Urban inland inundation ponding point analysis and prediction algorithm and system based on large model and big data technology

Abstract

The invention relates to the technical field of urban inland inundation monitoring and prediction, and discloses an urban inland inundation ponding point analysis and prediction algorithm and system based on a large model and a big data technology. Collecting multi-source heterogeneous data including hydrological data, real-time weather forecast data, topographic data and historical waterlogging data of key nodes of the rainwater pipe network; s2, cleaning, standardizing and fusing the multi-source heterogeneous data to form a structured input data matrix; through multi-source big data fusion and big model dynamic simulation, by means of an advanced urban inland inundation early warning and monitoring system and in combination with big data analysis and machine learning technologies, the prediction error of the inland inundation occurrence time can be controlled within 10 minutes, and meanwhile, the prediction precision of the inundation range is improved by more than 40%; accurate time and space references are provided for urban managers and emergency response teams, and the efficiency of coping with waterlogging disasters is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of urban flooding monitoring and prediction technology, specifically to an algorithm and system for analyzing and predicting urban flooding points based on large models and big data technology. Background Technology

[0002] With the increasing severity of global climate change and the continued advancement of urbanization, extreme rainfall events are becoming more frequent, and urban flooding has become one of the key disasters restricting the safe operation of cities. Urban flooding is characterized by its sudden onset, wide impact, and great destructive power, easily leading to traffic paralysis, infrastructure damage, property loss, and even casualties. Therefore, accurate analysis and early prediction of flood accumulation points are of great significance for improving urban disaster prevention and mitigation capabilities and ensuring urban safety.

[0003] Currently, urban flooding monitoring and prediction technologies mainly rely on traditional hydrological models, such as the SWMM model, the MIKE model, or simple statistical analysis methods, which have the following significant drawbacks: First, the data collection dimensions are limited, mostly relying on data from local rain gauges or a small number of water level monitoring points, failing to comprehensively integrate hydrological information from key nodes of the stormwater drainage network, high-precision topographic data, and real-time weather forecasts, resulting in limited monitoring coverage and poor data timeliness. Second, the prediction models have poor adaptability; traditional models are mostly static parameter models, making it difficult to dynamically adapt to real-time dynamic factors such as changes in rainfall intensity and the operational status of the drainage network, and thus unable to accurately simulate the evolution of urban flooding. Third, the prediction results lack practicality, only able to roughly determine the flooded area, unable to accurately predict the time of water accumulation, the inundation range, and the trend of water depth changes, and also unable to combine historical water accumulation data to conduct risk level assessments of key areas such as underpasses, low-lying areas, and residential communities, resulting in a lack of targeted disaster prevention and mitigation responses and an inability to effectively gain emergency response time.

[0004] In recent years, the rapid development of big data and large-scale modeling technologies has provided new technical approaches for urban flooding prediction. However, there is currently no technical solution that can fully integrate multi-source big data and achieve dynamic simulation and accurate prediction of the flooding evolution process based on large-scale models, while simultaneously completing risk identification and level assessment of key locations. To address this, we propose an algorithm and system for analyzing and predicting urban flooding points based on large-scale models and big data technologies. Summary of the Invention

[0005] The purpose of this invention is to provide an algorithm and system for analyzing and predicting urban waterlogging points based on large model and big data technology, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an algorithm for analyzing and predicting urban flooding points based on large model and big data technology, comprising the following steps: S1. Construct an all-weather, three-dimensional monitoring network to collect multi-source heterogeneous data, including hydrological data of key nodes in the rainwater pipe network, real-time weather forecast data, topographic data, and historical waterlogging data. S2. Clean, standardize, and fuse multi-source heterogeneous data to form a structured input data matrix; S3. Construct a large-scale urban flooding forecast model. When the monitored rainfall intensity reaches the preset threshold, use the large-scale urban flooding forecast model to dynamically simulate the urban flooding evolution process based on the input data matrix, and predict the occurrence time, inundation range and water depth change trend of the urban flooding event. S4. Based on the inundation range, the system automatically identifies the preset key locations affected by the inundation through spatial overlay analysis, and assesses the risk level of each key location based on a multi-dimensional indicator system that includes inundation depth, duration, location importance, and historical disaster severity. S5 outputs a report containing the prediction results and risk assessment results, and synchronizes it to the emergency management platform.

[0007] Preferably, in S1, the all-weather three-dimensional monitoring network adopts an edge computing architecture, and the edge computing devices deployed locally on the sensor nodes are used to complete the local preprocessing of data.

[0008] Preferably, in S3, the large-scale urban flooding forecast model is constructed using a deep neural network based on the Transformer architecture and combined with hydrodynamic principles. The model is trained using historical multi-source heterogeneous data and strengthens the weight allocation of core influencing factors through its self-attention mechanism.

[0009] Preferably, in S4, the preset key locations include, but are not limited to, underpasses, low-lying areas, and residential areas; the risk levels are divided into three levels: high, medium, and low.

[0010] This invention also provides a system for analyzing and predicting urban waterlogging points based on large model and big data technology, including a multi-source data acquisition module, a data preprocessing module, a large model module for waterlogging forecasting, a risk assessment module, and a result output and early warning module; The multi-source data acquisition module is connected to the data preprocessing module, the data preprocessing module is connected to the urban flooding forecasting large model module, the urban flooding forecasting large model module is connected to the risk assessment module, and the risk assessment module is connected to the result output and early warning module. The multi-source data acquisition module is used to acquire and aggregate multi-source heterogeneous data in real time; The data preprocessing module is used to clean, standardize, and fuse multi-source heterogeneous data to generate high-quality input data. The large-scale urban flood forecasting module is used to dynamically extrapolate and predict the evolution of urban flooding based on input data. The risk assessment module is used to identify key locations and assess their risk levels based on the prediction results; The result output and early warning module is used to visualize the assessment results, generate early warning information, and push it to the emergency management platform.

[0011] Preferably, the multi-source data acquisition module includes a hydrological sensing unit, a meteorological data access unit, a topographic data integration unit, and a historical data management unit; The hydrological sensing unit is used to sense and collect data on the hydrological conditions of key areas. The meteorological data access unit is configured to connect to the API interface of the meteorological department to obtain real-time weather forecast data; The terrain data integration unit is used to load and process high-precision terrain elevation datasets; The historical data management unit is used to store and manage historical flooding event data.

[0012] Preferably, the data preprocessing module includes a data cleaning unit, a data standardization unit, a data fusion unit, and a data quality verification unit; The data cleaning unit is connected to the data standardization unit, the data standardization unit is connected to the data fusion unit, and the data fusion unit is connected to the data quality verification unit. The data cleaning unit is used to remove outliers and invalid values ​​from the original data; The data standardization unit is used to unify the format and dimensions of data from different sources; The data fusion unit is used to construct the spatiotemporal correlation between multi-source data to form a structured input data matrix; The data quality verification unit is used to verify the preprocessed data.

[0013] Preferably, the large-scale flood forecasting model module includes a model training unit, a real-time simulation unit, and a model adaptation unit. The model training unit is connected to the real-time inference unit, and the real-time inference unit is connected to the model adaptation unit; The model training unit is used to optimize and train model parameters using historical data. The real-time simulation unit is used to initiate the simulation of urban flooding evolution when the rainfall intensity reaches a preset threshold. The model adaptive unit is used to dynamically adjust model parameters based on real-time data and prediction feedback.

[0014] Preferably, the risk assessment module includes a key location identification unit, an assessment indicator calculation unit, and a risk level determination unit; The key location identification unit is connected to the evaluation index calculation unit, and the evaluation index calculation unit is connected to the risk level determination unit. The key location identification unit is used to automatically identify the affected preset key locations within the predicted flooding range through spatial overlay analysis. The evaluation index calculation unit is used to calculate the value and weight of each index in the multi-dimensional index system. The risk level determination unit is used to output three risk levels: high, medium, and low, based on the comprehensive assessment results.

[0015] Preferably, the result output and early warning module includes a visualization display unit, a platform integration unit, and an early warning push unit; The visualization unit is used to generate flood forecast charts and risk level distribution maps; The platform interface unit is configured to synchronize data with the urban emergency management system; The early warning push unit is used to push early warning information to relevant management departments.

[0016] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects: By integrating multi-source big data and dynamic simulation using large models, and leveraging an advanced urban flooding early warning and monitoring system, combined with big data analysis and machine learning technologies, the prediction error of flooding occurrence time can be controlled within 10 minutes, while the prediction accuracy of the flooded area is improved by more than 40%. This provides city managers and emergency response teams with accurate time and space references, greatly improving the efficiency of responding to flooding disasters.

[0017] This invention relies on edge computing and an all-weather three-dimensional monitoring network. The data acquisition and preprocessing delay is no more than 30 seconds. The model starts simulation immediately after the rainfall reaches the standard, which can predict the risk of urban flooding 30-120 minutes in advance, and gain valuable time window for emergency response.

[0018] This invention automatically identifies key affected areas and classifies them into risk levels, enabling emergency response to focus on high-risk areas, improving the targeting and efficiency of disaster prevention and mitigation measures, and reducing disaster losses.

[0019] This invention is applicable to cities of different sizes and with different terrain features. It can be seamlessly integrated with existing urban emergency management platforms without requiring large-scale modifications to existing infrastructure, and has good prospects for widespread application. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a connection diagram of the algorithm modules of the present invention; Figure 2 This is a flowchart of the prediction method of the present invention; Figure 3 This is a schematic diagram of the large-scale model for urban flooding forecasting of the present invention. Detailed Implementation Example

[0022] Please see Figure 1-3 This invention provides a technical solution: an algorithm for analyzing and predicting urban flooding points based on large model and big data technology, comprising the following steps: S1. Construct an all-weather three-dimensional monitoring network to collect multi-source heterogeneous data, including hydrological data of key nodes in the rainwater pipe network, real-time weather forecast data, topographic data, and historical waterlogging data. The all-weather three-dimensional monitoring network adopts an edge computing architecture, with edge computing devices deployed locally on sensor nodes to perform local data preprocessing and reduce transmission latency.

[0023] S2. Clean, standardize, and fuse multi-source heterogeneous data to form a structured input data matrix; S3. Construct a large-scale urban flooding forecast model. When the monitored rainfall intensity reaches a preset threshold, the large-scale urban flooding forecast model is used to dynamically simulate the urban flooding evolution process based on the input data matrix, and predict the occurrence time, inundation range and water depth change trend of the urban flooding event. The large-scale urban flooding forecast model adopts a deep neural network based on the Transformer architecture and is constructed in combination with hydrodynamic principles. The model is trained using historical multi-source heterogeneous data and strengthens the weight allocation of core influencing factors through its self-attention mechanism.

[0024] S4. Based on the inundation range, the system automatically identifies the pre-defined key locations affected by the inundation through spatial overlay analysis, and assesses the risk level of each key location based on a multi-dimensional indicator system that includes inundation depth, duration, location importance, and historical disaster severity. The pre-defined key locations include, but are not limited to, underpasses, low-lying areas, and residential areas. The risk level is divided into three levels: high, medium, and low.

[0025] S5 outputs a report containing the prediction results and risk assessment results, and synchronizes it to the emergency management platform. The assessment indicators include flooding depth, duration, location importance, and historical disaster severity.

[0026] This invention also provides a system for analyzing and predicting urban waterlogging points based on large model and big data technology, including a multi-source data acquisition module, a data preprocessing module, a large model module for waterlogging forecasting, a risk assessment module, and a result output and early warning module; The multi-source data acquisition module is connected to the data preprocessing module, the data preprocessing module is connected to the large-scale urban flood forecasting model module, the large-scale urban flood forecasting model module is connected to the risk assessment module, and the risk assessment module is connected to the result output and early warning module. A multi-source data acquisition module is used to acquire and aggregate heterogeneous data from multiple sources in real time. This module includes a hydrological sensing unit, a meteorological data access unit, a topographic data integration unit, and a historical data management unit. Hydrological sensors are scientifically deployed at key nodes of the urban stormwater drainage network, including but not limited to network junctions, outlets, low-lying sections, and pump station inlets and outlets. The sensors employ LoRa / Wi-Fi dual-mode communication to collect core hydrological parameters such as rainfall, water level, flow velocity, and network pressure in real time, with a sampling frequency set at 2-5 minutes per sampling to ensure data timeliness. Simultaneously, it accesses high-precision meteorological forecasts issued by the meteorological department via a standardized API interface, covering rainfall intensity, duration, range, and type for the next 24 hours, updated hourly. The module also loads 1:5000 high-precision urban topographic data into the GIS system, encompassing elevation, slope, topographic classification, and land cover type data. It also analyzes data from the past 10-15 years. Historical flooding data from previous years was used to establish a historical database, which includes information such as the location coordinates of historical flooding points, inundation depth, duration, impact range, corresponding rainfall event parameters, emergency response measures, and damage, ultimately forming a multi-source heterogeneous big data set.

[0027] The hydrological sensing unit is used to sense and collect data on the hydrological conditions of key areas. The preferred hydrological sensing unit is an ultrasonic water level sensor, a rainfall sensor, a flow velocity sensor, etc. The meteorological data access unit is configured to connect to the meteorological department's API interface to obtain real-time weather forecast data. The terrain data integration unit is used to load and process high-precision terrain elevation datasets; The historical data management unit is used to store and manage historical flooding event data. It integrates hydrological sensor data of key nodes in the rainwater pipe network, real-time weather forecast data, high-precision topographic data and historical flooding data in a full-dimensional way, and builds a full-chain data system covering "rainfall-pipe network-topography-history", which solves the problems of single data dimension and insufficient timeliness of traditional technologies.

[0028] The data preprocessing module is used to clean, standardize, and fuse multi-source heterogeneous data to generate high-quality input data. The module includes a data cleaning unit, a data standardization unit, a data fusion unit, and a data quality verification unit. The module employs a "layered cleaning + standardized mapping + spatiotemporal fusion" processing strategy. In the cleaning stage, the 3σ criterion is first used to remove sensor fault data, followed by box plotting to remove outliers in meteorological data, and finally, logical checks are used to remove invalid records from historical data, such as water accumulation records without corresponding rainfall events. In the standardization stage, the min-max standardization method is used to convert data of different dimensions and units into a unified standard format within the [0, 1] interval. Continuous data such as rainfall and water level are directly standardized, while categorical data such as topography and rainfall type are converted using independent encoding. In the fusion stage, a data association algorithm is established based on an improved D-S evidence theory. This algorithm uses "timestamp + geographic coordinates" as its core. Based on the core correlation points, a spatiotemporal correlation matrix is ​​constructed to achieve deep integration of hydrological, meteorological, topographic, and historical data, ultimately forming a structured input data matrix. After data fusion, a quality verification module is used to verify the data, ensuring that data integrity is no less than 95% and the error does not exceed 3%.

[0029] like Figure 2 The flowchart is a linear process diagram, using rectangles to identify each core step. These steps include multi-source data acquisition, multi-source data preprocessing, rainfall threshold determination, urban flooding evolution simulation and prediction, key location identification and risk assessment, and results output and early warning. The steps are connected by arrowed lines, clearly indicating their sequence. For the rainfall threshold determination step… Figure 2 The code is marked with two branches: "Yes" and "No". The "No" branch will return to the multi-source data collection step, while the "Yes" branch will enter the urban flooding evolution simulation and prediction step, thus fully presenting the logical relationship between dynamic monitoring and inference.

[0030] The data cleaning unit is connected to the data standardization unit, the data standardization unit is connected to the data fusion unit, and the data fusion unit is connected to the data quality verification unit. The data cleaning unit is used to remove outliers and invalid values ​​from the original data. Data standardization units are used to unify the format and units of data from different sources; The data fusion unit is used to construct the spatiotemporal correlation between multi-source data and form a structured input data matrix; The data quality verification unit is used to verify the preprocessed data.

[0031] The large-scale urban flooding forecasting model module is used to dynamically extrapolate and predict the evolution of urban flooding based on input data. This module includes a model training unit, a real-time extrapolation unit, and a model adaptation unit. Based on preprocessed multi-source data, the module constructs a large-scale urban flooding forecasting model with a "Transformer architecture + hydrodynamic constraints." Figure 3 The model's input layer is a 64×64-dimensional structured data matrix, encompassing topographic feature vectors, real-time hydrological vectors, weather forecast vectors, and historical similar event vectors. The hidden layer employs eight attention heads and utilizes residual connections and layer normalization techniques to enhance model training stability. The output layer outputs the core parameters of urban flooding evolution via a fully connected layer. During model training, multi-source data corresponding to historical rainfall events are used as input samples, selecting 120 typical historical rainfall events, while historical urban flooding evolution data are used as label samples. Next, the AdamW optimizer is employed, with a learning rate set to 1e-4 and 150 training iterations. An early stopping strategy is used to prevent overfitting, ultimately converging the model's loss function to below 0.015. When the monitoring system detects rainfall intensity reaching a preset threshold (which can be dynamically adjusted based on the rainfall characteristics of different cities, defaulting to ≥20mm of 1-hour rainfall or ≥50mm of 24-hour rainfall), the model automatically initiates the inference process. During this process, the model will dynamically simulate the entire process of the generation, spread and receding of urban flooding based on real-time hydrological data and the latest meteorological forecast data, and output the time of occurrence of urban flooding, the boundary coordinates of the flooded area (using the WGS84 coordinate system), and the water depth change trend curve data at a time node of 10 minutes.

[0032] The model training unit is connected to the real-time inference unit, and the real-time inference unit is connected to the model adaptation unit; The model training unit is used to optimize and train model parameters using historical data. The real-time simulation unit is used to initiate the simulation of urban flooding evolution when the rainfall intensity reaches a preset threshold. The model adaptive unit is used to dynamically adjust model parameters based on real-time data and prediction feedback, breaking through the limitations of traditional static hydrological models. By using a large deep learning model and combining hydrodynamic principles, it can achieve dynamic simulation and accurate prediction of the waterlogging evolution process. It can adapt to changes in different rainfall intensities and topographic features, significantly improving the prediction accuracy of waterlogging occurrence time, inundation range, and water depth changes.

[0033] The risk assessment module is used to identify key locations and assess their risk levels based on prediction results. The module includes a key location identification unit, an assessment indicator calculation unit, and a risk level determination unit. In identifying key locations and assessing risk levels, the module uses GIS spatial analysis technology to overlay real-time flooding prediction results with a spatial database of key urban locations, automatically marking key locations potentially affected by flooding. These key locations include underpasses, underground parking garages, low-lying areas, residential communities, schools, hospitals, transportation hubs, and important enterprises—areas with high population density or critical infrastructure. A risk assessment indicator system combining objective indicators and subjective weights is constructed. Objective indicators include inundation depth, duration, and water accumulation rate; subjective weights are determined based on the analytic hierarchy process (AHP), specifically: inundation depth weight 0.4, duration weight 0.3, location importance weight 0.2, and historical disaster severity weight 0.1. A combination of large-scale model scoring and threshold determination is used to classify key locations into risk levels. The specific classification criteria are as follows: high risk level (score ≥ 80 points), medium risk level (score ≤ 60 points < 80 points), and low risk level (score < 60 points). After the classification is completed, a risk assessment report is generated that includes the location name, location coordinates, risk level, expected disaster situation, and recommended response measures.

[0034] The key location identification unit is connected to the assessment indicator calculation unit, which in turn is connected to the risk level determination unit. Utilizing spatial overlay analysis and a large-scale model scoring mechanism, key locations such as underpasses and low-lying residential areas are automatically identified, constructing a multi-dimensional risk assessment indicator system to accurately classify risk levels. This approach solves the problem of traditional technologies struggling to specifically identify high-risk areas, enhancing the targeting and effectiveness of emergency response.

[0035] The key location identification unit is used to automatically identify the preset key locations affected within the predicted flooding range through spatial overlay analysis; The evaluation index calculation unit is used to calculate the values ​​and weights of each index in the multi-dimensional index system. The risk level determination unit is used to output three risk levels: high, medium, and low, based on the comprehensive assessment results.

[0036] The results output and early warning module is used to visualize assessment results, generate early warning information, and push it to the emergency management platform. This module employs multiple output methods, including visual charts, text reports, and API push notifications. Visual outputs include dynamic simulation maps of flooding evolution, heat maps of inundation areas, risk level distribution maps of key locations, and line graphs showing water depth changes. Text reports include summaries of flooding predictions, risk level statistics, lists of key prevention areas, and emergency response recommendations. Synchronization to the city's emergency management platform is achieved via a RESTful API interface, ensuring a response time of ≤5 seconds to meet the high efficiency requirements of urban emergency response. Furthermore, this interface supports pushing early warning information to relevant departments such as emergency management, transportation, water resources, and communities, including SMS alerts, app push notifications, and platform pop-ups, providing accurate and timely data support for subsequent response measures such as emergency dispatch, personnel evacuation, drainage facility scheduling, and resource allocation.

[0037] The results output and early warning module includes a visualization unit, a platform integration unit, and an early warning push unit; The visualization unit is used to generate flood forecast charts and risk level distribution maps; The platform interface unit is configured to synchronize data with the city's emergency management system; The early warning push unit is used to push early warning information to relevant management departments; it adopts an edge computing architecture to preprocess monitoring data locally, reduce data transmission latency, and ensure the real-time nature of urban flooding prediction; at the same time, it uses cloud-based large models to complete complex simulations and risk assessments, balancing real-time performance and computational accuracy.

[0038] The all-weather three-dimensional monitoring network of this invention adopts an edge computing architecture, which completes the preprocessing and preliminary analysis of data locally at the sensor nodes, and only uploads key data and abnormal data to the cloud platform, thereby reducing data transmission latency and bandwidth consumption and ensuring real-time performance. Example

[0039] The difference between this embodiment and Embodiment 1 is that: a system for analyzing and predicting urban waterlogging points in a prefecture-level city is selected for construction. Step 1: Monitoring Network Deployment: Ultrasonic water level sensors and rainfall sensors are deployed at 120 key nodes of the city's stormwater drainage network, including intersections of main road networks, outlets of low-lying areas, and networks surrounding old residential areas. The sensor sampling frequency is set to once every 5 minutes. At the same time, high-precision weather forecast data from the Municipal Meteorological Bureau is accessed, with an update frequency of once every hour. High-precision topographic elevation data of 1:5000 and historical waterlogging data from the past 10 years are obtained, covering the location of waterlogging points, inundation depth, and duration of 86 rainfall-related waterlogging events.

[0040] Step 2: Data preprocessing. A data cleaning script is written using Python to remove abnormal water level data caused by sensor malfunctions, such as data exceeding the normal range and extreme outliers in meteorological data; data of different dimensions such as rainfall (mm), water level (m), and elevation (m) are standardized to the [0, 1] interval; with the help of a spatiotemporal correlation algorithm, hydrological data, meteorological data, topographic data and historical data are integrated using "timestamp + geographic coordinates" as the correlation key to form a structured data matrix.

[0041] Step 3: Construction of a large-scale urban flood forecast model: The Transformer-Base architecture is adopted. The input layer is the fused multi-source data matrix (128×128 dimension), the hidden layer is set with 6 attention heads, and the output layer is the urban flood evolution parameters, including the time of water accumulation, the coordinates of the inundation range, and the water depth change curve. Historical urban flood data from the past 10 years are used as training samples, with 70% used as the training set and 30% as the test set. The Adam optimizer is used to train the model, and the number of training iterations is set to 100 rounds. The model loss function converges to below 0.02.

[0042] Step 4: Threshold Setting and Model Launch: A rainfall intensity threshold of 25 mm or more per hour is set. When rainfall in a certain area reaches this threshold, the model automatically starts simulation. For example, according to the 2025 Urban Flooding Risk Assessment Report, a certain area in the city recorded 28 mm of rainfall in one hour on July 15th. Using an advanced flooding prediction model, combined with real-time water level data, topographic data, and subsequent rainfall trends predicted by weather forecasts, the model quickly simulated the flooding process. The prediction results indicate that flooding will occur in the area within 35 minutes, affecting two densely populated older residential communities and an underpass serving as a key traffic node, with a maximum expected inundation depth of 1.2 meters.

[0043] Step 5: Key Location Identification and Risk Assessment: First, using spatial overlay analysis, the two densely populated old residential areas and the underpass, which is a key traffic node, are automatically identified as key locations. Then, based on the risk assessment index system, where the weight of inundation depth is set to 0.4, the weight of duration is set to 0.3, and the weight of location importance is set to 0.3, the risk level of the underpass is calculated to be high, and the risk level of the two old residential areas is medium, and a risk assessment report is generated.

[0044] Step 6: Results Output and Early Warning: On the one hand, the prediction results, key location markers, and risk levels are displayed through a visual interface and synchronized to the municipal emergency management platform; on the other hand, based on the early warning information, the emergency department organizes traffic control in the underpass, evacuates people from old residential areas, and dispatches surrounding drainage pumping stations to start drainage operations in advance, thereby effectively reducing the losses caused by urban flooding.

[0045] Through practical verification, this study demonstrated that by constructing a multi-factor linked dataset and applying the LightGBM algorithm, the prediction error for the occurrence time of urban flooding was only 8 minutes, the prediction accuracy for the flooded area was as high as 85%, and the accuracy rate for risk level assessment was as high as 90%. This significantly improved the early warning capability for urban flooding, provided a valuable 35-minute time window for emergency response, and effectively enhanced the city's disaster prevention and mitigation capabilities.

[0046] In summary, by integrating multi-source big data and dynamic simulation using large models, and leveraging an advanced urban flooding early warning and monitoring system, combined with big data analysis and machine learning technologies, the prediction error of the flooding occurrence time can be controlled within 10 minutes, while the prediction accuracy of the inundation range is improved by more than 40%. This provides city managers and emergency response teams with accurate time and space references, greatly improving the efficiency of responding to flooding disasters.

[0047] This invention relies on edge computing and an all-weather three-dimensional monitoring network. The data acquisition and preprocessing delay is no more than 30 seconds. The model starts simulation immediately after the rainfall reaches the standard, which can predict the risk of urban flooding 30-120 minutes in advance, and gain valuable time window for emergency response.

[0048] This invention automatically identifies key affected areas and classifies them into risk levels, enabling emergency response to focus on high-risk areas, improving the targeting and efficiency of disaster prevention and mitigation measures, and reducing disaster losses.

[0049] This invention is applicable to cities of different sizes and with different terrain features. It can be seamlessly integrated with existing urban emergency management platforms without requiring large-scale modifications to existing infrastructure, and has good prospects for widespread application.

[0050] Those skilled in the art will understand that the features described in the various embodiments and / or claims of the present invention can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments and / or claims of the present invention can be combined or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

Claims

1. An algorithm for analyzing and predicting urban waterlogging points based on large-scale models and big data technology, characterized in that, Includes the following steps: S1. Construct an all-weather, three-dimensional monitoring network to collect multi-source heterogeneous data, including hydrological data of key nodes in the rainwater pipe network, real-time weather forecast data, topographic data, and historical waterlogging data. S2. Clean, standardize, and fuse multi-source heterogeneous data to form a structured input data matrix; S3. Construct a large-scale urban flooding forecast model. When the monitored rainfall intensity reaches the preset threshold, use the large-scale urban flooding forecast model to dynamically simulate the urban flooding evolution process based on the input data matrix, and predict the occurrence time, inundation range and water depth change trend of the urban flooding event. S4. Based on the inundation range, the system automatically identifies the preset key locations affected by the inundation through spatial overlay analysis, and assesses the risk level of each key location based on a multi-dimensional indicator system that includes inundation depth, duration, location importance, and historical disaster severity. S5 outputs a report containing the prediction results and risk assessment results, and synchronizes it to the emergency management platform.

2. The urban waterlogging point analysis and prediction algorithm based on large model and big data technology according to claim 1, characterized in that, In S1, the all-weather three-dimensional monitoring network adopts an edge computing architecture, with edge computing devices deployed locally on the sensor nodes used to complete local data preprocessing.

3. The urban waterlogging point analysis and prediction algorithm based on large model and big data technology according to claim 1, characterized in that, In S3, the large-scale urban flooding forecast model is constructed using a deep neural network based on the Transformer architecture and combined with hydrodynamic principles. The model is trained using historical multi-source heterogeneous data and strengthens the weight allocation of core influencing factors through its self-attention mechanism.

4. The urban waterlogging point analysis and prediction algorithm based on large model and big data technology according to claim 1, characterized in that, In S4, the preset key locations include, but are not limited to, underpasses, low-lying areas, and residential areas; the risk levels are divided into three levels: high, medium, and low.

5. A system for analyzing and predicting urban waterlogging points based on large model and big data technology according to any one of claims 1-4, characterized in that, It includes a multi-source data acquisition module, a data preprocessing module, a large-scale urban flood forecasting model module, a risk assessment module, and a results output and early warning module; The multi-source data acquisition module is connected to the data preprocessing module, the data preprocessing module is connected to the urban flooding forecasting large model module, the urban flooding forecasting large model module is connected to the risk assessment module, and the risk assessment module is connected to the result output and early warning module. The multi-source data acquisition module is used to acquire and aggregate multi-source heterogeneous data in real time; The data preprocessing module is used to clean, standardize, and fuse multi-source heterogeneous data to generate high-quality input data. The large-scale urban flood forecasting module is used to dynamically extrapolate and predict the evolution of urban flooding based on input data. The risk assessment module is used to identify key locations and assess their risk levels based on the prediction results; The result output and early warning module is used to visualize the assessment results, generate early warning information, and push it to the emergency management platform.

6. The system for analyzing and predicting urban waterlogging points based on large model and big data technology according to claim 5, characterized in that, The multi-source data acquisition module includes a hydrological sensing unit, a meteorological data access unit, a topographic data integration unit, and a historical data management unit; The hydrological sensing unit is used to sense and collect data on the hydrological conditions of key areas. The meteorological data access unit is configured to connect to the API interface of the meteorological department to obtain real-time weather forecast data; The terrain data integration unit is used to load and process high-precision terrain elevation datasets; The historical data management unit is used to store and manage historical flooding event data.

7. The system for analyzing and predicting urban waterlogging points based on large model and big data technology according to claim 5, characterized in that, The data preprocessing module includes a data cleaning unit, a data standardization unit, a data fusion unit, and a data quality verification unit; The data cleaning unit is connected to the data standardization unit, the data standardization unit is connected to the data fusion unit, and the data fusion unit is connected to the data quality verification unit. The data cleaning unit is used to remove outliers and invalid values ​​from the original data; The data standardization unit is used to unify the format and dimensions of data from different sources; The data fusion unit is used to construct the spatiotemporal correlation between multi-source data to form a structured input data matrix; The data quality verification unit is used to verify the preprocessed data.

8. The system for analyzing and predicting urban waterlogging points based on large model and big data technology according to claim 5, characterized in that, The large-scale model module for urban flooding forecasting includes a model training unit, a real-time simulation unit, and a model adaptation unit. The model training unit is connected to the real-time inference unit, and the real-time inference unit is connected to the model adaptation unit; The model training unit is used to optimize and train model parameters using historical data. The real-time simulation unit is used to initiate the simulation of urban flooding evolution when the rainfall intensity reaches a preset threshold. The model adaptive unit is used to dynamically adjust model parameters based on real-time data and prediction feedback.

9. The system for analyzing and predicting urban waterlogging points based on large model and big data technology according to claim 5, characterized in that, The risk assessment module includes a key location identification unit, an assessment indicator calculation unit, and a risk level determination unit. The key location identification unit is connected to the evaluation index calculation unit, and the evaluation index calculation unit is connected to the risk level determination unit. The key location identification unit is used to automatically identify the affected preset key locations within the predicted flooding range through spatial overlay analysis. The evaluation index calculation unit is used to calculate the value and weight of each index in the multi-dimensional index system. The risk level determination unit is used to output three risk levels: high, medium, and low, based on the comprehensive assessment results.

10. The system for analyzing and predicting urban waterlogging points based on large model and big data technology according to claim 5, characterized in that, The result output and early warning module includes a visualization display unit, a platform integration unit, and an early warning push unit; The visualization unit is used to generate flood forecast charts and risk level distribution maps; The platform interface unit is configured to synchronize data with the urban emergency management system; The early warning push unit is used to push early warning information to relevant management departments.

Images