CN-122021073-A - River evolution model construction method and device, electronic equipment and readable storage medium

CN122021073ACN 122021073 ACN122021073 ACN 122021073ACN-122021073-A

Abstract

The invention provides a river evolution model construction method, a river evolution model construction device, electronic equipment and a readable storage medium, and relates to the technical field of intelligent water conservancy. By constructing the normal priori distribution of the river bed roughness parameters and combining the customized likelihood function and the Bayesian inference framework, the accurate estimation of the posterior distribution of the parameters is realized, the defects that the traditional method only outputs a single optimal solution, is easy to sink and locally optimal and cannot quantify uncertainty are overcome, and the parameter calibration precision is remarkably improved. And the self-adaptive and multi-chain parallel dynamic simulation algorithm is adopted to sample in parallel and dynamically adjust the step length, so that the sampling efficiency and the convergence speed in the complex posterior space are greatly improved. The consistency of the model and the actual hydraulic process is enhanced through the deep coupling of the hydrodynamic error model and the likelihood function. And the parameter uncertainty is transmitted to a simulation result, and the mean value, variance and confidence interval of the prediction results such as water level, flow and the like are constructed, so that a scientific and reliable uncertainty quantification basis is provided for flood early warning and scheduling decision.

Inventors

WANG GUOMIAO
ZHAO HAORAN
WU JIANMING

Assignees

浙江远算科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The river evolution model construction method is characterized by comprising the following steps of: Obtaining original data in a target area, and carrying out standardization processing, missing value filling and abnormal value detection on the original data to form standardized time sequence data, wherein the original data at least comprises water level historical monitoring data and flow historical monitoring data; Based on river section sweep data, river bed soil type, bank slope structure and basic topography data of vegetation coverage, constructing a river evolution model by combining an external hydraulic driving field of water level and flow, and simulating the river evolution model by utilizing the driving field to obtain a river flow response process under a multi-rainfall scene; Constructing prior distribution of river bed roughness parameters according to the river bed soil type, the grain size composition, the vegetation coverage and a preset hydraulic empirical formula, wherein the prior distribution adopts normal distribution, and the average value and the variance of the normal distribution are determined by consulting an empirical parameter interval, a historical calibration result and a literature statistical value so as to describe a reasonable value range of the river bed roughness parameters under the condition of no observation constraint; Introducing observation data into a Bayesian inference framework on the basis of the river evolution model and the prior distribution, constructing a hydrodynamic error model and combining hydrodynamic characteristics, adopting a preset customized likelihood function to reflect actual observation errors, sampling by a self-adaptive and multi-chain parallel dynamic simulation algorithm, and updating posterior distribution of the river bed roughness parameters; And transmitting uncertainty of posterior distribution of the river bed roughness parameters to a river channel hydrodynamic simulation result, sampling from the posterior distribution to generate parameter samples, inputting each parameter sample into the river channel evolution model to operate, obtaining simulation results of multiple groups of water levels, flow rates and flow velocities, carrying out aggregate statistics on the simulation results, and constructing confidence intervals of mean value prediction, variance and preset confidence level to realize uncertainty quantification of model prediction.
2. The river evolution model construction method according to claim 1, wherein the steps of performing standardization processing, missing value filling and abnormal value detection on the raw data comprise: The method comprises the steps of carrying out dimension unified processing on the original data based on a Min-Max standardization method, and scaling the data to a [0,1] range to obtain standardized data through the following formula: ; Wherein, the The original value of the ith variable at time t; Normalized value at time t for the ith variable; As a variable A minimum value over the whole acquisition period; As a variable Maximum value in the whole acquisition period; Filling up the missing value in the standardized data by adopting the following linear interpolation formula to obtain complete time sequence data, wherein the calculation formula is as follows: ; Wherein, the Is the padded value; Is the time of A known standard value at; Is the time of A known standard value at; Is the missing moment; Is the known time a; is the known time b; detecting abnormal values of the complete time sequence data based on Z-Score statistics, identifying data points which are remarkably inconsistent with data distribution, and replacing abnormal points exceeding a preset threshold value by using a neighborhood mean value, wherein the calculation formula of the Z-Score value is as follows: ; ; ; Wherein, the A sliding window with a preset fixed time interval; is the mean value within the sliding window; Standard deviation within a sliding window; is a Z-Score value.
3. The river evolution model construction method according to claim 1, wherein the river evolution model is constrained by the following system of san velnan equations: ; ; wherein Q is the instantaneous section flow of the river reach in the unit of m 3 /s, A is the water cross-section area of the river reach in the unit of m 2 , x is the length of the river reach in the unit of m, t is the time in the unit of s; b is the width of the water surface, the unit is m, g is the gravity acceleration; Is the slope of the river bed; Is the friction coefficient; is the lateral inflow velocity.
4. The river evolution model construction method of claim 1, wherein the prior distribution of the riverbed roughness parameters is constrained based on the following formula: ; Wherein, the Is the river bed roughness parameter, mu is the average value of the river bed roughness parameter, Is the variance of the river bed roughness parameters; Is the prior probability density of the river bed roughness parameter.
5. The river evolution model construction method of claim 1, wherein the customized likelihood function is constrained by the following formula: ; Wherein, the -Customizing likelihood functions for the user; Is the error variance; For the i-th observation; The i output quantity of the river evolution model is obtained, and N is the number of samples.
6. The method of claim 1, wherein the step of updating the posterior distribution of the river bed roughness parameters by sampling with an adaptive and multi-chain parallel dynamic simulation algorithm comprises: Determining a posterior distribution of the riverbed roughness parameters without normalization based on the prior distribution of the riverbed roughness parameters and the customized likelihood function; Sampling by adopting a multi-chain parallel-based self-adaptive Markov chain Monte Carlo algorithm, running all sampling chains in parallel, and gradually adjusting the step length based on the residual value and the preset acceptance rate in the sampling process until all the sampling chains are converged, wherein the initial parameters of each sampling chain are obtained by sampling from prior distribution; after all the sampling chains are converged, constructing posterior distribution of the river bed roughness parameters based on sampling results of all the sampling chains.
7. The method for constructing a model of river evolution according to claim 1, wherein the step of propagating uncertainty of posterior distribution of the river bed roughness parameters into a river hydrodynamic simulation result, generating parameter samples by sampling from posterior distribution, inputting each of the parameter samples into the model of river evolution to operate, obtaining simulation results of a plurality of groups of water levels, flow rates and flow velocities, and performing aggregate statistics on the simulation results to construct confidence intervals of mean prediction, variance and preset confidence levels to realize uncertainty quantization of model prediction comprises: Sampling from posterior distribution of the river bed roughness parameters to generate a parameter sample set; Inputting each sample into the river evolution model to operate, so as to obtain a hydrodynamic simulation result; and carrying out statistical analysis on all sample results at each moment and space position, wherein the analysis results comprise mean prediction, variance and confidence intervals of the confidence level.
8. The river evolution model construction device is characterized by comprising: The observation data processing module is used for obtaining original data in a target area, and carrying out standardization processing, missing value filling and abnormal value detection on the original data to form standardized time sequence data, wherein the original data at least comprises water level historical monitoring data and flow historical monitoring data; The river flow evolution module is used for constructing a river flow evolution model based on basic topography and topography data of river section sweep data, river bed soil types, bank slope structures and vegetation coverage and combining an external hydraulic driving field of water level and flow, and simulating the river flow evolution model by utilizing the driving field to obtain a river flow response process under a multi-rainfall scene; The prior information configuration module is used for constructing prior distribution of the river bed roughness parameters according to the river bed soil type, the grain size composition, the vegetation coverage and a preset hydraulic empirical formula, wherein the prior distribution adopts normal distribution, and the average value and the variance of the normal distribution are determined by consulting an empirical parameter interval, a historical calibration result and a literature statistical value so as to describe a reasonable value range of the river bed roughness parameters under the condition of no observation constraint; The multi-chain parallel sampling module is used for introducing observation data into a Bayesian inference framework on the basis of the river evolution model and the prior distribution, adopting a preset customized likelihood function to reflect the actual observation error by constructing a hydrodynamic error model and combining with hydrodynamic characteristics, sampling by a self-adaptive and multi-chain parallel dynamic simulation algorithm, and updating posterior distribution of the river bed roughness parameters; The uncertainty analysis module is used for spreading uncertainty of posterior distribution of the river bed roughness parameters to a river channel hydrodynamic force simulation result, sampling from the posterior distribution to generate parameter samples, inputting each parameter sample into the river channel evolution model for operation, obtaining simulation results of multiple groups of water levels, flow rates and flow velocities, carrying out aggregate statistics on the simulation results, and constructing confidence intervals of mean value prediction, variance and preset confidence level so as to realize uncertainty quantification of model prediction.
9. An electronic device comprising a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the river evolution model construction method of any one of the preceding claims 1-7.
10. A readable storage medium storing computer executable instructions which, when invoked and executed by a processor, cause the processor to implement the river evolution model construction method of any one of claims 1 to 7.

Description

River evolution model construction method and device, electronic equipment and readable storage medium Technical Field The invention relates to the technical field of intelligent water conservancy, in particular to a river evolution model construction method, a river evolution model construction device, electronic equipment and a readable storage medium. Background In the field of hydraulic engineering, the river evolution model is widely applied to decision scenes such as flood prediction, water resource scheduling and the like. The river bed roughness is taken as a key hydrodynamic parameter, has a decisive influence on the model output precision, is difficult to directly measure, and is usually calibrated through historical data. The existing parameter calibration method mainly adopts a deterministic optimization technology (such as a least square method) and can only output single optimal parameter values, and has the technical problems that firstly, the parameter estimation precision and uncertainty characterization capability are insufficient, when facing to complex posterior distribution of multimodal, nonlinear and high-dimensional correlation, local optimization is easy to fall in, confidence intervals of prediction results cannot be provided, reliable basis is difficult to be provided for high-risk decisions such as flood early warning and the like, secondly, an error modeling mode and hydraulic characteristic coupling are insufficient, a fixed variance noise assumption is adopted in the traditional method, water level-flow conversion errors, response hysteresis effects and heteroscedastic characteristics are ignored, an error model and a hydrodynamic equation are disjointed, thirdly, the sampling efficiency and convergence performance are difficult to meet engineering requirements, a multi-chain parallel and self-adaptive adjustment mechanism is lacked, and the calculation efficiency is low and the convergence speed is low under complex scenes such as large flow fields, long sequences and the like. Therefore, a river evolution model construction method capable of coupling hydrodynamic characteristics, quantifying parameter uncertainty and improving sampling efficiency is needed. Disclosure of Invention The invention aims to provide a river evolution model construction method, a device, electronic equipment and a readable storage medium, which are used for realizing accurate estimation of parameter posterior distribution by constructing normal priori distribution of river bed roughness parameters and combining a customized likelihood function and a Bayesian inference framework, overcoming the defects that the traditional method only outputs a single optimal solution, is easy to sink and is locally optimal and can not quantify uncertainty, and obviously improving parameter calibration precision. And the self-adaptive and multi-chain parallel dynamic simulation algorithm is adopted to sample in parallel and dynamically adjust the step length, so that the sampling efficiency and the convergence speed in the complex posterior space are greatly improved. The consistency of the model and the actual hydraulic process is enhanced through the deep coupling of the hydrodynamic error model and the likelihood function. And the parameter uncertainty is transmitted to a simulation result, and the mean value, variance and confidence interval of the prediction results such as water level, flow and the like are constructed, so that a scientific and reliable uncertainty quantification basis is provided for flood early warning and scheduling decision. In a first aspect, the present invention provides a method for constructing a river evolution model, including: Obtaining original data in a target area, and carrying out standardized processing, missing value filling and abnormal value detection on the original data to form standardized time sequence data, wherein the original data at least comprises water level historical monitoring data and flow historical monitoring data; Based on river section sweep data, river bed soil type, bank slope structure and basic topography data of vegetation coverage, constructing a river evolution model by combining an external hydraulic driving field of water level and flow, and simulating the river evolution model by utilizing the driving field to obtain a river flow response process under a multi-rainfall scene; constructing prior distribution of river bed roughness parameters according to the river bed soil type, the grain size composition, the vegetation coverage and a preset hydraulic empirical formula, wherein the prior distribution adopts normal distribution, and the average value and the variance of the normal distribution are determined by consulting an empirical parameter interval, a historical calibration result and a literature statistical value so as to describe a reasonable value range of the river bed roughness parameters under the condition of no observation constraint