Emergency event handling auxiliary method and system for urban rail transit driving

Abstract

The invention discloses an emergency event handling auxiliary method and system for urban rail transit driving, and the method comprises the steps: obtaining the operation monitoring information of a target station, carrying out the time alignment and space alignment of the operation monitoring information of the target station, and obtaining the operation monitoring information of the aligned target station; event feature detection and analysis are carried out based on the aligned target station operation monitoring information, and multi-modal fusion features of the target station operation monitoring information are constructed; establishing a fuzzy rule database, and performing event level analysis and event influence analysis on the current event according to the multi-modal fusion features of the operation monitoring information of the target site to obtain event level analysis information and emergency event influence analysis information; and constructing an emergency event disposal auxiliary model, and inputting the emergency event influence analysis information and the event level analysis information into the emergency event disposal auxiliary model to generate an emergency event disposal scheme for recommendation. And the real-time performance and reliability of emergency event handling are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic emergency event handling, and in particular to an emergency event handling auxiliary method and system for urban rail transit operation. Background Art

[0002] As a crucial component of modern urban public transportation systems, urban rail transit's safety and operational efficiency are directly linked to the convenience of urban residents and socioeconomic stability. However, due to the complexity of rail transit systems and the uncertainty of the external environment, train operations can be impacted by factors such as equipment failures, sudden natural disasters, and abnormal passenger flows, triggering various emergency events. These events are often sudden, contagious, and multi-factor coupled. If not handled properly, they can not only disrupt the normal operation of trains but also cause significant economic losses and casualties.

[0003] Existing rail transit emergency response primarily relies on manual judgment and pre-set rule-based emergency plans. However, this approach presents problems such as delayed response and reliance on experience for decision-making. In actual operation, multi-source data, such as train operating status, platform passenger flow information, track equipment monitoring data, environmental perception data, and video surveillance information, can provide a wealth of event characteristics. However, how to effectively integrate this heterogeneous data to build an intelligent, automated emergency event detection and decision support system is a key focus of current technological research. Therefore, how to combine the actual operational characteristics of rail transit, fuse multi-source data for event detection and decision optimization, and build an efficient and reliable emergency event handling auxiliary method to improve the safety and operational efficiency of rail transit systems remains an urgent research issue. Summary of the Invention

[0004] The present invention overcomes the defects of the prior art and provides an emergency event handling auxiliary method and system for urban rail transit operation.

[0005] To achieve the above-mentioned objectives, the present invention provides, in a first aspect, an auxiliary method for handling emergency incidents in urban rail transit, comprising:

[0006] Acquiring target site operation monitoring information, performing time alignment and spatial alignment processing on the target site operation monitoring information, and obtaining aligned target site operation monitoring information;

[0007] Performing event feature detection and analysis based on the aligned target site operation monitoring information, and constructing a multimodal fusion feature of the target site operation monitoring information;

[0008] Establish a fuzzy rule database, conduct event level analysis and event impact analysis on the current event based on the multimodal fusion characteristics of the target site operation monitoring information, and obtain event level analysis information and emergency event impact analysis information;

[0009] An emergency event handling auxiliary model is constructed based on the MOEA / D algorithm, and the emergency event impact analysis information and event level analysis information are input into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation.

[0010] In this solution, the acquisition of target site operation monitoring information, performing time alignment and spatial alignment processing on the target site operation monitoring information, and obtaining aligned target site operation monitoring information specifically includes:

[0011] Acquire target site operation monitoring information based on a multi-source heterogeneous data acquisition module, wherein the target site operation monitoring information includes site equipment operation information, site environment monitoring information, site video monitoring information, site passenger flow monitoring information, and site vehicle monitoring information;

[0012] Using a global clock synchronization protocol to perform time correction processing on the target site operation monitoring information, extracting the operation monitoring data with time differences based on the time correction results, and setting a time compensation mechanism based on a sliding window algorithm;

[0013] Acquire the operation monitoring data of the previous moment and the next moment of the operation monitoring data with time difference to form a time difference operation monitoring data sequence, and divide the time difference operation monitoring data sequence into sliding windows of fixed length based on a preset time compensation;

[0014] In each window, the time deviation between each data source and the global clock is calculated. The average time deviation of the calculated data points in the window is used to compensate the timestamps of all data in the current window to obtain timing alignment information.

[0015] Inputting the timing alignment information into a Kalman filter for dynamic estimation, generating a state equation and an observation equation based on the dynamic estimation result, and performing real-time clock deviation estimation through the state equation and the observation equation to suppress the influence of network jitter;

[0016] Obtaining a line distribution map of the target site and setting a number of reference marking points based on the line distribution map and establishing a line engineering coordinate system, measuring the reference points to obtain corresponding absolute geographic coordinates and relative coordinates in the line engineering coordinate system, forming coordinate pairs for each reference point;

[0017] The centroid of each reference point is determined by using the least squares method through the coordinates of each reference point and a covariance matrix is constructed, and a singular value decomposition algorithm is introduced to decompose the covariance matrix to obtain a rotation matrix;

[0018] The translation vector is calculated by combining the rotation matrix and the center of mass, and a coordinate transformation model is constructed. A Kalman filter is introduced into the coordinate transformation model to reduce the coordinate conversion error of the dynamic equipment. The operation monitoring information of the target site is spatially aligned through the optimized coordinate change model.

[0019] In this solution, the event feature detection and analysis based on the aligned target site operation monitoring information and the construction of multimodal fusion features of the target site operation monitoring information specifically include:

[0020] An anomaly detection model is constructed using the isolation forest algorithm as a framework. The aligned target site operation monitoring information is obtained and input into the anomaly detection model to analyze the anomalies of the operating equipment within the site.

[0021] Randomly select the data input into the anomaly detection model to obtain a number of sample data, use the RANSAC linear fitting algorithm to perform linear fitting on the selected sample data, and calculate the difference between the actual observation value and the fitted prediction value to generate residual features;

[0022] Extracting features from the input target site operation monitoring information to generate an original feature matrix, fusing the residual features with the original feature matrix to generate a fused feature matrix, setting the total number of isolated forests based on the fused feature matrix, and constructing a feature space;

[0023] Using a random binary search tree to search for isolated points in the feature space and calculate the average path length of each isolated point, defining isolated points smaller than a preset threshold as outliers and generating anomaly detection information;

[0024] The target station's passenger flow monitoring features and station video monitoring features are extracted from the target station's operation monitoring information after time alignment and spatial alignment. A passenger flow intensity heat map of the target station is generated based on the target station's passenger flow monitoring features. The spatiotemporal graph convolutional network is used to detect abnormal behaviors based on the station video monitoring features to generate abnormal behavior detection information.

[0025] Combining the anomaly detection information, the passenger flow intensity heat map and the abnormal behavior detection information, a multimodal feature fusion is performed through a weighted fusion method to obtain a multimodal fusion feature of the target site operation monitoring information.

[0026] In this solution, a fuzzy rule database is established to perform event level analysis and event impact analysis on the current event based on the multimodal fusion features of the target site operation monitoring information, specifically including:

[0027] Based on data retrieval, several historical emergency events are obtained, event features of the historical emergency events are extracted, an attention mechanism is introduced to calculate the attention score of each event feature, and each event feature is weighted by the attention score to generate an event feature matrix;

[0028] Preset emergency event assessment rules, perform grade assessment on each historical emergency event using the emergency event assessment rules to obtain grade assessment results, and associate the grade assessment results with the event feature matrix to obtain associated result information;

[0029] Preset a membership function, calculate the membership of the event features corresponding to each historical emergency event through the membership function, and construct a historical emergency event data set in combination with the association result information;

[0030] The Apriority algorithm is introduced to mine fuzzy rules through the historical emergency event data set, the minimum membership degree of each fuzzy rule is defined as the matching degree, and a fuzzy rule database is established;

[0031] Building an event level analysis model based on a dual-channel neural network, the event level analysis model includes a temporal channel and a spatial channel, wherein the temporal channel is constructed by an LSTM network and the spatial channel is constructed by a CNN network;

[0032] The event level analysis model is deep-learned and trained by constructing a training data set through the fuzzy rule database, and a loss function is preset to balance the event classification. After the training is completed, the model parameters are retained to output the final event level analysis model;

[0033] The multimodal fusion features of the target site operation monitoring information are obtained and input into the trained event level analysis model for analysis to determine the event level of the target site under the current real-time monitoring status and obtain event level analysis information.

[0034] In this solution, the establishment of a fuzzy rule database and the performance of event level analysis and event impact analysis on the current event based on the multimodal fusion features of the target site operation monitoring information also include:

[0035] Obtaining an infrastructure distribution map and a line distribution map of the target site, and extracting facility distribution features and line distribution features of the target site based on the infrastructure distribution map and the line distribution map of the target site to obtain first feature information;

[0036] Abstracting the rail transit network of the target site into a space-time propagation graph model according to the first feature information, wherein the space-time propagation graph model includes an entity node dimension, an edge dimension, and a time dimension;

[0037] Acquiring event level analysis information, extracting event features of the emergency event at the current moment based on the event level analysis information, initializing nodes of the spatiotemporal propagation graph model based on the extracted event features, and defining abnormal nodes and node states;

[0038] The state of each node at any time is regarded as a discrete variable. The propagation dynamics equation is used to model the change of abnormal node state over time. The propagation equation is solved by the RK4 method to obtain the state change trend of each node within the preset time and generate node state change information.

[0039] Calculating a network vulnerability index and a cascading failure risk index based on the node state change information, wherein the network vulnerability index is represented by the importance of the node, and the cascading failure risk is represented by the ratio of the nodes with the current fault level to the total nodes;

[0040] Emergency event impact analysis information is generated by combining the node state change information, the network vulnerability index and the cascading failure risk index.

[0041] In this solution, the emergency event handling auxiliary model is constructed based on the MOEA / D algorithm, and the emergency event impact analysis information and event level analysis information are input into the emergency event handling auxiliary model to generate an emergency event handling solution for recommendation, specifically including:

[0042] An emergency response auxiliary model is constructed based on the MOEA / D algorithm to obtain emergency impact analysis information and event level analysis information. The objective function is set to minimize passenger delay time, operational economic losses, and response energy consumption as the optimization goal, and constraints are set.

[0043] Inputting the emergency event impact analysis information and event level analysis information into the emergency event handling auxiliary model for analysis, decomposing the optimization target to generate uniformly distributed weights, and initializing the population based on the input data to obtain an initial population;

[0044] Calculating the objective function value of each independent individual in the initial population by using a preset objective function, and screening elite individuals in the initial population according to the objective function value corresponding to each independent individual;

[0045] Calculate the Euclidean distance between each weight vector, compare the calculated Euclidean distance with the preset threshold, select the weight vectors smaller than the preset threshold to build a neighborhood, and randomly select several individuals from the neighborhood to perform crossover and mutation operations with the elite individuals;

[0046] Obtain the new individual after crossover and mutation operations, calculate the objective function value of the new individual in the neighborhood, determine whether it is the optimal individual in the neighborhood, and if it is the optimal individual, replace the neighborhood individual to generate a new neighborhood population for iterative update;

[0047] When the maximum number of iterations is reached or the Pareto front converges, the optimal solution set is output, and a number of candidate emergency event handling plans are generated based on the optimal solution set. The characteristics of the handling plan are extracted for each candidate emergency event handling plan, including delay time, operating loss and handling energy consumption;

[0048] The candidate emergency response plans are sorted according to their characteristics, and three ranking tables are generated and pushed to provide auxiliary suggestions for emergency response.

[0049] A second aspect of the present invention provides an emergency incident handling assistance system for urban rail transit operation, the system comprising: a memory and a processor, the memory containing an emergency incident handling assistance method program for urban rail transit operation, the emergency incident handling assistance method program for urban rail transit operation, when executed by the processor, implementing the following steps:

[0050] Acquiring target site operation monitoring information, performing time alignment and spatial alignment processing on the target site operation monitoring information, and obtaining aligned target site operation monitoring information;

[0051] Performing event feature detection and analysis based on the aligned target site operation monitoring information, and constructing a multimodal fusion feature of the target site operation monitoring information;

[0052] Establish a fuzzy rule database, conduct event level analysis and event impact analysis on the current event based on the multimodal fusion characteristics of the target site operation monitoring information, and obtain event level analysis information and emergency event impact analysis information;

[0053] An emergency event handling auxiliary model is constructed based on the MOEA / D algorithm, and the emergency event impact analysis information and event level analysis information are input into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation.

[0054] The present invention discloses an auxiliary method and system for handling emergency events in urban rail transit, comprising the following steps: obtaining target station operation monitoring information, performing time alignment and spatial alignment processing on the target station operation monitoring information to obtain aligned target station operation monitoring information; performing feature extraction and analysis based on the aligned target station operation monitoring information, and constructing a multimodal fusion feature of the target station operation monitoring information based on the analysis results; establishing a fuzzy rule database, performing event level analysis and event impact analysis on the current event based on the multimodal fusion feature of the target station operation monitoring information to obtain event level analysis information and emergency event impact analysis information; and constructing an auxiliary model for handling emergency events, inputting the emergency event impact analysis information and event level analysis information into the auxiliary model to generate and recommend emergency event handling plans. This improves the real-time and reliability of emergency event handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or exemplary descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to these drawings without paying any creative work.

[0056] Figure 1 A flow chart of an auxiliary method for handling emergency incidents in urban rail transit provided by one embodiment of the present invention;

[0057] Figure 2 A flow chart of a method for analyzing urban rail transit emergency events provided by one embodiment of the present invention;

[0058] Figure 3 A block diagram of an emergency event handling assistance system for urban rail transit operation provided by one embodiment of the present invention;

[0059] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0060] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0062] Figure 1 A flow chart of an auxiliary method for handling emergency incidents in urban rail transit provided by one embodiment of the present invention;

[0063] like Figure 1 As shown, the present invention provides a flow chart of an auxiliary method for handling emergency events in urban rail transit operation, including:

[0064] S102, acquiring target site operation monitoring information, performing time alignment and spatial alignment processing on the target site operation monitoring information, and obtaining aligned target site operation monitoring information;

[0065] S104, performing event feature detection and analysis based on the aligned target site operation monitoring information, and constructing a multimodal fusion feature of the target site operation monitoring information;

[0066] S106, establishing a fuzzy rule database, performing event level analysis and event impact analysis on the current event based on the multimodal fusion features of the target site operation monitoring information, and obtaining event level analysis information and emergency event impact analysis information;

[0067] S108, constructing an emergency event handling auxiliary model based on the MOEA / D algorithm, inputting the emergency event impact analysis information and the event level analysis information into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation.

[0068] Furthermore, in a preferred embodiment of the present invention, the step of obtaining target site operation monitoring information and performing time alignment and spatial alignment processing on the target site operation monitoring information to obtain aligned target site operation monitoring information specifically includes:

[0069] Acquire target site operation monitoring information based on a multi-source heterogeneous data acquisition module, wherein the target site operation monitoring information includes site equipment operation information, site environment monitoring information, site video monitoring information, site passenger flow monitoring information, and site vehicle monitoring information;

[0070] Using a global clock synchronization protocol to perform time correction processing on the target site operation monitoring information, extracting the operation monitoring data with time differences based on the time correction results, and setting a time compensation mechanism based on a sliding window algorithm;

[0071] Acquire the operation monitoring data of the previous moment and the next moment of the operation monitoring data with time difference to form a time difference operation monitoring data sequence, and divide the time difference operation monitoring data sequence into sliding windows of fixed length based on a preset time compensation;

[0072] In each window, the time deviation between each data source and the global clock is calculated. The average time deviation of the calculated data points in the window is used to compensate the timestamps of all data in the current window to obtain timing alignment information.

[0073] Inputting the timing alignment information into a Kalman filter for dynamic estimation, generating a state equation and an observation equation based on the dynamic estimation result, and performing real-time clock deviation estimation through the state equation and the observation equation to suppress the influence of network jitter;

[0074] Obtaining a line distribution map of the target site and setting a number of reference marking points based on the line distribution map and establishing a line engineering coordinate system, measuring the reference points to obtain corresponding absolute geographic coordinates and relative coordinates in the line engineering coordinate system, forming coordinate pairs for each reference point;

[0075] The centroid of each reference point is determined by using the least squares method through the coordinates of each reference point and a covariance matrix is constructed, and a singular value decomposition algorithm is introduced to decompose the covariance matrix to obtain a rotation matrix;

[0076] The translation vector is calculated by combining the rotation matrix and the center of mass, and a coordinate transformation model is constructed. A Kalman filter is introduced into the coordinate transformation model to reduce the coordinate conversion error of the dynamic equipment. The operation monitoring information of the target site is spatially aligned through the optimized coordinate change model.

[0077] It should be noted that during rail transit operation monitoring, due to the presence of time stamp deviations between different data sources, and the time synchronization accuracy of each monitoring device is affected by factors such as network latency and data transmission mechanisms, a global clock synchronization protocol is used to time-correct the operation monitoring information of the target station to ensure data consistency and accuracy. This time correction process extracts operation monitoring data with time discrepancies, and a time compensation mechanism is established based on a sliding window algorithm to mitigate data alignment issues caused by time deviations. During the time compensation process, the previous and next moments of the time-discrepant data are obtained to form a time-difference data sequence. This sequence is then divided into sliding windows of fixed duration according to a preset time compensation strategy. Within each sliding window, the time deviation of each data source relative to the global clock is calculated. The average time deviation is then calculated using the data points within the window. This is then used to compensate the timestamps of all data within the window, ultimately generating time-series alignment information. To further mitigate time errors caused by network jitter or data drift, the time-series alignment information is fed into a Kalman filter for dynamic estimation. The state equation and observation equation are constructed, and clock deviation estimation is performed through real-time calculation of these two equations. The introduction of a Kalman filter effectively reduces data noise and improves the stability and accuracy of temporal alignment, thereby providing higher-quality time series data for subsequent spatial alignment and analysis. Regarding spatial alignment, to ensure consistency of monitoring data across different coordinate systems, an engineering coordinate system for the rail transit line is established, and several benchmark markers are set based on a line distribution map of the target stations. These benchmarks are measured to obtain their absolute geographic coordinates and their relative coordinates in the engineering coordinate system, thereby forming coordinate pairs for each benchmark point. Based on this, the centroid of the benchmark points is calculated using the least squares method, and a covariance matrix is constructed to evaluate the transformation relationship between the different coordinate systems. To optimize the accuracy of the coordinate transformation, a singular value decomposition algorithm is introduced to decompose the covariance matrix, obtaining a rotation matrix. This is then combined with the centroid to calculate the translation vector to construct a complete coordinate transformation model. Given that rail transit operation monitoring information includes data from dynamic equipment, traditional static coordinate transformation methods may be subject to error. Therefore, a Kalman filter is introduced into the coordinate transformation model to reduce the coordinate transformation error of dynamic equipment and improve the accuracy of spatial alignment. Finally, through the optimized coordinate transformation model, the operation monitoring information of the target site is spatially aligned to ensure that the information of each data source can be integrated and analyzed under a unified time and space benchmark, providing accurate data support for intelligent monitoring and emergency response of rail transit.

[0078] It should be noted that station equipment operation information primarily refers to the status data of key equipment within the station, including rail transit signaling systems, power supply systems, braking systems, elevators and escalators, platform screen doors, and station ventilation systems. Station environmental monitoring information is primarily used to assess the impact of environmental factors both inside and outside the station on rail transit operations. This information includes temperature and humidity sensor data, air quality monitoring data, noise levels, light intensity, and fire and smoke detection. Station video monitoring information primarily comes from the station's surveillance camera system, which uses video streams to monitor the operating status of areas such as platforms, waiting areas, train cars, and entrances and exits. This is primarily used for passenger behavior analysis, security incident detection, and emergency incident tracing. Station passenger flow monitoring information is crucial data for rail transit operations management, primarily covering station passenger flow statistics, passenger entry and exit data, transfer volume, and platform and car congestion. This information is typically provided by equipment such as automatic ticket vending systems and smart cameras. Station train operation monitoring information primarily monitors train operating status, including train arrival and departure times, current speed, stop duration, and train load factor. This type of data is usually obtained by the train automatic operation system, train positioning system, trackside equipment, etc.

[0079] Furthermore, in a preferred embodiment of the present invention, the event feature detection and analysis based on the aligned target site operation monitoring information and the construction of multimodal fusion features of the target site operation monitoring information specifically include:

[0080] An anomaly detection model is constructed using the isolation forest algorithm as a framework. The aligned target site operation monitoring information is obtained and input into the anomaly detection model to analyze the anomalies of the operating equipment within the site.

[0081] Randomly select the data input into the anomaly detection model to obtain a number of sample data, use the RANSAC linear fitting algorithm to perform linear fitting on the selected sample data, and calculate the difference between the actual observation value and the fitted prediction value to generate residual features;

[0082] Extracting features from the input target site operation monitoring information to generate an original feature matrix, fusing the residual features with the original feature matrix to generate a fused feature matrix, setting the total number of isolated forests based on the fused feature matrix, and constructing a feature space;

[0083] Using a random binary search tree to search for isolated points in the feature space and calculate the average path length of each isolated point, defining isolated points smaller than a preset threshold as outliers and generating anomaly detection information;

[0084] The target station's passenger flow monitoring features and station video monitoring features are extracted from the target station's operation monitoring information after time alignment and spatial alignment. A passenger flow intensity heat map of the target station is generated based on the target station's passenger flow monitoring features. The spatiotemporal graph convolutional network is used to detect abnormal behaviors based on the station video monitoring features to generate abnormal behavior detection information.

[0085] Combining the anomaly detection information, the passenger flow intensity heat map and the abnormal behavior detection information, a multimodal feature fusion is performed through a weighted fusion method to obtain a multimodal fusion feature of the target site operation monitoring information.

[0086] It should be noted that during real-time site monitoring, after performing spatiotemporal alignment on the monitoring data, event recognition is performed based on the processed data to determine whether an emergency event has occurred. An anomaly detection model is constructed using the Isolation Forest algorithm to detect anomalous data from the daily operation of site equipment at the target site. Isolation Forest is an unsupervised anomaly detection algorithm based on decision trees. It isolates outliers to detect operational data that does not conform to normal patterns. Based on this, the RANSAC (Random Sample Consensus) linear fitting algorithm is introduced for optimization. RANSAC iteratively selects subsets from the dataset and calculates linear fit results to identify and remove outliers, thereby ensuring model robustness. Residual features are generated by calculating the difference between the actual observed values of the selected samples and the fitted predicted values. This supplements the information dimension of the original data and improves the accuracy of subsequent feature analysis. Subsequently, feature extraction is performed on the target site's operational monitoring information to generate a raw feature matrix. This matrix contains data from various sensors and monitoring systems, including equipment status, environmental information, and passenger flow. The residual features are combined with the original feature matrix to form a fused feature matrix. The total number of isolated forests is set based on the fused feature matrix to generate anomaly detection information. Key information is further extracted through the passenger flow monitoring features and station video monitoring features of the target station. First, based on the passenger flow monitoring data, the passenger flow distribution in different time periods and different areas is calculated, and a passenger flow intensity heat map of the target station is generated to intuitively display the passenger flow density and distribution trends. At the same time, based on the station video monitoring features, a spatiotemporal graph convolutional network (ST-GCN) is used for abnormal behavior detection. This network uses spatiotemporal relationship modeling to analyze the human behavior patterns in the video data and detect whether there are abnormal events such as falls, collisions with screen doors, or illegal intrusions into the track area, thereby generating abnormal behavior detection information.

[0087] Furthermore, during the multimodal feature fusion process, for the discrete anomaly marker sequence generated by anomaly detection information, the frequency of anomaly triggering for each device within a set time period is counted using a sliding time window, and the device anomaly index is calculated as the initial weight factor for that modality. For the two-dimensional spatial distribution matrix generated by the passenger flow intensity heat map, the deviation between the peak passenger flow density in each area and the historical data for the same period is extracted, and the spatial weight coefficient is calculated in combination with the maximum platform load constraint. The behavior detection weight is directly mapped to the anomaly probability value output by the spatiotemporal graph convolutional network, and the sigmoid function is used to convert the behavior anomaly confidence as the third modality weight. In the feature fusion stage, the two-dimensional matrix of the passenger flow intensity heat map is expanded into a spatial sequence and tensor-joined with the device dimension sequence of the anomaly detection information to form a spatiotemporal joint feature base. The behavior anomaly probability vector is then expanded to the same dimension using a spatial interpolation algorithm and superimposed on the base feature. The fused three-dimensional tensor is input into a hybrid neural network consisting of a dilated convolutional layer and a channel attention module. The dilated convolutional layer extracts cross-region correlation features through a kernel function, while the channel attention module adaptively calibrates the contribution ratio of each modal channel through a squeeze-excitation mechanism, and finally outputs a multimodal fusion feature vector with spatiotemporal perception capabilities, providing an analytical basis for subsequent hierarchical analysis.

[0088] Furthermore, in a preferred embodiment of the present invention, the establishment of a fuzzy rule database and the performance of event level analysis and event impact analysis on the current event based on the multimodal fusion features of the target site operation monitoring information specifically include:

[0089] Based on data retrieval, several historical emergency events are obtained, event features of the historical emergency events are extracted, an attention mechanism is introduced to calculate the attention score of each event feature, and each event feature is weighted by the attention score to generate an event feature matrix;

[0090] Preset emergency event assessment rules, perform grade assessment on each historical emergency event using the emergency event assessment rules to obtain grade assessment results, and associate the grade assessment results with the event feature matrix to obtain associated result information;

[0091] Preset a membership function, calculate the membership of the event features corresponding to each historical emergency event through the membership function, and construct a historical emergency event data set in combination with the association result information;

[0092] The Apriority algorithm is introduced to mine fuzzy rules through the historical emergency event data set, the minimum membership degree of each fuzzy rule is defined as the matching degree, and a fuzzy rule database is established;

[0093] Building an event level analysis model based on a dual-channel neural network, the event level analysis model includes a temporal channel and a spatial channel, wherein the temporal channel is constructed by an LSTM network and the spatial channel is constructed by a CNN network;

[0094] The event level analysis model is deep-learned and trained by constructing a training data set through the fuzzy rule database, and a loss function is preset to balance the event classification. After the training is completed, the model parameters are retained to output the final event level analysis model;

[0095] The multimodal fusion features of the target site operation monitoring information are obtained and input into the trained event level analysis model for analysis to determine the event level of the target site under the current real-time monitoring status and obtain event level analysis information.

[0096] It should be noted that after extracting the multimodal features of the target site, it is necessary to determine whether an emergency event exists under the current circumstances and the level of the emergency event. First, several historical emergency events are retrieved based on data retrieval, and event features are extracted from these historical events. Because different event features have varying degrees of influence on emergency decision-making, an attention mechanism is introduced during the feature extraction process to dynamically adjust the weights of each event feature. The attention mechanism calculates an attention score for each feature and weights the features based on this score to generate an event feature matrix that effectively represents historical emergency events, thereby increasing the focus on key features. Regarding event classification, a pre-defined emergency event assessment rule is used to rank each historical emergency event, generating a rating result. The emergency event assessment rule utilizes data retrieval to obtain different historical cases, thereby setting evaluation indicators and rules, and ultimately obtaining reliable rating results. Subsequently, a fuzzy membership function is introduced to calculate the membership of each historical emergency event feature. Combined with the aforementioned correlation results, a historical emergency event dataset is constructed to more accurately describe the feature distribution of events of different levels. Based on this dataset, the Apriori algorithm was used to mine fuzzy rules to identify potential association rules for various emergency events under different conditions. During the rule mining process, the minimum membership degree of each fuzzy rule was defined as the matching degree to measure the rule's effectiveness and applicability. All valid fuzzy rules were stored in a fuzzy rule database to provide rule support for subsequent event level analysis. To further improve the accuracy of emergency level analysis, a dual-channel neural network model was constructed, consisting of a temporal channel and a spatial channel. The temporal channel was constructed using a long short-term memory (LSTM) network to capture temporal dependencies during event evolution, thereby better understanding the dynamic characteristics of events. The spatial channel was constructed using a convolutional neural network (CNN) to extract spatial feature patterns from event data, thereby enhancing the classification of different emergency events. After training, the model parameters were retained and the final event level analysis model was output, enabling it to classify and evaluate newly occurring emergency events. Finally, multimodal fusion features of the target site's operational monitoring information were obtained and input into the trained event level analysis model for analysis. The model uses the learned feature patterns and classification rules to determine the real-time monitoring status of the current target site, identify the level of events that have occurred, and ultimately generate event level analysis information.

[0097] Furthermore, in a preferred embodiment of the present invention, the step of establishing a fuzzy rule database and performing event level analysis and event impact analysis on the current event based on the multimodal fusion features of the target site operation monitoring information further includes:

[0098] Obtaining an infrastructure distribution map and a line distribution map of the target site, and extracting facility distribution features and line distribution features of the target site based on the infrastructure distribution map and the line distribution map of the target site to obtain first feature information;

[0099] Abstracting the rail transit network of the target site into a space-time propagation graph model according to the first feature information, wherein the space-time propagation graph model includes an entity node dimension, an edge dimension, and a time dimension;

[0100] Acquiring event level analysis information, extracting event features of the emergency event at the current moment based on the event level analysis information, initializing nodes of the spatiotemporal propagation graph model based on the extracted event features, and defining abnormal nodes and node states;

[0101] The state of each node at any time is regarded as a discrete variable. The propagation dynamics equation is used to model the change of abnormal node state over time. The propagation equation is solved by the RK4 method to obtain the state change trend of each node within the preset time and generate node state change information.

[0102] Calculating a network vulnerability index and a cascading failure risk index based on the node state change information, wherein the network vulnerability index is represented by the importance of the node, and the cascading failure risk is represented by the ratio of the nodes with the current fault level to the total nodes;

[0103] Emergency event impact analysis information is generated by combining the node state change information, the network vulnerability index and the cascading failure risk index.

[0104] It should be noted that when an emergency occurs, since the response cannot be immediate, further consideration must be given to the extent of the emergency's impact on the station. The rail transit network at the target station is abstracted into a spatiotemporal propagation graph model. This spatiotemporal propagation graph model comprises three main dimensions: entity node dimension, edge dimension, and time dimension. The entity node dimension represents key infrastructure, track sections, and important transfer points within the station. The edge dimension describes the connections between nodes, reflecting the physical or functional relationships between facilities. The time dimension characterizes the evolution of station status over time, thus supporting dynamic analysis of the event propagation process. After the spatiotemporal propagation graph model is constructed, the emergency event characteristics at the current moment are obtained based on event level analysis information. These event characteristics reflect the specific impact scope, severity, and evolution trend of the emergency event. Based on these characteristics, the nodes of the spatiotemporal propagation graph model are initialized, defining the states of abnormal and normal nodes. Abnormal nodes are critical areas or facilities affected by the current emergency event, whose states may experience abnormal fluctuations, while other nodes not directly affected remain normal at the initial state. Using propagation dynamics equations, the node state change process is modeled, simulating the time-dependent evolution of abnormal node states. This method obtains the state change trends of each node within a preset timeframe and generates node state change information. This information is used to analyze how abnormal states impact the entire rail transit network and predict the development of emergency events. Based on this node state change information, a network vulnerability index and a cascading failure risk index are further calculated. The network vulnerability index measures the stability of the rail transit network under the impact of an emergency event. Its calculation is based on a weighted approach based on node importance. Highly important nodes experience a significant increase in vulnerability index when impacted by an abnormality. The cascading failure risk index measures the impact of an emergency event on the overall network operation. This index is calculated by calculating the ratio of the number of nodes currently affected by a fault to the total number of nodes. If this ratio exceeds a certain threshold, the rail transit network may be at risk of cascading failure. Finally, by integrating the node state change information, the network vulnerability index, and the cascading failure risk index, emergency event impact analysis information is generated, comprehensively assessing the scope of the current emergency event's impact on the rail transit network. Among them, the network vulnerability index, that is, the importance of the node, can be defined by three aspects: the hubness of the node in the network, the ratio of the current node load to the maximum capacity, and the difficulty of repair. The difficulty of repair can be defined by indicators such as idle maintenance personnel and node maintenance area.

[0105] Furthermore, in the modeling of abnormal propagation of rail transit network, we first define the state variable x of each node in the network at time t: i (t)∈(0,1), when x i When (t) is equal to 0, it means the node is completely normal. iWhen (t) is equal to 1, it indicates that the node has completely failed, and the intermediate value represents the degree of partial abnormality. The propagation dynamics equation consists of three parts: the self-recovery term, the neighborhood propagation term, and the external disturbance term. The self-recovery term describes the node's repair ability, the neighborhood propagation term describes the fault diffusion effect between nodes, and the external disturbance term simulates external interference such as sudden passenger flow or equipment impact. Subsequently, the propagation dynamics equation is numerically solved using the RK4 method. The time step is preset and the state of each node is initialized. Then, four sets of slopes are calculated in sequence within each time step. The node state value is updated by weighted average. The iterative calculation is performed until the preset time is reached. The state evolution sequence of each node is output, and the node state change information is generated, thereby describing the affected state of different nodes in the current emergency event.

[0106] The dynamic propagation equation is specifically:

[0107]

[0108] Among them, γ i is the self-recovery rate of node i, such as the maintenance efficiency of equipment, λ i is the node propagation sensitivity coefficient, which is determined by the node type (communication system, ordinary equipment, etc.), A ij is the adjacency matrix element, A ij ∈(0,1),A ij =1 indicates that nodes i and j are physically connected, μ ij is the coupling strength, which is determined by the line load ratio, I e (t) is the external disturbance function, η i is the node's response coefficient to external disturbances.

[0109] Furthermore, in a preferred embodiment of the present invention, the emergency event handling auxiliary model is constructed based on the MOEA / D algorithm, and the emergency event impact analysis information and event level analysis information are input into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation, specifically including:

[0110] An emergency response auxiliary model is constructed based on the MOEA / D algorithm to obtain emergency impact analysis information and event level analysis information. The objective function is set to minimize passenger delay time, operational economic losses, and response energy consumption as the optimization goal, and constraints are set.

[0111] Inputting the emergency event impact analysis information and event level analysis information into the emergency event handling auxiliary model for analysis, decomposing the optimization target to generate uniformly distributed weights, and initializing the population based on the input data to obtain an initial population;

[0112] Calculating the objective function value of each independent individual in the initial population by using a preset objective function, and screening elite individuals in the initial population according to the objective function value corresponding to each independent individual;

[0113] Calculate the Euclidean distance between each weight vector, compare the calculated Euclidean distance with the preset threshold, select the weight vectors smaller than the preset threshold to build a neighborhood, and randomly select several individuals from the neighborhood to perform crossover and mutation operations with the elite individuals;

[0114] Obtain the new individual after crossover and mutation operations, calculate the objective function value of the new individual in the neighborhood, determine whether it is the optimal individual in the neighborhood, and if it is the optimal individual, replace the neighborhood individual to generate a new neighborhood population for iterative update;

[0115] When the maximum number of iterations is reached or the Pareto front converges, the optimal solution set is output, and a number of candidate emergency event handling plans are generated based on the optimal solution set. The characteristics of the handling plan are extracted for each candidate emergency event handling plan, including delay time, operating loss and handling energy consumption;

[0116] The candidate emergency response plans are sorted according to their characteristics, and three ranking tables are generated and pushed to provide auxiliary suggestions for emergency response.

[0117] It should be noted that after analyzing the emergency level and impact of the target station, the emergency needs to be handled. An optimization framework is established using the MOEA / D (Multi-Objective Evolutionary Algorithm with Decomposition) algorithm. With the optimization objectives of minimizing passenger delay time, operational economic losses, and handling energy consumption, this framework assists in the decision-making of emergency response plans. Constraints can include indicators such as power supply capacity safety thresholds, passenger evacuation safety thresholds, and equipment switching delays. By inputting emergency impact analysis information and event level analysis information, a comprehensive description of the current abnormal state of the rail transit system is obtained. Subsequently, the optimization objective is decomposed to generate a uniformly distributed weight vector, which is then initialized to obtain an initial population. The initial population represents multiple different emergency response plans, with each individual corresponding to a possible optimized solution. Next, the objective function value of each individual in the initial population is calculated using a preset objective function. Based on the calculated results, elite individuals are selected from the initial population. Elite individuals are the solutions that perform best in the current population and serve as the core driving force of population evolution in the subsequent evolutionary process, improving optimization results. Subsequently, the Euclidean distance between each weight vector is calculated and compared with a preset threshold. Weight vectors with a value less than the threshold are selected to construct a neighborhood. After constructing the neighborhood, several individuals are randomly selected from the neighborhood and subjected to crossover and mutation operations with elite individuals to generate new individuals. The crossover operation facilitates information exchange, encouraging the new individuals to inherit superior genes, while the mutation operation introduces a certain amount of random perturbation to increase population diversity and prevent the algorithm from falling into a local optimum. For each newly generated individual, its objective function value in the neighborhood is recalculated to determine whether it is the optimal individual in the neighborhood. If the objective function value of the new individual is better than that of an individual in the current neighborhood, the corresponding individual in the neighborhood is replaced with the new individual. The neighborhood population is continuously optimized iteratively, allowing the overall population to gradually approach the Pareto optimal frontier, improving the quality and distribution of solutions, until the maximum number of iterations is reached or the Pareto front converges. At this point, the final optimal solution set is output. Based on the obtained optimal solution set, several candidate emergency event handling plans are generated, and the key features of each candidate plan are extracted, including passenger delay time, operating loss and handling energy consumption. The candidate plans are further ranked, and three types of ranking tables based on delay time, operating loss and handling energy consumption are generated respectively. The ranking results are pushed to relevant management personnel to provide reasonable emergency event handling auxiliary suggestions, thereby improving the response speed and handling efficiency of the rail transit system in emergencies.

[0118] Figure 2 A flow chart of a method for analyzing urban rail transit emergency events provided by one embodiment of the present invention;

[0119] like Figure 2 As shown, the present invention provides a flow chart of a method for analyzing an urban rail transit emergency event, including:

[0120] S202: Monitor the target site to obtain target site operation monitoring information, perform spatiotemporal alignment processing on the target site operation monitoring information, and construct a multimodal fusion feature of the target site operation monitoring information at the current moment based on the spatiotemporal alignment processing result;

[0121] S204: Perform event level analysis and event impact analysis based on the multimodal fusion features to analyze whether an emergency event exists at the target site and determine the impact range of the emergency event, thereby obtaining event level analysis information and emergency event impact analysis information.

[0122] S206: Input the emergency event impact analysis information and event level analysis information into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation.

[0123] It should be noted that a multi-source heterogeneous data acquisition module acquires real-time data on equipment status, environmental monitoring, video images, and passenger flow. A global clock synchronization protocol and sliding window compensation algorithm are used for temporal and spatial alignment. For equipment status data, the system employs an improved isolation forest algorithm for fault identification. Video data is detected using a spatiotemporal graph convolutional network (ST-GCN). Passenger flow data is then fused using a weighted splicing method to integrate multimodal features, calculating the number of ticket checks, platform density, and transfer flow intensity. Initial classification is performed using a fuzzy logic rule base, combined with a dual-channel deep learning model for fine-grained event classification. A spatiotemporal propagation graph model is constructed and the propagation dynamics equation is solved to quantify the impact of the event. The network vulnerability index is calculated using node importance. Subsequently, an emergency response support model is constructed. With the goal of minimizing passenger delays, operational economic losses, and response energy consumption, constraints are established to generate a set of candidate strategies. This set of strategies is then used to assist in emergency response, ensuring the scientific and feasible nature of the emergency response plan.

[0124] Figure 3 An embodiment of the present invention provides an emergency incident handling assistance system 3 for urban rail transit operation. The system includes: a memory 31 and a processor 32. The memory 31 contains an emergency incident handling assistance method program for urban rail transit operation. When the emergency incident handling assistance method program for urban rail transit operation is executed by the processor 32, the following steps are implemented:

[0125] Acquiring target site operation monitoring information, performing time alignment and spatial alignment processing on the target site operation monitoring information, and obtaining aligned target site operation monitoring information;

[0126] Performing event feature detection and analysis based on the aligned target site operation monitoring information, and constructing a multimodal fusion feature of the target site operation monitoring information;

[0127] Establish a fuzzy rule database, conduct event level analysis and event impact analysis on the current event based on the multimodal fusion characteristics of the target site operation monitoring information, and obtain event level analysis information and emergency event impact analysis information;

[0128] An emergency event handling auxiliary model is constructed based on the MOEA / D algorithm, and the emergency event impact analysis information and event level analysis information are input into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation.

[0129] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0130] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0131] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0132] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0133] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0134] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. An auxiliary method for handling emergency incidents in urban rail transit, characterized in that: include: Acquiring target site operation monitoring information, performing time alignment and spatial alignment processing on the target site operation monitoring information, and obtaining aligned target site operation monitoring information; Performing event feature detection and analysis based on the aligned target site operation monitoring information, and constructing a multimodal fusion feature of the target site operation monitoring information; Establish a fuzzy rule database, conduct event level analysis and event impact analysis on the current event based on the multimodal fusion characteristics of the target site operation monitoring information, and obtain event level analysis information and emergency event impact analysis information; An emergency event handling auxiliary model is constructed based on the MOEA / D algorithm, and the emergency event impact analysis information and event level analysis information are input into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation.

2. The method for assisting in handling emergency incidents in urban rail transit according to claim 1, characterized in that: The acquiring of target site operation monitoring information and performing time alignment and space alignment processing on the target site operation monitoring information to obtain aligned target site operation monitoring information specifically includes: Acquire target site operation monitoring information based on a multi-source heterogeneous data acquisition module, wherein the target site operation monitoring information includes site equipment operation information, site environment monitoring information, site video monitoring information, site passenger flow monitoring information, and site vehicle monitoring information; Using a global clock synchronization protocol to perform time correction processing on the target site operation monitoring information, extracting the operation monitoring data with time differences based on the time correction results, and setting a time compensation mechanism based on a sliding window algorithm; Acquire the operation monitoring data of the previous moment and the next moment of the operation monitoring data with time difference to form a time difference operation monitoring data sequence, and divide the time difference operation monitoring data sequence into sliding windows of fixed length based on a preset time compensation; In each window, the time deviation between each data source and the global clock is calculated. The average time deviation of the calculated data points in the window is used to compensate the timestamps of all data in the current window to obtain timing alignment information. Inputting the timing alignment information into a Kalman filter for dynamic estimation, generating a state equation and an observation equation based on the dynamic estimation result, and performing real-time clock deviation estimation through the state equation and the observation equation to suppress the influence of network jitter; Obtaining a line distribution map of the target site and setting a number of reference marking points based on the line distribution map and establishing a line engineering coordinate system, measuring the reference points to obtain corresponding absolute geographic coordinates and relative coordinates in the line engineering coordinate system, forming coordinate pairs for each reference point; The centroid of each reference point is determined by using the least squares method through the coordinates of each reference point and a covariance matrix is constructed, and a singular value decomposition algorithm is introduced to decompose the covariance matrix to obtain a rotation matrix; The translation vector is calculated by combining the rotation matrix and the center of mass, and a coordinate transformation model is constructed. A Kalman filter is introduced into the coordinate transformation model to reduce the coordinate conversion error of the dynamic equipment. The operation monitoring information of the target site is spatially aligned through the optimized coordinate change model.

3. The method for assisting in handling emergency incidents in urban rail transit according to claim 1, characterized in that: The detecting and analyzing event features based on the aligned target site operation monitoring information, and constructing multimodal fusion features of the target site operation monitoring information, specifically includes: An anomaly detection model is constructed using the isolation forest algorithm as a framework. The aligned target site operation monitoring information is obtained and input into the anomaly detection model to analyze the anomalies of the operating equipment within the site. Randomly select the data input into the anomaly detection model to obtain a number of sample data, use the RANSAC linear fitting algorithm to perform linear fitting on the selected sample data, and calculate the difference between the actual observation value and the fitted prediction value to generate residual features; Extracting features from the input target site operation monitoring information to generate an original feature matrix, fusing the residual features with the original feature matrix to generate a fused feature matrix, setting the total number of isolated forests based on the fused feature matrix, and constructing a feature space; Using a random binary search tree to search for isolated points in the feature space and calculate the average path length of each isolated point, defining isolated points smaller than a preset threshold as outliers and generating anomaly detection information; The target station's passenger flow monitoring features and station video monitoring features are extracted from the target station's operation monitoring information after time alignment and spatial alignment. A passenger flow intensity heat map of the target station is generated based on the target station's passenger flow monitoring features. The spatiotemporal graph convolutional network is used to detect abnormal behaviors based on the station video monitoring features to generate abnormal behavior detection information. Combining the anomaly detection information, the passenger flow intensity heat map and the abnormal behavior detection information, a multimodal feature fusion is performed through a weighted fusion method to obtain a multimodal fusion feature of the target site operation monitoring information.

4. The method for assisting in handling emergency incidents in urban rail transit according to claim 1, characterized in that: The establishment of a fuzzy rule database and the analysis of event level and event impact of the current event based on the multimodal fusion characteristics of the target site operation monitoring information specifically include: Based on data retrieval, several historical emergency events are obtained, event features are extracted for the historical emergency events, an attention mechanism is introduced to calculate the attention score of each event feature, and each event feature is weighted by the attention score to generate an event feature matrix; Preset emergency event assessment rules, perform grade assessment on each historical emergency event using the emergency event assessment rules to obtain grade assessment results, and associate the grade assessment results with the event feature matrix to obtain associated result information; Preset a membership function, calculate the membership of the event features corresponding to each historical emergency event through the membership function, and construct a historical emergency event data set in combination with the association result information; The Apriority algorithm is introduced to mine fuzzy rules through the historical emergency event data set, the minimum membership degree of each fuzzy rule is defined as the matching degree, and a fuzzy rule database is established; Building an event level analysis model based on a dual-channel neural network, the event level analysis model includes a temporal channel and a spatial channel, wherein the temporal channel is constructed by an LSTM network and the spatial channel is constructed by a CNN network; The event level analysis model is deep-learned and trained by constructing a training data set through the fuzzy rule database, and a loss function is preset to balance the event classification. After the training is completed, the model parameters are retained to output the final event level analysis model; The multimodal fusion features of the target site operation monitoring information are obtained and input into the trained event level analysis model for analysis to determine the event level of the target site under the current real-time monitoring status and obtain event level analysis information.

5. The method for assisting in handling emergency incidents in urban rail transit according to claim 1, characterized in that: The establishment of a fuzzy rule database and the performance of event level analysis and event impact analysis on the current event based on the multimodal fusion features of the target site operation monitoring information also include: Obtaining an infrastructure distribution map and a line distribution map of the target site, and extracting facility distribution characteristics and line distribution characteristics of the target site based on the infrastructure distribution map and the line distribution map of the target site to obtain first feature information; Abstracting the rail transit network of the target site into a space-time propagation graph model according to the first feature information, wherein the space-time propagation graph model includes an entity node dimension, an edge dimension, and a time dimension; Acquiring event level analysis information, extracting event features of the emergency event at the current moment based on the event level analysis information, initializing nodes of the spatiotemporal propagation graph model based on the extracted event features, and defining abnormal nodes and node states; The state of each node at any time is regarded as a discrete variable. The propagation dynamics equation is used to model the change of abnormal node state over time. The propagation equation is solved by the RK4 method to obtain the state change trend of each node within the preset time and generate node state change information. Calculating a network vulnerability index and a cascading failure risk index based on the node state change information, wherein the network vulnerability index is represented by the importance of the node, and the cascading failure risk is represented by the ratio of the nodes with the current fault level to the total nodes; Emergency event impact analysis information is generated by combining the node state change information, the network vulnerability index and the cascading failure risk index.

6. The method for assisting in handling emergency incidents in urban rail transit according to claim 1, characterized in that: The emergency event handling auxiliary model is constructed based on the MOEA / D algorithm, and the emergency event impact analysis information and event level analysis information are input into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation, specifically including: An emergency response auxiliary model is constructed based on the MOEA / D algorithm to obtain emergency impact analysis information and event level analysis information. The objective function is set to minimize passenger delay time, operational economic losses, and response energy consumption as the optimization goal, and constraints are set. Inputting the emergency event impact analysis information and event level analysis information into the emergency event handling auxiliary model for analysis, decomposing the optimization target to generate uniformly distributed weights, and initializing the population based on the input data to obtain an initial population; Calculating the objective function value of each independent individual in the initial population by using a preset objective function, and screening elite individuals in the initial population according to the objective function value corresponding to each independent individual; Calculate the Euclidean distance between each weight vector, compare the calculated Euclidean distance with the preset threshold, select the weight vectors smaller than the preset threshold to build a neighborhood, and randomly select several individuals from the neighborhood to perform crossover and mutation operations with the elite individuals; Obtain the new individual after crossover and mutation operations, calculate the objective function value of the new individual in the neighborhood, determine whether it is the optimal individual in the neighborhood, and if it is the optimal individual, replace the neighborhood individual to generate a new neighborhood population for iterative update; When the maximum number of iterations is reached or the Pareto front converges, the optimal solution set is output, and a number of candidate emergency event handling plans are generated based on the optimal solution set. The characteristics of the handling plan are extracted for each candidate emergency event handling plan, including delay time, operating loss and handling energy consumption; The candidate emergency response plans are sorted according to their characteristics, and three ranking tables are generated and pushed to provide auxiliary suggestions for emergency response.

7. An emergency incident handling auxiliary system for urban rail transit operation, characterized in that: The system includes: a memory and a processor, wherein the memory contains an emergency incident handling auxiliary method program for urban rail transit operation, and when the emergency incident handling auxiliary method program for urban rail transit operation is executed by the processor, the following steps are implemented: Acquiring target site operation monitoring information, performing time alignment and spatial alignment processing on the target site operation monitoring information, and obtaining aligned target site operation monitoring information; Performing event feature detection and analysis based on the aligned target site operation monitoring information, and constructing a multimodal fusion feature of the target site operation monitoring information; Establish a fuzzy rule database, conduct event level analysis and event impact analysis on the current event based on the multimodal fusion characteristics of the target site operation monitoring information, and obtain event level analysis information and emergency event impact analysis information; An emergency event handling auxiliary model is constructed based on the MOEA / D algorithm, and the emergency event impact analysis information and event level analysis information are input into the emergency event handling auxiliary model to generate an emergency event handling plan for recommendation.

Images