Underneath bridge waterlogging risk classification assessment and multi-channel linkage early warning system and method

Abstract

The application discloses a kind of underpass bridge waterlogging risk classification assessment and multi-channel linkage early warning system and method, belong to underpass bridge waterlogging technical field, including obtaining first water level value and second water level value;Difference of water level value is calculated, and first credibility mark or second credibility mark is generated according to difference;Current risk level is obtained by inputting water level value and credibility mark together into risk classification assessment model, and risk level is promoted when credibility mark is abnormal;According to risk level, trigger multi-channel linkage early warning including local early warning control, remote cooperative early warning and upstream blocking early warning.Local early warning equipment is directly connected with edge computing unit by hard line, does not depend on communication network, when communication interruption, edge computing unit uses built-in lightweight risk classification assessment model to complete risk assessment independently in local and directly drive local early warning equipment;The application is helpful to improve monitoring reliability, maintain early warning function under extreme conditions, and expand early warning coverage.

Description

Technical Field

[0001] This invention relates to the field of underpass flooding technology, and in particular to a risk classification assessment and multi-channel linkage early warning system and method for underpass flooding. Background Technology

[0002] Underpasses and culverts are critical nodes in urban traffic. Due to their low-lying location, they are prone to flooding during short-term heavy rainfall, seriously threatening the safety of vehicles and pedestrians. Currently, the monitoring and early warning systems for flooding in underpasses face the following main technical challenges: Current underpass flood monitoring typically relies on a single type of sensor, such as a radar level gauge or a submersible pressure level gauge. In real-world operating environments, sensors can be affected by various factors: radar level gauges may experience measurement errors due to floating debris, waves, or deformation of the mounting bracket; submersible level gauges may suffer from data distortion due to silt blockage or foreign object accumulation. Existing systems generally lack real-time diagnostic capabilities for the sensor's own operational status. When a sensor experiences a hidden fault, the system may interpret erroneous data as the true water level for early warning purposes, increasing the risk of missed or false alarms.

[0003] Existing flood warning systems primarily rely on a backend central platform for risk assessment and coordinated decision-making. Frontend monitoring equipment mainly handles data collection and uploading, lacking local decision-making capabilities. When heavy rain is accompanied by power outages, fiber optic cable interruptions, or public network base station congestion, the communication link between the frontend equipment and the backend platform is broken, rendering the system ineffective in its early warning function. Even if the water level at the underpass has risen rapidly, critical protective actions such as information board warnings and gate closures may fail to execute automatically, creating a safety vacuum.

[0004] Existing systems typically issue alerts through a single channel (such as sending SMS messages to on-duty personnel or simply illuminating on-site warning lights) after triggering an alert, failing to achieve differentiated alert coverage for different entities, such as on-site passing vehicles, remote management personnel, and vehicles approaching from upstream. The alert information also fails to fully integrate with on-site control equipment, remote collaboration platforms, and traffic guidance systems, resulting in insufficient timeliness and coverage. Summary of the Invention

[0005] The purpose of this invention is to provide a risk classification assessment and multi-channel linkage early warning system and method for water accumulation in underpasses, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a system and method for risk classification assessment and multi-channel linkage early warning of water accumulation in underpasses, comprising the following steps: S1. Obtain the first water level value collected by the radar water level gauge and the second water level value obtained by visual analysis of the water accumulation image of the underpass. S2. Calculate the difference between the first water level value and the second water level value. If the difference is not greater than the preset threshold, the first water level value and the second water level value are used as the output water level value according to the preset fusion rule, and a first confidence mark representing normal monitoring data is generated. If the difference is greater than the preset threshold, the first water level value and the second water level value are selected as the output water level value according to the preset verification rule, and a second confidence mark representing abnormal monitoring data is generated. S3. Input the output water level value and the generated credibility tag into the risk grading assessment model to obtain the current risk level; wherein, the risk grading assessment model is configured such that: when the input credibility tag is the second credibility tag, the risk level is adjusted to at least one level higher than the basic risk level corresponding to the output water level value. S4. Trigger a multi-channel linkage early warning mechanism based on the current risk level. The multi-channel linkage early warning mechanism includes: local early warning control directly driven by the edge computing unit deployed at the front end through hard-wired connection; remote collaborative early warning that pushes structured early warning information to the remote management platform; and upstream blocking early warning that sends detour suggestions to upstream traffic guidance equipment. Among these, local early warning control has the highest execution priority and does not depend on the communication network. When communication between the edge computing unit and the remote management platform is interrupted, the edge computing unit uses a locally pre-deployed lightweight risk classification assessment model to independently execute S2, S3 and local early warning control, and caches all operation process data in local storage. Once communication is restored, the data is synchronized to the remote management platform.

[0007] A risk assessment and multi-channel linkage early warning system for water accumulation in underpasses includes: The front-end AI intelligent monitoring layer includes a radar water level gauge, a camera, and an edge computing unit; the edge computing unit is connected to the radar water level gauge and the camera to acquire the first water level value and the second water level value. The data transmission layer establishes a two-way communication link with the front-end AI intelligent monitoring layer and the back-end platform layer; The backend platform layer is configured with a risk classification and assessment model, which is used to receive data uploaded from the frontend and issue control instructions. The early warning release layer includes local early warning equipment, remote early warning modules, and upstream guidance modules; Among them, the local early warning equipment is directly connected to the edge computing unit via hard wiring, and the edge computing unit has a built-in backup power supply and a lightweight risk classification assessment model; The edge computing unit is used to respond to risk level instructions issued by the backend platform layer to drive the local early warning device when the communication link is normal, or to independently complete the risk level assessment locally using a lightweight risk grading assessment model when a communication link interruption is detected, and directly drive the local early warning device based on the assessment results.

[0008] Therefore, the present invention employs the above-mentioned risk classification assessment and multi-channel linkage early warning system and method for underpass water accumulation, which has the following beneficial effects: (1) This invention forms a dual-link independent water level data source by simultaneously acquiring a first water level value collected by a radar water level gauge and a second water level value obtained based on camera visual analysis. When the consistency between the two data sources is good, a dynamic weighted fusion method based on the historical stability of each source is used to calculate the output water level value, and the fusion accuracy is better than that of a fixed weighted average. When the deviation between the two data sources exceeds a preset threshold, a relatively reliable data source is selected according to preset verification rules (floating object detection, jump detection, and coefficient of variation comparison), and a second credibility marker is generated, thereby triggering a forced upgrade of the risk level. This mechanism incorporates data credibility into risk assessment, overcomes the shortcomings of existing systems that simply alarm or require manual confirmation when data conflicts occur, and effectively reduces the probability of missed or false judgments caused by hidden faults of a single sensor.

[0009] (2) This invention introduces real-time calculation of the water level rise rate into the risk grading assessment model. When the water level rises rapidly beyond a preset rate threshold, the system automatically raises the basic risk level by one level; if an abnormal data credibility (i.e., a second credibility marker) is detected at the same time, the system raises the level by two levels (e.g., directly from the safe level to the warning level, or directly from the attention level to the danger level). This mechanism enables the system to respond in advance to the instantaneous confluence process of rainstorms, significantly improving the advance warning time compared to existing systems that rely solely on absolute water levels, thus providing more time for downstream vehicles to avoid danger.

[0010] (3) This invention incorporates a backup power supply and a lightweight risk grading assessment model within the front-end edge computing unit. When the communication link is interrupted, the edge computing unit can independently complete tasks such as water level data acquisition, dual-link verification, water level rise rate calculation, credibility marker generation, risk level superposition assessment, and local early warning device drive control locally, without relying on the back-end platform. The local early warning device is directly connected via hardwired connections, without going through any network protocol stack. This feature enables the system to still perform critical protective actions such as information board warnings, audible and visual alarms, and gate closure under extreme conditions of power outages and network interruptions, eliminating the "safety vacuum" that occurs when existing systems experience communication interruptions.

[0011] (4) The early warning release layer of this invention integrates local early warning equipment, remote early warning modules, and upstream guidance modules to form a layered early warning coverage for on-site vehicles, remote management departments, and upstream oncoming vehicles. Local early warning control has the highest priority and does not rely on the network; remote collaborative early warning pushes structured information to municipal platforms, traffic platforms, and responsible persons; upstream obstruction early warning pushes detour suggestions to upstream intersection information boards and navigation map service providers. The three together constitute a multi-channel linkage system of "on-site control - remote collaboration - upstream guidance", which significantly improves the early warning coverage and emergency response efficiency.

[0012] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0013] Figure 1 This is a flowchart of a method for risk classification assessment and multi-channel linkage early warning of water accumulation in underpasses according to the present invention; Figure 2 This is a schematic diagram illustrating the basic risk level classification and superposition rules of the method for risk classification assessment and multi-channel linkage early warning of water accumulation in underpasses according to the present invention; Figure 3 This is a diagram illustrating the architecture of a multi-channel linkage early warning system for risk classification assessment of water accumulation in underpasses, as described in this invention. Figure 4 This is a deployment diagram of a multi-channel linkage early warning system for risk classification assessment of water accumulation in underpasses, as described in this invention. Detailed Implementation

[0014] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely illustrates selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0015] Example like Figure 1 As shown, this invention provides a method for risk classification assessment and multi-channel linkage early warning of water accumulation in underpasses. This method is collaboratively completed by an edge computing unit deployed at the underpass site and a backend platform, and independently completed by the edge computing unit when communication is interrupted. The method includes the following steps: S1. Obtain the first water level value collected by the radar water level gauge and the second water level value obtained by visual analysis of the water accumulation image of the underpass.

[0016] The radar water level gauge is fixedly installed on the bridge wall or pillars near the lowest point of the underpass. It emits frequency-modulated continuous waves into the water surface in a non-contact manner, receives the reflected waves, calculates the frequency difference to obtain the distance to the water surface, and then converts it into the first water level value. A high-definition camera is installed on the side of the underpass, with a viewing angle covering the entire waterlogged area and pre-set calibration references on the bridge wall. The calibration references are scales marked at fixed intervals (e.g., every 10 centimeters), used for proportional conversion during subsequent visual water level calculations.

[0017] The second water level value is obtained by visually analyzing the water accumulation image of the underpass. Specifically, this involves: capturing water accumulation images using cameras deployed on the underpass; inputting these images into a pre-trained image semantic segmentation network; classifying each pixel in the water accumulation image and extracting a mask for the water accumulation region; and determining the pixel distance between the upper edge of the water accumulation region mask and the reference object, given that the water accumulation image contains a calibration reference of known physical size. The second water level value is calculated based on the ratio of the actual physical distance of the reference object to the number of pixels it occupies in the water accumulation image. This visual analysis process is performed locally by the edge computing unit, eliminating the need to upload the water accumulation image to the backend platform layer.

[0018] S2. Calculate the difference between the first water level value and the second water level value. If the difference is not greater than the preset threshold, the average of the first water level value and the second water level value is used as the output water level value, and a first confidence mark representing normal monitoring data is generated. If the difference is greater than the preset threshold, the first water level value and the second water level value are selected as the output water level value according to the preset verification rules, and a second confidence mark representing abnormal monitoring data is generated.

[0019] Under normal communication conditions, the edge computing unit uploads the first and second water level values ​​to the backend platform layer, which then executes steps S2 and S3. Under communication interruption conditions, steps S2 and S3 are executed locally by the edge computing unit. The following explanation uses the normal communication condition as an example.

[0020] The preset threshold is 2 cm. The absolute value of the difference between the first and second water level values ​​is calculated. When the absolute value is not greater than 2 cm, it indicates that the radar measurement value and the visual measurement value are highly consistent, and the current measurement data is reliable. This embodiment uses the minimum variance adaptive weighted fusion method to calculate the output water level value. Let the standard deviation of the radar water level gauge measurements within the past 30-second historical time window be... The standard deviation of the water level values ​​obtained from visual analysis is The weight w of the radar level gauge is calculated using the following formula: ; The formula for calculating the output water level is as follows: ; in, To output the water level value, This is the first water level value. This is the second water level value. As a reliability weight for radar level gauges, when historical data is insufficient, The default value of 0.5 is used, which gives higher weight to the data with a smaller standard deviation (i.e., more stable).

[0021] Variance estimation of output water level values for This is used to generate uncertainty markers for the output water level values.

[0022] Take data during system startup or when historical data is less than 30 seconds old. This is called equal-weighted averaging (which degenerates into an arithmetic mean, but this only occurs at startup). Compared to fixed-weight arithmetic mean, adaptive weighted fusion can dynamically adjust according to the real-time performance of the sensor. When the radar level gauge fluctuates greatly due to wave interference, its weight is automatically reduced, and the contribution of the visual level is increased, thereby obtaining a more stable and accurate output level value.

[0023] The variance estimate This reflects the uncertainty of the fusion result and can be used to generate an "uncertainty flag". When the uncertainty flag exceeds a preset threshold (e.g., ...), ... If the variance is greater than twice the historical average, the risk grading assessment model can treat it as an anomaly similar to the second confidence level label, raising the basic risk level by one level to avoid missed detections due to potential systemic biases in the stable fusion results. Simultaneously, a first confidence level label is generated.

[0024] When the absolute value is greater than 2 cm, it indicates that at least one water level data point may be abnormal. In this case, the preset verification rules are activated, and the preset verification rules are executed in the following order: (1) Review the video frames within the previous 10 seconds to detect whether there are floating objects obstructing the water accumulation area or human interference. The detection method is as follows: compare the shape of the water accumulation area mask extracted in the current frame with the mask in the normal state. If the mask area or shape changes abruptly and lasts for more than 2 seconds, it is determined that there are floating objects obstructing the water accumulation area. At the same time, use the trained lightweight image classification model to classify the water accumulation area image blocks to determine whether there are human-placed obstructions. If floating objects obstructing the water accumulation area or human interference is detected, the visual water level data is unreliable, and the first water level value is selected as the output water level value.

[0025] (2) If no floating objects or human interference are detected, the time series of the first water level value is checked to see if there is a jump exceeding the preset rate of change. The preset rate of change is 5 cm / s. If the difference between the current first water level value and the first water level value of the previous second exceeds 5 cm, it is determined that the radar water level gauge has a jump fault, and the second water level value is selected as the output water level value.

[0026] (3) If the first water level value does not change by more than the preset rate of change, compare the coefficients of variation of the first and second water levels within the past 30-second historical time window. The coefficient of variation is the ratio of the standard deviation to the mean. Select the water level value with the smaller coefficient of variation as the output water level value. A smaller coefficient of variation means that the data is more stable in the recent period and has a relatively higher reliability.

[0027] (4) If the duration of historical data is less than 30 seconds, or if the coefficients of variation of the first water level value and the second water level value both exceed the preset upper limit, a reliable judgment cannot be made by comparing the coefficients of variation. In this case, the first water level value is selected as the output water level value by default. At the same time, the edge computing unit triggers a local audio-visual prompt and sends an equipment abnormality notification to the remote management platform to prompt the maintenance personnel to conduct on-site inspection.

[0028] When the first water level value and the second water level value are selected as the output water level value according to the above verification rules, a second confidence mark is generated to indicate that there is an anomaly in the current monitoring data and the confidence is reduced.

[0029] S3. Input the output water level value and the generated credibility tag into the risk grading assessment model to obtain the current risk level; wherein, the risk grading assessment model is configured such that when the input credibility tag is the second credibility tag, the risk level is adjusted to at least one level higher than the basic risk level corresponding to the output water level value.

[0030] Basic risk levels are classified according to the following rules: A water level of less than 5 cm is considered safe. A water level output of 5 cm or more but less than 15 cm is classified as a warning level. A water level output of 15 cm or more but less than 30 cm is considered a warning level. A water level of 30 cm or more is considered dangerous. The above thresholds take into account the design standards for urban underpasses and the regulations for safe passage of vehicles through water: water depth of less than 5 cm has no substantial impact on the passage of the vast majority of vehicles; when the water depth reaches 15 cm, water may enter the exhaust pipes of small vehicles, and pedestrians will have difficulty passing; when the water depth reaches 30 cm, most small vehicles face the risk of water entering their air intakes, the possibility of vehicles floating increases significantly, and safe passage is no longer possible.

[0031] When the input is the first confidence level marker, the risk grading assessment model directly outputs the basic risk level as the current risk level. When the input is the second confidence level marker, the risk grading assessment model raises the risk level by one level based on the basic risk level. Specifically: the safe level is raised to the attention level, the attention level is raised to the alert level, the alert level is raised to the danger level, and the danger level remains unchanged. The reason for this setting is that the second confidence level marker indicates that the data is abnormal. Although a relatively reliable water level value has been selected through verification rules, this value still has uncertainty. Adopting a more conservative safety control strategy can leave a safety margin for emergency response. Figure 2 As shown, this embodiment further defines the specific threshold for classifying basic risk levels, the calculation formula for water level change rate, and the rules for superimposing risk levels. The basic risk levels are divided into four categories based on the output water level value: safe level (<5cm), caution level (5~15cm), warning level (15~30cm), and danger level (≥30cm).

[0032] This embodiment introduces dynamic correction for the water level rise rate into the risk grading assessment model. Edge computing units use fixed sampling time intervals. (In this embodiment, 2 seconds are used) Record and output the water level value. Assume the current time... The output water level value is Previous sampling time The output water level value is Then the rate of change of water level Calculate using the following formula: ; The unit is centimeters per second. Preset rise rate threshold. The value is 3 cm / s, calculated based on the typical catchment area of ​​urban underpasses and the return period of rainstorms. At that time, the system determined that the water level was rising rapidly, and even if the current absolute value of the water level had not yet reached the danger level, there was still a potential risk. Therefore, the risk classification assessment model temporarily raised the basic risk level by one level and generated a "rapidly rising" label.

[0033] Specifically, when the input confidence level is set to the second confidence level and is simultaneously detected... At this time, the risk level is upgraded by two levels, specifically: the basic safety level is upgraded to the alert level, the basic attention level is upgraded to the danger level, and the basic alert level and danger level remain unchanged at the danger level. This superposition mechanism ensures that in the most unfavorable operating condition where sensor data is uncertain and the water level rises sharply, the system can directly take the most stringent control measures (such as closing the gate) across multiple levels. Water level change rate It is also cached in local storage and uploaded to the backend platform after communication is restored for subsequent catchment model calibration and threshold adaptive optimization.

[0034] Table 1 shows the final risk level results under three scenarios: "only credibility anomaly", "only rapid increase", and "both occurring simultaneously". As can be seen from the table, when credibility anomaly and rapid increase occur simultaneously, the risk level is increased by two levels from the original base level (safe level → alert level, caution level → danger level). This superposition mechanism ensures a conservative safety strategy under the most unfavorable operating conditions.

[0035] Table 1 Risk Level Determination

[0036] S4. Trigger a multi-channel linkage early warning mechanism based on the current risk level. The multi-channel linkage early warning mechanism includes: local early warning control directly driven by the edge computing unit deployed at the front end through hard-wired connection, remote collaborative early warning pushing structured early warning information to the remote management platform, and upstream blocking early warning sending detour suggestions to upstream traffic guidance equipment. Among them, local early warning control has the highest execution priority and does not depend on the communication network.

[0037] Local early warning control includes: when the backend platform layer is communicating normally, or when the edge computing unit is interrupted, issuing control commands based on the current risk level.

[0038] The edge computing unit directly drives the local early warning device to perform the following actions via a hardwired output interface: When the risk level is at the attention level, the variable message sign installed at the entrance of the underpass is controlled to display a deceleration warning through the hard-wired output interface of the edge computing unit. When the risk level is alert, the variable message sign is controlled to display "Caution: Proceed with Caution" via the hard-wired output interface, and the audible and visual alarm installed at the underpass site is activated. When the risk level is dangerous, the variable message sign is controlled via the hard-wired output interface to display "No Entry", activate the audible and visual alarm, and control the gate installed at the entrance of the underpass to close.

[0039] Because the local early warning device is directly connected to the edge computing unit via hardwired connections, the generation and execution of control signals do not pass through the data transmission layer and the back-end platform layer. Therefore, the local early warning control can continue to operate regardless of the communication status.

[0040] Remote collaborative early warning includes: when communication is normal, the backend platform sends early warning work orders to the municipal management platform and traffic management platform. The early warning work order includes the current risk level, output water level value, first or second confidence mark, and real-time screenshot of the site. At the same time, the backend platform layer sends SMS messages or initiates automatic voice calls to the preset responsible person's mobile phone number through the SMS notification module and automatic voice call module to notify of the water accumulation early warning information.

[0041] Upstream traffic disruption warnings include: when the risk level reaches the alert or danger level, the backend platform automatically identifies key upstream intersections requiring traffic diversion based on the underpass's geographical location and surrounding road network topology, and pushes detour information to variable message signs at these upstream intersections: at the alert level, it displays "Underpass ahead is flooded, detour recommended"; at the danger level, it displays "Underpass ahead is closed, please detour". Simultaneously, the backend platform pushes road closure status and suggested detour routes to navigation map service providers via navigation map service interfaces, guiding passing vehicles to detour in advance.

[0042] When communication between the edge computing unit and the remote management platform is interrupted, the edge computing unit utilizes a locally pre-deployed lightweight risk grading assessment model to perform the following operations locally: calculate the difference between the first and second water level values ​​and generate an output water level value and a first or second confidence mark according to preset verification rules; input the output water level value and confidence mark into the lightweight risk assessment model to obtain the current risk level; and execute local early warning control based on the current risk level. Simultaneously, the edge computing unit calculates the output water level value using the same adaptive weighted fusion method as when communication is normal (if the dual-source data are consistent), and calculates the water level change rate in real time, using the water level change rate as a risk level overlay correction factor. All operation process data is cached in local storage, including raw water level data, verification logs, risk level calculation results, and early warning device driver records. When the edge computing unit detects that the communication link has been restored, it automatically synchronizes the cached data in batches to the remote management platform for archiving and analysis by the data traceability module. During the communication interruption, remote collaborative early warning and upstream blockage early warning are temporarily suspended due to communication failure, but local early warning control remains effective, ensuring uninterrupted critical safety protection functions on site.

[0043] like Figure 3 As shown, a risk classification assessment and multi-channel linkage early warning system for water accumulation in underpasses includes a front-end AI intelligent monitoring layer, a data transmission layer, a back-end platform layer, and an early warning release layer.

[0044] like Figure 4 As shown, radar water level gauges and cameras are installed on both sides of the underpass culvert. The edge computing unit is deployed in a waterproof box on site and is directly connected to the variable message signs, audible and visual alarms, and gates via hardwired connections (relay outputs or RS485 bus). Simultaneously, the edge computing unit connects to the backend platform via a communication network, and the backend platform then interacts with the upstream variable message signs, navigation map service providers, and remote management platforms. Figure 3 It intuitively demonstrates the distinction between hard-wired connections and network communication, as well as the logical relationship of edge computing units independently driving local early warning devices via hard-wired connections when communication is interrupted.

[0045] The front-end AI intelligent monitoring layer includes a radar level gauge, a camera, and an edge computing unit. The radar level gauge, preferably an 80GHz frequency-modulated continuous wave radar level gauge with a measurement accuracy of ±2 mm, is used to acquire the first water level value. The camera, preferably a network camera with at least 2 megapixels, is used to capture images of water accumulation under the underpass. The edge computing unit is an industrial-grade embedded computer with a processor and a graphics processor. The edge computing unit connects to the radar level gauge via an RS485 bus and to the camera via an Ethernet interface. The edge computing unit incorporates a visual analysis algorithm to calculate the second water level value from the water accumulation images; the specific calculation method has been described in the method embodiments.

[0046] The data transmission layer establishes a bidirectional communication link between the front-end AI intelligent monitoring layer and the back-end platform layer. It consists of an optical fiber communication module and a 4G / 5G wireless communication module, with optical fiber as the primary link and 4G / 5G as the backup link. The edge computing unit monitors the link status in real time, determining a communication interruption when both the primary and backup links become unreachable.

[0047] The backend platform layer is deployed in the municipal management center or cloud server and is configured with a risk grading assessment model. When communication is normal, it receives the first and second water level values ​​uploaded from the front end, performs the dual-link verification and risk level calculation described in steps S2 and S3, and sends control commands to the edge computing unit. The backend platform layer is also configured with a data traceability module to store the full-process parameters and early warning action records for each risk assessment.

[0048] The early warning distribution layer includes local early warning equipment, a remote early warning module, and an upstream guidance module. The local early warning equipment is directly connected to the edge computing unit via hardwired connections. Specifically, it includes: variable message signs installed at the underpass entrance to display warning text to approaching vehicles; audible and visual alarms installed at the underpass site to alert surrounding vehicles and pedestrians with sound and flashing lights; and a barrier gate installed at the underpass entrance to physically close the lane in dangerous situations. The variable message signs, audible and visual alarms, and barrier gates are all directly connected to the edge computing unit via relay output modules or an RS485 bus, ensuring that the generation and execution of control signals are completely independent of the communication status of the data transmission layer.

[0049] The remote early warning module connects to the backend platform layer, including data interfaces with the municipal management platform and the traffic management platform, as well as an SMS notification module and a voice call module. The upstream guidance module also connects to the backend platform layer, including data interfaces with upstream variable message signs at intersections and with navigation map service providers.

[0050] The edge computing unit incorporates a backup power supply and a lightweight risk grading assessment model. The backup power supply automatically switches to power when the main power supply fails, ensuring the edge computing unit can still operate even during power outages. The lightweight risk grading assessment model is derived from the risk grading assessment model in the backend platform layer through model pruning and parameter quantization, and is pre-stored in the edge computing unit's local memory. Model pruning removes network connections and neurons that contribute little to inference accuracy from the risk grading assessment model in the backend platform layer, and parameter quantization compresses model parameters from 32-bit floating-point numbers to 8-bit integers, significantly reducing the model's computational resource requirements and storage space. This allows the lightweight risk grading assessment model to run smoothly on the limited hardware resources of the edge computing unit.

[0051] During system operation, when the communication link is normal, the edge computing unit responds to the risk level instructions issued by the backend platform layer and drives the local early warning device to perform corresponding actions through the hard-wired output interface. When the edge computing unit detects an interruption in the communication link with the backend platform layer, it automatically loads the locally stored lightweight risk classification assessment model, independently completes the calculation of the difference between the first and second water level values, the generation of the first or second confidence mark, the calculation of the water level change rate, and the risk level assessment (including superposition correction) locally, and directly drives the local early warning device based on the assessment results to ensure that the core early warning function on site is not interrupted in the event of a communication interruption.

[0052] In the above embodiments, the preset verification difference threshold is 2 cm, the basic risk level water level thresholds are 5 cm, 15 cm, and 30 cm, the water level change rate sampling interval is 2 seconds, and the rise rate threshold is 3 cm / s. These values ​​can be adjusted according to the specific design parameters of the underpass, drainage capacity, and local rainfall characteristics. The image semantic segmentation network used for visual water level calculation can be replaced by traditional computer vision methods based on edge detection and horizontal line recognition. In addition to obtaining the lightweight risk grading assessment model through model pruning and parameter quantization, it can also be directly trained from scratch using a more streamlined lightweight neural network, or a rule-equivalent lightweight model can be constructed using traditional machine learning methods such as decision trees and logistic regression.

[0053] Therefore, this invention adopts the above-mentioned risk classification assessment and multi-channel linkage early warning system and method for underpass flooding. It achieves a conservative improvement in risk level through dual-source water level acquisition and credibility marking mechanism, realizes network outage autonomy and hard-wired direct-drive early warning through edge computing unit built-in backup power and lightweight model, and expands the early warning coverage through local, remote and upstream three-level linkage, providing a complete monitoring-assessment-early warning technical solution for emergency response to underpass flooding.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for risk classification assessment and multi-channel linkage early warning of water accumulation in underpasses, characterized in that, Includes the following steps: S1. Obtain the first water level value collected by the radar water level gauge and the second water level value obtained by visual analysis of the water accumulation image of the underpass. S2. Calculate the difference between the first water level value and the second water level value. If the difference is not greater than the preset threshold, the first water level value and the second water level value are used as the output water level value according to the preset fusion rule, and a first confidence mark representing the normality of the monitoring data is generated. If the difference is greater than the preset threshold, the first water level value and the second water level value are selected as the output water level value according to the preset verification rules, and a second confidence mark representing the abnormality of the monitoring data is generated. S3. Input the output water level value and the generated credibility tag into the risk grading assessment model to obtain the current risk level; wherein, the risk grading assessment model is configured such that: when the input credibility tag is the second credibility tag, the risk level is adjusted to at least one level higher than the basic risk level corresponding to the output water level value. S4. Trigger a multi-channel linkage early warning mechanism based on the current risk level. The multi-channel linkage early warning mechanism includes: local early warning control directly driven by the edge computing unit deployed at the front end through hard-wired connection; remote collaborative early warning that pushes structured early warning information to the remote management platform; and upstream blocking early warning that sends detour suggestions to upstream traffic guidance equipment. Among these, local early warning control has the highest execution priority and does not depend on the communication network. When communication between the edge computing unit and the remote management platform is interrupted, the edge computing unit uses a locally pre-deployed lightweight risk classification assessment model to independently execute S2, S3 and local early warning control, and caches all operation process data in local storage. Once communication is restored, the data is synchronized to the remote management platform.

2. The method according to claim 1, characterized in that, In S1, the second water level value is obtained through visual analysis of the water accumulation image of the underpass, specifically including: Images of accumulated water are captured using cameras deployed on the underpass. A semantic image segmentation network is used to extract a mask of the accumulated water area. Based on the pixel ratio between the mask of the accumulated water area and a preset calibration reference in the image, the second water level value is calculated.

3. The method according to claim 1, characterized in that, The preset fusion rule in S2 is an adaptive weighted fusion method. When the difference is not greater than the preset threshold, the output water level value is calculated using the following adaptive weighted formula: ； in, To output the water level value, This is the first water level value. This is the second water level value. As the reliability weight of the radar level gauge, , and These are the standard deviations of radar level gauges and visual analysis within their respective historical time windows; When the historical data is insufficient, Taking the default value 0.5, the variance estimate of the output water level value For , for generating the uncertainty label of the output water level value.

4. The method for risk classification assessment and multi-channel linkage early warning of water accumulation in underpasses according to claim 1, characterized in that, The preset verification rules in S2 include: The system traces back video frames within a preset time period to detect whether there are floating objects obstructing the water accumulation area or human interference. If there are floating objects obstructing the water or human interference, the first water level value is selected as the output water level value. If there are no floating objects obstructing the view or human interference, the system detects whether the time series of the first water level value changes by more than a preset rate of change; if such a change occurs, the second water level value is selected as the output water level value. If the first water level value does not change by more than the preset rate of change, then compare the coefficients of variation of the first water level value and the second water level value within their respective historical time windows, and select the water level value with the smaller coefficient of variation as the output water level value. If historical data is insufficient, or if the coefficients of variation of both the first and second water level values ​​exceed the preset upper limit, the first water level value will be selected as the output water level value by default. The edge computing unit will trigger a local audio-visual prompt and generate a device abnormality notification to be sent to the remote management platform.

5. The method according to claim 1, wherein, The basic risk level in S3 is divided according to the following rules: A water level of less than 5 cm is considered safe. A water level output of 5 cm or more but less than 15 cm is classified as a warning level. A water level output of 15 cm or more but less than 30 cm is considered a warning level. A water level of 30 cm or more is considered dangerous. When the input is a second confidence level, the risk level is adjusted to at least one level higher than the security control level. Specifically: when the basic risk level is safe, it is adjusted to the attention level; when the basic risk level is attention, it is adjusted to the alert level; when the basic risk level is alert, it is adjusted to the danger level; when the basic risk level is already danger, the danger level is maintained.

6. The method according to claim 4, characterized in that, The risk grading assessment model in S3 is configured as follows: Get the continuous sequence before the current time Calculate the rate of change of water level based on the output water level values ​​at each time point. When the rate of change of water level Greater than the preset rate of increase threshold At that time, the basic risk level will be temporarily raised by one level based on the original rules, and a "rapid increase" marker will be generated; When the input confidence level is set to the second confidence level and the water level change rate Greater than At that time, the risk level is increased by one level on the basis of the second credibility mark, that is, the safety level is increased to the alert level, the attention level is increased to the danger level, and the alert level and the danger level remain at the danger level; Rate of water level change The calculation formula is: ； wherein, is the output water level value at the current time instant, is is the output water level value at the previous time instant, is the sampling time interval.

7. The method according to claim 4, characterized in that, Local early warning control in S4 includes: When the risk level is at the attention level, the variable message sign installed at the entrance of the underpass is controlled to display a deceleration warning through the hard-wired output interface of the edge computing unit. When the risk level is alert, the variable message sign is controlled to display "Caution: Proceed with Caution" via the hard-wired output interface, and the audible and visual alarm installed at the underpass site is activated. When the risk level is dangerous, the variable message sign is controlled via the hard-wired output interface to display "No Entry", activate the audible and visual alarm, and control the gate installed at the entrance of the underpass to close.

8. The method according to claim 1, wherein, S4 remote collaborative early warning includes: sending early warning work orders to municipal management platforms and traffic management platforms. The early warning work order includes the current risk level, output water level value, first or second confidence mark, and real-time screenshot of the site; sending text messages and / or automatic voice calls to preset responsible persons; Upstream disruption warnings include: sending detour information to variable message signs at upstream intersections, and sending road closure status and suggested detour routes to navigation map service interfaces.

9. A system for risk classification and multi-channel early warning of water accumulation under a bridge, applied to the method for risk classification and multi-channel early warning of water accumulation under a bridge according to any one of claims 1-8, characterized in that, include: The front-end AI intelligent monitoring layer includes radar water level gauges, cameras, and edge computing units; The edge computing unit is connected to the radar water level gauge and the camera to obtain the first water level value and the second water level value; The data transmission layer establishes a two-way communication link with the front-end AI intelligent monitoring layer and the back-end platform layer; The backend platform layer is configured with a risk classification and assessment model, which is used to receive data uploaded from the frontend and issue control instructions. The early warning release layer includes local early warning equipment, remote early warning modules, and upstream guidance modules; Among them, the local early warning equipment is directly connected to the edge computing unit via hard wiring, and the edge computing unit has a built-in backup power supply and a lightweight risk classification assessment model; The edge computing unit is used to respond to risk level instructions issued by the backend platform layer to drive the local early warning device when the communication link is normal, or to independently complete the risk level assessment locally using a lightweight risk grading assessment model when a communication link interruption is detected, and directly drive the local early warning device based on the assessment results.

10. The risk classification assessment and multi-channel early warning system for underpass bridge water according to claim 9, characterized in that: The hardwired circuit is either a relay output line or an RS485 bus; the local early warning equipment includes a variable message sign installed at the entrance of the underpass, an audible and visual alarm installed at the underpass site, and a barrier gate installed at the entrance of the underpass, all of which are directly connected to the edge computing unit via hardwired circuits. The remote early warning module includes data interfaces for connecting with municipal management platforms and traffic management platforms, as well as SMS notification and voice call modules; The upstream guidance module includes a data interface with the variable message sign at the upstream intersection and a data interface with the navigation map service provider.

Images