CN113344305A - Rapid prediction method for rainstorm waterlogging event - Google Patents
Rapid prediction method for rainstorm waterlogging event Download PDFInfo
- Publication number
- CN113344305A CN113344305A CN202110895780.0A CN202110895780A CN113344305A CN 113344305 A CN113344305 A CN 113344305A CN 202110895780 A CN202110895780 A CN 202110895780A CN 113344305 A CN113344305 A CN 113344305A
- Authority
- CN
- China
- Prior art keywords
- model
- layer
- neural network
- waterlogging
- rainfall
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
- G06Q50/265—Personal security, identity or safety
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Abstract
The invention provides a rapid prediction method for a rainstorm waterlogging event, which comprises the following steps of acquiring various rainfall processes: constructing an urban flood simulation refinement model; checking the urban flood simulation refinement model by utilizing the various rainfall processes to obtain a rainstorm-waterlogging sample; constructing a neural network model of each water accumulation point by using 90% of the rainstorm-waterlogging samples, and training the neural network model; predicting the ponding process which does not participate in model training by using the trained neural network model; by calculating the coefficients of determination of the predicted data and the simulated dataR 2To judge the similarity of the prediction results. The invention provides a riotA rapid prediction method for a rainfall waterlogging event combines a traditional numerical simulation model and an artificial neural network, generates a large number of rainstorm-waterlogging samples under different rainfall conditions through the numerical simulation model, and trains a BP neural network waterlogging prediction model according to the samples.
Description
Technical Field
The invention relates to the technical field of hydraulic engineering, in particular to a rapid prediction method for a rainstorm waterlogging event.
Background
With the change of global climate environment, the frequency of extreme rainfall is higher and higher, and urban waterlogging disasters are frequent. Along with the rapid development of urban scale, population and urban economy become more and more centralized, losses caused by extreme rainfall and associated secondary disasters are multiplied, a plurality of cities such as Beijing, Guangzhou, Shenzhen, Wuhan all suffer from larger flood disasters, and urban heavy rainstorm becomes a main disaster affecting urban construction and normal life of China. The urban flood control has important scientific value and strategic significance for guaranteeing national water safety and supporting sustainable development of socioeconomic performance.
Before the heavy rainstorm comes, different project scheduling schemes are arranged according to the rainstorm condition issued by a meteorological department, the urban waterlogging risk is estimated in advance, and the disaster prevention, reduction and relief measures can be more pertinent.
In the past, urban inland inundation risk analysis methods utilize a hydrodynamic model to simulate and evaluate inland inundation risks generated by upcoming rainstorm in advance according to the rain condition, water condition and working condition. At present, the urban waterlogging simulation model technology is developed more mature, such as a one-dimensional drainage model, a two-dimensional shallow water model, a hydrologic drive hydrodynamic model based on a dotted boundary and the like, and the urban waterlogging condition can be well simulated. However, the mathematical model is complex to construct and the calculation time is slow, so that the requirement of urban flood prevention emergency is difficult to meet in terms of time efficiency. It is necessary to find a new way to set different rain conditions, water conditions, work conditions and other conditions in advance for the upcoming heavy rain weather, analyze different water conditions and work conditions under the current rainfall condition, quickly analyze and predict the urban waterlogging risk condition, and provide scientific basis for timely adjusting the dispatching operation condition of the water conservancy facilities and making emergency rescue measures.
In recent ten years, Artificial Intelligence (AI) technology has been developed in the fields of computer vision, natural language processing, machine translation, robotics, and biological information control. Currently, AI is also primarily applied successfully in weather forecast aid decision making and some extreme disaster identification. However, AI techniques do not depart from the training of large numbers of high quality data samples. In fact, both from the rainfall field and the actually measured waterlogging data, the demands of the AI technology on the samples can not be met far from the quality and quantity, so that the effective application of the AI technology in flood control and disaster reduction is limited.
Disclosure of Invention
In order to solve the technical problem, the invention provides a rapid rainstorm and waterlogging event prediction method, which combines a traditional numerical simulation model and an artificial neural network, generates a large number of rainstorm-waterlogging samples under different rainfall conditions through the numerical simulation model, and trains a BP neural network waterlogging prediction model according to the samples. Compared with the actual measurement accumulated water monitoring data and the numerical simulation result, the prediction result of the method has small errors, can effectively replace the prediction result of the numerical simulation, greatly shortens the waterlogging risk prediction time, and effectively solves the timeliness problem of urban waterlogging prediction and early warning
The invention aims to provide a rapid prediction method for a rainstorm waterlogging event, which comprises the following steps of acquiring various rainfall processes:
step 1: constructing an urban flood simulation refinement model;
step 2: checking the urban flood simulation refinement model by utilizing the various rainfall processes to obtain a rainstorm-waterlogging sample;
and step 3: constructing a neural network model of each water accumulation point by using 90% of the rainstorm-waterlogging samples, and training the neural network model;
and 4, step 4: predicting the ponding process which does not participate in model training by using the trained neural network model;
and 5: by calculating the coefficients of determination of the predicted data and the simulated dataR 2To judge the similarity of the prediction results.
Preferably, the rainfall process includes a short duration heavy rainfall process, a characteristic rainfall process of different temporal and spatial distributions, and a design rainfall process designed according to a rainstorm intensity formula.
In any of the above aspects, it is preferred that the short duration heavy rainfall event is a short duration rainstorm event within 3 hours of the actual occurrence of the duration of rainfall.
In any of the above solutions, it is preferable that the characteristic rainfall process of different space-time distributions is a rainfall process which conforms to different types of space-time distribution characteristics and has a 3-hour single-station accumulated rainfall from 30mm to 500 mm.
In any of the above aspects, it is preferable that the designed rainfall process designed according to the rainstorm intensity formula is a chicago type rainfall process in which the rainfall is generally downtown and the 3-hour cumulative rainfall is from 30mm to 500 mm.
In any of the above schemes, preferably, the urban flood simulation refinement model is composed of a one-dimensional river network model, a two-dimensional earth surface model and an underground pipe network model.
In any of the above schemes, preferably, the one-dimensional river network model is simulated by a one-dimensional hydrodynamic model method, including a river, a section, a gate dam and a water storage zone.
In any of the above schemes, preferably, the two-dimensional earth surface model is subdivided by using irregular grids, and the water-blocking micro-terrain in an important area is used as a grid in a control line construction range.
In any of the above schemes, preferably, the important area includes an intersection, a road, a water body, a river, a railway, a cell and a bridge.
In any of the above schemes, preferably, the underground pipe network model comprises a rainwater pipe section, an inspection well, a rain grate and a pump station.
In any of the above schemes, preferably, the step 2 includes simulating a production and convergence process of rainwater in a river, the earth surface and an underground pipe network by the one-dimensional river network model, the two-dimensional earth surface model and the underground pipe network model through coupling calculation, and describing an evolution process of waterlogging and ponding in an urban underlying surface in detail.
In any of the above aspects, preferably, the neural network model haskA number of input elements which are each arranged in a row,sa hidden layer, each hidden layer having neurons asn 1 、n 2 、……、n s The number of the main components is one,mand outputting the vector.
In any of the above aspects, preferably, the calculation method of the neural network model includes the following substeps:
step 31: determining the number of input layer nodes of the neural network modelkImplicit layer by layer numbersAnd number of nodes in each hidden layernAnd number of output layer nodesmPreliminarily determining the weight between layersω ij Threshold of hidden layerb ij ；
Step 32: computing hidden layer output
，
Wherein the content of the first and second substances,
is as followssFirst layer of neuronsjThe output of the first and second processors is,n s is as followssThe number of layer neurons in the layer,
is as followss-1 layer ofjThe nerve cell and the firstsLayer oneiThe weight of the link between the individual neurons,
is as followssLayer 1 neuronal secondjThe output of the first and second processors is,
is composed ofS thLayer oneiThe bias of the individual neurons is such that,
is an activation function of the neuron;
step 33: calculating an output variableO k ，
Wherein the content of the first and second substances,
is as followssFirst layer of neuronsjThe output of the first and second processors is,b k as an output layerkThe bias of the individual neurons is such that,
is as followssLayer onejThe individual neuron and the output layerkThe weight of the link to each of the neurons, k=1,2,…,m，
is an activation function of the neuron;
step 34: substituting the initial weight and the threshold value into the neural network to carry out normal feedforward training to obtain the error between the actual output value and the expected output value of the calculated sample
，
Wherein the content of the first and second substances,Nthe number of neurons in the output layer is,y k to a desired outputThe value of the one or more of the one,O k is the actual output value;
step 35: adjusting the weight and threshold of the network according to the normal training principle, and performing feedforward training again to obtain
，
Difference between the error of the output value and the expected value of the previous and subsequent training, if
Then updating the weight and the threshold value, if yes
Then the training is ended, wherein,βthe training precision of the neural network.
In any of the above embodiments, it is preferable that the input variables of the input layer of the neural network model are 6, and the input variables are the first 3 time-series rainfall amountR t-3, R t-2, R t-1Great depth of accumulated waterH t-3, H t-2, H t-1}; the output variable is 1, and the depth of the accumulated water point is 1 time sequence in the futureH t (ii) a The hidden layer neurons were set to 20.
In any of the above schemes, preferably, before the training, the raw data is normalized,
wherein the content of the first and second substances,x *i for the values after normalization of a certain kind of influencing factor,x i the value of the original numerical value is obtained,
is the minimum value of the influence factors，
Is the maximum value among the influence factors.
In any of the above aspects, preferably, the determination coefficient is determined based on a value of the coefficient of massR 2Is calculated by the formula
Wherein the content of the first and second substances,f i represents the predicted outcome of the neural network model,y i represents the predicted result of the numerical simulation model,
represents the mean of the prediction results of the numerical simulation model.
In any of the above aspects, preferably, the determination coefficient is determined based on a value of the coefficient of massR 2The closer to 1, the closer the prediction result representing the neural network model and the prediction result of the numerical simulation model are.
The invention provides a rapid prediction method for a rainstorm waterlogging event, which utilizes a method combining artificial intelligence and a numerical simulation model, overcomes the defect of low calculation efficiency of the numerical simulation model, meets the requirement of an artificial intelligence algorithm on large demand of a training sample, has high prediction precision and high calculation speed, provides a new thought for realizing rapid prediction of urban waterlogging, and develops a new research direction for application of the artificial intelligence technology in flood control and disaster reduction directions.
Drawings
Fig. 1 is a flow chart of a preferred embodiment of a method of automatically identifying session floods in accordance with the present invention.
Fig. 2 is a flow chart of another preferred embodiment of the method of automatically identifying session floods in accordance with the present invention.
FIG. 3 is a schematic diagram of the construction scope of an embodiment of the one-dimensional river network model of the Shenzhen bay watershed of Guangdong Shenzhen, according to the method for automatically identifying the field flood of the invention.
FIG. 4 is a schematic diagram of an embodiment of a two-dimensional surface grid of the Shenzhen bay streamer region in Guandong Shenzhen City according to the method for automatically identifying farm level floods of the present invention.
FIG. 5 is a distribution diagram of an exemplary water accumulation point in the Shenzhen bay watershed of Guandong Shenzhen City according to the method for automatically identifying the field flood of the present invention.
Fig. 6 is a schematic structural diagram of an embodiment of a BP neural network model of the method for automatically identifying a session flood according to the present invention.
Fig. 7 is a diagram illustrating comparison of simulation results of an embodiment of a rainfall designing scheme based on a rainstorm intensity formula according to the method of automatically identifying a flood of a storm according to the present invention.
Fig. 8 is a diagram illustrating comparison of simulation results of an embodiment of a characteristic rainfall pattern of the method for automatically identifying a flood according to the present invention.
Detailed Description
The invention is further illustrated with reference to the figures and the specific examples.
Example one
As shown in fig. 1, step 100 is executed to obtain a plurality of types of rainfall events. The rainfall process comprises a short-duration heavy rainfall process, characteristic rainfall processes with different space-time distributions and a design rainfall process designed according to a rainstorm intensity formula, wherein: the short-duration heavy rainfall process is a short-duration rainstorm process within 3 hours of the actual rainfall duration; the characteristic rainfall process with different space-time distributions is a rainfall process which accords with different types of space-time distribution characteristics and has the rainfall capacity of 30mm to 500mm accumulated in a single station within 3 hours; the designed rainfall process designed according to the rainstorm intensity formula is a Chicago rain type rainfall process with the general descent in the whole city and the 3-hour accumulated rainfall from 30mm to 500 mm.
Executing step 110, constructing an urban flood simulation refinement model, wherein the urban flood simulation refinement model is composed of a one-dimensional river network model, a two-dimensional earth surface model and an underground pipe network model, and the method comprises the following steps: the one-dimensional river network model is simulated by adopting a one-dimensional hydrodynamic model method and comprises a river, a section, a gate dam and a storage water body; the two-dimensional earth surface model is subdivided by adopting irregular grids, and water-blocking microtopography of important areas is used as a control line to construct grids in a range, wherein the important areas comprise bridges, roads, water bodies, riverways, railways, districts and bridges and culverts; the underground pipe network model comprises a rainwater pipe section, an inspection well, a rainwater grate and a pump station.
And step 120, checking the urban flood simulation refinement model by utilizing the various rainfall processes to obtain a rainstorm-waterlogging sample. The one-dimensional river network model, the two-dimensional earth surface model and the underground pipe network model simulate the production and convergence process of rainwater in the river channel, the earth surface and the underground pipe network through coupling calculation, and describe the evolution process of waterlogging and ponding of the urban underlying surface in detail.
Step 130 is executed to construct a neural network model for each water accumulation point using 90% of the rainstorm-waterlogging samples, and the neural network model is trained. The neural network model haskA number of input elements which are each arranged in a row,sa hidden layer, each hidden layer having neurons asn 1 、n 2 、……、n s The number of the main components is one,man output vector with 6 input variables of input layer of neural network model and a rainfall of the first 3 time sequencesR t-3, R t-2, R t-1Great depth of accumulated waterH t-3, H t-2, H t-1}; the output variable is 1, and the depth of the accumulated water point is 1 time sequence in the futureH t (ii) a The hidden layer neurons were set to 20. Before the training, the raw data is normalized,
wherein the content of the first and second substances,x *i for the values after normalization of a certain kind of influencing factor,x i the value of the original numerical value is obtained,
is the smallest value among the influencing factors of this type,
is the maximum value among the influence factors.
The calculation method of the neural network model comprises the following substeps:
step 131: determining the number of input layer nodes of the neural network modelkImplicit layer by layer numbersAnd number of nodes in each hidden layernAnd number of output layer nodesmPreliminarily determining the weight between layersω ij Threshold of hidden layerb ij ；
Step 132: computing hidden layer output
，
Wherein the content of the first and second substances,
is as followssFirst layer of neuronsjThe output of the first and second processors is,n s is as followssThe number of layer neurons in the layer,
is as followss-1 layer ofjThe nerve cell and the firstsLayer oneiThe weight of the link between the individual neurons,
is as followssLayer 1 neuronal secondjThe output of the first and second processors is,
is composed ofS thLayer oneiThe bias of the individual neurons is such that,
is an activation function of the neuron;
step 133: calculating an output variableO k ，
Wherein the content of the first and second substances,
is as followssFirst layer of neuronsjThe output of the first and second processors is,b k as an output layerkThe bias of the individual neurons is such that,
is as followssLayer onejThe individual neuron and the output layerkThe weight of the link to each of the neurons, k=1,2,…,m，
is an activation function of the neuron;
step 134: substituting the initial weight and the threshold value into the neural network to carry out normal feedforward training to obtain the error between the actual output value and the expected output value of the calculated sample
，
Wherein the content of the first and second substances,Nthe number of neurons in the output layer is,y k in order to be able to output the desired value,O k is the actual output value;
step (ii) of135: adjusting the weight and threshold of the network according to the normal training principle, and performing feedforward training again to obtain
，
Difference between the error of the output value and the expected value of the previous and subsequent training, if
Then updating the weight and the threshold value, if yes
Then the training is ended, wherein,βthe training precision of the neural network.
And step 140, predicting the water accumulation process which does not participate in model training by using the trained neural network model.
Step 150 is executed to calculate the decision coefficients of the predicted data and the simulation dataR 2To determine the similarity of the prediction results and determine the coefficientsR 2The closer to 1, the closer the prediction result representing the neural network model and the prediction result of the numerical simulation model are. Determining coefficientsR 2Is calculated by the formula
Wherein the content of the first and second substances,f i represents the predicted outcome of the neural network model,y i represents the predicted result of the numerical simulation model,
represents the mean of the prediction results of the numerical simulation model.
Example two
The traditional numerical simulation model and the neural network technology are combined, a novel method for predicting urban inland inundation risks is provided, and a Shenzhen river bay area is taken as an example, and the method is used for simulating and predicting the process of inland inundation water in rainstorm. The result shows that compared with the actually-measured accumulated water monitoring data and the numerical simulation result, the prediction result of the method has small errors, the prediction result of the numerical simulation can be effectively replaced, the waterlogging risk prediction time is greatly shortened, and the timeliness problem of urban waterlogging prediction and early warning is effectively solved.
The Shenzhen river bay basin is one of the five large basins of the Shenzhen city and is a typical high-speed developing region of the Shenzhen city, the hardening degree of the underlying surface is high, the confluence time is short, the construction of drainage facilities is not perfect, and in recent years, the damage and loss of inland water logging are in an ascending trend. Because the river in Shenzhen gulf river territory is shorter, and the relief is low-lying, and drainage time is shorter, accomplishes urban drainage process of sluicing within 3h takes place for the rainfall. Therefore, the 3h short-duration strong rainfall process in the Shenzhen gulf streaming domain is used as a rainfall scheme, a numerical simulation model is used for simulation, and a simulation result is used as data drive to construct a neural network prediction model of each ponding point. And finally, taking a rainstorm-waterlogging sample which does not participate in training as a prediction sample, predicting the water accumulation point by using the trained neural network model, and testing the effect of the neural network prediction model. The technical flow is shown in fig. 2.
First, rainfall scheme
The training of the neural network model requires a large number of learning samples of various types, the larger the training sample size is, the more information is contained, and the more intelligent the trained model is. Therefore, the rainfall-waterlogging samples under various rainfall conditions are obtained by adopting various rainfall processes as input rainfall conditions and simulating through a numerical model. The neural network model trained by the samples is more intelligent, and the water accumulation process of the water accumulation point under various rainfall conditions can be accurately predicted.
The rainfall regime primarily includes three sources: the short-duration heavy rainfall process actually occurs in the Shenzhen city 2008 + 2018 in the last 10 years; secondly, designing a rainfall process according to the rainfall space-time distribution characteristics of Shenzhen city; and thirdly, designing a rainfall process according to the Shenzhen city rainstorm intensity formula.
(1) Actual short-duration heavy rainfall scheme
Through analyzing the rainfall data of 63 meteorological stations in 2018 in 2008 & Shenzhen city, 5min by 5min, the short-time rainstorm process 178 fields within the actual rainfall duration time of 3h are screened out.
(2) Characteristic rainfall scheme with different space-time distributions
The research result of the short-term strong rainfall space-time distribution characteristic in the Shenzhen city 2008-2018 is adopted: according to different rainstorm space-time distribution laws, the short-duration heavy rainfall process in Shenzhen city can be divided into three types, namely: the rainstorm centre moves rapidly from the west to the south of the east; the rainfall center starts from the southeast and spreads to the west and the north; the rainfall is concentrated and basically no movement occurs. Based on these three types of rainfall characteristics, a rainfall process 708 field of 3h single station accumulated rainfall from 30mm to 500mm was designed in accordance with the above three types of space-time distribution characteristics.
(3) Rainfall design scheme according to rainstorm intensity formula
According to the Shenzhen city rainstorm intensity formula, the whole city prisoner is designed, the Chicago rain type rainfall process with 3h accumulated rainfall from 30mm to 500mm (2 mm interval) is designed, and 236 fields are totaled.
The urban flood simulation model is used for simulation, 1122 rainfall schemes lasting for 3 hours are calculated in total, and the requirements of the neural network model on learning samples are met in terms of the number and types of the samples.
Second, urban flood simulation model
The urban flood simulation refinement model is composed of a one-dimensional river network model, a two-dimensional surface model, an underground pipe network model and the like, and comprehensively considers the hydrodynamic combination of the urban surface and various objects such as an underground pipe network and a flood drainage channel related to the urban surface, so as to simulate the process of rainfall runoff and waterlogging water on the urban underlying surface.
(1) One-dimensional river network model
And (4) researching the flood routing of the inland river network, and simulating by adopting a one-dimensional hydrodynamic model method. The constructed one-dimensional river network model comprises 25 rivers, 1325 sections, 25 gate dams and 22 stagnant water bodies (20 reservoirs and 2 stagnant flood areas) in the bay area, and the construction range is shown in fig. 3.
(2) Two-dimensional earth surface model
The surface flood evolution is simulated by adopting a two-dimensional hydrodynamic model, the surface model is subdivided by adopting irregular grids with the side length of 5-10m, water-blocking microtopography in important areas such as overpasses, roads, water bodies, riverways, railways, districts, bridges and culverts is taken as a control line, the number of grids in a construction range is 304 thousands, the number of the grids is 564 thousands, and the total simulated area is 226.76km 2. The model mesh layout is shown in fig. 4.
(3) Pipe network model
The underground drainage pipe network model comprises 12.7 ten thousand rainwater pipe sections, 7 ten thousand inspection wells, 5 ten thousand rain grates and 12 pump stations.
The three models simulate the production and confluence process of rainwater in a river channel, the earth surface and an underground pipe network through coupling calculation, and particularly depict the evolution process of waterlogging and ponding on the urban underlying surface. However, due to the fact that the degree of model refinement is high, the calculated amount of the model is large, and the calculation time is more than 12 hours for a 3-hour rainfall scheme.
Third, training sample generation
And checking the numerical simulation model by utilizing the actual rainfall process to obtain a more accurate training sample. Through checking, compared with the measured value, the simulation result of the numerical model has smaller prediction error, and can meet the requirement on learning sample quality. For the sake of space, the explanation is made only by taking 2018.8.29 Shenzhen heavy storm as an example. Selecting 6 typical ponding points in the gulf area, wherein the distribution of the ponding points is shown in figure 5, and comparing the measured maximum ponding depth data with the model simulated ponding depth, which is shown in table 1. As can be seen from table 1, the simulation value has an average prediction error of 2.92% compared with the measured value, and meets the requirement of the neural network model on learning sample quality.
TABLE 12018.8.29 comparison chart of measured water depth and simulated water depth at waterlogging and ponding point of heavy rainstorm
The numerical simulation model is utilized to analyze and calculate the three rainfall schemes of 1122 total rainfall fields to obtain 40392 rainstorm-waterlogging samples, and the simulation result is used as a drive to construct a neural network model of each water accumulation point.
Four, BP neural network model
The waterlogging depth of the ponding point is related to the rainfall condition and the ponding condition of the front time sequence of the ponding point, and belongs to the nonlinear regression problem of multiple influence factors. Therefore, a feedback neural network model, namely a BP (back propagation) neural network model is selected to train and predict the ponding situation.
The BP neural network model is firstly proposed by Rumelhart, is the most mature neural network model which is also the most widely applied at present, and can approximate the nonlinear mapping relation with any precision as long as enough hidden layers and hidden nodes exist. The BP neural network structure is mainly divided into an input layer, a hidden layer and an output layer, and a model structure diagram is shown in fig. 6.
As shown in FIG. 6, the model haskA number of input elements which are each arranged in a row,sa hidden layer, each hidden layer having neurons asn 1 、n 2 、……、n s The number of the main components is one,mand outputting the vector.
The algorithm comprises the following calculation steps:
(1) first, the number of nodes of the input layer of the network model is determinedkImplicit layer by layer numbersAnd number of nodes in each hidden layernAnd the number m of nodes of the output layer, and preliminarily determining the weight between layersω ij Threshold of hidden layerb ij 。
(2) Computing hidden layer output
Wherein the content of the first and second substances,
is as followssFirst layer of neuronsjThe output of the first and second processors is,n s is as followssNumber of layer neurons. Is provided with
Is as followss-1 layer ofjThe nerve cell and the firstsLayer oneiThe weight of the link between the individual neurons,
is as followssLayer oneiThe bias of the individual neurons is such that,
is the activation function of the neuron, typically a sigmod function or a logistic function.
(3) Calculating an output variableO k
(4) Substituting the initial weight and the threshold value obtained in the above steps into a BP network to carry out normal feedforward training to obtain the error between the actual output value and the expected output value of the calculated sample
。
(5) Adjusting the weight and threshold of the network according to the normal training principle, and performing feedforward training again to obtain
. Is provided with
The difference value of the error between the output value and the expected value is obtained by training in two times. If it is
(βThe training precision of the neural network), the weight and the threshold are updated, if so
And the training is finished.
The BP algorithm is divided into a feedforward process and a feedback process, and in the feedforward process, training samples are input in an input layer
Calculating the output variable according to the formulas (1) and (2)
Then calculating the actual output from equation (3)
And desired output
The mean square error of (d). Adjusting weight by gradient descent methodωAnd a threshold valuebAnd finishing the training process when the mean square error value of the actual output value and the expected output value of the network reaches a preset target value.
Fifth, construction of waterlogging prediction model
6 typical water accumulation points in the Bay area in the figure 5 are taken as research objects to respectively establish a BP neural network model. The depth of the accumulated water in the accumulated water point is mainly related to the rainfall of the drainage area and the accumulated water depth of the time sequence before the accumulated water point. Thus, factors considered by the BP neural network prediction model include the amount of rainfall at the previous time within the ponding point drainage area
And the water accumulation depth of the time sequence before the water accumulation point
。
The input variables of the input layer of the neural network model are 6, and the input variables are rainfall of the first 3 time sequences
{R t-3, R t-2, R t-1Great depth of accumulated waterH t-3, H t-2, H t-1The output variable is 1, and the depth of water accumulation of the water accumulation point is 1 time sequence in the future
The hidden layer neurons were set to 20. Through continuous circulation and training of the model, the depth change of the accumulated water in the whole rainfall process is predicted. In order to avoid the difference of numerical value intervals of various types of data, before training and learning, the original data is normalized, and the formula is as follows:
in the formula (I), the compound is shown in the specification,x *i for the values after normalization of a certain kind of influencing factor,x i the value of the original numerical value is obtained,
is the smallest value among the influencing factors of this type,
is the maximum value thereof.
During training, 90% of the rainstorm-waterlogging samples are respectively selected as training samples, and the rest 10% are used as test samples. And predicting the water accumulation process without participating in model training by using the trained model. And finally, judging the similarity degree of the two curves by calculating a decision coefficient R2 of the predicted data and the simulation data, wherein the closer the decision coefficient R2 is to 1, the higher the fitting degree of the two curves is, namely, the closer the predicted result of the neural network model is to the predicted result of the numerical model.
The calculation formula of R2 is shown in formula 5. In the formulaf i Andy i respectively representing the prediction result of the neural network model and the prediction result of the numerical simulation model.
Sixthly, result analysis
As shown in fig. 7 and 8, the water accumulation process at each water accumulation point is predicted by using a neural network prediction model.
As shown in FIG. 6, the model haskA number of input elements which are each arranged in a row,sa hidden layer, each hidden layer having neurons asn 1 、n 2 、……、n s The number of the main components is one,mand outputting the vector.
The algorithm comprises the following calculation steps:
first, the number of nodes of the input layer of the network model is determinedkImplicit layer by layer numbersAnd number of nodes in each hidden layernAnd number of output layer nodesmPreliminarily determining the weight between layersω ij Threshold of hidden layerb ij 。
Computing hidden layer output
Wherein the content of the first and second substances,
is as followssFirst layer of neuronsjAn output，n s Is as followssNumber of layer neurons. Is provided with
Is as followss-1 layer ofjThe nerve cell and the firstsLayer oneiThe weight of the link between the individual neurons,
is as followssLayer oneiThe bias of the individual neurons is such that,
is the activation function of the neuron, typically a sigmod function or a logistic function.
Calculating an output variableO k
Substituting the initial weight and the threshold value obtained in the above steps into a BP network to carry out normal feedforward training to obtain the error between the actual output value and the expected output value of the calculated sample
。
Adjusting the weight and threshold of the network according to the normal training principle, and performing feedforward training again to obtain
. Is provided with
The difference value of the error between the output value and the expected value is obtained by training in two times. If it is
(βThe training precision of the neural network), the weight and the threshold are updated, if so
And the training is finished.
The BP algorithm is divided into a feedforward process and a feedback process, and in the feedforward process, training samples are input in an input layer
Calculating the output variable according to the formulas (1) and (2)
Then calculating the actual output from equation (3)
And desired output
The mean square error of (d). Adjusting weight by gradient descent method
And a threshold value
And finishing the training process when the mean square error value of the actual output value and the expected output value of the network reaches a preset target value.
Fifth, construction of waterlogging prediction model
6 typical water accumulation points in the Bay area in the figure 5 are taken as research objects to respectively establish a BP neural network model. The depth of the accumulated water in the accumulated water point is mainly related to the rainfall of the drainage area and the accumulated water depth of the time sequence before the accumulated water point. Thus, factors considered by the BP neural network prediction model include the amount of rainfall at the previous time within the ponding point drainage area
And the accumulated waterWater accumulation depth of pre-spot time sequence
。
The input variables of the input layer of the neural network model are 6, and are the rainfall amount of the first 3 time sequencesR t-3, R t-2,R t-1Great depth of accumulated waterH t-3, H t-2, H t-1The output variable is 1, and the depth of water accumulation of the water accumulation point is 1 time sequence in the future
The hidden layer neurons were set to 20. Through continuous circulation and training of the model, the depth change of the accumulated water in the whole rainfall process is predicted. In order to avoid the difference of numerical value intervals of various types of data, before training and learning, the original data is normalized, and the formula is as follows:
in the formula (I), the compound is shown in the specification,x *i for the values after normalization of a certain kind of influencing factor,x i the value of the original numerical value is obtained,
is the smallest value among the influencing factors of this type,
is the maximum value thereof.
During training, 90% of the rainstorm-waterlogging samples are respectively selected as training samples, and the rest 10% are used as test samples. And predicting the water accumulation process without participating in model training by using the trained model. And finally, judging the similarity degree of the two curves by calculating a decision coefficient R2 of the predicted data and the simulation data, wherein the closer the decision coefficient R2 is to 1, the higher the fitting degree of the two curves is, namely, the closer the predicted result of the neural network model is to the predicted result of the numerical model.
The calculation formula of R2 is shown in formula 5. In the formulaf i Andy i respectively representing the prediction result of the neural network model and the prediction result of the numerical simulation model.
Sixthly, result analysis
As shown in fig. 7 and 8, the water accumulation process at each water accumulation point is predicted by using a neural network prediction model.
Because the continuity of the actually measured rainfall water monitoring data is poor, the method is only selected to be compared with the actually monitored maximum water depth. The maximum ponding depth result of the ponding point of 8.29 rainstorms in 2018 without participation in training is selected for prediction, and the result is shown in table 2.
As can be seen from table 2, the error between the maximum water accumulation depth and the measured value simulated by the numerical model is 2.92%, the error between the maximum water accumulation depth and the measured value simulated by the BP neural network model is 3.84%, and the prediction error is increased by 0.92%, which is mainly because the neural network model is trained on the basis of the numerical simulation result, and the prediction result has accumulated errors, but the error between the two models is only 1.98%.
TABLE 22018.8.29 comparison chart of measured water depth and simulated water depth at waterlogging and ponding point of heavy rainstorm
As can be seen from fig. 7 and 8, for each water accumulation point, the water accumulation process simulated by the numerical model and predicted by the BP neural network model have very small fitting error, and the determination coefficient R2 is above 0.9, fig. 7 is simulated by the designed rainfall scheme based on the rainstorm intensity formula, and the rain model is the chicago rain model in the whole city. It can be seen that the water accumulation depth change trends of 6 water accumulation points from the point A to the point F distributed in the west, the middle and the east of the tablet area are basically similar. While the figure 7 feature designs the simulation results of a rain scenario in which the rainstorm centre moves from west to east. The change of the water accumulation depth of 6 water accumulation points from the point A to the point F in the figure 8 reflects the trend of the space-time change of the rainfall peak movement. The peak rainfall occurred early at A, B points of water accumulation located west of the parcel, and the corresponding maximum water depth occurred early, while the peak rainfall occurred late at E, F points of water accumulation located east of the parcel, and the corresponding maximum water depth occurred late. And C, D water accumulation point is located in the middle of the plate area, the rainfall peak value is in the middle, and the corresponding maximum water depth also correspondingly appears at the middle moment. Therefore, the numerical model utilized by the method and the BP neural network model trained based on the numerical model result reflect the relation between the depth change of each ponding water spot product and the rainfall space-time change, and the prediction result is close to the actual situation.
In the aspect of forecasting the aging, at a typical water accumulation point of an estuary district, a BP neural network model calculates a rainfall scheme with the water accumulation point lasting for 3 hours, the calculation time is only 0.01 second, and the numerical model calculates the same rainfall scheme, and the time is several hours. The calculation speed of the BP neural network model is improved by tens of thousands of times compared with that of a numerical model. If the number of the computational grids is increased, the computational simulation time consumption of the numerical model is increased, and the computational efficiency cannot meet the real-time simulation requirement. The method provided by the invention greatly saves the calculation time and can effectively meet the demand of flood prevention emergency work.
The numerical simulation model and the artificial neural network are combined, a large number of rainstorm-waterlogging samples under different rainfall conditions are generated through the numerical simulation model, and the BP neural network ponding prediction model is trained through the samples. Compared with the prediction result of the BP neural network ponding prediction model and the simulation result of the numerical model, the fitting error is very small, and the determination coefficient R2 is more than 0.9. Compared with the actual measurement data, the predicted maximum water accumulation depth of the BP neural network prediction model has the average error of 3.84%; the average error was 1.98% compared to the simulation results of the numerical model. In the aspect of computational efficiency, taking Shenzhen and river bay plate regions as an example, the computation speed of the BP neural network prediction model is tens of thousands of times higher than that of the numerical model. The flood prevention emergency requirement can be met in the aspects of precision and time efficiency.
The method combining artificial intelligence and the numerical simulation model overcomes the defect of low calculation efficiency of the numerical simulation model, meets the requirement of an artificial intelligence algorithm on large demand of training samples, has high prediction precision and high calculation speed, provides a new idea for realizing rapid prediction of urban waterlogging, and develops a new research direction for application of the artificial intelligence technology in flood control and disaster reduction directions.
For a better understanding of the present invention, the foregoing detailed description has been given in conjunction with specific embodiments thereof, but not with the intention of limiting the invention thereto. Any simple modifications of the above embodiments according to the technical essence of the present invention still fall within the scope of the technical solution of the present invention. In the present specification, each embodiment is described with emphasis on differences from other embodiments, and the same or similar parts between the respective embodiments may be referred to each other.
Claims (10)
1. A rapid prediction method for rainstorm waterlogging events comprises the steps of acquiring various types of rainfall processes, and is characterized by further comprising the following steps:
step 1: constructing an urban flood simulation refinement model;
step 2: checking the urban flood simulation refinement model by utilizing the various rainfall processes to obtain a rainstorm-waterlogging sample;
and step 3: constructing a neural network model of each water accumulation point by using 90% of the rainstorm-waterlogging samples, and training the neural network model;
and 4, step 4: predicting the ponding process which does not participate in model training by using the trained neural network model;
and 5: by calculating the coefficients of determination of the predicted data and the simulated dataR 2To judge the similarity of the prediction results.
2. The method for rapid prediction of a rainstorm flood event according to claim 1, wherein said rainfall events comprise short duration heavy rainfall events, characteristic rainfall events of different temporal and spatial distributions, and design rainfall events designed according to a rainstorm intensity formula.
3. The method for rapid prediction of a stormwater waterlogging event according to claim 2, wherein the urban flood simulation refinement model is comprised of a one-dimensional river network model, a two-dimensional surface model and an underground pipe network model.
4. The method for rapidly predicting the rainstorm waterlogging event according to claim 3, wherein the step 2 comprises simulating the production and convergence process of rainwater on the river channel, the earth surface and the underground pipe network by the one-dimensional river network model, the two-dimensional earth surface model and the underground pipe network model through coupling calculation, and describing the evolution process of waterlogging on the urban underlying surface in detail.
5. The method for rapid prediction of a stormwater waterlogging event of claim 4, wherein the neural network model compriseskA number of input elements which are each arranged in a row,sa hidden layer, each hidden layer having neurons asn 1 、n 2 、……、n s The number of the main components is one,mand outputting the vector.
6. The method for rapid prediction of a stormwater waterlogging event according to claim 5, wherein the calculation method of the neural network model comprises the sub-steps of:
step 31: determining the number of input layer nodes of the neural network modelkImplicit layer by layer numbersAnd number of nodes in each hidden layernAnd number of output layer nodesmPreliminarily determining the weight between layersω ij Threshold of hidden layerb ij ；
Step 32: computing hidden layer output
，
Wherein the content of the first and second substances,
is as followssFirst layer of neuronsjThe output of the first and second processors is,n s is as followssThe number of layer neurons in the layer,
is as followss-1 layer ofjThe nerve cell and the firstsLayer oneiThe weight of the link between the individual neurons,
is as followssLayer 1 neuronal secondjThe output of the first and second processors is,
is composed ofS thLayer oneiThe bias of the individual neurons is such that,
is an activation function of the neuron;
step 33: calculating an output variableO k ，
Wherein the content of the first and second substances,
is as followssFirst layer of neuronsjThe output of the first and second processors is,b k as an output layerkThe bias of the individual neurons is such that,
is as followssLayer onejThe individual neuron and the output layerkThe weight of the link to each of the neurons, k=1,2,…,m，
is an activation function of the neuron;
step 34: substituting the initial weight and the threshold value into the neural network to carry out normal feedforward training to obtain the error between the actual output value and the expected output value of the calculated sample
，
Wherein the content of the first and second substances,Nthe number of neurons in the output layer is,y k in order to be able to output the desired value,O k is the actual output value;
step 35: adjusting the weight and threshold of the network according to the normal training principle, and performing feedforward training again to obtain
，
Difference between the error of the output value and the expected value of the previous and subsequent training, if
Then updating the weight and the threshold value, if yes
Then the training is ended, wherein,βthe training precision of the neural network.
7. As in claimThe method for rapidly predicting a rainstorm waterlogging event according to claim 6, wherein the input variables of the input layer of the neural network model are 6, and the input variables are a rainfall amount of the first 3 time sequencesR t-3, R t-2, R t-1Great depth of accumulated waterH t-3, H t-2, H t-1}; the output variable is 1, and the depth of the accumulated water point is 1 time sequence in the futureH t (ii) a The hidden layer neurons were set to 20.
8. The method for rapid prediction of a stormwater waterlogging event according to claim 7, wherein prior to said training, raw data is normalized,
wherein the content of the first and second substances,x *i for the values after normalization of a certain kind of influencing factor,x i the value of the original numerical value is obtained,
is the smallest value among the influencing factors of this type,
is the maximum value among the influence factors.
9. The method for rapid prediction of a stormwater waterlogging event of claim 8, wherein the decision coefficientR 2Is calculated by the formula
Wherein the content of the first and second substances,f i represents the predicted outcome of the neural network model,y i represents the predicted result of the numerical simulation model,
represents the mean of the prediction results of the numerical simulation model.
10. The method for rapid prediction of a stormwater waterlogging event of claim 9, wherein the decision coefficientR 2The closer to 1, the closer the prediction result representing the neural network model and the prediction result of the numerical simulation model are.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110895780.0A CN113344305B (en) | 2021-08-05 | 2021-08-05 | Rapid prediction method for rainstorm waterlogging event |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110895780.0A CN113344305B (en) | 2021-08-05 | 2021-08-05 | Rapid prediction method for rainstorm waterlogging event |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113344305A true CN113344305A (en) | 2021-09-03 |
| CN113344305B CN113344305B (en) | 2021-10-29 |
Family
ID=77480854
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110895780.0A Active CN113344305B (en) | 2021-08-05 | 2021-08-05 | Rapid prediction method for rainstorm waterlogging event |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113344305B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113780668A (en) * | 2021-09-15 | 2021-12-10 | 泰华智慧产业集团股份有限公司 | Urban ponding waterlogging prediction method and system based on historical data |
| CN114925923A (en) * | 2022-05-27 | 2022-08-19 | 杭州电子科技大学 | Method for improving depth prediction precision of ponding of non-monitoring unit |
| CN115114842A (en) * | 2022-04-27 | 2022-09-27 | 中国水利水电科学研究院 | Rainstorm waterlogging event prediction method based on small sample transfer learning algorithm |
| CN115471078A (en) * | 2022-09-15 | 2022-12-13 | 镇江市勘察测绘研究院有限公司 | Flood risk point assessment method and device based on urban water affair system |
| CN116861317A (en) * | 2023-09-04 | 2023-10-10 | 北京建筑大学 | Cell waterlogging early warning method and system based on BP neural network |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108647778A (en) * | 2018-05-09 | 2018-10-12 | 天津大学 | Dynamic prediction method for drainage flow of drainage port of urban rainwater system |
| CN109978235A (en) * | 2019-03-04 | 2019-07-05 | 宁波市气象服务中心 | A kind of flooded water level prediction method of product based on sample learning |
| CN110738355A (en) * | 2019-09-19 | 2020-01-31 | 河源职业技术学院 | urban waterlogging prediction method based on neural network |
| CN111144254A (en) * | 2019-12-18 | 2020-05-12 | 江河瑞通（北京）技术有限公司 | Urban waterlogging degree identification method and device |
| CN111369059A (en) * | 2020-03-05 | 2020-07-03 | 山东大学 | Urban waterlogging prediction method and system based on rapid waterlogging simulation coupling model |
| CN111582541A (en) * | 2020-03-27 | 2020-08-25 | 合肥学院 | Firefly algorithm-based inland inundation model prediction method |
| CN112528563A (en) * | 2020-12-08 | 2021-03-19 | 丹华水利环境技术(上海)有限公司 | Urban waterlogging early warning method based on SVM algorithm |
| CN112699599A (en) * | 2020-12-17 | 2021-04-23 | 南京大学 | Flood disaster early warning method based on BP-GEO |
| CN112711825A (en) * | 2021-03-26 | 2021-04-27 | 中国水利水电科学研究院 | Direct bidirectional coupling method for ground surface and pipe network distribution |
-
2021
- 2021-08-05 CN CN202110895780.0A patent/CN113344305B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108647778A (en) * | 2018-05-09 | 2018-10-12 | 天津大学 | Dynamic prediction method for drainage flow of drainage port of urban rainwater system |
| CN109978235A (en) * | 2019-03-04 | 2019-07-05 | 宁波市气象服务中心 | A kind of flooded water level prediction method of product based on sample learning |
| CN110738355A (en) * | 2019-09-19 | 2020-01-31 | 河源职业技术学院 | urban waterlogging prediction method based on neural network |
| CN111144254A (en) * | 2019-12-18 | 2020-05-12 | 江河瑞通（北京）技术有限公司 | Urban waterlogging degree identification method and device |
| CN111369059A (en) * | 2020-03-05 | 2020-07-03 | 山东大学 | Urban waterlogging prediction method and system based on rapid waterlogging simulation coupling model |
| CN111582541A (en) * | 2020-03-27 | 2020-08-25 | 合肥学院 | Firefly algorithm-based inland inundation model prediction method |
| CN112528563A (en) * | 2020-12-08 | 2021-03-19 | 丹华水利环境技术(上海)有限公司 | Urban waterlogging early warning method based on SVM algorithm |
| CN112699599A (en) * | 2020-12-17 | 2021-04-23 | 南京大学 | Flood disaster early warning method based on BP-GEO |
| CN112711825A (en) * | 2021-03-26 | 2021-04-27 | 中国水利水电科学研究院 | Direct bidirectional coupling method for ground surface and pipe network distribution |
Non-Patent Citations (1)
| Title |
|---|
| 曾鹏: "机器学习技术在城市洪涝快速预报中的应用研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113780668A (en) * | 2021-09-15 | 2021-12-10 | 泰华智慧产业集团股份有限公司 | Urban ponding waterlogging prediction method and system based on historical data |
| CN115114842A (en) * | 2022-04-27 | 2022-09-27 | 中国水利水电科学研究院 | Rainstorm waterlogging event prediction method based on small sample transfer learning algorithm |
| CN114925923A (en) * | 2022-05-27 | 2022-08-19 | 杭州电子科技大学 | Method for improving depth prediction precision of ponding of non-monitoring unit |
| CN115471078A (en) * | 2022-09-15 | 2022-12-13 | 镇江市勘察测绘研究院有限公司 | Flood risk point assessment method and device based on urban water affair system |
| CN115471078B (en) * | 2022-09-15 | 2024-02-02 | 镇江市勘察测绘研究院有限公司 | Flood risk point assessment method and device based on urban water affair system |
| CN116861317A (en) * | 2023-09-04 | 2023-10-10 | 北京建筑大学 | Cell waterlogging early warning method and system based on BP neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113344305B (en) | 2021-10-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113344305B (en) | Rapid prediction method for rainstorm waterlogging event | |
| CN107992961B (en) | Adaptive drainage basin medium and long term runoff forecasting model architecture method | |
| CN111310968B (en) | LSTM neural network circulating hydrologic forecasting method based on mutual information | |
| Bae et al. | Monthly dam inflow forecasts using weather forecasting information and neuro-fuzzy technique | |
| Minglei et al. | Classified real-time flood forecasting by coupling fuzzy clustering and neural network | |
| CN111199298B (en) | Flood forecasting method and system based on neural network | |
| KR20230023688A (en) | Flood forecasting method using recurrent neural network | |
| KR102584619B1 (en) | Water pollution accident prediction system based on hydraulic hydrology analysis with distributed model applied | |
| CN116485584B (en) | Method and system for cooperative regulation and control of WEE of river with large bottom slope in alpine region | |
| Madaeni et al. | Ice jam formation, breakup and prediction methods based on hydroclimatic data using artificial intelligence: A review | |
| CN114841417B (en) | High-precision salt tide forecasting method and system and readable storage medium | |
| CN116050628A (en) | Reservoir flood control optimization scheduling method considering forecast uncertainty based on digital twinning | |
| CN115329656A (en) | Urban waterlogging water distribution rapid prediction method based on improved convolutional neural network | |
| Anilan et al. | Application of Artificial Neural Networks and regression analysis to L-moments based regional frequency analysis in the Eastern Black Sea Basin, Turkey | |
| Wang et al. | A statistical hydrological model for Yangtze river watershed based on stepwise cluster analysis | |
| Lee et al. | Improving applicability of neuro-genetic algorithm to predict short-term water level: a case study | |
| Khan et al. | A comparison of Quadratic Regression and Artificial Neural Networks for the estimation of quantiles at ungauged sites in regional frequency analysis. | |
| Oluwatobi et al. | Modeling and simulation of river discharge using artificial neural networks | |
| Kang et al. | A case study for ANN-based rainfall–runoff model considering antecedent soil moisture conditions in Imha Dam watershed, Korea | |
| Ogale et al. | Modelling and short term forecasting of flash floods in an urban environment | |
| CN115114842A (en) | Rainstorm waterlogging event prediction method based on small sample transfer learning algorithm | |
| CN117494586B (en) | Mountain torrent space-time prediction method based on deep learning | |
| Supharatid | The Hat Yai 2000 flood: the worst flood in Thai history | |
| Zeng et al. | Development of a model-based flood emergency management system in Yujiang River Basin, South China | |
| Bakute et al. | THE LAND USE AND CLIMATE CHANGE IMPACT ON LAKE USMA HYDROLOGICAL REGIME |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |