CN-121980246-A - Mechanism data double-drive flood forecasting method and system

CN121980246ACN 121980246 ACN121980246 ACN 121980246ACN-121980246-A

Abstract

The application discloses a mechanism data double-drive flood forecasting method and system, which belong to the technical field of computers, and effectively solve the problems of low calculation efficiency of a physical mechanism model and lack of physical constraint of a data drive model in the traditional method by combining a physical mechanism model forecasting and a generated state error model analysis through a central routing intelligent agent dynamic distribution calculation mode and assisting in physical consistency forced verification. Therefore, the depth coordination of real-time physical verification and dynamic error compensation in the forecasting process is realized, and the physical reliability, the calculation efficiency and the adaptability to extreme flood events are improved simultaneously in the flood forecasting.

Inventors

Bao Changchi
CHEN PENG
DU YANPENG
YANG FUMING
CHEN JIANLING
ZHANG SHUN
WU JIANGPING
WU MENGYANG
ZHOU TIAN

Assignees

浙江安澜工程技术有限公司
泰顺县水利局
文成县水利局

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. A method of mechanism data double-driven flood forecasting, the method comprising: Responding to a flood forecasting task of a target river basin, analyzing and making a routing decision on real-time multi-source flow domain characteristic data of a current period through a central routing intelligent agent to obtain a model routing diagram of the target river basin, wherein the model routing diagram is used for dynamically distributing calculation modes to a plurality of calculation units respectively; Based on the model route diagram and the model state corrected in the previous period, a calculation unit distributed into a coupling mode carries out forecast calculation through a physical mechanism model of a drainage basin scale to obtain a mechanism forecast flow result and a corresponding internal physical state sequence; Performing state error generation analysis on the internal physical state sequence and the real-time observation data through a generated state error model to obtain a physical state correction quantity, and performing verification and filtration on the physical state correction quantity through a physical consistency forced verifier based on a physical conservation law to obtain an effective state correction quantity; correcting the state variable of the physical mechanism model based on the effective state correction quantity, and carrying out fusion calculation on the forecast results of all calculation units by utilizing the corrected state variable based on the model route diagram to obtain the flood forecast result of the target river basin.
2. The method of claim 1, wherein the analyzing and routing decisions of the real-time multisource domain feature data for the current time period by the central routing agent results in a model routing graph for the target domain, comprising: performing multidimensional feature analysis on the real-time multisource domain feature data, and extracting a key feature set, wherein the key feature set is used for representing hydrologic response urgency and mechanism model applicability of different areas in the target domain; Based on the key feature set, performing pattern matching and decision through a routing decision network in the central routing agent, and generating a corresponding pattern allocation label for each computing unit, wherein the pattern allocation label is used for indicating a computing pattern; The model route map is generated and output based on the pattern assignment labels of all the computing units.
3. The method of claim 2, wherein the generating a corresponding pattern assignment tag for each computing unit based on the set of key features by pattern matching and decision through a routing decision network in the central routing agent comprises: For each computing unit, performing spatial context correlation analysis through the routing decision network based on the key feature set of the computing unit and the key feature set of a spatially adjacent computing unit of the computing unit, and generating a spatially enhanced decision feature of the computing unit; Based on the space enhancement decision feature, carrying out multi-factor collaborative decision through the routing decision network, and generating preliminary mode allocation probability of the computing unit, wherein factors of the multi-factor collaborative decision at least comprise a real-time rainfall intensity factor and a mechanism model historical error trend factor; and carrying out space consistency optimization processing on the preliminary mode allocation probability based on a physical constraint rule of the target river basin to obtain the mode allocation label of the computing unit, wherein the physical constraint rule is used for enabling mode allocation among units communicated by hydrology to accord with a physical rule.
4. The method according to claim 1, wherein the calculating unit allocated as a coupling mode based on the model routing graph and the model state corrected in the previous period performs a prediction calculation through a physical mechanism model of a drainage basin scale to obtain a mechanism prediction flow result and a corresponding internal physical state sequence, and the method comprises: Screening out all calculation units allocated as coupling modes from all calculation units of the target drainage basin based on the model route diagram; For each screened computing unit, reading the corresponding model state corrected in the previous period from a model state library subjected to physical verification correction, and taking the model state corrected in the previous period as a forecast initial condition of the computing unit in the current period; And driving the physical mechanism model of the river basin scale to execute forecast calculation based on the forecast initial conditions, and outputting a mechanism forecast flow result of the calculation unit and an internal physical state sequence generated according to a preset time resolution.
5. The method of claim 4, wherein driving the basin-scale physical mechanism model to perform a predictive calculation based on the predictive initial conditions, outputting a mechanism predictive traffic result of the calculation unit and an internal physical state sequence generated at a preset time resolution, comprises: Setting a calculation time step and an output frequency of the physical mechanism model running in a current forecast period based on the configuration parameters of the coupling mode; Based on the forecasting initial conditions, the physical mechanism model is operated step by step in an iterative mode according to the calculated time step until simulation of the current forecasting period is completed; In the iterative operation process, recording the flow value of the computing unit at the corresponding moment of each output frequency to form the mechanism forecast flow result; and capturing all key state variable snapshots of the physical mechanism model at corresponding moments to assemble the internal physical state sequence according to the preset time resolution.
6. The method of claim 1, wherein said performing a state error generation analysis on said internal physical state sequence and real-time observation data by generating a state error model to obtain a physical state correction comprises: performing space-time alignment and feature fusion on the internal physical state sequence and the corresponding real-time observation data to form a fusion data sample for error analysis; Inputting the fusion data sample into the generated state error model, learning a deviation mode between the internal physical state sequence and the real-time observation data through the generated state error model, and generating an initial physical state correction quantity; And carrying out standardized conversion on the initial physical state correction quantity to obtain the physical state correction quantity matched with the state variable of the physical mechanism model in the dimension and physical sense.
7. The method of claim 6, wherein the inputting the fused data samples into the generated state error model, learning a pattern of deviations between the internal physical state sequence and the real-time observed data via the generated state error model, and generating an initial physical state correction, comprises: Based on the fusion data sample, constructing an explicit state deviation field between the internal physical state sequence and the real-time observation data as a conditional input of the generated state error model; Deducing and generating a high-dimensional state correction latent field through a condition generation mechanism of the generated state error model based on the explicit state deviation field and the space-time context of the internal physical state sequence; mapping the high-dimensional state correction latent field to a physical space and dimension corresponding to a state variable of the physical mechanism model to form the initial physical state correction.
8. The method of claim 1, wherein performing check filtering based on physical conservation law on the physical state correction amount by a physical consistency forced checker to obtain an effective state correction amount comprises: Inputting the physical state correction amount and the current state variable of the physical mechanism model into the physical consistency forced checker; in the physical consistency forced checker, a check equation set concerning the physical state correction amount is constructed based on the law of conservation of mass and conservation of momentum; solving the check equation set to obtain theoretical conservation deviation corresponding to the physical state correction quantity; And comparing the theoretical conservation deviation with a preset tolerance threshold, screening or correcting the physical state correction amount based on a comparison result, and outputting a part meeting the tolerance threshold as the effective state correction amount.
9. The method according to claim 1, wherein the performing, based on the model routing graph, fusion calculation on the forecast results of all the calculation units by using the corrected state variables to obtain the flood forecast results of the target river basin includes: Based on the model route diagram, identifying all calculation units respectively adopting different calculation modes in the target flow domain and a first space topological relation among the calculation units; For each calculation unit, obtaining a prediction result corresponding to the calculation unit, wherein for the calculation unit of the coupling mode, the prediction result is calculated based on the corrected state variable; Integrating the forecasting results of all the computing units into a spatially continuous watershed state field through a spatial fusion strategy based on the first spatial topological relation and the computing modes adopted by the computing units; and performing hydrologic evolution calculation on the drainage basin state field to obtain and output an outlet section flow process of the target drainage basin as the flood forecast result.
10. A mechanism data dual-driven flood forecast system, the system comprising: The analysis module is used for responding to flood forecast tasks of the target river basin, analyzing and routing decision is carried out on the real-time multi-source flow domain characteristic data of the current period through the central routing agent, and a model route diagram of the target river basin is obtained, and is used for dynamically distributing calculation modes for a plurality of calculation units respectively; The forecast calculation module is used for carrying out forecast calculation on the calculation unit distributed into a coupling mode based on the model route map and the model state corrected in the previous period, and obtaining a mechanism forecast flow result and a corresponding internal physical state sequence through a physical mechanism model of a river basin scale; The error analysis module is used for carrying out state error generation analysis on the internal physical state sequence and the real-time observation data through a generated state error model to obtain a physical state correction quantity, and carrying out verification and filtration based on a physical conservation law on the physical state correction quantity through a physical consistency forced verifier to obtain an effective state correction quantity; and the fusion calculation module is used for correcting the state variable of the physical mechanism model based on the effective state correction quantity, and carrying out fusion calculation on the forecasting results of all calculation units by utilizing the corrected state variable based on the model route diagram to obtain the flood forecasting result of the target river basin.

Description

Mechanism data double-drive flood forecasting method and system Technical Field The application relates to the technical field of computers, in particular to a mechanism data double-drive flood forecasting method and system. Background In the intelligent water conservancy field, flood forecast is used as a core support for flood control decision, and two technical routes, namely a physical mechanism model and a data driving model, are relied on for a long time. The physical mechanism model is constructed based on a hydrologic physical equation, and can reflect the internal mechanism of the watershed hydrologic process by explicitly expressing basic rules such as mass conservation, momentum conservation and the like, so that the model has stronger physical interpretability. On the other hand, the pure data driving model represented by the generated artificial intelligence can efficiently mine nonlinear association in the historical data to realize rapid forecasting, but the internal operation mechanism of the model is lack of transparency, like a black box, and under rare situations (such as ultra-standard storm or sudden dam break event) outside the distribution of training data, forecasting results against physical rules are often output, such as non-physical oscillation or unbalanced water balance of a flow process line, and reliability of forecasting are seriously damaged. In view of the above problems, improvements are needed in the related art. Disclosure of Invention The embodiment of the application provides a flood forecasting method and a flood forecasting system driven by mechanism data, and the technical scheme is as follows: In one aspect, a method for mechanism data dual-driven flood forecasting is provided, the method comprising: Responding to a flood forecasting task of a target river basin, analyzing and making a routing decision on real-time multi-source flow domain characteristic data of a current period through a central routing intelligent agent to obtain a model routing diagram of the target river basin, wherein the model routing diagram is used for dynamically distributing calculation modes to a plurality of calculation units respectively; Based on the model route diagram and the model state corrected in the previous period, a calculation unit distributed into a coupling mode carries out forecast calculation through a physical mechanism model of a drainage basin scale to obtain a mechanism forecast flow result and a corresponding internal physical state sequence; Performing state error generation analysis on the internal physical state sequence and the real-time observation data through a generated state error model to obtain a physical state correction quantity, and performing verification and filtration on the physical state correction quantity through a physical consistency forced verifier based on a physical conservation law to obtain an effective state correction quantity; correcting the state variable of the physical mechanism model based on the effective state correction quantity, and carrying out fusion calculation on the forecast results of all calculation units by utilizing the corrected state variable based on the model route diagram to obtain the flood forecast result of the target river basin. In one aspect, there is provided a mechanism data dual-driven flood forecast system, the system comprising: The analysis module is used for responding to flood forecast tasks of the target river basin, analyzing and routing decision is carried out on the real-time multi-source flow domain characteristic data of the current period through the central routing agent, and a model route diagram of the target river basin is obtained, and is used for dynamically distributing calculation modes for a plurality of calculation units respectively; The forecast calculation module is used for carrying out forecast calculation on the calculation unit distributed into a coupling mode based on the model route map and the model state corrected in the previous period, and obtaining a mechanism forecast flow result and a corresponding internal physical state sequence through a physical mechanism model of a river basin scale; The error analysis module is used for carrying out state error generation analysis on the internal physical state sequence and the real-time observation data through a generated state error model to obtain a physical state correction quantity, and carrying out verification and filtration based on a physical conservation law on the physical state correction quantity through a physical consistency forced verifier to obtain an effective state correction quantity; and the fusion calculation module is used for correcting the state variable of the physical mechanism model based on the effective state correction quantity, and carrying out fusion calculation on the forecasting results of all calculation units by utilizing the corrected state variable based on the model route diagram to obta