Urban extreme rainfall risk assessment and drainage optimization method fusing composite threshold and Chicago method

Abstract

The invention provides an urban extreme rainfall risk assessment and drainage optimization method fusing a composite threshold and a Chicago method, and relates to the field of power grid disaster prevention and reduction planning and meteorological and hydrological analysis, and the method comprises the following steps: employing a quartile method and a composite threshold method to recognize and classify extreme rainfall events; constructing an extreme rainfall pattern model, and determining a minute-level rainfall intensity curve by using the extreme rainfall pattern model; generating an extreme rainfall spatial-temporal distribution characteristic spectrum according to the extreme rainfall event set in combination with a spatial interpolation technology fused with a deep learning model; the drainage capacity of the urban drainage system is evaluated, and the urban drainage system is optimized according to the evaluation result. The method has high adaptability, can be popularized to urban areas with different terrains and climate conditions, provides an efficient technical tool for coping with extreme weather events under the climate change background and improving the disaster prevention and reduction capacity of a power system, and has important social and economic values.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power grid disaster prevention and reduction planning and meteorological and hydrological analysis, in particular, especially relates to a city extreme rainfall risk assessment and drainage optimization method fusing composite threshold and Chicago method. BACKGROUND

[0002] Under the background of global warming, extreme rainfall events are becoming more frequent and intense, which undoubtedly poses a major threat to urban power grid infrastructure. As the "lifeline" of the city, the stable operation of the urban power grid is of great importance. However, secondary disasters such as urban waterlogging and landslides caused by extreme rainfall can easily cause damage to power grid equipment and line short circuits, seriously affecting the reliable power supply of the power grid. In order to effectively evaluate the operation risk of urban power grid equipment under extreme rainfall conditions and ensure the safety and stability of the power grid, it is necessary to conduct in-depth analysis on the characteristics of extreme rainfall, such as rainfall intensity, duration, coverage, etc., to provide a basis for developing scientific and reasonable response measures.

[0003] However, the current methods for identifying and analyzing the impact of extreme rainfall have limitations, and the dynamic process of short-duration rainfall is not well analyzed, and the identification of extreme rainfall is insufficient. The traditional percentile method relies on the length of historical data, and the fixed threshold method ignores regional climate differences (such as in arid regions, rain may cause power grid failure), resulting in a risk of missing high-destructive short-duration heavy rainfall. At the same time, there is a lack of dynamic process analysis, for example, existing rain type simulation (such as Chicago method) lacks minute-level precision, making it difficult to support power grid flood control design (such as substation drainage capacity) and operation warning (such as relay protection action timing optimization).

[0004] In view of the problems in the related art, an effective solution has not been proposed yet. SUMMARY

[0005] Therefore, the present application provides a city extreme rainfall risk assessment and drainage optimization method fusing composite threshold and Chicago method to solve the above-mentioned problems.

[0006] In order to solve the above problems, the specific technical scheme adopted by the present application is as follows:

[0007] A city extreme rainfall risk assessment and drainage optimization method fusing composite threshold and Chicago method, comprising the following steps:

[0008] S1, obtaining daily rainfall data of meteorological stations, and using quartile range method and composite threshold method to identify and classify extreme rainfall events, obtaining an extreme rainfall event set;

[0009] S2, constructing an extreme rainfall pattern model based on the Chicago method and a generative adversarial network, and determining a minute-level rainfall intensity curve by using the extreme rainfall pattern model and combining the extreme rainfall event set;

[0010] S3, generating an extreme rainfall spatiotemporal distribution feature map according to the extreme rainfall event set and combining a spatial interpolation technology of a fusion deep learning model;

[0011] S4, evaluating the drainage capacity of the urban drainage system based on the minute-level rainfall intensity curve and combining the extreme rainfall spatiotemporal distribution feature map, and optimizing the urban drainage system according to the evaluation result.

[0012] Preferably, the daily rainfall data of the meteorological station is obtained, and the extreme rainfall event set is obtained by identifying and classifying the extreme rainfall events by using the interquartile range method and the composite threshold method, including the following steps:

[0013] S11, collecting daily rainfall data of the meteorological station, arranging the daily rainfall data set in ascending order, and determining the quartiles of the daily rainfall data set according to the arrangement result;

[0014] S12, calculating the quartile moment of the daily rainfall data set according to the quartiles of the daily rainfall data set, determining the screening range according to the interquartile range, determining the screening condition according to the screening range, and deleting the data points of the daily rainfall data set that do not meet the screening condition;

[0015] S13, filling the deleted data points based on a long short-term memory network model to obtain a standard data set;

[0016] S14, identifying the extreme rainfall events by using the composite threshold method according to the standard data set, and classifying the extreme rainfall events to obtain the extreme rainfall event set.

[0017] Preferably, the filling of the deleted data points based on the long short-term memory network model to obtain the standard data set includes the following steps:

[0018] S131, determining a target data point to be filled according to the deleted data point, and inputting the historical rainfall sequence of the target data point and the rainfall data of the adjacent stations thereof into the long short-term memory network model;

[0019] S132, based on the historical rainfall sequence of the target data point and the rainfall data of the adjacent stations thereof, and combining the long short-term memory network model, outputting the time series prediction value of the deleted data point by calculating and minimizing the mean square error between the predicted value and the true value through multiple rounds of training of the long short-term memory network model;

[0020] S133, fill in the data points deleted by using the time series prediction value, and arrange in time series format to obtain a standard data set.

[0021] Preferably, the extreme rainfall event is identified and classified by using the composite threshold method according to the standard data set, and the extreme rainfall event set is obtained, including the following steps:

[0022] S141, arrange the standard data set according to the size of rainfall, calculate the percentile of the standard data set by using linear interpolation method, and take the percentile of the standard data set as a dynamic threshold;

[0023] S142, determine the intensity standard of the extreme rainfall event based on the preset rainfall grade standard, and take the intensity standard of the extreme rainfall event as a fixed threshold;

[0024] S143, integrate the dynamic threshold and the fixed threshold, and construct a composite threshold judgment rule;

[0025] S144, identify and classify the extreme rainfall event by using the composite threshold judgment rule to obtain the extreme rainfall event set.

[0026] Preferably, the extreme rainfall pattern model is constructed based on the Chicago method and the generative adversarial network, and the minute-level rainfall intensity curve is determined by combining the extreme rainfall event set by using the extreme rainfall pattern model, including the following steps:

[0027] S21, according to the daily rainfall data of the meteorological station, based on the preset rainfall time, extract the heavy rainfall event, and generate a heavy rainfall event data set;

[0028] S22, based on the generative adversarial network composed of a generator and a discriminator, construct an extreme rainfall pattern model;

[0029] S23, according to the heavy rainfall event data set, use the Worshtein loss function to train the extreme rainfall pattern model, and combine the total rainfall constraint to optimize the parameters of the generator and the discriminator in the extreme rainfall pattern model;

[0030] S24, based on the parameter-optimized extreme rainfall pattern model, combine the Chicago method storm intensity formula to generate a probabilistic minute-level rainfall intensity curve;

[0031] S25, compare the probabilistic minute-level rainfall intensity curve with the extreme rainfall event set, calculate the mean square error, calibrate and verify the probabilistic minute-level rainfall intensity curve, and obtain the final minute-level rainfall intensity curve.

[0032] Preferably, generating a spatiotemporal distribution feature map of extreme rainfall based on a set of extreme rainfall events in combination with a spatial interpolation technique integrating a deep learning model comprises the following steps:

[0033] S31. Calculate the dynamic threshold grid based on the set of extreme rainfall events using thin plate spline interpolation and spatiotemporal adaptive interpolation network;

[0034] S32. Calculate the spatial variation coefficient based on the dynamic threshold grid, and calculate the extreme rainfall frequency and average rainfall intensity of each grid cell using the kriging method based on the set of extreme rainfall events;

[0035] S33. Generate a spatiotemporal distribution characteristic map of extreme rainfall based on the extreme rainfall frequency and average rainfall intensity of each grid cell and a preset dimension.

[0036] Preferably, the step of calculating the dynamic threshold grid based on the set of extreme rainfall events by using a thin plate spline interpolation method and a spatiotemporal adaptive interpolation network comprises the following steps:

[0037] S311, mapping the extreme rainfall event set to a grid system of a first resolution, and generating a rasterized dataset using a thin plate spline interpolation method to obtain the extreme rainfall frequency and the extreme rainfall average intensity of the first resolution;

[0038] S312. Based on the set of extreme rainfall events, using a spatiotemporal adaptive interpolation network, combined with a convolutional neural network, a long short-term memory network, and an attention mechanism, a multi-layer perceptron is used to generate the second-resolution extreme rainfall frequency and the extreme rainfall average intensity;

[0039] S313: Fusing the extreme rainfall frequency and the extreme rainfall average intensity of the first resolution with the extreme rainfall frequency and the extreme rainfall average intensity of the second resolution to obtain a fused extreme rainfall frequency and extreme rainfall average intensity, and using the fused extreme rainfall frequency and extreme rainfall average intensity as a dynamic threshold grid.

[0040] Preferably, the step of calculating the spatial variation coefficient based on the dynamic threshold grid and calculating the extreme rainfall frequency and average rainfall intensity of each grid cell by the Kriging method based on the set of extreme rainfall events comprises the following steps:

[0041] S321. Calculate percentile rainfall thresholds based on the dynamic threshold grid, and determine the spatial mean and spatial standard deviation of the percentile rainfall thresholds;

[0042] S322. Calculate the spatial coefficient of variation based on the spatial mean and the spatial standard deviation, and determine whether the dynamic threshold grid requires threshold correction based on the spatial coefficient of variation. If necessary, adjust the dynamic threshold grid using the correction term formula to obtain a corrected dynamic threshold. If not, use the dynamic threshold grid as the corrected dynamic threshold.

[0043] S323, calculating the initial extreme rainfall frequency and rainfall average intensity of each grid cell based on the corrected dynamic threshold and the set of extreme rainfall events;

[0044] S324, optimizing the extreme rainfall frequency and rainfall average intensity of each grid cell by a spatio-temporal adaptive interpolation network to obtain the optimized extreme rainfall frequency and rainfall average intensity of each grid cell;

[0045] S325, performing spatial interpolation and visualization processing on the optimized extreme rainfall frequency and rainfall average intensity of each grid cell based on a geographic information system tool and in combination with the Kriging method to obtain the final extreme rainfall frequency and rainfall average intensity of each grid cell.

[0046] Preferably, the minute-level rainfall intensity curve is combined with the spatio-temporal distribution characteristic map of extreme rainfall to evaluate the drainage capacity of the urban drainage system, and the urban drainage system is optimized according to the evaluation result, including the following steps:

[0047] S41, based on the minute-level rainfall intensity curve, the grid rainfall intensity and frequency in the spatio-temporal distribution characteristic map of extreme rainfall are combined to calculate the rainfall intensity of each grid cell under different return periods, and compared with the design capacity of the drainage system to evaluate the drainage capacity of each grid cell, and the visualization result of the spatio-temporal distribution characteristic map is used to identify the area with insufficient drainage capacity;

[0048] S42, according to the spatial variability of the spatio-temporal distribution characteristic map, the drainage capacity difference of each grid cell is calculated, and the rainfall intensity distribution in the spatio-temporal distribution characteristic map is combined to determine the risk area;

[0049] S43, according to the drainage capacity difference of the drainage system under different rainfall intensities, the risk area in the spatio-temporal distribution characteristic map is combined to determine the optimization scheme of the waterlogging prevention facility, and the optimization scheme of the waterlogging prevention facility is adjusted through waterlogging risk analysis to obtain the final optimization scheme of the waterlogging prevention facility.

[0050] Preferably, the optimization scheme of the waterlogging prevention facility is determined according to the drainage capacity difference of the drainage system under different rainfall intensities, in combination with the risk area in the spatio-temporal distribution characteristic map, and the optimization scheme of the waterlogging prevention facility is adjusted through waterlogging risk analysis to obtain the final optimization scheme of the waterlogging prevention facility, including the following steps:

[0051] S431, according to the drainage capacity difference of the drainage system under different rainfall intensities, and based on the design basis of the expanded drainage pipeline, the risk area in the spatio-temporal distribution characteristic map is combined to determine the optimization scheme of the waterlogging prevention facility;

[0052] S432, adjust the diameter of the drainage pipe and add new rainwater collection facilities according to the waterlogging prevention facility optimization scheme, and calculate the waterlogging risk of each area in the drainage system;

[0053] S433, based on the waterlogging risk of each area in the drainage system, determine the optimized drainage area by optimizing the area formula, and determine the optimized area design scheme of the waterlogging prevention facility;

[0054] S434, based on the optimized area design scheme of the waterlogging prevention facility, adjust the waterlogging prevention facility optimization scheme to obtain the final waterlogging prevention facility optimization scheme.

[0055] The beneficial effects of the present application are: the extreme rainfall characteristic analysis of the present application improves the precision and practicality, and provides a scientific basis for urban waterlogging prevention planning and resilience infrastructure construction. By integrating the composite threshold method and the Chicago method, combining the ANUSPLIN spline interpolation of the deep learning model and the terrain correction technology, the method can accurately identify extreme rainfall events and generate high-resolution minute-level rainfall intensity curves, breaking through the limitations of traditional methods in time and space distribution simulation and long return period accuracy; the analysis results can provide reliable data support for power system planners, optimize the design of flood control measures, and protect power facilities from flood damage; in addition, the method has strong adaptability and can be popularized to urban areas with different topography and climate conditions, providing an efficient technical tool for coping with extreme weather events under the background of climate change and improving the disaster prevention and reduction capacity of the power system, which has important social and economic value. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. In the drawings:

[0057] Figure 1 is one of the flowcharts of a city extreme rainfall risk assessment and drainage optimization method according to an embodiment of the present application, which is a fusion of composite threshold and Chicago method;

[0058] Figure 2 is a graph of the average annual frequency exceeding the 99th percentile in a city extreme rainfall risk assessment and drainage optimization method according to an embodiment of the present application, which is a fusion of composite threshold and Chicago method;

[0059] Figure 3 is a graph of the average annual frequency exceeding 50mm / d in a city extreme rainfall risk assessment and drainage optimization method according to an embodiment of the present application, which is a fusion of composite threshold and Chicago method;

[0060] Figure 4 This is a schematic diagram of the annual average frequency exceeding 100 mm / d in an urban extreme rainfall risk assessment and drainage optimization method integrating a composite threshold and the Chicago method according to an embodiment of the present invention;

[0061] Figure 5 A map of annual average precipitation exceeding the 99th percentile in an urban extreme rainfall risk assessment and drainage optimization method integrating a composite threshold and the Chicago method according to an embodiment of the present invention;

[0062] Figure 6 A map of annual average rainfall exceeding 50 mm / d in an urban extreme rainfall risk assessment and drainage optimization method integrating a composite threshold and the Chicago method according to an embodiment of the present invention;

[0063] Figure 7 A map of annual average precipitation exceeding 100 mm / d in an urban extreme rainfall risk assessment and drainage optimization method integrating a composite threshold and the Chicago method according to an embodiment of the present invention;

[0064] Figure 8 A rainfall process diagram with different return periods in an urban extreme rainfall risk assessment and drainage optimization method integrating a composite threshold and the Chicago method according to an embodiment of the present invention;

[0065] Figure 9 This is the second flowchart of a method for urban extreme rainfall risk assessment and drainage optimization that integrates a composite threshold and the Chicago method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0067] According to an embodiment of the present invention, a method for urban extreme rainfall risk assessment and drainage optimization is provided that integrates a composite threshold and the Chicago method.

[0068] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 and Figure 9 As shown, the urban extreme rainfall risk assessment and drainage optimization method integrating the composite threshold and the Chicago method according to an embodiment of the present invention includes the following steps:

[0069] S1, obtain daily rainfall data of a weather station, and identify and classify extreme rainfall events by using the interquartile range method and the composite threshold method to obtain an extreme rainfall event set;

[0070] As a preferred embodiment, the obtaining daily rainfall data of a weather station, and identifying and classifying extreme rainfall events by using the interquartile range method and the composite threshold method to obtain an extreme rainfall event set comprises the following steps:

[0071] S11, collect daily rainfall data of a weather station, arrange the daily rainfall data set in ascending order, and determine the quartiles of the daily rainfall data set according to the arrangement result;

[0072] S12, calculate the interquartile range of the daily rainfall data set according to the quartiles of the daily rainfall data set, determine the screening range according to the interquartile range, determine the screening condition according to the screening range, and delete the data points of the daily rainfall data set that do not meet the screening condition;

[0073] Specifically, when collecting daily rainfall data of a weather station, long-time series historical records based on daily rainfall data of a national weather station are collected to ensure the representativeness and statistical reliability of the data. The interquartile range (IQR) method is used to identify outliers. By calculating the quartiles of the data, the range of normal data is determined, and the values outside the range are marked as outliers. The calculation steps are as follows:

[0074] First, arrange the daily rainfall data set in ascending order. The first quartile Q1 is the quartile of the first 25% of the data set (i.e., the 25th percentile), and the third quartile Q3 is the quartile of the first 75% of the data set (i.e., the 75th percentile). The calculation formula of the interquartile range IQR is:

[0075] IQR = Q3 - Q1;

[0076] Then determine the range of outliers. The lower bound of the abnormal range is Q1-k·IQR, and the upper bound is Q1+k·IQR, where k usually takes a value of 1.5 (for identifying "general outliers") or 3 (for identifying "extreme outliers").

[0077] If a data point p i satisfies one of the following conditions, it is marked as an outlier:

[0078] p i <Q1-k·IQR or p i >Q3+k·IQR;

[0079] For the rainfall data, negative values are directly considered as outliers and removed, as rainfall cannot be negative, while other datasets identified as outliers are marked as missing values to be imputed.

[0080] S13, based on the long short-term memory network model, the deleted data points are imputed to obtain a standard dataset;

[0081] As a preferred embodiment, the long short-term memory network model based on the deleted data point imputation processing to obtain a standard dataset includes the following steps:

[0082] S131, according to the deleted data points, determine the target data points to be imputed, and input the historical rainfall sequence of the target data points and the rainfall data of the adjacent stations into the long short-term memory network model;

[0083] S132, based on the historical rainfall sequence of the target data points and the rainfall data of the adjacent stations, and combined with the long short-term memory network model, the mean square error between the predicted value and the true value is minimized by calculation and multiple rounds of training, and the long short-term memory network model is used to output the time series prediction value of the deleted data points;

[0084] S133, using the time series prediction value of the deleted data points for imputation processing, and arranging in time sequence format to obtain a standard dataset.

[0085] Specifically, for the dataset to be imputed, a long short-term memory network (LSTM) model is used to dynamically impute the time series characteristics (periodicity, trend, seasonality) of the rainfall data to generate high-precision imputed values. The model first inputs the historical rainfall sequence of the target station and the rainfall data of the adjacent stations;

[0086] The historical rainfall sequence of the target station is combined as: t ={p t-n ,p t-n+1 ,...,p t-1};

[0087] The rainfall data set of the adjacent stations is:

[0088] In the formula, n is the length of the time window, is the rainfall data of the adjacent stations;

[0089] Wherein, the LSTM unit contains a forgetting gate, an input gate, an output gate, an updated hidden state h t and a cell state c t . After the calculation of the LSTM unit and the multiple rounds of training to minimize the mean square error between the predicted value and the true value, the final output is the missing or abnormal value p t , and the time series prediction value Replace p t , and record the fill-in value.

[0090] Then the daily rainfall data is sorted into a unified time series format to generate a data set P d ={p1, p2, ···, pN}, where pi represents the daily rainfall on the i-th day (unit: mm), and N is the total number of days. N i

[0091] S14, according to the standard data set, using the composite threshold method to identify extreme rainfall events, and classifying the extreme rainfall events to obtain an extreme rainfall event set.

[0092] As a preferred embodiment, the step of using the composite threshold method to identify extreme rainfall events according to the standard data set, and classifying the extreme rainfall events to obtain an extreme rainfall event set comprises the following steps:

[0093] S141, arranging the standard data set according to the size of rainfall, using linear interpolation method to calculate the percentile of the standard data set, and taking the percentile of the standard data set as a dynamic threshold;

[0094] S142, determining the intensity standard of the extreme rainfall event based on the preset rainfall amount grade standard, and taking the intensity standard of the extreme rainfall event as a fixed threshold;

[0095] S143, integrating the dynamic threshold and the fixed threshold, and constructing a composite threshold judgment rule;

[0096] S144, using the composite threshold judgment rule to identify and classify the extreme rainfall events to obtain an extreme rainfall event set.

[0097] Specifically, based on the processed rainfall data set P d , the 99th percentile is calculated as a dynamic threshold to reflect the extreme characteristics under the regional climate background, and the specific steps are as follows:

[0098] P d is arranged in ascending order according to the size of rainfall to generate an ordered sequence P d '={p1', p2', ···, pN'}, and satisfies p1'≤p2'≤···≤pN'. (1) (2) (N) (1) (2) (N)

[0099] Then the linear interpolation method is used to calculate the 99th percentile T 99 , which satisfies P(p i ≤T 99 ​​​​​​​​) = 0.99, with the formula as follows:

[0100]

[0101] where, is the floor function, T 99 is the dynamic threshold, representing the 99th percentile of daily rainfall, only 1% of rainfall reaches or exceeds this threshold.

[0102] where, the fixed thresholds are set in reference to the national standard GB / T 28592-2012 "Rainfall Grading", aiming to clarify the intensity standards of extreme rainfall events. Specifically, the heavy rain threshold is set to daily rainfall of 50 mm, corresponding to the lower limit of the "heavy rain" grade in the national standard, which is usually considered as the critical point triggering urban waterlogging and traffic paralysis; the heavy storm threshold is set to daily rainfall of 100 mm, corresponding to the "heavy storm" grade, used to assess more severe flood disaster risks. These fixed thresholds are based on meteorological industry specifications and provincial flood emergency response standards, and can directly reflect the potential impact of rainfall events on social and economic activities and infrastructure, ensuring the scientificity and practicality of risk assessment.

[0103] In addition, in the construction of composite threshold determination rules by integrating dynamic thresholds and fixed thresholds to ensure that the recognition results take into account both the climate background and the disaster impact, including:

[0104] A rainfall event is identified as an extreme rainfall event if and only if its daily rainfall p i satisfies both: p i ≧ T 99 , i.e., exceeding the dynamic threshold, embodying the extremity under regional climate background; and p i ≧ 50 mm or p i ≧ 100 mm, i.e., reaching or exceeding the fixed threshold, reflecting the disaster response level.

[0105] The identification process traverses the rainfall dataset p d , performs a double threshold check on each daily rainfall p i . If p i ≧ T 99 and p i ≧ 50 mm, it is marked as an extreme rainfall event and its date, rainfall and site information are recorded.

[0106] where, in generating the sequence of extreme rainfall events E = {(t k , loc k , R k ) | k = 1, 2,..., M}, (loc k = (lon k , lat k) is the site coordinate, t k is the date of the extreme rainfall event, R k is the rainfall intensity feature, M is the total number of identified extreme rainfall events, which can provide input for subsequent analysis. The identified extreme rainfall events are compared with historical disaster records to verify the rationality of the threshold setting and analyze the dynamic threshold T 99 distribution, verify whether it reflects the degree of urbanization (such as heat island effect) and topographic differences (such as low flat river network area).

[0107] S2, based on the Chicago method and the generative adversarial network, an extreme rainfall pattern model is constructed, and the minute-level rainfall intensity curve is determined by combining the extreme rainfall event set using the extreme rainfall pattern model;

[0108] As a preferred embodiment, the Chicago method and the generative adversarial network are used to construct an extreme rainfall pattern model, and the minute-level rainfall intensity curve is determined by combining the extreme rainfall event set using the extreme rainfall pattern model, which includes the following steps:

[0109] S21, according to the daily rainfall data of the meteorological station, based on the preset rainfall time, the heavy rainfall event is extracted, and the heavy rainfall event data set is generated;

[0110] S22, based on the generative adversarial network composed of a generator and a discriminator, an extreme rainfall pattern model is constructed;

[0111] S23, according to the heavy rainfall event data set, the Worshtein loss function is used to train the extreme rainfall pattern model, and the parameters of the generator and the discriminator in the extreme rainfall pattern model are optimized in combination with the total rainfall constraint;

[0112] S24, based on the parameter-optimized extreme rainfall pattern model, the probabilistic minute-level rainfall intensity curve is generated by combining the Chicago method storm intensity formula;

[0113] S25, the probabilistic minute-level rainfall intensity curve is compared with the extreme rainfall event set, the probabilistic minute-level rainfall intensity curve is calibrated and verified by calculating the mean square error, and the final minute-level rainfall intensity curve is obtained.

[0114] Specifically, short duration (≤120 minutes) heavy rainfall events are extracted from historical rainfall data to generate a data set R={r1, r2, ···, r K}, where r K contains the time series of rainfall events, the cumulative rainfall and the peak intensity. The GAN consists of a generator G and a discriminator D, and the goal is to generate realistic rainfall intensity sequences, where the mathematical expression of the discriminator D is:

[0115]

[0116] wherein, represents the discriminator D, represents a random noise vector, X i represents a conditional feature vector (such as peak intensity, total rainfall), θ G is a generator parameter, the generator parameter adopts an LSTM network, and an LSTM unit update formula is:

[0117] f t = σ(W f · [h t-1 , x t ] + b f ) ;

[0118] i t = σ(W i · [h t-1 , x t ] + b i ) ;

[0119]

[0120] o t = σ(W o · [h t-1 , x t ] + b o ), h t = o t · tanh (C t ) ;

[0121] wherein, f t , i t , o t are forget gate, input gate, output gate, C t is a cell state of a previous time step, is a cell state of a previous time step. h t is a hidden state, h t-1 is a hidden state of a previous time step x t is a current time step input, W f , W i , W C , W o are weight matrices of the forget gate, the input gate, the candidate state and the output gate, b f , b i , b C , b o are bias vectors corresponding to the gates, and σ is a Sigmoid activation function,

[0122] A mathematical expression of the discriminator D is:

[0123] D (I; θD ) = σ(CNN(I) or LSTM(I));

[0124] where θ D is the discriminator parameter, a convolutional neural network (CNN) is used to handle the sequence, and I is the input rainfall intensity sequence.

[0125] In addition, the loss function uses the Wasserstein GAN (WGAN) loss to improve the training stability:

[0126]

[0127] where L D is the discriminator loss, the optimization objective is to minimize L G is the generator loss, the optimization objective is to minimize L is the linear interpolation of the real and generated sequences, is the mathematical expectation, which calculates the average value on the distribution, and z ~ p z is random noise, and X is the conditional feature.

[0128] In addition, an additional physical constraint needs to be added to ensure that the total rainfall of the generated sequence is reasonable, and a loss term is added:

[0129]

[0130] where L PHY is the total rainfall, L target is the target total rainfall, is the intensity (mm / min) of the jth time step of the generated sequence, m is the sequence length (120 minutes), and Δt = 1 minute.

[0131] In addition, when training the GAN model, the Adam optimizer is used to update θ G and θ D , the learning rate η ≈ 10 -4 , the number of discriminator training is more than that of the generator (such as 5:1), and the training is stopped when the dynamic time warping (DTW) distance between the generated sequence and the real sequence converges:

[0132]

[0133] where, is the dynamic time warping (DTW) distance, π is the time alignment path, and i, j are both time steps.

[0134] When generating the minute-level rainfall intensity curve in combination with the Chicago method rainfall intensity formula, the Chicago method formula is as follows:

[0135]

[0136] where a, b, c represent the storm intensity parameters under the set return period, r represents the position coefficient of the integrated rain peak, i(t b ) represents the pre-peak rainfall intensity (mm / min); i(t a ) represents the post-peak rainfall intensity (mm / min); and t represents the rainfall duration (min). For each sample, the corresponding conditional rainfall intensity i(x i ,y i |T) is calculated, and the fitting adjustment parameters a, b, and c are fitted according to the return period T and the GAN output to generate the minute-level rainfall intensity curve.

[0137] In the curve calibration and verification, the generated probabilistic rainfall intensity curve is compared with the identified extreme rainfall event set, and the mean square error (MSE) is calculated, and the calculation formula is:

[0138]

[0139] where M represents the number of historical extreme rainfall events, k represents each event in the extreme rainfall event set, R k represents the actual observed rainfall intensity of the kth historical extreme rainfall event, i(t k ,T,q) - R k represents the difference between the predicted rainfall intensity and the actual observed rainfall intensity, and if the MSE exceeds a preset threshold, the Chicago method parameters a, b, and c are adjusted, or the GAN output fitting adjustment parameters are refitted to improve the fitting degree of the curve and the historical data.

[0140] S3, generating an extreme rainfall spatiotemporal distribution feature map according to the extreme rainfall event set and combining a spatial interpolation technology of a fusion deep learning model;

[0141] As a preferred embodiment, the step of generating an extreme rainfall spatiotemporal distribution feature map according to the extreme rainfall event set and combining a spatial interpolation technology of a fusion deep learning model comprises the following steps:

[0142] S31, calculating a dynamic threshold grid according to the extreme rainfall event set by a thin-plate spline interpolation method and a spatiotemporal adaptive interpolation network;

[0143] As a preferred embodiment, the step of calculating a dynamic threshold grid according to the extreme rainfall event set by a thin-plate spline interpolation method and a spatiotemporal adaptive interpolation network comprises the following steps:

[0144] S311, mapping the extreme rainfall event set into a grid system of a first resolution, and generating a gridded data set by a thin-plate spline interpolation method to obtain the extreme rainfall frequency and the extreme rainfall average intensity of the first resolution;

[0145] S312. Based on the set of extreme rainfall events, using a spatiotemporal adaptive interpolation network, combined with a convolutional neural network, a long short-term memory network, and an attention mechanism, a multi-layer perceptron is used to generate the second-resolution extreme rainfall frequency and the extreme rainfall average intensity;

[0146] S313: Fusing the extreme rainfall frequency and the extreme rainfall average intensity of the first resolution with the extreme rainfall frequency and the extreme rainfall average intensity of the second resolution to obtain a fused extreme rainfall frequency and extreme rainfall average intensity, and using the fused extreme rainfall frequency and extreme rainfall average intensity as a dynamic threshold grid.

[0147] S32. Calculate the spatial variation coefficient based on the dynamic threshold grid, and calculate the extreme rainfall frequency and average rainfall intensity of each grid cell using the kriging method based on the set of extreme rainfall events;

[0148] Specifically, the extreme rainfall event set is mapped to a grid system with a resolution of Δx×Δy=0.1°×0.1°, and the ANUSPLIN thin plate spline interpolation method is used to generate a gridded dataset:

[0149]

[0150] In the formula, (x i ,y j ) is the coordinates (latitude and longitude) of the target grid center point, λ k is the basis function coefficient of the kth site, φ(d)=d 2 lnd is the thin plate spline basis function (d is the Euclidean distance), F(x i ,y i ) is the extreme rainfall frequency at grid (i, j) (the extreme rainfall frequency of the first resolution), R(x i ,y i ) is the average intensity of extreme rainfall at grid (i, j) (mm) (average intensity of extreme rainfall at the first resolution), v k =[1,lon k ,lat k ,Z k ] T is the covariate vector (Z k is the site elevation), λ k and a=[a0,a1,a2,a3] T It is determined by solving a system of linear equations. The expression for solving a system of linear equations is:

[0151]

[0152] Where Φ is the basis function matrix, and V is the covariate matrix, is the basis function coefficient vector, F obsis the frequency vector of extreme rainfall measured at the station, and b is the regression coefficient vector of covariates.

[0153] In addition, to overcome the static limitations of the ANUSPLIN method, the spatiotemporal adaptive interpolation network (STAI-Net) is introduced, which combines convolutional neural networks (CNN), long short-term memory networks (LSTM), and attention mechanisms to enhance interpolation accuracy:

[0154] Use CNN to process geographic feature X geo , extract spatial features:

[0155] F spatial =CNN(X geo );

[0156] Where, is the extracted spatial feature, H×W is the spatial dimension of the feature map, and C is the number of channels.

[0157] Then use LSTM to process the historical rainfall time series R hist =[R t-1 ,R t-2 ,…,R t-T ]:

[0158] h t =LSTM(R hist );

[0159] Among them, h t is the time feature vector;

[0160] Then concatenate the spatial and temporal features:

[0161] F st =[flatten(F spatial ),h t ];

[0162] Where, F st It represents the concatenation of spatial and temporal features. Flatten means flattening the three-dimensional tensor of multi-dimensional features into a one-dimensional vector and concatenating it with the time feature vector to obtain a one-dimensional feature vector.

[0163] Then we introduce the spatial attention mechanism to dynamically focus on high-risk areas:

[0164] A spatial =σ(Conv(F spatial ));

[0165]

[0166] Where A spatialSpatial Attention Weights, representing the importance of each spatial location (e.g. grid cell), with values in [0, 1], for dynamically focusing on high-risk areas, Weighted Spatial Features, representing the spatial features after being enhanced or suppressed by applying the attention mechanism. σ represents the sigmoid function, and ⊙ represents element-wise multiplication.

[0167] The rainfall frequency and intensity of the high-resolution (0.05° × 0.05°) grid are generated by a multi-layer perceptron (MLP):

[0168]

[0169] where, Extreme Rainfall Frequency at High Resolution (Second Resolution), Extreme Rainfall Mean Intensity at High Resolution (Second Resolution).

[0170] The ANUSPLIN interpolation results and the STAI-Net prediction results are weighted and fused:

[0171]

[0172] where, F final (x i ,y j ) represents the fused extreme rainfall frequency, R final (x i ,y j ) represents the fused extreme rainfall mean intensity, and α ∈ [0, 1] is the weight coefficient, which is optimized through cross-validation.

[0173] S33, according to the extreme rainfall frequency and rainfall mean intensity of each grid cell, and according to the preset dimension, an extreme rainfall spatiotemporal distribution feature map is generated.

[0174] As a preferred embodiment, the calculation of the spatial variation coefficient according to the dynamic threshold grid, and the calculation of the extreme rainfall frequency and rainfall mean intensity of each grid cell by the Kriging method according to the set of extreme rainfall events include the following steps:

[0175] S321, based on the dynamic threshold grid, the percentile rainfall threshold is calculated, and the spatial mean and spatial standard deviation of the percentile rainfall threshold are determined;

[0176] S322、According to the spatial average value and the spatial standard deviation, a spatial coefficient of variation is calculated, and whether the dynamic threshold grid needs threshold correction is determined according to the spatial coefficient of variation. If needed, the dynamic threshold grid is adjusted by using a correction term formula to obtain a corrected dynamic threshold; if not needed, the dynamic threshold grid is taken as the corrected dynamic threshold.

[0177] S323、Based on the corrected dynamic threshold and the set of extreme rainfall events, the initial extreme rainfall frequency and rainfall average intensity of each grid cell are calculated.

[0178] S324、The spatial and temporal adaptive interpolation network optimizes the extreme rainfall frequency and rainfall average intensity of each grid cell to obtain the optimized extreme rainfall frequency and rainfall average intensity of each grid cell.

[0179] S325、Based on the geographic information system tool and combined with the Kriging method, the spatial interpolation and visualization processing are performed on the optimized extreme rainfall frequency and rainfall average intensity of each grid cell to obtain the final extreme rainfall frequency and rainfall average intensity of each grid cell.

[0180] Specifically, based on the generated dynamic threshold grid F final (x i ,y j ), the 99th percentile rainfall threshold F 99％ (x i ,y i ) is calculated, and the spatial coefficient of variation CV d is calculated to quantify the spatial variability of the dynamic threshold in the study area. Wherein, F 99％ (x i ,y i ) represents the 99th percentile rainfall threshold of each grid cell in the generated grid data set based on the ANUSPLIN interpolation method. The coefficient of variation formula is:

[0181]

[0182] In the formula, is the spatial average value of the 99th percentile threshold extreme rainfall frequency, and the calculation formula is:

[0183]

[0184] In the formula, is the spatial standard deviation of the 99th percentile threshold extreme rainfall frequency, and the calculation formula is:

[0185]

[0186] In the formula, N grid is the total number of grid cells, that is, the number of all grids in the study area.

[0187] Then, according to the calculated coefficient of variation CV d , it is determined whether threshold partition correction is needed: when CV d ≥ 5%, it indicates that there is significant spatial variation in the dynamic threshold grid, and the threshold partition correction module needs to be triggered. By correcting the dynamic threshold, the analysis error caused by spatial variability is reduced, and the accuracy of the extreme rainfall distribution characteristics is improved. If the threshold partition correction module is triggered, a terrain correction term Δ t h is generated to adjust the dynamic threshold grid. The basic correction term formula is:

[0188] Δ Γk (x i ,y j )=Γ·Z(x i ,y j )+θ·S(x i ,y j );

[0189] In the formula, Δ Nk (x i ,y j ) is the dynamic threshold grid correction term, Z(x i ,y i ) is the terrain elevation value of the grid cell (x, y), S(x i ,y j ) is the slope, and Γ and θ are correction coefficients representing the degree of influence of terrain elevation on the dynamic threshold, which are determined by regression analysis of historical rainfall data output by a deep learning model (STAI-Net). Γ and θ satisfy:

[0190] [Γ,θ]=MLP(F st );

[0191] The calculation formula for correcting the dynamic threshold is:

[0192]

[0193] In the formula, h represents the corrected dynamic threshold.

[0194] In addition, based on the corrected dynamic threshold and the set of extreme rainfall events, the extreme rainfall frequency F i,j and the average rainfall intensity I i,j of each grid cell are calculated, and their calculation formulas are respectively:

[0195]

[0196] In the formula, i and j are the row and column indices of the grid cell, M is the total number of extreme rainfall events, and R kis the rainfall intensity (unit: mm) of the kth extreme rainfall event, I k (i,j) represents whether the kth extreme rainfall event occurs within the grid cell (i,j).

[0197] Then, the STAI-Net is further optimized to generate a higher-precision grid by upsampling:

[0198]

[0199] wherein, is the higher-resolution extreme rainfall frequency, is the higher-resolution average rainfall intensity, and Deconv is a deconvolution layer, is the weighted spatiotemporal feature.

[0200] The rasterized datasets of the higher-resolution extreme rainfall frequency and the higher-resolution average rainfall intensity are spatially interpolated and visualized using the GIS tool ArcGIS, wherein the Kriging method is used to ensure spatial continuity, and the Kriging interpolation is based on the following formula:

[0201]

[0202] wherein, Z(s0) is the value of the point to be interpolated, ε k is the weight, and Z(s k ) is the value of the known point.

[0203] The spatial distribution of or is represented by a color gradient, and the contours of the frequency or intensity are drawn to highlight the spatial variation characteristics. To show the spatiotemporal characteristics of the extreme rainfall, the corresponding spatial distribution maps of and are generated for each period calculated in years, as shown in Figures 2-7 .

[0204] S4, based on the minute-level rainfall intensity curve, combined with the spatial distribution characteristics map of the extreme rainfall, the drainage capacity of the urban drainage system is evaluated, and according to the evaluation result, the urban drainage system is optimized.

[0205] As a preferred embodiment, the step of evaluating the drainage capacity of the urban drainage system based on the minute-level rainfall intensity curve, combined with the spatial distribution characteristics map of the extreme rainfall, and optimizing the urban drainage system according to the evaluation result comprises the following steps:

[0206] S41, based on the minute-level rainfall intensity curve, combined with the grid rainfall intensity and frequency in the spatial and temporal distribution characteristic map of extreme rainfall, the rainfall intensity of each grid unit under different return periods is calculated, and compared with the design capacity of the drainage system, the drainage capacity of each grid unit is evaluated, and the visualization result of the spatial and temporal distribution characteristic map is used to identify the area with insufficient drainage capacity;

[0207] S42, according to the spatial variability of the spatial and temporal distribution characteristic map, the drainage capacity difference of each grid unit is calculated, and combined with the rainfall intensity distribution in the spatial and temporal distribution characteristic map, the risk area is determined;

[0208] S43, according to the drainage capacity difference of the drainage system under different rainfall intensities, combined with the risk area in the spatial and temporal distribution characteristic map, the optimization scheme of the waterlogging prevention facility is determined, and the optimization scheme of the waterlogging prevention facility is adjusted through the waterlogging risk analysis, and the final optimization scheme of the waterlogging prevention facility is obtained.

[0209] Specifically, according to the drainage capacity difference of the drainage system under different rainfall intensities, and based on the high-risk area and waterlogging risk in the spatial and temporal distribution characteristic map, the optimization scheme of the waterlogging prevention facility is formulated, and the drainage pipe diameter is adjusted and the rainwater collection facility is added. Through waterlogging risk analysis, combined with the spatial distribution characteristics of the map, the drainage area is optimized, and the final optimization scheme of the waterlogging prevention facility is generated.

[0210] As a preferred embodiment, the optimization scheme of the waterlogging prevention facility is determined according to the drainage capacity difference of the drainage system under different rainfall intensities, combined with the risk area in the spatial and temporal distribution characteristic map, and the final optimization scheme of the waterlogging prevention facility is obtained by adjusting the optimization scheme of the waterlogging prevention facility through waterlogging risk analysis, comprising the following steps:

[0211] S431, according to the drainage capacity difference of the drainage system under different rainfall intensities, and based on the design basis of the expanded drainage pipe, combined with the risk area in the spatial and temporal distribution characteristic map, the optimization scheme of the waterlogging prevention facility is determined;

[0212] S432, according to the optimization scheme of the waterlogging prevention facility, the drainage pipe diameter is adjusted and the rainwater collection facility is added, and the waterlogging risk of each area in the drainage system is calculated;

[0213] S433, based on the waterlogging risk of each area in the drainage system, the optimized drainage area is determined by optimizing the area formula, and the optimization area design scheme of the waterlogging prevention facility is determined;

[0214] S434, based on the optimization area design scheme of the waterlogging prevention facility, the optimization scheme of the waterlogging prevention facility is adjusted, and the final optimization scheme of the waterlogging prevention facility is obtained.

[0215] Specifically, the rainfall intensity I(T) under different return periods T is calculated using the minute-level rainfall intensity curve generated based on the Chicago method, where the formula for calculating the rainfall intensity is:

[0216] I(T) = A T -Ψ ;

[0217] In the formula, I(T) is the rainfall intensity under the return period T (unit: mm / min), A and Ψ are parameters fitted by historical rainfall data, and T is the return period (year).

[0218] Generate a high-resolution rainfall atlas:

[0219]

[0220] In the formula, represents the high-resolution rainfall intensity distribution, covering the spatial and temporal distribution characteristics.

[0221] The formula for identifying the risk of water accumulation is:

[0222]

[0223] In the formula, represents the mean value of the high-resolution rainfall intensity distribution, is the standard deviation, and κ is the risk factor coefficient.

[0224] Then, according to the calculated rainfall intensity I(T) and the water accumulation risk value, the drainage capacity Q drain (T) of the urban drainage system is evaluated. The formula for calculating the drainage capacity of the drainage system is:

[0225] Q drain (T) = v A drain d(T);

[0226] In the formula, v is the flow rate of the drainage pipe (unit: m / s), A drain is the cross-sectional area of the drainage pipe (unit: m 2 ), and d(T) is the pipe diameter under different return periods (unit: m), which is calculated by the design diameter Q design (T) of the pipe, and the calculation formula is:

[0227]

[0228] In the formula, Q drain (T) is the design drainage volume (unit: m 3 / s).

[0229] Based on the calculated rainfall intensity and drainage capacity, the area of the drainage system that is insufficient under different rainfall intensities is determined. The difference in drainage capacity of this area is calculated, and the formula is:

[0230] ΔQ(T) = Q design (T) - Q drain (T);

[0231] wherein ΔQ(T) is the drainage capacity difference, reflecting the deficiency of the drainage system under a certain return period.

[0232] The calculated drainage capacity difference ΔQ(T) outputs the optimization scheme of the waterlogging prevention facilities, which includes but is not limited to measures such as expanding the drainage pipeline, increasing the rainwater collection facilities, and improving the storage capacity, and the design basis of the expanded drainage pipeline is:

[0233]

[0234] wherein Q new (T) is the optimized drainage capacity (unit: m 3 / s), and the calculation formula is:

[0235] Q new (T) = Q design (T) + ΔQ(T);

[0236] The Q new (T) calculated by the above formula ensures that the drainage capacity of the drainage pipeline meets the design requirements.

[0237] Then, based on the output optimization scheme, the waterlogging risk of different regions is further analyzed, and the waterlogging risk R flood (T) of each region is calculated, and the calculation formula is:

[0238]

[0239] wherein R flood (T) is the waterlogging risk, indicating the waterlogging depth under a certain return period, A flood is the area of the affected region (unit: m 2 ), and Q drain (T) is the drainage capacity.

[0240] After calculating the waterlogging risk, the optimization regional design scheme of the waterlogging prevention facilities is output. Specifically, the optimization regional area Aopt(T) is calculated according to the following formula:

[0241]

[0242] wherein A opt (T) is the newly added drainage regional area after optimization (unit: m 2 ), and RiskFactor iThe risk factor for each area is calculated based on the relationship between the depth of water accumulation and drainage capacity. flood,i is the actual waterlogging area of ​​each region, and U represents the number of drainage areas.

[0243] Finally, combined with the optimized drainage area A opt (T) and new drainage pipes, rainwater collection facilities, storage capacity and other measures, prioritize grid cells with high rainfall intensity and frequency, optimize the design of flood control facilities, and output urban flood control design plans under different recurrence periods to ensure that the drainage system can effectively respond to extreme rainfall events and reduce the risk of urban waterlogging and flood disasters.

[0244] It should be noted that the present invention provides strong technical support for building resilient power grids and key infrastructure in high-density urban areas under the background of extreme weather, especially in improving the flood prevention and disaster resistance capabilities of power grids. By accurately identifying and simulating extreme rainfall events, this method can effectively assess the potential threats of urban drainage systems to power facilities and accurately identify high-risk areas where power facilities may be flooded due to insufficient drainage. Based on this, the method can propose targeted optimization solutions (such as expanding drainage pipes, adding rainwater collection facilities, etc.), significantly reducing the risk of power grid operation interruption caused by urban floods. More importantly, the method can scientifically assess the flood risk level of a specific area and plan and deploy effective power grid flood prevention and disaster reduction measures based on this (such as optimizing substation site selection, improving flood prevention standards for key power facilities, and formulating emergency response plans). Compared with traditional methods, the present invention has significant advantages in spatiotemporal feature extraction and rainfall simulation accuracy, making it particularly suitable for power grid flood prevention planning in areas with complex terrain and high urbanization. Therefore, the present invention has important scientific guiding significance and broad application and promotion prospects for improving the disaster prevention and mitigation resilience of urban power grids, ensuring the security and stability of power supply, and maintaining the normal operation of the social economy.

[0245] Based on the above-mentioned urban extreme rainfall risk assessment and drainage optimization technology integrating the composite threshold and the Chicago method, a simulation analysis is performed in this embodiment, as follows:

[0246] This example takes a certain city area as an example, uses the rainfall data of national meteorological stations from 1961 to 2020, and constructs a high-precision raster dataset by integrating the ANUSPLIN spline interpolation method (0.5°×0.5°) of the deep learning model and combining it with elevation correction.

[0247] The spatial distribution of extreme rainfall frequency extracted according to different extreme rainfall thresholds is as follows: Figures 2-4 As shown, in Figure 2In the 99th percentile threshold method, the frequency of extreme rainfall in Shanghai presents a homogeneous distribution, with an annual average frequency of approximately 3.63-3.645 days, and a small spatial difference, with only slight fluctuations in local areas. Figure 3 When the threshold is set at daily rainfall exceeding 50 mm, the spatial distribution is more obvious, with the highest frequency of extreme rainfall events in the central urban area, especially along the Huangpu River, at approximately 2.5-3 times per year, and significantly lower in peripheral areas such as Chongming Island and the western part of the city, with a frequency of 0.5-1.5 days, showing a decreasing trend from the city center to the periphery. Figure 4 When the daily rainfall threshold is increased to 100 mm, the frequency of extreme rainfall in the city decreases significantly, with an annual average frequency of less than 0.4 days, and the spatial difference is significantly reduced, but the trend of higher frequency in the central urban area than in the peripheral areas is still observed, indicating that extreme heavy rainfall tends to concentrate in the core urban area.

[0248] The spatial distribution of the annual average extreme rainfall in Shanghai extracted based on different extreme rainfall thresholds is shown in Figures 5-7 As can be seen from the figure, there is a trend of gradually decreasing from the city center to the periphery. This phenomenon may be closely related to the concentration of local heavy rainfall events caused by the urban heat island effect. Specifically, Figure 5 In the 99th percentile threshold method, the annual extreme rainfall mainly ranges from 180 to 240 mm, with relatively high rainfall in the central urban area and along the Huangpu River, with some areas even exceeding 240 mm. Figure 6 The distribution of extreme rainfall with a daily rainfall exceeding 50 mm as the threshold is more complex, with high values in the urban core area, with rainfall ranging from 160 to 220 mm, and significantly lower values in the peripheral areas, mostly within the range of 100-140 mm. In Figure 7 When the extreme rainfall threshold is further increased to 100 mm / d, the extreme rainfall significantly decreases, with an annual average extreme rainfall of less than 150 mm in most areas of the city, and only a few central or southwestern local areas reaching around 170 mm, with a significantly reduced overall spatial difference, reflecting the local nature of extreme heavy rainfall events in the urban center.

[0249] Based on the spatial analysis of the historical period, combined with the characteristics of Shanghai's climate and urbanization, the Chicago method was used to simulate 120-minute rainstorm patterns under different return periods (2-100 years).

[0250] From Figure 8The simulation results show that, as the return period increases from 2 years to 100 years, the peak rainfall intensity increases sharply from 3.3 mm / min to 7.2 mm / min, and the 120-minute cumulative rainfall is significantly increased from 60.8 mm to 129.9 mm. The 100-year return period simulation value (129.9 mm) is highly consistent with the 115 mm / 120-minute rainfall record measured in the center of Shanghai during Typhoon "Yanhuai" in 2021, fully verifying the model's ability to simulate short-duration extreme rainfall events. In addition, all return period rain types are of the pre-peak type (rain peak in the first 30 minutes), which is consistent with the observed characteristics of historical rainfall processes in Shanghai (such as Typhoon "Fit" in 2013). This rain type leads to rapid runoff convergence, causing instantaneous overload pressure on the drainage network in the city center (design standards are mostly 5-10 years), highlighting the urgency of network expansion and storage facility construction.

[0251] In summary, with the above technical solutions of the present application, the extreme rainfall characteristic analysis of the present application improves the accuracy and practicality, providing a scientific basis for urban flood control planning and resilient infrastructure construction. By integrating the composite threshold method and the Chicago method, combined with the ANUSPLIN spline interpolation and terrain correction techniques of the deep learning model, this method can accurately identify extreme rainfall events and generate high-resolution minute-level rainfall intensity curves, breaking through the limitations of traditional methods in temporal and spatial distribution simulation and long return period accuracy; the analysis results can provide reliable data support for power system planners, optimize the design of flood control measures, and protect power facilities from flood damage; in addition, this method has strong adaptability and can be popularized to urban areas with different terrain and climate conditions, providing an efficient technical tool for coping with extreme weather events under the background of climate change and improving the disaster prevention and reduction capacity of the power system, which has important social and economic value.

[0252] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0253] The above specific embodiments further illustrate the purpose, technical solutions and advantages of the present application, and it should be understood that the above description is only for specific embodiments of the present application and is not intended to limit the protection scope of the present application, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for urban extreme rainfall risk assessment and drainage optimization that integrates composite thresholds and the Chicago method, characterized in that: The following steps are involved: S1. Obtain daily rainfall data from meteorological stations and use the interquartile range method and composite threshold method to identify and classify extreme rainfall events to obtain a set of extreme rainfall events. S2. Based on the Chicago method and generative adversarial networks, an extreme rainfall pattern model is constructed. The extreme rainfall pattern model is then combined with a set of extreme rainfall events to determine the minute-level rainfall intensity curve. S3. Based on the extreme rainfall event set, combined with the spatial interpolation technology of the fusion deep learning model, generate the spatiotemporal distribution characteristic map of extreme rainfall; S4. Based on the minute-level rainfall intensity curve and the spatial and temporal distribution characteristic map of extreme rainfall, the drainage capacity of the urban drainage system is evaluated, and the urban drainage system is optimized based on the evaluation results.

2. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 1 is characterized in that: The method of obtaining daily rainfall data from meteorological stations and identifying and classifying extreme rainfall events using the interquartile range method and the composite threshold method to obtain a set of extreme rainfall events comprises the following steps: S11. Collect daily rainfall data from meteorological stations, arrange the daily rainfall dataset in ascending order, and determine the quartiles of the daily rainfall dataset based on the arrangement result; S12. Calculate the interquartile range of the daily rainfall dataset based on the quartiles of the daily rainfall dataset, determine a screening range based on the interquartile range, determine a screening condition based on the screening range, and delete data points in the daily rainfall dataset that do not meet the screening condition; S13. Based on the long short-term memory network model, the deleted data points are filled in to obtain a standard data set; S14. Based on the standard data set, the composite threshold method is used to identify extreme rainfall events, and the extreme rainfall events are classified and processed to obtain a set of extreme rainfall events.

3. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 2 is characterized in that: The method of filling in the deleted data points based on the long short-term memory network model to obtain a standard data set includes the following steps: S131. Determine a target data point to be filled based on the deleted data point, and input the historical rainfall sequence of the target data point and the rainfall data of its neighboring stations into the long short-term memory network model; S132. Based on the historical rainfall sequence of the target data point and the rainfall data of its neighboring stations, and in combination with the long short-term memory network model, the mean square error between the predicted value and the true value is minimized through calculation and multiple rounds of training, and the long short-term memory network model is used to output the time series prediction value of the deleted data point; S133. Fill in the data points deleted by the time series prediction values ​​and organize them according to the time series format to obtain a standard data set.

4. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 2 is characterized in that: The method of identifying extreme rainfall events using a composite threshold method based on a standard data set and classifying the extreme rainfall events to obtain an extreme rainfall event set includes the following steps: S141, arranging the standard data set according to the size of rainfall, calculating the percentile of the standard data set using linear interpolation, and using the percentile of the standard data set as a dynamic threshold; S142. Determine an intensity standard for an extreme rainfall event based on a preset precipitation level standard, and use the intensity standard for the extreme rainfall event as a fixed threshold; S143, integrating the dynamic threshold and the fixed threshold, and constructing a composite threshold determination rule; S144. Identify and classify extreme rainfall events using a composite threshold judgment rule to obtain an extreme rainfall event set.

5. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 1 is characterized in that: The method of constructing an extreme rainfall pattern model based on the Chicago method and a generative adversarial network, and determining a minute-level rainfall intensity curve by using the extreme rainfall pattern model in combination with a set of extreme rainfall events, includes the following steps: S21. Extract heavy rainfall events based on daily rainfall data from meteorological stations and preset rainfall times, and generate a heavy rainfall event dataset; S22. Build an extreme rainfall pattern model based on a generative adversarial network consisting of a generator and a discriminator. S23. Based on the heavy rainfall event dataset, the extreme rainfall pattern model is trained using the Warschelstein loss function. Combined with the total rainfall constraint, the parameters of the generator and discriminator in the extreme rainfall pattern model are optimized. S24. Based on the extreme rainfall pattern model after parameter optimization and combined with the Chicago rainstorm intensity formula, a probabilistic minute-level rainfall intensity curve is generated; S25. Compare the probabilistic minute-level rainfall intensity curve with the set of extreme rainfall events, calibrate and verify the probabilistic minute-level rainfall intensity curve by calculating the mean square error, and obtain the final minute-level rainfall intensity curve.

6. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 1 is characterized in that: The generation of a spatiotemporal distribution feature map of extreme rainfall based on a set of extreme rainfall events and a spatial interpolation technique integrating a deep learning model comprises the following steps: S31. Calculate the dynamic threshold grid based on the set of extreme rainfall events using thin plate spline interpolation and spatiotemporal adaptive interpolation network; S32. Calculate the spatial variation coefficient based on the dynamic threshold grid, and calculate the extreme rainfall frequency and average rainfall intensity of each grid cell using the kriging method based on the set of extreme rainfall events; S33. Generate a spatiotemporal distribution characteristic map of extreme rainfall based on the extreme rainfall frequency and average rainfall intensity of each grid cell and a preset dimension.

7. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 6 is characterized in that: The method of calculating the dynamic threshold grid based on the extreme rainfall event set by thin plate spline interpolation and spatiotemporal adaptive interpolation network includes the following steps: S311, mapping the extreme rainfall event set to a grid system of a first resolution, and generating a rasterized dataset using a thin plate spline interpolation method to obtain the extreme rainfall frequency and the extreme rainfall average intensity of the first resolution; S312. Based on the set of extreme rainfall events, using a spatiotemporal adaptive interpolation network, combined with a convolutional neural network, a long short-term memory network, and an attention mechanism, a multi-layer perceptron is used to generate the second-resolution extreme rainfall frequency and the extreme rainfall average intensity; S313: Fusing the extreme rainfall frequency and the extreme rainfall average intensity of the first resolution with the extreme rainfall frequency and the extreme rainfall average intensity of the second resolution to obtain a fused extreme rainfall frequency and extreme rainfall average intensity, and using the fused extreme rainfall frequency and extreme rainfall average intensity as a dynamic threshold grid.

8. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 6 is characterized in that: The method of calculating the spatial variation coefficient based on the dynamic threshold grid and calculating the extreme rainfall frequency and average rainfall intensity of each grid cell by the Kriging method based on the extreme rainfall event set includes the following steps: S321. Calculate percentile rainfall thresholds based on the dynamic threshold grid, and determine the spatial mean and spatial standard deviation of the percentile rainfall thresholds; S322. Calculate the spatial coefficient of variation based on the spatial mean and the spatial standard deviation, and determine whether the dynamic threshold grid requires threshold correction based on the spatial coefficient of variation. If necessary, adjust the dynamic threshold grid using the correction term formula to obtain a corrected dynamic threshold. If not, use the dynamic threshold grid as the corrected dynamic threshold. S323, calculating the initial extreme rainfall frequency and average rainfall intensity of each grid cell based on the modified dynamic threshold and the extreme rainfall event set; S324, using a spatiotemporal adaptive interpolation network to optimize the extreme rainfall frequency and average rainfall intensity of each grid cell, thereby obtaining optimized extreme rainfall frequency and average rainfall intensity of each grid cell; S325. Based on geographic information system tools and combined with the Kriging method, the optimized extreme rainfall frequency and average rainfall intensity of each grid cell are spatially interpolated and visualized to obtain the final extreme rainfall frequency and average rainfall intensity of each grid cell.

9. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 1 is characterized in that: The method of evaluating the drainage capacity of the urban drainage system based on the minute-level rainfall intensity curve and combining the temporal and spatial distribution characteristic map of extreme rainfall, and optimizing the urban drainage system according to the evaluation results, includes the following steps: S41. Based on minute-level rainfall intensity curves and combined with gridded rainfall intensity and frequency from the extreme rainfall spatiotemporal distribution characteristic map, calculate the rainfall intensity for each grid cell at different return periods. Compare this with the drainage system design capacity to assess the drainage capacity of each grid cell. Use the visualization of the spatiotemporal distribution characteristic map to identify areas with insufficient drainage capacity. S42. Calculate the drainage capacity difference of each grid cell based on the spatial variability of the spatiotemporal distribution characteristic map, and determine the risk area based on the rainfall intensity distribution in the spatiotemporal distribution characteristic map; S43. Based on the differences in drainage capacity of the drainage system under different precipitation intensities and combined with the risk areas in the spatiotemporal distribution characteristic map, the optimization plan for flood control facilities is determined. Through waterlogging risk analysis, the optimization plan for flood control facilities is adjusted to obtain the final optimization plan for flood control facilities.

10. The urban extreme rainfall risk assessment and drainage optimization method integrating composite threshold and Chicago method according to claim 9 is characterized in that: The method of determining the optimal flood control facility plan based on the drainage capacity differences of the drainage system under different precipitation intensities and the risk areas in the spatiotemporal distribution characteristic map, and adjusting the optimal flood control facility plan through waterlogging risk analysis to obtain the final optimal flood control facility plan includes the following steps: S431. Determine the optimization plan for flood control facilities based on the drainage capacity differences of the drainage system under different precipitation intensities, the design basis for expanding drainage pipes, and the risk areas in the spatiotemporal distribution characteristic map; S432. Based on the flood control facility optimization plan, adjust the diameter of drainage pipes and add rainwater collection facilities, and calculate the risk of waterlogging in each area of ​​the drainage system; S433. Based on the waterlogging risk of each area within the drainage system, determine the optimized drainage area by optimizing the regional area formula and determine the optimized regional design plan for flood control facilities; S434. Based on the optimized regional design plan of flood control facilities, the flood control facility optimization plan is adjusted to obtain a final flood control facility optimization plan.

Images