Metro operation decision optimization method based on knowledge graph and cross-modal association

Abstract

The invention discloses a metro operation decision optimization method based on a knowledge graph and cross-modal association, and relates to the related technical field of metro operation, and the method comprises the steps: deploying an edge computing node at edge equipment of a metro, and configuring a universal weak classifier; computing preference analysis is executed, and a preference database is generated; reinforcement learning is carried out on the universal weak classifier; after communication data is received, fault decision recognition is carried out, and an edge fault decision recognition result is established; and obtaining subway operation environment data and user feedback data, and after synchronizing the data to the subway knowledge graph, generating a subway operation and maintenance decision optimization result. The technical problems that in the prior art, multi-modal big data cannot be efficiently processed, potential correlation of different-modal data cannot be fully mined, and consequently subway fault decision recognition accuracy is insufficient and operation efficiency is poor are solved, dynamically-updated and multi-modal fused subway operation knowledge graph construction is achieved, and the metro operation knowledge graph construction efficiency is improved. The technical effect of improving subway operation decision accuracy and operation efficiency is achieved.

Description

Technical Field

[0001] The present application relates to the technical field related to subway operation, and in particular to a subway operation decision optimization method based on knowledge graph and cross-modal association. Background Art

[0002] As cities continue to expand, the efficiency and stability of subway operations, as the core of urban transportation, are becoming increasingly critical. Traditional subway operation decision-making methods are gradually revealing their limitations when faced with complex and changing operating conditions. On the one hand, subway systems contain numerous sensors, video surveillance equipment, and inspection drones, generating massive and diverse data types, covering multimodal information such as equipment status, environmental parameters, and passenger behavior. Efficiently processing and analyzing this data is a challenge. On the other hand, traditional isolated decision-making methods struggle to fully consider various factors in subway operations and cannot fully explore the potential connections between different modal data. This leads to inaccurate and inaccurate decisions, which in turn affects the overall operational efficiency and reliability of the subway.

[0003] At present, relevant technologies have technical problems such as difficulty in efficiently processing multimodal big data and inability to fully explore the potential correlations between different modal data, resulting in insufficient accuracy in subway fault decision-making and identification and poor operational efficiency. Summary of the Invention

[0004] This application provides a subway operation decision optimization method based on knowledge graph and cross-modal association, which solves the technical problems in the existing technology that it is difficult to efficiently process multimodal big data and cannot fully explore the potential associations of different modal data, resulting in insufficient accuracy in subway fault decision identification and poor operational efficiency. It realizes the construction of a subway operation knowledge graph with dynamic update and multimodal fusion, and achieves the technical effect of improving the accuracy of subway operation decision-making and operational efficiency.

[0005] The present application provides a subway operation decision optimization method based on knowledge graph and cross-modal association, including: deploying edge computing nodes on edge devices of the subway, and configuring universal weak classifiers for the edge computing nodes, wherein the edge computing nodes are communicatively connected with sensors, video surveillance, and inspection drones; performing computational preference analysis on the edge computing nodes to generate a preference database; using the preference database to perform reinforcement learning on the universal weak classifier, and integrating the reinforcement learning classifier and the universal weak classifier into an edge decision classifier and then setting it on the corresponding edge computing node, wherein the reinforcement learning includes constructing a reinforcement learning environment using the preference database, loading the universal weak classifier to perform dynamic reward-based learning iterations, and constructing a reinforcement learning classifier; after the edge computing node receives communication data, using the edge decision classifier to perform fault decision identification, wherein the fault decision identification includes respectively obtaining fault decision identification results corresponding to the reinforcement learning classifier and the universal weak classifier, performing trust weighted calculation, and establishing an edge fault decision identification result; obtaining subway operating environment data and user feedback data, constructing cross-border data, synchronizing the cross-border data and the edge fault decision identification result to the subway knowledge graph, and generating a subway operation and maintenance decision optimization result.

[0006] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association also performs the following processing: obtaining the monitoring coverage area of ​​the edge computing node and configuring the location sensitivity of the monitoring coverage area; reading the fault events in the monitoring coverage area, and establishing a first preference constraint after weighted calculation of the probability of occurrence of the fault events through the location sensitivity; performing equipment aging analysis on the monitoring coverage area under regional environment fitting, and establishing a second preference constraint using the equipment aging analysis results; generating the preference database based on the first preference constraint and the second preference constraint.

[0007] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association further performs the following processing: reading the preference database, performing reinforcement data matching based on the preference database, and constructing a reinforcement learning environment; after loading the universal weak classifier, using the reinforcement learning environment to iterate the parameters of the universal weak classifier, and setting dynamic reward feedback in each round of parameter iteration; when the prediction result meets the expected threshold, the optimized universal weak classifier is output as a reinforcement learning classifier.

[0008] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association also performs the following processing: evaluating the current iteration stage of the general weak classifier and establishing a stage influence coefficient; obtaining the prediction result of this round of general weak classifier, and identifying the prediction deviation of the prediction result based on the reinforcement learning environment to generate a first error influence coefficient; establishing a linkage window, performing a prediction backtracking analysis of the general weak classifier based on the linkage window, and generating a second error influence coefficient according to the prediction backtracking analysis result and the parameter iteration direction; and using the stage influence coefficient, the first error influence coefficient, and the second error influence coefficient to set dynamic reward feedback.

[0009] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association also performs the following processing: using the reinforcement learning classifier and the general weak classifier in the edge decision classifier to perform fault decision identification based on communication data respectively, and establish a first fault decision identification result and a second fault decision identification result, the first fault decision identification result is the identification result of the reinforcement learning classifier, and the second fault decision identification result is the identification result of the general weak classifier; performing identification trust analysis on the first fault decision identification result and the reinforcement learning classifier to establish a first trust factor; performing identification trust analysis on the second fault decision identification result and the reinforcement learning classifier to establish a second trust factor; establishing an edge fault decision identification result based on the first fault decision identification result, the first trust factor, the second fault decision identification result, and the second trust factor.

[0010] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association also performs the following processing: after trust-weighting the first fault decision identification result using the first trust factor, a third fault decision identification result is established; after trust-weighting the second fault decision identification result using the second trust factor, a fourth fault decision identification result is established; and trust conflict analysis is performed on the third fault decision identification result and the fourth fault decision identification result to establish a marginal fault decision identification result.

[0011] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association also performs the following processing: extracting user voice data and complaint data, using the voice data and complaint data as user feedback data, and using natural language processing technology to extract key information of the user feedback data to construct user cross-border data; reading the subway's time-series operation noise data, using the time-series operation noise data as operation environment data, and constructing environmental cross-border data based on the operation environment data; constructing cross-border data using the user cross-border data and the environmental cross-border data.

[0012] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association also performs the following processing: establishing an emergency plan set; performing plan matching of the emergency plan set based on the subway operation and maintenance decision optimization result, and establishing a plan matching result; and using the plan matching result to perform emergency response management.

[0013] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association also performs the following processing: based on the cross-border data and the edge fault decision identification results, the subway knowledge graph is anomaly identified to establish anomaly identification results; the subway knowledge graph is used to perform association identification of the anomaly identification results, and the fault decision is reconstructed according to the association identification results to generate a subway operation and maintenance decision optimization result.

[0014] In a possible implementation, the subway operation decision optimization method based on knowledge graph and cross-modal association also performs the following processing: performing data integrity verification on the communication data and establishing a data integrity verification result; if the data integrity verification result is a verification failure result, reporting a data anomaly.

[0015] The proposed subway operation decision optimization method based on knowledge graph and cross-modal association in this application is to deploy edge computing nodes on the edge devices of the subway and configure general weak classifiers; perform computational preference analysis to generate a preference database; conduct reinforcement learning on the general weak classifiers, integrate them into edge decision classifiers and set them on the corresponding edge computing nodes; perform fault decision identification after receiving communication data and establish edge fault decision identification results; obtain subway operating environment data and user feedback data, construct cross-border data, synchronize them to the subway knowledge graph, and generate subway operation and maintenance decision optimization results. This solves the technical problems in the existing technology that it is difficult to efficiently process multimodal big data and cannot fully explore the potential associations of different modal data, resulting in insufficient accuracy in subway fault decision identification and poor operational efficiency. It realizes the construction of a subway operation knowledge graph that is dynamically updated and multimodally integrated, achieving the technical effect of improving the accuracy of subway operation decisions and operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0017] Figure 1A flowchart of a subway operation decision optimization method based on knowledge graph and cross-modal association provided in an embodiment of the present application.

[0018] Figure 2 A schematic diagram of the process of generating a preference database in the subway operation decision optimization method based on knowledge graph and cross-modal association provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0020] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0021] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms “first\second” involved are merely to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, product, or server comprising a series of steps is not necessarily limited to those steps clearly listed, but may include other steps that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0022] The embodiment of the present application provides a subway operation decision optimization method based on knowledge graph and cross-modal association, such as Figure 1 As shown, the method includes:

[0023] Step S100: deploy edge computing nodes on edge devices of the subway and configure universal weak classifiers for the edge computing nodes. The edge computing nodes are communicatively connected with sensors, video surveillance, and inspection drones.

[0024] Preferably, edge devices refer to various types of equipment used for subway operations, such as gates, ticket machines, ventilation equipment, power supply equipment in subway stations, and various control devices on subway trains, which are distributed in all corners of the subway system and are directly related to the operation and services of the subway. Edge computing nodes are then deployed on the edge devices of the subway. Edge computing nodes contain devices with certain computing capabilities, which can process and analyze data close to the data source, reduce the bandwidth pressure and delay of data transmission to the central server, and realize real-time or near real-time monitoring and response of subway operation status. Specifically, the requirements for edge computing in different business scenarios in subway operations are clarified, such as real-time monitoring of equipment status, analysis of video images to detect abnormal behavior, etc., and the number, distribution location and computing power requirements of the required edge computing nodes are determined based on the scale and layout of the subway. The network topology of the edge computing nodes is designed according to the physical structure and network layout of the subway, and it is ensured that the edge computing nodes can achieve efficient and stable communication connections with the existing sensors, video surveillance, inspection drones and other related equipment in the subway. For example, a combination of wired networks (such as optical fiber and Ethernet) and wireless networks (such as Wi-Fi and 5G) is used to meet the connection requirements of devices in different locations.

[0025] Preferably, appropriate edge computing devices are selected based on computing requirements. These edge computing devices should have low power consumption, high reliability and certain computing capabilities, and be able to operate stably in harsh subway environments. They may include industrial-grade servers, embedded computers, edge computing gateways, etc. For example, for scenarios that require processing large amounts of video data, edge computing servers with powerful graphics processing capabilities (GPU) are selected; and appropriate storage devices are selected based on data storage requirements. Considering the importance and real-time nature of subway operation data, high-speed, large-capacity solid-state drives (SSDs) or disk arrays are usually selected to ensure fast data reading and writing and security; according to the communication requirements with external devices, corresponding communication modules are selected, such as Ethernet interface modules for connecting to wired networks, and Wi-Fi or 5G modules for realizing wireless network connections, to ensure that edge computing nodes can communicate seamlessly with various devices; then, according to the planned location, the edge computing node equipment is installed near the edge equipment of the subway. For the equipment room of the subway station, the equipment box in the tunnel, etc., it is necessary to select suitable mounting brackets or cabinets to ensure that the equipment is firmly installed and easy to maintain and manage. At the same time, attention should be paid to issues such as heat dissipation, dust and moisture resistance of the equipment to provide a good operating environment for the equipment; finally, the deployment of the edge computing node is completed, and its performance testing and optimization are carried out.

[0026] Preferably, a general weak classifier is configured for the edge computing node, wherein the general weak classifier is a relatively simple classification model with limited classification capability but certain versatility. When processing data, it can usually only perform relatively basic classification judgments based on some features. Its classification accuracy may not be high, but the computing cost is low and the processing speed is fast. It is suitable for running on edge computing nodes with relatively limited resources. For example, preliminary classification of various types of data from subway operations (such as equipment status data collected by sensors, image data in video surveillance, etc.) is performed; specifically, based on the data characteristics and business needs generated by subway operations, a suitable general weak classifier model is selected, such as a decision tree stump, a naive Bayes classifier, etc., to collect and organize multimodal data from subway sensors, video surveillance, inspection drones and other equipment, and perform data cleaning and feature extraction, such as extracting personnel behavior characteristics from video surveillance data and extracting equipment temperature, pressure and other features from sensor data; then use these data to train the general weak classifier model, obtain the general weak classifier and deploy it to the edge computing node of the subway edge device, avoiding the transmission of a large amount of raw data, thereby reducing network bandwidth pressure and reducing data transmission costs.

[0027] Preferably, the edge computing node is connected to sensors (such as temperature sensors, pressure sensors, current sensors, etc.), which can obtain data collected by the sensors in real time and understand the operating parameters and environmental conditions of the subway equipment; the edge computing node is connected to video surveillance, which can obtain video stream data (including monitoring passenger behavior, station environment and train operation), and use general weak classifiers to analyze the image information in the video, such as detecting whether there is abnormal behavior, signs of equipment failure, etc.; the edge computing node is connected to the inspection drone, which can receive images and video data taken during the drone inspection and the collected environmental data, and integrate data from different data sources, thereby expanding the monitoring range and improving the safety and reliability of subway operations.

[0028] Step S200: Execute computing preference analysis of the edge computing node to generate a preference database.

[0029] Preferably, computing preference analysis of edge computing nodes is performed, which may include analyzing the number of prone faults and fault detection, aging prediction based on environmental data, and equipment status detection. Specifically, edge computing nodes collect information from various data sources such as sensors, video surveillance, and inspection drones to count the number of faults that have occurred in various equipment or areas in the subway system. For example, the number of equipment failures is recorded through the equipment's own fault alarm sensor, or the number of abnormal equipment conditions is analyzed from video surveillance. The number of failures that have occurred reflects the reliability and stability of the equipment or area. Equipment or areas with a large number of failures may have potential problems; fault-related features are extracted from data obtained from sensors and other data sources. For example, for motor equipment, current, voltage, speed and other features can be extracted. For track equipment, deformation, wear and other features of the track can be extracted. Machine learning, deep learning and other technologies are used to analyze and process the extracted features to identify possible failure modes. For example, by analyzing the motor current data, it is possible to identify whether the motor has short circuit, overload and other faults.

[0030] Preferably, sensors are used to collect data related to the subway's operating environment, such as temperature, humidity, dust concentration, and vibration. These environmental factors can affect the aging rate of subway equipment. For example, high temperature and high humidity can accelerate the aging of electronic equipment, while frequent vibration can increase the wear of mechanical components. Based on the collected environmental data, the equipment's status is assessed. For example, the equipment's operating status can be categorized into different levels, such as normal, warning, and fault. When a parameter of the equipment exceeds the normal range, a warning signal is promptly issued. An aging prediction model is then established, combining the equipment's design parameters and historical operating data. This model can predict the degree of aging and remaining service life of the equipment under the current environment. For example, based on the equipment's temperature-aging curve, the aging rate of the equipment can be predicted to be accelerated in a high-temperature environment. Finally, the data obtained from the computational preference analysis is integrated, including the number of faults that have occurred, aging prediction results, fault detection results, and equipment status monitoring data. This data is stored in a specific format and structure to form a preference database. The data in the database is annotated with information such as the data source, acquisition time, and data type. A data index is also established to facilitate rapid query and retrieval of required data.

[0031] Further, such as Figure 2As shown, step S200 also includes step S210, obtaining the monitoring coverage area of ​​the edge computing node and configuring the location sensitivity of the monitoring coverage area; step S220, reading the fault events in the monitoring coverage area, and establishing a first preference constraint after weighted calculation of the probability of occurrence of the fault events through the location sensitivity; step S230, performing equipment aging analysis on the monitoring coverage area under regional environment fitting, and establishing a second preference constraint using the equipment aging analysis results; step S240, generating the preference database based on the first preference constraint and the second preference constraint.

[0032] Preferably, each edge computing node is connected to sensors, video surveillance, inspection drones and other equipment in the subway, and is distributed in different locations of the subway. The range that can be monitored constitutes the monitoring coverage area of ​​the edge computing node. For example, an edge computing node installed at a specific location on a subway platform, the sensors connected to it can monitor the temperature, humidity, passenger flow and other data of the platform, and the camera can monitor the personnel activities and equipment status of the platform. Then the platform-related area is the monitoring coverage area of ​​the edge computing node, which helps to clarify the scope of monitoring that the edge computing node is responsible for; within the monitoring coverage area, the impact of failures at different locations on subway operations is different, and a corresponding sensitivity is configured for each location, where the sensitivity reflects the importance or attention of the failure at that location. For example, the sensitivity of the location where the key braking equipment of the subway train is located will be set higher, and the sensitivity of the location where the auxiliary lighting equipment is located will be relatively low.

[0033] Preferably, in the monitoring coverage area, various monitoring devices are used to read possible fault events in real time, which may be equipment failures (such as motor failures, signal system failures), environmental anomalies (such as excessive temperature, abnormal humidity), etc. For example, when a sensor detects that the temperature of a device exceeds the normal range, it is recorded as a fault; the probability of occurrence of the read fault event is weighted by the configured location sensitivity, that is, the fault event occurring at a location with high sensitivity has a larger weight and accounts for a higher proportion in the overall calculation; while the fault event occurring at a location with low sensitivity has a smaller weight. For example, the probability of occurrence of a fault event at a high-sensitivity location is 10%, and its weight is 0.8; the probability of occurrence of a fault event at a low-sensitivity location is 20%, and its weight is 0.2. After weighted calculation, the fault occurrence situation that comprehensively considers the importance of the location is obtained; and based on the result of the weighted calculation, a first preference constraint is established, which reflects the relationship between the possibility and importance of the fault event when considering the location sensitivity. For example, when allocating resources or making decisions, priority is given to those locations and events with a higher weighted probability of failure.

[0034] Preferably, regional environmental fitting is performed on the monitoring coverage area, that is, the relationship between the environmental factors (such as temperature, humidity, vibration, dust, etc.) in the area and the aging of the equipment is analyzed. Specifically, by collecting historical data and real-time monitoring data, a mathematical model of environmental factors and equipment aging is established to predict the aging degree of the equipment. For example, in a high temperature and high humidity environment, the aging rate of electronic equipment may be accelerated. This model is then used to analyze the aging of equipment in different locations under the current environment. Then, based on the results of the equipment aging analysis, a second preference constraint is established, which takes into account the impact of equipment aging on subway operations and gives priority to those equipment with a high degree of aging and that may be about to fail. For example, for equipment with severe aging, maintenance or replacement plans are arranged in a timely manner. Finally, a preference database is generated based on the first and second preference constraints, which integrates information such as the occurrence of failures and equipment aging that considers location sensitivity. When optimizing subway operation decisions, equipment maintenance plans can be reasonably arranged and resource allocation can be adjusted based on the information in the preference database to improve the safety and reliability of subway operations. Table 1 is an example data of the preference database:

[0035] Table 1 Example data of preference database

[0036]

[0037] Step S300, using the preference database to perform reinforcement learning on the general weak classifier, and integrating the reinforcement learning classifier and the general weak classifier into an edge decision classifier and setting it on the corresponding edge computing node. Reinforcement learning includes using the preference database to build a reinforcement learning environment, loading the general weak classifier to perform learning iteration based on dynamic rewards, and constructing a reinforcement learning classifier.

[0038] Preferably, the preference database is used to perform reinforcement learning on the general weak classifier, wherein reinforcement learning is a machine learning method. The intelligent agent (i.e., the general weak classifier) ​​learns the optimal strategy based on the reward signal fed back by the environment by interacting with the environment to maximize the long-term cumulative reward. In subway operation, the environment is the various operating states and fault events of the subway, and the reward signal is related to the accuracy and timeliness of the fault decision; specifically, the data in the preference database is used as input, including the fault event characteristics of different locations, equipment aging characteristics, etc. The general weak classifier performs preliminary classification based on these input features, and gives the general weak classifier corresponding rewards or penalties based on the comparison between the classification results and the actual situation. If the classification is accurate and the fault event can be identified in time or the equipment status can be accurately assessed, a positive reward will be given; otherwise, a negative reward will be given. For example, if the classifier correctly identifies an impending fault at a highly sensitive location, a higher positive reward will be given; if a fault is misjudged or missed, a negative reward will be given; then the general weak classifier continuously adjusts its classification strategy based on the reward feedback, and gradually improves the accuracy and reliability of the classification through multiple iterative learning, so that it can better adapt to the actual situation of subway operations; the classifier after reinforcement learning (reinforcement learning classifier) ​​is integrated with the original general weak classifier, such as using a simple voting The classification results of the reinforcement learning classifier and the general weak classifier are integrated by the methods of weighted average method, etc. to obtain the edge decision classifier, give full play to the advantages of both, and improve the overall classification performance and decision-making ability; finally, the integrated edge decision classifier is deployed on the corresponding edge computing node. The edge computing node is close to the data source (such as sensors, video surveillance, etc.) and can obtain subway operation data in real time. The edge decision classifier is used to quickly identify fault decisions to reduce data transmission delays, improve the timeliness and response speed of decisions, and provide strong guarantees for the safety and efficiency of subway operations. For example, when the edge computing node receives equipment status data from the sensor, the edge decision classifier can quickly determine whether the equipment has failed, and issue an early warning or take corresponding treatment measures in time.

[0039] Furthermore, step S300 also includes step S310, reading the preference database, performing reinforcement data matching based on the preference database, and constructing a reinforcement learning environment; step S320, after loading the universal weak classifier, using the reinforcement learning environment to iterate the parameters of the universal weak classifier, and setting dynamic reward feedback in each round of parameter iteration; step S330, when the prediction result meets the expected threshold, outputting the optimized universal weak classifier as a reinforcement learning classifier.

[0040] Preferably, various information related to the edge computing node in the preference database is read, including fault event data, device status data, location sensitivity information, and device aging analysis results within the monitoring coverage area of ​​the edge computing node, and data related to reinforcement learning is extracted from the preference database and sorted and matched. For example, the characteristics of the fault event are associated with the corresponding location sensitivity, device aging degree, and other information to provide comprehensive input features for the general weak classifier. Through reinforcement data matching, the data in the preference database is converted into a format suitable for processing by the reinforcement learning algorithm; then, based on the matched reinforcement data, an environment suitable for reinforcement learning of the general weak classifier is created, that is, a reinforcement learning environment, which defines the action space that the agent (general weak classifier) ​​can take (such as different classification decisions), the state space of the environment (represented by the input reinforcement data), and the reward mechanism (used to feedback the quality of the classification decision). For example, the environmental state can be the currently monitored device status characteristics and location information. The action of the agent is to classify and judge whether the device is faulty, and the reward is determined based on the accuracy and importance of the classification result.

[0041] Preferably, a pre-trained universal weak classifier is loaded into a reinforcement learning environment, wherein the universal weak classifier is a classification model with certain initial parameters, which can perform preliminary classification of the input data, but the classification performance may not be ideal; in the reinforcement learning environment, the universal weak classifier makes classification decisions based on the current environmental state (i.e., the input reinforcement data) and adjusts its own parameters based on the reward signal fed back by the environment. By continuously repeating this process, i.e., multiple iterations, the parameters of the universal weak classifier are gradually optimized to improve its classification performance. For example, in each iteration, based on the difference between the classification result and the actual situation, the classifier parameters are updated using algorithms such as gradient descent, so that the classifier can more accurately judge fault events in subsequent classifications; in each round of parameter iteration, dynamic reward feedback is given to the universal weak classifier based on the current classification result and environmental state. The reward setting is determined according to specific business needs and goals. For example, if the classifier accurately identifies a fault event in a high-sensitivity location, a higher reward is given; if a normal device in a low-risk location is misjudged as a fault event, a certain penalty is given. The size and nature of the reward will change dynamically with different situations in order to guide the general weak classifier to learn the optimal classification strategy. Table 2 shows the exemplary data of the reinforcement learning environment:

[0042] Table 2 Reinforcement learning environment related data

[0043]

[0044] Preferably, during the parameter iteration process, the prediction results of the general weak classifier are continuously evaluated. The evaluation indicators can be accuracy, recall rate, F1 value, etc. The current prediction results are compared with the pre-set expected threshold to determine whether the classifier has achieved a satisfactory performance level. When the prediction results of the general weak classifier meet the expected threshold, it means that after reinforcement learning, the classifier has been sufficiently optimized and can show good performance in the given task. The optimized general weak classifier is output as a reinforcement learning classifier and used for actual tasks such as subway fault detection and equipment status monitoring to ensure the accuracy and reliability of subway operation decisions. As shown in Table 3, exemplary general weak classifier parameter iteration data is shown:

[0045] Table 3 Example data of parameter iteration of general weak classifier

[0046]

[0047] Furthermore, step S320 also includes step S321, evaluating the current iteration stage of the general weak classifier and establishing a stage influence coefficient; step S322, obtaining the prediction result of the current round of the general weak classifier, and identifying the prediction deviation of the prediction result based on the reinforcement learning environment to generate a first error influence coefficient; step S323, establishing a linkage window, performing a prediction backtracking analysis of the general weak classifier based on the linkage window, and generating a second error influence coefficient according to the prediction backtracking analysis result and the parameter iteration direction; step S324, using the stage influence coefficient, the first error influence coefficient, and the second error influence coefficient to set dynamic reward feedback.

[0048] Preferably, in the reinforcement learning process, different iteration stages have different significance for improving the performance of the classifier. For example, in the early iteration stage, the classifier may be in a rapid learning and exploration stage, with a strong adaptability to new features and patterns, but the accuracy may be relatively low; in the later iteration stage, the classifier is more focused on fine-tuning existing knowledge to improve accuracy; by evaluating the current iteration stage, we can understand the progress of the classifier in the entire learning process, which may include evaluation based on indicators such as the number of iterations, the trend of changes in the loss function, and the speed of improvement of classification accuracy, and establish a stage influence coefficient to measure the impact of the current iteration stage on the final classification The relative importance of the performance of the classifier; in each round of iteration, the general weak classifier will make a prediction result based on the input reinforcement data, that is, classify and judge the equipment status or fault event. Based on the known real labels or actual conditions in the reinforcement learning environment, the prediction result can be compared with the true value to identify the prediction deviation; then, by calculating the degree of difference between the prediction result and the true value, the first error influence coefficient is generated, which reflects the influence of the accuracy of the prediction result of this round on the overall learning process. The larger the error, the larger the first error influence coefficient, which means that the prediction result of this round has a more important role in improving the classifier, and stronger feedback is needed to adjust the parameters of the classifier.

[0049] Preferably, the linkage window is a time window used to observe the prediction of the general weak classifier over a period of time or in several consecutive rounds of iterations. By establishing a linkage window, the prediction results of the classifier can be retrospectively analyzed, that is, the prediction results of several rounds in the past and their changing trends are reviewed, and then the prediction results of the classifier are analyzed under different parameter iteration directions. How does it change? For example, when the parameters are adjusted in a certain direction, does the prediction result become more accurate or worse? Combining the prediction retrospective analysis results and the parameter iteration direction, a second error influence coefficient is generated, which can more comprehensively reflect the learning dynamics and stability of the classifier. If the prediction results of the classifier in the linkage window gradually improve under a certain parameter iteration direction, then the second error influence coefficient may be small, indicating that the current parameter adjustment direction is correct, otherwise it is large.

[0050] Preferably, the stage influence coefficient, the first error influence coefficient, and the second error influence coefficient are combined to set dynamic reward feedback for the general weak classifier to guide the classifier to learn in the right direction, that is, to improve classification performance by adjusting parameters. Specifically, a reward function is designed based on the different values ​​of these three influence coefficients. For example, when the stage influence coefficient is large (in the key iteration stage), the first error influence coefficient is small (the current round of prediction is accurate), and the second error influence coefficient is also small (the parameter iteration direction is correct and the historical performance is stable), the classifier is given a higher reward to encourage it to continue to maintain the current learning state and parameter adjustment direction; conversely, if a coefficient is large (such as the first error influence coefficient is large, indicating that the current round of prediction error is large), a lower reward or even a penalty is given to encourage the classifier to adjust the parameters to improve the prediction results. Through this dynamic reward feedback mechanism, the general weak classifier can more effectively perform reinforcement learning and gradually optimize its own performance to meet the needs of practical tasks such as subway fault detection and equipment status monitoring.

[0051] In step S400, after the edge computing node receives the communication data, the edge decision classifier is used to perform fault decision identification and establish an edge fault decision identification result. The fault decision identification includes respectively obtaining the fault decision identification results corresponding to the reinforcement learning classifier and the general weak classifier, and performing trust weighted calculation.

[0052] Preferably, the subway operation-related data collected by monitoring equipment such as sensors, video surveillance equipment, and inspection drones, such as sensors that collect equipment operating parameters (such as temperature, pressure, current, voltage, etc.), video surveillance equipment that records real-time images in stations and on tracks, and inspection drones that may collect environmental conditions along the tracks, are transmitted to edge computing nodes through wired or wireless communications. The edge computing nodes serve as the center for data reception and preliminary processing, and are responsible for receiving these communication data from different devices to ensure that the subway operation status can be quickly monitored and analyzed; these communication data are then input into the edge decision classifier, which extracts and analyzes features of the data. For example, for equipment operation parameter data, the classifier will determine whether the current parameter value is within the normal range and whether there are abnormal fluctuations or trend changes; for video data, it will analyze whether there are abnormal behaviors, equipment damage, etc. in the picture. By comprehensively analyzing different types of data, the edge decision classifier can make decisions regarding the presence, type, and severity of a fault, generating edge fault identification results. These results may include information such as the type of fault (e.g., motor fault, signal fault, track fault), the location of the fault (e.g., a specific station, train car, or track section), the severity of the fault (minor, moderate, severe), and the possible cause of the fault. This can be used directly for local fault handling and response. For example, edge computing nodes can automatically trigger an alarm based on the severity of the fault, notifying nearby maintenance personnel for inspection and repair. Furthermore, this information can be uploaded to the central management platform for subway operations to optimize operational strategies and ensure the safe and stable operation of the subway system.

[0053] Furthermore, step S400 further includes step S410, performing a data integrity check on the communication data to establish a data integrity check result; and step S420, reporting a data anomaly if the data integrity check result is a check failure result.

[0054] Preferably, the communication data is checked for data integrity. Specifically, it is checked whether the data is lost, damaged or erroneous during transmission, storage, etc., such as by checking the data integrity through cyclic redundancy check (CRC), parity check, hash check, etc. Taking cyclic redundancy check as an example, before sending the data, the sender will calculate a CRC check code based on the data to be sent and send it together with the data. After receiving the data, the receiver (i.e., the edge computing node) will use the same algorithm to recalculate the CRC check code for the received data and compare the calculation result with the received check code. In addition, it may also check whether the data format is correct and whether the data fields are complete. For example, for the device status data sent by the sensor, it will check whether each data field (such as temperature, pressure, current, etc.) has a reasonable value and whether there are missing or erroneous fields.

[0055] Preferably, after completing the data integrity check of the communication data, a corresponding verification result will be obtained. The verification result is usually divided into two situations: verification passed and verification failed. If the data integrity check result is verification failed, it means that there is a problem with the received communication data. It may be that part of the data is lost during transmission, or the data is tampered with or damaged, resulting in an inability to correctly reflect the actual operating status of the subway system. The edge computing node will report data anomalies to avoid using incorrect or incomplete data for subsequent analysis and decision-making; for example, the system log records abnormal information, including the time when the anomaly occurred, the source of the data involved (such as which sensor or device sent the data), the specific reason for the failure of the verification, etc., and reminds relevant operation and maintenance personnel to pay attention to data anomalies so that they can take timely measures, such as re-acquiring data, checking communication lines or equipment, etc., to ensure that subsequent data can be accurately and completely received and processed.

[0056] Furthermore, step S400 also includes step S430, using the reinforcement learning classifier and the universal weak classifier in the edge decision classifier to perform fault decision identification based on communication data respectively, and establish a first fault decision identification result and a second fault decision identification result, the first fault decision identification result is the identification result of the reinforcement learning classifier, and the second fault decision identification result is the identification result of the universal weak classifier; step S440, performing identification trust analysis on the first fault decision identification result and the reinforcement learning classifier to establish a first trust factor; step S450, performing identification trust analysis on the second fault decision identification result and the reinforcement learning classifier to establish a second trust factor; step S460, establishing an edge fault decision identification result based on the first fault decision identification result, the first trust factor, the second fault decision identification result, and the second trust factor.

[0057] Preferably, the edge decision classifier includes a reinforcement learning classifier and a general weak classifier. When the edge computing node receives the communication data, it will use these two classifiers to process the data respectively to identify whether there is a fault and the related circumstances of the fault. Specifically, the reinforcement learning classifier analyzes and processes the communication data according to the learned strategies and patterns to obtain a decision identification result about the fault, that is, the first fault decision identification result; the general weak classifier will also perform fault decision identification on the same communication data, and the result obtained is called the second fault decision identification result. Then, a recognition confidence analysis is performed to evaluate the reliability of the recognition results of the two classifiers. For the first fault decision recognition result, the confidence level of this result is determined by analyzing the performance of the reinforcement learning classifier in previous similar data processing, the stability of its parameters, and the degree of conformity with the actual situation. This confidence level is expressed as a first trust factor. If the reinforcement learning classifier has high accuracy and good stability on historical data, the first trust factor will be high, indicating that there is high confidence in its recognition result. Similarly, for the second fault decision recognition result, various relevant factors of the general weak classifier when processing communication data for fault recognition, such as its initial performance and adaptability to the current task, are analyzed to determine the second trust factor, which represents the confidence level in the general weak classifier's recognition result. Finally, the recognition results of the two classifiers and the trust factor are comprehensively considered to determine the marginal fault decision recognition result. That is, the two results are weighted and fused according to the trust factor, fully considering the reliability of each result. Ultimately, a more comprehensive, accurate, and reliable marginal fault decision recognition result is obtained, thereby ensuring the accuracy and reliability of subway operation decisions.

[0058] Furthermore, step S460 also includes step S461, establishing a third fault decision identification result after trust weighting the first fault decision identification result using the first trust factor; step S462, establishing a fourth fault decision identification result after trust weighting the second fault decision identification result using the second trust factor; step S463, performing trust conflict analysis on the third fault decision identification result and the fourth fault decision identification result, and establishing a marginal fault decision identification result.

[0059] Preferably, the first fault decision recognition result is trust-weighted using a first trust factor. Specifically, the first fault decision recognition result is adjusted based on the magnitude of the first trust factor. Specifically, the first fault decision recognition result is multiplied by the first trust factor. If the first trust factor is close to 1, it indicates high confidence in the recognition result of the reinforcement learning classifier, and the weighted result is not much different from the original first fault decision recognition result. If the first trust factor is small, such as 0.3, the weighted result is significantly reduced from the original result. A third fault decision recognition result is then established. Similar to establishing the third fault decision recognition result, a second trust factor is used to trust-weight the recognition result of the general weak classifier (the second fault decision recognition result). Specifically, the second fault decision recognition result is multiplied by the second trust factor to obtain a fourth fault decision recognition result. A trust conflict analysis is then performed on the third and fourth fault decision recognition results. This involves evaluating the difference between the two results, such as by calculating the difference or relative error between the two results. Depending on the degree of conflict and the specific circumstances, different processing strategies are employed to determine the final marginal fault decision recognition result, thereby ensuring the accuracy and reliability of subway operation decisions.

[0060] Step S500: Obtain subway operating environment data and user feedback data, construct cross-border data, synchronize the cross-border data and the edge fault decision identification results to the subway knowledge graph, and generate subway operation and maintenance decision optimization results.

[0061] Preferably, subway operating environment data is obtained. For example, temperature and humidity sensors and air quality monitoring equipment deployed in subway stations and subway tunnels can collect real-time data on ambient temperature, humidity, and harmful gas concentrations. Track stress sensors and track geometry monitors can be used to obtain data on track stress and track smoothness. Seismic monitoring equipment can be used to monitor seismic activity in areas along the subway line. User feedback data can be collected from subway passengers, for example, through social media platforms, feedback boxes set up in subway stations and subway cars, feedback functions developed in mobile apps, and online questionnaires, encouraging passengers to provide feedback on their riding experience. Feedback may include information on car crowding, air conditioning temperature comfort, train punctuality, and station sign clarity. The operating environment data and user feedback data are then integrated and correlated to generate cross-domain data. For example, user feedback on a station experiencing stuffy carriages during hot weather (operating environment data) can be correlated with user feedback on a station experiencing stuffy carriages (user feedback data) to form a cross-domain data record, thus breaking down data domain limitations and exploring potential relationships between different types of data.

[0062] Preferably, the subway knowledge graph is a knowledge collection stored in a graph structure, which is used to describe various entities in subway operations (such as equipment, stations, lines, passengers, etc.) and the relationships between them (such as the relationship between equipment and stations, the connection relationship between lines and stations, the association relationship between passengers and riding behaviors, etc.), and then synchronize the constructed cross-border data and edge fault decision identification results to the subway knowledge graph. For example, if the edge fault decision identification result shows that a certain equipment on a certain train has a fault, this fault information is associated with the corresponding entity of the equipment in the knowledge graph, and the cross-border data at the time of the fault is added as additional information, so that the entities in the knowledge graph have more comprehensive and rich information, including not only their own inherent attributes, but also dynamic information related to the operating environment, user feedback and fault conditions. Finally, the subway knowledge graph that synchronizes cross-border data and edge fault decision identification results is deeply mined and analyzed. For example, by analyzing the correlation between equipment failures and user feedback under specific operating environments (such as high temperature and high humidity) in the knowledge graph, it is found that a certain type of equipment fails frequently under specific environments, and when failures occur, user complaints about car comfort increase significantly, thereby generating a subway operation and maintenance decision optimization plan. For example, in response to the above-mentioned problems, decisions are made to strengthen the inspection frequency and maintenance intensity of related equipment in high temperature and high humidity weather; or adjust the operating parameters of the car air conditioner to improve passenger comfort, thereby improving the accuracy of subway operation decisions and the safety and reliability of subway operations.

[0063] Furthermore, step S500 also includes step S510, extracting the user's voice data and complaint data, using the voice data and complaint data as user feedback data, and using natural language processing technology to extract key information of the user feedback data to construct user cross-border data; step S520, reading the subway's time-series operation noise data, using the time-series operation noise data as operation environment data, and constructing environmental cross-border data based on the operation environment data; step S530, constructing cross-border data with the user cross-border data and the environmental cross-border data.

[0064] Preferably, voice data and complaint data are extracted from the user's interaction with the subway, wherein the voice data may come from the voice message device installed in the subway station, the call records of the customer service center, or the voice feedback of the passenger through the mobile phone application; the complaint data includes the complaint information submitted by the passenger through various channels such as written complaints and online platform complaints; the user feedback data is processed by natural language processing (NLP) technology, and the voice data is converted into text form by voice recognition using NLP, and then the text is segmented, tagged with parts of speech, and recognized by named entities. For example, from the passenger's complaint "I took the subway today, and the air conditioning at XX station was not cool at all, it was too hot", key information such as "XX station" (location entity) and "the air conditioning was not cool" (problem description) can be extracted through NLP technology; the extracted key information is integrated and structured to form user cross-border data.

[0065] Preferably, the time-series operation noise data of the subway is read as the operation environment data, that is, the time-series operation noise data is collected in real time by noise sensors installed along the subway track, at stations, etc., reflecting the change of noise generated during the operation of the subway over time, and constructing environmental cross-border data based on the collected time-series operation noise data. Specifically, the noise data is analyzed, such as calculating the statistical characteristics such as the average value, maximum value, and minimum value of noise intensity in different time periods and different locations. At the same time, combined with other relevant information of the subway operation, such as train speed, line type (underground, above ground), etc., this information is associated with the noise data to form environmental cross-border data that can comprehensively describe the noise conditions of the subway operation environment. Finally, the user cross-border data and the environmental cross-border data are integrated to form cross-border data. For example, it may be found that when the subway operation noise exceeds a certain threshold, users' complaints about the comfort of the car will increase, which will help to gain a deeper understanding of the subway operation from multiple angles, thereby formulating more reasonable operation strategies and optimization plans to improve the quality of subway operation.

[0066] Furthermore, step S500 also includes step S540, establishing an emergency plan set; step S550, performing plan matching of the emergency plan set based on the subway operation and maintenance decision optimization result, and establishing a plan matching result; step S560, using the plan matching result to perform emergency response management.

[0067] Preferably, a set of emergency plans is formulated for various emergency situations that may occur during subway operations, where emergency situations include but are not limited to equipment failures, natural disasters, public health incidents, safety accidents, etc. For example, an emergency plan for a train failure may include a fault diagnosis process, emergency rescue measures, and a passenger evacuation plan; an emergency plan for a fire will cover the fire alarm mechanism, fire extinguishing measures, and personnel escape routes; each emergency plan specifies in detail the specific actions to be taken in a specific emergency situation, the division of responsibilities, and resource allocation, etc., to ensure that an emergency can be responded to quickly and effectively when it occurs. The subway operation and maintenance decision optimization results are compared and analyzed with each plan in the emergency plan set to match the corresponding emergency plan. For example, if the subway operation and maintenance decision optimization results show that a key equipment has a potential failure risk, then the emergency plan related to the equipment failure is searched in the emergency plan set. That is, by matching factors such as the type, severity, and possible impact range of the equipment failure with the trigger conditions in the emergency plan, the most suitable plan is determined to ensure that the correct emergency plan can be quickly activated and effective response measures can be taken when an emergency occurs.

[0068] Preferably, after finding an emergency plan that suits the current situation, emergency response management is carried out according to the plan, including quickly organizing relevant personnel and resources to carry out emergency actions in accordance with the processes and responsibilities specified in the plan. For example, if the matching plan is a train failure emergency plan, then emergency response management may involve notifying maintenance personnel to rush to the scene to repair the fault, organizing station staff to guide passengers to evacuate safely or transfer to other trains, and promptly informing passengers of the situation and calming their emotions through broadcasts, electronic display screens, etc. During the emergency response process, the implementation of emergency actions must be monitored in real time, and resources must be adjusted and optimized according to actual conditions to ensure that emergency actions can proceed smoothly according to the plan, ultimately achieving effective control and proper handling of emergencies and ensuring the safety of subway operations.

[0069] Furthermore, step S500 also includes step S570, performing abnormal identification of the subway knowledge graph based on the cross-border data and the edge fault decision identification results, and establishing abnormal identification results; step S580, using the subway knowledge graph to perform association identification of the abnormal identification results, reconstructing the fault decision according to the association identification results, and generating subway operation and maintenance decision optimization results.

[0070] Preferably, the cross-border data and the edge fault decision identification results are mapped to the subway knowledge graph. When an abnormal situation is found in the cross-border data (such as a large number of user complaints about elevator failures at a certain station for several consecutive days, and the vibration parameters in the elevator operating environment data of the station exceed the normal range) or fault information in the edge fault decision identification results, the abnormality is marked on the corresponding entity in the knowledge graph (such as the elevator entity of the station), including adding abnormal state attributes and recording detailed information related to the abnormality, such as the time when the abnormality occurred, the abnormality description (specific description from the cross-border data or the fault identification results), etc., and finally the abnormality identification result is formed; then the subway knowledge graph uses semantic relationships for reasoning to identify other entities associated with the abnormal identification entity, for example, other signals on the train operation line that may be affected by the signal failure, the affected train numbers, etc., and then obtain the associated identification result.

[0071] Preferably, the fault decision is reconstructed based on the association identification result, that is, the original fault decision is adjusted. For example, it was originally considered that a single signal machine fault was based on the edge fault decision identification result, but through the association identification of the knowledge graph, it was found that the fault may have a chain reaction effect on the operation of multiple trains on multiple lines. The fault decision cannot be limited to the maintenance of the signal machine, but it is also necessary to consider how to adjust the train operation plan, and finally generate a subway operation and maintenance decision optimization plan, which may include a detailed equipment maintenance plan (maintenance personnel arrangement, maintenance tool preparation, maintenance time window determination), train operation adjustment plan (adjustment of train number, change of operation line, setting of temporary stop), and passenger service optimization measures (increase broadcast notification, strengthen platform guidance, provide transfer suggestions), etc., which can more comprehensively and effectively respond to abnormal situations in subway operations, improve the efficiency and service quality of subway operation and maintenance, and ensure the safe and stable operation of subway operations.

[0072] The above specific embodiments do not constitute a limitation to the scope of protection of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application should be included in the scope of protection of this application. In some cases, the actions or steps recorded in this application can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A subway operation decision optimization method based on knowledge graph and cross-modal association, characterized by: The method comprises: Deploy edge computing nodes at the edge of the subway and configure universal weak classifiers for the edge computing nodes, which are connected to sensors, video surveillance, and inspection drones. Performing computing preference analysis on the edge computing nodes to generate a preference database; The preference database is used to perform reinforcement learning on the general weak classifier, and the reinforcement learning classifier and the general weak classifier are integrated into an edge decision classifier and then set on the corresponding edge computing node. The reinforcement learning includes using the preference database to build a reinforcement learning environment, loading the general weak classifier to perform learning iterations based on dynamic rewards, and building a reinforcement learning classifier; After the edge computing node receives the communication data, the edge decision classifier is used to perform fault decision identification and establish an edge fault decision identification result, wherein the fault decision identification includes respectively obtaining the fault decision identification results corresponding to the reinforcement learning classifier and the general weak classifier, and performing a trust weighted calculation; Obtain the subway's operating environment data and user feedback data, construct cross-border data, synchronize the cross-border data and the edge fault decision identification results to the subway knowledge graph, and generate subway operation and maintenance decision optimization results.

2. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 1 is characterized in that: The performing of computing preference analysis of the edge computing node to generate a preference database includes: Obtaining a monitoring coverage area of ​​the edge computing node and configuring a location sensitivity of the monitoring coverage area; Reading fault events within the monitoring coverage area, and calculating the probability of occurrence of the fault events by weighting the location sensitivity, to establish a first preference constraint; Performing equipment aging analysis under regional environment fitting for the monitoring coverage area, and establishing a second preference constraint using the equipment aging analysis result; The preference database is generated based on the first preference constraint and the second preference constraint.

3. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 1 is characterized in that: The using the preference database to perform reinforcement learning on the general weak classifier includes: Reading the preference database, performing reinforcement data matching based on the preference database, and building a reinforcement learning environment; After loading the universal weak classifier, performing parameter iteration of the universal weak classifier using the reinforcement learning environment, and setting dynamic reward feedback in each round of parameter iteration; When the prediction result meets the expected threshold, the optimized general weak classifier is output as the reinforcement learning classifier.

4. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 3 is characterized in that: The method of using the reinforcement learning environment to iterate the parameters of the universal weak classifier and setting dynamic reward feedback in each round of parameter iteration includes: Evaluate the current iteration stage of the general weak classifier and establish the stage influence coefficient; Obtaining a prediction result of the current round of the universal weak classifier, and identifying a prediction deviation of the prediction result based on the reinforcement learning environment to generate a first error influence coefficient; Establishing a linkage window, performing a prediction backtracking analysis of a universal weak classifier based on the linkage window, and generating a second error influence coefficient according to the prediction backtracking analysis result and the parameter iteration direction; Dynamic reward feedback is set using the stage influence coefficient, the first error influence coefficient, and the second error influence coefficient.

5. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 1 is characterized in that: The method of using the edge decision classifier to perform fault decision identification and establish an edge fault decision identification result includes: Using the reinforcement learning classifier and the universal weak classifier in the edge decision classifier to perform fault decision identification based on communication data, respectively, to establish a first fault decision identification result and a second fault decision identification result, wherein the first fault decision identification result is the identification result of the reinforcement learning classifier, and the second fault decision identification result is the identification result of the universal weak classifier; Performing an identification trust analysis on the first fault decision identification result and the reinforcement learning classifier to establish a first trust factor; performing an identification trust analysis on the second fault decision identification result and the reinforcement learning classifier to establish a second trust factor; A marginal fault decision identification result is established based on the first fault decision identification result, the first trust factor, the second fault decision identification result, and the second trust factor.

6. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 5 is characterized in that: The establishing of the edge fault decision identification result based on the first fault decision identification result, the first trust factor, the second fault decision identification result, and the second trust factor includes: After trust-weighting the first fault decision identification result using the first trust factor, a third fault decision identification result is established; After trust-weighting the second fault decision identification result using the second trust factor, a fourth fault decision identification result is established; A trust conflict analysis is performed on the third fault decision identification result and the fourth fault decision identification result to establish a marginal fault decision identification result.

7. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 1 is characterized in that: The acquisition of subway operating environment data and user feedback data and the construction of cross-border data include: Extracting user voice data and complaint data, using the voice data and complaint data as user feedback data, and using natural language processing technology to extract key information from the user feedback data to construct user cross-border data; Reading the time-series operation noise data of the subway, using the time-series operation noise data as the operation environment data, and constructing the environmental cross-border data according to the operation environment data; Cross-border data is constructed using the user cross-border data and the environment cross-border data.

8. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 1 is characterized in that: After the subway operation and maintenance decision optimization result is generated, the following steps are included: Establish a collection of emergency response plans; Performing plan matching of the emergency plan set based on the subway operation and maintenance decision optimization result, and establishing a plan matching result; The plan matching result is used to perform emergency response management.

9. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 1 is characterized in that: The step of synchronizing the cross-border data and the edge fault decision identification result to the subway knowledge graph includes: Based on the cross-border data and the edge fault decision identification result, anomaly identification of the subway knowledge graph is performed to establish an anomaly identification result; The subway knowledge graph is used to perform association identification of the abnormal identification results, and fault decision reconstruction is performed based on the association identification results to generate subway operation and maintenance decision optimization results.

10. The subway operation decision optimization method based on knowledge graph and cross-modal association according to claim 1, characterized in that: After the edge computing node receives the communication data, the method includes: Performing a data integrity check on the communication data and establishing a data integrity check result; If the data integrity check result is a check failure result, a data anomaly is reported.

Images