Method for simulating flood process of river with lack of data under influence of city
By employing a hybrid learning algorithm based on one-dimensional river control equations and iterative optimization of boundary conditions in urban waterways, the problems of data dependence and physical interpretability in flood simulation in waterways lacking underwater topographic data were solved, achieving efficient and accurate flood process simulation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- POWERCHINA HUADONG ENG CORP LTD
- Filing Date
- 2026-03-18
- Publication Date
- 2026-07-21
AI Technical Summary
In waterways lacking underwater topographic data due to urban influences, existing hydrodynamic and machine learning models struggle to quickly and accurately simulate flood processes, exhibiting problems such as strong data dependence, poor physical interpretability, and unreasonable simulation results.
A hybrid learning algorithm based on one-dimensional river control equations is adopted. River boundary condition data is preprocessed, and multiple hybrid learning algorithms are combined to simulate flood processes. By iteratively optimizing the boundary conditions, a flood simulation method adapted to urban impacts is constructed to ensure that the simulation process follows the physical laws of flood evolution.
It enables efficient and accurate simulation of flood processes in the absence of underwater topographic data, adapts to river scenarios affected by urban conditions, improves simulation accuracy and physical interpretability, and avoids problems such as non-conservation of water volume and incorrect flow direction.
Smart Images
Figure CN122432477A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for simulating flood processes in rivers with limited data under urban influence. It is applicable to the field of hydrology, and particularly suitable for simulating flood processes and hydrological conditions in rivers affected by urban development and lacking underwater topographic data. Background Technology
[0002] In recent years, influenced by global climate change and urban development, the occurrence and evolution patterns of extreme hydrological events with significant destructive potential, such as sudden floods, droughts, and rapid shifts between drought and flood, have gradually changed, posing challenges to flood control efforts. Currently, the most damaging flood events are characterized by their suddenness and short warning periods, which poses a challenge to the numerical simulation of hydrodynamic processes upon which understanding the evolution of flood processes relies.
[0003] Hydrodynamic models are effective tools for numerically simulating flood evolution, but their accuracy is highly dependent on the quality of underwater topographic data. In rivers affected by urban development, municipal and transportation construction often impacts the underwater topography at multiple levels and in multiple ways, from macroscopic river morphology to microscopic river cross-sections. However, comprehensively understanding these impacts presents challenges due to policy, funding, and information constraints. This hinders the rapid and accurate prediction of flood events through hydrodynamic models. Therefore, alternative approaches using machine learning to simulate flood events have recently emerged.
[0004] According to literature review, machine learning methods for flood process simulation have made significant progress, but some shortcomings remain, mainly in areas such as data dependence, model generalization ability, and physical mechanism interpretation. These shortcomings hinder the effective deployment of machine learning models in flood early warning and flood control and mitigation. First, machine learning, especially deep learning models, requires a large amount of high-quality historical data, such as high-resolution topographic data, rainfall data, and water depth observations, for training to learn the complex patterns of flood processes. This places high demands on the observation quality and completeness of the data. Second, there is insufficient physical consistency. Due to the lack of hard constraints on physical laws such as hydrodynamic equations and mass conservation, simulation results may be physically unreasonable, such as non-conservation of water volume or incorrect flow direction. Third, physical interpretability is poor because machine learning models, especially deep learning models, are often considered "black box" models. They make predictions by learning statistical relationships in the data but lack the ability to explicitly encode and interpret the physical processes of floods.
[0005] To address the pain points of data-scarce river flood simulation under urban influence and the shortcomings of existing technologies, there is an urgent need for a flood process simulation method that balances physical interpretability, low data dependence, and adaptability to urban influence characteristics. Summary of the Invention
[0006] The technical problem to be solved by this invention is to provide a method for simulating river flood processes under urban influences with limited data, addressing the aforementioned problems.
[0007] The technical solution adopted in this invention is: a method for simulating flood processes in rivers with limited data under urban influence, comprising: Data on river boundary conditions under the influence of urban areas were obtained. Based on the confluence flow and undetermined coefficients within the river section, the boundary condition data of the upstream section of the river were preprocessed. Multiple sets of undetermined coefficients were set. Based on upstream cross-section data, each flood event is identified, and the flood events corresponding to the upstream flood events are identified in the downstream cross-section data. The upstream and downstream cross-section data corresponding to the flood events are spliced together in chronological order to form the time series of the boundary conditions of the flood events for the upstream and downstream cross-sections respectively. Based on the time series of flood event boundary conditions at upstream and downstream sections, a hybrid learning algorithm based on one-dimensional river control equations is used to simulate and calculate the water level and flow rate at the downstream section. The simulation results accuracy of each hybrid learning algorithm is measured, and the upstream section boundary conditions are iteratively updated using the simulation results accuracy as the weight. The iterated upstream section boundary conditions are then re-substituted into the various hybrid learning algorithms for simulation. The combination of undetermined coefficient values with the best accuracy in the simulation results within the screening period is selected. Then, the simulation results with the highest accuracy within the screening period are selected from the simulation results of multiple iterations. The simulation results are then optimized a second time by combining the water level-discharge relationship of the downstream section, and finally the optimal simulation calculation model for the river flood process is determined.
[0008] By employing the aforementioned technical means, a complete flood simulation process is constructed, encompassing "data preprocessing - time series integration - physical constraint-based hybrid learning simulation - boundary condition iteration - result selection and optimization." This process requires no underwater topographic data, relying solely on readily available river boundary hydrological data, precisely adapting to data-scarce river scenarios under urban influence. Through multi-algorithm collaborative simulation, iterative optimization of accuracy weights, and multi-layered result selection, the accuracy of flood simulation is significantly improved. Furthermore, the deep integration of one-dimensional river control equations with hybrid learning algorithms imbues the simulation process with explicit physical constraints, effectively addressing the "black box" problem of existing machine learning models and ensuring the physical interpretability of the simulation process.
[0009] As a preferred embodiment, the preprocessing of the upstream cross-sectional boundary condition data based on the confluence flow and undetermined coefficients within the river section includes: The measured flow at the upstream section, the total flow of the tributaries, and the flow of the confluence of the two sections are integrated and calculated according to the following formula; ; Among them, Q up′ is the upstream cross-sectional flow rate after pretreatment; Q tr It is the total flow of the tributary; q L It is the inter-regional flow rate; a1 and b1 are undetermined coefficients.
[0010] Through the above technical means, the multi-source water inflow data, including measured flow at the upstream section, tributary confluence flow, and inter-regional confluence flow, are normalized and integrated to form standardized upstream boundary input parameters. This simplifies the input dimensions of subsequent hybrid learning algorithms and improves computational efficiency. The contribution weights of each flow item to the upstream boundary can be flexibly adjusted using undetermined coefficients a1 and b1 to adapt to the differences in river confluence characteristics under urban influence and avoid the problem of poor adaptability of general formulas.
[0011] As a preferred embodiment, the one-dimensional river channel control equation is the difference form of the momentum conservation equation in the Saint-Venant equations.
[0012] By employing the aforementioned technical means, it is ensured that all calculations in the simulation process follow the natural physical laws of flood evolution, thus avoiding physical inconsistencies such as non-conservation of water volume and incorrect flow direction.
[0013] As a preferred embodiment, one of the various hybrid learning algorithms based on one-dimensional river control equations includes: ; ; A linear approximation of water level and flow rate is made based on the difference form of the momentum conservation equation in the Saint-Venant equations, introducing the identity matrix E and the column vector of coefficients to be solved. Construct a water level estimation formula. The parameters are estimated using known parameters within the calibration period; the identity matrix E and the column vector of coefficients to be solved are introduced. Construct a flow recursive formula and estimate the parameters using known parameters within a periodic period to obtain the utilization rate. ;
[0014] By combining the flow rate recursion formula and the water level estimation formula, and taking the water level at time n+1... ,flow The unknown quantity is used to solve for the water level and flow rate at the downstream section.
[0015] Using the above-mentioned technical means, a linear equation system is constructed based on Saint-Venant's momentum conservation equation to achieve synchronous and rapid solution of downstream cross-section water level and flow rate. The calculation efficiency is high and it meets the timeliness requirements of urban flood early warning. Each step of the simulation process is based on clear hydrodynamic physical laws, which has clear physical interpretability and gets rid of "black box" fitting.
[0016] As a preferred embodiment, one of the various hybrid learning algorithms based on one-dimensional river control equations includes: ; A water level calculation formula is constructed based on the difference form of the momentum conservation equation in the Saint-Venant equations, and a time step is introduced. Spatial distance from upstream section to downstream section Changes in upstream water level after pretreatment and undetermined coefficients , , ,Regulation , After determining the range of values, the optimal value is determined by sliding learning combined with least squares fitting within the pass rate period. Solve for the change in water level at the downstream section. Then, the water level Z at the downstream section is calculated; ; Based on the water level Z at the downstream section, we introduce the identity matrix E and the column vector of coefficients to be solved. Construct a flow recursive formula and estimate the parameters using known parameters within a periodic period to obtain the utilization rate. The flow rate Q at the downstream section is calculated recursively.
[0017] Through the aforementioned technical means, accurate step-by-step recursive calculations of downstream cross-section water level and flow rate are achieved. With water level change as the core variable to be solved, incremental calculations reduce simulation errors, enabling continuous recursive calculations of water level that align with the physical laws governing dynamic changes in river conditions. The flow rate recursive formula is based on the inherent physical relationship between water level and flow rate, and a linear equation is constructed using an identity matrix to achieve rapid flow rate recursion. This complements the water level calculation, completing the continuous simulation of "water level-flow rate". All parameters are readily available hydrological and geographical parameters (Δt, Δx, etc.), requiring no topographic data.
[0018] As a preferred embodiment, the method for measuring the accuracy of the simulation results of each hybrid learning algorithm includes: based on the simulation results of each hybrid learning algorithm, using the Nash efficiency coefficient as an error index to measure the accuracy of the simulation results of each hybrid learning algorithm.
[0019] As a preferred embodiment, the iterative update of the upstream section boundary conditions, weighted by the accuracy of the simulation results, includes: ; ; in, , This represents the upstream cross-section water level and flow rate after iteration; , This indicates the upstream cross-sectional water level and flow rate before the iteration; , The values represent the simulated downstream cross-section water level and flow rate corresponding to the s-th hybrid learning algorithm; n represents the current time step. Let represent the Nash efficiency coefficient of the simulation results obtained by the s-th hybrid learning algorithm.
[0020] By employing the aforementioned technical means, a linear combination is performed using the Nash efficiency coefficient as a weight to achieve the optimization logic that "the higher the simulation accuracy of the algorithm, the greater its contribution to the upstream boundary iteration," ensuring that the iteration direction always moves towards improving accuracy and avoiding ineffective iterations. The upstream boundary conditions before iteration are fused with the simulation results of the two algorithms, and the hydraulic characteristic information in the downstream simulation results is used to reverse-correct the upstream input, making up for the observation errors and preprocessing errors of the upstream boundary data, and strengthening the intrinsic correlation between upstream and downstream hydraulic parameters.
[0021] As a preferred embodiment, the secondary optimization of the simulation results by combining the water level-discharge relationship of the downstream section includes:
[0022] Based on the water level-flow relationship, another simulated value is re-estimated using a more accurate simulated water level or flow rate. The estimated result is compared with the original simulation result, and the more accurate result is selected as the final simulation result.
[0023] By utilizing the aforementioned technical means and taking advantage of the natural hydraulic correlation between water level and flow rate at river cross-sections, bidirectional mutual estimation and optimization of water level and flow rate simulation results can be achieved. This avoids the problem of high accuracy in simulation of one parameter but large deviation in the other parameter, thereby improving the overall accuracy of the simulation results. By comparing the estimated results with the original simulation results, it is ensured that the final output simulation value is the optimal value within the calibration period, providing high-quality basic results for the prediction period simulation.
[0024] A simulation device for river flood processes in areas with limited data under urban influence includes: The data acquisition and preprocessing module is used to acquire river boundary condition data of the study river section under the influence of urban areas. Based on the confluence flow and undetermined coefficients in the river section, the module performs preprocessing on the boundary condition data of the upstream section of the river. The undetermined coefficients are set with multiple combinations of values to be selected. The flood data extraction module is used to identify flood events based on upstream cross-section data and to identify flood events corresponding to upstream flood events in downstream cross-section data. It splices the upstream and downstream cross-section data corresponding to the flood events in chronological order to form the time series of flood event boundary conditions for each upstream and downstream cross-section. Multiple algorithm simulation modules are used for time series of flood events based on boundary conditions at upstream and downstream sections. Multiple hybrid learning algorithms based on one-dimensional river control equations are used to simulate and calculate water level and flow at downstream sections. The boundary condition iteration module is used to measure the accuracy of the simulation results of each hybrid learning algorithm. Using the accuracy of the simulation results as the weight, the boundary conditions of the upstream section are iteratively updated, and the iterated boundary conditions of the upstream section are resubmitted into the various hybrid learning algorithms for simulation. The results screening and optimization module is used to screen the combination of undetermined coefficient values with the best accuracy of the simulation results within the periodicity period. Then, it selects the simulation results with the highest accuracy within the periodicity period from the simulation results of multiple iterations. Combined with the water level-discharge relationship of the downstream section, the simulation results are optimized a second time to determine the optimal simulation calculation model for the river flood process.
[0025] A storage medium on which a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of simulating the data-deficient river flood process under urban influence.
[0026] The beneficial effects of this invention are: this invention does not rely on underwater topographic data throughout the entire process, and only uses easily obtainable boundary condition data such as upstream measured flow, tributary confluence flow, inter-regional confluence flow, and upstream and downstream water level and flow to complete the flood process simulation. At the same time, through upstream boundary condition iteration and simulation result optimization, it achieves accurate flood simulation without topographic data, and is suitable for river scenarios affected by urban areas and where underwater topographic data is difficult to obtain.
[0027] This invention improves the utilization efficiency of limited and scarce river hydrological data by integrating time series data from multiple flood events; the hybrid learning algorithm adopts a lightweight coefficient fitting method, which only requires a small amount of flood data within the calibration period to complete model training, without the need for massive amounts of high-resolution historical observation data, thus significantly reducing the requirements for data quantity and quality.
[0028] This invention uses the difference form of the momentum conservation equation in the Saint-Venant equations as the underlying physical constraint of the hybrid learning algorithm. All simulation calculations follow the momentum conservation law of flood evolution. At the same time, the simulation values are optimized collaboratively through the hydraulic correlation between water level and flow rate, which fundamentally ensures the physical consistency of the simulation results and avoids physical inconsistencies such as non-conservation of water volume and incorrect flow direction.
[0029] The hybrid learning algorithms of this invention are all based on well-defined hydrodynamic physical laws. Each step of the simulation process (coefficient estimation, water level-discharge recursion, boundary condition iteration) can be explained by the physical laws of flood evolution. The undetermined coefficients in the model are also directly related to the physical characteristics of flood evolution, breaking the black box nature of traditional machine learning and providing interpretable simulation results for flood control decision-making.
[0030] This invention continuously improves simulation accuracy even with limited data by employing multi-algorithm collaborative simulation, iterative boundary conditions with precision weights, and multi-level simulation result screening and optimization. It can accurately capture key flood characteristics such as peak occurrence time, peak water level, and peak flow. At the same time, the method of this invention has high computational efficiency, is suitable for the characteristics of urban floods, which are characterized by strong suddenness and short warning periods, and the corresponding devices and storage media can realize the automated and intelligent execution of the method, making it easy to promote and apply in actual flood control work. Attached Figure Description
[0031] Figure 1 This is a flowchart of the method for simulating river flood processes with insufficient data under urban influence in the embodiments.
[0032] Figure 2 This is a schematic diagram of the river section studied in the example.
[0033] Figure 3 The results of seven flood events simulated at Pengshan Station are shown in the examples.
[0034] Figure 4 The results of simulations of seven flood events at the Gaochang station are shown in the examples. Detailed Implementation
[0035] The following detailed description of the method for simulating river flood processes in areas lacking data under urban influences, in conjunction with the accompanying drawings and specific implementation examples, further illustrates the present invention.
[0036] Example 1: As Figure 1 As shown, this embodiment is a method for simulating flood processes in data-scarce rivers under urban influences, specifically including the following steps: S100. Obtain river boundary condition data for the study river section under urban influence. Perform preprocessing on the upstream section boundary condition data based on the confluence flow and undetermined coefficients within the river section. Multiple sets of undetermined coefficients are set.
[0037] This embodiment selects numerical simulation of flood events at two stations on the Minjiang River main stream, Pengshan and Gaochang, where data on river network and underwater topography are insufficient due to the influence of the Chengdu metropolitan area. The study river sections are the Zipingpu to Pengshan and Pengshan to Gaochang sections of the Minjiang River main stream. In the Zipingpu to Pengshan section, Zipingpu station is used as the upstream section and Pengshan station as the downstream section; in the Pengshan to Gaochang section, Pengshan station is used as the upstream section and Gaochang station as the downstream section. The simulation object in this case study is the hourly-scale flood process at the Pengshan and Gaochang hydrological stations.
[0038] In this embodiment, the water level and flow rate at the upstream section represent the upstream inflow, while the water level and flow rate at the downstream section represent the flood event. The flow rates of tributaries such as the Dadu River flowing into the Minjiang River are collected, and a distributed watershed hydrological model is used to simulate the inter-regional confluence flow. These flow rates represent tributary flow and inter-regional confluence flow, respectively.
[0039] The Chengdu Metropolitan Area is located on the Minjiang River from Zipingpu to Pengshan. During the city's infrastructure and transportation development, river channel improvement, wetland construction, and the construction of cross-river bridges have impacted flood events at the Pengshan and Gaochang hydrological stations. Therefore, the boundary condition data collected from the Pengshan and Gaochang hydrological stations can characterize Chengdu's influence on the boundary of the Minjiang River channel.
[0040] The river boundary condition data obtained in this embodiment mainly includes the measured water level Z at the upstream section. up and traffic Q up Total flow of tributaries Q tr Interval flow rate q L Measured water level Z at downstream section down and traffic Q down The water level and flow rate to be determined at the Pengshan and Gaochang hydrological stations are represented by Z and Q, respectively. The water level Z and flow rate Q at Pengshan and Gaochang stations are collected periodically, while the remaining data are collected periodically and during the forecast period.
[0041] After completing data acquisition, this embodiment will use the measured flow rate Q at the upstream section. up Tributary flow Q tr Interval flow rate q L Perform preprocessing according to the following formula: (1) Among them, Q up ′ is the upstream cross-sectional flow rate after pretreatment; Q tr It is the total flow of the tributary; q L It is the inter-regional flow rate; a1 and b1 are undetermined coefficients.
[0042] This embodiment conforms to the hydrodynamic physical laws of the river channel and sets multiple combinations of undetermined coefficients a1 and b1. Each combination is a linear adjustment coefficient used to eliminate the interference of tributary flow and inter-regional confluence on upstream water.
[0043] S200. Based on upstream cross-section data, identify each flood event and determine the flood events corresponding to the upstream flood events in downstream cross-section data. Then, splice the upstream and downstream cross-section data corresponding to the flood events in chronological order to form the time series of flood event boundary conditions for each upstream and downstream cross-section.
[0044] Analyze the raw (before preprocessing) upstream cross-section water level and flow data to identify each flood event, standardize the duration of each flood event (e.g., set the flood duration to 120 hours), and extract the corresponding data for each flood event from the upstream cross-section data after preprocessing.
[0045] Identify the flood events corresponding to each flood event in the downstream section data and extract the corresponding data for each flood event.
[0046] By concatenating the time series of each flood event in chronological order, time series of water levels and flows for both upstream and downstream sections are generated. The integrated time series are then divided into a calibration period and a forecast period, each of which includes several floods and does not overlap in time.
[0047] S300, based on the time series of flood event boundary conditions at upstream and downstream sections, employs a variety of hybrid learning algorithms based on one-dimensional river control equations to simulate and calculate the water level and flow rate at the downstream section.
[0048] This embodiment employs a hybrid learning algorithm based on two methods: the difference form of the momentum conservation equations from the Saint-Venant equations, which is used to adjust the fitting coefficients of periodic data and then analyze the water level at the downstream section during the prediction period. ,flow The recursive simulation uses the difference form of the momentum conservation equation in the Saint-Venant equations for the one-dimensional river channel control equations. The specific execution of the two algorithms is as follows: The first algorithm is based on the difference form of the momentum conservation equation in Saint-Venant's equations, and estimates the water level and flow rate at the downstream section according to the following formula: (2) (3) Wherein, the superscripts n and n+1 of the parameters represent the parameters at time n and n+1, respectively; E is the identity matrix; b3 and b4 are column vectors of coefficients to be solved, which are obtained by estimating the known parameters within the calibration period through sliding learning.
[0049] A linear approximation of water level and flow rate is made based on the difference form of the momentum conservation equation in the Saint-Venant equations, introducing the identity matrix E and the column vector of coefficients to be solved. Construct a water level estimation formula. The parameters are estimated using known parameters within the calibration period; the identity matrix E and the column vector of coefficients to be solved are introduced. Construct a flow recursive formula and estimate the parameters using known parameters within a periodic period to obtain the utilization rate. .
[0050] During the rate period, in equations (2) and (3), Z is excluded. n+1 Q n+1 Since all variables are known, the solution to the two equations with two unknown parameters can be achieved by solving a system of linear equations.
[0051] The second algorithm, based on the first, uses an iterative method to solve for the downstream cross-sectional water level. The change in downstream water level ΔZ within any time step Δt is calculated using the following formula: (4) Where b′, a2, and b2 are undetermined coefficients; Δt is the time step; Δx is the distance between the upstream and downstream sections; ΔZ up ΔZ represents the change in water level at the upstream and downstream cross sections after preprocessing within the time step.
[0052] Based on the collected data, only three unknown coefficients, b′, a2, and b2, exist in equation (4) during the calibration period, while the remaining parameters are known. First, the range of values for a2 and b2 is defined. Then, through sliding learning during the calibration period, the least squares method and other fitting methods are used to determine the most ideal b′, so that the water level achieves the best simulation results during the calibration period.
[0053] Subsequently, the downstream cross-sectional flow rate is calculated according to equation (3). Based on the water level Z at the downstream cross-section, the identity matrix E and the column vector of coefficients to be solved are introduced. Construct a flow recursive formula and estimate the parameters using known parameters within a periodic period to obtain the utilization rate. The flow rate Q at the downstream section is calculated recursively.
[0054] In this embodiment, after determining the estimated values of each coefficient within the calibration period according to the two algorithms mentioned above, Z and Q within the prediction period are recursively calculated using the pre-processed upstream cross-section data and the downstream cross-section data within the calibration period according to the first algorithm; according to the second algorithm, ΔZ at any time is calculated based on equation (4) starting from Z at the initial time, then Z at the next time is calculated, and Q at the next time is calculated in combination with equation (3), thereby recursively calculating Z and Q for the entire simulation period.
[0055] S400. Measure the accuracy of the simulation results of each hybrid learning algorithm, use the accuracy of the simulation results as the weight, iteratively update the boundary conditions of the upstream section, and resubmit the iterated boundary conditions of the upstream section into the various hybrid learning algorithms for simulation.
[0056] In this embodiment, after obtaining simulation results through two hybrid learning algorithms, the accuracy of the simulation results is measured using error metrics such as the Nash efficiency coefficient. (5) Among them, M and S and S represent the observed value and the average of the observed values, respectively; These represent the simulated value and the average of the simulated values, respectively.
[0057] Subsequently, this embodiment iterates the upstream cross-section water level and flow boundary by using a linear combination of the existing upstream cross-section water level and flow boundary, and the water level and flow simulation results obtained from the two algorithms, with the error index as the weight: (6) (7) in, , This represents the upstream cross-section water level and flow rate after iteration; , This indicates the upstream cross-sectional water level and flow rate before the iteration; , The values represent the simulated downstream cross-section water level and flow rate corresponding to the s-th hybrid learning algorithm; n represents the current time step. Let represent the Nash efficiency coefficient of the simulation results obtained by the s-th hybrid learning algorithm.
[0058] After the iteration is completed, the upstream cross-section water level and flow boundary will be simulated using the two algorithms mentioned above to simulate Z and Q. Based on the latest Z and Q simulation results and error indices, a new iteration can be performed on the upstream cross-section boundary conditions. The iteration will be performed 10 to 20 times, and the error indices of the Z and Q simulation results within the calibration period will be recorded after each iteration.
[0059] S500, the combination of undetermined coefficients with the best accuracy of the simulation results within the screening period, and then the simulation results with the highest accuracy within the screening period are selected from the simulation results of multiple iterations. The simulation results are then optimized a second time by combining the water level-discharge relationship of the downstream section, and finally the optimal simulation calculation model for the river flood process is determined.
[0060] S510. Traverse the multiple sets of a1 and b1 value combinations set in step S100, combine the simulation accuracy (Nash efficiency coefficient) within the periodic rate, and select the a1 and b1 combination that can make the simulation accuracy of downstream cross-section water level and flow rate optimal, and determine the optimal parameters for preprocessing.
[0061] S520. From the simulation results of multiple iterations, select the set of water level and flow simulation results with the highest Nash efficiency coefficient (optimal accuracy) within the rate period as the basic results.
[0062] S530. Utilize the inherent correlation between water level and flow rate within the downstream cross-sectional area period. If the accuracy of the water level simulation is higher than that of the flow rate simulation, then the flow rate simulation value is re-estimated using the more accurate water level simulation value. If the accuracy of the flow rate simulation is higher than that of the water level simulation, then the water level simulation value is re-estimated using the more accurate flow rate simulation value. Compare the estimation results with the original simulation results, and select the more accurate result as the final simulation result to achieve secondary optimization of the simulation results.
[0063] S540 integrates the optimal combination of a1 and b1, the fitted parameters b′, a2, b2, b3, and b4, and the complete logic of dual-algorithm simulation, boundary iteration, and result optimization to determine the optimal simulation calculation model for river flood processes. This model can be directly used for simulating river flood processes during the prediction period.
[0064] In this embodiment, when predicting river flood events, the optimal simulation model determined in step S500 is reused. No parameters need to be refitted or repeated iterations are required. Only the river boundary condition data for the prediction period is substituted into the preprocessing formula in step S100 (using the optimal combination of a1 and b1) to obtain the preprocessed data for the prediction period. Then, by substituting the values into the dual algorithm in the optimal model, the water level and flow rate of the downstream section during the prediction period are calculated, thus completing the simulation of the river flood process during the prediction period.
[0065] Taking the simulation case from 1980 to 1985 as an example, based on the hydrological model simulation results, seven floods lasting 120 hours were selected at each of the Pengshan and Gaochang stations as simulation objects. At each station, the calibration period consisted of the first four floods, and the validation period consisted of the last three floods. The simulation results of water level and flow processes for the flood events at the Pengshan and Gaochang stations are as follows: Figures 3-4 As shown.
[0066] Overall, the simulation results of water level and flow processes for flood events at Pengshan and Gaochang stations showed high accuracy, with NSE coefficients approaching 1.00. The timing of flood peaks, flow rates, and water levels were all simulated relatively accurately for each flood event. This demonstrates that the present invention can reproduce and predict hourly-scale flood events occurring at Pengshan and Gaochang stations.
[0067] In the aforementioned simulation case, underwater topographic data was not provided for the more than 300km stretch of the Minjiang River main stream from Zipingpu Station to Gaochang Station. The complex branching and confluence relationships of the Minjiang River near Chengdu were not considered, nor were the impacts of municipal and transportation construction in the Chengdu metropolitan area on flood conditions. However, the method proposed in this invention still effectively simulated flood events at Pengshan and Gaochang stations, accurately capturing key information such as peak flow time, water level, and flow rate. This demonstrates that this invention can overcome problems such as insufficient topographic data and a weak connection between machine learning algorithms and physical laws, exhibiting certain advantages compared to existing hydrodynamic and machine learning models. In conclusion, this invention can effectively conduct numerical simulations of flood events in river sections with insufficient topographic data and those influenced by urban areas.
[0068] Example 2: This example is a simulation device for flood processes in rivers with limited data under urban influence, comprising: The data acquisition and preprocessing module is used to acquire river boundary condition data of the study river section under the influence of urban areas. Based on the confluence flow and undetermined coefficients in the river section, the module performs preprocessing on the boundary condition data of the upstream section of the river. The undetermined coefficients are set with multiple combinations of values to be selected. The flood data extraction module is used to identify flood events based on upstream cross-section data and to identify flood events corresponding to upstream flood events in downstream cross-section data. It splices the upstream and downstream cross-section data corresponding to the flood events in chronological order to form the time series of flood event boundary conditions for each upstream and downstream cross-section. Multiple algorithm simulation modules are used for time series of flood events based on boundary conditions at upstream and downstream sections. Multiple hybrid learning algorithms based on one-dimensional river control equations are used to simulate and calculate water level and flow at downstream sections. The boundary condition iteration module is used to measure the accuracy of the simulation results of each hybrid learning algorithm. Using the accuracy of the simulation results as the weight, the boundary conditions of the upstream section are iteratively updated, and the iterated boundary conditions of the upstream section are resubmitted into the various hybrid learning algorithms for simulation. The results screening and optimization module is used to screen the combination of undetermined coefficient values with the best accuracy of the simulation results within the periodicity period. Then, it selects the simulation results with the highest accuracy within the periodicity period from the simulation results of multiple iterations. Combined with the water level-discharge relationship of the downstream section, the simulation results are optimized a second time to determine the optimal simulation calculation model for the river flood process.
[0069] Example 3: This example is a storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of simulating the data-deficient river flood process under urban influence as described in Example 1.
Claims
1. A method for simulating flood processes in data-scarce rivers under urban influence, characterized in that, include: Data on river boundary conditions under the influence of urban areas were obtained. Based on the confluence flow and undetermined coefficients within the river section, the boundary condition data of the upstream section of the river were preprocessed. Multiple sets of undetermined coefficients were set. Based on upstream cross-section data, each flood event is identified, and the flood events corresponding to the upstream flood events are identified in the downstream cross-section data. The upstream and downstream cross-section data corresponding to the flood events are spliced together in chronological order to form the time series of the boundary conditions of the flood events for the upstream and downstream cross-sections respectively. Based on the time series of flood event boundary conditions at upstream and downstream sections, a hybrid learning algorithm based on one-dimensional river control equations is used to simulate and calculate the water level and flow rate at the downstream section. The simulation results accuracy of each hybrid learning algorithm is measured, and the upstream section boundary conditions are iteratively updated using the simulation results accuracy as the weight. The iterated upstream section boundary conditions are then re-substituted into the various hybrid learning algorithms for simulation. The combination of undetermined coefficient values with the best accuracy in the simulation results within the screening period is selected. Then, the simulation results with the highest accuracy within the screening period are selected from the simulation results of multiple iterations. The simulation results are then optimized a second time by combining the water level-discharge relationship of the downstream section, and finally the optimal simulation calculation model for the river flood process is determined.
2. The method for simulating river flood processes in areas with limited data under urban influences according to claim 1, characterized in that, The preprocessing of the upstream cross-sectional boundary condition data based on the confluence flow and undetermined coefficients within the river section includes: The measured flow at the upstream section, the total flow of the tributaries, and the flow of the confluence of the two sections are integrated and calculated according to the following formula; ； Among them, Q up ′ is the upstream cross-sectional flow rate after pretreatment; Q tr It is the total flow of the tributary; q L It is the inter-regional flow rate; a1 and b1 are undetermined coefficients.
3. The method for simulating river flood processes in areas with limited data under urban influences according to claim 1, characterized in that, The one-dimensional river channel control equation is the difference form of the momentum conservation equation in the Saint-Venant equations.
4. The method for simulating river flood processes in areas with limited data under urban influences according to claim 3, characterized in that, One of the various hybrid learning algorithms based on one-dimensional river control equations includes: ； ； A linear approximation of water level and flow rate is made based on the difference form of the momentum conservation equation in the Saint-Venant equations, introducing the identity matrix E and the column vector of coefficients to be solved. Construct a water level estimation formula. The parameters are estimated using known parameters within the calibration period; the identity matrix E and the column vector of coefficients to be solved are introduced. Construct a flow recursive formula and estimate the parameters using known parameters within a periodic period to obtain the utilization rate. ; By combining the flow rate recursion formula and the water level estimation formula, and taking the water level at time n+1... ,flow The unknown quantity is used to solve for the water level and flow rate at the downstream section.
5. The method for simulating river flood processes in areas with limited data under urban influences according to claim 3, characterized in that, One of the various hybrid learning algorithms based on one-dimensional river control equations includes: ； A water level calculation formula is constructed based on the difference form of the momentum conservation equation in the Saint-Venant equations, and a time step is introduced. Spatial distance from upstream section to downstream section Changes in upstream water level after pretreatment and undetermined coefficients , , ,Regulation , After determining the range of values, the optimal value is determined by sliding learning combined with least squares fitting within the pass rate period. Solve for the change in water level at the downstream section. Then, the water level Z at the downstream section is calculated; ； Based on the water level Z at the downstream section, we introduce the identity matrix E and the column vector of coefficients to be solved. Construct a flow recursive formula and estimate the parameters using known parameters within a periodic period to obtain the utilization rate. The flow rate Q at the downstream section is calculated recursively.
6. The method for simulating river flood processes in areas with limited data under urban influences according to claim 1, characterized in that, The method for measuring the accuracy of the simulation results of each hybrid learning algorithm includes: using the Nash efficiency coefficient as an error index to measure the accuracy of the simulation results of each hybrid learning algorithm based on the simulation results of each hybrid learning algorithm.
7. The method for simulating river flood processes in areas with limited data under urban influences according to claim 1, characterized in that, The iterative update of the upstream section boundary conditions, weighted by the accuracy of the simulation results, includes: ； ； in, , This represents the upstream cross-section water level and flow rate after iteration; , This indicates the upstream cross-sectional water level and flow rate before the iteration; , The values represent the simulated downstream cross-section water level and flow rate corresponding to the s-th hybrid learning algorithm; n represents the current time step. Let represent the Nash efficiency coefficient of the simulation results obtained by the s-th hybrid learning algorithm.
8. The method for simulating river flood processes in areas with limited data under urban influences according to claim 1, characterized in that, The secondary optimization of the simulation results by combining the water level-discharge relationship of the downstream section includes: Based on the water level-flow relationship, another simulated value is re-estimated using a more accurate simulated water level or flow rate. The estimated result is compared with the original simulation result, and the more accurate result is selected as the final simulation result.
9. A device for simulating flood processes in rivers with limited data under urban influence, characterized in that, include: The data acquisition and preprocessing module is used to acquire river boundary condition data of the study river section under the influence of urban areas. Based on the confluence flow and undetermined coefficients in the river section, the module performs preprocessing on the boundary condition data of the upstream section of the river. The undetermined coefficients are set with multiple combinations of values to be selected. The flood data extraction module is used to identify flood events based on upstream cross-section data and to identify flood events corresponding to upstream flood events in downstream cross-section data. It splices the upstream and downstream cross-section data corresponding to the flood events in chronological order to form the time series of flood event boundary conditions for each upstream and downstream cross-section. Multiple algorithm simulation modules are used for time series of flood events based on boundary conditions at upstream and downstream sections. Multiple hybrid learning algorithms based on one-dimensional river control equations are used to simulate and calculate water level and flow at downstream sections. The boundary condition iteration module is used to measure the accuracy of the simulation results of each hybrid learning algorithm. Using the accuracy of the simulation results as the weight, the boundary conditions of the upstream section are iteratively updated, and the iterated boundary conditions of the upstream section are resubmitted into the various hybrid learning algorithms for simulation. The results screening and optimization module is used to screen the combination of undetermined coefficient values with the best accuracy of the simulation results within the periodicity period. Then, it selects the simulation results with the highest accuracy within the periodicity period from the simulation results of multiple iterations. Combined with the water level-discharge relationship of the downstream section, the simulation results are optimized a second time to determine the optimal simulation calculation model for the river flood process.
10. A storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of simulating a data-deficient river flood process under urban influence as described in any one of claims 1-8.