Tunnel anomaly detection and response method and equipment based on rail robot

Abstract

The invention provides a tunnel anomaly detection and response method and device based on an orbital robot, and the method comprises the steps: obtaining scoring and grading data from a quantitative scoring model through a preset model optimization module, carrying out the dynamic adjustment of the parameters of the scoring model through a deep Q learning network in combination with historical response records, optimizing a grading threshold value to match an actual threat level, and carrying out the detection of an actual threat level. Transmitting the adjusted hierarchical data to a response scheduling system; and according to the adjusted hierarchical data, automatically matching a preset response strategy combination, if at least one selectable strategy combination exists, determining an optimal strategy combination through a weighted optimization algorithm, a comprehensive response speed, a resource utilization rate and a disposal cost, and issuing an execution instruction to a related execution unit.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a tunnel anomaly detection and response method and device based on a track robot. Background Art

[0002] In tunnel inspection systems, the 360° video streams collected by rail robots are massive, posing real-time processing challenges. Edge computing nodes have limited computing power, making it difficult to run complex AI models. Lightweight models can also affect detection accuracy. The quantitative metrics used in abnormal event scoring models struggle to comprehensively cover all scenarios, and fixed thresholds are unsuitable for dynamic environments. Training deep Q-learning networks requires a large amount of sample data, but abnormal events are relatively rare, potentially leading to model overfitting. The contingency plan selection problem under multi-objective optimization is NP-hard (Non-deterministic Polynomial Time), making it difficult to find the optimal solution in real time. Summary of the Invention

[0003] To achieve the purpose of the present invention, in a first aspect, the present invention provides a tunnel anomaly detection and response method based on a track robot, which mainly includes: The track robot performs scheduled inspection tasks along the preset track, and is equipped with a camera with a preset resolution to collect panoramic video stream data in the tunnel. The acquired video stream is transmitted to the edge computing node in real time, and the video stream is analyzed for anomaly detection through a lightweight artificial intelligence visual model to obtain anomaly detection results; if there is an abnormal condition in the anomaly detection result, the abnormal feature data is extracted to generate a structured message, and the structured message is input into the preset quantitative scoring model to calculate the score value, and the level of the abnormal condition is determined according to the score value. Among them, there are four levels of abnormal conditions, with the processing priority of level one being the highest and level four being the lowest; the deep Q learning network is used to dynamically adjust the scoring threshold of the preset quantitative scoring model, and the abnormal conditions are re-classified according to the adjusted scoring threshold to obtain adjusted classification data; the preset response strategy combination is matched according to the adjusted classification data, and if there is at least one optional strategy combination, the weighted optimization algorithm is started. , a comprehensive calculation is performed from the three dimensions of response time, resource utilization and disposal cost, the optimal strategy combination is determined, and the execution instructions are sent to the relevant execution units; the rail robot obtains the execution instruction, moves to the nearest pile number position of the anomaly point, collects the on-site abnormal image data and uploads it to the cloud server, and compares it with the pre-stored image database through the deep learning image feature matching algorithm to confirm the type of anomaly. At the same time, a navigation path is generated based on the synchronous positioning and map construction algorithm to drive the robot to move to the anomaly point for the second time; after the rail robot arrives at the anomaly point, it activates the nearest broadcast node and plays the pre-recorded voice command. At the same time, it connects to the surveillance cameras around the anomaly point to obtain real-time video stream data, which is transmitted to the command center through the real-time communication protocol to establish a multi-party video communication channel; during the anomaly handling process, the message queue telemetry transmission protocol is used to transmit command data, and all message transmission records are stored in the preset data analysis module; Obtain indicator data including response time and handling success rate from message transmission records, dynamically adjust the abnormal situation level, and use the log analysis stack to statistically evaluate the alarm push effect, optimize the push strategy, and regularly update the deep Q learning network parameters based on data accumulation, and continuously improve the response strategy through offline training.

[0004] In a second aspect, the present invention further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0005] The technical solution provided by the embodiment of the present invention may have the following beneficial effects: The present invention discloses a tunnel anomaly detection and response method and device based on a track robot. The method utilizes a track robot to perform inspection tasks along a preset track, collects panoramic video streams in the tunnel, and uses edge computing nodes to analyze and detect abnormal conditions in real time. When an anomaly is detected, feature data is extracted and graded through a quantitative scoring model, and then the classification threshold is dynamically optimized by a deep Q learning network. The response scheduling system matches the optimal strategy combination based on the classification data, directs the track robot to quickly return to the anomaly point, activates the broadcast node and establishes multi-party video communication. The entire process uses a message queue telemetry transmission protocol to ensure reliable delivery of instructions, and the data analysis module continuously optimizes the anomaly classification and response strategy. The present invention realizes intelligent detection, rapid response and continuous optimization of tunnel anomalies, thereby improving the efficiency and accuracy of tunnel safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 This is a flow chart of the tunnel anomaly detection and response method based on a track robot of the present invention.

[0007] Figure 2 The figure is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0008] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0009] like Figure 1 The tunnel anomaly detection and response method based on the track robot in this embodiment may specifically include: S101, the track robot performs scheduled inspection tasks along the preset track, and is equipped with a camera with a preset resolution to collect panoramic video stream data in the tunnel. The acquired video stream is transmitted to the edge computing node in real time, and anomaly detection and analysis of the video stream is performed through a lightweight artificial intelligence visual model to obtain anomaly detection results.

[0010] The track robot conducts inspections along a pre-set track. During the inspection, its onboard camera with a preset resolution captures panoramic video data of the tunnel environment. This raw video stream is transmitted to the edge computing node for initial storage and processing, generating basic video data. Based on this basic video data, a lightweight AI vision model decomposes the video stream into frames, extracting image features from each frame. Based on these features, key image regions are identified. Anomaly detection is then performed on this key region to generate results.

[0011] For example, in a tunnel inspection scenario involving a rail robot, the robot, as the core device, moves at a constant speed of 5 kilometers per hour along a pre-set track, equipped with a high-definition camera to capture panoramic video data from the tunnel in real time. This camera, with 1080p resolution, covers multiple angles within the tunnel, including the roof, sidewalls, and floor, ensuring comprehensive video data. The captured raw video stream is transmitted to an edge computing node at 30 frames per second. This process effectively reduces data transmission latency to the cloud and improves real-time performance. Specifically, during initial storage and processing of the raw video stream, the edge computing node compresses and converts the video data into a basic video data format for subsequent analysis. For example, the node may store the video stream as a temporary file of 500MB per minute and perform preliminary denoising to remove blur caused by low light levels in the tunnel or camera shake. This approach improves the accuracy of subsequent image analysis and reduces the possibility of misjudgment. In one possible implementation, based on the basic video data, the system uses a lightweight AI vision model to decompose the video stream into frames. This lightweight AI vision model, based on the MobileNetV3 or YOLOv8-nano architecture, is adaptively tuned for tunnel scene characteristics (such as low light and fixed backgrounds), retaining core feature extraction layers (such as the first five layers of the CSPDarknet backbone) to reduce computational overhead. The lightweight AI vision model generation process involves pre-training a base model using a public tunnel anomaly dataset (such as TunnelDefect-10K). A structured pruning algorithm (such as L1 regularized channel pruning) is then used to remove redundant convolution kernels. This algorithm calculates the sum of the absolute values ​​of each channel's weights and prunes 30% of the low-weight channels. Furthermore, knowledge distillation techniques are employed, using the original complex model (such as YOLOv8-large) as a teacher model to guide the lightweight model in learning feature maps, ensuring a pruning accuracy loss of ≤5%. After optimization, the model was tested on edge computing nodes (such as the NVIDIA Jetson Nano platform) to verify that the frame processing time remains stable at under 50 milliseconds (including image preprocessing). This model is usually based on deep learning technology, but in order to adapt to the limited computing power of edge computing nodes, the model will be pruned and optimized to ensure that the processing time of each frame of image is controlled within 50 milliseconds. After decomposition, the system extracts the features of each frame of image, such as color distribution, edge contour and other information, and determines the key area information in the image based on these features, such as crack areas on the tunnel wall or foreign objects on the ground. This method significantly improves the efficiency of data processing while reducing computing resource consumption. For example, for the analysis of key area information, if the system detects the presence of spilled objects in the image, such as scattered plastic bottles, or pedestrians entering the tunnel's restricted area, or even illegally parked vehicles, it will trigger the anomaly detection process.The system will compare these abnormal features with pre-set abnormal templates, such as the shape characteristics of the spilled objects and the dynamic trajectory of pedestrians, and generate anomaly detection results. This result may include information such as the type of anomaly, location coordinates, and severity. For example, if a spilled object is located 200 meters from the entrance of a tunnel and covers an area of ​​approximately 0.5 square meters, the system will mark it as a medium risk. Such a detection process can promptly identify potential safety hazards and provide an accurate basis for subsequent processing. In one possible implementation, after the anomaly detection result is generated, the system will automatically send an alarm message to the management center and capture the relevant video clips and images. Figure 1 And upload it so that staff can quickly verify it. For example, for vehicles that park illegally, the system will provide the specific location and duration of the vehicle's parking. If the vehicle stays in the tunnel for more than 10 minutes, it will be upgraded to a high-level alarm. This multi-dimensional data support helps managers make quick decisions and reduce the probability of safety accidents in the tunnel. Through the above technical means, the rail robot inspection system has achieved full process automation from data collection to anomaly detection, which not only improves inspection efficiency, but also significantly reduces the risks and costs of manual inspections. For example, traditional manual inspections may take several hours to complete a tunnel inspection, while the robotic system can complete the same task within 1 hour with higher accuracy. This technical effect is of great significance to ensuring tunnel operation safety and improving management efficiency.

[0012] S102. If an abnormal condition exists in the abnormality detection result, the abnormal feature data is extracted to generate a structured message, and the structured message is input into a preset quantitative scoring model to calculate a score value. The level of the abnormal condition is determined according to the score value. Among them, there are four levels of abnormal conditions, and the processing priority is level 1 with the highest priority and level 4 with the lowest priority.

[0013] If an abnormality is detected in the anomaly detection results, the system extracts abnormal feature data, including the abnormal image, stake location, and time stamp, to generate a structured message. This structured message is then fed into a pre-set quantitative scoring model, which calculates a score ranging from 0 to 100. The abnormality's grade is determined based on the pre-set scoring threshold and the score value, resulting in graded abnormality data. A scoring threshold is set every 25 points.

[0014] For example, in the scenario of a rail robot inspecting a tunnel, the processing flow for anomaly detection results can be analyzed in detail and illustrated from multiple perspectives to ensure logical rigor and alignment with actual business needs. For example, extracting feature information from abnormal images, pile locations, and timestamps can be understood as structuring key abnormal segments within the raw video data. Suppose, during a tunnel inspection, the system detects a spilled object on the ground. The image shows scattered debris, the pile location is recorded as 300 meters from the tunnel entrance, and the time stamp is 10:15 AM on the same day. After this information is extracted, a data packet is generated containing an image screenshot, the specific location, and the time. This extraction method benefits by consolidating scattered abnormality information into a traceable record, facilitating subsequent analysis and archiving, and providing a basis for rapid problem location. Specifically, the system may compress the image data into a small file, while embedding the pile location and timestamp in text within the message. This message data acts like a brief event report, containing a basic description of the anomaly, making it easier for the system to access and transmit. This processing method improves data transmission efficiency and lays the foundation for further evaluation. For example, when using a quantitative scoring model to assess anomalies, analysis can be conducted from two dimensions: threat level and impact range. The threat level of anomalies is determined by a predefined rule base and categorized into three levels: high, medium, and low, with corresponding base scores: high threat: 70 points; medium threat: 50 points; and low threat: 30 points. The impact range is determined by analyzing anomaly images. The predefined mapping between impact range and score is as follows: impact range ≤ 1 square meter: 20 points; 1 square meter < impact range ≤ 5 square meters: 6 points per square meter; 5 square meters < impact range ≤ 10 square meters: 50 points; > 10 square meters: 70 points. Suppose a rail robot detects spilled debris 300 meters from a tunnel entrance. The image shows scattered debris located in the center of the lane. The rule base matches: anomaly type: spilled debris; location parameter: center of lane. According to the rule base, the anomaly matching "spilled debris - center of lane" has a medium threat level and a base score of 50 points. The impact range, estimated based on image analysis, covers a 2-square-meter area, resulting in a score of 2 × 6 = 12 points. The weight ratio of the threat level and impact range of an abnormality is 7:3 (with the threat level taking precedence). The system combines these two parameters with the structured message to calculate the score using the following formula: Threat Level Score × 70% + Impact Range Score × 30% = 50 × 0.7 + 12 × 0.3 = 35 + 3.6 = 38.6. This scoring mechanism benefits from converting subjective abnormalities into objective numerical values, facilitating unified standards and tiered management. For example, when categorizing abnormalities based on the score, 76-100 points correspond to a Level 1 event, 51-75 points to a Level 2 event, 26-50 points to a Level 3 event, and 0-25 points to a Level 4 event. Therefore, a score of 38.6 corresponds to a Level 3 event, indicating that the abnormality requires prompt attention but is not an emergency.The generation of hierarchical data can help the management center allocate resources appropriately, such as prioritizing Level 1 incidents and arranging routine cleanup for Level 3 incidents. This hierarchical approach effectively improves the targetedness and efficiency of exception handling.

[0015] S103. Apply a deep Q learning network to dynamically adjust the scoring threshold of a preset quantitative scoring model, and re-classify abnormal conditions according to the adjusted scoring threshold to obtain adjusted classification data.

[0016] A deep Q-learning network analyzes the pre-set quantitative scoring model to obtain initial configuration data for the scoring thresholds and determine the baseline range for the scoring thresholds. Based on this initial configuration data, a dynamic adjustment mechanism is used to update the scoring thresholds in real time, generating adjusted scoring thresholds. Abnormal conditions are rescored based on the adjusted scoring thresholds. Based on the rescoring results, adjusted classification data for the abnormal conditions is obtained.

[0017] For example, in the tunnel inspection business scenario, the process of analyzing a quantitative scoring model using a deep Q-learning network can be discussed from both the theoretical and practical perspectives. Deep Q-learning networks are a reinforcement learning method whose core focus is optimizing the threshold configuration of the scoring model through trial and error and feedback. In tunnel inspections, the system might analyze the distribution of abnormality scores based on historical inspection data and initially determine a baseline range for the scoring thresholds. For example, the 0 to 100 score range might be divided into four levels: 0-25, 26-50, 51-75, and 76-100, corresponding to different levels of abnormal severity. This initial configuration lays the foundation for subsequent dynamic adjustments. For example, the process of using a dynamic adjustment mechanism to update the scoring thresholds in real time based on the initial configuration data can be understood as the system continuously revising the threshold range based on new data from actual inspections. In one possible implementation, suppose that during a tunnel inspection, the system detects a certain type of anomaly, such as a minor crack on the tunnel wall, with a score frequently falling between 26 and 30 points. However, during actual processing, the system finds the threat level to be low. In this case, the system might adjust the lower threshold for the corresponding level from 26 to 30 points. This dynamic adjustment mechanism allows the scoring threshold to better align with actual business needs and improve the accuracy of classification. For example, the reassessment of anomalies based on the adjusted scoring threshold can be explained in detail from a specific implementation perspective. Suppose, during an inspection, a water accumulation anomaly is detected in a tunnel. The initial score is 55 points, falling within the 51-75 range. However, after the upper threshold is adjusted to 70 points, the system reassesses the anomaly, considering factors such as the area and location of the water accumulation, to determine whether it still falls within the original classification range. This reassessment process helps avoid classification bias caused by threshold adjustments and ensures more reasonable classification of anomalies. For example, the process of obtaining adjusted classification data based on the reassessment results can be explained from a business logic perspective. During tunnel inspections, if the aforementioned water accumulation anomaly is reassessed and the score is adjusted to 65 points, still falling within the range of 51-70 points, the system will confirm its classification as a Level 3 anomaly and require routine maintenance. This adjustment and confirmation process of graded data can provide a more accurate basis for subsequent resource scheduling, help optimize the priority arrangement of inspection work, and improve overall management efficiency. For example, in the implementation of the entire process, the close connection of each link is particularly important. From initial threshold configuration to dynamic adjustment, to reassessment and updating of graded data, each step is centered around the actual needs of tunnel inspections, ensuring that the determination and handling of abnormal conditions are more scientific and reasonable. This approach can reduce misjudgments and waste of resources while ensuring inspection quality, providing strong support for tunnel safety management.

[0018] S104. Match the preset response strategy combination according to the adjusted hierarchical data. If there is at least one optional strategy combination, start the weighted optimization algorithm, perform comprehensive calculations from the three dimensions of response time, resource utilization, and disposal cost, determine the optimal strategy combination, and send execution instructions to the relevant execution units.

[0019] The abnormality level is obtained from the adjusted classification data, matched with the preset response strategy combination, and a candidate strategy set is generated. Based on the candidate strategy set, all strategy combinations in the set are scored. The score is calculated using a weighted optimization algorithm based on the three dimensions of response time, resource utilization, and disposal cost of each strategy combination to obtain the optimal strategy combination. If the response time of the optimal strategy combination exceeds the preset response time threshold, a supplementary strategy is obtained from the preset backup strategy library and the candidate strategy set is updated. Based on the updated candidate strategy set, the weighted optimization score is recalculated to determine the final strategy combination. Execution instructions are extracted from the final strategy combination to generate an instruction sequence. The instruction sequence is sent to the corresponding execution unit through the preset communication interface.

[0020] For example, in a tunnel inspection scenario, adjusted classification data provides a crucial basis for subsequent responses. After obtaining the event level, the system matches it with pre-defined response strategy combinations to form a candidate strategy set. For example, suppose an inspection discovered a structural deformation in a tunnel, scored 88 points, and classified it as a high-level anomaly. Based on the level, the system then matches three strategy combinations, focusing on rapid response, resource conservation, and cost control, respectively. This matching process ensures targeted strategy selection. For example, within the candidate strategy set, the system uses a weighted optimization algorithm to evaluate the performance of each combination. Suppose the first strategy has a faster response time but lower resource utilization; the second strategy has higher resource utilization but higher costs; and the third strategy has lower costs but lower response speed. The system uses a weighted calculation, comprehensively considering response speed (40%), resource utilization (30%), and handling costs (30%), ultimately selecting the first strategy as the optimal combination. This process demonstrates a balanced strategy selection process. For example, if the response speed of the optimal strategy combination falls below a preset threshold—for example, if the required response time is within 30 minutes but the actual estimated time is 40 minutes—the system will extract a supplementary strategy from the backup strategy library. Suppose one of the strategies in the backup strategy library involves invoking external emergency resources. The system will add it to the candidate set, update it, and recalculate the weighted score to ultimately determine a new strategy that includes invoking external resources. This mechanism ensures efficient response. For example, extracting execution instructions from the final strategy combination and generating an instruction sequence is key to implementing the strategy. In the aforementioned structural deformation case, the instruction sequence might include dispatching two maintenance equipment to a designated location within one hour and notifying monitoring personnel for real-time tracking. This sequence is sent to the execution unit via a pre-set communication interface to ensure accurate instruction delivery. For example, after the instructions are issued, triggering the response process is the final step in the entire process. Suppose, after the instructions are issued to the on-site equipment, the maintenance equipment arrives at the designated location in the tunnel on schedule and begins repair work. Monitoring personnel also confirm progress through system feedback. This process ensures a seamless transition from classification to execution, improving the timeliness of exception handling. For example, it should be noted that the implementation of each of the above steps is closely designed around tunnel inspection requirements. For example, the parameters of the weighted optimization algorithm can be dynamically adjusted based on historical data. If the system finds that the weighting of response speed needs to be increased, it will automatically adjust it to 50% to accommodate emergency situations. This flexibility provides a more reliable guarantee for tunnel safety. For example, in one possible implementation, updating the backup strategy library is also crucial. The system can regularly analyze historical response cases, incorporate effective strategies into the library, and eliminate outdated solutions, ensuring that the strategies in the library always meet actual needs. This continuous optimization provides more possibilities for response in complex scenarios.

[0021] S105. The track robot obtains the execution instruction, moves to the nearest pile number position of the abnormal point, collects the on-site abnormal image data and uploads it to the cloud server, compares it with the pre-stored image database through the deep learning image feature matching algorithm, confirms the abnormality type, and generates a navigation path based on the synchronous positioning and map construction algorithm to drive the robot to move to the abnormal point for the second time.

[0022] The track robot receives an execution instruction, extracts relevant information about the anomaly location and stake position from the instruction, generates task initialization data, and transmits it to the track robot's internal control module, generating a task initiation signal. Based on the task initiation signal, the track robot moves to the stake position closest to the anomaly. Built-in sensors collect on-site environmental data, generate preliminary environmental perception data, and transmit it to the cloud server. An image acquisition module captures the anomaly location from multiple angles, generating an on-site image dataset. This dataset is uploaded to a cloud storage unit and then determines whether the image data clarity meets a preset clarity threshold. If the image data clarity meets the preset clarity threshold, image comparison technology is used to perform feature matching with a pre-stored image database, extract key features of the anomaly type, and generate an anomaly classification result. Based on the anomaly classification result, a navigation path is generated for the anomaly location using simultaneous positioning and mapping technologies. This path data is transmitted to the track robot's drive unit, which drives the robot for a second movement to the anomaly point.

[0023] For example, in a tunnel inspection scenario, after receiving an anomaly alert, the response dispatch system generates an execution instruction including a return mission. This instruction may specifically instruct the rail robot to proceed to stake number K12+300 in the tunnel to address a suspected crack. After interpreting the instruction, the system identifies the anomaly location as K12+300 and generates mission initialization data, including target coordinates and mission priority. This data is transmitted via wireless communication to the rail robot's internal control module, triggering a mission start signal. Upon receiving the signal, the robot activates its drive motors and moves along the track to K12+300. Upon arrival, internal sensors such as lidar and temperature and humidity sensors begin collecting environmental data, such as a temperature of 28°C and a humidity of 65%, generating preliminary environmental perception data and uploading it to a cloud server. The cloud server verifies the data integrity to ensure there are no missing or incorrect values. In one possible implementation, after receiving the environmental perception data, the cloud server activates the image acquisition module. The rail robot, equipped with a 1080p HD camera, captures the anomaly location from multiple angles, such as the front, side, and overhead views of the crack, generating a field image dataset. Suppose 10 images are captured with a resolution of 1920x1080 and uploaded to a cloud storage unit. The system uses an algorithm to determine image clarity, setting a clarity threshold of 80%. If all images meet this threshold, the system proceeds to the next step. Image comparison technology matches the on-site images with crack features stored in a database, such as crack width, depth, and extension direction, to generate an anomaly classification result, determining whether the crack is "minor" or "major." For example, based on the anomaly classification result, the system combines simultaneous positioning and mapping technology to generate a navigation path. Assuming the crack is located on the left side of the tunnel wall, the system calculates a path from the current location to the crack point, avoiding obstacles such as pipes or cables. The path data includes a 30° turning angle and a 5-meter straight-line distance. This path data is transmitted to the robot drive unit, which drives the robot to move to the crack point a second time with centimeter-level positioning accuracy. Once the robot arrives, it can perform further inspection or marking tasks. As can be seen, the above process ensures a closed-loop operation from command parsing to path navigation. Each link is tightly integrated. For example, sensor data provides environmental context for image acquisition, image clarity assessment provides reliable input for anomaly classification, and navigation paths ensure the robot reaches its target point with precision. This design improves inspection efficiency, reduces the need for manual intervention, and provides technical support for tunnel safety management.

[0024] S106. After the track robot arrives at the abnormal point, it activates the nearest broadcast node and plays the pre-recorded voice command. At the same time, it connects to the surveillance cameras around the abnormal point to obtain real-time video stream data, transmits it to the command center through the real-time communication protocol, and establishes a multi-party video communication channel.

[0025] Upon arriving at the anomaly location, the rail robot automatically scans the surrounding environment, identifies the nearest broadcast node, and sends an activation signal via its built-in communication module, triggering the broadcast node to play a pre-recorded voice message, providing preliminary on-site prompts. Based on the broadcast node's activation status, the rail robot simultaneously connects to surveillance equipment around the anomaly location, acquires real-time video data, generates an initial video stream, and verifies the integrity of the video stream. If the initial video stream is complete, it compresses the video data using a pre-set communication protocol, generating a compressed video stream and uploading it to the command center. The command center receives the compressed video stream, decompresses it, and generates restored video data, which is then connected to the video interaction system, establishing a multi-party video interaction channel. Based on the connection status of the video interaction channel, the command center generates command data based on on-site coordination needs and transmits it to the rail robot via the communication protocol, updating task execution parameters. After receiving the command data, the rail robot analyzes the task execution parameters and, based on the real-time video data from the anomaly location, adjusts its position and acquisition angle to generate an updated on-site information stream. The rail robot continuously transmits this updated on-site information stream to the command center, completing a closed-loop process of command transmission and on-site coordination.

[0026] For example, upon arriving at an anomaly, a rail robot automatically scans the surrounding environment and identifies the nearest broadcast node, providing a highly efficient on-site interaction method. For example, in a tunnel inspection scenario, upon arriving at a suspected water seepage point, the robot scans for broadcast nodes within a 5-meter radius and locks onto the nearest node, located at coordinates X:120, Y:50. The robot then sends an activation signal via its built-in 4G module, triggering the node to play a pre-recorded 10-second voice message, "Attention, anomaly detected, do not approach." This method ensures that on-site personnel receive timely warnings. It should be noted that the activation of the broadcast node relies on the robot's precise positioning capabilities, typically based on an environmental map generated by LiDAR with an accuracy of up to 0.1 meter, ensuring accurate signal transmission. The robot then simultaneously connects to monitoring equipment surrounding the anomaly, acquiring real-time video data and generating an initial video stream. For example, a fixed camera installed in the tunnel covers the anomaly. The robot connects to it via Wi-Fi, acquiring a video stream with a resolution of 1280x720 and a frame rate of 30 fps. Preferably, integrity assessment is based on timestamp verification to confirm that the video stream covers the entire anomaly location, providing reliable data for subsequent analysis. Video data compression is a key step, reducing transmission bandwidth requirements. For example, the robot uses the H.264 encoding protocol to compress the initial video stream to one-third its size, generating a compressed video stream, reducing the file size from 100MB to approximately 30MB. The compressed video stream is uploaded to the command center via a 5G network, with a transmission rate of 50Mbps, ensuring real-time data delivery. Upon receipt, the command center decompresses the video data and connects it to the video interaction system. For example, the command center uses dedicated software to decompress the video stream and display it on a multi-screen interactive interface, allowing remote experts to analyze the water seepage point in real time. The command center generates command data based on the video interaction system and updates the robot's task parameters. For example, if the expert discovers a water seepage point at the tunnel roof, the robot's camera angle needs to be adjusted to a 45-degree elevation angle and video needs to be recaptured. The command data is sent via an encrypted protocol and contains specific parameters such as "adjust angle: 45 degrees, capture duration: 15 seconds." After the robot parses the instructions, it adjusts the angle of the robotic arm and regenerates the on-site information stream, maintaining a resolution of 1280x720 and covering the details of the seepage point. Preferably, the information stream is transmitted through the real-time streaming protocol RTSP to maintain low-latency interaction with the command center, with the delay controlled within 200 milliseconds. Through the updated on-site information stream, the robot continues to maintain data interaction with the command center. For example, the 15-second video stream collected by the robot based on the instructions clearly shows the flow rate and range of the seepage point. The command center judges the seepage level based on this and generates further disposal instructions. This closed-loop processing method relies on the efficiency of the video interaction channel to ensure real-time synchronization of instructions and on-site data. It should be noted that the stable communication between the robot and the command center depends on the 5G base stations deployed in the tunnel, with a signal coverage rate of 95%, to ensure the continuity of data transmission.

[0027] S107. During the exception handling process, a message queue telemetry transmission protocol is used to transmit command data, and all message transmission records are stored in a preset data analysis module.

[0028] The service quality level is determined based on the level of the abnormal condition. When the abnormal condition is level 1 or 2, the corresponding service quality level is level 2. When the abnormal condition is level 3 or 4, the corresponding service quality level is level 1. Based on the service quality level, command data is transmitted via a preset message queue telemetry transmission protocol. When the service quality level is level 2, a dedicated transmission path is established. The command data for the abnormal event is distributed based on the dedicated transmission path. Distribution status information is obtained to determine whether the command data has been delivered to the target node. If the command data has not been delivered to the target node, a new transmission request is generated. For the undelivered command data, the command data is distributed via a backup transmission channel to obtain new distribution status information. When the service quality level is level 1, a shared transmission channel is generated. The command data is distributed using preset message routing rules, and distribution status information is obtained. The distribution status information is integrated into a transmission record, which is then stored in a preset data analysis module.

[0029] For example, in a rail robot's abnormality handling scenario, the command transmission mechanism for different event levels can be managed hierarchically through a pre-set message queue telemetry transmission protocol. The core of the message queue telemetry transmission protocol is the structured processing of command data and status information to ensure the orderly and reliable transmission of information. For abnormality handling scenarios, abnormal conditions are classified into specific levels. A level 1 event might be a sudden structural crack in a tunnel, a level 2 event might be an equipment operational anomaly, and level 3 and 4 events might be minor deviations or non-urgent alarms during routine inspections. For example, regarding command priority, the system automatically allocates resources based on the event level. For level 1 and level 2 events, the quality of service level is set to 2, which requires higher transmission priority and dedicated paths. For example, in a level 1 event, a crack is discovered in the tunnel, and the command data is "Immediately stop all equipment in the area." The system generates a dedicated transmission path, prioritizes bandwidth resources, and ensures that the command is delivered in the shortest possible time. If a path is established, the command is distributed along this path to the target node, such as the rail robot or field control equipment, and real-time feedback on the distribution status is provided. If a command is not delivered, for example due to signal interference, the system activates a message retention mechanism, regenerates a delivery request, and switches to a backup dedicated channel to ensure the command ultimately reaches the destination node. For example, for command data for Level 3 and Level 4 events, the quality of service level is set to 1, and the system distributes it via a shared transmission channel. For example, in a Level 3 event, an inspection reveals minor dust accumulation, and the command data is "Start Cleaning Mode." Based on pre-set message routing rules, the system sends the command to the robot via a shared channel. While this approach has a lower priority, it is sufficient for non-emergency scenarios while conserving dedicated resources. Delivery results are recorded in real time to confirm command delivery. Specifically, the integration and storage of delivery records are crucial components of the entire process. The system aggregates delivery status information for each level of event into delivery records. For example, for a Level 1 event, the command delivery time is 5 seconds, while for a Level 3 event, the command delivery time is 10 seconds. These records are stored in the data analysis module for subsequent tracing and optimization of delivery strategies. This layered transmission mechanism not only ensures the priority delivery of emergency commands, but also ensures the stable transmission of routine commands, improving overall efficiency in emergency response scenarios. In one possible implementation, the design of dedicated transmission paths and shared transmission channels fully considers the complex communication environment within tunnels. Dedicated paths prioritize base stations with high signal coverage, while shared channels dynamically allocate remaining bandwidth resources to ensure maximum resource utilization. This approach effectively improves the reliability of command delivery, especially in emergency situations, enabling rapid response to on-site needs.

[0030] S108. Obtain indicator data including response time and handling success rate from message delivery records, dynamically adjust the abnormal situation level, and use the log analysis stack to statistically evaluate the alarm push effect, optimize the push strategy, and regularly update the deep Q learning network parameters based on data accumulation, and continuously improve the response strategy through offline training.

[0031] Obtain the response time consumption and disposal success rate data from the message passing records, calculate the mean response time consumption and success rate distribution, and obtain the real-time performance indicators of abnormal events. If the response time consumption in the real-time performance indicators is higher than the preset time consumption threshold, adjust the abnormal condition level according to the correlation between the time consumption and the success rate, and generate new classification data. Among them, the preset time consumption threshold is determined based on historical data statistics and safety standards: for level 1 events, take the 80th percentile value of the response time consumption of similar events (such as structural cracks) in the past 6 months (such as 0.3 seconds), and calibrate it according to the requirements of 'Emergency Event Response Time Limit' in the 'Safety Regulations for Highway Tunnel Operations'; the thresholds for level 2 / 3 / 4 events are 0.5 seconds, 1.0 seconds, and 2.0 seconds respectively, which are also set by combining historical percentile values with industry standards, and are recalculated and updated quarterly based on new data. The correlation between the time consumption and the success rate is a negative correlation. For example, through Pearson coefficient calculation, the correlation coefficient r = -0.7, that is, for every 0.1 second increase in time consumption, the success rate decreases by an average of 5%, etc. The rule for adjusting the abnormal condition level is: if the real-time response time consumption t > the preset threshold t0, and the corresponding disposal success rate s < the preset success rate threshold s0 (s0 for level 1 = 95%, s0 for level 2 = 90%), then trigger a level adjustment; the specific adjustment method: when t > t0 and s < s0, if the current event is a level 2 event, then upgrade to level 1; if it is a level 3 event and t > 1.2×t0, then upgrade to level 2; after adjustment, generate new classification data, and record the levels, time consumption, success rate, and trigger conditions before and after the adjustment. Extract the alarm push effect indicators from the push log data through log analysis stack, analyze the coverage rate and response rate of the push strategy, and obtain the optimized push strategy configuration. According to the optimized push strategy configuration, update the execution parameters of the alarm push, and generate new push log data. Obtain the push log data and classification data in the past three months, extract the abnormal event characteristics, and obtain the offline training dataset. Through the deep Q-learning algorithm, update the network parameters based on the offline training dataset, and generate an optimized response strategy model. According to the optimized response strategy model, adjust the push strategy configuration to obtain new business execution parameters. Specifically, the optimized response strategy model is a deep Q-network (DQN) model, whose input is abnormal feature data (such as type, location, current level), real-time performance indicators (response time consumption, success rate), and the output is the adjustment threshold of the quantization scoring model (such as the correction value of the scoring threshold for level 1 / level 2 events); the network structure includes 3 fully connected layers (the hidden layer dimensions are 128, 64, and 32 respectively), the activation function uses ReLU, and the output layer uses linear activation to output a continuous threshold adjustment amount.The process of generating a model using the deep Q-learning algorithm is as follows: the state space includes the current anomaly level, the historical average response time, and the success rate of handling the last three similar events; the action space includes the adjustment step size of the scoring threshold (e.g., ±5 points, with each adjustment not exceeding 20% ​​of the original threshold); the reward function is R = α×(1 - response time / threshold) + β×success rate (α=0.6, β=0.4, reward range [-1,1]); the training process is as follows: using a historical three-month offline dataset (containing 5000+ anomaly event records), an ε-greedy strategy (ε linearly decays from 0.9 to 0.1) is used to select actions, the target network is updated every 1000 steps, and the Q value loss is minimized using the Adam optimizer (learning rate 0.001). It is iteratively trained until the loss function (MSE) stabilizes below 0.01, and finally a converged response strategy model is generated.

[0032] For example, in the abnormal event handling scenario of tunnel inspections, obtaining response time and handling success rate data from message transmission logs is a key step in optimizing system performance. The average response time can be determined by analyzing the duration from sending to receiving each command. For example, suppose the records show an average response time of 0.5 seconds over a certain period. The success rate distribution reflects the effectiveness of handling abnormal events at different levels, such as a 98% success rate for level 1 events and a 90% success rate for level 4 events. These real-time performance metrics provide a basis for subsequent adjustments. If the response time exceeds the preset threshold of 0.3 seconds, the system analyzes correlations based on the success rate data and dynamically adjusts the abnormal classification threshold. For example, some level 2 events may be upgraded to level 1 to ensure faster response, and new classification rule data is generated. For example, by extracting alarm push performance metrics from push logs using the ELK stack, the coverage and response rate of the push policy can be analyzed. The coverage rate reflects whether the alarm reaches all relevant nodes, while the response rate measures the actual percentage of nodes that respond to the alarm. If the analysis finds a coverage rate of 95% but a response rate of only 80%, the system will optimize the push policy configuration, such as increasing the push frequency or adjusting the push channel, generating new execution parameters, and updating the log data. This approach can effectively improve alert reach and feedback efficiency. For example, by acquiring push logs and classification rule data from the past three months, the system extracts abnormal event features to form an offline training dataset, which serves as the foundation for further optimization. Features may include information such as event type, occurrence time, and node location. Using a deep Q-learning algorithm, the system updates network parameters based on the dataset and generates an optimized response strategy model. For example, if model training reveals that responses to nighttime events take longer, the system will adjust its strategy to prioritize more resources for nighttime tasks. This approach allows the system to better adapt to real-world scenarios. For example, by adjusting abnormality classification rules and push strategy configuration based on the optimized response strategy model, new service execution parameters can be generated. Specifically, if the model indicates a low response rate for Level 3 events, the system may prioritize push notifications in certain scenarios and optimize push routing to ensure rapid delivery of instructions. This adjustment can significantly improve overall processing efficiency, especially in resource-constrained environments. For example, in one possible implementation, the combined application of the above methods can form a complete optimization loop. From real-time performance indicator monitoring to offline training dataset construction, to policy model updates and parameter adjustments, each step is closely centered around the tunnel inspection scenario. For example, if an analysis reveals that the response time to a Level 1 water seepage event is extended to 0.8 seconds due to signal issues, the system will review historical data and, combined with model recommendations, prioritize base stations with more stable signals for command transmission. This closed-loop mechanism continuously improves the system's adaptability and reliability, providing strong support for rapid response to abnormal events.

[0033] Reference Figure 2In the embodiment of the present application, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 2 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as on-site abnormal image data and message delivery records. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements the personalized recommendation method based on real-time user behavior of any of the above embodiments.

[0034] Those skilled in the art will understand that Figure 2 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.

[0035] It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above and that the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the present application is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A tunnel anomaly detection and response method based on a track robot, characterized in that: The method comprises: The track robot performs scheduled inspection tasks along the preset track, and is equipped with a camera with a preset resolution to collect panoramic video stream data in the tunnel. The acquired video stream is transmitted to the edge computing node in real time, and the video stream is analyzed for anomaly detection through a lightweight artificial intelligence visual model to obtain anomaly detection results; if there is an abnormal condition in the anomaly detection result, the abnormal feature data is extracted to generate a structured message, and the structured message is input into the preset quantitative scoring model to calculate the score value, and the level of the abnormal condition is determined according to the score value. Among them, there are four levels of abnormal conditions, with the processing priority of level one being the highest and level four being the lowest; the deep Q learning network is used to dynamically adjust the scoring threshold of the preset quantitative scoring model, and the abnormal conditions are re-classified according to the adjusted scoring threshold to obtain adjusted classification data; the preset response strategy combination is matched according to the adjusted classification data, and if there is at least one optional strategy combination, the weighted optimization algorithm is started. , a comprehensive calculation is performed from the three dimensions of response time, resource utilization and disposal cost, the optimal strategy combination is determined, and the execution instructions are sent to the relevant execution units; the rail robot obtains the execution instruction, moves to the nearest pile number position of the anomaly point, collects the on-site abnormal image data and uploads it to the cloud server, and compares it with the pre-stored image database through the deep learning image feature matching algorithm to confirm the type of anomaly. At the same time, a navigation path is generated based on the synchronous positioning and map construction algorithm to drive the robot to move to the anomaly point for the second time; after the rail robot arrives at the anomaly point, it activates the nearest broadcast node and plays the pre-recorded voice command. At the same time, it connects to the surveillance cameras around the anomaly point to obtain real-time video stream data, which is transmitted to the command center through the real-time communication protocol to establish a multi-party video communication channel; during the anomaly handling process, the message queue telemetry transmission protocol is used to transmit command data, and all message transmission records are stored in the preset data analysis module; Obtain indicator data including response time and handling success rate from message transmission records, dynamically adjust the abnormal situation level, and use the log analysis stack to statistically evaluate the alarm push effect, optimize the push strategy, and regularly update the deep Q learning network parameters based on data accumulation, and continuously improve the response strategy through offline training.

2. The tunnel anomaly detection and response method based on a track robot according to claim 1 is characterized in that: The track robot performs scheduled inspection tasks along the preset track, and is equipped with a camera with a preset resolution to collect panoramic video stream data in the tunnel. The acquired video stream is transmitted to the edge computing node in real time, and anomaly detection and analysis of the video stream is performed using a lightweight artificial intelligence vision model to obtain anomaly detection results, including: The track robot conducts inspections along the preset track. During the inspection, the track robot uses a camera with a preset resolution to collect panoramic video data in the tunnel environment, obtain raw video stream information, and transmit it to the edge computing node for preliminary storage and processing to obtain basic video data. Based on the basic video data, a lightweight artificial intelligence visual model is used to decompose the video stream frame by frame, extract the image features of each frame, and determine the key area information in the image based on the image features; For key area information, the anomaly detection process is executed to obtain anomaly detection results.

3. The tunnel anomaly detection and response method based on a track robot according to claim 1 is characterized in that: If an abnormality is found in the abnormality detection result, the abnormality feature data is extracted to generate a structured message, and the structured message is input into a preset quantitative scoring model to calculate a score value. The level of the abnormality is determined according to the score value. There are four levels of abnormality, with the processing priority level 1 being the highest and level 4 being the lowest, including: If an abnormal condition is detected in the abnormality detection result, the abnormal feature data including the abnormal image, stake position and time stamp is extracted to generate a structured message; Input the structured message into the preset quantitative scoring model to calculate the score value ranging from 0 to 100 points; The level of the abnormal situation is determined according to the preset scoring threshold and scoring value, and the graded data of the abnormal situation is obtained, wherein a scoring threshold is set every 25 points.

4. The tunnel anomaly detection and response method based on a track robot according to claim 1 is characterized in that: The deep Q learning network is applied to dynamically adjust the scoring threshold of the preset quantitative scoring model, and the abnormal conditions are re-classified according to the adjusted scoring threshold to obtain adjusted classification data, including: Analyze the preset quantitative scoring model through the deep Q learning network to obtain the initial configuration data of the scoring threshold and determine the baseline range of the scoring threshold; Based on the initial configuration data, a dynamic adjustment mechanism is used to update the scoring threshold in real time to obtain the adjusted scoring threshold; Re-score abnormal conditions based on the adjusted scoring threshold; Based on the results of the re-scoring, the adjusted classification data of the abnormal condition is obtained.

5. The tunnel anomaly detection and response method based on a track robot according to claim 1 is characterized in that: The preset response strategy combination is matched according to the adjusted hierarchical data. If at least one optional strategy combination exists, a weighted optimization algorithm is started to perform a comprehensive calculation based on the three dimensions of response time, resource utilization, and disposal cost to determine the optimal strategy combination, and an execution instruction is sent to the relevant execution unit, including: Obtain the abnormal condition level from the adjusted classification data, match it with the preset response strategy combination, and generate a candidate strategy set; Based on the candidate strategy set, all strategy combinations in the set are scored. The score is calculated comprehensively from the three dimensions of response time, resource utilization and disposal cost of each strategy combination using a weighted optimization algorithm to obtain the optimal strategy combination. If the response time of the optimal strategy combination is greater than the preset response time threshold, a supplementary strategy is obtained from the preset backup strategy library to update the candidate strategy set; Based on the updated candidate strategy set, recalculate the weighted optimization score and determine the final strategy combination; Extract execution instructions from the final strategy combination and generate an instruction sequence; The instruction sequence is sent to the corresponding execution unit through the preset communication interface.

6. The tunnel anomaly detection and response method based on a track robot according to claim 1 is characterized in that: The track robot receives the execution instruction, moves to the nearest stake number position of the abnormal point, collects the abnormal image data on site and uploads it to the cloud server, compares it with the pre-stored image database through the deep learning image feature matching algorithm, confirms the abnormal type, and generates a navigation path based on the synchronous positioning and map construction algorithm to drive the robot to move to the abnormal point for the second time, including: The track robot obtains the execution instruction, extracts the relevant information of the abnormal position and the pile number position from the execution instruction, generates the task initialization data, transmits it to the internal control module of the track robot, and obtains the task start signal; According to the mission start signal, the track robot moves to the nearest pile number position of the abnormal situation, uses the built-in sensor to collect on-site environmental data, generates preliminary environmental perception data, and transmits it to the cloud server; The image acquisition module is used to shoot the abnormal location from multiple angles to generate an on-site image data set, which is uploaded to the cloud storage unit to determine whether the clarity of the image data reaches the preset clarity threshold; If the clarity of the image data reaches the preset clarity threshold, feature matching is performed with the pre-stored image database through image comparison technology to extract the key features of the abnormal type and generate the abnormal classification result; Based on the anomaly classification results, combined with synchronous positioning and map construction technology, a navigation path for the abnormal location is generated, and the path data is transmitted to the track robot drive unit to drive the robot to move to the abnormal point for the second time.

7. The tunnel anomaly detection and response method based on a track robot according to claim 1 is characterized in that: When the track robot reaches the abnormal point, it activates the nearest broadcast node, plays pre-recorded voice commands, and simultaneously connects to the surveillance cameras around the abnormal point to obtain real-time video stream data, which is transmitted to the command center via a real-time communication protocol to establish a multi-party video communication channel, including: After the track robot arrives at the abnormal point, it automatically scans the surrounding environment, identifies the nearest broadcast node, sends an activation signal through the built-in communication module, drives the broadcast node to play a pre-recorded voice message, and completes the initial on-site prompt; Based on the activation status of the broadcast node, the track robot synchronously connects to the monitoring equipment around the abnormal point, obtains real-time video data, generates the initial video stream, and determines whether the video stream is complete; If the initial video stream is complete, the video data is compressed using a preset communication protocol to generate a compressed video stream, which is then uploaded to the command center. The command center receives compressed video streams, decompresses them, generates restored video data, connects to the video interaction system, and builds a multi-party video interaction channel. Based on the connection status of the video interaction channel, the command center generates command data based on on-site coordination needs, sends it to the track robot through the communication protocol, and updates the task execution parameters; After receiving the command data, the track robot analyzes the task execution parameters, combines the real-time video data of the abnormal point, adjusts its own position or acquisition angle, and generates an updated on-site information flow; Through the updated on-site information flow, the rail robot continuously transmits it to the command center, completing the closed-loop processing of command transmission and on-site coordination.

8. The tunnel anomaly detection and response method based on a track robot according to claim 1, characterized in that: During the exception handling process, the message queue telemetry transmission protocol is used to transmit command data, and all message transmission records are stored in a preset data analysis module, including: Determine the service quality level according to the level of the abnormal state, wherein when the abnormal state level is level 1 or level 2, the corresponding service quality level is level 2, and when the abnormal state level is level 3 or level 4, the corresponding service quality level is level 1; According to the service quality level, the command data is transmitted through the preset message queue telemetry transmission protocol, where: When the service quality level is level 2, a dedicated transmission path is established. The command data for abnormal events is distributed based on the dedicated transmission path, and the distribution status information is obtained to determine whether the command data has been delivered to the target node. If the command data has not been delivered to the target node, a new transmission request is generated. The command data that has not been delivered is distributed through the backup transmission channel to obtain new distribution status information. When the service quality level is level 1, a shared transmission channel is generated, command data is distributed through the preset message routing rules, and distribution status information is obtained; The distribution status information is integrated into a delivery record, and the delivery record is stored in a preset data analysis module.

9. The tunnel anomaly detection and response method based on a track robot according to claim 1, characterized in that: The system obtains indicator data including response time and handling success rate from message transmission records, dynamically adjusts the abnormal situation level, and uses the log analysis stack to statistically evaluate the alarm push effect, optimize the push strategy, and regularly update the deep Q learning network parameters based on data accumulation. The response strategy is continuously improved through offline training, including: Obtain response time and handling success rate data from message delivery records, calculate the mean response time and success rate distribution, and obtain real-time performance indicators for abnormal events; If the response time in the real-time performance indicator is higher than the preset time threshold, the abnormal situation level is adjusted according to the correlation between the time and the success rate, and new classification data is generated; Extract alert push effect indicators from push log data through the log analysis stack, analyze the coverage and response rate of the push strategy, and obtain the optimized push strategy configuration; According to the optimized push policy configuration, update the execution parameters of the alarm push and generate new push log data; Obtain push log data and classification data from the past three months, extract abnormal event features, and obtain an offline training dataset; Through the deep Q-learning algorithm, the network parameters are updated based on the offline training data set to generate an optimized response strategy model; According to the optimized response strategy model, the push strategy configuration is adjusted to obtain new business execution parameters.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images