (congestion) becomes a key task.

The traffic health diagnosis is to realize fault separation by utilizing various state information and prior knowledge in a traffic system and identify the position of a fault source causing the service capability reduction of a traffic network; the failure of the traffic system is represented as congestion, and the generation of congestion can be mainly divided into internal accidents and external threats according to the source of congestion, wherein the internal accidents can be from technical failures, traffic accidents or other errors and events caused in the traffic system. External threats include extreme weather effects, terrorist activities and even war. Whether from internal accidents or external threats, congestion is generated and transmitted by the internal organization structure of the traffic system. Traditional research focuses on identification of key roads, and on the basis of congestion propagation relations among different connected edges (road sections) in a traffic network, roads with strong congestion propagation capacity on surrounding roads are mined, bottlenecks of existing traffic systems are identified through the road sections, and the roads are considered as causes of traffic congestion, so that service capacity of the urban traffic system is restricted, and health of the urban traffic system is affected.

However, urban traffic congestion has the characteristics of dynamic propagation, cascade failure and multiple faults, and the complex topological structure and the internal nonlinear dynamic characteristics of a traffic system also bring congestion uncertainty. The existing traffic diagnosis method is often applied to scenes with single fault sources and influences among road sections, and the characteristics of structural coupling, regional propagation and congestion block association of traffic congestion make it difficult to effectively realize fault location analysis by only analyzing congestion propagation strength among roads.

The method comprises the steps of searching a stable congestion coupling area from a congestion sub-group in a congestion form, identifying a congestion mode by using a Bayesian network correlation theory and combining methods such as complex network analysis, establishing a traffic congestion area characteristic Bayesian diagnosis network to infer an abnormal congestion mode, and finally realizing the traffic health diagnosis based on the Bayesian network.

The invention aims to provide theoretical and technical guidance for urban traffic health diagnosis by combining a complex network theory and a Bayesian network technology, thereby providing support for urban traffic congestion management, intelligent traffic construction and other contents.

Disclosure of Invention

The invention aims to provide a Bayesian network-based traffic health diagnosis method, which can realize urban traffic health diagnosis.

In order to achieve the purpose, the invention provides the following scheme:

a Bayesian network-based traffic health diagnosis method comprises the following steps:

equally dividing the measurement day into a plurality of time segments;

acquiring traffic flow speeds of all road sections in each time period;

constructing a traffic operation network at each moment in a time period according to the traffic flow speed of each road section in each time period to obtain a first road network;

acquiring a first threshold value of traffic jam;

screening congested roads in the first road network according to the traffic congestion first threshold value to obtain a second road network;

calculating the probability that two roads in the second road network belong to a congestion sub-group, and recording the probability as a first probability, wherein the congestion sub-group is a set formed by congestion roads which can be mutually communicated;

constructing a congestion association weight matrix among roads according to the first probability, wherein the rows and the columns of the weight matrix represent the roads, and the matrix elements are values of the first probability;

performing dimensionality reduction and clustering on the weight matrix to obtain multiple types of congestion modes;

acquiring the characteristics of each type of congestion mode;

constructing a Bayesian network with the characteristics of the congestion mode as nodes;

acquiring a second threshold value of the normal congestion mode and a third threshold value of the abnormal congestion mode;

screening a normal congestion mode and an abnormal congestion mode in the plurality of congestion modes according to the second threshold value and the third threshold value;

respectively training the Bayesian network by using data in a normal congestion mode and data in an abnormal congestion mode to obtain a normal congestion Bayesian network and an abnormal congestion Bayesian network;

calculating the joint probability of the characteristics of a plurality of congestion modes when a certain congestion mode is a normal congestion mode according to the normal congestion Bayesian network, and calculating the joint probability of the characteristics of a plurality of congestion modes when the certain congestion mode is an abnormal congestion mode according to the abnormal congestion Bayesian network;

calculating the normal probability and the abnormal probability of the congestion mode according to the combined probability of the characteristics of the congestion modes in the normal congestion mode and the combined probability of the characteristics of the congestion modes in the abnormal congestion mode;

and judging whether the congestion mode is abnormal or not according to the normal probability and the abnormal probability.

Optionally, the first road network is a directed network, and the directed network takes intersections as nodes, roads as connecting edges, and road vehicle flow speed as edge weight.

Optionally, the traffic flow speed is a relative speed.

Optionally, using the formula r_ij＝v_ij/v_i ^95％Calculating a relative velocity, wherein r_ijIndicating the relative speed, v, of the road flow i at time j_ijRepresenting the real speed, v, observed by the road flow i at time j_i ^95％And the corresponding speed of 95% of quantiles of the all-day speed distribution of the road traffic flow i is shown.

Optionally, after the step of "obtaining the traffic flow speed of each road segment in each time period", before the step of "constructing a traffic operation network at each moment in each time period according to the traffic flow speed of each road segment in each time period to obtain a first road network", the method further includes: and carrying out data compensation on the traffic flow speed of the missing road section.

Optionally, screening congested roads in the first road network according to the traffic congestion first threshold to obtain a second road network specifically includes: and obtaining a second road network by removing the roads with the traffic flow speed larger than the traffic jam first threshold value in the first road network.

Optionally, using formulasCalculating a first probability, wherein CC (t)_k) Represents t_kCongestion sub-cluster at time, t_k|i，j∈CC(t_k) Indicating i, j road at t_kThe time belongs to a certain congestion sub-cluster, T represents the time length of the observation period, N (T)_k|i，j∈CC(t_k) I, j links belong to a certain congestion subgroup.

Optionally, the performing dimension reduction and clustering on the weight matrix to obtain a plurality of congestion modes includes:

performing column dimensionality reduction on the weight matrix by adopting a principal component analysis method, wherein each row of the matrix after the dimensionality reduction represents the characteristics of a congested road;

and clustering the features of the congested roads by adopting a K-means clustering algorithm to obtain a plurality of congestion modes.

Optionally, the obtaining the characteristics of each type of congestion mode includes:

constructing a sub-network by taking roads in the same congestion mode as edges and taking intersections as nodes;

and calculating the node degree mean value, the node degree variance, the betweenness and the clustering coefficient of the sub-network to obtain the characteristics of the congestion mode.

Optionally, the normal congestion mode and the abnormal congestion mode are divided according to the range of traffic congestion.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects:

according to the method, the congestion mode is identified by using a Bayesian network correlation theory and combining methods such as complex network analysis, the Bayesian network is established for the traffic congestion area characteristic, the abnormal congestion mode is deduced, and finally the traffic health diagnosis based on the Bayesian network is realized.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

Fig. 1 is a flow chart of a bayesian network-based traffic health diagnosis method according to the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

The invention aims to provide a Bayesian network-based traffic health diagnosis method and system, which can realize urban traffic health diagnosis.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

The invention comprises the following main steps:

step 1: and dividing time intervals, collecting road speed data, and constructing a traffic operation time sequence network.

Step 2: and (3) extracting a congestion network according to a congestion threshold value based on the traffic running time sequence network obtained in the step (1), and counting the probability that two roads belong to a congestion communication sub-group.

And step 3: and (3) constructing a congestion association weight matrix between the roads based on the probability data of the congestion subgroups of the two roads of the same genus and the same genus counted in the step (2), and mining the congestion mode.

And 4, step 4: and (3) extracting the characteristics of the congestion mode in the step (3), and identifying the abnormal congestion mode by constructing a Bayesian network with the characteristics as nodes.

Specifically, the traffic health diagnosis method based on the bayesian network disclosed by the invention specifically comprises the following steps as shown in fig. 1:

step 101: the measurement day is divided equally into a plurality of time periods.

The method specifically comprises the following steps: one day is divided into L time segments with the same time span: for a specific date of research, 24 hours a day is equally divided into L time periods with equal duration, and the step is actually to carry out coarse granulation treatment on the traffic network to some extent, so that the relationship between roads with the same follow-up statistical congestion mode is facilitated.

Step 102: and acquiring the traffic flow speed of each road section in each time period.

The method specifically comprises the following steps: for each time segment in step 101, road traffic flow velocity data is collected: and collecting traffic flow speed data information of each road section in a time period. And performing data compensation on the data with the missing data. The specific compensation method is supported by the known technology and literature in the traffic field, and is not described herein again.

Step 103: and constructing a traffic operation network at each moment in the time period according to the traffic flow speed of each road section in each time period to obtain a first road network.

The method specifically comprises the following steps: for each time period in the step 101, a traffic operation network at each moment in the time period is constructed by using the traffic flow speed collected in the step 102: for a researched traffic system, a first road network is constructed by taking intersections as nodes, roads as connecting edges and traffic flow speed as edge weight, wherein the first road network is a directed network; wherein the speed data of the vehicle flow is relative speed, and is obtained by normalizing the speed of 95 percent quantile points of the speed of the road all day, and the speed data is specifically expressed as r_ij＝v_ij/v_i ^95％Wherein r is_ijIndicating the relative speed, v, of the road flow i at time j_ijRepresenting the real speed, v, observed by the road flow i at time j_i ^95％And the corresponding speed of 95% of quantiles of the all-day speed distribution of the road traffic flow i is shown.

Step 104: a first threshold value of traffic jam is obtained.

Step 105: and screening congested roads in the first road network according to the traffic congestion first threshold value to obtain a second road network.

The method specifically comprises the following steps: for the directed network constructed in step 103, extracting a congestion network according to a congestion threshold: and for the constructed directed networks at different moments, setting a first threshold sigma of traffic jam, removing the roads with the speed greater than the jam threshold, then analyzing the connectivity of the rest networks, and counting the inclusion relation between the jam connected sub-groups and the roads.

Here, "removing" a road refers to setting the road to a non-connectable state; "connected" here means that when any two adjacent roads are not removed, the two adjacent areas are in a connected state; "adjacent" here refers to other roads that coincide with a certain road passing start or end point; the "connected sub-cluster" refers to a set composed of connected roads, each road in the set is directly connected with another road in the set or indirectly connected through a transitive connection relationship, and any road in the set does not have any direct or indirect connection relationship with the road outside the set. For example, if road a is in communication with road B whose end point coincides, and road B is in communication with road C whose start point coincides, road a is in indirect communication with road C; if the road A, B, C no longer has direct or indirect connectivity with other roads, the three roads form a connected sub-cluster.

Step 106: and calculating the probability that two roads in the second road network belong to the same congestion sub-group, and recording the probability as a first probability, wherein the congestion sub-group is a set formed by congestion roads which can be mutually communicated.

The method specifically comprises the following steps: counting the probability that two roads belong to the same congestion communication sub-group: counting the number of times that two roads belong to the same congestion communication sub-group at different moments in a target time period by using the inclusion relation between the congestion communication sub-group and the roads obtained by the statistics in the step 105, wherein the ratio of the number of times to the time period length is the probability P that two roads belong to the same congestion communication sub-group_ijI.e. the first probability.

Wherein the content of the first and second substances,wherein CC (t)_k) Represents t_kCongestion sub-cluster at time, t_k|i，j∈CC(t_k) Indicating i, j road at t_kThe time belongs to a certain congestion sub-cluster, T represents the time length of the observation period, N (T)_k|i，j∈CC(t_k) I, j links belong to a certain congestion subgroup.

Step 107: and constructing an inter-road congestion association weight matrix according to the first probability, wherein the rows and the columns of the weight matrix represent roads, and the matrix elements are values of the first probability.

The method specifically comprises the following steps: constructing an inter-road congestion association weight matrix W based on statistical probability data of congestion subgroups of the same genus and the same category of two roads: the rows and columns in the weight matrix represent roads, and the matrix elements represent congestion association weights among different roads, wherein the congestion association weights refer to the probability of congestion subgroups of the same genus and the same category of the different roads, namely the value of a first probability; the association weight between the road and the road itself is set to 0, which is to highlight the degree of congestion association between different roads.

Step 108: and performing dimensionality reduction and clustering on the weight matrix to obtain multiple congestion modes.

The method specifically comprises the following steps: and (3) carrying out congestion mode excavation: firstly, Principal Component Analysis (PCA) dimension reduction (column) is carried out on the weight matrix obtained in the step 107, and each row of the matrix after dimension reduction represents the characteristics of a congested road. And (4) carrying out dimensionality reduction on the high-dimensional data by extracting the main characteristic components of the data by PCA. Here we choose the concept of maximum variance, use less data dimension to retain more original data characteristics (the variance of the reduced sample is as large as possible), and the best projection direction with the maximum data variance in the new space is formed by M eigenvectors u of the covariance matrix S₁，u₂，…u_kDefinitions which correspond to the M maximum eigenvalues, respectively, whereinx_nIn order to be the D-dimensional data,is x_nThe mean, N, is the number of samples, thus reconstructing the k-dimensional features based on the original N-dimensional features. The specific process of PCA dimension reduction is supported by the known technology and literature in the field of computational science, and is not described herein in detail.

Secondly, based on the congestion characteristics, road clustering is carried out by using a K-means clustering algorithm (K-means), and a congestion mode is excavated. For sample x_nDefining a clustering label r_nIf x_nBelongs to the kth cluster, then r_nk1, otherwise r_nk0; the purpose of the K-means algorithm is to minimize the intra-group distance J.

Where N represents the number of samples, K represents the number of clusters, μ_kRepresents the clustering center of class k, | x_n-μ_k||²Represents a sample x_nAnd cluster center mu_kThe distance between them. K-means clustering needs to include three elements: clustering number, clustering center and clustering termination condition; for a given congestion feature data, the process is: according to the given clustering number, randomly appointing initial clustering centers, calculating Euclidean distances between data and each clustering center, distributing the clustering centers according to the distances, wherein the distribution results and the clustering centers represent a cluster, recalculating various clustering centers according to the data in the clustering results in the next distribution, repeating the iteration process until a termination condition is met, namely the position distance fluctuation of two rounds before and after the clustering centers is less than a threshold value delta, and finishing the clustering. The specific process of K-means clustering is supported by the known technology and literature in the field of computational science, and is not described herein any more.

Step 109: and acquiring the characteristics of each type of congestion mode.

The method specifically comprises the following steps: extracting features of the multiple types of congestion patterns in step 108: calculating congestion characteristics such as the number of roads in the same congestion mode (clustering result), the node degree mean, the node degree variance, the betweenness, the clustering coefficient and the like of the sub-networks formed by the roads according to the road clustering result in the step 108; the road forming sub-network is a network formed by taking roads in the same congestion mode as edges and taking intersections as nodes, the degree is the number of the connecting edges of a certain node in the network, the node degree mean value is the mean value of the node degrees in the sub-network, and the node degree variance is the mean value of the square value of the difference value between each node degree in the sub-network and the node degree mean value.Wherein N is the number of network nodes, X is the variable of node degree, mu is the mean value of node degree, sigma²Is the node degree variance. The point betweenness is the ratio of the number of shortest paths passing through a certain node in the network to the number of all shortest paths in the network, B_i＝∑_{j，k≠im∈N}n_jk(i)/n_jkWherein n is_jkIndicates the number of shortest paths between nodes j, k, n_jk(i) Represents the number of shortest paths between nodes j, k through node i, B_iDisplay sectionThe betweenness of the points i, the betweenness of the sub-networks is the mean value of the betweenness of the points, and the clustering coefficient is the ratio of the number of the closed triplets in the sub-networks to the number of all the triplets.

Step 110: and constructing a Bayesian network with the characteristics of the congestion mode as nodes.

The method specifically comprises the following steps: firstly, a Bayesian network with the characteristics of the congestion mode as nodes is constructed, and the Bayesian network learning comprises the following steps: directed acyclic network structure, conditional probability distribution; given congestion pattern feature data, the above "network structure" is specifically a dependency topology relationship between features of congestion patterns, which can be learned based on a score search method, where we select a minimum description length MDL as a bayesian structure score criterion, MDL (G, D) — logP (D | G, θ) + r (G) log (N)/2, where D denotes data, G denotes a network structure, N denotes the number of training data, θ denotes a parameter to be estimated by a bayesian model under a given structure G, and r (G) is a free variable in the structure G. Then searching a dependent topological relation between the characteristics of the network structure with the highest MDL score as a congestion mode by using a heuristic search algorithm, such as a simulated annealing algorithm, a hill climbing method and the like; the "conditional probability distribution" refers to a conditional probability between nodes estimated by maximum likelihood according to a bayesian network structure.

Step 111: and acquiring a second threshold value of the normal congestion mode and a third threshold value of the abnormal congestion mode.

The method specifically comprises the following steps: and judging a normal congestion mode or an abnormal congestion mode according to experience or a congestion range, wherein the normal congestion mode or the abnormal congestion mode is judged, for example, the abnormal congestion mode is judged when the distance is more than 10 kilometers, and the normal congestion mode is judged when the distance is less than 5 kilometers.

Step 112: and screening a normal congestion mode and an abnormal congestion mode in the plurality of congestion modes according to the second threshold value and the third threshold value.

The method specifically comprises the following steps: according to the model in step 110 and the characteristics in step 109, the congestion mode is judged through the threshold in step 111, and data of one type including the abnormal congestion mode and the characteristics thereof, and data of another type including the normal congestion mode and the characteristics thereof, that is, data of the normal congestion mode and data of the abnormal congestion mode are generated.

Step 113: and training the Bayesian network by respectively utilizing the data of the normal congestion mode and the data of the abnormal congestion mode to obtain the normal congestion Bayesian network and the abnormal congestion Bayesian network.

The process of training the Bayesian network is to perform a structure and parameter learning process of the Bayesian network.

Step 114: and calculating the joint probability of the characteristics of the multiple congestion modes when a certain congestion mode is a normal congestion mode according to the normal congestion Bayesian network, and calculating the joint probability of the characteristics of the multiple congestion modes when the certain congestion mode is an abnormal congestion mode according to the abnormal congestion Bayesian network.

The method specifically comprises the following steps: calculating a plurality of feature joint probabilities of the congestion mode in the normal or abnormal congestion mode based on the learned normal congestion bayesian network and abnormal congestion bayesian network, wherein the probabilities represent feature condition probabilities under abnormal or normal congestion conditions and are represented as P (D | C), C represents a state of the congestion mode, C ═ 1 is abnormal, and C ═ 0 is normal; d is represented as a set of congestion features in step 109.

Step 115: and calculating the normal probability and the abnormal probability of the congestion mode according to the combined probability of the characteristics of the congestion modes in the normal congestion mode and the combined probability of the characteristics of the congestion modes in the abnormal congestion mode.

The method specifically comprises the following steps: and deducing the abnormal probability of the congestion mode and the corresponding normal or abnormal state by Bayesian inference P (C | D) ═ P (C) P (D | C)/P (D) · P (C) P (D | C) and P (D | C). The two calculation results correspond to C ═ 1 and C ═ 0, where C ═ 1 indicates an abnormal probability and C ═ 0 indicates a normal probability.

Step 116: and judging whether the congestion mode is abnormal or not according to the normal probability and the abnormal probability.

The method specifically comprises the following steps: selecting one of the normal jam probability and the abnormal jam probability with higher probability as a final diagnosis result: normal congestion or abnormal congestion, and outputting the corresponding probability. The specific parameter adjustment and training process of bayesian network learning is supported by the known technology and literature in the field of computer science, and is not described herein again.

The invention also discloses the following technical effects:

based on a complex network theory and a Bayesian network learning method, the invention aims to construct a congestion network recognition congestion subgroup through the complex network theory, count congestion association strength among roads, recognize a congestion mode by using an unsupervised clustering algorithm, and infer an abnormal congestion mode through the Bayesian network method, thereby solving the health diagnosis pain point problem of an urban level traffic system.

The invention has the advantages that: firstly, the classical system health diagnosis is mainly based on a congestion propagation mode among road sections, so that a scene with a single fault source and influence among the road sections can be solved, a relatively accurate result can be obtained only by relying on high-quality and long-term data, the influence of the structural coupling, the regional propagation and the congestion block correlation characteristics of traffic congestion on health is difficult to consider, and the complex network theory and Bayesian network learning method can calculate an accurate result only by road network topology information and speed data within a certain time. Secondly, the invention adopts a complex network theory analysis method, can perform modeling description on the characteristics of structural coupling, regional propagation and congestion block association of traffic congestion, and excavates different congestion modes. Finally, the Bayesian network learning method is adopted, the Bayesian network learning capability and the reasoning capability are outstanding, the method can adapt to complex and dynamic external environments, and can deduce and find abnormal modes, namely abnormal road sets instead of simple abnormal roads based on prior experience and observation results, so that the method can be used for diagnosing the urban-level traffic health.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.