Monitoring and early warning management system and method for intelligent traffic signal box

Abstract

The application discloses a traffic intelligent signal box monitoring and early warning management system and method, and belongs to the field of intelligent traffic equipment and Internet of Things monitoring. In view of the technical problems of low data collection and transmission efficiency under abnormal events, over-reliance on manual review resulting in response lag, limited camera field of view and operation and maintenance resource constraints resulting in inflexible scheduling, the system comprises an end-side collection module, an adaptive communication module, a man-machine cooperative review module and an intelligent scheduling module; through the posture change amount, frequency sampling is triggered, and key posture data is downgraded and transmitted in a weak network; a fault diagnosis list with a priority score is generated by a first artificial intelligence agent; man-machine cooperative review is realized by cooperating with review countdown and automatic processing of overtime; and by a second artificial intelligence agent, state uncertainty determination is performed based on the camera field of view limitation and the operation and maintenance resource state, and formal maintenance work orders or preliminary verification tasks are distributed to corresponding terminals. The application is used for traffic signal box fault monitoring and early warning and operation and maintenance work order intelligent scheduling.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation equipment and Internet of Things monitoring technology, and more specifically, to a monitoring and early warning management system and method for intelligent traffic signal boxes. Background Technology

[0002] Traffic signal boxes are an indispensable infrastructure in urban traffic management systems, integrating a large number of communication, power supply, and control devices. With the expansion of urban traffic networks, the number of traffic signal boxes is constantly increasing, and their operational status directly affects the safety and smooth flow of road traffic.

[0003] Traditional traffic signal box management mainly faces the following pain points: First, traditional equipment has a low level of intelligence and lacks effective real-time monitoring methods. When the cabinet is tilted, collided, flooded, leaking electricity, or the door is opened abnormally, it often needs to be detected by manual inspection or public alarms, which has a serious lag and can easily lead to facility damage or even safety accidents.

[0004] Secondly, although some upgraded intelligent traffic signal boxes have introduced sensor and network communication functions, enabling them to send alarm data to the monitoring center, low data processing and transmission efficiency still exists in practical applications. Especially in the event of transient abnormal events such as collisions, conventional low-frequency fixed sampling often fails to capture critical abnormal state data. Furthermore, when abnormal events cause communication network signal attenuation, existing full-data (such as video stream) transmission methods are prone to network congestion or data loss, preventing the cloud from obtaining crucial early warning information in a timely manner.

[0005] Third, existing monitoring systems rely excessively on manual monitoring and verification. When faced with large-scale fault alarms, fatigue of night shift staff, or shift handover, problems such as untimely processing of alarm information and delayed work order processing are likely to occur. Furthermore, relying solely on the system's fully automated work order dispatch is prone to ineffective consumption of maintenance resources due to false alarms from sensors, lacking an effective human-machine collaboration and fallback mechanism.

[0006] Fourth, regarding fault verification and work order dispatch, existing intelligent traffic signal boxes are typically only equipped with cameras facing the door opening area to record images of people opening the door. When the door is closed, the monitoring center cannot directly confirm the physical damage to the exterior of the box through the internal cameras. When a suspected collision or tipping alarm occurs, the system can usually only passively dispatch the task to official maintenance personnel for on-site verification. In situations where official maintenance resources are limited or the incident location is far away, this single dispatch mode leads to excessively long response times, making it difficult to meet the actual need for rapid confirmation and handling of traffic facility faults.

[0007] In summary, how to provide a traffic intelligent signal box monitoring and early warning management system that can adaptively acquire and transmit key data when anomalies occur, has an efficient human-machine collaborative review mechanism, and can flexibly schedule traffic based on hardware field-of-view limitations and operational resource status is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to provide a monitoring and early warning management system and method for intelligent traffic signal boxes, in order to solve the technical problems in the prior art, such as low data collection and transmission efficiency of traffic signal boxes under abnormal events, excessive reliance on manual review leading to delayed response, and inflexible fault verification scheduling due to limited field of view of internal cameras and constraints of official operation and maintenance resources.

[0009] To achieve the above objectives, the technical solution of the present invention is as follows: In a first aspect, the present invention provides a monitoring and early warning management system for intelligent traffic signal boxes, comprising an end-side data acquisition module, an adaptive communication module, a human-machine collaborative review module, and an intelligent scheduling module; The end-side acquisition module is installed inside the intelligent traffic signal box and includes an attitude sensor and an environmental sensor. It is configured to synchronously acquire environmental parameters and attitude data at a first sampling frequency, and when it is determined that the change in the attitude data exceeds a preset frequency conversion trigger threshold, it triggers the frequency conversion sampling control logic to instruct the attitude sensor to acquire attitude data during the abnormal event at a second sampling frequency greater than the first sampling frequency. The adaptive communication module is connected to the end-side acquisition module and is configured to acquire the signal strength indication value of the current wireless network, and execute a downgrade transmission command when the signal strength indication value is lower than a preset communication threshold to block video stream data, encapsulate the attitude data during the abnormal event into a payload in variable-length byte array format, and send it to the cloud server through the Internet of Things communication protocol; The human-machine collaborative review module is deployed on the cloud server and is configured to receive the payload and call the first artificial intelligence agent to generate a fault diagnosis list with anomaly priority scores. Based on the anomaly priority scores, it renders visually prominent elements corresponding to the alarm level on the large screen visualization display terminal and simultaneously starts the review countdown cycle corresponding to the alarm level. The intelligent scheduling module is connected to the human-machine collaborative review module and is configured to convert the fault diagnosis list into a work order based on the manual intervention instruction when a manual intervention instruction is received within the review countdown period, and to trigger an automatic timeout processing mechanism when the review countdown period is reached and no manual intervention instruction is received, automatically adopt the fault diagnosis list and generate the work order, and call the second artificial intelligence agent to perform automated allocation of the work order.

[0010] Furthermore, the edge acquisition module also includes a microcontroller unit and a four-wire bus. The environmental sensor is connected to the microcontroller unit via the four-wire bus. The microcontroller unit is configured to acquire and store the initial zero bias data of the attitude sensor when the device is first powered on, and use the initial zero bias data as a reference for subsequent calculation of attitude changes. When the adaptive communication module executes a degradation transmission command, the microcontroller unit sends a power cut-off signal via the four-wire bus to stop the power supply to unnecessary environmental sensors.

[0011] Furthermore, the cloud server is also equipped with a large-screen visualization display terminal, and the system also includes an official maintenance terminal and a third-party crowdsourcing terminal that establish a two-way communication connection with the cloud server; The large-screen visualization display terminal includes a map rendering submodule and an anomaly details interaction submodule. The map rendering submodule is configured to load a geographic information layer and generate corresponding spatial positioning anchor points based on the registration ledger of the intelligent traffic signal box. The anomaly details interaction submodule is connected to the human-machine collaborative review module and is configured to parse alarm icons of different levels, drive the corresponding spatial positioning anchor points to present preset color features and scaling ratios, and overlay a digital timer containing the review countdown period and a trigger control for receiving manual intervention commands. The official operation and maintenance terminal includes a QR code parsing submodule, a satellite positioning submodule, and a work order verification submodule. The QR code parsing submodule is configured to read and parse the two-dimensional matrix code attached to the outside of the intelligent traffic signal box through photosensitive hardware to generate a device authentication request. The satellite positioning submodule is configured to obtain the measured latitude and longitude coordinates on site and perform geofence matching verification between the measured latitude and longitude coordinates and the reference coordinates corresponding to the dispatched work order. The work order verification submodule is configured to obtain real-scene image data after the device authentication is passed and the geofence matching verification is passed, and send the real-scene image data and closed-loop status instructions back to the intelligent dispatch module. The third-party crowdsourcing terminal is configured to receive preliminary verification tasks and send back external status confirmation data.

[0012] Furthermore, the intelligent traffic signal box is equipped with a camera facing the door opening area. The camera is used to collect image information of the person opening the door when the door is open. When the door is closed, the field of view of the camera is insufficient to cover the external state of the box.

[0013] A second aspect of the present invention provides a monitoring and early warning management method for a traffic intelligent signal box, applied to the above-mentioned system, the method comprising the following steps: The system receives multi-source sensor data uploaded by the end-side acquisition module of the intelligent traffic signal box, and calls the first artificial intelligence agent to perform feature analysis on the multi-source sensor data to generate a fault diagnosis list containing anomaly priority scores. Based on the anomaly priority score, the corresponding visual highlighting element for the alarm level is rendered on the large screen visualization display terminal, and the review countdown cycle corresponding to the alarm level is started simultaneously. If a manual intervention instruction is received within the review countdown period, a work order will be initiated based on the manual intervention instruction; if the review countdown period is reached and no manual intervention instruction is received, the fault diagnosis list will be automatically adopted and a work order will be initiated. After the work order is triggered, the second AI agent is invoked to determine the uncertainty of the work order execution status based on the fault diagnosis list, the field of view of the camera inside the traffic intelligent signal box when the box door is closed, and the current availability status of the official maintenance terminal. If the external status of the fault can be confirmed and the official maintenance terminal resources meet the conditions for immediate dispatch, a formal maintenance work order will be generated. If the external status of the fault can be confirmed but the official maintenance terminal resources are insufficient, or if the external status of the fault cannot be directly confirmed by data collected inside the box, a preliminary verification task is generated. The formal maintenance work order or the preliminary verification task is matched and dispatched to the corresponding terminal.

[0014] Furthermore, the step of the first artificial intelligence agent performing feature analysis and generating a fault diagnosis list specifically includes: based on the hardware error features in the multi-source sensor data, combined with a preset equipment model and fault map knowledge base, generating structured data containing checklist-style on-site maintenance steps.

[0015] Furthermore, when generating the preliminary verification task, the method further includes: extracting verification sub-items from the checklist-style on-site maintenance steps that are pre-marked as being used only for external observation and status confirmation and can be executed without opening the door; encapsulating the verification sub-items into crowdsourced verification instructions and dispatching them to the third-party crowdsourcing terminal.

[0016] Furthermore, the official maintenance terminal displays the checklist-style on-site maintenance steps on the display interface. The checklist-style on-site maintenance steps include multiple maintenance sub-items. Each maintenance sub-item is configured with a first identifier for representing the inspection completion status and a second identifier for representing the actual fault resolution node. The official maintenance terminal obtains the triggering instructions from the on-site maintenance personnel for the first identifier and the second identifier. Under the condition that the first identifier corresponding to all maintenance sub-items is confirmed and the second identifier corresponding to at least one maintenance sub-item is triggered, the maintenance sub-item corresponding to the triggered second identifier is locked as the target node that actually caused the problem. A structured feedback list is generated and sent to the cloud server so that the first artificial intelligence agent can update the generation basis of the subsequent fault diagnosis list.

[0017] Furthermore, the method also includes a task closure confirmation step, which includes: receiving external status confirmation data returned by the third-party crowdsourcing terminal and handing it over to the second artificial intelligence agent for further analysis; when it is determined that official maintenance personnel need to intervene, extracting the remaining unpacking repair and component replacement sub-items from the checklist-style on-site maintenance steps to generate the formal maintenance work order, and dispatching the formal maintenance work order to the official maintenance terminal.

[0018] Furthermore, the method also includes role-based access control and operation auditing steps, as well as historical fault data aggregation and trend analysis steps. In the step of triggering work order flow, only users with the monitoring center auditing role are allowed to issue manual intervention commands, and the issuance time, user identifier, and corresponding content of the structured feedback list are recorded in the system log. Within a preset time period, closed-loop archived work orders are aggregated, a statistical analysis report is generated, and pushed to the large screen visualization display terminal.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention addresses the problem of insufficient data capture capability in conventional systems during sudden anomalies by introducing threshold-triggered variable-frequency sampling logic at the edge. When an abnormal event such as a collision or tipping over occurs in a traffic intelligent signal box, the system can automatically increase the sampling frequency based on changes in attitude data. This allows the system to maintain low-power, low-frequency monitoring in normal operation, while acquiring high-density attitude data at the moment an anomaly occurs. Simultaneously, the adaptive communication module can actively block the video stream when the network signal weakens, transmitting only the high-density attitude data as a variable-length payload. This degraded transmission mechanism ensures that even in harsh network environments, the most critical early warning data can still be stably and quickly delivered to the cloud, avoiding link congestion and alarm omissions caused by full data transmission.

[0020] This invention constructs a fault-tolerant human-machine collaborative review mechanism. After receiving data in the cloud, an AI agent performs preliminary feature analysis and generates a fault list with priority scores. This list is then presented on a large screen using different levels of visual elements, and a corresponding countdown cycle is initiated. This approach allows monitoring center staff to manually confirm high-priority alarms, thus filtering out potential false alarms from sensors. More importantly, the system includes an automatic timeout handling mechanism. In scenarios where human response may be delayed, such as night shift fatigue, large-scale fault outbreaks, or shift changes, the system automatically adopts the diagnostic list and triggers work order processing once the countdown threshold is reached. This design preserves the accuracy of human decision-making while providing a time safety net through automation, preventing critical faults from being indefinitely postponed.

[0021] This invention proposes a dynamic work order dispatch strategy based on a combination of hardware physical limitations and resource availability. Considering that the internal cameras of intelligent traffic signal boxes typically face the door opening, and cannot capture images of external damage when the door is closed, the system, after triggering a work order flow, comprehensively assesses whether the external state of the fault can be directly confirmed and the availability of current official maintenance resources. If the external state cannot be confirmed, or if the external state can be confirmed but official maintenance resources are insufficient to support immediate dispatch, the system generates a preliminary verification task limited to external observation and dispatches it to a third-party crowdsourcing terminal. This mechanism uses non-professional crowdsourcing personnel as rapid external sensing nodes, leveraging their rapid response to complete preliminary state confirmation without involving dangerous opening operations. This effectively breaks through the scheduling bottleneck caused by relying solely on official maintenance personnel to verify each instance in the traditional model, achieving efficient hierarchical utilization of maintenance resources.

[0022] This invention also achieves continuous iteration of diagnostic logic through a closed-loop verification and feedback mechanism. When executing on-site tasks, the official maintenance terminal performs dual geofence verification via scanning a two-dimensional matrix code and satellite positioning, ensuring the authenticity of on-site handling. Simultaneously, by setting dual identifiers for the inspection completion status and the actual fault resolution node in the checklist-style maintenance steps, the system can accurately pinpoint the target node that actually caused the fault. This closed-loop data, confirmed by on-site personnel, is transmitted back to the cloud and written into the fault graph knowledge base, enabling the system to continuously refine the feature analysis rules of the AI ​​agent using real engineering feedback, thereby gradually improving the accuracy of fault diagnosis and dispatch decisions over long-term operation. Attached Figure Description

[0023] Figure 1 Example 1: Framework diagram of the monitoring and early warning management system for intelligent traffic signal boxes.

[0024] Figure 2 Example 1: Schematic diagram of a large-screen visualization display terminal.

[0025] Figure 3 Flowchart of the entire Example 2.

[0026] Figure 4 Example 2: Schematic diagram of the first stage process.

[0027] Figure 5 Example 2: Flowchart of the third stage.

[0028] Figure 6 Example 2: Schematic diagram of the official operation and maintenance terminal display interface.

[0029] Figure 7 Example 2: Cockpit module display diagram. Detailed Implementation

[0030] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0031] Example 1.

[0032] like Figure 1 As shown, this embodiment provides a monitoring and early warning management system for intelligent traffic signal boxes. The system is built on a cloud-native architecture, with its cloud server using a backend program developed based on Java and the Spring Boot framework. It communicates with edge devices via IoT communication protocols such as MQTT in full-duplex mode and provides services to large-screen visualization terminals and official maintenance terminals through RESTful application programming interfaces. The system mainly includes an edge data acquisition module, an adaptive communication module, a human-machine collaborative review module, and an intelligent scheduling module.

[0033] The end-side data acquisition module is located inside the intelligent traffic signal box. Its core hardware includes a microcontroller unit and attitude and environmental sensors connected in series with the microcontroller unit via a four-wire bus. To comprehensively cover the operating conditions inside and around the box, the environmental sensors are specifically configured to include temperature and humidity sensors, door magnetic sensors, water immersion sensors, electrical parameter detection modules, and fire smoke sensors, among other expansion components. Accordingly, the environmental parameters synchronously acquired by the microcontroller unit include not only the signal strength indication value of the wireless network, but also the real-time temperature, relative humidity, door opening / closing status, overflow or leakage status, power supply voltage and current, and smoke concentration inside the box. For example, the normal operating temperature range can be set to -20°C to 60°C. The electrical interface of the four-wire bus is specifically defined as positive power, ground power, positive data, and negative data.

[0034] During the initial power-on initialization of the device, the microcontroller acquires and stores the initial zero-bias data of the attitude sensor in non-volatile memory, serving as a reference for subsequent calculations of attitude changes. Under normal, non-abnormal operating conditions, the microcontroller synchronously acquires the aforementioned multi-dimensional environmental parameters and attitude data (i.e., multi-source sensor data) at a first sampling frequency, for example, set to one hertz, to maintain low-power operation. The microcontroller continuously calculates the change between the current attitude data and the reference reference, and triggers the frequency conversion sampling control logic when it determines that the change in attitude data exceeds a preset frequency conversion trigger threshold. For example, when an instantaneous change in tilt angle in any axis is detected to be greater than fifteen degrees, the microcontroller instructs the attitude sensor to acquire attitude data during the abnormal event at a second sampling frequency greater than the first sampling frequency, for example, set to twenty hertz, thereby acquiring high-density transient attitude information.

[0035] The adaptive communication module communicates with the edge-side acquisition module to ensure data transmission reliability under extreme conditions. The adaptive communication module is configured to acquire the signal strength indicator value of the current wireless network in real time. When the signal strength indicator value is lower than a preset communication threshold, for example, a received signal strength indicator value less than -85 dB / mW, the system determines that the current network is in a weak or congested state. At this time, the adaptive communication module executes a degraded transmission command. When this command is executed, the microcontroller sends a power cut-off signal through a four-wire bus to stop the power supply to unnecessary environmental sensors, namely the temperature and humidity sensor, smoke sensor, and the camera inside the enclosure, thereby actively blocking the real-time streaming protocol video stream data that occupies a large bandwidth. Subsequently, the adaptive communication module encapsulates the attitude data acquired during the high-frequency abnormal event into a payload in a variable-length byte array format and sends it to the cloud server first through the IoT communication protocol.

[0036] The human-machine collaborative auditing module is deployed on a cloud server and configured to receive the payload in the aforementioned variable-length byte array format and invoke the first AI agent for feature parsing. The first AI agent is implemented through an inference engine running on the cloud server. It receives structured input fields, such as attitude offset angle, equipment online status, door status, temperature and humidity values, and network signal strength. The first AI agent processes the input fields according to a preset diagnostic model and generates a fault diagnosis list with anomaly priority scores. This fault diagnosis list is output in a lightweight data exchange format, and the output content includes at least an anomaly priority score field, an alarm level field, and a field maintenance step field containing multiple maintenance sub-items.

[0037] like Figure 2As shown, the cloud server is also equipped with a large-screen visualization display terminal, which is built based on a responsive front-end framework such as Vue or React. The large-screen visualization display terminal includes a map rendering submodule and an anomaly details interaction submodule. The map rendering submodule is configured to load a geographic information layer and generate corresponding spatial positioning anchor points based on the registration ledger of the traffic intelligent signal box. The anomaly details interaction submodule is connected to the human-machine collaborative review module and receives fault diagnosis list data pushed from the cloud through a status management tool. When rendering the visually highlighted elements corresponding to the alarm level, the front-end program parses the alarm level field and uses parameters to drive view updates. For example, when the alarm is determined to be the highest level, the front-end program enlarges the corresponding spatial positioning anchor point by modifying the scaling ratio parameter of the map component, and at the same time controls the hexadecimal color value of the anchor point to alternate between red and transparent by dynamically binding the keyframe animation property in the Cascading Style Sheet, thereby achieving a visual red flashing effect. In addition, the anomaly details interaction submodule renders an overlaid digital timer component in the interface DOM tree, starts the review countdown cycle corresponding to the alarm level, and renders a button trigger control for receiving manual intervention instructions.

[0038] The intelligent scheduling module is connected to the human-machine collaborative review module. It is configured to convert the fault diagnosis list into a work order based on a manual intervention command issued by a user at the monitoring center via a button trigger control within the review countdown period. If the digital timer reaches the review countdown period and no manual intervention command is received, the intelligent scheduling module triggers an automatic timeout handling mechanism. The background program automatically changes the status of the fault diagnosis list to "accepted," generates a work order, and calls a second AI agent to automatically allocate the work order.

[0039] To achieve a closed-loop process for work order dispatching, the system also includes an official maintenance terminal and a third-party crowdsourcing terminal that establish bidirectional communication connections with the cloud server. The official maintenance terminal is a mobile maintenance page built on Hypertext Markup Language 5, comprising a QR code parsing submodule, a satellite positioning submodule, and a work order verification submodule. The QR code parsing submodule is configured to read the two-dimensional matrix code attached to the outside of the intelligent traffic signal box using the mobile device's photosensitive hardware, and initiate a device authentication request to the cloud. The satellite positioning submodule is configured to call the mobile terminal's underlying interface to obtain the measured latitude and longitude coordinates at the site, and perform geofence matching verification against the reference coordinates corresponding to the dispatched work order. The work order verification submodule is configured to, after successful device authentication and geofence matching verification, open the image upload interface to obtain real-scene image data after on-site handling, and then send the real-scene image data and closed-loop status instructions back to the intelligent dispatch module.

[0040] In addition, the intelligent traffic signal box is equipped with a camera facing the door opening area. This camera is used to capture image information of the person opening the door when it is open. When the door is closed, due to the physical installation position, the camera's field of view is insufficient to cover the external environment of the box. To address this physical limitation, when the second AI agent performs automated allocation, if it determines that external status confirmation data is needed, it dispatches the preliminary verification task to a third-party crowdsourcing terminal. The third-party crowdsourcing terminal receives the task and then uses it to transmit the external status confirmation data back.

[0041] Example 2.

[0042] This embodiment provides a monitoring and early warning management method for intelligent traffic signal boxes, applied to the intelligent traffic signal box monitoring and early warning management system described in Embodiment 1. This embodiment uses a real-world attitude anomaly event as an example to fully illustrate the process of multi-source data acquisition at the edge, loading of device configuration data, diagnostic processing by the first artificial intelligence agent, and generation of structured diagnostic results. The specific operation flow is as follows: Figure 3 As shown.

[0043] In this embodiment, there is a traffic intelligent signal box in the system, whose unique device identifier in the cloud system is: device_id = "TSB-HZ-GS-000357" The basic registration information for this device in the system includes the installation location coordinates (longitude: 120.15507, latitude: 30.27408), the device model is "Intelligent Traffic Signal Box - Standard Type," and the types of sensors used include attitude sensors, temperature and humidity sensors, door magnetic sensors, water immersion sensors, electrical parameter detection modules, and fire smoke sensors. The initial installation attitude reference values ​​for this device were collected and stored during initial deployment. The initial attitude reference values ​​are as follows: initial_tilt_x = 0.8° initial_tilt_y = 1.2° initial_tilt_z = 0.5° Meanwhile, the key threshold parameters configured for this device in the cloud system include: a 15° attitude change trigger threshold, a normal internal temperature range of -20°C to 70°C, and a normal relative humidity range of 10% to 95%.

[0044] Phase 1: Multi-source data analysis and generation of the first AI agent diagnosis.

[0045] like Figure 4As shown, at a certain runtime, the edge acquisition module collected and reported the following multi-source sensor data within the same time window, and sent it to the cloud server via the MQTT protocol. The data is represented in structured form as follows: { "device_id": "TSB-HZ-GS-000357", "timestamp": "2026-03-11 14:32:18", "tilt": { "x": 14.6, "y": 2.9, "z": 1.1 }, "tilt_delta": { "x": 13.8, "y": 1.7, "z": 0.6 }, "door_status": "closed", Temperature: 35.2 "humidity": 55.4, "water_leak": false, "smoke_detected": false, "voltage": 221.3, "current": 1.87, "rssi": -88 } After receiving the above data, the cloud server loads the configuration data corresponding to the device based on the device_id, and merges the real-time collected data with the device configuration data to form a complete diagnostic input object for the first AI agent. The merged diagnostic input object is as follows: { "device_id": "TSB-HZ-GS-000357", "gps_coordinates": { "longitude": 120.15507, "latitude": 30.27408 }, "device_model": "Traffic Intelligent Signal Box - Standard Type", "initial_tilt": { "x": 0.8, "y": 1.2, "z": 0.5 }, "current_tilt": { "x": 14.6, "y": 2.9, "z": 1.1 }, "tilt_delta": { "x": 13.8, "y": 1.7, "z": 0.6 }, "door_status": "closed", Temperature: 35.2 "humidity": 55.4, "water_leak": false, "smoke_detected": false, "voltage": 221.3, "current": 1.87, "rssi": -88, "tilt_threshold": 15.0, "temperature_range": [-20.0, 70.0], "humidity_range": [10.0, 95.0] } The first AI agent processes the input object based on a preset diagnostic prompt template. The prompt template is a set of instruction text used to constrain the output format and judgment logic, an example of which is as follows: "You are a traffic facility operation and maintenance diagnostic expert. You will receive real-time sensor data and equipment configuration data from a smart traffic signal box. Please complete the following tasks: First, determine if there are any anomalies based on the attitude change, door status, and environmental parameters; second, calculate the urgency score of the anomaly, ranging from 0 to 100; third, provide a clear alarm level; fourth, generate a complete checklist of on-site maintenance steps, which must include the entire process from external verification to internal testing and repair, and internal repairs must be broken down into specific atomic operation steps. Each item in the checklist must clearly state whether the door needs to be opened and whether it can be performed by third-party crowdsourced personnel. Please strictly follow the specified JSON structure for output results and do not output any additional explanatory text." After completing the reasoning, the first AI agent outputs a structured fault diagnosis result file as shown below. This result file is directly stored on a cloud server and serves as the unified input source for subsequent human-machine collaborative review, decision-making by the second AI agent, and the work order system.

[0046] { "device_id": "TSB-HZ-GS-000357", "diagnosis_time": "2026-03-11 14:32:22", "abnormal_detected": true, "abnormal_type": "cabinet_tilt", "priority_score": 78, "alarm_level": "level_2", "diagnosis_summary": "A significant change in the attitude of the intelligent traffic signal box was detected, with an X-axis tilt angle change of 13.8 degrees, approaching the safety threshold of 15.0 degrees, indicating a potential risk to structural stability. Further verification of the external physical condition is required." "maintenance_steps": [ { "step_id": 1, "description": "Take a full-view photo from the front." "requires_open_door": false, "allow_crowdsourcing": true, "safety note": "The entire case door and top must be included during photography. Maintain a safe distance when taking pictures and do not touch the case."}, { "step_id": 2, "description": "Take a panoramic photo of the back view." "requires_open_door": false, "allow_crowdsourcing": true, "safety_note": "Used to check the status of the back attachments or fixed poles. Pay attention to road traffic safety behind you." }, { "step_id": 3, "description": "Take photos from both the left and right sides." "requires_open_door": false, "allow_crowdsourcing": true, "safety_note": "Used to check the condition of the side vents and the outer casing. Observe the surrounding area for any fallen metal debris or parts." }, { "step_id": 4, "description": "A close-up photo of the bottom." "requires_open_door": false, "allow_crowdsourcing": true, "safety_note": "Used to check for flooding, siltation, or a loose base. If water is found at the bottom, do not step into the puddle to prevent electric shock." }, { "step_id": 5, "description": "A close-up photo of the damaged area." "requires_open_door": false, "allow_crowdsourcing": true, "safety note": "Details of surface dents or impact damage must be clearly displayed. Do not touch the edges of damage with your hands to avoid cuts from metal burrs." }, { "step_id": 6, "description": "Set up a cordon at the scene." "requires_open_door": false, "allow_crowdsourcing": false, "safety note": "Set up safety warning cones around the enclosure to isolate the area and prevent pedestrians or vehicles from approaching. This must be carried out by official maintenance personnel wearing reflective vests." }, { "step_id": 7, "description": "Casing leakage current detection." "requires_open_door": false, "allow_crowdsourcing": false, "safety_note": "Use a non-contact voltage tester to check the metal casing of the enclosure and the door handle. Only proceed to the next step after confirming that the voltage tester does not sound an alarm." }, { "step_id": 8, "description": "Open the box door." "requires_open_door": true, "allow_crowdsourcing": false, "safety_note": "Use officially authorized commands to unlock. When opening the door, personnel should stand to the side of the door's opening arc to avoid loose items falling and causing injury." }, { "step_id": 9, "description": "Disconnect the main power supply." "requires_open_door": true, "allow_crowdsourcing": false, "safety note": "Immediately disconnect the main power circuit breaker inside the box and hang a 'Do Not Reconnect' safety sign. This must be performed by officially certified electrician personnel." }, { "step_id": 10, "description": "Inspect and tighten internal cables." "requires_open_door": true, "allow_crowdsourcing": false, "safety note": "Check each microcontroller unit and four-wire bus terminal one by one, and reconnect any loose wires. Take precautions against static electricity during operation to avoid damaging delicate electronic components." }, { "step_id": 11, "description": "Correcting damaged physical structural components." "requires_open_door": true, "allow_crowdsourcing": false, "safety note": "Inspect and correct or replace damaged mounting brackets, rails, and module clips. Take care to avoid touching exposed circuit boards with tools during operation." }, { "step_id": 12, "description": "Power on again and check status." "requires_open_door": true, "allow_crowdsourcing": false, "safety note": "Remove the safety sign and close the main power circuit breaker. Observe whether the indicator light is solid green. Personnel should maintain a safe distance when power is applied, and pay attention to any abnormal sounds or burning smells." }, { "step_id": 13, "description": "Perform sensor zero bias calibration." "requires_open_door": true, "allow_crowdsourcing": false, "safety_note": "Issue calibration commands via the official maintenance terminal to solidify the new initial zero bias value. During calibration, ensure the enclosure is absolutely stationary and do not touch it." } ] } It is important to note that the "allow_crowdsourcing" field plays a crucial technical control role in the aforementioned structured fault diagnosis result file. This field is automatically generated by the first AI agent based on safety specifications and operational attributes; in other words, it is pre-marked and clearly defines the execution permission boundaries for different maintenance steps. When this field is true, it indicates that the corresponding step only involves external non-contact observation or photography, posing no risk of electric shock or damage to the internal structure of the equipment, and can be safely separated and handed over to non-professional third-party crowdsourcing personnel. When this field is false, it indicates that the corresponding step involves on-site traffic control, contact voltage testing, unpacking and power disconnection, and internal component repair, with strict safety qualification requirements, and must be performed by official maintenance personnel. This permission boundary field provides clear data basis for the subsequent intelligent scheduling module to decompose and dynamically dispatch tasks.

[0047] Phase Two: Human-Machine Collaborative Review and Work Order Flow Triggering Phase.

[0048] After the first AI agent generates the structured fault diagnosis result file, the human-machine collaborative review module on the cloud server immediately parses the anomaly priority scoring field (78) and alarm level field (level 2 alarm) in the file. In this embodiment, the large-screen visualization display terminal is built using the Vue.js responsive front-end framework and integrates WebGL-based electronic map components (such as ECharts or Mapbox). After the large-screen front-end listens to the JSON data updates pushed from the cloud in real time through a WebSocket long connection, it stores them in the front-end state management tree (such as Vuex or Pinia), thereby triggering the automatic re-rendering of the view.

[0049] When rendering visually highlighted elements, the front-end program locates the corresponding spatial anchor point in the map layer based on the device identifier. To achieve visual prominence for secondary alerts, the front-end program dynamically modifies the attribute parameters of the anchor point component through a data binding mechanism: First, it dynamically increases the scaling parameter (e.g., symbolSize) of the anchor point from the default 1.0 to 2.0; second, it dynamically attaches a preset alert class name (e.g., class="alert-level-2") to the document object model (DOM) node of the anchor point. This class name is bound to a keyframe animation (@keyframes) in the Cascading Style Sheets (CSS). Specifically, the animation defines the background color as 100% pure orange (hexadecimal color value #FFA500) with zero transparency, reduces the transparency to 20% at 50%, and restores pure orange at 100%, with an animation period of 1.0 seconds and infinite loop. Through the combination of the above parameter-driven approach and CSS animation, the orange high-frequency pulse flashing effect of the device anchor point is accurately implemented on the large screen.

[0050] Meanwhile, the abnormal details interaction submodule on the large-screen visualization display pops up an audit panel with a digital timer on the front end of the interface. For level 2 alarms, the system's default audit countdown period configured on the backend is, for example, thirty minutes. After receiving this timestamp, the digital timer component starts counting down from thirty minutes. The panel uses Vue's data interpolation syntax to synchronously display a fault diagnosis summary, "Attitude change is approaching the threshold, indicating a potential structural stability risk," as well as trigger controls for receiving manual intervention instructions, such as "Confirm Dispatch," "Adjust Priority," or "Reject False Alarm" buttons.

[0051] During the countdown of the digital timer, the system remains in a suspended state awaiting manual intervention. If the monitoring center staff, after reviewing the attitude data and diagnostic summary transmitted from the field, deems the alarm accurate, they can issue a manual intervention command by clicking the "Confirm Dispatch" button. Upon receiving this command, the human-machine collaborative review module immediately changes the status of the fault diagnosis list to "Confirmed" and triggers the work order workflow.

[0052] However, in real-world scenarios, situations may arise where night shift workers are fatigued, staff are briefly absent during shift changes, or large-scale alarm outbreaks occur during extreme weather, overloading manual processing capacity. To prevent potential faults from being shelved due to lack of response, the system introduces an automatic timeout handling mechanism. If the human-machine collaborative review module has not received any manual intervention instructions from the front-end controls when the digital timer counts down to zero, the back-end program will trigger timeout interruption logic. At this point, the system no longer waits indefinitely for manual confirmation but automatically adopts the fault diagnosis list generated by the first AI agent, records an "automatic timeout confirmation" audit log in the system database, and forcibly triggers the work order workflow process.

[0053] Whether triggered manually or automatically by a countdown, once the work order workflow is activated, the fault diagnosis result file, which includes complete multi-source sensor data, equipment configuration data, and structured maintenance steps, will be transferred to the intelligent scheduling module. This will then formally enter the process of determining state uncertainty and making dynamic work order decisions, which is led by the second artificial intelligence agent.

[0054] Through the above process, this embodiment completes the entire process in the first and second stages, from the complete acquisition of multi-source data from the edge side, loading of device configuration data, diagnostic reasoning by the first artificial intelligence agent, to the output of a structured fault diagnosis result file that can be directly used by subsequent system modules. This result file is then used in the human-machine collaborative review process and serves as the basic input for the second artificial intelligence agent to determine state uncertainty and make dispatch decisions.

[0055] Phase 3: Determining Uncertainty of State and Making Dynamic Order Dispatch Decisions.

[0056] like Figure 5 As shown, once the work order workflow is activated, the fault diagnosis result file, containing complete multi-source sensor data, equipment configuration data, and structured maintenance steps, is transferred to the intelligent scheduling module. To avoid blindly dispatching high-cost official maintenance resources when information is incomplete, the intelligent scheduling module invokes a second artificial intelligence agent to determine the uncertainty of the work order execution status based on the fault diagnosis list, the hardware physical limitations of the cameras in the traffic intelligent signal box, and the current availability of the official maintenance terminal.

[0057] In this embodiment, the intelligent scheduling module first queries the hardware constraint attributes of the target device (TSB-HZ-GS-000357) through the system interface, and simultaneously calls the operation and maintenance resource management system to obtain the real-time availability status of the current official operation and maintenance terminal. Subsequently, the intelligent scheduling module merges the above information with the fault diagnosis result file generated in the first stage to form a complete scheduling input object for the second artificial intelligence agent. The merged scheduling input object is as follows: { "device_id": "TSB-HZ-GS-000357", "alarm_level": "level_2", "diagnosis_summary": "A significant change in the attitude of the intelligent traffic signal box was detected, with an X-axis tilt angle change of 13.8 degrees, approaching the safety threshold of 15.0 degrees, indicating a potential risk to structural stability. Further verification of the external physical condition is required." "maintenance_steps_summary": [ {"step_id": 1, "allow_crowdsourcing": true}, {"step_id": 2, "allow_crowdsourcing": true}, {"step_id": 3, "allow_crowdsourcing": true}, {"step_id": 4, "allow_crowdsourcing": true}, {"step_id": 5, "allow_crowdsourcing": true}, {"step_id": 6, "allow_crowdsourcing": false}, {"step_id": 7, "allow_crowdsourcing": false}, {"step_id": 8, "allow_crowdsourcing": false}, {"step_id": 9, "allow_crowdsourcing": false}, {"step_id": 10, "allow_crowdsourcing": false}, {"step_id": 11, "allow_crowdsourcing": false}, {"step_id": 12, "allow_crowdsourcing": false}, {"step_id": 13, "allow_crowdsourcing": false} ], "hardware_constraints": { "camera_installed": true, "camera_direction": "internal_door_facing", "current_door_status": "closed", "external_view_coverage": 0.0 }, "official_resource_status": { "available_terminals_nearby": 0, "nearest_terminal_distance_km": 12.5, "meets_immediate_dispatch_condition": false } } The second AI agent processes the input objects based on a preset scheduling decision prompt template. The prompt template is a set of instruction text used to constrain the output format and dispatch logic, as shown in the following example: "You are a traffic facility intelligent dispatching expert. You will receive a summary of a fault diagnosis list, equipment hardware constraint information, and official maintenance resource status. Please complete the following tasks: First, based on camera field of view coverage and door sensor status, determine whether the external physical status can be directly confirmed through the cloud; Second, considering the availability and immediate dispatch conditions of official maintenance resources, determine whether there is status uncertainty in the current fault scenario; Third, if there is status uncertainty and official resources are insufficient, extract the sub-items with the 'allow_crowdsourcing' field set to true from the maintenance steps and generate preliminary verification task instructions for third-party crowdsourcing terminals; Fourth, mark high-risk steps with the 'allow_crowdsourcing' field set to false as pending official processing. Please strictly follow the specified JSON structure to output the dispatching decision file." After completing the inference, the second AI agent outputs the following structured scheduling decision file. This result file clarifies the task routing path and serves as the standard instruction payload for subsequent automated agent frameworks to execute cross-application dispatching.

[0058] { "work_order_id": "WO-20260311-001", "device_id": "TSB-HZ-GS-000357", "decision_timestamp": "2026-03-11 14:47:25", "uncertainty_evaluation": { "external_state_confirmable": false, "official_resource_sufficient": false, "uncertainty_detected": true, "evaluation_reason": "The closed door limited the camera's field of view, making it impossible to directly confirm the external physical damage, and there are no official maintenance resources nearby that meet the conditions for immediate delivery." }, "dispatch_decision": { "action_type": "preliminary_verification", "target_channel": "third_party_crowdsourcing", "target_app_package": "com.sankuai.meituan", "task_title": "Emergency Verification of External Status of Intelligent Traffic Signal Box on Shangtang Elevated Road", "task_bounty": 15.0, "extracted_steps": [ { "step_id": 1, "description": "Take a full-view photo from the front." }, { "step_id": 2, "description": "Take a panoramic photo of the back." }, { "step_id": 3, "description": "Take photos from both the left and right sides." }, { "step_id": 4, "description": "A close-up photo of the bottom." }, { "step_id": 5, "description": "A close-up photo of the damaged area." } ], "safety_warning": "Maintain a safe distance when filming, do not touch the container, and pay attention to road traffic safety." }, "pending_official_steps": [6, 7, 8, 9, 10, 11, 12, 13] } It is worth noting that in the aforementioned structured scheduling decision file, the second AI agent accurately executed the uncertainty judgment logic (uncertainty_detected is true) by parsing hardware constraints (external_view_coverage is 0.0) and resource status (meets_immediate_dispatch_condition is false). Based on this judgment, the system successfully segmented the permission boundaries of the maintenance list generated in the first phase: only steps 1 to 5 were extracted to generate third-party crowdsourced verification tasks with a bounty of 15.0 yuan, and the target application package name was specified (e.g., com.sankuai.meituan); simultaneously, high-risk operations involving unpacking and power verification, such as steps 6 to 13, were safely placed in the pending array (pending_official_steps). This mechanism effectively breaks through the scheduling bottleneck caused by relying solely on official maintenance personnel to conduct on-site verification in the traditional model, achieving efficient hierarchical utilization of maintenance resources, and providing precise data support for the next phase of autonomously distributing crowdsourced tasks through automated terminals.

[0059] Phase 4: The phase of autonomous distribution and closed-loop acquisition of third-party crowdsourcing tasks.

[0060] After generating the structured scheduling decision file, the system faces a technical barrier: third-party crowdsourcing platforms (such as Meituan) typically do not provide open application programming interfaces (APIs) for external systems to directly dispatch orders. To achieve fully automated work order flow, this embodiment introduces an automated agent framework (such as AutoGLM) based on a graphical user interface (GUI) to complete the autonomous dispatch of crowdsourcing tasks.

[0061] In the hardware and network topology of this embodiment, the cloud server establishes an Android Debug Bridge (ADB) connection with a dedicated Android scheduling terminal (which can be a physical smartphone or a cloud-based Android emulator) via a local area network or virtual private network (VPN). This Android scheduling terminal is pre-installed with the third-party crowdsourcing application specified in the scheduling decision file (i.e., target_app_package: "com.sankuai.meituan") and maintains a logged-in status.

[0062] The intelligent scheduling module parses the scheduling decision file generated in the third stage, concatenating structured fields such as task location, extracted verification items (extracted_steps), bounty (task_bounty), and safety warnings (safety_warning) into automated operation prompts that conform to natural language usage. The prompt template input to the AutoGLM agent framework is as follows: "Please control your device to open the application with the package name 'com.sankuai.meituan', enter the errand or shopping service module, and publish a custom on-site verification task. The task pickup / occurrence location should be filled in as 'Northeast corner of the intersection of Shangtang Elevated Road and Wensan Road, Gongshu District, Hangzhou City'. In the task requirements remarks column, enter the following in full: 'Emergency verification of the external status of the Shangtang Elevated Road traffic intelligent signal box. Riders, please take 5 photos upon arrival: 1. Front panoramic view (including the box door and top); 2. Back panoramic view (confirm the back hanging); 3. Left and right sides (observe any fallen parts); 4. Bottom close-up (confirm the base is loose); 5. Close-up of the damaged area. Safety warning: Maintain a safe distance when taking photos, do not touch the box, and pay attention to road traffic safety!'. Set the task tip / reward to 15.0 yuan, then click 'Publish Order' and complete the default payment process." After receiving the aforementioned prompts, the AutoGLM proxy framework uses its built-in multimodal large model to understand the current screen's view hierarchy, automatically planning and generating a series of ADB control commands, including clicks, swipes, and text input, to drive the Android scheduling terminal to autonomously complete order filling, price increases, and publishing operations within the third-party crowdsourcing application. Since the specific UI node recognition and touch mapping logic of AutoGLM is a mature existing technology, it will not be elaborated upon here.

[0063] After a task order is successfully published, the intelligent scheduling module starts a status polling daemon in the background to achieve closed-loop management of crowdsourced tasks. This daemon wakes up the Android scheduling terminal screen via ADB commands at preset time intervals (e.g., every two minutes) and calls AutoGLM or an automated script to refresh the order details page of the third-party application.

[0064] The status polling daemon continuously reads the order status text. When it detects a change in order status to "Rider delivered" or "Task completed," the daemon triggers a closed-loop data acquisition command. At this point, the automated script automatically clicks to enter the text-and-image chat dialog box between the rider and the order placer or the order delivery voucher area, and downloads in batches the on-site verification photos taken and uploaded by the rider. Subsequently, the daemon packages these image files, including the front, back, sides, bottom, and damaged areas, and sends them back to the cloud server along with the corresponding device identifier (device_id). Thus, the system successfully acquires the actual physical status data of the traffic smart signal box without requiring official maintenance personnel or third-party platform API access, providing crucial visual evidence for the next stage of closed-loop confirmation and knowledge base iteration.

[0065] Phase 5: Task closure confirmation and knowledge base iteration phase.

[0066] After receiving the on-site verification photos from the third-party crowdsourced terminal, the cloud server invokes a second AI agent to perform secondary visual analysis of the images. At this point, the system possesses dynamic verification capabilities based on visual evidence and resource status. If the second AI agent determines that the photo only shows a slight tilt of the container, with no structural damage, no risk of electrical leakage or fire, and given the current severe shortage of official maintenance personnel, the second AI agent can directly cancel steps 6 to 13 in the third stage, directly archiving the alarm or transferring it to a long-term observation queue. This avoids the rigid scheduling logic that would otherwise require official maintenance personnel to be present after verification by third-party crowdsourced personnel.

[0067] However, in the specific context of this embodiment, although the photos returned by the crowdsourced system confirmed that there was no severe impact dent on the exterior and that the base was not severely loose, ruling out the risk of an emergency external physical collapse, the attitude fluctuation characteristics continuously reported by the end-side indicated that there might be loose internal cables, and the system determined that step 13 (performing sensor zero bias calibration) must be executed to correct the 13.8-degree tilt deviation. Therefore, the second AI agent did not completely cancel the subsequent steps, but instead downgraded the original Level 1 emergency response to a regular maintenance work order, and dispatched the formal maintenance work order, including steps 6 to 13, to its mobile terminal when the surrounding official maintenance personnel finished their preliminary tasks and resources were idle.

[0068] Upon arrival at the site, official maintenance personnel first perform dual geofence verification via the official maintenance terminal. Specifically, the personnel use the terminal's barcode scanning submodule to scan the two-dimensional matrix code attached to the outside of the intelligent traffic signal box, generating a device authentication request (device_id = "TSB-HZ-GS-000357"). Simultaneously, the terminal's satellite positioning submodule obtains the measured latitude and longitude coordinates on-site and compares them with the work order's reference coordinates. Only after both the device identity and spatial location are successfully matched can the system unlock subsequent operation permissions.

[0069] When step 8 (opening the box door) is executed, the door magnetic sensor inside the intelligent traffic signal box changes its state from "closed" to "open." This state change signal is captured by the microcontroller unit, immediately triggering the linked image capture logic. If the internal camera is not damaged in the fault, it will automatically capture an image containing the facial features of the person opening the door and the surrounding background the moment the door opens. This image is uploaded to the cloud server in real time. The system compares and binds it with the identity information of the official maintenance personnel handling the work order, and permanently archives it as an operation audit log, thus achieving strict security traceability for high-risk opening operations.

[0070] like Figure 6 As shown, the official maintenance terminal displays a checklist of on-site maintenance steps. To achieve accurate fault tracing, the interface employs a dual-identifier interaction design mechanism. To the right of each maintenance sub-item, two independent UI controls are placed side-by-side: the first control is a checkbox, serving as the "first identifier" to indicate the "check completion status" of the step (e.g., displayed as "[ ] executed"); the second control is a radio button or a special highlight switch, serving as the "second identifier" to indicate whether the step is an "actual fault resolution node" (e.g., displayed as "( ) confirmed as the fault source").

[0071] In practice, maintenance personnel must strictly follow the checklist to troubleshoot item by item. For example, after completing step 10 (checking and tightening internal cables), it is found that a minor collision earlier caused the four-wire bus terminal connecting the microcontroller to the attitude sensor to become loose, resulting in abnormal attitude data fluctuations. At this point, the maintenance personnel check the "Executed" checkbox for step 10 on the interface and simultaneously select the "Confirmed as Fault Source" radio button for step 10. The official maintenance terminal's logical verification mechanism requires that all maintenance sub-items' corresponding first identifiers must be checked and confirmed (ensuring compliant and complete operation), and at least one maintenance sub-item's corresponding second identifier must be triggered before clicking the "Submit and Close Loop" button.

[0072] After submission, the official operations and maintenance terminal generates a structured feedback list marked with dual identifiers and sends it back to the cloud server. Upon receiving the feedback list, the cloud server triggers the automatic iteration mechanism of the two-layer knowledge base: The first layer involves updating device-level maintenance records. The system appends the actual result of the current issue—"slight tilt combined with loose four-wire bus terminals causing abnormal posture"—along with the processing time and troubleshooting path, to the lifecycle file of the device (TSB-HZ-GS-000357). When this device reports multi-source sensor data and requests diagnosis again in the future, the first AI agent will automatically import the device's historical maintenance records within the context of the prompt. This injection of specific context allows the first AI agent to prioritize recurring "chronic problems" of the device (such as aging and loose wiring terminals), thereby generating more targeted diagnostic conclusions.

[0073] The second layer involves iterating the global equipment fault graph knowledge base. The system extracts the mapping relationship between the "initial characterization (X-axis tilt angle change of 13.8 degrees)" and the "actual fault node (step 10: loose internal cable)" of the current fault, transforming it into a new knowledge graph edge weight data. After aggregating similar data from multiple devices of the same model, the system updates the global fault graph knowledge base. When faced with similar alarms from all traffic intelligent signal boxes of the same model in the future, the first AI agent will use the updated knowledge base to automatically optimize the generation strategy of checklist-style on-site maintenance steps. For example, it may prioritize the "check internal cables" step or directly indicate a high probability of cable looseness in the diagnostic summary. Through this closed-loop iteration based on real engineering feedback, the system achieves continuous self-evolution of diagnostic logic accuracy.

[0074] Phase 6: Access Control Audit and Data Aggregation Analysis.

[0075] To ensure system security and data traceability, this embodiment strictly implements role-based access control and operation auditing steps throughout the entire lifecycle. In the second stage, the work order flow triggering phase, although the large-screen visualization displays manual intervention controls such as "Confirm Order Dispatch" and "Adjust Priority," the system's backend interface gateway is configured with a strict role-based access control (RBAC) interceptor. When a user attempts to click these controls to issue a manual intervention command, the system verifies the currently logged-in user's identity token. The command is only allowed and executed if the user has been granted the "Monitoring Center Reviewer" role; any unauthorized operation is directly blocked with an insufficient permission message, thus preventing unauthorized personnel from making mistakes.

[0076] Meanwhile, for the structured feedback list with dual identifiers submitted by the official maintenance terminal in the fifth phase, the system performed a detailed operation audit log. Upon receiving the feedback list, the cloud server automatically extracted the sending time of the structured feedback list (e.g., 2026-03-11 16:30:45), the operator's user identifier (e.g., User_ID_007), and the corresponding core investigation content (e.g., step 10 confirmed as the source of the fault). This triplet data was then encrypted using an anti-tampering hash algorithm and persistently recorded in the system's underlying operation audit log database for subsequent security reviews and accountability.

[0077] Furthermore, to achieve a closed loop from micro-level fault handling to macro-level operation and maintenance management, the system also performs historical fault data aggregation and trend analysis. The system backend is configured with scheduled tasks; within a preset time period (e.g., 23:59 at the end of each month), the data aggregation engine automatically extracts all closed-loop archived work order data for that period. The system performs multi-dimensional statistical calculations on fault types such as "abnormal posture" and "loose cable" involved in this embodiment, mean time to repair (MTTR), and the proportion of third-party crowdsourcing resources used.

[0078] After the calculation is complete, the system generates a structured statistical analysis report containing multi-dimensional charts (such as pie charts, bar charts, and line charts), and actively pushes it to the dashboard module of the large-screen visualization display via a WebSocket long connection. Figure 7 As shown, after receiving the report data, the large screen front-end dynamically updates the statistics panel using a chart rendering library (such as ECharts). For example, the device (TSB-HZ-GS-000357) in this embodiment is marked as a high-frequency warning node for the current month, and a heat map distribution is generated on the map layer. Through this global trend analysis, the system not only completes the closed loop of handling a single fault, but also provides management with intuitive and quantitative decision support for optimizing subsequent maintenance personnel scheduling, equipment procurement selection, and crowdsourcing budget allocation.

[0079] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0080] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A monitoring and early warning management system for intelligent traffic signal boxes, characterized in that, It includes an edge-side data acquisition module, an adaptive communication module, a human-machine collaborative review module, and an intelligent scheduling module; The end-side acquisition module is installed inside the intelligent traffic signal box and includes an attitude sensor and an environmental sensor. It is configured to synchronously acquire environmental parameters and attitude data at a first sampling frequency, and when it is determined that the change in the attitude data exceeds a preset frequency conversion trigger threshold, it triggers the frequency conversion sampling control logic to instruct the attitude sensor to acquire attitude data during the abnormal event at a second sampling frequency greater than the first sampling frequency. The adaptive communication module is connected to the end-side acquisition module and is configured to acquire the signal strength indication value of the current wireless network, and execute a downgrade transmission command when the signal strength indication value is lower than a preset communication threshold to block video stream data, encapsulate the attitude data during the abnormal event into a payload in variable-length byte array format, and send it to the cloud server through the Internet of Things communication protocol; The human-machine collaborative review module is deployed on the cloud server and is configured to receive the payload and call the first artificial intelligence agent to generate a fault diagnosis list with anomaly priority scores. Based on the anomaly priority scores, it renders visually prominent elements corresponding to the alarm level on the large screen visualization display terminal and simultaneously starts the review countdown cycle corresponding to the alarm level. The intelligent scheduling module is connected to the human-machine collaborative review module and is configured to convert the fault diagnosis list into a work order based on the manual intervention instruction when a manual intervention instruction is received within the review countdown period, and to trigger an automatic timeout processing mechanism when the review countdown period is reached and no manual intervention instruction is received, automatically adopt the fault diagnosis list and generate the work order, and call the second artificial intelligence agent to perform automated allocation of the work order.

2. The monitoring and early warning management system for intelligent traffic signal boxes according to claim 1, characterized in that, The end-side acquisition module also includes a microcontroller unit and a four-wire bus. The environmental sensor is connected to the microcontroller unit through the four-wire bus. The microcontroller unit is configured to acquire and store the initial zero bias data of the attitude sensor when the device is first powered on, and use the initial zero bias data as a reference for subsequent calculation of attitude change. When the adaptive communication module executes the degradation transmission command, the microcontroller sends a power cut-off signal through the four-wire bus to stop the power supply to the non-essential environmental sensors.

3. The monitoring and early warning management system for intelligent traffic signal boxes according to claim 1, characterized in that, The cloud server is also equipped with a large-screen visualization display terminal, and the system also includes an official operation and maintenance terminal and a third-party crowdsourcing terminal that establish a two-way communication connection with the cloud server. The large-screen visualization display terminal includes a map rendering submodule and an anomaly details interaction submodule. The map rendering submodule is configured to load a geographic information layer and generate corresponding spatial positioning anchor points based on the registration ledger of the intelligent traffic signal box. The anomaly details interaction submodule is connected to the human-machine collaborative review module and is configured to parse alarm icons of different levels, drive the corresponding spatial positioning anchor points to present preset color features and scaling ratios, and overlay a digital timer containing the review countdown period and a trigger control for receiving manual intervention commands. The official operation and maintenance terminal includes a QR code parsing submodule, a satellite positioning submodule, and a work order verification submodule; The third-party crowdsourcing terminal is configured to receive preliminary verification tasks and send back external status confirmation data.

4. The monitoring and early warning management system for intelligent traffic signal boxes according to claim 3, characterized in that, The intelligent traffic signal box is equipped with a camera facing the door opening area. The camera is used to collect image information of the person opening the door when the door is open. When the door is closed, the field of view of the camera is insufficient to cover the external state of the box.

5. A monitoring and early warning management method for a traffic intelligent signal box, characterized in that, Applied to the system according to claim 4, the method includes the following steps: The system receives multi-source sensor data uploaded by the end-side acquisition module of the intelligent traffic signal box, and calls the first artificial intelligence agent to perform feature analysis on the multi-source sensor data to generate a fault diagnosis list containing anomaly priority scores. Based on the anomaly priority score, the corresponding visual highlighting element for the alarm level is rendered on the large screen visualization display terminal, and the review countdown cycle corresponding to the alarm level is started simultaneously. If a manual intervention instruction is received within the review countdown period, a work order will be initiated based on the manual intervention instruction; if the review countdown period is reached and no manual intervention instruction is received, the fault diagnosis list will be automatically adopted and a work order will be initiated. After the work order is triggered, the second AI agent is invoked to determine the uncertainty of the work order execution status based on the fault diagnosis list, the field of view of the camera inside the traffic intelligent signal box when the box door is closed, and the current availability status of the official maintenance terminal. If the fault is determined to be verifiable in the external environment and the official maintenance terminal resources meet the conditions for immediate dispatch, a formal maintenance work order will be generated. If the external status of the fault can be confirmed but the official maintenance terminal resources are insufficient, or if the external status of the fault cannot be directly confirmed by data collected inside the box, a preliminary verification task is generated. The formal maintenance work order or the preliminary verification task is matched and dispatched to the corresponding terminal.

6. The method according to claim 5, characterized in that, The steps of the first artificial intelligence agent to perform feature analysis and generate a fault diagnosis list specifically include: based on the hardware error features in the multi-source sensor data, combined with a preset equipment model and fault map knowledge base, generating structured data containing checklist-style on-site maintenance steps.

7. The method according to claim 6, characterized in that, When generating the preliminary verification task, the method further includes: extracting verification sub-items from the checklist-style on-site maintenance steps that are pre-marked as being used only for external observation and status confirmation and can be executed without opening the door; encapsulating the verification sub-items into crowdsourced verification instructions and dispatching them to the third-party crowdsourcing terminal.

8. The method according to claim 6, characterized in that, The official maintenance terminal displays the checklist-style on-site maintenance steps on its interface. The checklist-style on-site maintenance steps include multiple maintenance sub-items. Each maintenance sub-item is configured with a first identifier to represent the completion status of the inspection and a second identifier to represent the actual fault resolution node. The official maintenance terminal obtains the triggering instructions from the on-site maintenance personnel for the first identifier and the second identifier. Under the condition that the first identifier corresponding to all maintenance sub-items is confirmed and the second identifier corresponding to at least one maintenance sub-item is triggered, the maintenance sub-item corresponding to the triggered second identifier is locked as the target node of the actual problem, a structured feedback list is generated and sent to the cloud server, so that the first artificial intelligence agent can update the basis for generating the subsequent fault diagnosis list.

9. The method according to claim 7, characterized in that, The method further includes a task closure confirmation step, which includes: receiving external status confirmation data returned by the third-party crowdsourcing terminal and handing it over to the second artificial intelligence agent for further analysis; when it is determined that official maintenance personnel need to intervene, extracting the remaining unpacking repair and component replacement sub-items from the list-style on-site maintenance steps to generate the formal maintenance work order, and dispatching the formal maintenance work order to the official maintenance terminal.

10. The method according to claim 8, characterized in that, The method also includes role-based access control and operation auditing steps, as well as historical fault data aggregation and trend analysis steps. In the step of triggering work order flow, only users with the monitoring center audit role are allowed to issue manual intervention commands, and the issuance time, user identifier and corresponding content of the structured feedback list are recorded in the system log. Within a preset time period, closed-loop archived work orders are aggregated, a statistical analysis report is generated and pushed to the large screen visualization display terminal.

Images