Comprehensive simulation and regulation model of basin water system based on multi-model fusion, coupling and parameter rolling correction

Abstract

The application discloses a kind of based on multi-model fusion, coupling and parameter rolling correction watershed water system comprehensive simulation and regulation model, comprising: multi-type model construction, multi-model coupling, model hierarchical fusion application and model parameter rolling correction, the four parts of the application are closely linked, layer by layer progressive, jointly constitute a set from physical digitization, process integration, application flexibility to model self-adapting complete, closed-loop watershed water system comprehensive simulation and regulation technical system.This system provides strong whole-process technical support for realizing fine watershed water resources management, flood control and disaster reduction intelligent scheduling and water engineering collaborative optimization.

Description

Technical Field

[0001] This invention relates to the field of hydrological and water resources engineering and watershed system simulation, and to a comprehensive simulation and regulation model for watershed water systems based on multi-model fusion, coupling and parameter rolling correction. Background Technology

[0002] With climate change and intensifying human activities, watershed water systems face multiple pressures, including floods, water shortages, and water pollution. Traditional single hydrological or hydraulic models are insufficient to comprehensively simulate the multi-process coupling and dynamic regulation of complex water systems, especially in terms of multi-scale and multi-type unit coupling. Current technologies often limit model coupling to one-way data transfer, lacking true dynamic interaction and joint solution; parameter calibration also relies heavily on historical static calibration, making it difficult to adapt to real-time changes in hydrological and hydraulic processes. Therefore, there is an urgent need to construct a comprehensive watershed water system simulation and regulation model that can integrate multiple models, achieve dynamic coupling, and possess adaptive parameter correction capabilities to improve simulation accuracy and the scientific basis of regulation decisions. Summary of the Invention

[0003] To address the aforementioned issues, embodiments of the present invention propose a comprehensive simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction.

[0004] The present invention provides a comprehensive simulation and regulation model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction, comprising:

[0005] Multiple types of models are constructed, including watershed hydrological models, regional hydrological models, river and canal hydraulic models, complex water network hydraulic models, and gate and pump unit control models;

[0006] Multi-model coupling, which includes external coupling, internal coupling and full coupling techniques, achieves seamless connection between hydrological processes, hydraulic processes and control processes through coupling, thereby improving the accuracy and practicality of simulation.

[0007] The model hierarchical fusion application includes a downscaling method based on a large-scale model and an upscaling method based on a small-scale model.

[0008] The rolling correction of model parameters is based on rolling optimization and feedback correction, forming a strict closed-loop control process.

[0009] The multi-model coupling requires the establishment of a unified technical framework, including data standardization, coupling interface design, time step coordination, feedback mechanism, verification and calibration.

[0010] The implementation process of the downscaling method based on the large-scale model includes constructing a "general-specific-general" physical framework and staggered water balance inversion. The upscaling method based on the small-scale model refers to aggregating the fine simulation results of multiple regions into a large-scale response along the topological relationship, while maintaining physical laws and engineering constraints.

[0011] The rolling correction of model parameters is based on rolling optimization and feedback correction, forming a strict closed-loop control process. Its core is to solve a dynamic parameter optimization problem that integrates historical data fitting and future prediction stability at each real-time step.

[0012] The objective function for rolling correction of model parameters aims to balance historical simulation accuracy and future prediction performance, and takes the following form:

[0013] Minimize J(θ) = α×Σ [Q_sim(ti) - Q_obs(ti)]² + β×Σ [Q_pred(t+j-1) - Q_ref(t+j-1)]² (1)

[0014] In the formula, J(θ) is the objective function, Minimize J(θ) means taking the minimum value of J(θ), θ is the vector of key parameters of the model to be corrected in real time, Q_sim(ti) is the flow at time ti obtained by simulation using parameters θ and measured rainfall R(ti), Q_obs(ti) is the actual observed flow at time ti, Q_pred(t+j-1) is the predicted flow at time t+j-1, which refers to the flow at time t+j-1 predicted forward using the corrected model state at time t as the initial condition, using parameters θ and future forecast rainfall R_f(t+j-1), j=1~P, Q_ref(t+j-1) refers to the predicted reference value at time t+j-1, α and β are the weight coefficients of historical fitting and future prediction, respectively, used to balance the weight between historical fitting accuracy and future prediction stability, α + β = 1.

[0015] The physical constraint of the parameter θ is: θ_min ≤ θ ≤ θ_max, to ensure that the optimized parameter does not deviate from its physical meaning range. θ_min and θ_max represent the lower limit and upper limit of the parameter θ, respectively.

[0016] The rate of change constraint for the predicted flow rate Q_pred is:

[0017] |Q_pred(t+1) - Q_pred(t)| ≤ ΔQ_max (2)

[0018] Wherein, ΔQ_max is the maximum allowable variation in the predicted flow rate between two consecutive time points.

[0019] The final state of parameter θ at time t is S_sim(t, θ), and the optimal state of parameter θ after feedback correction and independent estimation is... ,but:

[0020] (3)

[0021] Here, ε is the minimum allowable error, and this constraint ensures the consistency between the optimization parameters and the current optimal system state.

[0022] The prediction error formula for the rolling correction of the model parameters is:

[0023] e(t+1) = Q_obs(t+1) - Q_pred(t+1) (4)

[0024] Where e(t+1) is the difference between the observed flow and the predicted flow at time t+1, Q_obs(t+1) is the observed flow at time t+1, and Q_pred(t+1) represents the predicted flow at time t+1.

[0025] The rolling correction of model parameters employs a state update algorithm to instantly correct the model's internal state variable S(t+1) using the prediction error, thereby obtaining the corrected state estimate. The formula is:

[0026] (5)

[0027] Where S_pred(t+1) is the model based on θ * (t) from the previous state The predicted state at time t+1, K is the gain matrix. θ*(t) represents the hydrological parameters at time t.

[0028] The beneficial effects of this invention are:

[0029] 1. The four parts of this application are interconnected and progressively build upon each other, together forming a complete and closed-loop technology system for comprehensive simulation and regulation of watershed water systems, encompassing physical digitization, process integration, application flexibility, and model adaptation. This system provides strong end-to-end technical support for achieving refined watershed water resource management, intelligent flood control and disaster reduction scheduling, and collaborative optimization of water projects.

[0030] 2. Parameter correction is transformed from "post-hoc fitting" to "dynamic adjustment guided by pre-hoc prediction". By introducing an optimization objective function for future prediction accuracy, parameter correction can not only "explain" the observed errors that have occurred, but also actively "adapt" to the future changing trends of the system.

[0031] 3. In view of the high nonlinearity and time-varying nature of hydrological systems, special emphasis is placed on the differentiated design of parameter correction strategies for lumped hydrological models (such as the Xin'anjiang model) and hydrodynamic models (such as one-dimensional and two-dimensional hydrodynamic coupling models): the former focuses on the rolling optimization of empirical parameters of runoff generation and confluence, while the latter focuses on the local feedback correction of hydraulic parameters such as roughness and cross-sectional water level, thereby achieving a coordinated improvement in the simulation accuracy of the entire process from watershed hydrological response to river hydraulic evolution. Detailed Implementation

[0032] The embodiments of the present invention are described in detail below, and are intended to explain the present invention, but should not be construed as limiting the present invention.

[0033] The present invention provides a comprehensive simulation and regulation model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction, comprising:

[0034] 1. Construction of multiple model types.

[0035] To achieve a holistic characterization of complex water systems, it is first necessary to construct a standardized basic simulation unit system and establish its spatial topological relationships based on physical laws, forming the "skeleton framework" of the simulation network.

[0036] Complex water systems are divided into multi-scale watershed units, regional hydrological units such as urban areas, towns, and irrigation districts, multi-level water systems (rivers and canals) or virtual river sections hydraulic transmission channel units, and control nodes such as reservoirs (virtual reservoirs), pumping stations, sluice gates, and weirs. Based on this, the various units and nodes are combined and correlated to form different comprehensive water system simulation and control units.

[0037] For example, a multi-level water system (river and canal) hydraulic transmission channel unit plus sluice gate control node constitutes a river and canal hydraulic control system; a multi-scale watershed unit plus a multi-level water system (river and canal) hydraulic transmission channel unit constitutes a watershed hydrological and hydraulic simulation system; and an urban town hydrological simulation unit plus a multi-level water system (river and canal) hydraulic transmission channel unit plus control nodes such as reservoirs, pumping stations, sluice gates, and weirs constitutes an urban hydrological and hydraulic simulation and control system.

[0038] Watersheds, regional hydrological units, hydraulic transmission units, and control nodes are associated according to confluence topology, spatial affiliation, and location, which can be achieved using unified river, watershed, and node codes.

[0039] (1) Watersheds of different scales are associated with each other according to their spatial affiliation and spatial location; the watershed is divided into different sub-watersheds, and the sub-watersheds are associated with each other according to their confluence topology.

[0040] (2) The control nodes such as hydraulic transmission channel units, reservoirs, pumping stations, sluice gates, and weirs in multi-level water systems (rivers and canals) are associated according to the confluence topology.

[0041] (3) The control nodes such as multi-level water system (river canal) hydraulic transmission channel units, reservoirs, pumping stations, sluice gates, weirs and dams are associated with the watershed, urban area and town units according to their affiliation and spatial location.

[0042] Based on the above units, a comprehensive water system is constructed that includes watersheds, urban areas, towns, irrigation districts, multi-level water systems (rivers and canals), reservoirs, pumping stations, sluice gates, and weirs.

[0043] Based on the construction and association of simulation units and nodes of different types and scales, and based on watershed runoff generation and confluence, hydraulic equations, and control equations, we first construct independent watershed hydrological models, regional hydrological models such as urban towns and irrigation districts, river and canal hydraulic models, complex water network hydraulic models, and gate and pump station unit control models. Then, based on the combination and association of units and nodes, we construct different water system hydrological, hydraulic and control models, including urban hydrological, hydraulic and control models, watershed hydrological, hydraulic and control models, river and canal hydraulic and control models, and irrigation district hydrological, hydraulic and control models.

[0044] 2. Multi-model coupling.

[0045] Based on the construction of independent watershed hydrological models, regional hydrological models for urban areas, towns, and irrigation districts, river and canal hydraulic models, complex water network hydraulic models, and sluice gate and pumping station unit control models, these models need to be effectively coupled to simulate the dynamic behavior of the entire water system. Coupling relationships are based on data exchange and interaction between models, mainly including external coupling, internal coupling, and full coupling techniques. Through coupling, seamless integration of hydrological, hydraulic, and control processes can be achieved, improving the accuracy and practicality of the simulation.

[0046] 2.1 The dependencies and interactions between the models are as follows:

[0047] (1) Watershed hydrological model and river-canal hydraulic model: The watershed model simulates the rainfall-runoff process, and the calculated outlet flow or water level is used as the input boundary condition for the river-canal model to realize the transition from hydrological to hydraulic process.

[0048] (2) Urban hydrological model and complex water network hydraulic model: The urban model calculates surface runoff and drainage flow, which are used as inputs to the water network model to simulate urban flooding, pipeline water flow and flood propagation; the water level information returned by the water network model can be used for inundation analysis of the urban model.

[0049] (3) Irrigation district hydrological model and river and canal hydraulic model: The irrigation district model simulates irrigation water use, return flow and drainage process, and its output serves as the lateral inflow or outflow condition of the river and canal model, affecting the water flow state of the canal.

[0050] (4) Gate and pump station unit control model and river and canal hydraulic model, complex water network hydraulic model: The gate and pump station model adjusts the gate opening or pump station operation according to real-time water level or flow data. Its operation status serves as the internal control point or boundary condition of the hydraulic model to realize water flow scheduling.

[0051] (5) The integrated model (such as the urban hydrology and hydraulics and regulation model, the watershed hydrology and hydraulics and regulation model) integrates multiple sub-models, but the internal sub-models still need to achieve data sharing and collaborative calculation through coupling technology.

[0052] 2.2 Based on the degree of interaction and data sharing methods between models, the following three coupling methods are adopted:

[0053] (1) External coupling

[0054] This approach is suitable for situations where the input-output relationship between models is clear and data is transferred in a one-way manner. The output of one model serves as the input of another model, and the models operate independently.

[0055] Example 1: Watershed hydrological model and river-canal hydraulic model

[0056] The flow process line or water level process line calculated by the watershed model is used as the boundary condition of the river and canal model. The river and canal model simulates water flow, but does not feed back data to the watershed model.

[0057] Example 2: Urban hydrological model and complex water network hydraulic model

[0058] The urban model provides surface runoff and drainage flow as input node flows for the water network model. The water network model calculates the water level of the pipe network or river and outputs it for urban flood assessment.

[0059] (2) Internal coupling

[0060] This approach is suitable for situations where models share parameters, boundary conditions, or internal data, but maintain independent solution processes. Models exchange information in real time through a shared data interface, but the solution process remains separate.

[0061] The irrigation district hydrological model and the river-canal hydraulic model share data such as canal geometric parameters, soil properties, and irrigation plans. The irrigation district model calculates irrigation water demand and regression flow, while the river-canal model simulates canal flow based on these data and updates water level information through a shared interface, thus influencing the drainage decisions of the irrigation district model.

[0062] The control model for the gate pumping station and the hydraulic model for the river canal share real-time water level and flow data. The gate pumping station model adjusts the gate opening according to the water level threshold, and the river canal model calculates the water flow change according to the opening. However, the two are solved independently, and dynamic interaction is achieved through data synchronization.

[0063] (3) Fully coupled technology

[0064] This method is suitable for situations where models are tightly integrated and require joint solution of governing equations. It combines the equations of multiple models into a single system and solves for all variables simultaneously.

[0065] For complex water network hydraulic models and gate pump station control models, the basic equations of the water network (such as the Saint-Venant equation) and the control equations of the gate pump station (such as the gate flow equation and pump station performance curve) are combined to construct a complete set of equations, and the water flow state and gate pump operation are solved as a whole to achieve real-time control.

[0066] For watershed hydrological, hydraulic and control models, the watershed runoff generation and confluence equations (such as the Xin'anjiang model) are integrated with the river and canal hydraulic equations to solve the hydrological and hydraulic processes simultaneously, avoiding data transmission errors and improving simulation efficiency.

[0067] 2.3 To ensure the smooth implementation of coupling, a unified technical framework needs to be established:

[0068] (1) Data standardization: Define common data formats, unit systems and coordinate systems to ensure data compatibility between models, such as using NetCDF or HDF5 formats to store and exchange data.

[0069] (2) Coupling interface design: Develop application programming interfaces (APIs) or middleware (such as OpenMI, FMI standards) to handle data transfer, time step synchronization and error handling between models.

[0070] (3) Time step coordination: For dynamic simulation, an adaptive time step or a fixed time step strategy is adopted according to the numerical stability requirements of each model to achieve synchronous calculation between models.

[0071] (4) Feedback mechanism: In internal coupling and full coupling, a feedback loop is established so that the model can respond to each other's changes in real time, such as water level changes triggering gate pump control adjustments.

[0072] (5) Verification and calibration: Verify the accuracy of the coupled model using historical or experimental data, and calibrate the coupling parameters to ensure the reliability of the simulation results.

[0073] Through the above coupling relationships and methods, an integrated water system simulation platform can be constructed to realize the comprehensive management and control of multiple types of water systems, providing technical support for water resource planning, flood control and disaster reduction, and engineering design.

[0074] 3. Model hierarchical fusion application.

[0075] Hydrological models, hydraulic models, and regulation models of different scales vary in their modeling scope and data precision, resulting in different scales and application ranges of their results. For example, large-scale hydrological models often require large-scale data to obtain hydrological processes at the basin outlet, without focusing on the hydrological processes of sub-basins or river sections within the basin. Small-scale hydrological models, on the other hand, often require more refined data to obtain hydrological process results for sub-basins or local areas. Therefore, they are suitable for different application scenarios.

[0076] The hierarchical application of hydrological and hydraulic models at different scales is introduced, taking hydrological forecasting models as an example.

[0077] (1) Application of distributed models for small watershed units (for flash floods)

[0078] To address the need for flash flood forecasting in small watersheds, and considering the heterogeneity of underlying surface conditions, a grid-distributed model is constructed for each sub-watershed. Independent simulations are performed for each small watershed, focusing on the outflow flood process of the small watershed, and the model for each watershed is calibrated.

[0079] (2) Application of distributed models for larger watershed outlets

[0080] For a large watershed, considering the numerous influencing factors within the watershed and ignoring other internal nodes due to underlying surface factors, the number of sub-watersheds is simplified, an overall grid-distributed model of the watershed is constructed, the flood process at the outlet of the large watershed is calculated, and the entire watershed model is calibrated.

[0081] 3.1 Downscaling methods based on large-scale models

[0082] Traditional downscaling methods primarily focus on scale transformation of data or driving forces. For example, they transform the output of large-scale climate or hydrological models into hydrological and meteorological inputs with smaller spatial resolutions (such as sub-basins or grids) through statistical relationships or dynamic methods, for use in local, refined models. The core of these methods lies in establishing mathematical relationships between variables at different scales, but they have limitations in terms of the clarity of physical mechanisms and the utilization of existing observation facilities within the watershed.

[0083] This application proposes an innovative downscaling method based on a framework that integrates physical structure and observational constraints, namely a watershed structure downscaling method under observational constraints. It is both a unique upscaling path and a physical structure downscaling strategy.

[0084] The core of this method lies in abandoning the traditional approach of "a single model covering the entire region" or "mechanical geographical division," and instead making full use of existing hydrological monitoring stations within the watershed that have different functions (such as small reservoir stations that can monitor rainfall and runoff, and downstream control hydrological stations). Based on the actual control range of these stations, the large watershed is naturally decomposed into a series of sub-computational units directly constrained by the stations (i.e., source small reservoir control areas, watersheds between stations, etc.). In other words, by utilizing the confluence topology and observational or model information from some sub-watersheds, the output of the large-scale model is decomposed into more refined spatial units.

[0085] Implementation process:

[0086] 1) Construct a "general-specific-general" physical framework:

[0087] "General": A well-calibrated large-scale watershed hydrological model (lumped or distributed) that can reliably simulate flood processes at the watershed outlet section (the watershed outlet control hydrological station). .

[0088] "Partially Known": Within the watershed, some sub-watersheds (such as areas controlled by small reservoirs A and B) have established reliable sub-watershed models or have directly observable outflow processes. , .

[0089] "Division" (Unknown Target): The target is to solve for the independent flow generation and merging processes in the remaining unknown intervals (such as AD, BD, and the intervals between AB) through physical relationships. .

[0090] 2) Time-sharing water balance inversion

[0091] Upstream contribution stripping: This involves separating known upstream sub-basin outflow processes (such as...) The contribution was calculated by using a river confluence model (such as the Muskingan method) to downstream key sections (such as station D). .

[0092] Interval process solution: The overall process simulated by the large-scale model at the entire watershed outlet control station. By subtracting the calculated contributions from all known upstream sources, the flood process resulting from the combined contribution of all unknown intervals can be derived. If the watershed structure is simple, the water balance at intermediate nodes can be used to calculate more specific interval processes step by step.

[0093]

[0094] In the formula, This represents a flood process with unknown joint contributions from different regions. This indicates a flood event in the basin that has been monitored at or above the existing monitoring points.

[0095] Interval model construction and validation: using the inversion obtained The corresponding spatial precipitation data can be used to calibrate hydrological models representing these unknown regions. These models maintain physical consistency with the results of large watershed models at the outlet section.

[0096] Downscaling methods based on large-scale models fully utilize observational data from existing internal stations (small reservoirs, hydrological stations) as constraints, ensuring physical consistency and strictly adhering to watershed topology and water balance. This solves the "black box" problem of previous watershed downscaling simulations, decomposing the "black box" output of large watershed models into physically meaningful sub-regional processes, which is particularly helpful in understanding the sources of inter-regional floods and assessing local flood control risks.

[0097] 3.2 Upscaling Method Based on Small-Scale Model

[0098] Upscaling from small-scale models requires aggregating detailed simulation results from multiple regions into large-scale responses along topological relationships, while preserving physical laws and engineering constraints. Spatially, based on the directed topology of river / pipeline networks, outflows from sub-basins, river segments, or urban zones are superimposed level by level according to confluence relationships, simultaneously considering the effects of node diversion, flood diversion, and pumping. For key structures such as gates, weirs, and pumping stations, real-time operating conditions or rules are embedded to ensure that flow capacity and water level control meet engineering feasibility.

[0099] In terms of timing, the output time series of each small scale are first aligned, and the rationality of the data is verified. The large-scale response after upscaling not only serves the forecast of the watershed outlet, but also provides a reliable set of parameters and boundary conditions for the upper-level model through parameter regionalization and state representativeness transfer, supporting joint regulation and control decisions at the watershed level.

[0100] 4. Rolling correction of model parameters.

[0101] This module aims to construct a closed-loop, adaptive real-time parameter correction framework for hydrological-hydraulic models. The core idea is to deeply integrate the "rolling time-domain optimization" and "feedback correction" mechanisms of predictive control into the hydrological simulation process. Traditional parameter calibration is an "open-loop" static optimization, while this application transforms it into a "closed-loop" dynamic process: at each real-time simulation step, the system not only dynamically estimates key model parameters based on the current and past n time periods' measured rainfall-runoff sequences, but also focuses on using this information to make advance predictions of hydrological conditions (such as flow and water level) for one or more future time periods. Its innovation lies in transforming parameter correction from "post-hoc fitting" to "dynamic adjustment guided by pre-hoc prediction." By introducing an optimization objective function oriented towards future prediction accuracy, parameter correction can not only "explain" existing observation errors but also proactively "adapt" to future system changes. In response to the high nonlinearity and time-varying nature of hydrological systems, this module particularly emphasizes the differentiated design of parameter correction strategies for lumped hydrological models (such as the Xin'anjiang model) and hydrodynamic models (such as one-dimensional and two-dimensional hydrodynamic coupling models): the former focuses on the rolling optimization of empirical parameters for runoff generation and confluence, while the latter focuses on the local feedback correction of hydraulic parameters such as roughness and cross-sectional water level, thereby achieving a synergistic improvement in the simulation accuracy of the entire process from watershed hydrological response to river hydraulic evolution.

[0102] The implementation of rolling correction of model parameters is based on rolling optimization and feedback correction, forming a strict closed-loop control process. Its core is to solve a dynamic parameter optimization problem that integrates historical data fitting and future prediction stability at each real-time step.

[0103] Implementation steps:

[0104] S401. System Initialization and Parameter Definition

[0105] 1) Model construction: Complete the coupling and integration of the basin hydrological model (such as the Xin'anjiang model) and the downstream channel one- and two-dimensional hydrodynamic model (such as the basic model of the river and canal hydraulic model and the gate and pump station unit control model) to ensure the continuous transmission of the water flow process in the generation, confluence and evolution stages.

[0106] 2) Variable definition:

[0107] Control variables (decision variables): defined as the vector of key model parameters θ to be corrected in real time. For example, θ = This may include the water storage capacity WM and evaporation coefficient C of the hydrological model, as well as the comprehensive roughness coefficient n of the hydrodynamic model.

[0108] State variables: defined as the internal storage set S of the model, such as the average soil moisture content of the watershed, the water storage of the river unit, etc.

[0109] Input and output sequences: The real-time rainfall input sequence is set as R(1, t), and the measured flow (or water level) sequence of the key section is set as Q_obs(1, t).

[0110] 3) Window settings: Determine the length N of the rolling optimization window (e.g., 6 hours) and the prediction step size P (usually set to 1 step, i.e., predicting the next time period).

[0111] S402. Optimization of Rolling Time Domain Parameters (Prediction Compensation Stage)

[0112] At each current time t, initiate a constrained optimization calculation cycle:

[0113] 1) Constructing the objective function (predictive control core): The objective function J(θ) aims to balance historical fitting accuracy with future prediction performance, and its general form is as follows:

[0114] Minimize J(θ) = α×Σ [Q_sim(ti) - Q_obs(ti)]² + β×Σ [Q_pred(t+j-1) - Q_ref(t+j-1)]² (1)

[0115] In the formula, J(θ) is the objective function, Minimize J(θ) means that J(θ) takes the minimum value, and θ is the vector of key parameters of the model to be corrected in real time;

[0116] Q_sim(ti) is the flow rate at time ti, which is simulated using parameter θ and measured rainfall R(ti); Q_obs(ti) is the actual observed flow rate at time ti. The difference between the two represents the difference in historical fitting, which is used to measure the simulation error of the model for the past N steps of observations under parameter θ, i = 1~N.

[0117] Q_pred(t+j-1) is the predicted flow rate at time t+j-1, which refers to the flow rate at time t+j-1 predicted forward using the current (time t) corrected model state as the initial condition, the parameter θ, and the future forecast rainfall R_f(t+j-1), where j=1~P; Q_ref(t+j-1) refers to the predicted reference value at time t+j-1 (such as the rainfall at time t+j-1 based on the rainfall prediction); the difference between the two represents the difference in future predictions, which is used to control the direction of parameter optimization so that it can produce physically reasonable and trend-stable future predictions.

[0118] α and β are the weighting coefficients for historical fitting and future prediction, respectively, used to balance the weight between historical fitting accuracy and future prediction stability, α + β = 1. By adjusting the ratio of the two, emphasis can be placed on parameter tracking ability (larger α) or prediction robustness (larger β).

[0119] S403. Constraints

[0120] 1) Physical constraints on parameter θ: θ_ min ≤ θ ≤ θ_ max To ensure that the optimized parameters do not deviate from their physical meaning, θ_ min θ_ max These represent the lower and upper limits of the parameter θ, respectively.

[0121] 2) Model state consistency constraint: The optimization process must ensure that the internal state of the model at time t (such as soil moisture content, river storage) is smoothly connected with the state after the previous feedback correction.

[0122] 3) Output smoothness constraint: The rate of change constraint can be applied to the predicted flow rate Q_pred to avoid unreasonable oscillations.

[0123] |Q_pred(t+1) - Q_pred(t)| ≤ ΔQ_max (2)

[0124] Wherein, ΔQ_max is the maximum allowable variation in the predicted flow rate between two consecutive time points.

[0125] 4) State continuity constraint: Ensure that the model corresponding to parameter θ, whose simulated final state S_sim(t,θ) at time t, should be consistent with the independently estimated optimal state after feedback correction. Get as close as possible.

[0126] (3)

[0127] Here, ε is the minimum allowable error, and this constraint ensures the consistency between the optimization parameters and the current optimal system state.

[0128] S404. Optimization Solution

[0129] The constrained nonlinear optimization problem described above is solved using efficient optimization algorithms (such as sequential quadratic programming (SQP) or differential evolution algorithm (DE) with global search capabilities). The optimal parameter estimate θ(t) for the current time t is finally obtained.

[0130] S405. Feedback Correction and Rolling Progress

[0131] 1) Calculate the error and update the status:

[0132] This error is not only used for performance evaluation, but more importantly, it drives the feedback correction of the model state. The formula for calculating the prediction error is as follows:

[0133] e(t+1) = Q_obs(t+1) - Q_pred(t+1) (4)

[0134] Where e(t+1) is the difference between the observed flow and the predicted flow at time t+1, Q_obs(t+1) is the observed flow at time t+1, and Q_pred(t+1) represents the predicted flow at time t+1.

[0135] A state update algorithm (such as ensemble Kalman filter EnKF or direct correction method) is used to instantly correct the model's internal state variable S(t+1) using the prediction error e(t+1), thus obtaining the corrected state estimate. The formula is:

[0136] (5)

[0137] Where S_pred(t+1) is the model based on θ * (t) from the previous state The predicted state at time t+1, K is the gain matrix. θ*(t) represents the hydrological parameters at time t, which are general parameter estimates.

[0138] 2) Scrolling time window:

[0139] Update the current time t to t+1. Add the latest data pair {R(t+1), Q_obs(t+1)} to the rolling window, while removing the oldest data pair, keeping the window length constant at N.

[0140] 3) Iterative loop:

[0141] Update the model state The data from the scrolling window is brought back to "1) Calculate the error and update the status", and the "optimization-prediction-feedback" loop at time (t+1) is started.

[0142] Through continuous iteration of the above five steps, the model parameters θ and internal state S are continuously and dynamically adjusted in coordination according to the latest observation information, thus forming a complete adaptive real-time simulation system with feedforward (prediction compensation) and feedback (error correction) capabilities.

[0143] Although the above embodiments have been shown and described, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Any changes, modifications, substitutions and variations made to the above embodiments by those skilled in the art are within the protection scope of the present invention.

Claims

1. A comprehensive simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction, characterized in that, include: Multiple types of models are constructed, including watershed hydrological models, regional hydrological models, river and canal hydraulic models, complex water network hydraulic models, and gate and pump unit control models; Multi-model coupling, which includes external coupling, internal coupling and full coupling techniques, achieves seamless connection between hydrological processes, hydraulic processes and control processes through coupling, thereby improving the accuracy and practicality of simulation. The model hierarchical fusion application includes a downscaling method based on a large-scale model and an upscaling method based on a small-scale model. The rolling correction of model parameters is based on rolling optimization and feedback correction, forming a strict closed-loop control process.

2. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 1, is characterized in that, The multi-model coupling requires the establishment of a unified technical framework, including data standardization, coupling interface design, time step coordination, feedback mechanism, verification and calibration.

3. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 1, is characterized in that, The implementation process of the downscaling method based on the large-scale model includes constructing a "general-specific-general" physical framework and staggered water balance inversion. The upscaling method based on the small-scale model refers to aggregating the fine simulation results of multiple regions into a large-scale response along the topological relationship, while maintaining physical laws and engineering constraints.

4. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 1, is characterized in that, The rolling correction of model parameters is based on rolling optimization and feedback correction, forming a strict closed-loop control process. Its core is to solve a dynamic parameter optimization problem that integrates historical data fitting and future prediction stability at each real-time step.

5. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 4, is characterized in that, The objective function for rolling correction of model parameters aims to balance historical simulation accuracy and future prediction performance, and takes the following form: Minimize J(θ) = α×Σ [Q_sim(ti) - Q_obs(ti)]² + β×Σ [Q_pred(t+j-1) -Q_ref(t+j-1)]² (1) In the formula, J(θ) is the objective function, Minimize J(θ) means taking the minimum value of J(θ), θ is the vector of key parameters of the model to be corrected in real time, Q_sim(ti) is the flow rate at time ti obtained by simulation using parameters θ and measured rainfall R(ti), Q_obs(ti) is the actual observed flow rate at time ti, Q_pred(t+j-1) is the predicted flow rate at time t+j-1, j=1~P, Q_ref(t+j-1) refers to the predicted reference value at time t+j-1, α and β are the weight coefficients of historical fitting and future prediction, respectively, used to balance the weight between historical fitting accuracy and future prediction stability, α + β = 1.

6. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 5, is characterized in that, The physical constraint of the parameter θ is: θ_min ≤ θ ≤ θ_max, to ensure that the optimized parameter does not deviate from its physical meaning range. θ_min and θ_max represent the lower limit and upper limit of the parameter θ, respectively.

7. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 5, is characterized in that, The rate of change constraint for the predicted flow rate Q_pred is: |Q_pred(t+1) - Q_pred(t)| ≤ ΔQ_max (2) Wherein, ΔQ_max is the maximum allowable variation in the predicted flow rate between two consecutive time points.

8. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 5, is characterized in that, The final state of parameter θ at time t is S_sim(t, θ), and the optimal state of parameter θ after feedback correction and independent estimation is... ,but: （3） Here, ε is the minimum allowable error, and this constraint ensures the consistency between the optimization parameters and the current optimal system state.

9. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 8, is characterized in that, The prediction error formula for the rolling correction of the model parameters is: e(t+1) = Q_obs(t+1) - Q_pred(t+1) (4) Where e(t+1) is the difference between the observed flow and the predicted flow at time t+1, Q_obs(t+1) is the observed flow at time t+1, and Q_pred(t+1) represents the predicted flow at time t+1.

10. The integrated simulation and control model for watershed water systems based on multi-model fusion, coupling, and parameter rolling correction as described in claim 9, is characterized in that, The rolling correction of model parameters employs a state update algorithm to instantly correct the model's internal state variable S(t+1) using the prediction error, thereby obtaining the corrected state estimate. The formula is: （5） Where S_pred(t+1) is the model based on θ * (t) from the previous state The predicted state at time t+1, K is the gain matrix. θ*(t) represents the hydrological parameters at time t.

Images