Urban waterlogging disaster monitoring and early warning method and system

Abstract

The application discloses a kind of urban waterlogging disaster situation monitoring and early warning method and system, wherein the method includes: obtaining initial signal from multiple types of sensors, and generating stable acquisition data by adaptive processing.According to stable acquisition data, high-precision multidimensional data set is generated by fusing multi-source environmental information.If high-precision multidimensional data set exceeds the preset threshold, then generate risk distribution map by extracting spatial features through neural network.Based on the amount of regional data corresponding to the risk distribution map, the data transmission path of the region is optimized through the communication network to generate a low-latency transmission channel.Real-time data stream is obtained through the low-latency transmission channel, and real-time data stream is processed based on the spatial features of the risk distribution map to generate a disaster trend distribution.Based on the disaster trend distribution, a real-time situation map layer is generated by visual mapping.The application can improve the stability of monitoring data, enhance the accuracy of disaster prediction, and optimize the efficiency of early warning response.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of urban environmental monitoring and disaster early warning technology, and in particular to a method and system for monitoring and early warning of urban flooding disasters. Background Technology

[0002] Urban flood control, as a core issue in urban management and disaster prevention and mitigation, is directly related to the safety of residents' lives and property and the normal operation of urban functions; its criticality is self-evident. With the intensification of climate change and the acceleration of urbanization, the frequency and intensity of urban flooding disasters pose a severe test to existing technological systems, urgently requiring efficient and intelligent solutions to enhance urban resilience.

[0003] Currently, many urban flooding monitoring and early warning methods rely on single sensors or traditional hydrological models, resulting in limited data collection range, insufficient real-time performance, and inefficient transmission of early warning information due to a lack of precise location. These limitations make it difficult for city managers to accurately predict risk areas before disasters occur, to promptly grasp the dynamic disaster situation during disasters, let alone quickly restore normal order after disasters. Despite this, the field still faces several core challenges, particularly in overcoming technological bottlenecks in sensor technology, data transmission efficiency, and intelligent analysis capabilities.

[0004] First, sensor technology is susceptible to interference in complex urban environments, making it difficult to continuously and stably collect multi-dimensional data. Second, wireless communication technology often experiences delays or packet loss in high-density urban areas, affecting real-time performance. Finally, while the integration of big data analytics and hydrological and hydrodynamic models shows potential, the accuracy and speed of existing algorithms in processing dynamic multi-source data still need improvement. These unresolved technical factors result in inaccurate predictions of urban flooding risk points, inaccurate location of warning areas, and a failure to fully reflect the actual situation in real-time disaster updates, creating a series of pressing technical challenges that need to be overcome.

[0005] Therefore, how to achieve efficient deployment and stable operation of sensor technology in urban flood control, improve the reliability of wireless communication data transmission, and optimize the real-time fusion of big data analysis and hydrological models to ensure accurate risk prediction before disasters, dynamic monitoring of disaster situations during disasters, and rapid lifting of warnings after disasters have become key issues that urgently need to be addressed. Summary of the Invention

[0006] This invention provides a method and system for monitoring and early warning of urban flooding disasters, which can improve the stability of monitoring data, enhance the accuracy of disaster prediction, and optimize the efficiency of early warning response.

[0007] To achieve the above objectives, in a first aspect, the present invention provides a method for monitoring and early warning of urban flooding disasters, comprising: acquiring initial signals from multiple types of sensors and generating stable collected data through adaptive processing; generating a high-precision multidimensional dataset by fusing multi-source environmental information based on the stable collected data; if the high-precision multidimensional dataset exceeds a preset threshold, extracting spatial features through a neural network to generate a risk distribution map; optimizing the data transmission path of the region corresponding to the risk distribution map through a communication network to generate a low-latency transmission channel based on the amount of regional data corresponding to the risk distribution map; acquiring real-time monitoring data streams from multiple types of sensors through the low-latency transmission channel and processing the real-time data streams based on the spatial features of the risk distribution map to generate a disaster trend distribution; generating a real-time situation layer through visualization mapping based on the disaster trend distribution; simulating the disaster evolution path to generate a pre-disaster warning range based on the real-time situation layer; optimizing risk prediction through distributed computing based on the pre-disaster warning range and the real-time situation layer to generate prediction results; and analyzing the disaster receding rate to generate a recovery time window based on the prediction results.

[0008] Secondly, this invention provides an urban flooding disaster monitoring and early warning system, comprising: a data acquisition module, a multidimensional dataset generation module, a distribution map generation module, a transmission channel generation module, a trend distribution generation module, a situation layer generation module, an early warning range generation module, a prediction result generation module, and a time window generation module. The data acquisition module acquires initial signals from multiple types of sensors and generates stable acquisition data through adaptive processing. The multidimensional dataset generation module generates a high-precision multidimensional dataset by fusing multi-source environmental information based on the stable acquisition data. The distribution map generation module extracts spatial features through a neural network to generate a risk distribution map if the high-precision multidimensional dataset exceeds a preset threshold. The transmission channel generation module optimizes the data transmission path of the area corresponding to the risk distribution map through a communication network to generate a low-latency transmission channel. The trend distribution generation module acquires real-time monitoring data streams from multiple types of sensors through the low-latency transmission channel and processes the real-time data streams based on the spatial features of the risk distribution map to generate a disaster trend distribution. The situation layer generation module generates a real-time situation layer through visualization mapping based on the disaster trend distribution. The module for generating the early warning range is used to generate a pre-disaster early warning range based on the real-time situation layer and by simulating the disaster evolution path. The module for generating prediction results is used to generate prediction results by optimizing risk prediction through distributed computing based on the pre-disaster early warning range and the real-time situation layer. The module for generating a time window is used to generate a recovery time window based on the prediction results and by analyzing the disaster's rate of decline.

[0009] Thirdly, the present invention provides an electronic device, comprising:

[0010] At least one processor; and

[0011] A memory that is communicatively connected to the at least one processor;

[0012] The memory stores instructions that can be executed by the at least one processor, which are then executed by the at least one processor to enable the at least one processor to perform the urban flooding disaster monitoring and early warning method as described above.

[0013] Fourthly, the present invention provides a computer-readable storage medium including a computer program and instructions, which, when the computer program or the instructions are run on a computer, cause the computer to perform the urban flooding disaster monitoring and early warning method as described above.

[0014] Compared with existing technologies, the urban flooding disaster monitoring and early warning method and system according to the present invention achieves comprehensive monitoring and accurate early warning of urban flooding disasters. Through the fusion and adaptive processing of multi-source data, the stability and reliability of data acquisition are improved. The establishment of a low-latency transmission channel ensures the real-time transmission of key data, providing timely support for decision-making. The application of neural networks and distributed computing enhances the accuracy of risk prediction, making early warning more precise. Real-time situational mapping and disaster evolution simulation provide intuitive decision-making basis for emergency response. Disaster recession rate analysis provides scientific guidance for post-disaster recovery work. The comprehensive application of these technologies effectively improves the city's ability to cope with urban flooding disasters, reduces potential loss of life and property, and provides strong support for urban resilience building. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a method for monitoring and early warning of urban flooding disasters according to Embodiment 1 of the present invention;

[0016] Figure 2 This is a schematic diagram of the structure of an urban flooding disaster monitoring and early warning system according to Embodiment 2 of the present invention;

[0017] Figure 3 This is a schematic diagram of the structure of an electronic device according to Embodiment 3 of the present invention. Detailed Implementation

[0018] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the embodiments of the present invention, and not all structures.

[0019] To facilitate understanding, the main implementation concepts of the various embodiments of the present invention will be briefly described first.

[0020] In traditional urban flood control systems, various types of sensors suffer from signal distortion due to electromagnetic interference and physical obstruction, making it difficult to guarantee the spatiotemporal continuity of collected data. Wireless communication networks experience link congestion in high-density building areas due to multipath effects, causing delays in critical disaster data transmission. During dynamic multi-source heterogeneous data fusion, traditional interpolation algorithms cannot effectively eliminate outliers in the spatiotemporal dimension, leading to a decline in the quality of input data for risk prediction models. For example, a monitoring system deployed in a megacity during the flood season includes 200 pressure sensors distributed in underground pipe networks and 50 rain gauges on the ground. When rainfall reaches 50 mm / h, the underground sensors generate impulse noise due to turbulent impact, causing the signal acquisition module to misclassify 12% of node data as outliers. Simultaneously, ground base stations experience an average transmission delay of 380 ms due to 4G network channel contention, resulting in spatiotemporal misalignment of the real-time data stream received by the central processing system. When processing noisy time-series data, the traditional LSTM model's gating mechanism cannot effectively separate high-frequency interference components, ultimately resulting in a flood evolution path prediction error exceeding the actual value by 23%. If the above problems are not addressed, the city's emergency command center will be unable to accurately identify flooding risk points in the early stages of a disaster, leading to delayed evacuation orders and missed golden response windows; the dynamic disaster assessment system may misjudge risk levels due to data quality defects, causing the allocation of rescue resources to deviate from actual needs; and the lack of accurate recession rate analysis during the post-disaster recovery phase will prolong the cycle of traffic control and power restoration, causing secondary economic losses.

[0021] Faced with the aforementioned problems, this invention first considers how to fundamentally solve the data quality issues caused by sensor signal distortion and communication delays. Regarding impulse noise caused by environmental interference in sensors, traditional filtering methods can only eliminate noise in fixed frequency bands, but cannot cope with dynamic noise patterns under turbulent impacts. To address this, this invention proposes using a pre-trained model to dynamically identify signal types and matching adaptive filters according to signal characteristics, improving the signal-to-noise ratio while preserving key features. Regarding the multipath effect of wireless networks in high-density areas, existing routing algorithms struggle to achieve a dynamic balance between latency and bandwidth, leading to delays in disaster data transmission. This invention generates a candidate path set by real-time monitoring of network performance indicators and combining it with a dynamic routing algorithm, and dynamically allocates priorities based on weighted scoring to achieve load balancing and low-latency transmission. Regarding prediction errors caused by spatiotemporal misalignment of input data in traditional LSTM models, existing technologies lack a spatiotemporal alignment mechanism for multi-source heterogeneous data. This invention generates a high-precision multidimensional dataset as model input by fusing standardized processing of multi-source environmental information, combining spatiotemporal feature extraction and interpolation correction, and introduces a channel attention mechanism to enhance spatial feature weights, thereby improving the accuracy of risk prediction.

[0022] Example 1, Figure 1 This is a flowchart illustrating an urban flooding disaster monitoring and early warning method according to Embodiment 1 of the present invention. Embodiment 1 provides an urban flooding disaster monitoring and early warning method, including: Step S100, acquiring initial signals from multiple types of sensors and generating stable collected data through adaptive processing; Step S200, generating a high-precision multidimensional dataset by fusing multi-source environmental information based on the stable collected data; Step S300, if the high-precision multidimensional dataset exceeds a preset threshold, extracting spatial features through a neural network to generate a risk distribution map; Step S400, optimizing the data transmission path of the region corresponding to the risk distribution map through a communication network to generate a low-latency... The transmission channel includes the following steps: Step S500: Real-time monitoring data streams from multiple types of sensors are acquired through the low-latency transmission channel, and the real-time data streams are processed based on the spatial characteristics of the risk distribution map to generate a disaster trend distribution; Step S600: Based on the disaster trend distribution, a real-time situation layer is generated through visualization mapping; Step S700: Based on the real-time situation layer, the disaster evolution path is simulated to generate a pre-disaster warning range; Step S800: Based on the pre-disaster warning range and the real-time situation layer, risk prediction is optimized through distributed computing to generate prediction results; Step S900: Based on the prediction results, the disaster recession rate is analyzed to generate a recovery time window. The real-time data stream includes dynamic environmental data collected in real-time by sensors such as water level, rainfall, and flow velocity through the low-latency transmission channel.

[0023] Specifically, adaptive processing to generate stable acquired data refers to the process of noise filtering, feature extraction, and data compression of the raw signals acquired by sensors. This can be achieved using pre-trained models to classify signal types, adaptive filters to improve the signal-to-noise ratio, time-frequency feature correlation analysis, and coding compression methods. This addresses the issue of sensor data being easily interfered with in complex urban environments, ensuring the stability of data acquisition. Fusion of multi-source environmental information to generate high-precision multidimensional datasets involves integrating environmental data from different sources into a unified dataset through standardization, dynamic weighting, and anomaly correction. This can be achieved using sensor weight dynamic fusion algorithms, Mahalanobis distance anomaly interpolation correction, and random forest missing value detection methods. This eliminates errors caused by the heterogeneity of multi-source data and improves data fusion accuracy. Neural network extraction of spatial features to generate risk distribution maps involves using deep learning models to perform spatial topology modeling on multidimensional environmental data. This can be achieved using U-Net convolutional neural networks to extract three-dimensional feature tensors, channel attention weighting, and DBSCAN clustering algorithms to delineate risk regions. This addresses the problem that traditional hydrological models cannot capture dynamic spatial relationships, achieving accurate location of risk regions. Optimizing data transmission paths in communication network areas to generate low-latency transmission channels refers to dynamically adjusting data transmission strategies based on network performance indicators. Specifically, this can be achieved using dynamic routing algorithms to generate candidate paths, weighted scoring to select the optimal path, and load balancing to distribute traffic. This addresses communication latency or packet loss issues in high-density urban areas, improving the real-time transmission efficiency of disaster data. Processing real-time data streams based on risk distribution maps to generate disaster trend distribution involves dividing real-time data by risk areas and extracting spatiotemporal variation characteristics. This can be achieved using spatial partitioning data cleaning, dynamic feature sequence analysis, and period-trend decomposition methods. This identifies disaster events and predicts their evolution trends, solving the problem of traditional methods being unable to dynamically track disasters. Visualizing and mapping to generate real-time situational layers involves converting disaster data into interactive geospatial information. This can be achieved using spatiotemporal correlation data projection, distribution density threshold calculation, and dynamic layer updates. This visually displays disaster dynamics, assisting decision-makers in quickly grasping the overall situation. Simulating disaster evolution paths to generate pre-disaster warning ranges involves predicting disaster spread paths using hydrological models. This can be achieved using SWAT models to simulate flood evolution, spatiotemporal correlation data mapping, and dynamic boundary expansion methods. This accurately delineates warning areas and allows for advance deployment of disaster prevention resources. Distributed computing optimization for risk prediction generation refers to using a parallel computing framework to improve model computation efficiency. Specifically, it can be achieved by using MapReduce to divide tasks into grids, integrating hybrid prediction models, and using Kriging interpolation space smoothing methods. This is used to solve the problem of insufficient speed in large-scale data analysis and improve the real-time performance and accuracy of prediction results.Analyzing the rate of disaster decline to generate a recovery time window refers to estimating the recovery period based on the trend of changes in disaster intensity. Specifically, it can be achieved by using a decline rate threshold judgment, exponential smoothing denoising, and linear extrapolation prediction methods. This is used to formulate post-disaster recovery plans and optimize resource scheduling priorities.

[0024] This invention achieves precise location of disaster risks through multi-source sensor data fusion and neural network spatial feature extraction, improves the real-time performance of early warning by combining dynamic communication path optimization and distributed computing framework, and forms a complete monitoring-early warning-recovery closed loop based on disaster evolution simulation and recession rate analysis, systematically solving the problems of unstable data collection, delayed early warning and low recovery efficiency in urban flood control.

[0025] In practical applications, initial signals are first acquired from multiple types of sensors, and stable data is generated through adaptive processing. This step effectively eliminates environmental interference and noise by analyzing the characteristics of signals from different types of sensors and employing corresponding adaptive filtering algorithms, ensuring the stability and reliability of the data. Next, based on the stable data, multi-source environmental information is fused to generate a high-precision multidimensional dataset. This process involves data standardization, dynamic fusion, and outlier correction, improving data accuracy and completeness through collaborative analysis of multi-source data. When the high-precision multidimensional dataset exceeds a preset threshold, the system extracts spatial features through a neural network to generate a risk distribution map. The application of neural networks can effectively capture complex spatial relationships, thereby more accurately identifying potential risk areas. Based on the amount of regional data corresponding to the risk distribution map, the system optimizes the data transmission path of the communication network, generating a low-latency transmission channel. This step ensures fast and reliable data transmission in critical areas by dynamically adjusting routing strategies through real-time monitoring of network performance. The system acquires real-time data streams through the low-latency transmission channel and processes these streams based on the spatial features of the risk distribution map to generate a disaster trend distribution. This process combines spatial features and real-time data to more accurately reflect the dynamic changes in the disaster situation. Based on the disaster trend distribution, the system generates a real-time situational layer through visualization mapping. This step transforms complex data into intuitive visual information, facilitating decision-makers' quick understanding of the current situation. Based on the real-time situational layer, the system simulates the disaster's evolution path and generates a pre-disaster warning range. This prediction process considers multiple factors and can identify potentially affected areas in advance. Based on the pre-disaster warning range and the real-time situational layer, the system optimizes risk prediction through distributed computing, generating prediction results. The application of distributed computing improves the efficiency of processing large-scale data, making the prediction results more accurate and timely. Finally, based on the prediction results, the system analyzes the disaster's decline rate and generates a recovery time window. This step provides a scientific basis for post-disaster recovery work, contributing to the rational allocation of resources and the orderly progress of recovery efforts.

[0026] In a preferred embodiment, the specific implementation of the present invention is as follows:

[0027] In urban flood monitoring systems, a multi-sensor network is deployed, including water level sensors, rain gauges, and flow velocity meters. Each sensor collects data every 5 minutes, which is initially processed by edge computing nodes. These edge nodes use pre-trained convolutional neural network models to identify signal types and apply appropriate adaptive filtering algorithms, such as Kalman filters or wavelet transforms, to denoise the signals. The processed data is then transmitted to a central data processing center via 4G or 5G networks. The data center uses a distributed database to store the received data and integrates multi-source data to generate a high-precision multidimensional dataset using data fusion algorithms, such as Bayesian fusion or Dempster-Shafer evidence theory. The system continuously monitors key indicators in the dataset, such as water level, rainfall intensity, and flow velocity. When these indicators exceed preset thresholds, deep learning models, such as U-Net or SegNet, are triggered to extract spatial features and generate a risk distribution map. Based on the risk distribution map, the system uses software-defined networking (SDN) technology to dynamically adjust network routing. By monitoring network performance indicators in real time, such as latency and bandwidth utilization, the system employs multi-path routing algorithms to optimize data transmission paths, ensuring priority transmission of data from high-risk areas. The system receives real-time data streams through an optimized network channel and processes the time-series data using Long Short-Term Memory (LSTM) networks or Gated Recurrent Unit (GRU) models. Combined with spatial information from the risk distribution map, it generates a disaster trend distribution. Using Geographic Information System (GIS) technology, the system maps the disaster trend distribution onto a city map, generating a real-time situation layer. This layer contains information such as risk level and affected area range, and supports dynamic updates. Based on the real-time situation layer, the system runs hydrological models, such as the SWMM (Storm Water Management Model), to simulate flood evolution paths and generate pre-disaster warning areas. The warning area considers factors such as topography and drainage system capacity. The system employs distributed computing frameworks such as Apache Spark, combined with machine learning algorithms such as random forests or gradient boosting trees, to perform refined risk predictions within the warning area. The prediction results include specific parameters such as water depth and flow velocity. Finally, the system uses time-series analysis methods, such as exponential smoothing or ARIMA models, to analyze the disaster recession rate and generate recovery time windows for each affected area. These time windows consider factors such as drainage capacity and surface water storage.

[0028] Based on the above analysis, this invention achieves comprehensive monitoring and precise early warning of urban flooding disasters. Through the fusion and adaptive processing of multi-source data, the stability and reliability of data acquisition are improved. The establishment of low-latency transmission channels ensures real-time transmission of critical data, providing timely support for decision-making. The application of neural networks and distributed computing enhances the accuracy of risk prediction, making early warnings more precise. Real-time situational mapping and disaster evolution simulation provide intuitive decision-making basis for emergency response. Disaster recession rate analysis provides scientific guidance for post-disaster recovery work. The comprehensive application of these technologies effectively improves the city's ability to cope with urban flooding disasters, reduces potential loss of life and property, and provides strong support for urban resilience building.

[0029] In this embodiment, step S100 includes: step S101, acquiring initial signals from multiple types of sensors, determining signal types using a pre-trained model, and generating a classified signal set; step S102, calculating the signal-to-noise ratio (SNR) for the classified signal set using an adaptive filter corresponding to the signal type; if the SNR is lower than a preset threshold, performing filtering processing to generate a denoised signal set; step S103, extracting time-frequency features from the denoised signal set, calculating the correlation coefficient between signal segments; if the correlation coefficient is lower than a preset threshold, using a smoothing method to generate a smoothed signal set; step S104, extracting key data points with significant time-frequency features from the smoothed signal set, compressing them using an encoding method to generate a compressed data set; and step S105, performing an integrity check on the compressed data set; if the check passes, generating the stable acquisition data.

[0030] Specifically, the pre-trained model, trained using historical data from multiple sensors, can identify signal types such as temperature, water level, and flow velocity. Adaptive filters select either Butterworth or Chebyshev filters based on the signal type to reduce high-frequency noise. Time-frequency feature extraction uses wavelet transform to decompose signal segments, and correlation coefficient calculation is based on Pearson coefficients to determine inter-segment consistency. The smoothing method uses a moving average algorithm to correct low-correlation signal segments. The encoding method uses Huffman coding to compress key data points while retaining peak time-frequency features. Integrity verification uses cyclic redundancy check (CRC) codes to verify data integrity. The raw signals from multiple sensors are first classified by the pre-trained model to separate different types of environmental parameter data and avoid signal mixing. After classification, a corresponding filter is matched for each signal type; for example, a low-pass filter is used to suppress electromagnetic interference for water level signals, and a band-pass filter is used to eliminate baseline drift for temperature signals. The filtered signals are then processed using wavelet transform to extract time-frequency features. If the correlation coefficient between adjacent signal segments is below 0.8, it indicates abnormal fluctuations, which are then smoothed using a fifth-order moving average algorithm. The smoothed signal retains data points with significant time-frequency characteristics, such as peak points where wavelet energy exceeds the mean, and compresses them to 30% of the original data volume using Huffman coding. Finally, the verification module checks the CRC32 checksum of the compressed data; if it matches the original data, it outputs stable acquired data. Through classification processing to reduce cross-interference, dynamic filtering to enhance the signal-to-noise ratio, smoothing correction to improve data consistency, compression coding to reduce redundancy, and integrity verification to ensure data reliability, a stable data set that meets the requirements of subsequent processing is ultimately generated.

[0031] In a preferred embodiment, the specific implementation of the present invention is as follows: Initial signals are acquired from multiple types of sensors, and a pre-trained model is used to determine the signal types to generate a classified signal set. Specifically, various environmental monitoring sensors, including water level sensors, rainfall sensors, and flow velocity sensors, can be deployed to collect data in real time via a wireless communication network. A deep learning model is used to classify the acquired raw signals, categorizing different types of signals separately. For the classified signal set, an adaptive filter corresponding to the signal type is used to calculate the signal-to-noise ratio (SNR). If the SNR is lower than a preset threshold, filtering is performed to generate a denoised signal set. For example, a Kalman filter can be used for water level sensor signals, and a median filter can be used for rainfall sensor signals. By calculating the SNR and comparing it with a preset threshold, it is determined whether filtering and noise reduction processing is required. Time-frequency features are extracted from the denoised signal set, and the correlation coefficient between signal segments is calculated. If the correlation coefficient is lower than a preset threshold, a smoothing method is used to generate a smoothed signal set. Specifically, time-frequency features can be extracted using short-time Fourier transform, and the correlation coefficient of signals in adjacent time windows can be calculated. When the correlation coefficient is lower than a set threshold, exponential smoothing is used to smooth the signal. Key data points with significant time-frequency characteristics are extracted from a smooth signal set and compressed using encoding methods to generate a compressed dataset. Feature points in the signal can be identified using wavelet transform, and compression algorithms such as run-length encoding can be used to compress the data, reducing data transmission volume. Integrity checks are performed on the compressed dataset; if the check passes, stable acquisition data is generated. Cyclic Redundancy Check (CRC) and other methods can be used to verify the compressed data, ensuring its integrity and reliability.

[0032] Based on the above analysis, it can be seen that this invention achieves adaptive processing of signals from multiple types of sensors. This improves the stability and reliability of sensor data acquisition in complex urban environments. Furthermore, by classifying, filtering, smoothing, and compressing the raw signals, the data transmission load is reduced, improving the efficiency and accuracy of subsequent data analysis. Specifically, this invention can effectively address the interference problem of different types of sensor signals in urban flood monitoring, ensuring the quality of collected data and providing a reliable data foundation for subsequent disaster analysis and early warning.

[0033] In this embodiment, step S200 includes: step S201, extracting multi-source environmental data from the stable acquisition data and generating a standardized dataset through standardization processing; step S202, generating a preliminary multidimensional dataset from the standardized dataset using a dynamic fusion algorithm based on sensor weights; wherein, the dynamic fusion algorithm dynamically calculates weights based on the measurement variance of each sensor within a sliding time window, with a larger weight for smaller variances, and periodically updates the weights to respond to changes in sensor performance; step S203, extracting spatiotemporal features from the preliminary multidimensional dataset, calculating Mahalanobis distance between features, and correcting outliers by interpolation if the distance exceeds a preset threshold, generating a corrected multidimensional dataset; step S204, checking the distribution of missing values ​​in the corrected multidimensional dataset using a random forest algorithm, and performing t-SNE dimensionality reduction to generate a low-dimensional compressed dataset if the missing rate is less than 5%; step S205, verifying the integrity of the low-dimensional compressed dataset to generate the high-precision multidimensional dataset.

[0034] Specifically, the standardization process employs the Min-Max method to eliminate dimensional differences; the dynamic fusion algorithm dynamically adjusts weights based on the historical accuracy of the sensors; spatiotemporal features are extracted using a sliding window to correlate time series data with geographic coordinates; Mahalanobis distance is used to filter out features with significant statistical distribution differences; interpolation correction uses cubic spline functions to fill in outlier regions; the random forest algorithm identifies missing data patterns by ranking features by importance; t-SNE dimensionality reduction preserves local structural features; and integrity verification compares data consistency using hash checksums. Standardization unifies multi-source environmental data to the same numerical range, eliminating the impact of dimensional differences between different sensors on fusion. The sensor-weighted dynamic fusion algorithm dynamically allocates weights based on historical error rates, with high-precision sensor data playing a dominant role in the fusion process, thereby improving the overall reliability of the dataset. Spatiotemporal feature extraction uses a sliding window to divide the time series, establishes a spatiotemporal correlation matrix based on geographic coordinates, detects statistical distribution differences between features using Mahalanobis distance calculation, and locally corrects outliers exceeding a threshold using cubic spline interpolation to ensure data continuity. The Random Forest algorithm analyzes the distribution pattern of missing values, prioritizing the filling of missing regions with minimal global impact. When the missing value rate is below 5%, t-SNE dimensionality reduction is performed, preserving the local neighborhood structure of high-dimensional data through nonlinear mapping and avoiding information loss. In the integrity verification stage, the MD5 hash algorithm is used to generate data fingerprints, and the consistency of the checksums before and after compression is compared, ultimately outputting a complete and low-redundancy high-precision multidimensional dataset. For example, in missing value processing, when the temperature data missing rate in a certain area is 3%, Random Forest prioritizes imputation based on adjacent sensor data and time trends, and then t-SNE compresses the 128-dimensional features to 3 dimensions, retaining more than 95% of the information. The resulting standardized, low-noise, and complete multidimensional dataset provides high-quality input for subsequent risk feature extraction.

[0035] In a preferred embodiment, the specific implementation of the present invention is as follows: Multi-source environmental data is extracted from stable acquired data, and a standardized dataset is generated through standardization processing. A preliminary multidimensional dataset is generated from the standardized dataset using a dynamic fusion algorithm based on sensor weights. Spatiotemporal features are extracted from the preliminary multidimensional dataset, and Mahalanobis distances between features are calculated. If the distance exceeds a preset threshold, outliers are corrected through interpolation to generate a corrected multidimensional dataset. The distribution of missing values ​​in the corrected multidimensional dataset is checked using a random forest algorithm. If the missing value rate is less than 5%, t-SNE dimensionality reduction is performed to generate a low-dimensional compressed dataset. The low-dimensional compressed dataset is then validated for completeness to generate a high-precision multidimensional dataset.

[0036] Specifically, the process begins by extracting multi-source environmental data, including temperature, humidity, air pressure, and rainfall, from stable data collection. Standardization processes unify data of different dimensions to the same scale, generating a standardized dataset. Further, a dynamic fusion algorithm based on sensor weights is applied to this standardized dataset, assigning weights according to the accuracy and reliability of each sensor to generate a preliminary multidimensional dataset. Features in both time and space dimensions are extracted from this preliminary multidimensional dataset, and Mahalanobis distances between features are calculated. If the distance exceeds a preset threshold, such as three standard deviations, outliers are corrected using interpolation, generating a corrected multidimensional dataset. A random forest algorithm is then used to check the distribution of missing values ​​in the corrected multidimensional dataset. If the missing value rate is below 5%, a t-SNE dimensionality reduction algorithm is executed to map the high-dimensional data to a two- or three-dimensional space, generating a low-dimensional compressed dataset. Finally, the low-dimensional compressed dataset undergoes integrity verification to check the consistency and validity of the data. After verification, a high-precision multidimensional dataset is generated as the basis for subsequent analysis.

[0037] Based on the above analysis, this invention achieves efficient fusion and processing of multi-source environmental data. Through standardization and dynamic weight fusion, data consistency and comparability are improved. Mahalanobis distance and interpolation correction are used to effectively identify and process outlier data points. Combining random forest algorithm and t-SNE dimensionality reduction, data dimensionality is reduced while retaining key information. The resulting high-precision multidimensional dataset provides a reliable data foundation for subsequent risk analysis and early warning, improving the accuracy and real-time performance of urban flooding monitoring.

[0038] In this embodiment, step S300 includes: Step S301, when the environmental parameters in the high-precision multidimensional dataset exceed the corresponding threshold, using a U-Net convolutional neural network to extract spatial topological features and generate a three-dimensional feature tensor. Step S302, applying channel attention weighting to the three-dimensional feature tensor, and using principal component analysis to reduce the dimensionality of the weighted three-dimensional features to a two-dimensional geographic plane, generating a geographic projection feature map. Step S303, based on the spatial resolution of the geographic projection feature map, using a neighborhood radius of 0.5 km, using the DBSCAN clustering algorithm to divide risk areas and generate an initial risk heatmap. Step S304, using the Kriging interpolation algorithm to generate a continuous density distribution surface from the discrete risk values ​​of the initial risk heatmap, and calculating the gradient distribution of the surface. Step S305, based on the gradient distribution of the continuous density distribution surface, dividing risk level thresholds and generating a risk distribution map with graded labels.

[0039] Specifically, the U-Net convolutional neural network preserves shallow spatial details through a skip connection structure and generates a 3D feature tensor by combining deep semantic features. The encoder part uses four layers of convolutional downsampling, and the decoder part restores resolution through deconvolution. The channel attention weighting module obtains channel feature weights through global average pooling and enhances key feature channels, for example, amplifying the features of channels with weights higher than 0.8 by a factor of 1.5. Principal component analysis projects the 3D feature tensor along the height direction, retaining 2D features whose cumulative contribution rate of the first two principal components exceeds 85%. The DBSCAN clustering algorithm sets the neighborhood radius to 0.5km and the minimum number of samples to 10 to adapt to the scale of urban blocks. Kriging interpolation uses a spherical model to fit the semi-variogram function, with a search radius of 2km, generating a continuous density surface with a raster resolution of 50m. Gradient distribution calculation uses the Sobel operator to extract the derivatives of the surface in the longitude and latitude directions, and classifies three risk levels—high, medium, and low—based on the absolute value of the slope, with areas with a slope exceeding 0.3 defined as high risk.

[0040] Specifically, when environmental parameters exceed thresholds, the U-Net network first extracts spatial topological features, including terrain undulations and pipeline distribution, from high-precision multidimensional data, forming a 256×256×64 three-dimensional feature tensor. The channel attention weighting module enhances the channel response in rainfall-sensitive areas through adaptive learning; for example, it increases the channel weight of groundwater level monitoring data by 1.2 times and reduces the weight of temperature sensors by 0.7 times. After principal component analysis, the two-dimensional geographic projection feature map retains 97% of the original information, and its spatial resolution is aligned with the city's GIS base map. DBSCAN clustering with a neighborhood radius of 0.5 km can identify isolated waterlogging points and contiguous flooded areas at the urban block level; for example, it identifies 7 independent waterlogging points with a diameter less than 300 m and 3 contiguous areas with an area exceeding 1 km². Kriging interpolation converts the discrete clustering results into a continuous surface, and the gradient distribution is calculated to identify the direction of flood spread; for example, a slope of 0.35 is detected on the eastern side of the surface, triggering a high-risk indicator. The final risk distribution map divides the city into three zones: red, orange, and yellow, corresponding to warning levels for water depths exceeding 0.5m, 0.3m-0.5m, and 0.1m-0.3m, respectively, achieving a precise mapping between risk levels and geographic space.

[0041] In this embodiment, the preset threshold can be flexibly configured according to the actual needs of urban flood control. As an option, the system collects historical water level and rainfall data from the past three flood seasons, calculates the distribution of environmental parameters at each monitoring point under disaster-free conditions, and uses the mean plus three times the standard deviation as the initial threshold. Alternatively, the system refers to the water depth warning standards specified in the "Technical Specifications for Urban Flood Control" (e.g., 15cm for a yellow warning, 30cm for an orange warning, and 50cm for a red warning), converting the corresponding depths into sensor measurements as the threshold.

[0042] As a more efficient dynamic approach, the system employs an adaptive threshold algorithm: using monitoring data from the most recent 24 hours as a sliding window, it calculates the mean and standard deviation of each parameter within the window, sets the threshold to the mean plus twice the standard deviation, and updates it every 10 minutes. Simultaneously, the system can access real-time early warning data issued by meteorological departments (such as blue, yellow, orange, and red rainstorm warnings), dynamically adjusting the threshold according to the warning level: for example, when a yellow rainstorm warning is received, the threshold is lowered by 10%; when an orange rainstorm warning is received, it is lowered by 20%; and when a red rainstorm warning is received, it is lowered by 30%, thus triggering the risk monitoring process in advance. Furthermore, meteorological warning data can also serve as an independent trigger condition: even if the sensor's measured data does not exceed the conventional threshold, if the meteorological warning level reaches a preset level (such as orange or above), the system can still proactively initiate the subsequent risk distribution map generation step, achieving predictive early warning based on meteorological conditions.

[0043] When the actual measured value exceeds the current window threshold, or when a meteorological warning that meets the triggering conditions is received, the system triggers the subsequent risk distribution map generation step. Those skilled in the art can select appropriate threshold setting methods or triggering strategies based on factors such as city size, historical disaster frequency, sensor deployment density, and meteorological warning response mechanisms.

[0044] In practical applications, when environmental parameters in a high-precision multidimensional dataset exceed corresponding thresholds, a U-Net convolutional neural network is used to extract spatial topological features and generate a three-dimensional feature tensor. Specifically, the U-Net network consists of an encoder and a decoder. The encoder comprises four convolutional layers, each using a 3x3 kernel with a stride of 1, padding of 1, and ReLU activation. The decoder also contains four deconvolutional layers with 2x2 upsampling. Channel attention weighting is applied to the three-dimensional feature tensor, and principal component analysis (PCA) is used to reduce the dimensionality of the weighted three-dimensional features to a two-dimensional geographic plane, generating a geographic projection feature map. The channel attention mechanism is implemented using 1x1 convolutional layers, and PCA retains the first two principal components. Based on the spatial resolution of the geographic projection feature map, the DBSCAN clustering algorithm is used to divide risk areas with a neighborhood radius of 0.5 km, generating an initial risk heatmap. The MinPts parameter of the DBSCAN algorithm is set to 5. Kriging interpolation is used to generate a continuous density distribution surface from the discrete risk values ​​of the initial risk heatmap, and the gradient distribution of the surface is calculated. Kriging interpolation uses the ordinary kriging method, and the variogram model is a spherical model. Based on the gradient distribution of the continuous density distribution surface, risk level thresholds are defined, generating a risk distribution map with graded labels. Risk levels are divided into low, medium, and high, and the thresholds are determined using K-means clustering.

[0045] Based on the above analysis, this invention achieves accurate spatial feature extraction and risk distribution visualization of urban flooding disasters. The U-Net network effectively captures the spatial topological relationships of environmental parameters, and the channel attention mechanism highlights key features. The DBSCAN clustering algorithm adapts to irregularly shaped risk areas, and the continuous surface generated by Kriging interpolation improves the smoothness and interpretability of the risk distribution. Compared with traditional single-threshold judgment, this method can more comprehensively reflect the spatial distribution characteristics of risks, providing a more reliable basis for subsequent early warning and decision-making.

[0046] In this embodiment, step S400 includes: Step S401, acquiring transmission latency, bandwidth, and packet loss rate parameters from the communication network in real time, and extracting key performance indicators to generate a performance indicator set. Step S402, if the latency in the performance indicator set exceeds a preset threshold, generating a candidate path set based on the current network topology and real-time load status using a dynamic routing algorithm. Step S403, extracting latency, bandwidth, and hop count features from the candidate path set, and generating a preferred path sequence using a weighted scoring algorithm. Step S404, allocating data packet transmission priorities and performing load balancing based on the scoring results of the preferred path sequence to generate a balanced traffic distribution. Step S405, monitoring the real-time congestion status of the balanced traffic distribution; if congestion exists, dynamically switching to backup paths based on the preferred path sequence, reallocating data streams, and generating a low-latency transmission channel.

[0047] Specifically, key performance indicators are extracted by real-time collection of transmission latency, bandwidth, and packet loss rate parameters. A dynamic routing algorithm generates a set of candidate paths based on network topology and real-time load. A weighted scoring algorithm calculates a comprehensive score for candidate paths based on latency, bandwidth, and hop count weighting coefficients, generating a sequence of optimal paths. Balanced traffic distribution is achieved through priority allocation and load balancing; when congestion occurs, data flows are dynamically reallocated by switching to backup paths.

[0048] Specifically, after acquiring transmission latency, bandwidth, and packet loss rate parameters in real time, the communication network extracts them to form a set of performance indicators. If the latency exceeds a preset threshold, a dynamic routing algorithm is used to calculate candidate paths based on the current network topology and real-time load status. The latency, bandwidth, and hop count characteristics of each candidate path are weighted and scored, for example, latency accounts for 40%, bandwidth for 35%, and hop count for 25%, generating a preferred path sequence. According to the scoring results, high-priority data packets are assigned to low-latency paths, and low-priority data packets are assigned to high-bandwidth paths, performing load balancing to generate a balanced traffic distribution. When congestion is detected in the balanced traffic distribution, backup paths are dynamically switched based on the preferred path sequence, for example, switching the data flow from a congested path to a suboptimal path, reallocating the data flow to maintain low-latency transmission. Thus, through dynamic path selection and priority allocation, the risk of network congestion is reduced while ensuring the real-time transmission of critical data, forming a stable low-latency transmission channel.

[0049] It should be noted that the purpose of the low-latency transmission channel established in this step is not to repeatedly transmit the high-precision multidimensional dataset already generated in step S200, but rather to acquire a new round of continuous monitoring data streams from sensors in the high-risk areas identified by the risk distribution map in real time. By allocating dedicated, low-latency transmission paths to high-risk areas, it ensures that the real-time data used in subsequent disaster trend analysis (step S500) can reach the processing center in a timely and complete manner, avoiding prediction delays due to network congestion.

[0050] In practical applications, transmission latency, bandwidth, and packet loss rate parameters are obtained in real time from the communication network, and key performance indicators are extracted to generate a performance indicator set. Specifically, network probing tools periodically send probe packets to each node to measure round-trip time, available bandwidth, and packet loss, generating a performance data table containing these indicators. If the latency in the performance indicator set exceeds a preset threshold, a candidate path set is generated based on the current network topology and real-time load status using a dynamic routing algorithm. For example, when a link latency exceeding 100ms is detected, route recalculation is triggered, and multiple alternative paths are calculated using an improved Dijkstra algorithm. Latency, bandwidth, and hop count features are extracted from the candidate path set, and a preferred path sequence is generated using a weighted scoring algorithm. A comprehensive score is calculated for each candidate path: Score = 0.5 * Delay + 0.3 * Bandwidth + 0.2 * HopCount, with the path with the highest score ranked first in the sequence; where 0.5 is the latency weight, Delay is the path latency, 0.3 is the bandwidth weight, Bandwidth is the path bandwidth, 0.2 is the hop count weight, and HopCount is the hop count. Based on the scoring results of the preferred path sequence, data packet transmission priorities are assigned and load balancing is performed to generate a balanced traffic distribution. Specifically, high-priority data packets are assigned to the path with the highest score, and other data packets are distributed across paths proportionally to their scores. The real-time congestion status of the balanced traffic distribution is monitored. If congestion is detected, backup paths are dynamically switched based on the preferred path sequence to reallocate data flows and generate low-latency transmission channels. For example, when the utilization rate of a certain path is detected to exceed 80%, some traffic is switched to the next preferred path to achieve dynamic load balancing.

[0051] Based on the above analysis, this invention achieves intelligent routing optimization based on real-time network status, effectively reducing data transmission latency and improving network resource utilization. The dynamic routing algorithm and load balancing mechanism can respond promptly to network congestion and flexibly adjust transmission paths, ensuring low-latency data transmission. Furthermore, the priority allocation mechanism ensures timely transmission of critical data, enhancing the system's real-time performance and reliability.

[0052] In this embodiment, step S500 includes: step S501, acquiring real-time data stream through the low-latency transmission channel, cleaning high-risk area data based on the spatial characteristics of the risk distribution map, and generating a standardized data stream; step S502, spatially partitioning the standardized data stream according to the regional boundaries in the risk distribution map, and generating a fragmented data set based on a time window; step S503, extracting spatiotemporal change features from the fragmented data set to generate a dynamic feature sequence; step S504, if the rate of change of a certain region in the dynamic feature sequence exceeds a preset threshold for that region in the risk distribution map, it is identified as a disaster event; step S505, decomposing periodic and trend components according to the spatial weights of the dynamic feature sequence and the risk distribution map, and generating a geographically labeled disaster trend distribution.

[0053] Specifically, data cleaning employs standardization to eliminate differences in data dimensions, which may include normalization or z-score standardization. Spatial partitioning is based on predefined geographic grids on the risk distribution map, such as using geographic information system rasterization. The time window length can be set to 5-10 minutes to balance real-time performance and data volume. Dynamic feature sequence extraction combines moving average algorithms and discrete wavelet transform to identify abrupt changes in the data stream. The disaster event identification threshold is dynamically adjusted based on historical disaster data, for example, using a rate of change exceeding twice the standard deviation of the historical mean as a trigger condition. Trend decomposition uses empirical mode decomposition algorithms to separate periodic fluctuations from long-term trend components. The real-time data stream, received through a low-latency transmission channel, is first spatially filtered based on the high-risk area boundaries marked on the risk distribution map, retaining only the original data from key areas to reduce data processing overhead for irrelevant areas. The cleaned, standardized data stream is divided into multiple independent segments according to geographic grids. Each segment aggregates data points according to a fixed time window, generating a set of segments with spatiotemporal identifiers. Statistical features, including mean, variance, and slope, are extracted from each segment set, and a dynamic sequence is constructed by combining the feature differences between adjacent time windows. When the feature change rate of a segment exceeds a preset threshold, a disaster event is triggered, and the area is assigned a higher spatial weight based on the risk distribution map to enhance its influence on the overall trend. Finally, a decomposition algorithm separates short-term fluctuations from long-term trends in the dynamic sequence, generating disaster spread rate and direction prediction results including geographic coordinates. This process effectively solves the problems of noise interference and spatial positioning ambiguity in real-time data streams, while improving the sensitivity of disaster identification and the accuracy of trend prediction through dynamic thresholds and weight allocation.

[0054] In practical applications, this invention acquires real-time data streams through low-latency transmission channels and cleans high-risk area data based on the spatial characteristics of the risk distribution map to generate a standardized data stream. Specifically, the Kalman filtering algorithm is used to remove noise and detect outliers in the original data, preserving the effective components of the signal. Further, the standardized data stream is spatially partitioned according to the regional boundaries in the risk distribution map, and fragmented data sets are generated based on time windows. For example, a quadtree algorithm is used to divide urban areas into grids of different sizes, with each grid corresponding to a data fragment, and the time window is set to 10 minutes. Spatiotemporal variation features are extracted from the fragmented data sets to generate dynamic feature sequences. Wavelet transform is used to extract multi-scale features of the time series, while spatial autocorrelation analysis is applied to capture geographic location correlations. Therefore, if the rate of change of a certain area in the dynamic feature sequence exceeds a preset threshold for that area in the risk distribution map, it is identified as a disaster event. Specifically, the CUSUM algorithm is used to detect abrupt changes in the time series, combined with spatial clustering methods to determine abnormal areas. Based on the spatial weights of the dynamic feature sequence and the risk distribution map, periodic and trend components are decomposed to generate a geographically labeled disaster trend distribution. Among them, the seasonal trend decomposition (STL) method is used to separate the periodic, trend and residual components, and spatial heterogeneity is considered in combination with the geographically weighted regression model.

[0055] Through the above technical solution, this invention achieves real-time monitoring and trend analysis of urban flooding disasters. By fusing multi-source data and extracting spatiotemporal features, the accuracy of disaster identification is improved. Simultaneously, by utilizing dynamic feature sequences and spatial weights, accurate prediction of disaster trends is achieved, providing timely and reliable decision support for urban managers. Furthermore, this method can quickly identify abnormal events, effectively shortening disaster response time and enhancing urban disaster prevention and mitigation capabilities.

[0056] In this embodiment, step S600 includes: step S601, acquiring dynamic data from a sensor network and generating a structured data stream through standardization; step S602, performing multi-source integration on the structured data stream and generating a spatiotemporal correlated dataset by combining spatial coordinates; step S603, performing projection processing on the spatiotemporal correlated dataset to generate a preliminary layer; step S604, if the distribution density in the preliminary layer exceeds a preset threshold, calculating the disaster concentration area and generating a high-risk area distribution; step S605, updating the layer according to the high-risk area distribution to generate the real-time situation layer.

[0057] The standardization process generates structured data streams using a unified data format conversion rule, transforming raw sensor data into standardized fields with timestamps and geographic coordinates. Multi-source integration establishes a mapping relationship between time series and geographic locations by matching data points from different sensors using spatial coordinates. Projection processing uses an isometric conic projection algorithm to convert three-dimensional spatiotemporal data into a two-dimensional planar coordinate system. Distribution density calculation uses a kernel density estimation algorithm to estimate the density of discrete data points in the preliminary layer. High-risk area distribution uses the DBSCAN clustering algorithm to identify continuous areas with density higher than a preset threshold. Layer updates use an incremental data fusion method to overlay high-risk areas onto the preliminary layer.

[0058] Specifically, the dynamic data collected by the sensor network is first standardized to eliminate differences in data format, sampling frequency, and units among different sensors, generating a structured data stream containing time, space, and environmental parameters. Next, through multi-source integration, the time series data in the structured data stream is correlated with spatial coordinates to form a correlated dataset containing complete spatiotemporal information. Projection processing is then performed on the spatiotemporally correlated dataset, using an isometric conic projection algorithm to convert three-dimensional geographic coordinates into two-dimensional planar coordinates, generating a preliminary layer covering the target area. When the distribution density of data points in the preliminary layer exceeds a preset threshold, the density of data points per unit area is calculated based on a kernel density estimation algorithm to identify areas of concentrated density anomalies, and the DBSCAN clustering algorithm is used to delineate continuous high-risk area boundaries. Finally, the spatial boundaries and attribute data of high-risk areas are dynamically overlaid onto the preliminary layer, and an incremental fusion method is used to update the layer content, ensuring that the real-time situation layer accurately reflects the spatial distribution and evolution trend of disaster hotspots. Through these steps, rapid integration and dynamic updating of multi-source heterogeneous data are achieved, solving the problems of delayed identification of concentrated disaster areas and low layer update efficiency in existing technologies, thereby improving the real-time response capability of the early warning system.

[0059] In practical applications, this invention acquires dynamic data from sensor networks and generates a structured data stream through standardization. The structured data stream is then integrated from multiple sources and combined with spatial coordinates to generate a spatiotemporally correlated dataset. This dataset is then projected to generate a preliminary layer. If the distribution density in the preliminary layer exceeds a preset threshold, concentrated disaster areas are calculated, generating a high-risk area distribution. The layer is updated based on the high-risk area distribution to generate a real-time situation layer. Specifically, real-time data is first acquired from water level sensors, rainfall sensors, flow sensors, etc., distributed throughout the city. This raw data is standardized and converted into a structured data stream in a unified format. Then, structured data from different sources are integrated and combined with spatial coordinate information from a Geographic Information System (GIS) to generate a correlated dataset containing both temporal and spatial information. Further, map projection technology is used to convert the spatiotemporally correlated dataset into a two-dimensional planar layer. The system checks the data distribution density in this preliminary layer; if the density of a certain area exceeds a preset threshold (e.g., more than 10 data points per square kilometer), the area is identified as a concentrated disaster area, and the boundaries of high-risk areas are calculated using a clustering algorithm. Finally, based on the distribution information of high-risk areas, the system dynamically updates the initial layer to generate a real-time situational layer containing information such as risk level and disaster intensity. This layer can be directly overlaid on the city map, with different colors indicating risk levels, and can be updated in real time according to data changes.

[0060] Based on the above analysis, this invention enables real-time visualization of urban flooding disasters. This allows city managers to intuitively grasp the distribution and development trends of the disaster, providing strong support for timely decision-making and resource allocation. Furthermore, the generation of the real-time situation layer integrates multi-source data, improving the accuracy and comprehensiveness of disaster assessment. Simultaneously, by setting dynamic thresholds and an automatic update mechanism, it ensures that the situation layer can promptly reflect the latest changes in the disaster situation, providing more accurate and timely information support for disaster prevention and mitigation efforts.

[0061] In this embodiment, step S700 includes: Step S701, acquiring dynamic data streams from the real-time situation layer, extracting risk hotspots using a spatial clustering algorithm, and generating high-risk boundaries. Step S702, if the water level monitoring data within the high-risk boundaries exceeds a preset threshold, generating flood flow distribution characteristics based on time-series data to obtain the potential flood impact area. Step S703, simulating the flood evolution path using the SWAT hydrological model based on the potential flood impact area, generating a path distribution dataset. Step S704, performing geographic coordinate system transformation on the path distribution dataset, generating spatiotemporal correlation data of flood evolution through spatial mapping. Step S705, if the coverage of the flood evolution path in the spatiotemporal correlation data expands, dynamically updating the high-risk boundaries and determining the expanded risk area. Step S706, based on the expanded risk area, dividing the warning level and generating the pre-disaster warning range.

[0062] The spatial clustering algorithm employs the density-based DBSCAN method, identifying spatial clusters of risk hotspots with a neighborhood radius of 300 meters. The water level monitoring data threshold is set to 1.5 times the historical average water level, triggering the calculation of flood flow distribution characteristics. The SWAT hydrological model's input parameters include surface runoff coefficient, permeability, and topographic slope data, with an output path distribution data temporal resolution of 5 minutes. The geographic coordinate system transformation uses the WGS84 to UTM projection transformation algorithm, ensuring spatial mapping accuracy error is less than 0.5 meters. A dynamic update mechanism performs boundary correction every 2 minutes, triggering an adjustment of the warning level when the coverage growth rate exceeds 10%. After spatial clustering, high-risk boundaries are automatically identified as continuous risk areas through density thresholds. When water level sensors detect levels exceeding the set threshold, flood peak propagation characteristics are extracted through time series analysis, combined with the evolution path simulated by the SWAT model, generating path distribution data with a time dimension. Geographic coordinate transformation overlays the vector path output by the model with a rasterized layer, generating a continuous coverage surface through spatial interpolation. The system continuously monitors changes in the area of ​​the covered surface. When it detects an expansion of the coverage area, it automatically adjusts the geometry of the high-risk boundary and recalculates the warning level for each sub-region. For example, if an area expands at a rate exceeding 500 square meters per hour for three consecutive monitoring periods, the warning level for that area is raised from Level 3 to Level 2. This process is handled in parallel by distributed computing nodes, ensuring that boundary updates and warning range generation are completed within 120 seconds.

[0063] In practical applications, this invention acquires dynamic data from sensor networks and generates a structured data stream through standardization. Specifically, it uses multi-source sensors to collect data related to urban flooding, such as water levels, rainfall, and surface runoff. The raw data is standardized by unifying units and removing outliers to generate a structured data stream with a consistent format. This structured data stream is then integrated from multiple sources and combined with spatial coordinates to generate a spatiotemporal correlated dataset. Further, data from different types of sensors are aligned by timestamps and correlated with spatial coordinate information in a geographic information system (GIS) to form a correlated dataset containing time, space, and multi-dimensional monitoring indicators. The spatiotemporal correlated dataset is then projected to generate a preliminary layer. Using a GIS projection algorithm, the three-dimensional spatiotemporal data is projected onto a two-dimensional plane to generate a preliminary situational layer containing multiple data layers. If the distribution density in the preliminary layer exceeds a preset threshold, areas of concentrated disaster are calculated, generating a high-risk area distribution. Specifically, a data point density threshold is set, and cluster analysis is performed on areas exceeding the threshold to identify high-risk areas with concentrated disasters. The layer is updated based on the high-risk area distribution to generate a real-time situational layer. For example, high-risk areas can be overlaid on a preliminary layer, with different colors used to indicate risk levels, and the layers can be dynamically updated to form the final real-time situational awareness layer.

[0064] Based on the above analysis, this invention achieves real-time visualization of urban flooding monitoring data, improving the intuitiveness and timeliness of disaster situation assessment. By employing multi-source data fusion and spatial analysis techniques, the situation layer comprehensively reflects the dynamic changes in urban flooding, helping decision-makers quickly identify high-risk areas and providing a basis for timely prevention and control measures. Furthermore, the real-time update mechanism ensures the timeliness of situation information, effectively supporting emergency response and resource allocation during disasters.

[0065] In this embodiment, step S800 includes: step S801, extracting geographic boundary data from the pre-disaster warning range, fusing dynamic risk parameters from the real-time situation layer, and generating a distributed computing input dataset; step S802, dividing the input dataset into geographic grid sub-tasks using the MapReduce framework based on the spatiotemporal distribution characteristics, and generating a distributed task set; step S803, loading a pre-trained risk prediction model onto each sub-task, dynamically integrating the LSTM time-series prediction module and the random forest regression module using the Bayesian optimization algorithm, and generating a hybrid prediction model; step S804, performing distributed parallel computing based on the hybrid prediction model, using the Kriging interpolation algorithm to spatially smooth the prediction differences between adjacent areas, and generating preliminary prediction results; step S805, performing a reduction operation on the distributed node computing results, combining spatial weighted fusion with the warning level classification standard, and generating a comprehensive risk prediction surface; step S806, overlaying the comprehensive risk prediction surface with flood evolution spatiotemporal data, performing integrity verification, and generating prediction results.

[0066] Specifically, the geographic grid subtasks are divided based on the spatiotemporal distribution characteristics of the input dataset, with each subtask corresponding to an independent geographic grid unit. The hybrid prediction model dynamically adjusts the weight ratio of LSTM and random forest using a Bayesian optimization algorithm; for example, the weight of LSTM is increased when the time-series characteristics of regional hydrological data are significant, and the weight of random forest is increased when spatial heterogeneity is significant. The Kriging interpolation algorithm calculates the semi-variogram of the prediction results of adjacent grids based on geographic coordinates, and uses a Gaussian kernel function to achieve a smooth transition of the difference values. In the spatial weighted fusion process, an inverse distance weighting method is used, allocating weight coefficients according to the distance between the grid unit and the center point of the warning area; the closer the grid is to the warning center, the higher the weight. Specifically, geographic boundary data and real-time risk parameters are fused to form a spatiotemporally labeled input dataset. The MapReduce framework divides the geographic region into several grid units according to data density and computing resource distribution, and each unit is assigned as an independent computing task to distributed nodes. The hybrid prediction model loaded at each node dynamically adjusts the integration ratio of the two algorithms through Bayesian optimization. For example, for areas with strong temporal continuity of water level monitoring data, the weight of the LSTM module is increased to 70%, while the weight of the random forest module is increased to 65% for areas with high terrain complexity. After the nodes complete their calculations, Kriging interpolation constructs a covariance matrix using the coordinates of the grid center points, and Gaussian smoothing is applied to the prediction differences at the boundaries of adjacent grids to eliminate abrupt values ​​caused by differences in model adaptation. In the reduction phase, the calculation results of each node are assigned different weights according to the warning level classification standard. For example, the weight coefficient for the first-level warning area is set to 0.8, and for the second-level warning area, it is set to 0.5. A continuous risk prediction surface is generated through a spatial weighting algorithm. Finally, the spatiotemporal data of flood evolution and the risk prediction surface are spatially superimposed and verified to ensure that the prediction results are consistent between hydrodynamic characteristics and statistical prediction values.

[0067] In practical applications, the real-time situational data stream is input into a spatial clustering algorithm. The density-based OPTICS clustering method is used to identify risk hotspots and generate high-risk boundaries in polygonal vector format. When water level monitoring data exceeds a preset warning line for three consecutive collection periods, time-series data is input into the SWAT hydrological model. By setting watershed topographic parameters and rainfall intensity parameters, the model simulates the surface runoff evolution path of floods under obstructed drainage networks, generating a path distribution dataset containing velocity and direction attributes. The path distribution dataset is transformed from the WGS84 coordinate system to the UTM projection coordinate system, and a spatial interpolation algorithm maps discrete path points to continuous coverage areas, generating spatiotemporal correlation data of flood evolution. If the flood coverage area in the spatiotemporal correlation data increases by more than 15% compared to the previous period, the high-risk boundary is dynamically extended to the outer rectangle of the newly flooded area, while simultaneously overlaying spatial weight coefficients from historical flooding points. The expanded risk area is divided into three levels of warning zones: red, orange, and yellow. The red warning zone corresponds to areas where water depth may exceed 1 meter within 24 hours, the orange warning zone corresponds to areas where water depth may be 0.5-1 meter within 48 hours, and the yellow warning zone corresponds to areas where water depth may be less than 0.5 meters within 72 hours.

[0068] Through the above technical solutions, this invention effectively improves the accuracy of flood evolution path simulation and dynamic updating of risk areas. The SWAT hydrological model combined with real-time water level data can accurately reflect the flood diffusion pattern under the state of obstructed urban surface runoff. Spatial coordinate transformation ensures the spatial consistency of data from different sources. The dynamic boundary expansion mechanism can capture the flood diffusion trend in a timely manner. The multi-level early warning classification standard provides a scientific basis for differentiated emergency response.

[0069] In this embodiment, step S900 includes: step S901, obtaining time series data from the prediction results, calculating the rate of change of disaster intensity over time, and generating a recession rate dataset; step S902, if the recession rate of a certain region in the recession rate dataset is lower than a preset threshold for that region in the prediction results, then extracting the continuous stable time period characteristics of that region and generating a stable recession time period; step S903, applying an exponential smoothing algorithm to the stable recession time period to process noise and generating a smoothed recession trend curve; step S904, if the slope of the recession trend curve meets a preset stable range, then predicting the future disaster intensity through linear extrapolation and generating a preliminary recovery time window; step S905, generating a recovery priority distribution dataset based on the preliminary recovery time window and the distribution of key facilities in each region; step S906, adjusting the time window length using a weighted average algorithm based on the priority distribution dataset and generating a recovery time window with priority identifiers.

[0070] Specifically, the recession rate dataset is obtained through differential operations on time series data, including calculations of the change in disaster intensity at adjacent time points. A sliding window mechanism is used to determine stable recession periods, with the window length dynamically adjusted based on regional historical data. The exponential smoothing algorithm uses quadratic exponential smoothing with a smoothing coefficient set to 0.3-0.6 to balance noise suppression and trend preservation. Linear extrapolation prediction uses the derivative of the recession trend curve, combined with confidence intervals, to determine the start and end points of the recovery time window. The recovery priority distribution dataset uses hospitals and transportation hubs as key facilities, marking their spatial coordinates using a GIS system, with weight allocation based on the population density served by the facilities. In the weighted average algorithm, the weight coefficient of the area surrounding key facilities is increased to 1.5 times that of the regular area, ensuring that adjustments to the recovery time window prioritize coverage of key areas. Specifically, firstly, time series disaster intensity data is extracted from the prediction results generated by distributed computing, and the recession rate for each region is calculated through time-by-time differential operations. When the recession rate of a region is lower than a preset threshold for three consecutive time periods, a stable period extraction operation is triggered, using a sliding window mechanism to filter out consecutive stable periods. The filtered time-period data undergoes quadratic exponential smoothing to eliminate random fluctuation noise and generate a stable fading trend curve. Based on the linear extrapolation results of this curve, combined with the lower limit of the confidence interval, the starting time of the initial recovery time window is determined. Furthermore, according to the distribution location of critical facilities, areas within a 1-kilometer radius of hospitals and transportation hubs are marked as high-priority areas, assigned higher weight coefficients in the weighted average algorithm, and their recovery time window length is dynamically shortened. The final generated recovery time window includes priority indicators, enabling post-disaster resource allocation to prioritize the functional restoration of critical areas and improve the overall recovery efficiency of the city.

[0071] In a preferred embodiment, the present invention obtains time-series data of disaster intensity in each region from the prediction results, calculates the rate of change of disaster intensity for each geographic unit, and forms a regression rate dataset containing timestamps and intensity values. When the regression rate of a region is lower than a preset threshold for three consecutive time windows, the stable time period characteristics of the region are extracted to generate a stable regression time period containing the start time and duration. The data within the stable regression time period are processed using an exponential smoothing algorithm to handle high-frequency noise, and abnormal fluctuations are suppressed by adjusting the smoothing coefficient to generate a regression trend curve with a gently changing slope. When the slope of the curve is within a preset stable range for six consecutive time windows, a linear extrapolation method is used to predict the disaster intensity decay trajectory for the next six time windows based on the current slope value, generating a preliminary recovery time window containing confidence intervals. Combining the coordinate data of hospitals, substations, and transportation hubs in the urban geographic information system, a priority impact area with a radius of 500 meters is generated centered on the key facilities. The number of key facilities contained in each area within the preliminary recovery time window is counted to form a priority distribution dataset. Based on the weight coefficients of the priority distribution dataset, the length of the time window is dynamically adjusted by weighting, and finally a recovery time window with labeled priority level and optimized time range is generated.

[0072] Based on the above analysis, this invention achieves refined prediction and optimized resource allocation for post-disaster recovery. By dynamically identifying areas with abnormal disaster recession rates and analyzing their stability characteristics, it effectively avoids interference from noisy data in recovery time prediction. By combining linear extrapolation with key facility distribution data from a geographic information system, priority weights are embedded in the time window adjustment to ensure priority recovery in areas where important infrastructure is located. This method significantly improves the timeliness and rationality of post-disaster recovery planning, providing a reliable basis for urban management departments to quickly formulate phased recovery plans.

[0073] Example 2, Figure 2 This is a schematic diagram of the structure of an urban flooding disaster monitoring and early warning system according to Embodiment 2 of the present invention, as shown below. Figure 2As shown in Embodiment 2, an urban flooding disaster monitoring and early warning system is provided, including: a data acquisition module 201, a multidimensional dataset generation module 202, a distribution map generation module 203, a transmission channel generation module 204, a trend distribution generation module 205, a situation layer generation module 206, an early warning range generation module 207, a prediction result generation module 208, and a time window generation module 209. The data acquisition module 201 is used to acquire initial signals from multiple types of sensors and generate stable acquisition data through adaptive processing. The multidimensional dataset generation module 202 is used to generate a high-precision multidimensional dataset by fusing multi-source environmental information based on the stable acquisition data. The distribution map generation module 203 is used to extract spatial features through a neural network and generate a risk distribution map if the high-precision multidimensional dataset exceeds a preset threshold. The transmission channel generation module 204 is used to optimize the data transmission path of the area corresponding to the risk distribution map through a communication network based on the amount of regional data, generating a low-latency transmission channel. The trend distribution generation module 205 is used to acquire real-time monitoring data streams from multiple types of sensors through the low-latency transmission channel, and process the real-time data streams based on the spatial characteristics of the risk distribution map to generate a disaster trend distribution. The situation layer generation module 206 is used to generate a real-time situation layer based on the disaster trend distribution through visualization mapping. The early warning range generation module 207 is used to generate a pre-disaster early warning range by simulating the disaster evolution path according to the real-time situation layer. The prediction result generation module 208 is used to generate a prediction result by optimizing risk prediction through distributed computing based on the pre-disaster early warning range and the real-time situation layer. The time window generation module 209 is used to generate a recovery time window based on the prediction result and analyzing the disaster decline rate.

[0074] In this embodiment, the data acquisition generation module 201 includes: a unit for generating a classified signal set, a unit for generating a denoised signal set, a unit for generating a smoothed signal set, a unit for generating a compressed data set, and a unit for generating stable acquired data. The unit for generating a classified signal set acquires initial signals from multiple types of sensors, uses a pre-trained model to determine the signal type, and generates a classified signal set. The unit for generating a denoised signal set calculates the signal-to-noise ratio (SNR) for the classified signal set using an adaptive filter corresponding to the signal type. If the SNR is lower than a preset threshold, filtering is performed to generate a denoised signal set. The unit for generating a smoothed signal set extracts time-frequency features from the denoised signal set, calculates the correlation coefficient between signal segments, and uses a smoothing method if the correlation coefficient is lower than a preset threshold to generate a smoothed signal set. The unit for generating a compressed data set extracts key data points with significant time-frequency features from the smoothed signal set, compresses them using an encoding method, and generates a compressed data set. The unit for generating stable acquired data performs an integrity check on the compressed data set; if the check passes, the stable acquired data is generated.

[0075] In this embodiment, the multidimensional dataset generation module 202 includes: a standardized dataset generation unit, a preliminary multidimensional dataset generation unit, a corrected multidimensional dataset generation unit, a low-dimensional compressed dataset generation unit, and a low-dimensional compressed dataset generation unit. The standardized dataset generation unit extracts multi-source environmental data from the stable collected data and generates a standardized dataset through standardization processing. The preliminary multidimensional dataset generation unit uses a sensor weight-based dynamic fusion algorithm to generate a preliminary multidimensional dataset from the standardized dataset. The corrected multidimensional dataset generation unit extracts spatiotemporal features from the preliminary multidimensional dataset, calculates the Mahalanobis distance between features, and corrects outliers through interpolation if the distance exceeds a preset threshold, generating a corrected multidimensional dataset. The low-dimensional compressed dataset generation unit uses a random forest algorithm to check the missing value distribution of the corrected multidimensional dataset; if the missing value rate is less than 5%, t-SNE dimensionality reduction is performed to generate a low-dimensional compressed dataset. The low-dimensional compressed dataset generation unit performs integrity verification on the low-dimensional compressed dataset to generate the high-precision multidimensional dataset.

[0076] In this embodiment, the distribution map generation module 203 includes: a three-dimensional feature tensor generation unit, a geographic projection feature map generation unit, an initial risk heatmap generation unit, a surface gradient distribution calculation unit, and a risk distribution map generation unit. The three-dimensional feature tensor generation unit is used to extract spatial topological features and generate a three-dimensional feature tensor when environmental parameters in the high-precision multidimensional dataset exceed corresponding thresholds, using a U-Net convolutional neural network. The geographic projection feature map generation unit performs channel attention weighting on the three-dimensional feature tensor and reduces the dimensionality of the weighted three-dimensional features to a two-dimensional geographic plane using principal component analysis, generating a geographic projection feature map. The initial risk heatmap generation unit uses the DBSCAN clustering algorithm to divide risk areas based on the spatial resolution of the geographic projection feature map, with a neighborhood radius of 0.5 km, to generate an initial risk heatmap. The surface gradient distribution calculation unit uses the Kriging interpolation algorithm to generate a continuous density distribution surface from the discrete risk values ​​of the initial risk heatmap and calculates the gradient distribution of the surface. The risk distribution map generation unit is used to divide risk level thresholds based on the gradient distribution of the continuous density distribution surface and generate a risk distribution map with graded labels.

[0077] In this embodiment, the transmission channel generation module 204 includes: a performance indicator set generation unit, a candidate path set generation unit, a preferred path sequence generation unit, a traffic balance distribution generation unit, and a low-latency transmission channel generation unit. The performance indicator set generation unit is used to obtain transmission latency, bandwidth, and packet loss rate parameters from the communication network in real time, and extract key performance indicators to generate a performance indicator set. The candidate path set generation unit is used to generate a candidate path set based on the current network topology and real-time load status, using a dynamic routing algorithm, if the latency in the performance indicator set exceeds a preset threshold. The preferred path sequence generation unit is used to extract latency, bandwidth, and hop count features from the candidate path set and generate a preferred path sequence using a weighted scoring algorithm. The traffic balance distribution generation unit is used to allocate data packet transmission priorities and perform load balancing based on the scoring results of the preferred path sequence to generate a traffic balance distribution. The low-latency transmission channel generation unit is used to monitor the real-time congestion status of the traffic balance distribution; if congestion exists, it dynamically switches to backup paths based on the preferred path sequence, reallocates data streams, and generates a low-latency transmission channel.

[0078] In this embodiment, the trend distribution generation module 205 includes: a standardized data stream generation unit, a fragmented data set generation unit, a dynamic feature sequence generation unit, an identification unit, and a disaster trend distribution generation unit. The standardized data stream generation unit acquires real-time data streams through the low-latency transmission channel, cleans high-risk area data based on the spatial characteristics of the risk distribution map, and generates a standardized data stream. The fragmented data set generation unit spatially partitions the standardized data stream according to the regional boundaries in the risk distribution map and generates fragmented data sets based on time windows. The dynamic feature sequence generation unit extracts spatiotemporal change features from the fragmented data sets to generate dynamic feature sequences. The identification unit identifies a disaster event if the rate of change of a region in the dynamic feature sequence exceeds a preset threshold for that region in the risk distribution map. The disaster trend distribution generation unit decomposes periodic and trend components according to the spatial weights of the dynamic feature sequences and the risk distribution map, generating a geographically labeled disaster trend distribution.

[0079] In this embodiment, the situation layer generation module 206 includes: a structured data stream generation unit, a spatiotemporal correlation dataset generation unit, a preliminary layer generation unit, a high-risk area distribution generation unit, and a real-time situation layer generation unit. The structured data stream generation unit acquires dynamic data from the sensor network and generates a structured data stream through standardization. The spatiotemporal correlation dataset generation unit integrates the structured data stream from multiple sources and generates a spatiotemporal correlation dataset by combining spatial coordinates. The preliminary layer generation unit projects the spatiotemporal correlation dataset to generate a preliminary layer. The high-risk area distribution generation unit calculates the disaster concentration area and generates a high-risk area distribution if the distribution density in the preliminary layer exceeds a preset threshold. The real-time situation layer generation unit updates the layer based on the high-risk area distribution to generate the real-time situation layer.

[0080] In this embodiment, the early warning range generation module 207 includes: a high-risk boundary generation unit, an acquisition unit, a path distribution dataset generation unit, a spatiotemporal correlation data generation unit, a risk area determination unit, and a pre-disaster early warning range generation unit. The high-risk boundary generation unit acquires dynamic data streams from the real-time situation layer, extracts risk hotspots using a spatial clustering algorithm, and generates a high-risk boundary. The acquisition unit generates flood flow distribution characteristics based on time-series data to obtain the potential flood impact area if the water level monitoring data within the high-risk boundary exceeds a preset threshold. The path distribution dataset generation unit simulates the flood evolution path using the SWAT hydrological model based on the potential flood impact area, generating a path distribution dataset. The spatiotemporal correlation data generation unit performs geographic coordinate system transformation on the path distribution dataset, generating spatiotemporal correlation data of flood evolution through spatial mapping. The risk area determination unit dynamically updates the high-risk boundary and determines the expanded risk area if the coverage of the flood evolution path in the spatiotemporal correlation data expands. The pre-disaster early warning range generation unit divides the early warning level and generates the pre-disaster early warning range based on the expanded risk area.

[0081] In this embodiment, the prediction result generation module 208 includes: an input dataset generation unit, a distributed task set generation unit, a preliminary prediction result generation unit, a comprehensive risk prediction surface generation unit, and a prediction result generation unit. The input dataset generation unit extracts geographic boundary data from the pre-disaster warning area, integrates dynamic risk parameters from the real-time situation layer, and generates a distributed computing input dataset. The distributed task set generation unit divides the input dataset into geographic grid sub-tasks using the MapReduce framework based on the spatiotemporal distribution characteristics of the input dataset, generating a distributed task set. The hybrid prediction model generation unit loads a pre-trained risk prediction model onto each sub-task, dynamically integrates the LSTM time-series prediction module and the random forest regression module using a Bayesian optimization algorithm, and generates a hybrid prediction model. The preliminary prediction result generation unit performs distributed parallel computing based on the hybrid prediction model, uses the Kriging interpolation algorithm to spatially smooth the prediction differences between adjacent areas, and generates preliminary prediction results. The comprehensive risk prediction surface generation unit performs a reduction operation on the distributed node computing results, performs spatial weighted fusion based on the warning level classification standard, and generates a comprehensive risk prediction surface. The prediction result generation unit is used to overlay flood evolution spatiotemporal data onto the comprehensive risk prediction surface, perform integrity verification, and then generate prediction results.

[0082] In this embodiment, the time window generation module 209 includes: a unit for generating a recession rate dataset, a unit for generating a stable recession time period, a unit for generating a recession trend curve, a unit for generating a preliminary recovery time window, a unit for generating a recovery priority distribution dataset, and a unit for generating a recovery time window. The recession rate dataset generation unit is used to obtain time series data from the prediction results, calculate the rate of change of disaster intensity over time, and generate a recession rate dataset. The stable recession time period generation unit is used to extract the continuous stable time period features of a region if the recession rate of a certain region in the recession rate dataset is lower than a preset threshold for that region in the prediction results, and generate a stable recession time period. The recession trend curve generation unit is used to process noise in the stable recession time period using an exponential smoothing algorithm to generate a smoothed recession trend curve. The preliminary recovery time window generation unit is used to predict the future disaster intensity through linear extrapolation if the slope of the recession trend curve meets a preset stable range, and generate a preliminary recovery time window. The recovery priority distribution dataset generation unit is used to generate a recovery priority distribution dataset based on the preliminary recovery time window and the distribution of critical facilities in each region. The unit for generating recovery time windows is used to adjust the time window length based on the priority distribution dataset using a weighted average algorithm, thereby generating recovery time windows with priority identifiers.

[0083] The various variations and specific examples of the urban flooding disaster monitoring and early warning method provided in Embodiment 1 are also applicable to the urban flooding disaster monitoring and early warning system provided in this embodiment. Through the foregoing detailed description of an urban flooding disaster monitoring and early warning method, those skilled in the art can clearly understand the implementation method of an urban flooding disaster monitoring and early warning system in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0084] Example 3, Figure 3 This is a schematic diagram of the structure of an electronic device according to Embodiment 3 of the present invention, as shown below. Figure 3 As shown, Embodiment 3 also provides an electronic device 300, which may include a processor 301 and a memory 302.

[0085] Memory 302 is used to store programs. Memory 302 may include volatile memory, such as random-access memory (RAM), such as static random-access memory (SRAM), double data rate synchronous dynamic random-access memory (DDR SDRAM), etc.; memory may also include non-volatile memory, such as flash memory. Memory 302 is used to store computer programs (such as application programs, functional modules, etc. that implement the above methods), computer instructions, etc. The computer programs, computer instructions, etc., can be partitioned and stored in one or more memories 302. Furthermore, the computer programs, computer instructions, data, etc., can be accessed by processor 301.

[0086] The aforementioned computer programs and instructions can be stored in one or more partitions of memory 302. Furthermore, the aforementioned computer programs and instructions can be invoked by processor 301.

[0087] The processor 301 is configured to execute the computer program stored in the memory 302 to implement the various steps in the methods described in the above embodiments.

[0088] For details, please refer to the relevant descriptions in the preceding method embodiments.

[0089] The processor 301 and the memory 302 can be independent structures or integrated structures. When the processor 301 and the memory 302 are independent structures, the memory 302 and the processor 301 can be coupled together via bus 303.

[0090] The electronic device in this embodiment can execute the technical solution in the above method. Its specific implementation process and technical principle are the same, and will not be repeated here.

[0091] Example 4: Example 4 also provides a computer-readable storage medium including a computer program and instructions, which, when the computer program or instructions are run on a computer, cause the computer to execute the urban flooding disaster monitoring and early warning method of any embodiment of the present invention.

[0092] Computer-readable storage media include various media that can store program code, such as USB flash drives, external hard drives, ROM, RAM, magnetic disks, or optical disks.

[0093] This embodiment also provides a computer program product, which includes: a computer program stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and the at least one processor executes the computer program to cause the electronic device to perform the solution provided in any of the above embodiments.

[0094] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders.

[0095] This document does not impose any restrictions as long as the desired results of the technical solution disclosed in this invention can be achieved.

[0096] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method for monitoring and early warning of urban flooding disasters, characterized in that, include: Initial signals are acquired from multiple types of sensors, and stable data is generated through adaptive processing. Based on the stable collected data, high-precision multidimensional datasets are generated by fusing multi-source environmental information. If the high-precision multidimensional dataset exceeds a preset threshold, spatial features are extracted through a neural network to generate a risk distribution map; Based on the amount of regional data corresponding to the risk distribution map, the data transmission path of the region is optimized through the communication network to generate a low-latency transmission channel. The real-time monitoring data stream from multiple types of sensors is acquired through the low-latency transmission channel, and the real-time data stream is processed based on the spatial characteristics of the risk distribution map to generate a disaster trend distribution. Based on the aforementioned disaster trend distribution, a real-time situation layer is generated through visualization mapping; Based on the real-time situation layer, the disaster evolution path is simulated to generate the pre-disaster early warning range; Based on the pre-disaster warning range and the real-time situation layer, risk prediction is optimized through distributed computing to generate prediction results; Based on the prediction results, the disaster mitigation rate is analyzed to generate a recovery time window.

2. The urban flooding disaster monitoring and early warning method as described in claim 1, characterized in that, The process of acquiring initial signals from multiple types of sensors and generating stable acquired data through adaptive processing includes: Initial signals are acquired from multiple types of sensors, and a pre-trained model is used to determine the signal type, generating a set of classified signals. For the set of classified signals, an adaptive filter corresponding to the signal type is used to calculate the signal-to-noise ratio. If the signal-to-noise ratio is lower than a preset threshold, filtering is performed to generate a set of denoised signals. Extract time-frequency features from the denoised signal set, calculate the correlation coefficient between signal segments, and if the correlation coefficient is lower than a preset threshold, use a smoothing method to generate a smoothed signal set. Key data points with significant time-frequency characteristics are extracted from the smooth signal set and compressed using an encoding method to generate a compressed data set. An integrity check is performed on the compressed data set. If the check passes, the stable collected data is generated.

3. The urban flooding disaster monitoring and early warning method as described in claim 1, characterized in that, The process of generating a high-precision multidimensional dataset by fusing multi-source environmental information from stably collected data includes: Multi-source environmental data is extracted from the stable collected data, and a standardized dataset is generated through standardization processing. For the standardized dataset, a preliminary multidimensional dataset is generated using a dynamic fusion algorithm based on sensor weights; Spatiotemporal features are extracted from the initial multidimensional dataset, and Mahalanobis distance between features is calculated. If the distance exceeds a preset threshold, outliers are corrected by interpolation to generate a corrected multidimensional dataset. The random forest algorithm is used to check the missing value distribution of the modified cube. If the missing value rate is less than 5%, t-SNE dimensionality reduction is performed to generate a low-dimensional compressed dataset. The integrity of the low-dimensional compressed dataset is verified to generate the high-precision multidimensional dataset.

4. The urban flooding disaster monitoring and early warning method as described in claim 1, characterized in that, If the high-precision multidimensional dataset exceeds a preset threshold, the process of extracting spatial features through a neural network to generate a risk distribution map includes: When the environmental parameters in the high-precision multidimensional dataset exceed the corresponding threshold, the U-Net convolutional neural network is used to extract spatial topological features and generate a three-dimensional feature tensor. The three-dimensional feature tensor is subjected to channel attention weighting, and the weighted three-dimensional features are reduced to a two-dimensional geographic plane by principal component analysis to generate a geographic projection feature map. Based on the spatial resolution of the geographic projection feature map, the DBSCAN clustering algorithm is used to divide the risk areas with a neighborhood radius of 0.5km to generate an initial risk heat map. The discrete risk values ​​of the initial risk heatmap are used to generate a continuous density distribution surface using the Kriging interpolation algorithm, and the gradient distribution of the surface is calculated. Based on the gradient distribution of the continuous density distribution surface, risk level thresholds are defined, and a risk distribution map with graded labels is generated.

5. The urban flooding disaster monitoring and early warning method as described in claim 1, characterized in that, The process of optimizing data transmission paths in regions based on the amount of regional data corresponding to the risk distribution map, and generating low-latency transmission channels through communication networks, includes: Real-time transmission latency, bandwidth, and packet loss rate parameters are obtained from the communication network, and key performance indicators are extracted to generate a set of performance indicators. If the latency in the set of performance metrics exceeds a preset threshold, a set of candidate paths is generated based on the current network topology and real-time load status using a dynamic routing algorithm. The latency, bandwidth, and hop count features are extracted from the candidate path set, and a preferred path sequence is generated using a weighted scoring algorithm. Based on the scoring results of the preferred path sequence, data packet transmission priorities are assigned and load balancing is performed to generate a balanced traffic distribution. The system monitors the real-time congestion status of the balanced traffic distribution. If congestion is present, it dynamically switches to backup paths based on the preferred path sequence, reallocates data streams, and generates low-latency transmission channels.

6. The urban flooding disaster monitoring and early warning method as described in claim 1, characterized in that, The process of acquiring real-time data streams through a low-latency transmission channel and processing the real-time data streams based on the spatial characteristics of the risk distribution map to generate a disaster trend distribution includes: Real-time data streams are acquired through the low-latency transmission channel, and high-risk area data is cleaned based on the spatial characteristics of the risk distribution map to generate standardized data streams. The standardized data stream is spatially partitioned according to the regional boundaries in the risk distribution map, and fragmented data sets are generated based on time windows; Spatiotemporal variation features are extracted from the fragmented data set to generate a dynamic feature sequence; If the rate of change of a certain region in the dynamic feature sequence exceeds a preset threshold for that region in the risk distribution map, it is identified as a disaster event. Based on the spatial weights of the dynamic feature sequence and risk distribution map, the periodic and trend components are decomposed to generate a geographically labeled disaster trend distribution.

7. The urban flooding disaster monitoring and early warning method as described in claim 1, characterized in that, The generation of a real-time situation layer based on disaster trend distribution and visualization mapping includes: Dynamic data is acquired from sensor networks and structured data streams are generated through standardization; The structured data stream is integrated from multiple sources and combined with spatial coordinates to generate a spatiotemporally correlated dataset; The spatiotemporal correlated dataset is projected to generate a preliminary layer; If the distribution density in the preliminary layer exceeds a preset threshold, then the disaster concentration area is calculated, and a high-risk area distribution is generated; The layer is updated based on the distribution of high-risk areas to generate the real-time situation layer.

8. A monitoring and early warning system for urban flooding disasters, characterized in that, include: The data acquisition module is used to acquire initial signals from multiple types of sensors and generate stable acquisition data through adaptive processing. A multidimensional dataset generation module is used to generate a high-precision multidimensional dataset by fusing multi-source environmental information based on the stable collected data. The distribution map generation module is used to extract spatial features through a neural network and generate a risk distribution map if the high-precision multidimensional dataset exceeds a preset threshold. The transmission channel generation module is used to optimize the data transmission path of the region based on the amount of regional data corresponding to the risk distribution map and generate a low-latency transmission channel through the communication network. A trend distribution generation module is used to acquire real-time monitoring data streams from multiple types of sensors through the low-latency transmission channel, and process the real-time data streams based on the spatial characteristics of the risk distribution map to generate a disaster trend distribution. The situation layer generation module is used to generate a real-time situation layer through visualization mapping based on the disaster trend distribution. The early warning range generation module is used to generate the pre-disaster early warning range based on the real-time situation layer and the simulation of the disaster evolution path. The prediction result generation module is used to generate prediction results by optimizing risk prediction through distributed computing based on the pre-disaster warning range and the real-time situation layer. The time window generation module is used to generate a recovery time window based on the prediction results and the disaster mitigation rate.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the urban flooding disaster monitoring and early warning method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It includes computer programs and instructions, which, when run on a computer, cause the computer to perform the urban flooding disaster monitoring and early warning method as described in any one of claims 1-7.

Images