Metro network passenger flow anomaly detection method based on GCN-informer and gaussian bayesian model

Abstract

The application discloses a subway line network passenger flow anomaly detection method based on a GCN-informer and a Gaussian Bayesian model, and comprises the following steps: dividing a sample data set into a training set and a validation set without containing anomalies, and a test set containing abnormal and normal samples; constructing a GCN-informer model and training; inputting the test set into the trained GCN-informer model to obtain a predicted output, and then obtaining the reconstruction error of each test sample, and constructing a reconstruction error Gaussian distribution model; constructing a binary classifier for detecting subway line network inbound and outbound passenger flow anomalies, and obtaining the normal and abnormal probabilities of the test sample respectively. The application considers the space-time correlation of the station passenger flow, can realize the synchronous detection of the network level inbound and outbound passenger flow anomalies, is beneficial to the manager to timely master the state of the subway line network operation, and can make timely and effective response to the subway passenger flow anomaly, so as to promote the autonomous safety management level of the subway line network operation.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation information processing technology, and more specifically to a method for detecting abnormal passenger flow in subway networks based on GCN-informer and Gaussian Bayes models. Background Technology

[0002] Passenger flow anomalies refer to sudden increases or decreases in passenger flow at subway stations caused by external disturbances such as line interruptions, station closures, train delays, extreme weather, and large-scale events. In an increasingly complex and frequently disruptive operating environment, passenger flow anomalies can easily affect the safety and reliability of subway network operations. For example, sudden surges in passenger flow at subway stations caused by disturbances can lead to station overload, overcrowding, stampedes, and the failure of train operation scheduling strategies, severely impacting the efficiency of the subway network and passenger safety. Therefore, it is necessary to detect passenger flow anomalies at subway stations and utilize historical passenger flow fluctuation characteristics to achieve real-time monitoring of subway station operational status. This will improve the level of proactive safety management of the subway network and provide strong decision-making support for the formulation and optimization of subway passenger flow organization under disturbance scenarios.

[0003] In practice, identifying abnormal trends in subway passenger flow is typically done semi-manually. Its accuracy and reliability depend heavily on the experience of the inspectors, and it is inefficient and costly. With the continuous development of information technology, automated anomaly detection algorithms have also been applied to detect abnormal passenger flow in subway stations. These algorithms can be broadly categorized into four types: statistical methods, index prediction methods, unsupervised methods, and supervised methods. Statistical methods identify anomalies by utilizing the distribution of abnormal and normal data. Index prediction methods use historical data characteristics to predict data for the next moment, comparing the prediction with the true value to determine anomalies. Unsupervised and supervised methods typically learn from the historical evolution trends of data samples, using the actual evolution trend and model output to determine whether the data sample is abnormal. A common feature of these methods is that they all require defining discrimination rules between abnormal and normal data during anomaly detection. Using the magnitude of the reconstruction error between the true data value and the model's predicted output to determine whether data is abnormal is used in most methods.

[0004] However, in practical applications, these automated anomaly detection methods still have the following problems: (1) The accuracy of detection methods based on reconstruction error thresholds is limited by the rationality of the reconstruction error threshold determination. In actual detection tasks, due to the differences in data distribution, it is difficult to reasonably determine the reconstruction error threshold. (2) There are currently few methods for synchronous detection of passenger flow anomalies at multiple stations in the metro network. Most methods focus on the detection of passenger flow anomalies at a single station, ignoring the spatiotemporal correlation of passenger flow between stations under disturbance scenarios. At the same time, their anomaly detection accuracy is low and their robustness is poor.

[0005] Therefore, how to provide a high-precision method for detecting abnormal passenger flow in the subway network that can take into account the spatiotemporal correlation of passenger flow between stations is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method for detecting abnormal passenger flow in subway networks based on GCN-informer and Gaussian Bayes models, which aims to solve the problems of low accuracy and insufficient consideration of factors in existing detection methods.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for detecting passenger flow anomalies in subway networks based on GCN-informer and Gaussian Bayes models includes the following steps:

[0009] The data to be detected is input into the trained subway network passenger flow anomaly detection system. The subway network passenger flow anomaly detection system includes a GCN-informer model, a reconstruction error Gaussian distribution model, and a binary classifier. The data to be detected is input into the trained GCN-informer model to obtain the predicted output of the data to be detected. The reconstruction error Gaussian distribution model calculates the reconstruction error between the predicted output of the data to be detected and the true value of the data to be detected, and inputs it into the trained binary classifier to obtain the anomaly detection result.

[0010] The construction method of the subway network passenger flow anomaly detection system is as follows:

[0011] S1. Data preprocessing: Based on the text data of subway network disturbance events recorded by IC cards during the time period, determine whether there are any abnormalities in the passenger flow entering and exiting each station. Use the passenger flow entering and exiting each station and its abnormal situations as a sample dataset. Divide the sample dataset into a training set and a validation set that do not contain abnormalities, and a test set that contains abnormal and normal samples.

[0012] S2. Constructing the GCN-informer model: The GCN-informer model includes an improved GCN model and an informer model. The GCN model is improved by replacing the 0-1 adjacency matrix of the original GCN model with a weighted adjacency matrix of passenger flow entering and exiting stations. The improved GCN model is obtained by connecting the output of the improved GCN model and the output of the informer model through a fully connected layer, and the output of the fully connected layer is used as the final output of the GCN-informer model. The parallel structure of the GCN-informer model is trained using the training set and the validation set.

[0013] S3. Construct a Gaussian distribution model for reconstruction error: Input the test set into the trained GCN-informer model to obtain the prediction output, calculate the difference between the prediction output and the true value of the test set to obtain the reconstruction error of each test sample, construct a Gaussian distribution model for reconstruction error, and estimate the parameters of the Gaussian distribution model for reconstruction error.

[0014] S4. Construct a binary classifier for detecting abnormal passenger flow at subway stations: Using the Gaussian distribution of the estimated reconstruction error, construct a binary classifier for detecting abnormal passenger flow at subway stations, and obtain the normal and abnormal probabilities of the test samples respectively.

[0015] Preferably, the GCN model is improved using a passenger flow-weighted adjacency matrix. The improved GCN model is expressed as follows:

[0016]

[0017] H l+1 =f(H l ,M) (2)

[0018] f(H l ,M)=ReLU(MH l W l (3)

[0019] In the formula: M is the passenger flow weighted adjacency matrix; a ij For the elements in the passenger flow weighted adjacency matrix, a ij ∈M; q ij The passenger flow from station i to station j; δ ij Used to determine the adjacency relationship between stations i and j. If station i and station j are adjacent, then δ ij =1; otherwise δ ij =0; H l+1 and H l These are the output of layer l+1 and the input of layer l, respectively; W lHere is the weight matrix to be trained in the l-th layer; ReLU(·) is the non-linear activation function in GCN; f(H l M) is the output of the l-th layer.

[0020] Preferably, the informer model is an encoder-decoder composed of multiple stacked attention mechanisms. The informer model incorporates a ProbSparse self-attention mechanism, represented as follows:

[0021]

[0022] Where Q, K, and V are the query space, key space, and value space of the sparse self-attention mechanism, respectively; It is a sparse matrix with the same dimension as Q. Each element q i The condition satisfied by the value space is:

[0023]

[0024]

[0025] Where KL(q) i K) is a similarity measure between the query subspace and the key distribution. The encoder of the Informer model uses n layers of ProbSparse self-attention distillation to extract features from the original data.

[0026] A single-layer attention distillation structure can be represented as:

[0027]

[0028] in and These represent the output of layer k+1 and the input of layer k, respectively; Conv1d(·) represents 1D convolution; [·] AB This represents the computation of the multi-head ProbSparse self-attention mechanism; ELU(·) is the activation function; MaxPool(·) represents pooling computation.

[0029] Preferably, the final output of the GCN-informer model is:

[0030]

[0031] in Y is the final output of the GCN-informer model; W and b are the parameters to be trained in the fully connected layer; G Y is the output of the GCN model. I This is the output of the Informer model.

[0032] Preferably, the specific content of S3 includes:

[0033] (1) The reconstruction error of the test sample is:

[0034]

[0035] Where Y t Let be the true value of the test sample at time t. The predicted output obtained after inputting the test samples for time period t into the GCN-informer model;

[0036] (2) The Gaussian distribution of the reconstruction error of the test samples is expressed as:

[0037]

[0038]

[0039] Where e t E represents the reconstruction error of the test data over time period t. t E is used to indicate whether a test sample is a normal sample or an abnormal sample. t =1 indicates that the test data in time period t is an outlier sample, E t =0 indicates that the test data in time period t is a normal sample; P(e=e t |E t =0) means that when the test sample is a normal sample, the reconstruction error is e t The probability; σ0 and μ0 are the standard deviation and mean of the sample reconstruction error, respectively; σ1 and μ1 are the standard deviation and mean of the reconstruction error of the normal test sample, respectively; P(e=e t Let e ​​be the reconstruction error of the test sample obtained after merging normal and abnormal samples. t The probability of;

[0040] (3) The parameters in equations (10) and (11) are estimated using the maximum likelihood estimation method:

[0041]

[0042]

[0043]

[0044]

[0045] Where I{·} is an indicator function that returns 1 if the internal condition is true; otherwise, it returns 0.

[0046] Preferably, the specific content of S4 includes:

[0047] Using the Gaussian distribution of the estimated reconstruction error, a Naive Bayes binary classifier is constructed for anomaly detection in passenger flow entering and exiting subway stations. The Naive Bayes binary classifier is expressed as:

[0048]

[0049] P(E t =1|e=e t )=1-P(E t =0|e=e t (17)

[0050] Where P(E) t =0|e=e t ) and P(E t =1|e=e t ) represent the cases where the reconstruction error is e t When the test sample is normal or abnormal, P(e = e) represents the probability of the test sample being normal or abnormal. t |E t =0) means that when the test sample is a normal sample, its reconstruction error is e t The probability of.

[0051] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method for detecting abnormal passenger flow in subway networks based on GCN-informer and Gaussian Bayes models, which specifically includes the following beneficial effects:

[0052] (1) The present invention can effectively capture the spatiotemporal correlation of passenger flow entering and exiting the subway network under disturbance scenarios, and realize high-precision synchronous detection of abnormal passenger flow entering and exiting the station;

[0053] (2) This invention uses a Gaussian Bayes model to construct a binary classifier for anomaly detection, which can effectively solve the problem that the reconstruction error threshold is difficult to determine in traditional detection methods and can greatly improve the accuracy of anomaly detection.

[0054] (3) This invention introduces text data of subway events to determine whether the original data sample is abnormal, which can provide strong information support for the reliability and robustness of the anomaly detection model. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0056] Figure 1A flowchart illustrating the subway network passenger flow anomaly detection method based on GCN-informer and Gaussian Bayes model provided by this invention.

[0057] Figure 2 A schematic diagram of the structure of the GCN-informer model in the subway network passenger flow anomaly detection method based on GCN-informer and Gaussian Bayes model provided by the present invention;

[0058] Figure 3 A schematic diagram of the structure of the Informer model in the subway network passenger flow anomaly detection method based on GCN-informer and Gaussian Bayes model provided by the present invention;

[0059] Figure 4 A schematic diagram of the encoder of the subway network passenger flow anomaly detection method based on GCN-informer and Gaussian Bayes model provided by the present invention.

[0060] Figure 5 A map showing subway stations and routes in the study area provided for embodiments of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] This invention discloses a method for detecting passenger flow anomalies in subway networks based on GCN-informer and Gaussian Bayes models, such as... Figure 1 As shown, it includes the following steps:

[0063] The data to be detected is input into the trained subway network passenger flow anomaly detection system. The subway network passenger flow anomaly detection system includes a GCN-informer model, a reconstruction error Gaussian distribution model, and a binary classifier. The data to be detected is input into the trained GCN-informer model to obtain the predicted output of the data to be detected. The reconstruction error Gaussian distribution model calculates the reconstruction error between the predicted output of the data to be detected and the true value of the data to be detected, and inputs it into the trained binary classifier to obtain the anomaly detection result.

[0064] The construction method of the subway network passenger flow anomaly detection system is as follows:

[0065] S1. Data Preprocessing: Based on the text data of subway network disturbance events recorded by IC cards, determine whether there are any abnormalities in the passenger flow entering and exiting each station. Use the passenger flow entering and exiting each station and its abnormal situations as a sample dataset. Divide the sample dataset into a training set and a validation set that do not contain abnormalities, and a test set that contains abnormal and normal samples.

[0066] S2. Constructing the GCN-informer model: The GCN-informer model includes an improved GCN model and an informer model. The original GCN model's 0-1 adjacency matrix is ​​replaced with a weighted adjacency matrix based on passenger flow to obtain the improved GCN model. A fully connected layer connects the outputs of the improved GCN model and the informer model, and the output of the fully connected layer is used as the final output of the GCN-informer model. The parallel structure of the GCN-informer model is trained using training and validation sets. The specific structure of the GCN-informer model is as follows: Figure 2 As shown.

[0067] S3. Construct a Gaussian distribution model for reconstruction error: Input the test set into the trained GCN-informer model to obtain the predicted output. Calculate the difference between the predicted output and the true value of the test set to obtain the reconstruction error of each test sample. Construct a Gaussian distribution model for reconstruction error and estimate the parameters of the Gaussian distribution model for reconstruction error.

[0068] S4. Construct a binary classifier for detecting abnormal passenger flow at subway stations: Using the Gaussian distribution of the estimated reconstruction error, construct a binary classifier for detecting abnormal passenger flow at subway stations, and obtain the normal and abnormal probabilities of the test samples respectively.

[0069] To further implement the above technical solution, the GCN model is improved using a passenger flow-weighted adjacency matrix. The improved GCN model is expressed as follows:

[0070]

[0071] H l+1 =f(H l ,M) (2)

[0072] f(H l ,M)=ReLU(NH l W l (3)

[0073] In the formula: M is the passenger flow weighted adjacency matrix; a ij For the elements in the passenger flow weighted adjacency matrix, a ij ∈M; q ij The passenger flow from station i to station j; δ ijUsed to determine the adjacency relationship between stations i and j. If station i and station j are adjacent, then δ ij =1; otherwise δ ij =0; H l+1 and H l These are the output of layer l+1 and the input of layer l, respectively; W l Here is the weight matrix to be trained in the l-th layer; ReLU(·) is the non-linear activation function in GCN; f(H l M) is the output of the l-th layer.

[0074] To further implement the above technical solution, the Informer model is an encoder-decoder composed of multiple stacked attention mechanisms, such as... Figure 3 As shown, to address the issues of model complexity, limitations on input layer feature length, and a sharp drop in output speed, a ProbSparse self-attention mechanism is introduced into the Informer model, and is represented as:

[0075]

[0076] Where Q, K, and V are the query space, key space, and value space of the sparse self-attention mechanism, respectively; It is a sparse matrix with the same dimension as Q. Each element q i The regulation satisfied by the value space is:

[0077]

[0078]

[0079] Where KL(q) i K) is a similarity measure between the query subspace and the key distribution. The encoder of the Informer model uses n layers of ProbSparse self-attention distillation to extract features from the original data.

[0080] The Informer model's encoder uses multi-layer ProbSparse self-attention distillation to extract features from the original data, thereby reducing the dimensionality of long-sequence input features and addressing the issue of limited input feature length. Its structure is as follows: Figure 4 As shown. The single-layer attention distillation structure can be represented as:

[0081]

[0082] in and These represent the output of layer k+1 and the input of layer k, respectively; Conv1d(·) represents 1D convolution; [·] ABThis represents the computation of the multi-head ProbSparse self-attention mechanism; ELU(·) is the activation function; MaxPool(·) represents pooling computation.

[0083] In this embodiment, the decoder of the Informer model uses a two-layer stacked ProbSparse self-attention mechanism, where the input to the second attention layer includes the output of the first attention layer and the output of the encoder. The final output of the decoder is the output of a fully connected layer that takes the output of the second attention layer as input.

[0084] To further implement the above technical solution, the final output of the GCN-informer model is:

[0085]

[0086] in Y is the final output of the GCN-informer model; W and b are the parameters to be trained in the fully connected layer; G Y is the output of the GCN model. I This is the output of the Informer model.

[0087] To further implement the above technical solution, the specific content of S3 includes:

[0088] (1) The reconstruction error of the test sample is:

[0089]

[0090] Where Y t Let be the true value of the test sample at time t. The predicted output obtained after inputting the test samples for time period t into the GCN-informer model;

[0091] (2) The Gaussian distribution of the reconstruction error of the test samples is expressed as:

[0092]

[0093]

[0094] Where e t E represents the reconstruction error of the test data over time period t. t E is used to indicate whether a test sample is a normal sample or an abnormal sample. t =1 indicates that the test data in time period t is an outlier sample, E t =0 indicates that the test data in time period t is a normal sample; P(e=e t |E t =0) means that when the test sample is a normal sample, the reconstruction error is e tThe probability; σ0 and μ0 are the standard deviation and mean of the sample reconstruction error, respectively; σ1 and μ1 are the standard deviation and mean of the reconstruction error of the normal test sample, respectively; P(e=e t Let e ​​be the reconstruction error of the test sample obtained after merging normal and abnormal samples. t The probability of;

[0095] (3) The parameters in equations (10) and (11) are estimated using the maximum likelihood estimation method:

[0096]

[0097]

[0098]

[0099]

[0100] Where I{·} is an indicator function that returns 1 if the internal condition is true; otherwise, it returns 0.

[0101] To further implement the above technical solution, the specific content of S4 includes:

[0102] Using the Gaussian distribution of the estimated reconstruction error, a Naive Bayes binary classifier is constructed for anomaly detection in passenger flow entering and exiting subway stations. The Naive Bayes binary classifier is expressed as:

[0103]

[0104] P(E t =1|e=e t )=1-P(E t =0|e=e t (17)

[0105] Where P(E) t =0|e=e t ) and P(E t =1|e=e t ) represent the cases where the reconstruction error is e t When the test sample is normal or abnormal, P(e = e) represents the probability of the test sample being normal or abnormal. t |E t =0) means that when the test sample is a normal sample, its reconstruction error is e t The probability of.

[0106] S5. Perform anomaly detection on the test data: Input the test data into the trained GCN-informer model to obtain the predicted output. Calculate the reconstruction error of the test data using the predicted output and the ground truth of the test data, and input it into the binary classifier built in S4 to obtain the anomaly detection result.

[0107] The invention will be further illustrated below through specific examples:

[0108] In this embodiment, the analysis focuses on the subway IC card data and corresponding event text data of City A from September 1, 2019 to September 30, 2019. The IC card data used includes data records from 120 stations on 12 lines. The subway network topology is as follows: Figure 5 As shown, textual data from subway incidents revealed 12 days of disruption events between September 1st and 30th. Therefore, 18 days of IC card records without disruptions were used as the foundation for passenger flow sample extraction in both the training and validation sets, while IC card records containing disruption data were used for passenger flow sample extraction in the test set. Furthermore, the time window for extracting passenger flow in and out of the station was 5 minutes.

[0109] In this embodiment, two stacked GCN models and one Informer model are used. The outputs of the corresponding models are concatenated using a fully connected layer to obtain the final output of the GCN-informer model. The specific parameters used in the model are shown in Table 1.

[0110] Table 1. Parameters of the GCN-informer model

[0111]

[0112]

[0113] This invention is used to calculate the reconstruction error of each test sample, and the parameters of the Gaussian distribution of the reconstruction error are estimated using the maximum likelihood estimation method.

[0114] A binary classifier for anomaly detection was constructed, and the performance of the anomaly detection model of the present invention was evaluated using the benchmark model and parameters shown in Table 2. The evaluation results of the model performance are shown in Table 2.

[0115] Table 2 Model Performance Evaluation

[0116]

[0117]

[0118] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0119] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting passenger flow anomalies in subway networks based on GCN-informer and Gaussian Bayes models, characterized in that, Includes the following steps: The data to be detected is input into the trained subway network passenger flow anomaly detection system, which includes a GCN-informer model, a reconstruction error Gaussian distribution model, and a binary classifier. The data to be detected is input into the trained GCN-informer model to obtain the predicted output of the data to be detected. The reconstruction error Gaussian distribution model calculates the reconstruction error between the predicted output of the data to be detected and the true value of the data to be detected, and inputs it into the trained binary classifier to obtain the anomaly detection result. The construction method of the subway network passenger flow anomaly detection system is as follows: S1. Data preprocessing: Based on the text data of subway network disturbance events recorded by IC cards during the time period, determine whether there are any abnormalities in the passenger flow entering and exiting each station. Use the passenger flow entering and exiting each station and its abnormal situations as a sample dataset. Divide the sample dataset into a training set and a validation set that do not contain abnormalities, and a test set that contains abnormal and normal samples. S2. Constructing the GCN-informer model: The GCN-informer model includes an improved GCN model and an informer model. The GCN model is improved by replacing the 0-1 adjacency matrix of the original GCN model with a weighted adjacency matrix of passenger flow entering and exiting stations. The improved GCN model is obtained by connecting the output of the improved GCN model and the output of the informer model through a fully connected layer, and the output of the fully connected layer is used as the final output of the GCN-informer model. The GCN-informer model is trained using the training set and the validation set. The GCN model is improved by using a passenger flow-weighted adjacency matrix. The improved GCN model is expressed as follows: (1)； (2)； (3)； In the formula: i and j are the index numbers of the passenger's starting and ending stations. A weighted adjacency matrix for passenger flow; For the elements in the passenger flow weighted adjacency matrix, ; For the site to station The number of passengers; For the site to station The number of passengers; To the originating station Sum all departing passenger flows; To terminate the site Sum all departing passenger flows; Used to determine the site and The adjacency relationship of the stations and sites Adjacent, then ;otherwise ; and The first Layer output and Layer input; For the first The weight matrix that the layer needs to train; This refers to the nonlinear activation function in GCN; For the first Layer output; S3. Construct a Gaussian distribution model for reconstruction error: Input the test set into the trained GCN-informer model to obtain the prediction output, calculate the difference between the prediction output and the true value of the test set to obtain the reconstruction error of each test sample, construct a Gaussian distribution model for reconstruction error, and estimate the parameters of the Gaussian distribution model for reconstruction error. S4. Construct a binary classifier for detecting abnormal passenger flow at subway stations: Using the Gaussian distribution of the estimated reconstruction error, construct a binary classifier for detecting abnormal passenger flow at subway stations, and obtain the normal and abnormal probabilities of the test samples respectively.

2. The method for detecting abnormal passenger flow in subway networks based on GCN-informer and Gaussian Bayes models according to claim 1, characterized in that, The informer model is an encoder-decoder composed of multiple stacked attention mechanisms. The informer model introduces a ProbSparse self-attention mechanism, which is represented as follows: (4)； Among them, SP() is the ProbSparse self-attention mechanism. and These are the query matrix, key matrix, and value matrix in the ProbSparse self-attention mechanism, respectively. This is the sparse query matrix obtained by filtering from Q based on the query sparsity metric. Each element The conditions that the bond matrix must satisfy are: ； ； in To query the similarity measure of subspaces and key distributions, the encoder of the Informer model uses... Layer ProbSparse self-attentional distillation is used to extract features from raw data. A single-layer attention distillation structure can be represented as: (7)； in and They are respectively Layer output and Layer input; Represents 1D convolution; This represents the computation of the multi-head ProbSparse self-attention mechanism; For activation functions; This indicates pooling computation.

3. The method for detecting abnormal passenger flow in subway networks based on GCN-informer and Gaussian Bayes models according to claim 1, characterized in that, The final output of the GCN-informer model is: (8)； in This is the final output of the GCN-informer model; and These are the parameters to be trained for the fully connected layer; The output of the GCN model, This is the output of the Informer model.

4. The method for detecting abnormal passenger flow in subway networks based on GCN-informer and Gaussian Bayes models according to claim 1, characterized in that, The specific content of S3 includes: (1) The reconstruction error of the test sample is: （9）； in for The true value of the test sample over the time period. for The predicted output obtained after inputting the time period test samples into the GCN-informer model; (2) The Gaussian distribution of the reconstruction error of the test samples is expressed as: (10)； (11)； in for Reconstruction error of test data over a time period; Used to indicate whether a test sample is a normal sample or an abnormal sample. express The test data for that time period were outlier samples. express The test data for the specified time period represents normal samples; When the test sample is a normal sample, the reconstruction error is: The probability of; and These are the standard deviation and mean of the sample reconstruction error distribution, respectively; and These are the standard deviation and mean of the reconstruction error distribution for the normal test samples, respectively. The reconstruction error of the test sample obtained after merging normal and abnormal samples is taken as... The probability of; (3) The parameters in equations (10) and (11) are estimated using the maximum likelihood estimation method: (12)； (13)； (14)； (15)； in It is an indicator function that returns 1 if the internal condition is true, otherwise it returns 0.

5. The method for detecting abnormal passenger flow in subway networks based on GCN-informer and Gaussian Bayes models according to claim 4, characterized in that, The specific content of S4 includes: Using the Gaussian distribution of the estimated reconstruction error, a Naive Bayes binary classifier is constructed for anomaly detection in passenger flow entering and exiting subway stations. The Naive Bayes binary classifier is expressed as: (16)； (17)； in and When the reconstruction error is respectively The probability that a test sample is normal or abnormal at that time; When the test sample is a normal sample, its reconstruction error is: The probability of.

Images