Method and system for identifying critical state of urban flood control resilience considering multiple stress couplings

Abstract

The present invention discloses a method and system for identifying the critical state of urban flood control resilience considering multiple pressure couplings. The original meteorological, hydrological, engineering, and socioeconomic data are preprocessed in a spatio-temporal unified manner, and the pressure characteristics are decomposed after outlier detection and correction; based on mutual information entropy and Granger causality test, the coupling relationship between pressure elements is analyzed, a dynamic network is constructed, and the network evolution characteristics are extracted; then, combined with historical resilience assessment data, state indicators are constructed, and a resilience state evolution model is established through time series decomposition and scenario simulation; based on real-time monitoring data, critical feature analysis and threshold optimization are carried out to realize the dynamic identification of the critical state of flood control resilience. This method can accurately identify the critical state of the urban flood control system under multiple pressure couplings and improve the accuracy of early warning.

Description

Technical Field

[0001] The present invention relates to flood simulation technology, in particular to a method and system for identifying the critical state of urban flood control resilience considering multiple stress couplings. Background Art

[0002] Under the dual influence of global climate change and rapid urbanization, urban flood control systems are facing unprecedented challenges. The coupled effects of multiple stress factors such as frequent extreme rainfall events, expanding urban impervious areas, and aging drainage systems have made the occurrence mechanism of urban flood disasters increasingly complex. Research shows that the urban flood control resilience assessment method under a single stress source is no longer sufficient to meet the needs of the current complex urban system. There is an urgent need to establish a method for identifying the critical state of urban flood control resilience under the condition of multiple stress couplings to improve the urban flood control and disaster reduction capabilities.

[0003] Currently, urban flood control resilience research mainly focuses on assessment methods under a single stress source and the determination of static thresholds. Traditional methods usually use deterministic hydrological models to simulate rainfall-runoff processes or use statistical methods to analyze historical flood event data to establish an experience-based early warning index system. Some studies have tried to introduce socio-economic vulnerability indicators to construct a comprehensive assessment framework. However, these methods often analyze different stress sources separately and independently, ignoring the dynamic coupling effect between stress elements. At the same time, the existing early warning thresholds are mostly determined based on static statistical analysis, lacking consideration of the dynamic evolution characteristics of the system.

[0004] In practical applications, the existing technologies have the following key problems: First, the spatio-temporal scales of multi-source heterogeneous data are inconsistent, resulting in information distortion during the data integration process and making it difficult to accurately describe the spatio-temporal distribution characteristics of different stress elements. Second, traditional correlation analysis methods are difficult to capture the non-linear coupling relationship between stress elements, especially when considering time lag effects, and cannot effectively identify causal chains. Third, existing resilience assessment models are mostly based on the principle of linear superposition, ignoring the synergistic and antagonistic effects between stress elements, resulting in a large error in the judgment of the system's critical state. In addition, the dynamic optimization problem of early warning thresholds has not been effectively solved. Especially when considering the multi-stress coupling effect, how to establish an adaptive threshold update mechanism is still a challenge. The sensitivity analysis method of system parameters is relatively simple and difficult to accurately evaluate the influence degree of parameter perturbation on model results, affecting the reliability of early warning results. Summary of the Invention

[0005] The object of the invention is to provide a method and system for identifying the critical state of urban flood control resilience considering multiple stress couplings, in order to solve at least one technical problem existing in the prior art.

[0006] Technical solution. According to one aspect of the present application, a method for identifying the critical state of urban flood control resilience considering multiple pressure couplings is provided, including the following steps:

[0007] Step S1: According to the characteristics of the urban flood control system, identify the pressure sources affecting the resilience and performance of the urban flood control system; obtain research data and perform unified spatio-temporal preprocessing, outlier detection and correction to obtain a corrected data set; perform pressure feature decomposition on the corrected data set to obtain a pressure feature vector set including time dimension features, space dimension features and intensity dimension features;

[0008] Step S2: Obtain the pressure feature vector set and perform coupling relationship analysis to obtain a coupling strength matrix and a coupling direction matrix; construct a directed weighted network based on the coupling strength matrix and the coupling direction matrix to obtain a pressure coupling dynamic network; perform topological structure analysis on the pressure coupling dynamic network to obtain a network evolution feature set;

[0009] Step S3: Perform comprehensive analysis on the network evolution feature set and the pre-stored historical resilience evaluation data to obtain a resilience state index set; perform time series decomposition and scenario simulation to obtain scenario simulation data; construct a state evolution prediction model based on the scenario simulation data and the resilience state index set to obtain a resilience state evolution model;

[0010] Step S4: Perform critical feature analysis based on the resilience state evolution model and the pre-stored real-time monitoring data to obtain a critical feature index set; perform threshold optimization calculation according to the critical feature index set to obtain an early warning threshold set.

[0011] According to one aspect of the present application, step S1 is specifically as follows:

[0012] Step S11: According to the characteristics of the urban flood control system, identify the main pressure sources affecting the resilience and performance of the urban flood control system; including meteorological pressure sources, hydrological pressure sources, engineering pressure sources and socio-economic pressure sources;

[0013] Step S12: Obtain research data, perform Kriging interpolation calculation to obtain temporary spatial interpolation data; perform linear interpolation on the temporary spatial interpolation data according to the time series to obtain temporary time interpolation data; calculate the time weight function according to the reciprocal of the data time interval and perform weight correction on the temporary time interpolation data to obtain a unified spatio-temporal scale data set;

[0014] Step S13: Read the unified spatio-temporal scale data set and calculate the mean and standard deviation to obtain statistical feature data; perform preliminary outlier identification to obtain preliminary outlier data and compare it with physical constraint conditions to obtain confirmed outlier data and perform correction to obtain corrected data; merge the corrected data with the normal value data in the unified spatio-temporal scale data set to obtain a corrected data set;

[0015] Step S14: Extract the rainfall data from the corrected dataset and perform rainfall-runoff simulation to obtain runoff simulation data; perform evolution calculation on the runoff simulation data to obtain flood inundation data; merge the flood inundation data with the corrected dataset to obtain hydrodynamic characteristic data;

[0016] Step S15: Perform wavelet transform on the hydrodynamic characteristic data to obtain a wavelet coefficient set; perform threshold decomposition on the wavelet coefficient set to obtain characteristic component data of different scales; reorganize the characteristic component data to obtain multi-dimensional characteristic data; perform standardization processing on the multi-dimensional characteristic data to obtain standardized characteristic data; convert the standardized characteristic data into a vector form to obtain a pressure characteristic vector set.

[0017] According to one aspect of the present application, step S2 is specifically as follows:

[0018] Step S21: Obtain and calculate the mutual information entropy for each pair of characteristic vectors in the pressure characteristic vector set to obtain a mutual information entropy matrix; calculate the coupling strength value to obtain a coupling strength matrix; calculate the Granger causality test statistic for each pair of characteristic vectors in the pressure characteristic vector set to obtain causality test data; determine the coupling direction according to the causality test data to obtain a coupling direction matrix;

[0019] Step S22: Construct network edge weights based on the coupling strength matrix to obtain a weighted edge set; determine the direction of network edges based on the coupling direction matrix to obtain a directed edge set; combine the weighted edge set and the directed edge set, and add time marks to obtain a pressure coupling dynamic network;

[0020] Step S23: Calculate node degree characteristic data, clustering characteristic data, centrality characteristic data, and community evolution data based on the pressure coupling dynamic network; and combine them to obtain a network evolution characteristic set.

[0021] According to one aspect of the present application, step S3 is specifically as follows:

[0022] Step S31: Perform principal component analysis on the network evolution characteristic set to obtain principal component characteristic data; perform standardization processing on the pre-stored historical resilience evaluation data to obtain standardized resilience data; perform correlation analysis on the principal component characteristic data and the standardized resilience data to obtain a characteristic correlation matrix; select key characteristic combinations according to the characteristic correlation matrix to obtain a resilience characteristic candidate set; perform weighted combination on the resilience characteristic candidate set to obtain a resilience state index set;

[0023] Step S32: Perform time series decomposition on the resilience state index set to obtain trend component data and fluctuation component data; set different boundary conditions according to the trend component data and the fluctuation component data to obtain scenario boundary data; use the hydrodynamic model to perform simulation calculation on the scenario boundary data to obtain scenario simulation data;

[0024] Step S33: Construct an autoregressive model for the scenario simulation data and the resilience status index set to obtain trend prediction data; calculate the phase space reconstruction parameters based on the scenario simulation data to obtain reconstruction parameter data; construct a phase space according to the reconstruction parameter data to obtain state space data; combine the trend prediction data and the state space data to construct a hybrid prediction model to obtain a resilience status evolution model.

[0025] According to one aspect of the present application, step S4 is specifically as follows:

[0026] Step S41: Calculate the system state change rate for the resilience status evolution model to obtain state change rate data; perform normalization processing and fluctuation analysis on the pre-stored real-time monitoring data to obtain dynamic fluctuation characteristic data; calculate the time correlation of the normalized monitoring data to obtain time correlation data; perform feature fusion on the dynamic fluctuation characteristic data and the time correlation data to obtain a critical feature index set;

[0027] Step S42: Perform statistical distribution analysis on the critical feature index set to obtain distribution characteristic data; construct a probability density function based on the distribution characteristic data to obtain density function data; perform quantile analysis on the density function data to obtain quantile characteristic data; calculate the optimal segmentation point according to the quantile characteristic data to obtain segmentation point data; convert the segmentation point data into a warning threshold to obtain a warning threshold set.

[0028] According to another aspect of the present application, there is also provided a critical state identification system for urban flood control resilience considering multiple stress couplings, including:

[0029] At least one processor; and,

[0030] A memory communicatively connected to the at least one processor; wherein,

[0031] The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the method for identifying the critical state of urban flood control resilience considering multiple stress couplings in any one of the above technical solutions.

[0032] Beneficial effects: It can accurately identify the critical state of the urban flood control system under multiple stress couplings and improve the warning accuracy. The relevant technical effects will be described in detail below in combination with specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flowchart of the method of the present invention.

[0034] Figure 2 is a flowchart of step S1 of the present invention.

[0035] Figure 3It is the flowchart of step S2 of the present invention.

[0036] Figure 4 It is the flowchart of step S3 of the present invention.

[0037] Figure 5 It is the flowchart of step S4 of the present invention. Detailed implementation manners

[0038] As Figure 1 shown, according to one aspect of the present application, a method for identifying the critical state of urban flood control resilience considering multiple pressure couplings is provided, including the following steps:

[0039] Step S1: According to the characteristics of the urban flood control system, identify the pressure sources affecting the resilience and performance of the urban flood control system; obtain research data and perform unified spatio-temporal preprocessing to obtain a unified spatio-temporal scale data set; perform outlier detection and correction on the unified spatio-temporal scale data set to obtain a corrected data set; perform pressure feature decomposition on the corrected data set to obtain a pressure feature vector set including time dimension features, space dimension features, and intensity dimension features.

[0040] Step S2: Obtain the pressure feature vector set and perform coupling relationship analysis to obtain a coupling strength matrix and a coupling direction matrix; construct a directed weighted network based on the coupling strength matrix and the coupling direction matrix to obtain a pressure coupling dynamic network; perform topological structure analysis on the pressure coupling dynamic network to obtain a network evolution feature set.

[0041] Step S3: Perform comprehensive analysis on the network evolution feature set and the pre-stored historical resilience evaluation data to obtain a resilience state index set; perform time series decomposition and scenario simulation on the resilience state index set to obtain scenario simulation data; construct a state evolution prediction model based on the scenario simulation data and the resilience state index set to obtain a resilience state evolution model.

[0042] Step S4: Perform critical feature analysis based on the resilience state evolution model and the pre-stored real-time monitoring data to obtain a critical feature index set; perform threshold optimization calculation according to the critical feature index set to obtain an early warning threshold set.

[0043] A complete method for identifying the critical state of urban flood control resilience is established through systematic data processing, network analysis, model construction, and threshold identification. Starting from data preprocessing, the quality of basic data is ensured through the spatio-temporal unification and outlier correction of multi-source data; through pressure feature decomposition and coupling network construction, the complex interaction relationships among system pressure elements are revealed; then a resilience state prediction model is established based on network evolution characteristics to achieve dynamic prediction of system states; finally, the warning threshold is determined through critical feature analysis, providing a scientific basis for flood control decision-making. The technical solutions that progress step by step and support each other not only improve the accuracy and reliability of the resilience assessment of the urban flood control system but also enhance the real-time performance and adaptability of the warning system, providing an innovative solution for urban flood control and disaster reduction.

[0044] As Figure 2 shown, according to one aspect of the present application, step S1 is specifically as follows:

[0045] Step S11: Identify the main pressure sources affecting the resilience and performance of the urban flood control system according to the characteristics of the urban flood control system; including meteorological pressure sources, hydrological pressure sources, engineering pressure sources, and socio-economic pressure sources; meteorological pressure sources include rainfall intensity and rainfall duration; hydrological pressure sources include water level changes and flow fluctuations in rivers or drainage systems; engineering pressure sources include the states of sluice pumps, drainage system capacity, and pipe network states; socio-economic pressure sources include population density, intensity of economic activities, and traffic flow;

[0046] Step S12: Obtain research data, including original meteorological data, original hydrological data, original engineering data, and original socio-economic data; extract rainfall intensity and duration data from the original meteorological data, water level and flow data from the original hydrological data, states of sluice pumps, drainage system capacity, and pipe network state data from the original engineering data, and population density and asset distribution data from the original socio-economic data, perform Kriging interpolation calculation to obtain temporary spatial interpolation data; perform linear interpolation on the temporary spatial interpolation data according to the time series to obtain temporary time interpolation data; calculate the time weight function according to the reciprocal of the data time interval, and perform weight correction on the temporary time interpolation data to obtain a unified spatio-temporal scale data set;

[0047] Step S13: Read the unified spatio-temporal scale data set and calculate the mean and standard deviation to obtain statistical feature data; use the box plot method to preliminarily identify outliers in the unified spatio-temporal scale data set to obtain preliminary outlier data; compare the preliminary outlier data with physical constraint conditions to obtain confirmed outlier data; use the local polynomial regression method to correct the confirmed outlier data to obtain corrected data; merge the corrected data with the normal value data in the unified spatio-temporal scale data set to obtain a corrected data set;

[0048] Step S14: Extract the rainfall data from the corrected dataset and conduct rainfall-runoff simulation to obtain runoff simulation data; perform evolution calculation on the runoff simulation data using a one-dimensional / two-dimensional coupled hydrodynamic model to obtain flood inundation data; merge the flood inundation data with the corrected dataset to obtain hydrodynamic characteristic data;

[0049] Step S15: Conduct wavelet transform on the hydrodynamic characteristic data to obtain a wavelet coefficient set; perform threshold decomposition on the wavelet coefficient set according to the energy distribution characteristics to obtain characteristic component data of different scales; reorganize the characteristic component data according to the time dimension, space dimension, and intensity dimension to obtain multi-dimensional characteristic data; perform standardization processing on the multi-dimensional characteristic data to obtain standardized characteristic data; convert the standardized characteristic data into a vector form to obtain a pressure characteristic vector set.

[0050] Through a systematic method of spatio-temporal unified preprocessing, outlier detection and correction, and pressure characteristic decomposition of multi-source data, high-quality integration of multi-dimensional pressure data of the urban flood control system is achieved. Specifically, the Kriging interpolation method is used to perform spatial interpolation on meteorological, hydrological, engineering, and socio-economic data, and combined with a time weight function for time series interpolation, ensuring the spatial consistency and time continuity of data from different sources and scales. The box plot method is used for preliminary outlier identification, and combined with physical constraint conditions for outlier confirmation and correction, effectively avoiding the impact of data quality problems on subsequent analysis. Finally, wavelet transform is used to perform pressure characteristic decomposition on the data, extracting the characteristics of the time dimension, space dimension, and intensity dimension to form a standardized pressure characteristic vector set. Through systematic data preprocessing and feature extraction methods, not only the data quality and reliability are improved, but also a high-quality data basis is provided for subsequent coupling relationship analysis. Especially when dealing with multi-source heterogeneous data in the urban flood control system, the interaction relationship between different pressure factors can be effectively captured.

[0051] As Figure 3 shown, according to one aspect of the present application, step S2 is specifically as follows:

[0052] Step S21: Obtain and calculate the mutual information entropy for each pair of characteristic vectors in the pressure characteristic vector set to obtain a mutual information entropy matrix; calculate the coupling strength value according to the mutual information entropy matrix to obtain a coupling strength matrix; calculate the Granger causality test statistic for each pair of characteristic vectors in the pressure characteristic vector set to obtain causality test data; determine the coupling direction according to the causality test data to obtain a coupling direction matrix;

[0053] Step S22: Based on the coupling strength matrix, construct network edge weights to obtain a weighted edge set; based on the coupling direction matrix, determine the direction of network edges to obtain a directed edge set; combine the weighted edge set and the directed edge set, and add time marks to obtain a pressure coupling dynamic network;

[0054] Through the construction method of a directed weighted network, a dynamic network representation of the pressure coupling relationship is realized. This step constructs the network edge weights based on the coupling strength matrix to ensure that the weights of the network edges can accurately reflect the interaction strength between pressure elements. The direction of the network edges is determined according to the coupling direction matrix to construct a set of directed edges, so that the network structure can reflect the path and direction of pressure transmission. Finally, by adding time stamps, the static network is extended to a dynamic network, realizing the temporal evolution characterization of the pressure coupling relationship. The pressure coupling representation method based on complex network theory can not only intuitively display the pressure transmission path in the system, but also track the dynamic changes of the coupling relationship. Especially when analyzing the multiple pressure action mechanisms of the urban flood control system, it can effectively identify the key pressure paths and evolution characteristics, providing a scientific basis for the system vulnerability analysis.

[0055] Step S23: Calculate the node degree distribution of the pressure coupling dynamic network to obtain node degree characteristic data; calculate the clustering coefficient of the pressure coupling dynamic network to obtain clustering characteristic data; calculate the centrality index of the pressure coupling dynamic network to obtain centrality characteristic data; calculate the community structure change of the pressure coupling dynamic network at different time points to obtain community evolution data; combine the node degree characteristic data, clustering characteristic data, centrality characteristic data and community evolution data to obtain a network evolution feature set.

[0056] Through the coupling relationship analysis and dynamic network construction method, the precise quantification of the complex interactions between the pressure elements of the urban flood control system is realized. This step calculates the coupling strength between pressure feature vectors based on mutual information entropy, determines the coupling direction through Granger causality test, and constructs a directed weighted network reflecting the interaction of system pressure elements. The network construction method based on information theory and causal inference can accurately capture the non-linear interaction relationships between different pressure elements. By analyzing the topological structure of the network, including the calculation of node degree distribution, clustering coefficient and centrality index, and the extraction of the evolution characteristics of the community structure over time, the dynamic characteristics of the pressure coupling network of the flood control system are comprehensively characterized. The dynamic network analysis method can not only identify the key pressure elements and their influence paths, but also track the temporal evolution characteristics of the pressure coupling relationship, providing an important theoretical basis for understanding the vulnerability and resilience of the urban flood control system.

[0057] As Figure 4 shown, according to one aspect of the present application, step S3 is specifically as follows:

[0058] Step S31: Perform principal component analysis on the network evolution feature set to obtain principal component feature data; perform standardization processing on the pre-stored historical resilience evaluation data to obtain standardized resilience data; perform correlation analysis on the principal component feature data and the standardized resilience data to obtain a feature correlation matrix; select key feature combinations according to the feature correlation matrix to obtain a candidate set of resilience features; perform weighted combination on the candidate set of resilience features to obtain a set of resilience state indicators;

[0059] Step S32: Perform time series decomposition on the set of resilience state indicators to obtain trend component data and fluctuation component data; set different boundary conditions according to the trend component data and the fluctuation component data to obtain scenario boundary data; use a hydrodynamic model to perform simulation calculations on the scenario boundary data to obtain scenario simulation data;

[0060] Step S33: Construct an autoregressive model for the scenario simulation data and the set of resilience state indicators to obtain trend prediction data; calculate phase space reconstruction parameters based on the scenario simulation data to obtain reconstruction parameter data; construct a phase space according to the reconstruction parameter data to obtain state space data; combine the trend prediction data and the state space data to construct a hybrid prediction model to obtain a resilience state evolution model.

[0061] Through the comprehensive analysis of network evolution features and historical resilience evaluation data, a dynamic prediction model of the resilience state of the urban flood control system is established. This step performs principal component analysis on the network evolution feature set, extracts key feature combinations, and performs correlation analysis with historical resilience evaluation data to construct a resilience state index system. Based on the data-driven feature selection method, the feature dimension is effectively reduced, and the generalization ability of the model is improved. The resilience state indicators are separated into trend components and fluctuation components through time series decomposition, and scenario simulations are performed based on different boundary conditions to obtain the response characteristics of the system under different pressure scenarios. By combining the autoregressive model and the phase space reconstruction method, a hybrid prediction model is established, which can not only capture the long-term evolution trend of the system but also depict the short-term fluctuation characteristics, significantly improving the prediction accuracy. The hybrid modeling strategy that integrates multiple prediction methods not only improves the prediction ability of the model but also enhances the interpretability of the prediction results, providing a reliable decision-making support tool for urban flood control resilience assessment.

[0062] As Figure 5 shown, according to one aspect of the present application, step S4 is specifically as follows:

[0063] Step S41: Calculate the system state change rate for the resilience state evolution model to obtain state change rate data; perform normalization processing on the pre-stored real-time monitoring data to obtain normalized monitoring data; use the simplified hydrodynamic model to quickly simulate the normalized monitoring data to obtain quick simulation data; perform fluctuation analysis on the state change rate data and the quick simulation data to obtain dynamic fluctuation characteristic data; calculate the time correlation of the normalized monitoring data to obtain time correlation data; fuse the dynamic fluctuation characteristic data and the time correlation data to obtain a critical characteristic index set;

[0064] Step S42: Conduct statistical distribution analysis on the critical characteristic index set to obtain distribution characteristic data; construct a probability density function based on the distribution characteristic data to obtain density function data; perform quantile analysis on the density function data to obtain quantile characteristic data; calculate the optimal segmentation point according to the quantile characteristic data to obtain segmentation point data; convert the segmentation point data into warning thresholds to obtain a warning threshold set.

[0065] Through the critical characteristic analysis of the resilience state evolution model and real-time monitoring data, the accurate identification of the warning thresholds of the urban flood control system is achieved. This step calculates the system state change rate, and combines with real-time monitoring data for quick simulation, extracts dynamic fluctuation characteristics and time correlation characteristics, and constructs a critical characteristic index set. Based on the feature extraction method of multi-source data fusion, it can timely capture the abnormal changes of the system state. Through statistical distribution analysis and probability density function construction, perform quantile analysis on the critical characteristics, determine the optimal segmentation point, and finally obtain the warning threshold set. The threshold determination strategy combining statistical theory and optimization methods not only considers the probability distribution characteristics of the system state, but also takes into account the accuracy and timeliness of the warning, significantly improving the reliability and practicability of the warning system.

[0066] According to one aspect of the present application, it further includes:

[0067] Step S5: Conduct parameter sensitivity analysis on the intermediate calculation results obtained in Steps S1 to S4 to obtain a sensitivity assessment report; perform a comprehensive reliability assessment based on the sensitivity assessment report and all data processing results to obtain a final reliability assessment report.

[0068] Step S51: Extract the pressure feature vector set, network evolution feature set, resilience state evolution model, and warning threshold set to obtain the key result data set; conduct perturbation analysis on each calculation parameter in the key result data set to obtain parameter perturbation data; calculate the influence degree of the parameter perturbation data on the final result to obtain influence degree data; perform grade division on the influence degree data to obtain sensitivity grade data; organize the sensitivity grade data into a report form to obtain a sensitivity assessment report. This can not only accurately identify the key influencing parameters but also reveal the interaction relationships between parameters. Especially when evaluating the complex model of the urban flood control system, it can provide a scientific basis for parameter optimization and model simplification.

[0069] Step S52: Extract key indicators from the sensitivity assessment report to obtain key sensitivity indicators; conduct uncertainty analysis on all data processing results from Step S1 to Step S4 to obtain uncertainty data; perform a comprehensive score on the key sensitivity indicators and the uncertainty data to obtain comprehensive score data; establish a reliability assessment standard based on the comprehensive score data to obtain assessment standard data; conduct a reliability assessment on the entire data processing process according to the assessment standard data to obtain the final reliability assessment report.

[0070] Through systematic sensitivity analysis and reliability assessment methods, the quality control and reliability assurance of the entire technical solution are achieved. This step extracts the key result data set from the calculation results of S1 to S4, including the pressure feature vector set, network evolution feature set, resilience state evolution model, and warning threshold set, and constructs a complete evaluation object set. Through parameter perturbation analysis, a method combining local sensitivity analysis and global sensitivity analysis is adopted to quantitatively evaluate the influence degree of each calculation parameter on the final result. In terms of sensitivity grade division, a multi-level classification method based on fuzzy comprehensive evaluation is used, making the sensitivity assessment results more practically instructive. For reliability assessment, key sensitivity indicators are extracted, combined with the Monte Carlo simulation method for uncertainty analysis, and the weights of each indicator are determined by the entropy weight method to establish a scientific comprehensive scoring system. Through multi-level quality control methods, not only can the key influencing factors in the calculation process be identified, but also the reliability level of the evaluation results can be quantified. Especially when dealing with complex systems such as urban flood control, it can effectively improve the credibility of the warning results and provide reliable technical support for flood control decision-making. Specifically, this step identifies the key parameters affecting system performance through parameter sensitivity analysis, evaluates the confidence interval of the model prediction results through uncertainty analysis, ensures the stable operation of the warning system through reliability assessment, and finally forms a complete quality assurance system, significantly enhancing the practical value and promotion potential of the entire technical solution.

[0071] According to one aspect of the present application, Step S12 is specifically:

[0072] Step S121: Analyze the spatial distribution characteristics of rainfall intensity and duration data in the original meteorological data to obtain rainfall spatial characteristic data; analyze the spatial distribution characteristics of water level and flow data in the original hydrological data to obtain hydrological spatial characteristic data; analyze the spatial distribution characteristics of drainage system capacity and pipe network status data in the original engineering data to obtain engineering spatial characteristic data; analyze the spatial distribution characteristics of population density and asset distribution data in the original social and economic data to obtain social spatial characteristic data; perform variational interpolation calculation on the rainfall spatial characteristic data, hydrological spatial characteristic data, engineering spatial characteristic data, and social spatial characteristic data to obtain preliminary spatial interpolation data.

[0073] Step S122: Perform local polynomial fitting on the preliminary spatial interpolation data to obtain local fitting data; use the adaptive anisotropic Kriging method to optimize the interpolation of the local fitting data to obtain optimized interpolation data; perform geographically weighted regression analysis on the optimized interpolation data to obtain temporary spatial interpolation data.

[0074] Kriging interpolation model Z*(x) = Σ(λ i ·Z(x i )) + μ(x)·[1 - Σλ i ; where: variogram γ(h,θ) = C0 + C1·[1 - exp(-h² / a(θ)²)]; anisotropic range function a(θ) = a0·[1 + β·cos²(θ - φ)]; adaptive weight λ i = w i (x)·ρ i / Σ(w i (x)·ρ i ); local anisotropic parameter β(x) = [γmax(x) - γmin(x)] / [γmax(x) + γmin(x)]; where: Z*(x) is the predicted value at the point x to be estimated; Z(x i ) is the observed value at the known sampling point x i ; λ i is the Kriging weight; μ(x) is the local mean; h is the spatial distance; θ is the direction angle; C0 is the nugget effect; C1 is the sill value; a0 is the maximum range; β is the anisotropic coefficient; φ is the main direction angle; w i (x) is the local window weight; ρ i is the data reliability weight; γmax(x) and γmin(x) are the local maximum and minimum variograms respectively; this method improves the accuracy of spatial interpolation by introducing adaptive anisotropic parameters and local weight coefficients.

[0075] Step S123: Perform time series decomposition on the interpolated data in the temporary space to obtain trend sequence data and periodic sequence data; process the trend sequence data using cubic spline interpolation to obtain trend interpolated data; process the periodic sequence data using Fourier interpolation to obtain periodic interpolated data; combine the trend interpolated data and the periodic interpolated data to obtain temporary time interpolated data.

[0076] Step S124: Calculate the time weight function based on the sampling interval of the temporary time interpolated data to obtain time weight data; perform normalization processing on the pre-stored spatial correlation data to obtain spatial weight data; use the bilateral filtering algorithm to perform multi-source fusion on the temporary time interpolated data, time weight data, and spatial weight data to obtain a unified spatio-temporal scale dataset.

[0077] Bilateral filtering algorithm: Weighted fusion function F(x) = k(x)·Σ[Gs(||p - x||)·Gr(|Ip - Ix|)·Gt(|tp - tx|)·Ip] / Σ[Gs(||p - x||)·Gr(|Ip - Ix|)·Gt(|tp - tx|)]; where: Spatial weight function Gs(||p - x||) = exp(-||p - x||² / 2σs²); Numerical weight function Gr(|Ip - Ix|) = exp(-|Ip - Ix|² / 2σr²); Time weight function Gt(|tp - tx|) = exp(-|tp - tx|² / 2σt²); Adaptive kernel parameter k(x) = (1 + λ·v(x)) / (1 + λ); Local coefficient of variation v(x) = std(Wx) / mean(Wx); where: F(x) is the fused data value; p is the sampling point within the neighborhood; x is the current processing point; Ip is the data value of the sampling point; Ix is the data value of the current point; tp is the timestamp of the sampling point; tx is the timestamp of the current point; σs is the spatial smoothing parameter; σr is the numerical smoothing parameter; σt is the time smoothing parameter; λ is the adaptive adjustment coefficient; Wx is the dataset within the local window; std is the standard deviation operator; mean is the mean operator; This method realizes the smooth fusion of spatio-temporal data by introducing time weights and adaptive kernel parameters.

[0078] Through the spatial feature analysis and spatio-temporal interpolation calculation of multi-source data, the unified spatio-temporal scale conversion of heterogeneous data is realized. In this step, the spatial distribution characteristics of different types of data (meteorological, hydrological, engineering, socio-economic) are analyzed, and the variational interpolation method is used for preliminary spatial interpolation. The optimization interpolation is carried out by local polynomial fitting and adaptive anisotropic Kriging method, which improves the accuracy of spatial interpolation. In the time dimension, the time series decomposition method is used to extract the trend and periodic components, and the cubic spline interpolation and Fourier interpolation are used for processing respectively, ensuring the continuity of the time series and the maintenance of periodic characteristics. Finally, the bilateral filtering algorithm is used to perform multi-source fusion on the time weight and spatial weight to obtain a dataset with a unified spatio-temporal scale. Through the data unification method based on multiple interpolation and fusion, not only the spatio-temporal consistency of the data is ensured, but also the interpolation error is effectively reduced. Especially when dealing with multi-source heterogeneous data in the urban flood control system, it can accurately depict the spatio-temporal correlation characteristics between different data sources, providing a high-quality data basis for subsequent analysis.

[0079] According to one aspect of the present application, step S13 is specifically as follows:

[0080] Step S131: Calculate the mean and standard deviation of each variable for the unified spatio-temporal scale dataset to obtain basic statistical feature data; use the kernel density estimation method to calculate the probability distribution characteristics of each variable to obtain distribution feature data; perform a sliding window analysis on the unified spatio-temporal scale dataset to calculate the time series characteristics to obtain time series feature data; fuse the basic statistical feature data, distribution feature data and time series feature data to obtain statistical feature data.

[0081] Step S132: Use the isolation forest algorithm to perform a preliminary screening of outliers on the statistical feature data to obtain the preliminarily screened outlier data; use the local outlier factor method to perform spatial outlier detection on the unified spatio-temporal scale dataset to obtain spatial outlier data; use the dynamic time warping algorithm to perform time series outlier detection on the unified spatio-temporal scale dataset to obtain time series outlier data; perform ensemble learning on the preliminarily screened outlier data, spatial outlier data and time series outlier data to obtain preliminary outlier data.

[0082] Isolation forest algorithm: The outlier scoring function s(x) = 2^(-E[h(x)] / E[h(T)])·[1 + α·d(x)]; where: The expected path length E[h(x)] = Σ(c i ·h i (x)) / n; The expected tree height E[h(T)] = 2·(ln(ψ - 1) + 0.5772156649) - 2·(ψ - 1) / ψ; The distance weight coefficient d(x) = Σ[w(x i )·d i st(x, x i)] / Σw(x i ); The adaptive weight w(x i ) = exp(-s(x i )² / 2σ²); The splitting selection probability P(j) = Iθ(j) / ΣIθ(j); The information gain Iθ(j) = H(D) - Σ(|Dv| / |D|)·H(Dv); where: s(x) is the anomaly score; h(x) is the path length of sample x; h i (x) is the path length of sample x in the i-th tree; c i is the tree weight; n is the number of trees; ψ is the number of samples; d(x) is the distance weight term; α is the weight adjustment parameter; dist(x, x i ) is the distance between samples; σ is the kernel width parameter; P( j ) is the selection probability of feature j; Iθ( j ) is the information gain of feature j; H(D) is the entropy of dataset D; Dv is the sub-dataset; This method improves the accuracy of anomaly detection by introducing distance weights and adaptive feature selection.

[0083] The improvement of the Local Outlier Factor method is specifically as follows: The Local Outlier Factor is calculated as LOF(x) = [Σ(lrd(p) / lrd(x)) / k]·[1 + β·entropy(x)]; where: The local reachability density lrd(x) = 1 / [Σ(reach-distk(x,p)) / k]; The reachability distance reach-distk (x,p) = max{k-dist(p), d(x,p)}·w(x,p); The neighborhood entropy entropy(x)= -Σ[p i ·log(p i )]; The adaptive weight w(x,p) = exp(-||x-p||² / [2·σ(x)²]); The local scale parameter σ(x) = [Σ(k-dist(x,p)²) / k]^(1 / 2); The density probability p i = density(x, i ) / Σdensity(x, j ); where: LOF(x) is the Local Outlier Factor of sample x; lrd(x) is the local reachability density of point x; k is the number of nearest neighbors; β is the entropy weight coefficient; reach-distk(x,p) is the reachability distance; d(x,p) is the Euclidean distance; w(x,p) is the distance weight; k-dist(p) is the distance to the k-th nearest neighbor; entropy(x) is the neighborhood entropy; σ(x) is the local scale parameter; density(x, i) is the density value of point x in the i-th density interval; by introducing neighborhood entropy and adaptive weights, this method enhances the accuracy of local density estimation.

[0084] The dynamic time warping algorithm is specifically: the cumulative distance function DTW( i , j ) = d(x i ,y j ) + min{γ·DTW( i-1 , j-1 ), ω·DTW( i-1 , j ), ω·DTW( i , j-1 )}·exp(-λ·|v i - v j |); where: the local distance calculation d(x i ,y j ) = ||x i - y j ||·[1 + α·cos(θ ij )]; the velocity vector v i = (x i - x i-1 ) / Δt; the direction angle θ ij = arccos[(x i - x i-1 )·(y j - y j-1 ) / (||x i - x i-1 ||·||y j - y j-1 ||)]; the step size weight ω = 1 + β·|v i | / max(|v|); the path penalty γ = 1 - β·|v i - v j | / max(|v|); where: DTW( i , j ) is the cumulative distance between sequences x and y at position ( i , j ); d(x i ,y j ) is the local distance; x i ,y j are the values of the two sequences at the corresponding positions; v i ,v j are the corresponding velocity vectors; θ ijis the direction angle; α is the direction weight coefficient; β is the speed weight coefficient; λ is the time decay coefficient; ω is the step weight; γ is the path penalty factor; this method improves the accuracy of temporal matching by introducing speed and direction information.

[0085] The ensemble learning algorithm is specifically: the ensemble prediction function E(x) = Σ[w i ·f i (x)]·[1 + α·D(x)]·[1+ β·T(t)]; where: the model weight w i = exp(-E i / θ)·[1 + λ·V i (x)] / Σexp(-E j / θ); the diversity measure D(x) = Σ[I(f i (x)≠f j (x))·S ij ] / [M(M - 1)]; the temporal modulation T(t) = exp(-|t - t0|² / 2τ²)·cos(ω·|t - t0|); the model variance V i (x) = E[(f i (x)-f*(x))²]·[1 + μ·ρ i (x)]; the similarity matrix S ij = exp(-||f i - f j ||² / 2σ²)·[1 + γ·cos(θ ij )]; the local density ρ i (x) = Σexp(-||x - xk||² / 2η²)·I(f i (xk)=y i )]; where: E(x) is the ensemble prediction result; f i (x) is the base learner; α is the diversity weight; β is the temporal weight; λ is the variance weight; E i is the base learner error; θ is the temperature parameter; M is the number of models; τ is the time scale; ω is the period parameter; μ is the density weight; σ is the similarity scale; γ is the direction weight; η is the density scale; this method improves the stability of ensemble prediction by introducing diversity measure and temporal modulation.

[0086] Step S133: Constrained verification is performed on the preliminary outlier data according to hydro-physical constraint conditions to obtain physically constrained verification data; a water balance equation is used to perform a balance test on the preliminary outlier data to obtain balance test data; a hydrodynamic model is used to perform a kinetic rationality verification on the preliminary outlier data to obtain kinetic verification data; the physically constrained verification data, the balance test data, and the kinetic verification data are comprehensively evaluated through a multi-criteria decision-making method to obtain confirmed outlier data.

[0087] Step S134: The tensor decomposition method is used to reconstruct the missing values of the confirmed outlier data to obtain reconstructed data; the locally weighted scatter smoothing method is used to smooth the reconstructed data to obtain smoothed data; the wavelet threshold denoising method is used to suppress the noise of the smoothed data to obtain corrected data; the corrected data is fused with the normal value data in the unified spatio-temporal scale dataset to obtain the corrected dataset.

[0088] Improvement of the tensor decomposition method: Tensor reconstruction model T ≈ G ×1 U⁽ 1 ⁾ ×2 U⁽ 2 ⁾ ×3 U⁽ 3 ⁾ + λ·S; where: Core tensor update G = T ×1 (U⁽ 1 ⁾) T ×2 (U⁽ 2 ⁾) T ×3 (U⁽ 3 ⁾) T ; Factor matrix update U⁽ n ⁾ = T⁽ n ⁾·G⁽ n ⁾·[(G⁽ n ⁾)T·G⁽ n ⁾ + αI] -1 ; Sparse term update S = Sτ(T - G ×1 U⁽ 1 ⁾ ×2 U⁽ 2 ⁾ ×3U⁽ 3 ⁾); Soft threshold operator Sτ(x) = sign(x)·max(|x| - τ,0); Adaptive regularization parameter α = η·||T –T*||F / ||T||F; where: T is the original tensor; G is the core tensor; U⁽ n ⁾ is the factor matrix of the nth dimension; λ is the sparse term weight; S is the sparse tensor; T⁽ n ⁾ is the unfolding matrix along the nth dimension; G⁽ nThe mode expansion with the core tensor; τ is the threshold parameter; α is the regularization parameter; η is the scaling coefficient; ||·||F is the Frobenius norm; by introducing adaptive regularization and sparse constraints, the robustness of tensor decomposition is improved.

[0089] Through multi-level outlier detection and correction methods, the overall improvement of data quality is achieved. This step uses the kernel density estimation method to calculate the probability distribution characteristics, combines the sliding window analysis to extract the time series characteristics, and constructs a complete statistical feature system. The outlier pre-screening is carried out by the Isolation Forest algorithm, and the spatial and temporal outlier detection is realized by combining the Local Outlier Factor method and the Dynamic Time Warping algorithm, achieving multi-dimensional outlier recognition. In the outlier confirmation stage, multiple verifications are carried out through hydrological physical constraints, water balance equations, and hydrodynamic rationality verification to ensure the accuracy of outlier determination. Finally, the tensor decomposition method is used for missing value reconstruction, and the data is corrected by local weighted scatter smoothing and wavelet threshold denoising. It not only improves the reliability of the data but also maintains the physical rationality of the data. Especially when dealing with the monitoring data of complex urban flood control systems, it can effectively identify and correct various outliers, providing high-quality data support for subsequent analysis.

[0090] According to one aspect of the present application, step S14 is specifically as follows:

[0091] Step S141: Extract rainfall data from the corrected dataset and conduct spatio-temporal distribution analysis to obtain rainfall distribution characteristic data; use the Green-Ampt model to calculate the infiltration loss to obtain infiltration loss data; calculate the surface interception loss using surface roughness data to obtain interception loss data; conduct water balance calculation on the rainfall distribution characteristic data, infiltration loss data, and interception loss data to obtain effective rainfall data; use the nonlinear unit hydrograph method to conduct runoff concentration calculation on the effective rainfall data to obtain runoff simulation data;

[0092] Green-Ampt model: Infiltration flux equation f(t) = Ks·(1 + [ψf·Δθ] / F)·[1 + β·exp(-αt)]; where: Cumulative infiltration update F(t+Δt) = F(t) + f(t)·Δt + γ·[θs - θ i]·[1 -exp(-λ·I(t))]; effective suction head ψf = ψb·(Se)^(-1 / m); saturation Se = (θ - θr) / (θs - θr); soil hydraulic conductivity Ks = Ks0·[1 + ω·cos(2πt / T)]; where: f(t) is the infiltration flux at time t; F(t) is the cumulative infiltration; Ks is the saturated hydraulic conductivity; ψf is the wetting front suction head; Δθ is the soil water deficit; β and α are time-varying adjustment parameters; γ is the cumulative correction coefficient; θs is the saturated water content; θ i is the initial water content; θr is the residual water content; ψb is the bubble pressure; m is the pore distribution parameter; λ is the rainfall intensity influence coefficient; I(t) is the accumulated rainfall; Ks0 is the benchmark hydraulic conductivity; ω is the seasonal variation amplitude; T is the period; By introducing time-varying parameters and seasonal changes, the accuracy of infiltration calculation is improved.

[0093] Step S142: performing one-dimensional river hydrodynamic calculation based on the runoff simulation data to obtain river hydrodynamic data; performing urban terrain evolution calculation on the runoff simulation data using a two-dimensional overland flow model to obtain overland flow data; performing coupling calculation on the river hydrodynamic data and overland flow data using an interactive iterative algorithm to obtain one- and two-dimensional coupled data; performing inundation range calculation on the one- and two-dimensional coupled data according to pre-stored terrain elevation data to obtain flood inundation data;

[0094] Step S143, extracting spatiotemporal features from the corrected data set to obtain basic feature data; calculating hydrodynamic indicators from the flood and waterlogging inundation data to obtain inundation feature data; combining the basic feature data and the inundation feature data using a multi-source data fusion algorithm to obtain combined feature data; optimizing the combined feature data using an adaptive weight method to obtain hydrodynamic feature data;

[0095] Step S144, perform quality verification on the hydrodynamic characteristic data to obtain characteristic verification data; use physical constraints to verify the validity of the characteristic verification data to obtain characteristic verification data; use a multi-criteria evaluation method to perform reliability evaluation on the characteristic verification data to obtain characteristic evaluation data; compare and correct the characteristic evaluation data with the original hydrodynamic characteristic data to obtain final hydrodynamic characteristic data.

[0096] According to one aspect of the present application, step S21 is specifically:

[0097] Step S211, calculating the mutual information entropy for each pair of eigenvectors in the pressure eigenvector set to obtain preliminary mutual information data; correcting the preliminary mutual information data using a kernel density estimation method to obtain corrected mutual information data;

[0098] Perform matrix reconstruction on the corrected mutual information data to obtain the mutual information entropy matrix;

[0099] The density estimation function f(x) = [1 / (nh)]·Σ{K[(x - x i ) / h]·[1 + α·L(x, x i )]}·[1 + β·S(x)]; where: the local bandwidth h(x) = h o ·[f o (x) / g]^(-γ); the similarity measure L(x, x i ) = exp(-||x - x i ||² / 2σ²)·cos(θx i ); the spatial correlation S(x) = Σ[w i ·exp(-d i ² / 2τ²)]; the kernel function K(u) = (1 - u²)·I(|u| ≤ 1)·[1 + λ·|u|]; the adaptive weight w i = exp[-μ·(f o (x i ) – f* o )²]; where: f(x) is the estimated density; n is the number of samples; h is the bandwidth parameter; α is the similarity weight; β is the spatial weight; h o is the reference bandwidth; f o (x) is the initial density estimate; g is the geometric mean; γ is the adaptive coefficient; σ is the kernel width; θx i is the sample angle; d i is the spatial distance; τ is the spatial scale; λ is the kernel function adjustment parameter; μ is the weight adjustment coefficient; this method improves the accuracy of density estimation by introducing local adaptive bandwidth and spatial correlation.

[0100] Step S212, Obtain the mutual information entropy matrix and perform normalization processing to obtain the normalized mutual information data; Use the non - linear mapping method to perform intensity conversion on the normalized mutual information data to obtain the intensity mapping data; Perform matrix reconstruction on the intensity mapping data to obtain the coupling intensity matrix C ij = exp(-λI ij ), 1 ≤ i , j ≤ n; where, C ij is the coupling intensity matrix.

[0101] Among them, the mapping function M(x) = σ[W·φ(x) + b]·[1 + α·R(x)]·[1 + β·T(x)]; where: the non - linear transformation φ(x) = tanh(ωx)·[1 + λ·||grad x||]; the temporal regularization T(x) = exp(-|t - t o |² / 2σt²)·cos(θ·|t - t o |); the spatial regularization R(x) = Σ[w i ·exp(-||x - x i ||² / 2σs²)]; the adaptive weight w i = [1 + γ·cos(θ i )] / [1 + ||x - x i ||]; the activation function σ(z) = max(0,z) +μ·min(0,z); the gradient term grad x = (x – x*) / std(x); where: M(x) is the mapping result; W is the weight matrix; b is the bias term; α is the spatial weight coefficient; β is the temporal weight coefficient; ω is the non - linear degree parameter; λ is the gradient weight; t is the current time; t0 is the reference time; σt is the time scale parameter; θ is the period parameter; σs is the spatial scale parameter; γ is the direction weight; θ i is the spatial direction angle; μ is the negative slope; by introducing spatio - temporal regularization and adaptive weights, the stability of non - linear mapping is improved.

[0102] Step S213: Perform time - delay analysis on each pair of feature vectors in the pressure feature vector set to obtain time - delay feature data; τ ij =argmax τ corr(x i (t),x j (t+τ)); where, the time - delay τ ij can be calculated by maximizing the cross - correlation function.

[0103] Perform Granger causality test on the pressure feature vector set and the time - delay feature data to obtain preliminary causal data; X j (t)=α+∑ p k=1 βx j (t - k)+∑ p k=1 γ k x i (t - k)+o* t ; where, α is the constant term, β k and γ k are the coefficients to be estimated, ϵ t is the error term, p is the number of lag periods. If the coefficient γk If it is significantly non - zero, then reject the null hypothesis.

[0104] Use the confidence test method to perform a significance analysis on the preliminary causal data to obtain causal test data;

[0105] The test statistic Z(x)=[μ1(x) – μ2(x)] / [s(x)·√(1 / n1 + 1 / n2)]·[1 + α·V(x)]·[1 + β·C(x)]; where: the sample variance s²(x)=[(n1 - 1)s1²+(n2 - 1)s2²] / (n1 + n2 - 2)·[1+ λ·H(x)]; the variance non - homogeneity correction V(x)=max(s1² / s2², s2² / s1²)·exp(-γ·|r 12 |); the autocorrelation correction C(x)=[1 - ρ(x)]·exp(-ω·|Δt|); the heterogeneity measure H(x)=Σ|x i - x*| / [n·std(x)]; the autocorrelation coefficient ρ(x)=Σ[(x i -x*)(x i+1 -x*)] / Σ(x i -x*)²; where: Z(x) is the test statistic; μ1, μ2 are the sample means; s(x) is the pooled standard deviation; n1, n2 are the sample sizes; α is the variance weight; β is the autocorrelation weight; λ is the heterogeneity weight; γ is the correlation weight; ω is the time weight; r 12 is the sample correlation coefficient; Δt is the time interval; by introducing the variance non - homogeneity and autocorrelation corrections, the reliability of the test is improved.

[0106] Step S214: Perform threshold segmentation on the causal test data to obtain causal determination data; perform directional analysis on the causal determination data to obtain direction feature data; perform matrix reconstruction on the direction feature data to obtain a coupling direction matrix.

[0107] Through a multi-dimensional coupling relationship quantification method, an accurate characterization of the interaction between pressure elements is achieved. In this step, the kernel density estimation method is used to correct the mutual information entropy, improving the accuracy of information quantification. The mutual information is converted into the coupling strength through a non-linear mapping method, and a coupling strength matrix reflecting the interaction strength between elements is constructed. In terms of determining the coupling direction, time-delay analysis is carried out to determine the optimal time delay, and causal relationship judgment is performed based on the Granger causality test and confidence test methods. Finally, a coupling direction matrix is constructed. Combining the coupling relationship of information theory, non-linear mapping, and causal inference can not only accurately quantify the interaction strength between pressure elements but also identify the directionality of the interaction. Especially when dealing with the complex relationship between multiple pressures in the urban flood control system, it can effectively capture non-linear and time-varying coupling characteristics, providing a reliable data basis for constructing a pressure coupling network.

[0108] According to one aspect of the present application, step S23 is specifically as follows:

[0109] Step S231: Calculate the in-degree and out-degree of each node in the pressure coupling dynamic network to obtain basic metric data; use the local connection density algorithm to calculate the local connection characteristics of the pressure coupling dynamic network to obtain connection density data; perform weighted combination on the basic metric data and the connection density data to obtain node degree characteristic data.

[0110] The connection density function D(x) = [Σ(K(x,x i )·w(x i )) / V(x)]·[1 + α·L(x)]·[1 + β·T(x)]; where: the kernel function K(x,x i ) = exp(-||x-x i ||² / 2h²)·[1 + λ·cos(θx i )]; the adaptive volume V(x) = πr²(x)·[1 + γ·std(Nx)]; the local scale r(x) = r0·[f(x) / g]^(-ω); the topological feature L(x) = Σ[δ ij ·exp(-d ij / σ)] / k(k - 1); the time-varying feature T(x) = exp(-|t - t0|² / 2τ²)·cos(μ·|t - t0|); the weight function w(x i ) = [1 + η·ρ(x i )] / [1 + ||x - x i ||]; where: D(x) is the local connection density; K(x,x i ) is the kernel function; h is the bandwidth parameter; V(x) is the adaptive volume; α is the topological weight coefficient; β is the time-varying weight coefficient; λ is the direction weight; θxi is the spatial angle; γ is the variation weight; r(x) is the local radius; r0 is the reference radius; f(x) is the initial density estimate; g is the geometric mean; ω is the adaptive exponent; δ ij is the connection indicator; d ij is the path distance; σ is the scale parameter; k is the number of neighbors; τ is the time scale; μ is the period parameter; η is the density weight; ρ(x i ) is the point density; This method improves the accuracy of connection density estimation by introducing adaptive volume and topological features.

[0111] Step S232: Extract the triangular structure features from the pressure-coupled dynamic network to obtain triangular feature data; Use the local clustering coefficient algorithm to perform clustering analysis on the pressure-coupled dynamic network to obtain clustering analysis data; Perform feature fusion on the triangular feature data and the clustering analysis data to obtain clustering feature data.

[0112] Local clustering coefficient algorithm: Clustering coefficient C( i ) = [2·Σ(w ij ·w jk ·w ki )·(1 + α·θ ijk )]·[1 + β·T( i )] / [k i ·(k i-1 )]; where: Edge weight w ij = exp(-||x i -x j ||² / 2σ²)·[1 + λ·cos(φ ij )]; Time-varying feature T( i ) = exp(-|t i -t0|² / 2τ²)·cos(ω·|t i -t0|); Angular feature θ ij k = exp[γ·cos(ψ ij k)]; Local metric k i = Σ j [I(w ij > δ)·(1 + μ·ρ ij )]; Density correlation ρ ij = min(ρ i ,ρ j ) / max(ρ i ,ρ j ); where: C( i ) is the clustering coefficient of node i ; w ij ,w jk ,wki is the edge weight; α is the angle weight coefficient; β is the time-varying weight coefficient; λ is the direction weight; φ ij is the spatial direction angle; τ is the time scale parameter; ω is the period parameter; γ is the angle adjustment coefficient; ψ ijk is the triangle included angle; δ is the connection threshold; μ is the density weight; ρ i , ρ j is the node density; By introducing angle features and time-varying features, the accuracy of clustering coefficient calculation is improved in this method.

[0113] Step S233: Calculate the eigenvector centrality for the pressure-coupled dynamic network to obtain eigenvector data; calculate the betweenness centrality for the pressure-coupled dynamic network to obtain betweenness feature data; calculate the importance distribution of the pressure-coupled dynamic network using the PageRank algorithm to obtain importance distribution data; comprehensively evaluate the eigenvector data, betweenness feature data, and importance distribution data to obtain centrality feature data.

[0114] The PageRank algorithm is specifically: The importance score PR( i ) = (1 - d)·[1 + α·S( i )] + d·Σ[PR( j )·w ji ·T( j , i )] / Σw ji ; where: The edge weight calculation w ji = exp(-||x j - x i ||² / 2σ²)·[1 + β·cos(θ ji )]; The time series transfer function T( j , i ) = exp(-|t j - t i |² / 2τ²)·[1 + λ·v( j , i )]; The node strength S( i ) = (k i , i n·k i ,out)^(1 / 2) / max(k); The velocity correlation v( j , i ) = cos(φ j - φ i )·exp(-|v j - v i | / v0); The local importance k i , i n = Σwji · [1 + μ·ρ( i )]; Local influence k i ,out = Σw ij · [1 + μ·ρ( i )]; Where: PR( i ) is the importance score of the node i ; d is the damping coefficient; α is the strength weight; β is the direction weight; θ ji is the spatial direction angle; τ is the time scale parameter; λ is the velocity weight; φ j , φ i is the flow direction angle; v j , v i is the flow velocity; v0 is the reference flow velocity; μ is the density weight; ρ( i ) is the node density; This method improves the accuracy of importance assessment by introducing time series transmission and flow field characteristics.

[0115] Step S234, extract the community structure of the pressure-coupled dynamic network at different time points to obtain time series community data; use the dynamic community detection algorithm to perform evolutionary analysis on the time series community data to obtain evolutionary feature data; extract the stability and variability features from the evolutionary feature data to obtain community evolution data.

[0116] Dynamic community detection algorithm: Community quality function Q(t) = Σ[w ij (t)·δ(c i (t), c j (t)) - P ij (t)]· [1 + α·M(t)]· [1 + β·E(t)]; Where: Time-varying weight w ij (t) = w 0ij · exp(-|t - t0|² / 2τ²)· [1 + λ·R( i , j , t)]; Null model P ij (t) = k i (t)· k j (t) / 2m(t)· [1 + γ·C( i , j , t )]; Community migration M(t) = Σ|c i (t) - c i (t - 1)| / N· exp(-ω·Δt); Community entropy E(t) = -Σ[ns(t) / N· log(ns(t) / N)]; Correlation feature R( i , j , t) = cos(θ ij)·exp(-||v i -v j || / v0); Aggregation feature C( i , j , t ) = |N i ∩N j | / |N i ∪N j |; Where: Q(t) is the time-varying modularity; w ij (t) is the time-varying edge weight; c i (t) is the community label of the node i at time t; α is the migration weight; β is the entropy weight; λ is the correlation weight; γ is the aggregation weight; τ is the time scale; ω is the time decay coefficient; ns(t) is the size of community s; N is the total number of nodes; θ ij is the direction angle; v i ,v j is the node velocity; v0 is the reference velocity; N i ,N j is the node neighborhood; This method improves the dynamic adaptability of community division by introducing time-varying features and migration features.

[0117] Step S235, perform temporal sorting on the node degree feature data to obtain temporal degree feature data; perform temporal sorting on the clustering feature data to obtain temporal clustering data; perform temporal sorting on the centrality feature data to obtain temporal centrality data; perform multi-dimensional integration on the temporal degree feature data, temporal clustering data, temporal centrality data, and community evolution data to obtain a network evolution feature set.

[0118] According to one aspect of the present application, step S32 is specifically:

[0119] Step S321, obtain a resilience status index set and perform empirical mode decomposition to obtain initial component data; use a change point detection algorithm to identify mutation features in the initial component data to obtain mutation feature data; perform component reconstruction on the initial component data and mutation feature data to obtain trend component data and fluctuation component data.

[0120] Step S322, perform trend extrapolation on the trend component data to obtain trend prediction data; perform fluctuation intensity analysis based on the fluctuation component data to obtain fluctuation intensity data; use a scenario generation algorithm to convert the trend prediction data and fluctuation intensity data into boundary conditions to obtain boundary scenario data; perform feasibility verification on the boundary scenario data to obtain scenario boundary data.

[0121] Step S323: Construct a two-dimensional hydrodynamic model calculation grid for the scenario boundary data to obtain calculation grid data; use the adaptive time step algorithm to perform dynamic solution on the calculation grid data to obtain hydrodynamic calculation data; extract the results from the hydrodynamic calculation data to obtain simulation result data; organize the simulation result data according to scenarios to obtain scenario simulation data.

[0122] Through the comprehensive method of time series decomposition and scenario simulation, multi-scenario prediction of the resilient state is realized. In this step, the empirical mode decomposition and change point detection algorithm are used to decompose the resilient state index to accurately identify the trend component and the fluctuation component. Through trend extrapolation and fluctuation intensity analysis, combined with the scenario generation algorithm, different boundary condition schemes are constructed. Finally, the adaptive time step algorithm is used for hydrodynamic simulation to obtain the response characteristics of the system under different scenarios. Through the prediction method combining time series analysis, scenario generation and hydrodynamic simulation, not only can the long-term evolution trend of the system be accurately depicted, but also the system response under different pressure scenarios can be simulated. Especially when evaluating the resilience of urban flood control systems, it can provide comprehensive scenario analysis results and provide a scientific basis for resilience improvement decisions.

[0123] According to one aspect of the present application, step S41 is specifically as follows:

[0124] Step S411: Perform numerical differentiation on the resilient state evolution model to obtain initial change rate data; use the singular value decomposition algorithm to perform noise reduction processing on the initial change rate data to obtain noise-reduced change rate data; perform time scale analysis on the noise-reduced change rate data to obtain state change rate data;

[0125] Singular value decomposition algorithm: Signal reconstruction function S(X) = U·[Σ + α·W(σ)]·Vᵀ·[1 + β·T(t)]; where: Adaptive weight matrix W(σ) = diag[w(σ1), w(σ2),..., w(σ r )]; Weight function w(σ i ) = exp(-|σ i - σ*|² / 2η²)·[1 + λ·h(σ i )]; Time series modulation T(t) = exp(-|t - t0|² / 2τ²)·cos(ω·|t - t0|); Entropy function h(σ i ) = -Σ[p i ·log(p i )]; Probability distribution p i = σ i / Σσ j; The noise threshold η = σ1·√(2·log(n)) / n; where: S(X) is the reconstructed signal; U and V are the left and right singular vectors; Σ is the singular value matrix; α is the weight coefficient; β is the time series weight; λ is the entropy weight; τ is the time scale; ω is the period parameter; σ i is the singular value; n is the number of samples; This method improves the accuracy of signal reconstruction by introducing adaptive weights and time series modulation.

[0126] Step S412: Identify outliers in the pre-stored real-time monitoring data to obtain cleaned monitoring data; perform maximum-minimum normalization on the cleaned monitoring data to obtain preliminary normalized data; use the quantile mapping method to adjust the distribution of the preliminary normalized data to obtain normalized monitoring data;

[0127] Quantile mapping method: The distribution transformation function Q(x) = F -1 γ [Fx(x)]·[1 + α·K(x)]·[1 + β·T(t)]; where: The empirical distribution function Fx(x) = (1 / n)·Σ[I(x i ≤x)·w(x i )]; The kernel density correction K(x) = Σ[Kh(x - x i )·exp(-|v i | / v0)]; The time series modulation T(t) = exp(-|t - t0|² / 2τ²)·cos(ω·|t - t0|); The adaptive weight w(x i ) = [1 + λ·ρ(x i )] / [1 + ||x - x i ||]; The kernel function Kh(u) = (1 / h)·K(u / h)·[1 + γ·|u / h|]; where: Q(x) is the mapping result; F -1 γ is the inverse function of the target distribution; α is the kernel density weight; β is the time series weight; λ is the density weight; τ is the time scale; ω is the period parameter; v i is the local change rate; v0 is the reference change rate; h is the bandwidth parameter; γ is the scale weight; This method improves the smoothness of distribution mapping by introducing kernel density correction and time series modulation.

[0128] Step S413: Construct a simplified Saint-Venant equation set based on the normalized monitoring data to obtain simplified equation data; use the feature extraction network to quickly extract features from the normalized monitoring data to obtain feature extraction data; quickly solve the simplified equation data and the feature extraction data to obtain fast simulation data;

[0129] The feature extraction network is specifically: the feature mapping function F(x) = σ[W·φ(x) + b]·[1 + α·A(x)]·[1 + β·R(x)]; where: the non-linear transformation φ(x) = tanh(ωx)·[1 + λ·||grad x||]; the attention mechanism A(x) = softmax(Q·K^T / √d)·[1 + μ·cos(θx)]; the residual connection R(x) = x + γ·g(x)·[1 + η·ρ(x)]; the local density ρ(x) = Σexp(-||x - x i ||² / 2σ²); the gradient feature ||grad x|| = √(Σ(dx / dx i )²); where: F(x) is the extracted feature; σ is the activation function; W, b are network parameters; α is the attention weight; β is the residual weight; λ is the gradient weight; μ is the direction weight; γ is the residual coefficient; η is the density weight; σ is the kernel width; by introducing the attention mechanism and the residual connection, the efficiency of feature extraction is improved.

[0130] Step S414: Perform multi-scale decomposition on the state change rate data and the fast simulation data to obtain decomposed fluctuation data; use random matrix theory to extract features from the decomposed fluctuation data to obtain fluctuation feature data; reconstruct the fluctuation feature data to obtain dynamic fluctuation feature data;

[0131] The random matrix theory is specifically: the feature extraction function E(X) = Σ[λ i ·v i ·v i T ·I(λ i > λc)]·[1 + α·H(λ)]·[1 + β·T(t)]; where: the critical threshold λc = μ(1 + √(n / p))²·[1 + γ·σ(λ)]; the entropy correction H(λ) = -Σ[p i ·log(p i )]·exp(-|λ i -λ*| / λ0); the time series modulation T(t) = exp(-|t - t0|² / 2τ²)·cos(ω·|t - t0|); the feature probability p i = λ i / Σλ j ; the local variance σ(λ) = √[Σ(λ i -λ*)²·w(λ i )] / n; the weight function w(λ i ) = [1 + η·ρ(λ i )] / [1 + |λ i-λ*|]; where: E(X) is the extracted feature matrix; λ i is the eigenvalue; v i is the eigenvector; α is the entropy weight; β is the time series weight; γ is the variance weight; μ is the mean; λ0 is the reference eigenvalue; τ is the time scale; ω is the period parameter; η is the density weight; n, p are the matrix dimensions; By introducing entropy correction and time series modulation, the stability of feature extraction is improved.

[0132] Step S415: Calculate the autocorrelation function of the normalized monitoring data to obtain autocorrelation feature data; calculate the cross-correlation function of the normalized monitoring data to obtain cross-correlation feature data; fuse the autocorrelation feature data and the cross-correlation feature data to obtain time-related data;

[0133] Step S416: Extract the stability index from the dynamic fluctuation feature data to obtain stability index data; extract the correlation index from the time-related data to obtain correlation index data; use the deep feature fusion network to fuse the stability index data and the correlation index data to obtain the critical feature index set.

[0134] The deep feature fusion network is specifically: the feature fusion function F(X) = σ[W·(X + α·M(X)) + b]·[1+ β·A(X)]·[1 + λ·R(X)]; where: the multi-scale feature M(X) = Σ[w i ·Conv(X,k i )]·[1 + γ·S(X)]; the attention mechanism A(X) = softmax(Q·K^T / √d)·[1 + μ·cos(θx)]; the residual term R(X) = G(X)- X + η·L(X); the scale correlation S(X) = exp(-|s i -s j |² / 2σs²); the local feature L(X) = Σ[φ(x i )·exp(-||x-x i ||² / 2σ²)]; where: F(X) is the fusion feature; σ is the activation function; W, b are the network parameters; α is the multi-scale weight; β is the attention weight; λ is the residual weight; w i is the convolution weight; k i is the convolution kernel; γ is the scale weight; μ is the direction weight; η is the local weight; σs is the scale parameter; σ is the spatial parameter; G(X) is the non-linear transformation; By introducing multi-scale features and attention mechanisms, the effect of feature fusion is improved.

[0135] Through the multi-dimensional extraction and fusion method of critical features, the accurate characterization of system state changes is achieved. In this step, numerical differentiation is performed on the resilience state evolution model, and noise reduction is carried out using the singular value decomposition algorithm to obtain the accurate state change rate. Outlier identification and normalization are performed on the real-time monitoring data, and distribution adjustment is carried out through the quantile mapping method to ensure the quality of the monitoring data. In terms of rapid simulation, the rapid evaluation of the system state is realized through the simplified Saint-Venant equations and the feature extraction network. Finally, the fluctuation features are extracted through multi-scale decomposition and random matrix theory, and the time-related features are obtained by combining autocorrelation and cross-correlation analysis. The integration of features is achieved using the deep feature fusion network. Through the multi-dimensional critical feature extraction method, not only can the dynamic changes of the system state be accurately captured, but also the early signals of critical transitions can be effectively identified. Especially when monitoring the resilience state of the urban flood control system, potential risk factors can be detected in a timely manner, providing a reliable basis for early warning decisions.

[0136] According to one aspect of the present application, step S421 is further as follows:

[0137] Calculate the basic statistics for the critical feature index set to obtain the basic statistical data; use the kernel density estimation method to extract the distribution characteristics of the critical feature index set to obtain the empirical distribution data; perform distribution fitting on the basic statistical data and the empirical distribution data to obtain the distribution characteristic data.

[0138] The kernel density estimation method is specifically: probability density function f(x) = [1 / (nh(x))]·Σ[K((x - x i ) / h(x))·w(x i )]·[1 + α·L(x)]·[1 + β·T(t)]; where: adaptive bandwidth h(x) = h0·[f0(x) / g]^(-γ)·[1 + λ·v(x)]; local weight w(x i ) = exp(-||x - x i ||² / 2σ²)·[1 + μ·ρ(x i )]; time series feature T(t) = exp(-|t - t0|² / 2τ²)·cos(ω·|t - t0|); local variation v(x) = std(Nx) / mean(Nx); spatial correlation L(x) = Σ[exp(-d i ² / 2η²)·cos(θ i )]; density feature ρ(x i ) = Σexp(-||x i - x j||² / 2ξ²); where: f(x) is the estimated density; n is the number of samples; h(x) is the adaptive bandwidth; K is the kernel function; α is the spatial weight; β is the temporal weight; h0 is the reference bandwidth; f0 is the prior density; γ is the adaptive exponent; λ is the variation weight; μ is the density weight; σ, τ, η, ξ are scale parameters; ω is the period parameter; this method improves the accuracy of density estimation by introducing the adaptive bandwidth and spatio-temporal features.

[0139] Step S422: Estimate the parameters of the distribution feature data to obtain distribution parameter data; use the mixture Gaussian model to perform probability density modeling on the distribution feature data to obtain probability model data; perform function reconstruction on the distribution parameter data and the probability model data to obtain density function data.

[0140] The mixture Gaussian model is specifically: the probability model function P(x) = Σ[π i ·N(x|μ i ,Σ i )]·[1 + α·M(x)]·[1 + β·T(t)]; where: the adaptive weight π i = exp(-E i / θ)·[1 + λ·V i (x)] / Σexp(-E j / θ); the covariance matrix Σ i = S i + γ·R i ; the mixture metric M(x) = -Σ[p i ·log(p i )]·exp(-||x-μ*||² / 2σ²); the temporal modulation T(t) = exp(-|t-t0|² / 2τ²)·cos(ω·|t-t0|); the local variance V i (x) = Σ[(x-μ i )ᵀΣ i ⁻¹(x-μ i )]·w(x); where: P(x) is the mixture probability; N(·) is the Gaussian distribution; π i is the mixture weight; μ i is the mean vector; α is the mixture weight; β is the temporal weight; λ is the variance weight; θ is the temperature parameter; γ is the regularization coefficient; σ, τ are scale parameters; ω is the period parameter; this method improves the fitting ability of the model by introducing the adaptive weight and the temporal modulation.

[0141] Step S423: Calculate the cumulative distribution function for the density function data to obtain cumulative distribution data; perform quantile calculation on the cumulative distribution data to obtain preliminary quantile data; use the quantile regression method to correct the preliminary quantile data to obtain quantile feature data.

[0142] The quantile regression method is specifically as follows: The quantile function Q(τ|x) = argmin{Σ[ρτ(y - f(x,β))·w(x)] + α·||β||1}·[1 + λ·S(x)]·[1 + μ·T(t)]; where: The loss function ρτ(u) = u·(τ - I(u < 0)); The sample weight w(x) = exp(-||x - x*||² / 2σ²)·[1 + γ·ρ(x)]; The spatial feature S(x) = Σ[exp(-d i ² / 2η²)·cos(θ i )]; The temporal feature T(t) = exp(-|t - t0|² / 2τ²)·cos(ω·|t - t0|); The local density ρ(x) = Σexp(-||x - x i ||² / 2ξ²); where: Q(τ|x) is the τ quantile; f(x,β) is the regression function; α is the regularization parameter; λ is the spatial weight; μ is the temporal weight; γ is the density weight; σ, η, τ, ξ are scale parameters; ω is the period parameter; This method improves the robustness of quantile estimation by introducing weighted loss and spatio-temporal features.

[0143] Step S424: Conduct threshold sensitivity analysis on the quantile feature data to obtain sensitivity data; use the maximum inter-class variance method to optimize the segmentation point for the quantile feature data and the sensitivity data to obtain optimized segmentation data; perform feasibility verification on the optimized segmentation data to obtain segmentation point data.

[0144] The maximum inter-class variance method is specifically as follows: The segmentation function V(t) = [ω1(t)·ω2(t)·(μ1(t) - μ2(t))²]·[1 + α·H(t)]·[1 + β·S(t)]; where: The within-class weight ω i (t) = Σ[p(x)·I(x ∈ C i )]·[1 + λ·v i (t)]; The class mean μ i (t) = Σ[x·p(x)·I(x ∈ C i )] / ω i (t); The entropy feature H(t) = -Σ[p i ·log(p i)]·exp(-|t - t*|² / 2τ²); The smoothing term S(t) = exp(-|t - t0|² / 2σ²)·cos(ω·|t - t0|); The within-class variance v i (t) = Σ[(x - μ i (t))²·p(x)·I(x ∈ C i )] / ω i (t); where: V(t) is the between-class variance; ω i (t) is the class weight; μ i (t) is the class mean; α is the entropy weight; β is the smoothing weight; λ is the variance weight; p(x) is the probability density; C i is the category; τ, σ are scale parameters; ω is the period parameter; By introducing entropy features and a smoothing term, this method improves the stability of the selection of segmentation points.

[0145] Step S425: Perform hierarchical assignment on the segmentation point data to obtain early warning level data; conduct uncertainty analysis on the segmentation point data to obtain threshold interval data; integrate the early warning level data and the threshold interval data to obtain an early warning threshold set.

[0146] Through a multi-level early warning threshold optimization method, a scientific classification of early warning levels is achieved. This step performs parameter estimation and distribution fitting on the critical feature index set, constructs a probability density function using a mixture Gaussian model, and accurately depicts the distribution characteristics of the indices. Through the calculation of the cumulative distribution function and the quantile regression method, robust quantile features are obtained. In terms of threshold determination, the segmentation points are optimized through threshold sensitivity analysis and the maximum between-class variance method, and feasibility verification is carried out to ensure the rationality of the thresholds. Finally, through hierarchical assignment and uncertainty analysis, a complete early warning system including early warning levels and threshold intervals is constructed. The threshold determination strategy based on statistical theory and optimization methods not only considers the distribution characteristics of the indices but also takes into account the accuracy and practicality of early warning. Especially in the early warning practice of urban flood control systems, it can effectively reduce the false alarm rate and missed alarm rate and improve the reliability of the early warning system.

[0147] According to another aspect of the present application, there is also provided an urban flood control resilience critical state identification system considering multiple stress couplings, including:

[0148] At least one processor; and,

[0149] A memory communicatively connected to the at least one processor; wherein,

[0150] The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the method for identifying the critical state of urban flood control resilience considering multiple stress couplings according to any one of the above technical solutions.

[0151] It should be noted that, in the above specific embodiments, the various specific technical features described can be combined in any appropriate manner without conflict. To avoid unnecessary repetition, the present invention will not separately describe various possible combination manners.

Claims

1. A method for identifying the critical state of urban flood control resilience considering multiple stress couplings, characterized in that It includes the following steps: Step S1: According to the characteristics of the urban flood control system, identify the stress sources affecting the resilience and performance of the urban flood control system; obtain research data and perform unified spatio-temporal preprocessing to obtain a unified spatio-temporal scale dataset; perform outlier detection and correction on the unified spatio-temporal scale dataset to obtain a corrected dataset; Perform stress feature decomposition on the corrected dataset to obtain a stress feature vector set including time dimension features, space dimension features, and intensity dimension features; Step S2: Obtain the stress feature vector set and perform coupling relationship analysis to obtain a coupling strength matrix and a coupling direction matrix; construct a directed weighted network based on the coupling strength matrix and the coupling direction matrix to obtain a stress coupling dynamic network; Perform topological structure analysis on the stress coupling dynamic network to obtain a network evolution feature set; Step S3: Conduct comprehensive analysis on the network evolution feature set and the pre-stored historical resilience evaluation data to obtain a resilience state index set; perform time series decomposition and scenario simulation on the resilience state index set to obtain scenario simulation data; construct a state evolution prediction model based on the scenario simulation data and the resilience state index set to obtain a resilience state evolution model; Step S4: Based on the resilience state evolution model and the pre-stored real-time monitoring data, conduct critical feature analysis to obtain a critical feature index set; perform threshold optimization calculation according to the critical feature index set to obtain an early warning threshold set; Specifically, Step S1 is as follows: Step S11: According to the characteristics of the urban flood control system, identify the main stress sources affecting the resilience and performance of the urban flood control system; including meteorological stress sources, hydrological stress sources, engineering stress sources, and socio-economic stress sources; meteorological stress sources include rainfall intensity and rainfall duration; hydrological stress sources include water level changes and flow fluctuations in rivers or drainage systems; engineering stress sources include the state of sluice pumps, drainage system capacity, and pipe network state; socio-economic stress sources include population density, economic activity intensity, and traffic flow; Step S12: Obtain research data, including original meteorological data, original hydrological data, original engineering data, and original socio-economic data; extract rainfall intensity and duration data from the original meteorological data, water level and flow data from the original hydrological data, the state of sluice pumps, drainage system capacity, and pipe network state data from the original engineering data, and population density and asset distribution data from the original socio-economic data, perform Kriging interpolation calculation to obtain temporary spatial interpolation data; perform linear interpolation on the temporary spatial interpolation data according to the time series to obtain temporary time interpolation data; calculate the time weight function according to the reciprocal of the data time interval, and perform weight correction on the temporary time interpolation data to obtain a unified spatio-temporal scale dataset; Step S13: Read the unified spatio-temporal scale dataset and calculate the mean and standard deviation to obtain statistical feature data; Use the box plot method to preliminarily identify outliers in the unified spatio-temporal scale dataset to obtain preliminary outlier data; compare the preliminary outlier data with physical constraint conditions to obtain confirmed outlier data; Use the local polynomial regression method to correct the confirmed outlier data to obtain corrected data; Merge the corrected data with the normal value data in the dataset with unified spatio-temporal scale to obtain the corrected dataset; Step S14: Extract the rainfall data in the corrected dataset and perform rainfall-runoff simulation to obtain runoff simulation data; Use a one-dimensional / two-dimensional coupled hydrodynamic model to perform evolution calculation on the runoff simulation data to obtain flood inundation data; Merge the flood inundation data with the corrected dataset to obtain hydrodynamic characteristic data; Step S15: Perform wavelet transform on the hydrodynamic characteristic data to obtain a wavelet coefficient set; Perform threshold decomposition on the wavelet coefficient set according to the energy distribution characteristics to obtain characteristic component data at different scales; Recombine the characteristic component data according to the time dimension, space dimension, and intensity dimension to obtain multi-dimensional characteristic data; Perform standardization processing on the multi-dimensional characteristic data to obtain standardized characteristic data; Convert the standardized characteristic data into vector form to obtain a pressure characteristic vector set.

2. The method for identifying the critical state of urban flood control resilience considering multi-pressure coupling according to claim 1, wherein Step S2 is specifically as follows: Step S21: Obtain and calculate the mutual information entropy for each pair of characteristic vectors in the pressure characteristic vector set to obtain a mutual information entropy matrix; Calculate the coupling strength value according to the mutual information entropy matrix to obtain a coupling strength matrix; Calculate the Granger causality test statistic for each pair of characteristic vectors in the pressure characteristic vector set to obtain causality test data; Determine the coupling direction according to the causality test data to obtain a coupling direction matrix; Step S22: Construct network edge weights based on the coupling strength matrix to obtain a weighted edge set; Determine the direction of network edges based on the coupling direction matrix to obtain a directed edge set; Combine the weighted edge set and the directed edge set and add time marks to obtain a pressure coupling dynamic network; Step S23: Calculate the node degree distribution of the pressure coupling dynamic network to obtain node degree characteristic data; Calculate the clustering coefficient of the pressure coupling dynamic network to obtain clustering characteristic data; Calculate the centrality index of the pressure coupling dynamic network to obtain centrality characteristic data; Calculate the community structure change of the pressure coupling dynamic network at different time points to obtain community evolution data; Combine the node degree characteristic data, clustering characteristic data, centrality characteristic data, and community evolution data to obtain a network evolution characteristic set.

3. The method for identifying the critical state of urban flood control resilience considering multiple pressure couplings according to claim 2, characterized in that, Step S3 is specifically as follows: Step S31: Perform principal component analysis on the network evolution characteristic set to obtain principal component characteristic data; Perform standardization processing on the pre-stored historical resilience assessment data to obtain standardized resilience data; Perform correlation analysis on the principal component characteristic data and the standardized resilience data to obtain a characteristic correlation matrix; Select key characteristic combinations according to the characteristic correlation matrix to obtain a resilience characteristic candidate set; Perform weighted combination on the resilience characteristic candidate set to obtain a resilience state index set; Step S32: Perform time series decomposition on the resilience state index set to obtain trend component data and fluctuation component data; Set different boundary conditions according to the trend component data and the fluctuation component data to obtain scenario boundary data; Use the hydrodynamic model to perform simulation calculation on the scenario boundary data to obtain scenario simulation data; Step S33: Construct an autoregressive model for the scenario simulation data and the resilience state index set to obtain trend prediction data; Calculate the phase space reconstruction parameters based on the scenario simulation data to obtain the reconstructed parameter data; construct the phase space according to the reconstructed parameter data to obtain the state space data; combine the trend prediction data and the state space data to construct a hybrid prediction model to obtain the resilience state evolution model.

4. The method for identifying the critical state of urban flood control resilience considering multi-pressure coupling according to claim 3, wherein Step S4 is specifically as follows: In step S41, calculate the system state change rate of the resilience state evolution model to obtain the state change rate data; normalize the pre-stored real-time monitoring data to obtain the normalized monitoring data; use the simplified hydrodynamic model to quickly simulate the normalized monitoring data to obtain the quick simulation data; perform fluctuation analysis on the state change rate data and the quick simulation data to obtain the dynamic fluctuation characteristic data; calculate the time correlation of the normalized monitoring data to obtain the time correlation data; Fuse the dynamic fluctuation characteristic data and the time correlation data to obtain the critical characteristic index set; In step S42, perform statistical distribution analysis on the critical characteristic index set to obtain the distribution characteristic data; construct a probability density function based on the distribution characteristic data to obtain the density function data; perform quantile analysis on the density function data to obtain the quantile characteristic data; calculate the optimal segmentation point according to the quantile characteristic data to obtain the segmentation point data; convert the segmentation point data into warning thresholds to obtain the warning threshold set.

5. The method for identifying the critical state of urban flood control resilience considering multiple pressure couplings according to claim 4, characterized in that Step S14 is specifically as follows: In step S141, extract the rainfall data from the corrected dataset and perform spatio-temporal distribution analysis to obtain the rainfall distribution characteristic data; calculate the infiltration loss using the Green-Ampt model to obtain the infiltration loss data; calculate the surface interception loss using the surface roughness data to obtain the interception loss data; perform water balance calculation on the rainfall distribution characteristic data, the infiltration loss data and the interception loss data to obtain the effective rainfall data; perform runoff calculation on the effective rainfall data using the nonlinear unit hydrograph method to obtain the runoff simulation data; In step S142, perform one-dimensional river channel hydrodynamic calculation based on the runoff simulation data to obtain the river channel hydrodynamic data; use the two-dimensional overland flow model to perform urban surface evolution calculation on the runoff simulation data to obtain the overland flow data; use the interactive iteration algorithm to perform coupling calculation on the river channel hydrodynamic data and the overland flow data to obtain the one-two dimensional coupling data; calculate the flood inundation range according to the pre-stored terrain elevation data for the one-two dimensional coupling data to obtain the flood inundation data; In step S143, perform spatio-temporal feature extraction on the corrected dataset to obtain the basic feature data; perform hydrodynamic index calculation on the flood inundation data to obtain the inundation characteristic data; use the multi-source data fusion algorithm to perform feature combination on the basic feature data and the inundation characteristic data to obtain the combined feature data; use the adaptive weight method to perform feature optimization on the combined feature data to obtain the hydrodynamic feature data; In step S144, perform quality verification on the hydrodynamic feature data to obtain the feature verification data; use the physical constraint conditions to perform validity test on the feature verification data to obtain the feature test data; Use the multi-criteria evaluation method to perform reliability assessment on the feature test data to obtain the feature assessment data; Compare and correct the feature evaluation data with the original hydrodynamic feature data to obtain the final hydrodynamic feature data.

6. The method for identifying the critical state of urban flood control resilience considering multi-pressure coupling according to claim 4, characterized in that, Step S21 is specifically as follows: Step S211: Calculate the mutual information entropy for each pair of feature vectors in the pressure feature vector set to obtain preliminary mutual information data; use the kernel density estimation method to correct the preliminary mutual information data to obtain corrected mutual information data; Perform matrix reconstruction on the corrected mutual information data to obtain a mutual information entropy matrix; Step S212: Obtain the mutual information entropy matrix and perform normalization processing to obtain normalized mutual information data; Use the non-linear mapping method to perform intensity conversion on the normalized mutual information data to obtain intensity mapping data; Perform matrix reconstruction on the intensity mapping data to obtain a coupling intensity matrix; Step S213: Perform time-delay analysis on each pair of feature vectors in the pressure feature vector set to obtain time-delay feature data; Perform Granger causality test on the pressure feature vector set and the time-delay feature data to obtain preliminary causality data; use the confidence test method to perform significance analysis on the preliminary causality data to obtain causality test data; Step S214: Perform threshold segmentation on the causality test data to obtain causality determination data; perform directional analysis on the causality determination data to obtain direction feature data; Perform matrix reconstruction on the direction feature data to obtain a coupling direction matrix.

7. The method for identifying the critical state of urban flood control resilience considering multi-pressure coupling according to claim 4, wherein Step S32 is specifically as follows: Step S321: Obtain the toughness state index set and perform empirical mode decomposition to obtain initial component data; use the change point detection algorithm to identify mutation features in the initial component data to obtain mutation feature data; Perform component reconstruction on the initial component data and the mutation feature data to obtain trend component data and fluctuation component data; Step S322: Perform trend extrapolation on the trend component data to obtain trend prediction data; Perform fluctuation intensity analysis based on the fluctuation component data to obtain fluctuation intensity data; Use the scenario generation algorithm to convert the trend prediction data and the fluctuation intensity data into boundary conditions to obtain boundary scheme data; perform feasibility verification on the boundary scheme data to obtain scenario boundary data; Step S323: Construct a two-dimensional hydrodynamic model calculation grid for the scenario boundary data to obtain calculation grid data; use the adaptive time step algorithm to perform dynamic solution on the calculation grid data to obtain hydrodynamic calculation data; extract the results from the hydrodynamic calculation data to obtain simulation result data; organize the simulation result data according to scenarios to obtain scenario simulation data.

8. The method for identifying the critical state of urban flood control resilience considering multi-pressure coupling according to claim 4, wherein Step S41 is specifically as follows: Step S411: Perform numerical differentiation on the toughness state evolution model to obtain initial change rate data; use the singular value decomposition algorithm to perform noise reduction processing on the initial change rate data to obtain noise-reduced change rate data; perform time-scale analysis on the noise-reduced change rate data to obtain state change rate data; Step S412: Identify outliers in the pre-stored real-time monitoring data to obtain cleaned monitoring data; Perform maximum-minimum normalization on the cleaned monitoring data to obtain preliminary normalized data; use the quantile mapping method to adjust the distribution of the preliminary normalized data to obtain normalized monitoring data; Step S413: Construct a simplified Saint-Venant equation set based on the normalized monitoring data to obtain simplified equation data; Use a feature extraction network to quickly extract features from the normalized monitoring data to obtain feature extraction data; quickly solve the simplified equation data and the feature extraction data to obtain fast simulation data; Step S414: Perform multi-scale decomposition on the state change rate data and the fast simulation data to obtain decomposed fluctuation data; use random matrix theory to extract features from the decomposed fluctuation data to obtain fluctuation feature data; Reconstruct the fluctuation feature data to obtain dynamic fluctuation feature data; Step S415: Calculate the autocorrelation function of the normalized monitoring data to obtain autocorrelation feature data; calculate the cross-correlation function for the normalized monitoring data to obtain cross-correlation feature data; perform feature fusion on the autocorrelation feature data and the cross-correlation feature data to obtain time-related data; Step S416: Extract stability indicators from the dynamic fluctuation feature data to obtain stability indicator data; extract correlation indicators from the time-related data to obtain correlation indicator data; use a deep feature fusion network to perform feature fusion on the stability indicator data and the correlation indicator data to obtain a critical feature index set.

9. An urban flood control resilience critical state identification system considering multi-pressure coupling, characterized in that, Comprising: At least one processor; And, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the method for identifying the critical state of urban flood control resilience considering multiple pressure couplings according to any one of claims 1 to 8.

Images