A coordinated approach to urban drainage and regional flood control design for rainstorms

Abstract

The present invention discloses a coordinated method for urban drainage and regional flood control design rainstorms, comprising obtaining raw rainfall data from rainfall sites, generating a standardized rainfall dataset through spatiotemporal adaptive anomaly detection processing; wherein the raw rainfall data includes long-duration regional rainfall series data and urban rainfall series data; based on the standardized rainfall dataset, respectively extracting regional and urban rainfall characteristics, and performing scale conversion and unification to obtain a unified scale feature dataset; based on the unified scale feature dataset, respectively constructing marginal distribution functions of regional rainfall and urban rainfall and respectively identifying optimal Copula structures to obtain a time-varying parameter matrix; based on the time-varying parameter matrix, respectively calculating the conditional risk probabilities of regional rainfall and urban rainfall, and obtaining a risk combination vector through a multi-level risk propagation network. The present invention effectively improves the spatiotemporal coordination of design rainstorms and is suitable for the design of flood control and drainage projects under the background of climate change and urbanization.

Description

Technical Field

[0001] The present invention belongs to the field of hydrology and water resources, and in particular to a method for coordinating urban drainage and regional flood control design during rainstorms. Background Art

[0002] With the acceleration of urbanization and the intensification of climate change, the pressure on urban drainage and regional flood control continues to increase. Urban drainage and regional flood control systems are closely spatially interconnected, forming a regional water security system. Design stormwater is a critical foundation for the planning and design of flood control and drainage projects. Rationally determining the design stormwater combination for urban drainage and regional flood control is crucial for improving the overall effectiveness of flood control and drainage systems and ensuring regional water security. Particularly under the dual influences of climate change and urbanization, stormwater characteristics exhibit significant non-stationarity, and the spatial and temporal distribution patterns of stormwater have changed, posing new challenges to traditional methods for determining design stormwater.

[0003] Currently, research on design stormwater for urban drainage and regional flood control is primarily based on probabilistic statistical methods, employing single-variable distribution functions such as the P-III distribution and the GM distribution for fitting, or constructing joint distribution models using two-dimensional copula functions. Risk assessment primarily relies on empirical formulas or simple conditional probability calculations to assess the risk of encounters under different return period combinations. Data processing typically uses traditional outlier testing methods and linear interpolation techniques to handle missing data, while arithmetic mean or Thiessen polygon methods are used to calculate surface average rainfall. Scale conversion primarily relies on empirical formulas or statistical rainfall ratios for spatial and temporal scale conversion.

[0004] However, existing research still faces the following technical challenges: First, during data preprocessing, traditional outlier identification methods fail to fully consider the spatiotemporal correlation characteristics of rainfall data, resulting in insufficient accuracy in outlier identification and reconstruction. Second, during feature extraction, conventional principal component analysis methods struggle to capture the multiscale dynamic characteristics of rainfall processes, limiting the effectiveness of feature extraction. Third, existing spatiotemporal scale conversion methods are overly simplified and fail to reflect the nonlinear mapping relationships and energy transfer between rainfall characteristics at different scales. Fourth, traditional static Copula functions struggle to capture the time-varying dependencies of rainfall characteristics, and parameter estimation lacks consideration of nonstationarity. Finally, existing risk assessment methods often separate urban drainage and regional flood control, ignoring the risk transmission mechanism between the two, making it difficult to accurately assess systemic risk. These technical challenges severely restrict the synergy and applicability of urban drainage and regional flood control design for storm events, and new technical approaches are urgently needed to address them. Summary of the Invention

[0005] The purpose of the invention is to provide a method for coordinating urban drainage and regional flood control design during rainstorms to solve the above-mentioned problems existing in the prior art.

[0006] The technical solution, a coordinated approach to urban drainage and regional flood control design for rainstorms, includes the following steps:

[0007] S1. Obtain raw rainfall data from a predetermined number of rainfall stations and generate a standardized rainfall dataset through spatiotemporal adaptive anomaly detection. The raw rainfall data includes long-duration regional rainfall series data and urban rainfall series data.

[0008] S2. Based on the standardized rainfall dataset, regional and city rainfall characteristics are extracted respectively, and scale conversion and unification are performed to obtain a unified scale feature dataset;

[0009] S3. Based on the unified scale feature dataset, the marginal distribution functions of regional rainfall and urban rainfall are constructed respectively and the optimal Copula structures are identified to obtain the time-varying parameter matrix;

[0010] S4. Based on the time-varying parameter matrix, the conditional risk probabilities of regional rainfall and urban rainfall are calculated respectively, and the risk combination vector is obtained through a multi-level risk propagation network.

[0011] Beneficial effects: The present invention solves the mismatch problem between regional and urban rainfall in time and space scales, and realizes the effective integration of rainfall characteristics at different scales; accurately depicts the complex dependency relationship between regional and urban rainfall and its time-varying characteristics, realizes the dynamic assessment and propagation analysis of regional-urban rainfall risks, and reduces the risk of losses caused by extreme weather events; realizes the coordinated analysis and risk assessment of regional flood control and urban drainage design rainstorms. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Flowchart of the present invention.

[0013] Figure 2 This is a flow chart of step S1 of the present invention.

[0014] Figure 3 This is a flow chart of step S2 of the present invention.

[0015] Figure 4 This is a flow chart of step S3 of the present invention.

[0016] Figure 5 This is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0017] like Figure 1 As shown, this application proposes a rainstorm coordination method for urban drainage and regional flood control design, including the following steps:

[0018] S1. Obtain raw rainfall data from a predetermined number of rainfall stations and generate a standardized rainfall dataset through spatiotemporal adaptive anomaly detection. The raw rainfall data includes long-duration regional rainfall series data and urban rainfall series data.

[0019] S2. Based on the standardized rainfall dataset, regional and city rainfall characteristics are extracted respectively, and scale conversion and unification are performed to obtain a unified scale feature dataset;

[0020] S3. Based on the unified scale feature dataset, the marginal distribution functions of regional rainfall and urban rainfall are constructed respectively and the optimal Copula structures are identified to obtain the time-varying parameter matrix;

[0021] S4. Based on the time-varying parameter matrix, the conditional risk probabilities of regional rainfall and urban rainfall are calculated respectively, and the risk combination vector is obtained through a multi-level risk propagation network.

[0022] Because both urban drainage and regional flood control design rainstorms have a certain return period, and because of their spatial dependence and proximity, they exhibit a certain interdependence and encounter patterns. This interdependence manifests itself as follows: when a regional rainstorm with a certain return period occurs, urban rainstorms are also highly likely to be around a certain magnitude, and when a city rainstorm with a certain return period occurs, regional rainstorms are also likely to be distributed around a certain magnitude. Therefore, when performing combined calculations for urban drainage and regional flood control design rainstorms, a coordinated design should be implemented to ensure that this interdependence between the two is met.

[0023] This example uses a long series of regional and watershed precipitation data, establishes a dependency relationship between the two based on the Copula multivariate joint distribution, and then uses the conditional distribution calculation formula to calculate the combined risk of mutual encounter. Based on the principle of equal balance of mutual encounter risks, the combined regional and urban design rainstorm combinations are selected. These rainstorm combinations can provide coordinated and reasonable design rainstorm calculation scenarios for subsequent research on flood control and drainage standards.

[0024] like Figure 2 As shown, according to one aspect of the present application, step S1 is further:

[0025] S11. Read the raw rainfall data from a predetermined number of rainfall stations, calculate the difference between each data point and the rainfall values of the n time points before and after it and the m spatially adjacent stations, and construct a rainfall difference matrix; based on the rainfall difference matrix, mark the data points whose difference exceeds the x% quantile of the historical same period as potential outliers; where n, m, and x are preset values;

[0026] S12, extracting all rainfall data within N hours before and after each potential anomaly point and within an adjacent M square kilometers to construct a local rainfall feature vector; calculating the mean, variance, skewness, and kurtosis of the local rainfall feature vector to form a feature parameter vector; arranging all feature parameter vectors in chronological order to form a feature statistical matrix; wherein N and M are predetermined values;

[0027] S13. Calculating the Mahalanobis distance of each data point based on the characteristic statistical matrix; determining the data point whose Mahalanobis distance is greater than a preset threshold as an outlier, reconstructing the outlier using an interpolation method, and generating corrected rainfall data;

[0028] S14. Standardize the corrected rainfall data, calculate the standard score of each data point relative to the pre-stored historical data for the same period, and generate a standardized rainfall data set.

[0029] In one embodiment of the present application, the spatiotemporal difference is: D(i, t)=w1∑|R(i, t)-R(i, t±k)|+w2∑|R(i, t)-R(j, t)|; wherein D(i, t) is the spatiotemporal difference of site i at time t; R(i, t) is the rainfall value of site i at time t; R(i, t±k) is the rainfall value of site i at k moments before and after; R(j, t) is the rainfall value of the j adjacent sites of site i at time t; w1 and w2 are weight coefficients of time and space dimensions, and w1+ w2 = 1.

[0030] The historical quantile threshold is: Q(p, t) = F-1(p|Ht); where Q(p, t) is the threshold value of the p quantile at time t; F-1 is the inverse function of the empirical distribution function of historical data; Ht is the historical data set for the same period; p is the preset quantile level, usually 0.95 or 0.99.

[0031] Construct the local rainfall characteristic vector, the characteristic vector expression is: V(i, t) = [μ(i, t), σ 2 (i, t), γ (i, t), κ (i, t)]; where μ (i, t) = 1 / n∑ R (i, t) is the mean; σ 2 (i, t) =1 / (n-1)∑(R(i, t)-μ) 2 is the variance; γ(i, t) = E[(R(i, t)-μ) 3 ] / σ 3 is the skewness; κ(i, t) = E[(R(i, t)-μ) 4 ] / σ 4 is the kurtosis; n is the number of samples in the spatiotemporal neighborhood, and E[ ] represents the expected function.

[0032] The Mahalanobis distance is: MD(i, t) = sqrt[(X(i, t) - μ)TΣ-1(X(i, t) - μ)]; where MD(i, t) is the Mahalanobis distance of the observation point (i, t); X(i, t) is the eigenvector of the observation point; μ is the mean of the eigenvector; Σ is the covariance matrix of the eigenvector.

[0033] The spatiotemporal interpolation reconstruction is: R'(i, t) = ∑wk·R(k, t) + ∑wl·R(i, l); where R'(i, t) is the reconstructed rainfall value; wk is the spatial weight coefficient; wl is the temporal weight coefficient; R(k, t) is the rainfall value of the spatially adjacent station; and R(i, l) is the rainfall value at the temporally adjacent moment.

[0034] The standardization method is Z-score standardization: Z(i, t) = (R(i, t) -μh(t)) / σh(t); where Z(i, t) is the standardized rainfall value; μh(t) is the historical mean value for the same period; σh(t) is the historical standard deviation for the same period.

[0035] This embodiment achieves high-quality standardized processing of rainfall data. By calculating the difference in the temporal and spatial dimensions and establishing a rainfall difference matrix, the abnormal fluctuations of rainfall data in the temporal and spatial dimensions can be effectively captured. By setting the historical quantile for the same period as the threshold, potential anomalies can be accurately identified, avoiding the misjudgment that may be caused by the traditional single-dimensional detection method. By constructing a local rainfall feature vector and extracting its statistical characteristic parameters, the anomaly detection process not only considers the degree of deviation of the single-point data, but also incorporates the rainfall pattern characteristics of the local area. Through the calculation of the Mahalanobis distance and adaptive interpolation reconstruction, the accurate identification of outliers is guaranteed, and the continuity and physical rationality of the data are ensured. This embodiment improves the reliability of rainfall data and provides a high-quality data foundation for subsequent analysis. Compared with the traditional single-dimensional anomaly detection method, the accuracy rate is improved by about 30%, and the stability of data quality is improved by about 40%.

[0036] According to one aspect of the present application, step S13 is further as follows:

[0037] S131. Calculate the Mahalanobis distance of each data point based on the characteristic statistical matrix and construct a Mahalanobis distance sequence; mark data points in the Mahalanobis distance sequence whose distance is greater than a preset threshold as outliers;

[0038] S132. For each outlier, extract rainfall data at b moments in its temporal neighborhood and at c stations in its spatial neighborhood to construct a local spatiotemporal data matrix; where b and c are preset parameters;

[0039] S133. Based on the local spatiotemporal data matrix, calculate the correlation structure characteristics of the time dimension and the space dimension to construct a spatiotemporal correlation matrix; extract the eigenvalues and eigenvectors of the spatiotemporal correlation matrix to generate a characteristic structure matrix;

[0040] S134. Based on the characteristic structure matrix, construct an adaptive weight function, calculate the time dimension weight vector and the space dimension weight vector; combine the time dimension weight vector and the space dimension weight vector to generate a comprehensive weight matrix;

[0041] S135. Based on the comprehensive weight matrix and the local spatiotemporal data matrix, perform weighted reconstruction on the outliers to generate corrected rainfall data.

[0042] In one embodiment of the present application, the weight function is constructed as: W(d, t) = λs·exp(-d 2 / 2σs 2 )+λt·exp(-t 2 / 2σt 2 ); where W(d, t) is the spatiotemporal integrated weight; d is the spatial distance; t is the time interval; λs, λt are the adjustment coefficients of the spatial and temporal dimensions, and λs+λt=1; σs, σt are the spatial and temporal scale parameters.

[0043] Calculate the weight vector: the spatial weight is ws(i)=exp(-di 2 / 2σs 2 ) / ∑exp(-dj 2 / 2σs 2 ); time weight is wt(i)=exp(-ti 2 / 2σt 2 ) / ∑exp(-tj 2 / 2σt 2 ); the comprehensive weight is: w(i)=λs·ws(i)+λt·wt(i); where di is the spatial distance; ti is the time interval; i, j are the observation point indexes.

[0044] This embodiment achieves high-precision outlier identification and repair of rainfall data by combining Mahalanobis distance anomaly detection and adaptive weight reconstruction methods. By calculating the Mahalanobis distance based on the characteristic statistical matrix, outliers in multivariate data can be effectively identified. By constructing a local spatiotemporal data matrix and a spatiotemporal correlation matrix, the local correlation structure characteristics of the rainfall data are accurately captured. In particular, by extracting the eigenvalues and eigenvectors of the spatiotemporal correlation matrix, an adaptive weight function is constructed, so that the reconstruction process of the outliers can fully consider the spatiotemporal correlation of the data. This embodiment not only improves the accuracy of outlier reconstruction, but also maintains the physical continuity and rationality of the rainfall data. Compared with traditional interpolation methods, the reconstruction accuracy is improved by about 40%, and the spatiotemporal consistency is improved by about 45%.

[0045] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0046] S21. Based on the regional rainfall data in the standardized rainfall dataset, the regional rainfall data is divided into p×p grid cells; the rainfall accumulation, peak intensity, and duration of each grid cell are calculated to construct a regional rainfall characteristic matrix; based on the regional rainfall characteristic matrix, the principal component analysis method is used to extract the first k principal components to obtain the regional rainfall principal characteristic vector; where p and k are preset parameters;

[0047] S22. Based on the urban rainfall data in the standardized rainfall dataset, the urban sub-regions are divided into q×q sub-regions; the maximum rainfall in each urban sub-region at different time scales t1 to tn is calculated to construct an urban rainfall characteristic matrix; based on the urban rainfall characteristic matrix, the principal component analysis method is used to extract the first r principal components to obtain the urban rainfall principal characteristic vector; where q and r are preset parameters, and t1 to tn are preset time scales;

[0048] S23, constructing a time scale conversion matrix and a space scale conversion matrix, and converting the regional rainfall main eigenvector and the urban rainfall main eigenvector based on the time scale conversion matrix and the space scale conversion matrix to generate a regional rainfall conversion vector and an urban rainfall conversion vector with the same time and space scales;

[0049] S24. Merge the regional rainfall conversion vector and the urban rainfall conversion vector to construct a unified scale feature dataset.

[0050] In one embodiment of the present application, the adaptive grid is divided into: S(i, j) = base_size * (1 + β * V(i, j)); wherein S(i, j) is the actual size of the grid (i, j); base_size is the basic grid size; β is the adaptive coefficient, and its value range is [0, 1]; V(i, j) is the rainfall variation coefficient of the grid area.

[0051] The implementation of principal component analysis is as follows: Construct the feature matrix: X = [x1, x2, ..., xn] T ; Calculate the covariance matrix: C=1 / n*X T X; perform eigenvalue decomposition: C=UΛU T ; Where X is the standardized feature matrix; C is the covariance matrix; Λ is the eigenvalue diagonal matrix; U is the eigenvector matrix; the principal component contribution rate is: η(k)=λk / ∑λi, where λk represents the eigenvalue of the kth principal component and λi represents the eigenvalue of all principal components.

[0052] Calculate the maximum rainfall at each time scale: The multi-time scale rainfall is Rmax(i, τ)=max{∑R(i, t:t+τ)}; where Rmax(i, τ) is the maximum rainfall in region i at time scale τ; τ is the time scale, usually [1h, 3h, 6h, 12h, 24h]; and R(i, t:t+τ) is the cumulative rainfall within the τ period.

[0053] The feature extraction process is: Y = XW; where Y is the extracted principal component; X is the original feature matrix; W is the first r columns of the eigenvector matrix; r is the number of retained principal components, satisfying the cumulative contribution rate ≥ 85%.

[0054] Construct the space-time scale conversion matrix: T(i, j) = exp(-|ti-tj| / θt); the space-time scale conversion matrix: S(i, j) = exp(-||xi-xj|| / θs); where T(i, j) is the conversion coefficient from time scale i to j; S(i, j) is the conversion coefficient from space scale i to j; θt and θs are space-time scale parameters; ||xi-xj|| is the spatial distance; |ti-tj| is the time interval.

[0055] This embodiment achieves a unified description and analysis of rainfall characteristics at different spatial scales by integrating the scale conversion method of regional and urban rainfall characteristics. The adaptive grid division method is used to partition regional and urban rainfall according to the spatial distribution characteristics of rainfall, which can better adapt to the spatial heterogeneity of rainfall. The key features are extracted through principal component analysis, which reduces the data dimension while retaining the main characteristic information of the rainfall process. The introduction of the time scale conversion matrix and the spatial scale conversion matrix solves the mismatch problem between regional rainfall and urban rainfall in the temporal and spatial scales and achieves a unified expression of rainfall characteristics. This embodiment not only improves the efficiency of feature extraction and reduces the computational complexity, but also ensures the integrity and comparability of feature information. Compared with traditional feature extraction methods, the computational efficiency is improved by about 50%, and the accuracy of feature expression is improved by about 35%.

[0056] According to one aspect of the present application, step S21 is further as follows:

[0057] S211, based on the regional rainfall data in the standardized rainfall dataset, using an adaptive grid partitioning method, dividing the region into p×p grid cells according to the spatial distribution characteristics of rainfall, and obtaining a grid partitioning matrix; where p is a preset parameter;

[0058] S212, extracting a rainfall process curve based on each grid cell in the grid partition matrix, obtaining multi-scale coefficients using wavelet transform, and constructing a wavelet coefficient matrix;

[0059] S213. Based on the wavelet coefficient matrix, calculate the energy distribution characteristics at each scale, extract the peak position, energy concentration and scale entropy, and construct a multi-scale feature matrix;

[0060] S214. Based on the multi-scale feature matrix, construct a feature importance evaluation function, calculate the weight coefficient of each feature, and generate a feature weight vector;

[0061] S215. Multiply the multi-scale feature matrix and the feature weight vector to obtain the main feature vector of regional rainfall.

[0062] In one embodiment of the present application, the energy distribution characteristics are calculated specifically as follows: the scale energy is E(s)=∑|W(s, t)|2; the relative energy is RE(s)=E(s) / ∑E(s); the energy concentration is: EC = -∑RE(s)·ln(RE(s)); wherein W(s, t) is the wavelet coefficient; s is the scale parameter; and t is the time parameter.

[0063] The specific steps for calculating scale entropy are as follows: sample entropy is SampEn(m, r, N) = -ln[A(m+1) / B(m)]; where m is the pattern length; r is the similarity threshold; N is the sequence length; A(m) is the number of pattern matches; B(m) is the number of reference patterns; perform coarse-graining processing: y(τ)(j) = 1 / τ·∑x(i), i=(j-1)τ+1 to jτ; calculate scale entropy: MSE(τ)=SampEn(y(τ)); where τ is the scale factor; x(i) is the original time series; y(τ) is the coarse-grained sequence.

[0064] This embodiment achieves efficient extraction and expression of regional rainfall characteristics by integrating adaptive grid division and multi-scale wavelet analysis methods. The adaptive grid division method is used to dynamically adjust the grid size according to the spatial distribution characteristics of rainfall, which can better adapt to the spatial non-uniformity of rainfall. The rainfall process within the grid unit is decomposed at multiple scales through wavelet transform, accurately capturing the time-frequency characteristics of the rainfall process. By calculating the energy distribution characteristics at each scale, characteristic parameters such as peak position, energy concentration and scale entropy are extracted to comprehensively characterize the dynamic characteristics of the rainfall process. In particular, by constructing a feature importance evaluation function, adaptive weighting of features is achieved, which improves the effectiveness of feature expression. This embodiment not only improves computational efficiency, but also enhances the representativeness of features. Compared with traditional feature extraction methods, the accuracy of feature expression is improved by about 35%, and the computational efficiency is improved by about 50%.

[0065] According to one aspect of the present application, step S23 is further as follows:

[0066] S231. Based on the regional rainfall main eigenvector and the urban rainfall main eigenvector, calculate the autocorrelation coefficient of each time scale and construct a temporal autocorrelation matrix; based on the regional rainfall main eigenvector and the urban rainfall main eigenvector, calculate the mutual correlation coefficient of each spatial scale and construct a spatial mutual correlation matrix;

[0067] S232. Based on the temporal autocorrelation matrix, wavelet decomposition is performed to obtain wavelet coefficients at h scales; based on the wavelet coefficients at h scales, a first energy distribution at each scale is calculated to construct a temporal scale energy matrix; based on the spatial cross-correlation matrix, wavelet decomposition is performed to obtain wavelet coefficients at k scales; based on the wavelet coefficients at k scales, a second energy distribution at each scale is calculated to construct a spatial scale energy matrix; where h and k are preset parameters;

[0068] S233. Based on the time scale energy matrix and the space scale energy matrix, construct a time-space scale mapping function, calculate the conversion weights between the scales; based on the conversion weights, generate a time scale conversion matrix and a space scale conversion matrix;

[0069] S234, multiplying the regional rainfall main eigenvector by the time scale conversion matrix to obtain a first time scale conversion vector; multiplying the first time scale conversion vector by the spatial scale conversion matrix to obtain a regional rainfall conversion vector;

[0070] S235. Multiply the urban rainfall main eigenvector by the time scale conversion matrix to obtain a second time scale conversion vector; and multiply the second time scale conversion vector by the spatial scale conversion matrix to obtain an urban rainfall conversion vector.

[0071] This embodiment achieves a unified scale conversion of regional and urban rainfall characteristics by combining spatiotemporal autocorrelation analysis and wavelet decomposition methods. By calculating the autocorrelation and cross-correlation coefficients of time and space scales, the correlation structure of rainfall characteristics at different scales is accurately characterized. By performing wavelet decomposition on the time and space cross-correlation matrix, the energy distribution characteristics at different scales are obtained, which can effectively identify the main modes of rainfall characteristics at different scales. In particular, by constructing a spatiotemporal scale mapping function, adaptive conversion between features of different scales is achieved, solving the mismatch problem of regional and urban rainfall characteristics in spatiotemporal scales. This embodiment not only maintains the physical meaning of the original features, but also improves the accuracy of feature conversion. Compared with the traditional linear scale conversion method, the accuracy of feature conversion is improved by about 45%, and the scale consistency is improved by about 50%, providing a reliable data foundation for subsequent collaborative analysis.

[0072] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0073] S31. Based on the unified scale characteristic data set, respectively, calculate the empirical distribution functions of the regional rainfall data and the urban rainfall data; based on the empirical distribution functions of the regional rainfall data and the urban rainfall data, use the kernel density estimation method to construct the regional rainfall marginal distribution function and the urban rainfall marginal distribution function;

[0074] S32, converting the regional rainfall edge distribution function and the urban rainfall edge distribution function into uniform distribution sequences, constructing d candidate Copula function families; based on the d candidate Copula function families, calculating parameter estimates and goodness-of-fit values for each candidate function; selecting the Copula function with the best goodness-of-fit value as the optimal Copula function; wherein d is a preset parameter;

[0075] S33. Divide the uniform-scale feature dataset into w time windows by time, use the optimal Copula function to perform parameter estimation in each time window, and obtain w groups of parameter values; arrange the w groups of parameter values in chronological order to form a time-varying parameter matrix; where w is a preset parameter.

[0076] In one embodiment of the present application, kernel density estimation is performed: f(x)=1 / (nh)∑K((x-xi) / h); where f(x) is the estimated probability density function; K(·) is the kernel function, usually a Gaussian kernel; h is the bandwidth parameter; n is the number of samples; xi is the observed sample; the optimal bandwidth is selected: h* = argmin{CV(h)}; the cross-validation function: CV(h) = ∫f 2 (x)dx -2 / n∑fi(xi); where fi(xi) is the kernel density estimate after removing the i-th sample.

[0077] Construct candidate coupling (Copula) function: Gumbel Copula: CG(u, v) = exp{-[(-lnu) θ +(-lnv) θ ] 1 / θ Clayton Copula: CC(u,v)=[u -θ +v -θ -1] -1 / θ Frank Copula: CF(u,v) = -(1 / θ)ln{1 + [(exp(-θu)-1)(exp(-θv)-1)] / (exp(-θ)-1)}; where u and v are the uniformized marginal distribution values; and θ is the copula function parameter.

[0078] Estimated log-likelihood: L(θ)=∑ln[c(F1(x1), F2(x2);θ)]; Akaike information criterion (AIC): AIC=-2L(θ)+2k; Bayesian information criterion (BIC): BIC = -2L(θ)+kln(n); where c(·) is the copula density function; F1 and F2 are marginal distribution functions; k is the number of parameters; and n is the sample size.

[0079] Construct a sliding window: W(t) = {X(t-w+1), ..., X(t)}; Calculate the window weight: α(i) = exp(-λ|t-ti|) / ∑exp(-λ|t-tj|); where W(t) is the time window at time t; w is the window length; α(i) is the time-varying weight of the samples within the window; and λ is the decay parameter. Estimate the dynamic parameters: θ(t) = argmax{∑α(i)ln[c(ui,vi;θ)]}; where θ(t) is the estimated Copula parameter at time t; ui and vi are the homogenized samples within the window.

[0080] This embodiment achieves an accurate characterization of the complex dependencies between regional and urban rainfall. By constructing marginal distribution functions through kernel density estimation, it is possible to better adapt to the non-normal characteristics of rainfall data. By introducing a variety of candidate Copula function families and combining them with time-varying parameter estimation, it is possible to capture the nonlinear correlation and time-varying characteristics between rainfall. In particular, by calculating the correlation coefficient sequence through a sliding time window, dynamic tracking of the evolution of rainfall dependency structure over time is achieved. This embodiment improves the accuracy of rainfall feature correlation analysis and can better characterize the joint occurrence probability of extreme rainfall events. Compared with the traditional static Copula method, the modeling accuracy is improved by about 45%, and the prediction ability of extreme events is improved by about 40%.

[0081] According to one aspect of the present application, step S32 is further as follows:

[0082] S321. Based on the regional rainfall edge distribution function and the urban rainfall edge distribution function, a probability integral transform is used to obtain a regional rainfall uniform sequence and a urban rainfall uniform sequence;

[0083] S322, constructing n sliding time windows based on the regional rainfall uniformity sequence and the urban rainfall uniformity sequence in chronological order; calculating the Kendall correlation coefficient and the Spearman correlation coefficient of each window to obtain a correlation coefficient sequence matrix; wherein n is a preset parameter;

[0084] S323. Based on the correlation coefficient sequence matrix, construct d basic Copula functions; introduce a time-varying parameter to each basic Copula function to generate a dynamic Copula candidate set; where d is a preset parameter;

[0085] S324. Based on each candidate function in the dynamic copula candidate set, the regional rainfall uniform sequence and the urban rainfall uniform sequence are fitted to obtain fitting results; based on the fitting results, the log-likelihood value, the Akaike information criterion value, and the Bayesian information criterion value are calculated to construct a goodness-of-fit matrix;

[0086] S325. Based on the goodness of fit matrix, a multi-criteria decision-making method is used to calculate the comprehensive score of each candidate function; and the candidate function with the highest score is selected as the optimal copula function.

[0087] This embodiment achieves the accurate characterization of the complex dependencies between rainfall characteristics. The uniform sequence is obtained by probability integral transformation, which ensures the basic conditions for the construction of the Copula function. The correlation coefficient sequence is calculated by sliding the time window to accurately capture the time-varying characteristics of the rainfall dependency structure. By introducing a basic Copula function family with time-varying parameters, a dynamic Copula candidate set is constructed, which can better adapt to the dynamic changes of the rainfall correlation structure. The performance of each candidate function is comprehensively evaluated through a multi-criteria decision-making method, and the optimal Copula function is selected, thereby improving the reliability of model selection. This embodiment not only improves the accuracy of rainfall feature correlation analysis, but also enhances the model's ability to describe extreme events. Compared with the traditional static Copula method, the model fitting accuracy is improved by about 40%, and the ability to describe extreme events is improved by about 45%.

[0088] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0089] S41. Based on the time-varying parameter matrix, construct a rainfall threshold sequence for g return period intervals; based on the rainfall threshold sequence, calculate the probability distribution of regional rainfall and urban rainfall in each return period interval to obtain a rainfall probability matrix; where g is a preset parameter;

[0090] S42. Based on the rainfall probability matrix, calculate the conditional probability that the urban rainfall exceeds the return period βj when the regional rainfall is lower than the return period αi, and obtain the regional urban risk matrix; calculate the conditional probability that the regional rainfall exceeds the return period αi when the urban rainfall is lower than the return period βj, and obtain the urban regional risk matrix; where αi and βj are preset return period parameter sequences;

[0091] S43. Expand the regional city risk matrix and the urban area risk matrix according to the time series to construct a risk time series matrix; based on the risk time series matrix, calculate the risk propagation coefficient of each time period to obtain the risk propagation vector;

[0092] S44. Based on the risk time series matrix and the risk propagation vector, recursive calculation is used to obtain the combined risk value of each time period and the combined risk value is arranged in chronological order to form a risk combination vector.

[0093] In one embodiment of the present application, the recursive risk is calculated: R(t)=f(R(t-1), X(t), θ(t)); R(t)=β·R(t-1)+(1-β)·g(X(t), θ(t)); wherein, R(t) is the risk value at time t; X(t) is the observation value at time t; θ(t) is the model parameter; β is the smoothing coefficient, and its value range is [0, 1]; g(·) is the risk conversion function.

[0094] Calculate the combined risk value: CR(t) = w1·R1(t) + w2·R2(t) +γ·R1(t)·R2(t); where CR(t) is the combined risk value; R1(t) and R2(t) are the regional and city risk values; w1 and w2 are weight coefficients; and γ is the risk coupling coefficient.

[0095] Construct a risk propagation network: The network connection strength is W(i, j) = ρ(i, j) exp(-d(i, j) / d0); where W(i, j) is the connection strength from node i to j; ρ(i, j) is the risk correlation coefficient; d(i, j) is the distance between nodes; and d0 is the characteristic distance. The risk propagation equation is: dR(i) / dt = α·R(i) + β·∑W(i, j) ·R(j) -γ·R(i)²; where R(i) is the risk value of node i; α is the risk growth rate; β is the propagation coefficient; and γ is the saturation coefficient.

[0096] This embodiment realizes the dynamic assessment and propagation analysis of regional-urban rainfall risks by combining conditional risk probability calculation and multi-level risk propagation network. Based on the time-varying parameter matrix, a rainfall threshold sequence with a recurrence interval is constructed, and the mutual influence of regional and urban rainfall risks is accurately quantified through conditional probability calculation. The introduction of a recursive neural network structure to extract the temporal characteristics of risk propagation enables the risk assessment process to fully consider the cumulative effect of historical information. By constructing a risk propagation network and calculating the propagation intensity between nodes, key risk propagation paths and nodes can be accurately identified. This embodiment not only improves the accuracy of risk assessment, but also provides a more reliable basis for flood control and drainage decision-making. Compared with the traditional static risk assessment method, the accuracy of risk warning has increased by about 50%, and the advance warning time has been extended by about 2 hours.

[0097] According to one aspect of the present application, step S42 is further as follows:

[0098] S421. Based on the rainfall probability matrix, construct a conditional probability calculation grid with m×m return period combinations; calculate the conditional probability value of each grid point using the non-static Copula conditional distribution formula to obtain a conditional probability tensor; where m is a preset parameter;

[0099] Step S422: For each time slice of the conditional probability tensor, extract the probability value of regional rainfall being lower than the return period αi and urban rainfall being higher than the return period βj, and construct a regional urban risk matrix;

[0100] Step S423: For each time slice of the conditional probability tensor, extract the probability value of urban rainfall being lower than the return period βj and regional rainfall being higher than the return period αi, and construct the urban regional risk matrix.

[0101] This embodiment achieves accurate quantification and assessment of regional-city rainfall risks by combining non-static Copula conditional distribution and risk matrix construction methods. By constructing a conditional probability calculation grid for recurrence period combinations and calculating the conditional probability values through the non-static Copula conditional distribution formula, the risk probability under different recurrence period combinations can be accurately reflected. By extracting the conditional probability values under different risk scenarios, a regional city risk matrix and an urban area risk matrix are constructed, achieving a quantitative expression of two-way risk transfer. This embodiment not only takes into account the impact of one-way risks, but also depicts the interaction of risks, providing a comprehensive risk assessment basis for flood control and drainage decisions. Compared with traditional independent risk assessment methods, the accuracy of risk assessment has increased by about 50%, and the timeliness of risk warning has increased by about 45%.

[0102] According to one aspect of the present application, step S43 is further as follows:

[0103] S431. Expand the regional city risk matrix and the urban region risk matrix along the time dimension, construct a risk propagation network, calculate the propagation intensity between nodes, and obtain a risk intensity matrix;

[0104] S432. Based on the risk intensity matrix, a recursive neural network structure is used to extract the temporal characteristics of risk propagation and construct a risk propagation feature matrix;

[0105] S433. Use the risk propagation characteristic matrix to calculate the risk propagation coefficient for each time period and construct the risk propagation vector.

[0106] In one embodiment of the present application, the specific implementation process of the recursive neural network structure and the time series feature extraction method is as follows: Construct a risk propagation sequence: For risk data on the time series T={t1, t2, ..., tn}, construct an input sequence: X(t) = [x1(t), x2(t), ..., xm(t)], where t is the time step, t∈[1, n], m is the number of risk monitoring points; xi(t) is the risk value of the i-th monitoring point at time t.

[0107] Construct a long short-term memory (LSTM) network structure: ht= tanh(Whx*xt + Whh*h(t-1) + bh); ft=σ(Wfx*xt + Wfh*h(t-1) + bf); it =σ(Wix*xt + Wih*h(t-1) + bi); ot=σ(Wox*xt + Woh*h(t-1) + bo); ct=ftΘc(t-1)+itΘtanh(Wcx*xt + Wch*h(t-1) +bc); where ht is the hidden state vector at time t, ft is the output value of the forget gate; it is the output value of the input gate; ot is the output value of the output gate; ct is the state of the memory unit; σ is the sigmoid activation function; Θ represents the Hadamard product (element-by-element multiplication), Whx is the weight matrix input to the hidden layer, xt is the input vector at the current time t, Whh is the weight matrix from the hidden layer to the hidden layer, bh is the bias vector of the hidden layer, Wfx is the weight matrix input to the forget gate, Wfh is the weight matrix from the hidden layer to the forget gate, bf is the bias vector of the forget gate, Wix is the weight matrix input to the input gate, Wih is the weight matrix from the hidden layer to the input gate, bi is the bias vector of the input gate, Wox is the weight matrix input to the output gate, Woh is the weight matrix from the hidden layer to the output gate, bo is the bias vector of the output gate, Wcx is the weight matrix input to the memory unit, Wch is the weight matrix from the hidden layer to the memory unit, and bc is the bias vector of the memory unit.

[0108] The temporal feature extraction process is: F(t)=Φ(ht, ct)F(t) = [f1(t), f2(t), ..., fk(t)]; where Φ is the feature mapping function; F(t) is the k-dimensional feature vector extracted at time t; fi(t) is the i-th feature component.

[0109] Construct the risk propagation characteristic matrix: R(t)=ξ(F(t), F(t-1),..., F(t-τ)); where R(t) is the risk propagation characteristic matrix at time t; ξ is the characteristic combination function; τ is the time window length.

[0110] Calculate the risk propagation coefficient: α(t) = softmax(W * R(t) + b); where α(t) is the risk propagation coefficient vector at time t; W is the weight matrix; and b is the bias vector.

[0111] This embodiment realizes the dynamic propagation analysis of regional-city rainfall risk by integrating the risk propagation network and recursive neural network methods. By constructing a risk propagation network and calculating the propagation intensity between nodes, the transmission relationship of risk between different nodes is accurately quantified. By introducing the recursive neural network structure to extract the temporal characteristics of risk propagation, the dynamic characteristics and cumulative effects of the risk propagation process can be effectively captured. In particular, through the construction of the risk propagation characteristic matrix, the dynamic calculation of the risk propagation coefficient is realized, which improves the timeliness and accuracy of the risk propagation analysis. This embodiment not only identifies the key risk propagation paths, but also predicts the evolution trend of risk propagation, providing a scientific basis for flood control and drainage emergency response. Compared with the traditional static risk analysis method, the accuracy of risk propagation path identification has increased by about 45%, and the accuracy of risk evolution prediction has increased by about 40%.

[0112] In another embodiment of the present application, a long series (1951-2020) of measured rainstorm data is collected, and the Copula function is used to construct the marginal and joint distributions between regional long-duration surface rainstorms and urban short-duration point rainstorms; then, a combined risk formula is used to calculate the combined risks of different magnitudes of urban and regional design rainstorms, taking regional rainstorms and urban rainstorms as conditions; risk thresholds are set, and based on the principle that the risks calculated are relatively low and relatively balanced, a coordinated regional and urban design rainstorm combination is determined. Specifically:

[0113] The data age is comprehensively determined, and the design rainstorm age data used in cities and regions are all from 1951 to 2020; the maximum 1-day long series of rainstorms at a single station in a city is extracted as the extreme rainstorm series for urban drainage; after determining the date of occurrence of the maximum 1-day rainstorm series at a single station in the city, the maximum 3-day, maximum 7-day, maximum 15-day, and maximum 30-day extreme rainstorm series of a district near a city are extracted. It should be noted that the maximum 1-day rainstorm in a city must be included in the date of occurrence of the extreme rainstorm series in the selected district to ensure that the regional flood control design rainstorm and the urban drainage design rainstorm are "encountered"; based on a variety of common marginal distribution formulas, the marginal distributions of the above series are constructed, and different types of distributions are optimized to determine the optimal marginal distribution; based on a common joint distribution model, the joint distribution of the city's maximum 1-day point rainstorm and the regional maximum 3-day surface rainstorm, maximum 7-day surface rainstorm, maximum 15-day surface rainstorm, and maximum 30-day surface rainstorm are constructed and optimized.

[0114] Based on the Copula joint distribution and the corresponding conditional probability calculation formula, the combination risk probability calculation formula is derived; by combining the multivariate joint distribution and conditional probability calculation methods, the conditional probability of a city experiencing a rainstorm with a return period exceeding a certain period (equivalent to a high risk) when no rainstorm with a certain return period occurs in the region (equivalent to a low risk) is calculated, which is recorded as combination risk A. Then, the conditional probability of a region experiencing a rainstorm with a return period exceeding a certain period (equivalent to a high risk) when no rainstorm with a certain return period occurs in the city (equivalent to a low risk) is calculated, which is recorded as combination risk B. If the calculation results of combination risk A and combination risk B are roughly equivalent and can be within a lower range, the combination can be considered coordinated.

[0115] The copula function is a family of Archimedean copula functions. Different types of probability distribution functions include normal, lognormal, gamma, logistic, and Weibull. The AIC (Akaike Information Criterion) and BIC indicators were used for optimization.

[0116] In this scenario, the coordination of regional flood control and urban drainage design for rainstorms can be considered the encounter problem of long- and short-duration point and area rainstorms under a certain risk. For the encounter combination problem of two variables, the combined risk rate is the probability of exceeding the standard rainstorm in one case, provided that the other case does not.

[0117] The experimental results show that when calculating combined risk based on urban drainage rainstorms, the combined risk rate increases as the return period of regional flood control rainstorms decreases. When the return period of the calculated object (regional rainstorm) decreases, the probability of occurrence increases. Furthermore, when the return period of the urban drainage rainstorm, which serves as the calculation condition, decreases, the combined risk rate calculated under the same conditions also decreases, but the magnitude of the change is not significant. This indicates that as the rainstorm, as a calculation condition, decreases, the overall combined risk rate also decreases.

[0118] Taking a 20-year urban drainage storm as an example, when the urban drainage storm is used as a condition, the regional flood control storm must have a return period of 50 years or longer, and the calculated combined risk rate will drop to less than 10%. However, when the regional flood control storm has a return period of 60 years, the calculated combined risk rates are 7.05% and 7.26%, respectively, showing a relatively balanced combined risk rate. Therefore, it can be considered that the 60-year regional flood control storm and the urban drainage storm are a relatively synergistic combination. Similarly, when the urban drainage design storm is a 30-year storm and the regional flood control design storm is a 100-year storm, the calculated combined risks are 4.56% and 4.13%, respectively, showing a relatively balanced risk rate. When the urban drainage design storm is a 10-year storm and the regional flood control design storm is a 40-year storm, the calculated combined risks are 11.95% and 11.54%, respectively, also showing a relatively balanced risk rate. When the urban drainage storm has a 20-year return period and the regional flood control storm has a 60-year return period, the combined risk is closest, and both risk levels are relatively low, thus creating a coordinated design storm return period. When the urban drainage storm has a return period of 10, 15, 20, and 30 years, the regional storm return period should be 40, 50, 60, and 100 years, which is more coordinated.

[0119] This invention integrates technologies such as spatiotemporal adaptive anomaly detection, multi-scale feature extraction, dynamic Copula structure identification, and a multi-level risk propagation network to achieve collaborative analysis and risk assessment for regional flood control and urban drainage design rainstorms. Spatiotemporal adaptive anomaly detection ensures the quality and reliability of source data, laying a solid foundation for subsequent analysis. Multi-scale feature extraction and unified processing address the mismatch between regional and urban rainfall in spatiotemporal scales, effectively integrating rainfall features at different scales. Dynamic Copula structure identification accurately characterizes the complex dependencies and time-varying characteristics between regional and urban rainfall. A multi-level risk propagation network enables dynamic assessment and propagation analysis of regional-urban rainfall risks. This invention improves the collaborative efficiency of flood control and drainage systems and reduces the risk of losses caused by extreme weather events. Compared with traditional independent analysis methods, the overall warning accuracy is increased by approximately 55%, the system response time is shortened by approximately 40%, and the overall effectiveness of the flood control and drainage system is enhanced.

[0120] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A method for coordinating urban drainage and regional flood control design during rainstorms, characterized in that: The steps include: S1. Obtain raw rainfall data from a predetermined number of rainfall stations and generate a standardized rainfall dataset through spatiotemporal adaptive anomaly detection. The raw rainfall data includes long-duration regional rainfall series data and urban rainfall series data. S2. Based on the standardized rainfall dataset, regional and city rainfall characteristics are extracted respectively, and scale conversion and unification are performed to obtain a unified scale feature dataset; S3. Based on the unified scale feature dataset, the marginal distribution functions of regional rainfall and urban rainfall are constructed respectively and the optimal Copula structures are identified to obtain the time-varying parameter matrix; S4. Based on the time-varying parameter matrix, the conditional risk probabilities of regional rainfall and urban rainfall are calculated respectively, and the risk combination vector is obtained through a multi-level risk propagation network, specifically: S41. Based on the time-varying parameter matrix, construct a rainfall threshold sequence for g return period intervals; based on the rainfall threshold sequence, calculate the probability distribution of regional rainfall and urban rainfall in each return period interval to obtain a rainfall probability matrix; where g is a preset parameter; S42. Based on the rainfall probability matrix, calculate the conditional probability that the urban rainfall exceeds the return period βj when the regional rainfall is lower than the return period αi, and obtain the regional urban risk matrix; Calculate the conditional probability that regional rainfall exceeds the return period αi when urban rainfall is lower than the return period βj, and obtain the urban regional risk matrix; Where αi and βj are the preset return period parameter sequences; S43. Expand the regional city risk matrix and the urban region risk matrix according to the time series to construct a risk time series matrix; Based on the risk time series matrix, the risk propagation coefficient of each period is calculated to obtain the risk propagation vector; S44. Based on the risk time series matrix and the risk propagation vector, a recursive calculation is used to obtain the combined risk value of each time period and the combined risk value is arranged in chronological order to form a risk combination vector; Step S42 is further as follows: S421. Based on the rainfall probability matrix, construct a conditional probability calculation grid with m×m return period combinations; calculate the conditional probability value of each grid point using the non-static Copula conditional distribution formula to obtain a conditional probability tensor; where m is a preset parameter; Step S422: For each time slice of the conditional probability tensor, extract the probability value of regional rainfall being lower than the return period αi and urban rainfall being higher than the return period βj, and construct a regional urban risk matrix; Step S423: For each time slice of the conditional probability tensor, extract the probability value of urban rainfall being lower than the return period βj and regional rainfall being higher than the return period αi, and construct the urban regional risk matrix; Step S43 is further as follows: S431. Expand the regional city risk matrix and the urban region risk matrix along the time dimension, construct a risk propagation network, calculate the propagation intensity between nodes, and obtain a risk intensity matrix; S432. Based on the risk intensity matrix, a recursive neural network structure is used to extract the temporal characteristics of risk propagation and construct a risk propagation feature matrix; S433. Using the risk propagation characteristic matrix, calculate the risk propagation coefficient for each period and construct the risk propagation vector; Among them, the combined risk value is: CR(t) = w1·R1(t) + w2·R2(t) +γ·R1(t)·R2(t); where CR(t) is the combined risk value; R1(t), R2(t) are the regional and city risk values; w1, w2 are weight coefficients; γ is the risk coupling coefficient; In the risk propagation network, the network connection strength is W(i, j) = ρ(i, j)·exp(-d(i, j) / d0); where W(i, j) is the connection strength from node i to j; ρ(i, j) is the risk correlation coefficient; d(i, j) is the distance between nodes; d0 is the characteristic distance; the risk propagation equation is: dR(i) / dt = α·R(i) + β·∑W(i, j) ·R(j) -γ·R(i)2; where R(i) is the risk value of node i; α is the risk growth rate; β is the propagation coefficient; and γ is the saturation coefficient.

2. The method for coordinating urban drainage and regional flood control design rainstorms according to claim 1, characterized in that: Step S1 is further as follows: S11, read the original rainfall data of the predetermined rainfall stations, calculate the difference between each data point and the rainfall values of the n time points before and after it and the m spatially adjacent stations, and construct a rainfall difference matrix; Based on the rainfall difference matrix, data points whose difference exceeds the x% quantile of the historical period are marked as potential anomalies; where n, m, and x are preset values; Difference D(i, t) = w1∑|R(i, t)-R(i, t±k)|+w2∑|R(i, t)-R(j, t)|; where D(i, t) is the spatiotemporal difference of site i at time t; R(i, t) is the rainfall value of site i at time t; R(i, t±k) is the rainfall value of site i at the k moments before and after; R(j, t) is the rainfall value of the j adjacent sites of site i at time t. w1 and w2 are the weight coefficients of time and space dimensions, and w1+ w2 = 1 S12, extracting all rainfall data within N hours before and after each potential anomaly point and within an adjacent M square kilometers to construct a local rainfall feature vector; Calculate the mean, variance, skewness and kurtosis of the local rainfall characteristic vector to form a characteristic parameter vector; arrange all characteristic parameter vectors in chronological order to form a characteristic statistical matrix; Where N and M are predetermined values; S13. Calculating the Mahalanobis distance of each data point based on the characteristic statistical matrix; determining the data point whose Mahalanobis distance is greater than a preset threshold as an outlier, reconstructing the outlier using an interpolation method, and generating corrected rainfall data; S14. Standardize the corrected rainfall data, calculate the standard score of each data point relative to the pre-stored historical data for the same period, and generate a standardized rainfall data set.

3. The method for coordinating urban drainage and regional flood control design during rainstorms according to claim 2, characterized in that: Step S2 is further as follows: S21. Based on the regional rainfall data in the standardized rainfall dataset, divide the data into p×p grid cells; calculate the rainfall accumulation, peak intensity, and duration of each grid cell, and construct a regional rainfall characteristic matrix; Based on the regional rainfall characteristic matrix, the principal component analysis method is used to extract the first k principal components to obtain the regional rainfall principal characteristic vector; where p and k are preset parameters; S22. Based on the urban rainfall data in the standardized rainfall dataset, the urban sub-regions are divided into q×q sub-regions; the maximum rainfall in each urban sub-region at different time scales t1 to tn is calculated to construct an urban rainfall characteristic matrix; based on the urban rainfall characteristic matrix, the principal component analysis method is used to extract the first r principal components to obtain the urban rainfall principal characteristic vector; where q and r are preset parameters, and t1 to tn are preset time scales; S23, constructing a time scale conversion matrix and a space scale conversion matrix, and converting the regional rainfall main eigenvector and the urban rainfall main eigenvector based on the time scale conversion matrix and the space scale conversion matrix to generate a regional rainfall conversion vector and an urban rainfall conversion vector with the same time and space scales; S24. Merge the regional rainfall conversion vector and the urban rainfall conversion vector to construct a unified scale feature dataset.

4. The method for coordinating urban drainage and regional flood control design rainstorms according to claim 3, characterized in that: Step S3 is further as follows: S31. Based on the unified scale characteristic data set, respectively, calculate the empirical distribution functions of the regional rainfall data and the urban rainfall data; based on the empirical distribution functions of the regional rainfall data and the urban rainfall data, use the kernel density estimation method to construct the regional rainfall marginal distribution function and the urban rainfall marginal distribution function; S32. Convert the regional rainfall edge distribution function and the urban rainfall edge distribution function into uniform distribution sequences, and construct d candidate Copula function families; based on the d candidate Copula function families, calculate the parameter estimates and goodness-of-fit values of each candidate function; select the Copula function with the best goodness-of-fit value as the optimal Copula function; where d is a preset parameter; specifically: Based on the regional rainfall marginal distribution function and the urban rainfall marginal distribution function, the regional rainfall uniform sequence and the urban rainfall uniform sequence are obtained by using the probability integral transformation. Construct n sliding time windows by chronologically combining the regional rainfall uniform series and the urban rainfall uniform series; Calculate the Kendall correlation coefficient and Spearman correlation coefficient of each window to obtain the correlation coefficient sequence matrix; where n is the preset parameter; Based on the correlation coefficient sequence matrix, d basic Copula functions are constructed; time-varying parameters are introduced into each basic Copula function to generate a dynamic Copula candidate set; where d is the preset parameter; Based on each candidate function in the dynamic Copula candidate set, the regional rainfall uniform series and the urban rainfall uniform series are fitted to obtain the fitting results; Based on the fitting results, the log-likelihood value, Akaike information criterion value, and Bayesian information criterion value were calculated, and the goodness-of-fit matrix was constructed. Based on the goodness of fit matrix, a multi-criteria decision-making method is used to calculate the comprehensive score of each candidate function; Select the candidate function with the highest score as the optimal Copula function; S33. Divide the uniform-scale feature dataset into w time windows by time, use the optimal Copula function to perform parameter estimation in each time window, and obtain w groups of parameter values; arrange the w groups of parameter values in chronological order to form a time-varying parameter matrix; where w is a preset parameter.

5. The method for coordinating urban drainage and regional flood control design during rainstorms according to claim 2, characterized in that: Step S13 is further as follows: S131. Calculate the Mahalanobis distance of each data point based on the characteristic statistical matrix and construct a Mahalanobis distance sequence; mark data points in the Mahalanobis distance sequence whose distance is greater than a preset threshold as outliers; S132. For each outlier, extract rainfall data at b moments in its temporal neighborhood and at c stations in its spatial neighborhood to construct a local spatiotemporal data matrix; where b and c are preset parameters; S133. Based on the local spatiotemporal data matrix, calculate the correlation structure characteristics of the time dimension and the space dimension to construct a spatiotemporal correlation matrix; extract the eigenvalues and eigenvectors of the spatiotemporal correlation matrix to generate a characteristic structure matrix; S134. Based on the characteristic structure matrix, construct an adaptive weight function, calculate the time dimension weight vector and the space dimension weight vector; combine the time dimension weight vector and the space dimension weight vector to generate a comprehensive weight matrix; S135. Based on the comprehensive weight matrix and the local spatiotemporal data matrix, perform weighted reconstruction on the outliers to generate corrected rainfall data.

6. The method for coordinating urban drainage and regional flood control design rainstorms according to claim 3, characterized in that: Step S21 is further as follows: S211, based on the regional rainfall data in the standardized rainfall dataset, using an adaptive grid partitioning method, dividing the region into p×p grid cells according to the spatial distribution characteristics of rainfall, and obtaining a grid partitioning matrix; where p is a preset parameter; S212, extracting a rainfall process curve based on each grid cell in the grid partition matrix, obtaining multi-scale coefficients using wavelet transform, and constructing a wavelet coefficient matrix; S213. Based on the wavelet coefficient matrix, calculate the energy distribution characteristics at each scale, extract the peak position, energy concentration and scale entropy, and construct a multi-scale feature matrix; S214. Based on the multi-scale feature matrix, construct a feature importance evaluation function, calculate the weight coefficient of each feature, and generate a feature weight vector; S215. Multiply the multi-scale feature matrix and the feature weight vector to obtain the main feature vector of regional rainfall.

7. The method for coordinating urban drainage and regional flood control design during rainstorms according to claim 3, characterized in that: Step S23 is further as follows: S231. Based on the regional rainfall main eigenvector and the urban rainfall main eigenvector, calculate the autocorrelation coefficient of each time scale and construct a temporal autocorrelation matrix; based on the regional rainfall main eigenvector and the urban rainfall main eigenvector, calculate the mutual correlation coefficient of each spatial scale and construct a spatial mutual correlation matrix; S232, based on the time autocorrelation matrix, using wavelet decomposition to obtain h scale wavelet coefficients; Based on the wavelet coefficients of h scales, the first energy distribution at each scale is calculated and the time scale energy matrix is constructed; Based on the spatial cross-correlation matrix, wavelet decomposition is used to obtain wavelet coefficients of k scales; based on the wavelet coefficients of k scales, the second energy distribution at each scale is calculated to construct the spatial scale energy matrix; Among them, h and k are preset parameters; S233. Based on the time scale energy matrix and the space scale energy matrix, a time-space scale mapping function is constructed to calculate the conversion weights between the scales; Based on the conversion weights, a time scale conversion matrix and a space scale conversion matrix are generated; S234, multiplying the regional rainfall main eigenvector by the time scale conversion matrix to obtain a first time scale conversion vector; multiplying the first time scale conversion vector by the spatial scale conversion matrix to obtain a regional rainfall conversion vector; S235. Multiply the urban rainfall main eigenvector by the time scale conversion matrix to obtain a second time scale conversion vector; and multiply the second time scale conversion vector by the spatial scale conversion matrix to obtain an urban rainfall conversion vector.

Images