Urban inland inundation risk intelligent monitoring and early warning device based on short and temporary rainfall forecast

Abstract

The invention discloses an urban inland inundation risk intelligent monitoring and early warning device based on short and temporary rainfall forecast, and relates to the technical field of urban disaster prevention and reduction and smart cities. The device comprises a generative approach rainfall forecasting module, an urban hydrological underlying surface module, an intelligent waterlogging risk deduction module, a high-performance calculation and data base and an intelligent emergency decision support module, the generative approach rainfall forecasting module outputs high-temporal-spatial-resolution rainfall data, and the urban hydrological underlying surface module provides static and dynamic hydrological data. Inputting the two into an intelligent waterlogging risk deduction module to calculate a dynamic risk level; the high-performance computing and data base provides computing power and data support for all the modules, the intelligent emergency decision support module generates a dynamic risk avoiding path and an emergency plan based on a risk result, and the device improves waterlogging early warning accuracy and emergency decision effectiveness and assists active defense of urban waterlogging.

Description

Technical Field

[0001] This invention relates to the field of urban disaster prevention and mitigation and smart city technology, specifically to an intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasts. Background Technology

[0002] Against the backdrop of rapid global climate change and urbanization, extreme heavy rainfall events are becoming increasingly frequent, and urban flooding has become a prominent problem restricting the safe operation and sustainable development of cities. To address this challenge, various urban flooding monitoring and early warning technologies and systems have emerged. For example, the invention patent CN119904101A, a method for dynamic assessment of critical rainfall at waterlogging points and prediction of waterlogging risk, improves the accuracy and real-time performance of waterlogging prediction by dynamically tracking rainfall events and setting multi-level critical rainfall thresholds. This technology represents a current approach to risk assessment through threshold optimization. Another example is the invention patent CN119620243A, a method and system for forecasting short-term rainfall and floods based on deep learning, which constructs a radar echo extension prediction model based on deep learning, aiming to improve the accuracy of short-term rainfall and subsequent flood forecasts. These existing technological solutions have all made beneficial explorations into improving urban flooding early warning capabilities at specific stages.

[0003] However, existing technologies still face significant technical bottlenecks when dealing with rapidly evolving and complex urban flooding disaster chains. The main problem is that traditional urban flooding early warning systems suffer from insufficient spatiotemporal accuracy in short-term precipitation forecasts and static risk assessments, failing to match the rapid, dynamic, and precise triggering mechanisms of flooding disaster chains. This results in delayed warnings, ambiguous location data, and passive emergency responses. Specifically, on the one hand, forecasting technologies such as CN119620243A, while employing deep learning models, still lack sufficient spatiotemporal resolution to accurately capture minute-level precipitation processes at the urban street scale, failing to meet the core requirements of precise, accurate, and rapid forecasting of flooding sources. On the other hand, risk assessment methods, such as CN119904101A, while introducing the concept of dynamic thresholds, still rely on a pre-set rainfall threshold system in their risk assessment models. This fails to achieve deep coupling and dynamic simulation with multi-dimensional data such as high-precision precipitation forecasts, urban underlying surface physical characteristics, and real-time hydrological conditions, resulting in a disconnect between risk assessment results and the rapidly changing reality of urban flooding. Furthermore, existing technical solutions suffer from a gap in the transformation chain from risk warning to emergency decision-making, failing to form a closed loop and making it difficult for warning information to directly support precise and efficient emergency actions.

[0004] In summary, existing technologies have shortcomings in terms of forecast accuracy for urban flooding early warning, dynamic adaptability of risk assessment, and closed-loop linkage between early warning and response. Therefore, there is an urgent need for an innovative solution that can fundamentally improve the accuracy of short-term precipitation forecasts and, on this basis, achieve dynamic and refined risk extrapolation and intelligent decision support. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasts. By integrating generative nowcasting of precipitation, urban hydrological underlying surface, intelligent flooding risk simulation, high-performance computing and data base, and intelligent emergency decision support module, it can improve the accuracy of precipitation forecasts with potential diffusion models, conduct risk assessments by integrating static and dynamic data, and ensure real-time performance through heterogeneous computing, thereby achieving accurate monitoring and early warning of urban flooding and intelligent emergency decision-making.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent monitoring and early warning device for urban waterlogging risk based on short-term precipitation forecasting, comprising a generative near-term precipitation forecasting module, an urban hydrological underlying surface module, an intelligent waterlogging risk simulation module, a high-performance computing and data base, and an intelligent emergency decision support module;

[0007] The generative nowcasting module is used to generate high spatiotemporal resolution nowcasting data based on a potential diffusion model.

[0008] The urban hydrological underlying surface module is used to integrate static geographic elements and dynamic hydrological sensing data.

[0009] The intelligent urban flooding risk simulation module is used to calculate a dynamic risk index and classify early warning levels based on precipitation forecast data and underlying surface data.

[0010] The high-performance computing and data infrastructure is used to provide heterogeneous computing resources and real-time data processing pipelines;

[0011] The intelligent emergency decision support module is used to generate dynamic risk avoidance paths and emergency response plans based on risk simulation results.

[0012] The output of the generative nowcasting module is connected to the input of the intelligent urban flooding risk simulation module, the output of the urban hydrological underlying surface module is connected to another input of the intelligent urban flooding risk simulation module, the output of the intelligent urban flooding risk simulation module is connected to the input of the intelligent emergency decision support module, and the high-performance computing and data foundation provides computing and data support for all modules.

[0013] Furthermore, the generative nowcasting module includes a generative artificial intelligence engine based on a potential diffusion model. This engine employs a hybrid architecture of a three-dimensional earth transformer and a U-shaped network to process meteorological radar echo data and generate precipitation forecast sequences for the next 0 to 3 hours.

[0014] The generative artificial intelligence engine introduces a physical constraint loss function during the training process, which is embedded in the water vapor mass conservation equation as a regularization term.

[0015] The generative nowcasting module also includes a rolling forecasting mechanism, which receives the latest radar observation data in real time at fixed time intervals and drives the generative artificial intelligence engine to output precipitation forecast results with a spatial resolution of hundreds of meters.

[0016] The generative nowcasting module compresses the features of high-dimensional radar data through a variational autoencoder and performs a diffusion process of forward noise addition and reverse noise reduction in the latent space.

[0017] Furthermore, the hybrid architecture of the three-dimensional earth transformer and the U-shaped network is specifically composed of an earth transformer component and a U-shaped network component;

[0018] The Earth Transformer component is responsible for capturing the long-term spatiotemporal dependencies of radar echo sequences, while the U-shaped network component is responsible for encoding and decoding high-resolution features.

[0019] The specific form of the physical constraint loss function is a weighted sum of the mean square error loss and the water vapor mass conservation deviation;

[0020] The fixed time period for the rolling forecast mechanism is 6 minutes;

[0021] The generative now-near precipitation forecast module also includes a data preprocessing unit, which performs spatiotemporal benchmark alignment, ground clutter filtering, and multi-scale interpolation fusion of multi-source radar base data to unify the data into a high-resolution grid.

[0022] Furthermore, the urban hydrological underlying surface module includes a static geographic element digitization unit and a dynamic hydrological status real-time sensing unit;

[0023] The static geographic element digitization unit integrates a high-precision digital elevation model obtained by lidar, an impermeable surface distribution map interpreted from high-resolution remote sensing images, and a digital archive of the urban drainage system.

[0024] The dynamic hydrological status real-time sensing unit connects to IoT water level sensor data in real time and builds a city water level map, while managing a knowledge base of historical urban flooding disasters.

[0025] The digital archive of the urban drainage system includes the topological relationships and geometric properties of stormwater pipe networks, pumping stations, and sluice gates;

[0026] The historical flood disaster knowledge base stores structured event data, including time, location, water depth, and affected area.

[0027] Furthermore, the intelligent urban flooding risk simulation module includes a dynamic disaster-causing threshold optimization unit and a comprehensive risk index calculation unit;

[0028] The dynamic disaster threshold optimization unit defines multiple time windows and uses the TS score as the objective function to calculate the optimal disaster threshold for each time window through a traversal search algorithm.

[0029] The comprehensive risk index calculation unit calculates a standardized comprehensive risk index based on precipitation forecast data and the optimal disaster-causing threshold. The mathematical formula is as follows:

[0030]

[0031] in, This represents the overall risk index. Indicates the total number of time windows. Indicates the time window index. Indicates the first The cumulative precipitation over a time window, Indicates the first The optimal disaster-causing threshold for each time window;

[0032] The intelligent urban flooding risk simulation module also includes an early warning level classification unit, which classifies multiple early warning levels based on the statistical distribution of historical risk indices.

[0033] Furthermore, the total number of time windows in the dynamic disaster threshold optimization unit is 10, and the time windows cover the range from short duration to long duration.

[0034] The warning level classification unit divides the warning level into four levels based on the percentiles of the historical distribution of the risk index. The percentiles include the 25th percentile, the 50th percentile, the 75th percentile, and the 90th percentile.

[0035] The intelligent urban flooding risk simulation module also includes a risk space correction unit, which combines topographic elevation, impermeable surface ratio and drainage network density to divide the city into flood-sensitive areas and dynamically correct the comprehensive risk index.

[0036] The risk space correction unit adjusts the early warning level through the sensitivity level matrix.

[0037] Furthermore, the high-performance computing and data infrastructure includes a heterogeneous computing resource pool and an end-to-end real-time processing pipeline;

[0038] The heterogeneous computing resource pool consists of a GPU computing resource pool and a CPU computing cluster. The GPU computing resource pool is dedicated to parallel inference tasks of generative artificial intelligence models, while the CPU computing cluster is responsible for cleaning IoT sensor data streams and performing risk index matrix operations.

[0039] The end-to-end real-time processing pipeline uses a model optimization framework and a streaming data processing engine to ensure that the end-to-end latency from data access to early warning issuance is controlled within minutes.

[0040] The model optimization framework is TensorRT;

[0041] The streaming data processing engine is Apache Flink.

[0042] Furthermore, the intelligent emergency decision support module includes a dynamic traffic network risk avoidance unit and a tiered response and early warning release unit;

[0043] The dynamic traffic network risk avoidance unit maps the results of urban flooding risk simulation to the urban digital road network, dynamically adjusts the road segment toll costs, and uses an improved A-star algorithm to calculate the optimal dynamic risk avoidance path.

[0044] The tiered response early warning release unit automatically generates emergency response plans based on the early warning level and the distribution of waterlogging points, and uses geofencing technology to accurately push early warning information to terminals within the risk area;

[0045] The improved A* algorithm introduces a risk weighting factor in path planning;

[0046] The emergency response plan includes the deployment points of mobile drainage pumping vehicles, recommendations for closing risky road sections, and the dispatch direction of emergency rescue teams.

[0047] Compared with existing technologies, this intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasts has the following advantages:

[0048] I. This invention constructs a generative nowcasting precipitation module, employing a potential diffusion model combined with a physical constraint loss function to process meteorological radar data, generating high spatiotemporal resolution short-term precipitation forecast data. Simultaneously, it integrates static geographic elements and dynamic hydrological sensing data from an urban hydrological underlying surface module, providing multi-dimensional input for an intelligent urban flooding risk simulation module. This simulation module achieves dynamic and refined risk assessment through dynamic disaster threshold optimization and risk spatial correction, accurately capturing precipitation processes at the urban street scale. This ensures that risk assessment matches the actual evolution of urban flooding, avoiding delayed warnings and ambiguous positioning. It solves the problems of insufficient accuracy in short-term precipitation forecasts, static risk assessments, and gaps between early warning and emergency decision-making in existing technologies, providing accurate source data and dynamic assessment support for urban flooding early warning.

[0049] Second, this invention integrates heterogeneous computing resources and an end-to-end real-time processing pipeline through high-performance computing and a data foundation, ensuring the efficiency of data processing and model inference in each module and ensuring the real-time generation of early warning information. At the same time, the intelligent emergency decision support module maps the risk simulation results to the urban road network, generates dynamic risk avoidance paths through improved algorithms, and uses geofencing technology to accurately push early warning information and emergency response plans to risk areas. This enables a technical closed loop from precipitation forecasting and risk assessment to emergency decision-making, thereby improving the efficiency of urban flood emergency response and reducing the impact of flood disasters on traffic operations and public safety.

[0050] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0052] Figure 1 This is a flowchart illustrating the operation of the present invention;

[0053] Figure 2 This is a system architecture diagram of the present invention;

[0054] Figure 3 This is a dynamic risk deduction logic diagram for the present invention. Detailed Implementation

[0055] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0056] Example 1

[0057] like Figure 2 and Figure 3As shown, this embodiment aims to specifically implement an intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting. It addresses the problems of insufficient spatiotemporal accuracy, static risk assessment, and coarse-grained emergency decision support in traditional flooding early warning systems by constructing a complete technical system encompassing generative nowcasting of precipitation, urban hydrological underlying surface perception, intelligent flooding risk simulation, high-performance computing support, and smart emergency decision support. This embodiment optimizes the accuracy of short-term precipitation forecasting by introducing a potential diffusion model, integrates static geographic elements and dynamic hydrological data to construct a digital twin foundation, establishes a dynamic risk assessment model based on historical disaster data, and combines heterogeneous computing resources to ensure real-time performance. Ultimately, it outputs precise street-level early warning information and implementable emergency response plans, realizing a shift from passive response to proactive defense against urban flooding.

[0058] Overall Device Architecture: The intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting described in this embodiment mainly consists of a generative nowcasting precipitation forecasting module, an urban hydrological underlying surface module, an intelligent flooding risk simulation module, a high-performance computing and data base, and an intelligent emergency decision support module. These modules work together through data interfaces and communication protocols.

[0059] The output of the generative now-near precipitation forecast module is connected to the first input of the intelligent urban flooding risk simulation module, and is used to provide the risk simulation module with high spatiotemporal resolution now-near precipitation forecast data;

[0060] The output end of the urban hydrological underlying surface module is connected to the second input end of the intelligent urban flooding risk simulation module, which is used to provide the risk simulation module with urban static geographic elements and dynamic hydrological sensing data.

[0061] The output of the intelligent urban flooding risk simulation module is connected to the input of the intelligent emergency decision support module, and is used to provide the decision support module with dynamic risk index and early warning level results;

[0062] The high-performance computing and data base connects to the other four modules via an internal bus, providing heterogeneous computing resources and real-time data processing support to each module, ensuring the efficient operation of the entire device.

[0063] Implementation of the Generative Nowcasting Module for Precipitation: The core of the generative nowcasting module for precipitation is to provide precipitation forecasts for the next 0-3 hours with a spatial resolution of hundreds of meters and a time update frequency of minutes. The specific implementation steps are as follows:

[0064] Data preprocessing: Multi-source meteorological radar base data are collected. First, a spatiotemporal reference alignment operation is performed to unify the observation data from different radar stations into the same spatiotemporal coordinate system. Then, an adaptive filtering algorithm is used to filter out ground clutter and anomalous signals, eliminating the interference of non-precipitation factors on the data. Finally, a multi-scale interpolation fusion algorithm is used to map the processed radar data onto a high-resolution grid designed for urban flooding simulation, ensuring that the spatial accuracy of precipitation data can support street-level risk extrapolation.

[0065] Construction and training of the latent diffusion network: A variational autoencoder is used to compress the features of the preprocessed high-dimensional radar data, mapping the high-dimensional radar echo data to a low-dimensional latent space to reduce the computational complexity of subsequent operations. Based on the low-dimensional latent space, a 3D-Earthformer-Unet hybrid network is constructed. The Earthformer component is used to capture the long-term spatiotemporal dependencies of the radar echo sequence and accurately identify the evolution patterns of local severe convective weather. The U-shaped network component is used to encode and decode high-resolution features to ensure the spatial details of the forecast results.

[0066] During training, a physical constraint loss function is introduced. This function is a weighted sum of the mean square error loss and the water vapor mass conservation deviation, as shown in the following formula:

[0067]

[0068] in, For the total loss function, These are the weighting coefficients. To compensate for the mean square error loss between the forecast radar echo and the observed radar echo, and to ensure the overall accuracy of the forecast results, To mitigate the loss due to water vapor mass conservation bias, the water vapor mass conservation equation is embedded as a regularization term to constrain the model output to conform to atmospheric physical laws, thus avoiding physically unreasonable forecast results.

[0069] Rolling forecast mechanism operation: A rolling forecast mechanism with a fixed time period is established, which is set to 6 minutes, to ensure that the dynamic changes of precipitation processes can be captured in real time;

[0070] Every 6 minutes, the module receives the latest radar observation data in real time and inputs it into the trained 3D-Earthformer-Unet network, driving the network to generate precipitation forecast sequences for the next 0-3 hours. The output is gridded precipitation data with a spatial resolution of hundreds of meters, including the cumulative precipitation information of each grid at different time points.

[0071] Implementation of the Urban Hydrological Underlying Surface Module: The urban hydrological underlying surface module is used to construct a digital twin substrate for simulating urban flooding, realizing the digitization of static geographic elements and real-time perception of dynamic hydrological conditions. The specific implementation steps are as follows:

[0072] Digitalization of static geographic elements: Topographic data of urban areas are collected using lidar technology to construct a high-precision digital elevation model. This model is used to accurately identify flood-prone micro-topographic features such as urban depressions and underpasses. High-resolution remote sensing images are used to interpret the distribution of impervious surfaces in the city through semantic segmentation algorithms to generate an impervious surface distribution map, providing key parameters for subsequent runoff calculations.

[0073] Establish a digital archive of the urban drainage system, collect basic information on drainage facilities such as rainwater pipe networks, pumping stations, and sluice gates, including the topological relationships and geometric attributes of the facilities, and store this information in a structured manner to form a digital ledger of the drainage system.

[0074] Real-time dynamic hydrological status perception: IoT water level sensors are deployed at key urban flooding points, low-lying areas along roads, riverbanks, and manholes. The sensors upload water level data in real time via wireless communication protocols. The module connects to the data interfaces of all sensors to receive, clean, and integrate the uploaded water level data in real time, generating a single map of urban water levels that intuitively displays the real-time water level status of various areas of the city.

[0075] A historical flood disaster knowledge base was constructed, systematically sorting out the records of flood events in the target city over the years. Information such as the time, location, water depth, and scope of impact of the events was structured and processed to form a standardized historical disaster dataset, which was used for subsequent verification, calibration and optimization of risk models.

[0076] The intelligent urban flooding risk simulation module, based on precipitation forecast data and underlying surface data, realizes dynamic risk index calculation and early warning level classification. The specific implementation steps are as follows:

[0077] Dynamic disaster threshold optimization: 10 time windows covering short to long durations are defined to comprehensively characterize the disaster effects of different precipitation patterns; historical flooding event records of the target city are selected as positive samples, and historical events with heavy rainfall but no flooding are selected as negative samples, which together constitute the dataset for threshold optimization.

[0078] The TS score is used as the objective function. The TS score is used to measure the degree of agreement between the forecast results and the actual disaster situation. The formula is as follows:

[0079]

[0080] in, To determine the sample size for those predicted to experience flooding and those that actually did, To determine the sample size for those predicted to experience flooding but which did not actually occur, This represents the sample size of actual flooding that occurred but was not predicted or warned against.

[0081] By employing a traversal search algorithm, different cumulative precipitation thresholds are tested one by one for each time window. The TS score corresponding to each threshold is calculated, and the cumulative precipitation corresponding to the maximum TS score is determined as the optimal disaster-causing threshold for that time window. .

[0082] Comprehensive risk index calculation: based on the cumulative precipitation for each time window output by the generative nowcasting module. With the optimal disaster-causing threshold The comprehensive risk index is calculated using a standardized formula. The formula is as follows:

[0083]

[0084] in, This is a comprehensive risk index used to quantify the degree of urban flooding risk. For the first Forecast cumulative precipitation for each time window, in units of Maintain consistency; For the first The optimal disaster-causing threshold for a given time window represents the critical precipitation intensity that triggers the risk of urban flooding within that time window.

[0085] Warning level classification and risk correction: Data collection from multiple historical flooding events. The data is statistically analyzed to determine... The 25th percentile of the distribution ( ), 50th percentile ( ), 75th percentile ( ) and the 90th percentile ( Based on these percentiles, four warning levels are defined:

[0086] Level I Warning: This indicates that the area is at extremely high risk of flooding.

[0087] Level II Warning: This indicates a high risk of flooding in the area;

[0088] Level III Warning: This indicates a high risk of flooding in the area;

[0089] Level IV Warning: This indicates that the risk of flooding in the area is generally low.

[0090] Combining topographic elevation, impervious surface ratio, and drainage network density data provided by the urban hydrological underlying surface module, the analytic hierarchy process (AHP) is used to divide the city into three flood-sensitive zones: high, medium, and low. A sensitivity level matrix is ​​then constructed, and the cities are classified according to their regional sensitivity levels. The corresponding warning level is dynamically adjusted. For example, in highly sensitive areas, if the original warning level is Level III, it will be adjusted to Level II to improve the accuracy of warnings for flood-prone areas.

[0091] Implementation of the High-Performance Computing and Data Infrastructure: The high-performance computing and data infrastructure provides computing power and real-time data processing capabilities for the entire device. The specific implementation steps are as follows:

[0092] Construction of heterogeneous computing resource pool: Build a GPU computing resource pool, select high-performance graphics processors, and form a parallel computing cluster through distributed deployment. It is specifically used to process the inference task of the 3D-Earthformer-Unet network in the generative nowcasting module, and improve the model inference speed by utilizing the parallel computing capabilities of GPUs.

[0093] A CPU computing cluster was built, using a multi-core central processing unit. Task allocation was achieved through load balancing technology. This cluster was responsible for handling the high-concurrency data stream cleaning task of IoT water level sensors in the urban hydrological underlying surface module, as well as the risk index matrix calculation task covering tens of thousands of grids throughout the city in the intelligent urban flooding risk simulation module, ensuring the efficiency of data processing and calculation.

[0094] End-to-end real-time processing pipeline construction: The 3D-Earthformer-Unet network in the generative nowcasting precipitation forecast module is optimized using the TensorRT framework. Model inference latency is reduced through model quantization, layer fusion and other technologies to ensure that the time from receiving the latest radar data to outputting precipitation forecast results is controlled within a few seconds.

[0095] Using Apache Flink as the streaming data processing engine, an end-to-end real-time data processing pipeline is built to achieve fully automated processing from multi-source data access, cleaning, and fusion to model calculation, risk assessment, and early warning release. The end-to-end latency of the entire device is controlled at the minute level to meet the real-time requirements of urban flooding early warning.

[0096] A distributed storage system is deployed, combining distributed file storage with a database to store massive amounts of historical radar data, disaster data, and real-time structured data, ensuring the security and accessibility of data storage.

[0097] Implementation of the Intelligent Emergency Decision Support Module: The Intelligent Emergency Decision Support Module generates practical and actionable emergency plans based on risk simulation results. The specific implementation steps are as follows:

[0098] Dynamic traffic network risk avoidance: The regional risk level results output by the intelligent waterlogging risk simulation module are mapped to the urban digital road network in real time, and a correlation rule between risk and road segment passage cost is established—the higher the waterlogging risk level, the greater the weight of the corresponding road segment passage cost.

[0099] An improved A algorithm is used for path planning. A risk weight factor is introduced into the heuristic function of the traditional A algorithm, as shown in the following formula:

[0100]

[0101] in, For nodes The total cost, From the starting point to the node The actual cost, For the node The estimated cost to reach the destination. To estimate the cost weighting coefficient, For risk weighting coefficients, For nodes Risk weight of the road segment;

[0102] The algorithm calculates the optimal dynamic avoidance routes for both civilian vehicles and emergency rescue vehicles, and then publishes the route information to users through navigation applications, road traffic guidance screens, and other terminals to guide vehicles to avoid areas at risk of flooding.

[0103] Tiered response and early warning issuance: Based on the early warning level and distribution of flooding points output by the intelligent urban flooding risk simulation module, emergency response plans are automatically generated: For Level I warnings, the optimal deployment points of mobile drainage pumping vehicles are recommended, risky road sections are suggested to be closed, and the dispatch direction of emergency rescue teams is determined; for Level II and below warnings, the content of the plan is adjusted according to the degree of risk, and the scale of emergency resource input is reduced.

[0104] By utilizing geofencing technology and combining it with high-precision positioning data from user terminals, the corresponding level of early warning information is pushed point-to-point to public terminals within the risk area. At the same time, the early warning information and emergency response plans are pushed to the terminals of responsible persons in urban management, transportation, emergency response and other departments to ensure the accurate delivery of early warning information.

[0105] In summary, this embodiment achieves the complete implementation of an intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting by specifically constructing a generative nowcasting precipitation forecasting module, an urban hydrological underlying surface module, an intelligent urban flooding risk simulation module, a high-performance computing and data foundation, and an intelligent emergency decision support module. This embodiment improves the spatiotemporal accuracy of short-term precipitation forecasting through a potential diffusion model, achieves dynamic risk assessment by fusing static and dynamic data, ensures the real-time performance of the device through heterogeneous computing resources, and achieves precise emergency decision-making through improved algorithms and geofencing technology. It forms a complete technical closed loop from precipitation forecasting to risk assessment to emergency response, effectively solving the technical bottlenecks of traditional urban flooding early warning systems. It provides urban managers with forward-looking and refined means of urban flooding risk management, significantly improving the city's ability to cope with sudden extreme rainfall, which aligns with the core objective of this invention: empowering proactive urban flooding prevention.

[0106] Example 2

[0107] like Figure 1 As shown in Example 1, this example elaborates on the specific steps of an intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasts during operation. The specific steps are as follows:

[0108] 1. Multi-source data acquisition and preprocessing:

[0109] Real-time acquisition of multi-source meteorological radar base data, execution of spatiotemporal benchmark alignment, ground clutter filtering and multi-scale interpolation fusion operations, and unification of data into a high-resolution grid.

[0110] It integrates static geographic feature data, including high-precision digital elevation models acquired by lidar, impermeable surface distribution maps interpreted from high-resolution remote sensing images, and digital archives of urban drainage systems.

[0111] Real-time access to IoT water level sensor data is performed, data is cleaned and integrated to build a city water level map, and the historical flood disaster knowledge base is updated.

[0112] 2. Generative short-term precipitation forecast:

[0113] The preprocessed radar data is input into the generative nowcasting module, and inference is performed using a generative artificial intelligence engine based on a potential diffusion model.

[0114] Through a rolling forecast mechanism, the model is driven to generate a precipitation forecast sequence for the next 0-3 hours every 6 minutes, outputting gridded precipitation data with a spatial resolution of hundreds of meters.

[0115] 3. Intelligent Urban Flooding Risk Simulation:

[0116] Dynamic disaster threshold optimization: Based on historical flooding event data, multiple time windows are defined, and the TS score is used as the objective function. The optimal disaster threshold for each time window is calculated through a traversal search algorithm.

[0117] Comprehensive risk index calculation: Based on precipitation forecast data and the optimal disaster-causing threshold for each time window, the comprehensive risk index for each grid is calculated.

[0118] Early warning level classification: Based on the historical distribution percentile of the comprehensive risk index, the risk is classified into four levels of early warning.

[0119] Risk Space Correction: Combining topographic elevation, impermeable surface ratio and drainage network density data provided by the urban hydrological underlying surface module, the warning level is dynamically corrected through a sensitivity level matrix.

[0120] 4. Intelligent emergency decision support:

[0121] Dynamic traffic network risk avoidance: The results of urban flooding risk simulation are mapped to the city's digital road network, the road segment toll costs are dynamically adjusted, and the optimal dynamic risk avoidance path is calculated using an improved A* algorithm to provide path guidance for social vehicles and emergency rescue vehicles.

[0122] Tiered response and early warning issuance: Based on the warning level and the distribution of waterlogged areas, emergency response plans are automatically generated, including the deployment points of mobile drainage pumping vehicles, recommendations for closing risky road sections, and the dispatch direction of emergency rescue teams.

[0123] By using geofencing technology, early warning information and emergency response plans can be accurately pushed to public terminals and relevant department personnel terminals within the risk area.

[0124] 5. High-performance computing and data infrastructure support:

[0125] Use GPU computing resource pools to process inference tasks of generative artificial intelligence models in parallel.

[0126] CPU computing clusters are used to process IoT sensor data stream cleaning and risk index matrix operations.

[0127] By using an end-to-end real-time processing pipeline, the entire process from data access to early warning issuance is automated, with end-to-end latency controlled within minutes.

[0128] The entire device achieves a closed-loop process from data acquisition, precipitation forecasting, risk simulation to emergency decision-making through the above steps, thereby improving the monitoring and early warning capabilities of urban flooding risks and the efficiency of emergency response.

[0129] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasts, characterized in that, It includes a generative now-nearest precipitation forecast module, an urban hydrological underlying surface module, an intelligent urban flooding risk simulation module, a high-performance computing and data foundation, and an intelligent emergency decision support module; The generative nowcasting module is used to generate high spatiotemporal resolution nowcasting data based on a potential diffusion model. The urban hydrological underlying surface module is used to integrate static geographic elements and dynamic hydrological sensing data. The intelligent urban flooding risk simulation module is used to calculate a dynamic risk index and classify early warning levels based on precipitation forecast data and underlying surface data. The high-performance computing and data infrastructure is used to provide heterogeneous computing resources and real-time data processing pipelines; The intelligent emergency decision support module is used to generate dynamic risk avoidance paths and emergency response plans based on risk simulation results. The output of the generative nowcasting module is connected to the input of the intelligent urban flooding risk simulation module, the output of the urban hydrological underlying surface module is connected to another input of the intelligent urban flooding risk simulation module, the output of the intelligent urban flooding risk simulation module is connected to the input of the intelligent emergency decision support module, and the high-performance computing and data foundation provides computing and data support for all modules.

2. The intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting as described in claim 1, characterized in that, The generative now-nearest precipitation forecast module includes a generative artificial intelligence engine based on a potential diffusion model. This engine adopts a hybrid architecture of a three-dimensional earth transformer and a U-shaped network to process meteorological radar echo data and generate precipitation forecast sequences for the next 0 to 3 hours. The generative artificial intelligence engine introduces a physical constraint loss function during the training process, which is embedded in the water vapor mass conservation equation as a regularization term. The generative nowcasting module also includes a rolling forecasting mechanism, which receives the latest radar observation data in real time at fixed time intervals and drives the generative artificial intelligence engine to output precipitation forecast results with a spatial resolution of hundreds of meters. The generative nowcasting module compresses the features of high-dimensional radar data through a variational autoencoder and performs a diffusion process of forward noise addition and reverse noise reduction in the latent space.

3. The intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting as described in claim 2, characterized in that, The hybrid architecture of the three-dimensional earth transformer and U-shaped network is specifically composed of earth transformer components and U-shaped network components; The Earth Transformer component is responsible for capturing the long-term spatiotemporal dependencies of radar echo sequences, while the U-shaped network component is responsible for encoding and decoding high-resolution features. The specific form of the physical constraint loss function is a weighted sum of the mean square error loss and the water vapor mass conservation deviation; The fixed time period for the rolling forecast mechanism is 6 minutes; The generative now-near precipitation forecast module also includes a data preprocessing unit, which performs spatiotemporal benchmark alignment, ground clutter filtering, and multi-scale interpolation fusion of multi-source radar base data to unify the data into a high-resolution grid.

4. The intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting as described in claim 1, characterized in that, The urban hydrological underlying surface module includes a static geographic element digitization unit and a dynamic hydrological status real-time sensing unit. The static geographic element digitization unit integrates a high-precision digital elevation model obtained by lidar, an impermeable surface distribution map interpreted from high-resolution remote sensing images, and a digital archive of the urban drainage system. The dynamic hydrological status real-time sensing unit connects to IoT water level sensor data in real time and builds a city water level map, while managing a knowledge base of historical urban flooding disasters. The digital archive of the urban drainage system includes the topological relationships and geometric properties of stormwater pipe networks, pumping stations, and sluice gates; The historical flood disaster knowledge base stores structured event data, including time, location, water depth, and affected area.

5. The intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting as described in claim 1, characterized in that, The intelligent urban flooding risk simulation module includes a dynamic disaster threshold optimization unit and a comprehensive risk index calculation unit. The dynamic disaster threshold optimization unit defines multiple time windows and uses the TS score as the objective function to calculate the optimal disaster threshold for each time window through a traversal search algorithm. The comprehensive risk index calculation unit calculates a standardized comprehensive risk index based on precipitation forecast data and the optimal disaster-causing threshold. The mathematical formula is as follows: in, This represents the overall risk index. Indicates the total number of time windows. Indicates the time window index. Indicates the first The cumulative precipitation over a time window, Indicates the first The optimal disaster-causing threshold for each time window; The intelligent urban flooding risk simulation module also includes an early warning level classification unit, which classifies multiple early warning levels based on the statistical distribution of historical risk indices.

6. The intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting as described in claim 5, characterized in that, The total number of time windows in the dynamic disaster threshold optimization unit is 10, and the time windows cover the range from short duration to long duration. The warning level classification unit divides the warning level into four levels based on the percentiles of the historical distribution of the risk index. The percentiles include the 25th percentile, the 50th percentile, the 75th percentile, and the 90th percentile. The intelligent urban flooding risk simulation module also includes a risk space correction unit, which combines topographic elevation, impermeable surface ratio and drainage network density to divide the city into flood-sensitive areas and dynamically correct the comprehensive risk index. The risk space correction unit adjusts the early warning level through the sensitivity level matrix.

7. The intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting as described in claim 1, characterized in that, The high-performance computing and data infrastructure includes a heterogeneous computing resource pool and an end-to-end real-time processing pipeline. The heterogeneous computing resource pool consists of a GPU computing resource pool and a CPU computing cluster. The GPU computing resource pool is dedicated to parallel inference tasks of generative artificial intelligence models, while the CPU computing cluster is responsible for cleaning IoT sensor data streams and performing risk index matrix operations. The end-to-end real-time processing pipeline uses a model optimization framework and a streaming data processing engine to ensure that the end-to-end latency from data access to early warning issuance is controlled within minutes. The model optimization framework is TensorRT; The streaming data processing engine is Apache Flink.

8. The intelligent monitoring and early warning device for urban flooding risk based on short-term precipitation forecasting as described in claim 1, characterized in that, The intelligent emergency decision support module includes a dynamic traffic network risk avoidance unit and a graded response and early warning release unit. The dynamic traffic network risk avoidance unit maps the results of urban flooding risk simulation to the urban digital road network, dynamically adjusts the road segment toll costs, and uses an improved A-star algorithm to calculate the optimal dynamic risk avoidance path. The tiered response early warning release unit automatically generates emergency response plans based on the early warning level and the distribution of waterlogging points, and uses geofencing technology to accurately push early warning information to terminals within the risk area; The improved A* algorithm introduces a risk weighting factor in path planning; The emergency response plan includes the deployment points of mobile drainage pumping vehicles, recommendations for closing risky road sections, and the dispatch direction of emergency rescue teams.

Images