Intelligent flood routing method for river-lake regulation system based on dynamic storage-discharge relationship

Abstract

The application discloses a kind of river and lake regulation system flood evolution intelligent prediction method based on dynamic storage and discharge relationship, comprising: constructing the space-time dynamic correlation graph structure reflecting the hydraulic connection of basin, mapping multiple-source hydrological data to generate the model input set of unified space-time scale;Establish an aggregated reservoir parameterization structure model, including the dynamic water level-storage relationship driven by inflow variation characteristics, and the composite water level-flow relationship integrating the time lag effect and backwater jacking feedback;Construct flood regulation equation set, according to the real-time time lag state in the composite relationship, switch between different algorithm channels to solve the equation set using adaptive solving mechanism;Use real-time observation data to correct model state and perform rolling prediction.The application effectively solves the problems of difficult to describe the dynamic characteristics of storage and discharge relationship in complex river and lake system, unstable solution of non-same phase evolution process and difficult to determine high-dimensional parameters, and improves the continuity and physical consistency of flood prediction.

Description

Technical Field

[0001] This invention relates to the field of flood forecasting, and in particular to an intelligent forecasting method for flood evolution in river and lake regulation and storage systems based on dynamic storage and discharge relationships. Background Technology

[0002] With the deepening development of water resources in large river basins, the main river channels and connected lakes constitute a complex river-lake joint regulation and storage system. Its flood evolution process is subject to multiple constraints, including river channel storage, lake diversion, and downstream backwater, exhibiting highly nonlinear and time-varying characteristics. Accurately analyzing the water exchange and propagation patterns of this system under dynamic boundaries and constructing a high-fidelity flood evolution model is of significant engineering and technical value for improving the scientific rigor and timeliness of river basin flood control and dispatch decisions.

[0003] Existing techniques typically employ the Saint-Venant equations to construct one-dimensional or two-dimensional hydrodynamic models, or utilize hydrological methods such as the Muskingen model for generalized simulations. These methods generally treat rivers and lakes as static units with fixed geometric boundaries, pre-determining fixed water level-storage capacity curves based on topographic data to describe storage capacity, and often rely on single-valued processing or empirical loop curves to characterize the water level-discharge relationship at cross-sections. At the computational solution level, finite difference methods or fixed iterative algorithms are commonly used to perform time-step derivation of the discretized governing equations.

[0004] However, when faced with scenarios involving dramatic flood fluctuations or complex hydraulic connections, the aforementioned conventional solutions suffer from rigid representations of storage and discharge structures and insufficient adaptability of the solution mechanism. Specifically, firstly, traditional models neglect the real-time impact of flood wave dynamics on storage capacity, and the fixed water level-storage capacity function cannot reflect the adjustment of water surface gradient caused by sudden changes in inflow, leading to systematic biases in dynamic storage capacity calculations. Secondly, the water level and flow responses at control sections often exhibit significant out-of-phase (lag) characteristics. Due to the coupling effect of backwater backflow and propagation lag, traditional single solution algorithms struggle to ensure stable convergence of the water balance equation within a large lag interval, easily resulting in computational oscillations or divergences. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent forecasting method for flood evolution in river and lake regulation and storage systems based on dynamic storage and discharge relationships, in order to solve at least one of the aforementioned problems in the existing technology.

[0006] According to one aspect of this application, a method for intelligent forecasting of flood evolution in river and lake regulation systems based on dynamic storage and discharge relationships includes:

[0007] A spatiotemporal dynamic correlation graph structure reflecting the hydraulic connections of the watershed is constructed. The collected multi-source hydrological data is mapped to the spatiotemporal dynamic correlation graph structure. Through node feature aggregation and edge weight update, a model input dataset with a unified spatiotemporal scale is generated.

[0008] Based on the inflow change features extracted from the model input dataset, a parameterized structural model of the aggregated reservoir is established to characterize the overall storage and discharge features of the main stream and the lake. It includes a dynamic water level-storage capacity relationship driven by the inflow change features in the model input dataset, as well as a composite water level-discharge relationship integrating the lag effect and backwater backwater feedback.

[0009] A flood control calculation equation set is constructed based on the parameterized structural model of the aggregated reservoir. Based on the real-time lag state identified from the composite water level-discharge relationship, an adaptive solution mechanism is used to solve the flood control calculation equation set to obtain the calculation state variables.

[0010] Based on real-time observation data, the parameterized structural model of the aggregated reservoir is corrected for its state. The corrected model and calculated state variables are used to perform rolling predictions and output water level and flow process information for future periods.

[0011] Beneficial effects: This invention effectively solves the problems of difficulty in characterizing the dynamic characteristics of storage and discharge relationships in complex river and lake systems, instability in solving non-phase evolution processes, and difficulty in calibrating high-dimensional parameters, thereby improving the continuity and physical consistency of flood forecasting. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall process of the intelligent flood evolution forecasting method for river and lake regulation and storage systems based on dynamic storage and discharge relationships, according to an embodiment of this application.

[0013] Figure 2 This is a schematic diagram illustrating the process of constructing a spatiotemporal dynamic correlation graph structure that reflects the hydraulic connections of a watershed and generating a model input dataset with a unified spatiotemporal scale, according to an embodiment of this application.

[0014] Figure 3 This is a schematic diagram of the process for establishing a parameterized structural model of a reservoir that characterizes the overall storage and discharge characteristics of the main stream and the lake, according to an embodiment of this application.

[0015] Figure 4 This is a schematic diagram of the process for constructing the time lag correction model according to an embodiment of this application. Detailed Implementation

[0016] Example 1 details the overall framework of an intelligent flood evolution forecasting method for river and lake regulation systems based on dynamic storage-discharge relationships. It provides an intelligent forecasting scheme adaptable to the nonlinear and non-synchronous storage-discharge response characteristics of complex river and lake systems, such as... Figure 1 As shown.

[0017] Step 101: Construct a spatiotemporal dynamic correlation graph structure that reflects the hydraulic connections of the watershed, and map the collected multi-source hydrological data into the spatiotemporal dynamic correlation graph structure. Generate a model input dataset with a unified spatiotemporal scale through node feature aggregation and edge weight update.

[0018] In this embodiment, the selection of data sources is crucial for constructing accurate hydraulic connections. The study area encompasses a complex river and lake system, such as the section of the Yangtze River from the Three Gorges to Luoshan, which involves the confluence of multiple tributaries and the regulation function of large lakes (such as Dongting Lake). To comprehensively capture hydrological processes, the system first collects multi-source hydrological data, specifically including measured water level and flow data from ground hydrological stations, rainfall data retrieved from meteorological radar, water area data monitored by remote sensing satellites, and future rainfall forecast data output from numerical weather prediction models.

[0019] Furthermore, the process of constructing the spatiotemporal dynamic relational graph structure involves the digital mapping of geospatial information and hydraulic connectivity. The system generalizes hydrological stations, key control sections, and lake inlets and outlets on rivers into nodes in the graph structure. The connections between nodes are not based solely on simple Euclidean distance, but rather on the actual physical connectivity of the river channels. For example, there is a direct river channel connection between the upstream Three Gorges Reservoir outlet node and the downstream Zhicheng node, thus establishing a directed edge.

[0020] Based on this, multi-source hydrological data are mapped to corresponding nodes. Since data from different sources often differ in temporal resolution and spatial scale—for example, ground station data may be hourly, while numerical forecasts may be 6-hourly—it is necessary to unify the spatiotemporal scale.

[0021] Specifically, methods such as interpolation, resampling, or aggregation are used to align all data to a preset standard spatiotemporal scale, such as a 6-hour time step. Node feature aggregation fuses information about the node itself and its neighboring nodes, and a dynamic update mechanism for edge weights reflects the changes in the intensity of influence between different nodes during flood propagation. The final model input dataset is a collection of feature tensors for all nodes in the entire watershed within a unified spatiotemporal framework, providing a standardized input foundation for subsequent structural modeling.

[0022] In some alternative implementations, the computing device performing this step can be a server configured with a high-performance graphics processing unit (GPU) to support the training and inference of large-scale graph neural network models. If some data sources (such as radar rainfall) are unavailable, the system can be configured to construct a simplified static graph structure using only surface hydrological station data as a degraded operating mode.

[0023] Step 102: Extract inflow variation features based on the model input dataset and establish a parameterized structural model of the aggregated reservoir that characterizes the overall storage and discharge features of the main stream and the lake. The parameterized structural model of the aggregated reservoir includes a dynamic water level-storage capacity relationship driven by the inflow variation features in the model input dataset, as well as a composite water level-discharge relationship that integrates the lag effect and backwater backwater feedback.

[0024] Specifically, this step addresses the problem that traditional hydrological models struggle to describe the dynamic storage and retention characteristics of river-lake systems. In complex river-lake systems, frequent water exchange occurs between the main river channel and connected lakes, significantly influenced by water level fluctuations. To avoid the time-consuming and detailed two-dimensional hydrodynamic simulation of complex river branching and lake flow fields, this embodiment employs the concept of a convergent reservoir. A convergent reservoir refers to a system that generalizes closely hydraulically connected main river sections and lake areas into a single water storage unit. For example, the main river channel downstream of the Three Gorges Dam to the Luoshan section and the Dongting Lake area are considered as a unified virtual reservoir.

[0025] The dynamic water level-storage capacity relationship is a crucial component of this model. Traditional reservoir capacity curves are typically static, meaning a single water level strictly corresponds to a single reservoir capacity value. However, during floods, the actual water storage at the same water level changes due to the magnitude and rate of inflow, exhibiting a looped curve characteristic. This embodiment introduces inflow variation characteristics (such as the magnitude and volatility of inflow) as driving variables, allowing the water level-storage capacity function to dynamically deform over time. For example, during the rapid rise of the flood, the surge in inflow drives the reservoir capacity curve to shift towards higher storage levels, reflecting the increase in reservoir capacity caused by the additional gradient of the river channel and the backwater effect of the lake.

[0026] Correspondingly, the composite level-discharge relationship is used to describe the discharge capacity of the outlet section of the reservoir. In actual flood evolution, the outlet discharge depends not only on the current water level but also on the significant influence of flood propagation lag (delay effect) and downstream water level backwater feedback. This composite relationship model is not merely a simple level-discharge lookup table but a parameterized equation integrating multiple physical mechanisms. It can automatically adjust the delay parameter based on the current inflow gradient and introduce a negative feedback term based on the rate of change of the downstream water level, accurately simulating the non-phase level-discharge response process.

[0027] Step 103: Construct a flood control calculation equation set based on the parameterized structural model of the aggregated reservoir, and solve the flood control calculation equation set using an adaptive solution mechanism based on the real-time lag state identified from the composite water level-discharge relationship to obtain the calculation state variables.

[0028] In this step, the flood control calculation equations are typically composed of a water balance equation and a storage and discharge equation. The water balance equation describes the conservation relationship between the inflow, outflow, and storage capacity change of the reservoir per unit time, i.e., the inflow minus the outflow equals the increase in storage capacity. The storage and discharge equation is provided by the parameterized structural model of the reservoir established in step 102, specifically embodying the dynamic water level-storage capacity relationship and the composite water level-discharge relationship.

[0029] The adaptive solution mechanism addresses the convergence and stability issues of the equation system under different hydraulic conditions. During flood evolution, the system exhibits different response states. For example, during the steady period, the responses of water level and flow are relatively synchronous; however, during the flood peak or when severely affected by backwater, the responses of water level and flow show significant lag or anomalies, the so-called real-time lag state. If a single numerical solution method is always used, it may lead to computational divergence or oscillation when the anomalies are severe. Therefore, the system introduces a discrimination logic to monitor the lag state in real time and switch between different solution algorithms. When the response is determined to be synchronous, a more efficient synchronous iterative algorithm can be used; when the response is determined to be significantly lagging, a hybrid algorithm including a prediction and correction stage is switched to ensure that the equation system can be solved stably and output the calculated state variables such as water level, flow, and reservoir capacity for the next moment.

[0030] Step 104: Based on real-time observation data, the parameterized structural model of the aggregated reservoir is corrected for its state, and the corrected model and calculated state variables are used to perform rolling predictions to output water level and flow process information for future periods.

[0031] This step enables real-time updates and rolling releases of forecasts. As time progresses, new real-time observation data (such as the latest measured water levels and flow rates) continuously enters the system. The model compares the measured data with the simulated values ​​from the previous moment to calculate the error. Based on the error feedback mechanism, the system corrects the state variables (such as current water storage and current lag factor values) in the aggregated reservoir parameterized structural model, and even fine-tunes some sensitive parameters to make the model state as close as possible to the actual watershed hydrological state.

[0032] Based on this, starting from the corrected state and combining future meteorological and rainfall forecast data as input, the system uses the aforementioned flood control calculation framework to extrapolate into the future. For example, starting from the current time t, it progressively calculates the water level and flow rate at times t+1, t+2, and up to t+K (K being the time step). This extrapolation is rolling; that is, every time a time step, such as 6 hours, the forecast window slides forward one step, always providing a forecast sequence for a future period (such as the next 5 days). The final output information not only includes specific numerical values ​​but can also include key flood control indicators such as the peak arrival time and the highest water level, providing support for flood control scheduling decisions.

[0033] Example 2 details the specific process of constructing a spatiotemporal dynamic graph structure, its dynamic evolution, and feature reconstruction. It focuses on addressing the uneven spatiotemporal distribution and data quality issues of multi-source hydrological data, and utilizes graph embedding technology to enhance and repair the data. Figure 2 As shown.

[0034] Step 201: Time synchronization and spatial registration are performed on the collected multi-source hydrological data to construct a multimodal hydrological time series matrix, and an initial graph structure is established based on the physical connectivity of the river channels.

[0035] In this embodiment, multi-source hydrological data often exhibit heterogeneity. For example, surface hydrological stations provide single-point time-series data, meteorological radar provides gridded spatial data, while remote sensing imagery provides discontinuous isometric data. Time synchronization refers to resampling all data onto a unified time axis, such as unifying all data into a time series with 6-hour intervals. Spatial mapping is achieved through coordinate transformation, such as uniformly converting to the WGS84 geographic coordinate system or the Lambert projection coordinate system, mapping radar grid data and remote sensing imagery data to the geographical locations of the corresponding hydrological stations or river cross-sections.

[0036] The process of constructing the multimodal hydrological time series matrix M(t) is essentially the concatenation of the aligned data described above. For each node v, its eigenvector at time t contains not only its own water level and flow rate observations, but also information such as rainfall and water surface area at that location. These eigenvectors are arranged in chronological order to form the multimodal time series matrix.

[0037] The initial graph structure G0 is established based on the physical topology of the river channel. The node set V contains all key hydrological control points within the study area. For any two nodes v and u, an edge is established if there is a direct river channel connection between them. The edge weight w... vu The initial value calculation can take into account the correlation between river distance and historical flow.

[0038] Preferably, the formula for calculating the initial edge weights can be expressed as:

[0039] w vu =exp(-d vu / σ d )*(1+ρ vu ) / 2;

[0040] Among them, w vu d represents the initial weight between nodes v and u. vu σ represents the actual river channel distance between two nodes, in kilometers (km). d ρ is the distance decay coefficient, used to control the degree of influence of distance on weight. In this embodiment, this coefficient is preferably set to 80km. As distance increases, the influence between nodes decays exponentially. vu Let be the Pearson correlation coefficient between the historical flow sequences of nodes v and u, with a value ranging from -1 to 1. This formula suggests that nodes that are closer together and whose flow changes are more synchronized have a stronger hydraulic connection and should therefore have a larger initial weight.

[0041] Step 202: The initial graph structure is driven to undergo time-varying evolution using a multimodal hydrological time series matrix. The edge weights are updated by calculating the dynamic similarity between nodes to obtain a spatiotemporal dynamic association graph structure.

[0042] Although the physical river channel is fixed, the intensity of the water flow's influence varies with the flood process. For example, during the dry season, the upstream's influence on the downstream may be weaker and have a longer lag; while during the flood season, the wave velocity increases, and the connection between upstream and downstream is significantly strengthened. This step introduces a dynamic evolution mechanism. The system uses the node feature vectors at each time point, i.e., slices in the multimodal hydrological time series matrix, to calculate the feature similarity between nodes in real time.

[0043] Specifically, the dynamic edge weight w at time t vu (t) can be updated through an attention mechanism, and its calculation formula can be:

[0044] w vu (t)=Softmax u (α*sim(x v (t),x u (t)));

[0045] Where, x v (t) and x u (t) represents the feature vectors of nodes v and u at time t, respectively; sim represents the similarity function, such as cosine similarity, used to measure the directional proximity of two feature vectors; α is the self-attention coefficient, a learnable parameter used to adjust the sensitivity of the attention mechanism; Softmax... u This means that the weights of all neighboring nodes u connected to node v are normalized to ensure that the sum of the weights of all outgoing edges is 1.

[0046] Through the above mechanism, the graph structure can automatically adjust the information transmission path and intensity according to the current water situation, generating a spatiotemporally dynamic relational graph structure G. t .

[0047] Step 203: Extract the node embedding features of the spatiotemporal dynamic association graph structure, perform node missing completion, anomaly detection and residual feedback correction to obtain an optimized graph structure containing reconstruction features.

[0048] This step leverages the feature extraction capabilities of Graph Neural Networks (GNNs) to address data quality issues. Spatiotemporal embedding features are extracted for each node using graph convolution or graph attention networks. These features not only contain information about the node itself but also aggregate information from its neighboring nodes.

[0049] The specific steps for performing node missing completion, anomaly detection, and residual feedback correction include: aggregating the features of neighboring nodes in the spatiotemporally dynamic relational graph structure using a graph message passing mechanism to generate a node completion matrix; applying a sliding window residual threshold detection algorithm to the observation subset of the node completion matrix to identify and correct abnormal observation points; introducing a residual gated feedback unit to calculate the node prediction error and feeding the node prediction error back to the edge weight update process of the neighboring nodes to generate a feedback correction matrix; and fusing the corrected observation subset with the feedback correction matrix to obtain the optimized graph structure.

[0050] Specifically, to address data missingness, the system utilizes a graph message passing mechanism. If data for a node (such as the Qingjiang Gaobazhou Station) is missing at a certain time, its value can be inferred by aggregating the weighted features of its upstream and downstream neighboring nodes (such as Zhicheng and Yichang), thus generating a node completion matrix.

[0051] To address data anomalies, a sliding window residual detection algorithm is employed. Specifically, for the observed sequence, a time window is defined, for example, with a width of 12 hours. Within this window, the imputation value X is calculated. v_hat The residual r between (t) and the original observation value v (t)=|X v_hat (t)-x v (t)|。 The system presets a threshold, for example, 3 times the standard deviation 3σ. If the calculated residual exceeds this threshold, the observation is determined to be abnormal, possibly due to sensor failure or radar false alarm, and a replacement correction is performed using the completed value or weighted smoothed value.

[0052] Furthermore, to improve the accuracy of feature reconstruction, especially considering the non-stationarity of time-series data, this embodiment introduces a residual gated feedback mechanism. Specifically, this includes: introducing a residual gated feedback unit to calculate the node prediction error, and feeding the node prediction error back to the edge weight update process of neighboring nodes to generate a feedback correction matrix.

[0053] The specific calculation formula for this process can be expressed as follows:

[0054] e v (t)=g(t)*(X v_hat (t)-X v (t-1));

[0055] Among them, e v (t) represents the prediction error feedback amount of node v at time t. v_hat (t) is the current node completion value or prediction value, X v (t-1) is the node state value at the previous time step. g(t) is the gating coefficient, which is usually generated by an activation function (such as Sigmoid) and takes a value between 0 and 1. It is used to control the proportion of error feedback.

[0056] Calculated error e v (t) is not only used to correct the current node features, but also backpropagated to adjust the connection weights between the node and its neighbors, so that the model can focus on the region with large prediction error and adaptively optimize the feature extraction strategy in subsequent steps.

[0057] The system fuses the corrected subset of observations with the feedback correction matrix to obtain an optimized graph structure. The node features in this structure fill in missing information, eliminate anomalies, and are enhanced through spatiotemporal correlation, resulting in higher reliability.

[0058] Step 204: The optimized graph structure is transformed into a high-dimensional embedding matrix, and tensor expansion and spatiotemporal normalization are performed to generate the model input dataset.

[0059] After the above processing, each node possesses an enhanced high-dimensional feature vector. The feature vectors of all nodes are stacked to form a high-dimensional embedding matrix E(t). To accommodate the input requirements of time series prediction models (such as Long Short-Term Memory networks (LSTM) or Transformer models), the embedding matrix at consecutive time points needs to be expanded along the time dimension.

[0060] For example, if the forecast period is set to 5 days and the time step is 6 hours, then the number of time steps K=20. The system expands the embedding matrix sequence from time t to t+K into a three-dimensional tensor, whose dimensions can be represented as (K,N,d), where N is the number of nodes, such as 17, and d is the feature dimension, preferably 64 dimensions. Spatiotemporal standardization is performed on this tensor, for example, using the Z-Score standardization method, subtracting the mean and dividing by the standard deviation to eliminate the order-of-magnitude differences between different physical units (such as water level and flow rate). The final standardized tensor is the model input dataset, which is directly fed into the subsequent aggregated reservoir parameterized structure model.

[0061] Example 3 details the construction process of the parameterized structural model of the aggregated reservoir, especially the specific implementation of the dynamic water level-reservoir capacity relationship, focusing on solving the problem that traditional static hydrological methods cannot describe the dynamic deformation of river and lake storage capacity during flood rise and fall.

[0062] Step 301: The multi-level equivalent unit method is used to dynamically aggregate the main river channel and the lake flood storage area, determine the adaptive aggregation boundary and the equivalent dam site section, and generalize the water body within the adaptive aggregation boundary into aggregated reservoir units.

[0063] In this embodiment, considering the complex river-lake relationship in the middle reaches of the Yangtze River from the Three Gorges to Luoshan, while direct full-area two-dimensional hydrodynamic simulation offers high accuracy, its computational time is insufficient to meet the requirements of real-time rolling forecasts. This step introduces the physical generalization concept of aggregated reservoirs. The multi-level equivalent unit method is a technique for logically merging dispersed water bodies. Specifically, using clustering algorithms or elevation-based connectivity analysis, the main channel (such as the Zhicheng to Luoshan section), its laterally connected Dongting Lake area, and the flood storage and detention areas along the route are divided into the same hydraulic unit.

[0064] An adaptive convergence boundary refers to a spatial extent that is not fixed but dynamically adjusted according to the water level during the forecast period. For example, at low water levels, the convergence boundary only includes the main channel; while at high water levels, the boundary automatically expands to include the floodplain and flood storage area. The equivalent dam site section is determined as the controlling outflow section of the convergence reservoir. In this embodiment, the section where the Luoshan hydrological station is located is preferably selected as the equivalent dam site section.

[0065] Based on this, the system needs to calculate the total inflow to the aggregated reservoir. Merging the outflow from the upstream reservoir, the inflow from the tributaries, and the confluence of the inter-regional flow into a total inflow sequence for the aggregated reservoir is crucial for driving subsequent dynamic deformation. The total inflow sequence Q of the aggregated reservoir is... in The specific formula for calculating (t) can be expressed as:

[0066] Q in (t)=∑Q out_u (t)+∑Q trib_k (t)+Q int (t);

[0067] Where, ∑Q out_u (t) represents the sum of outflows from all upstream cascade reservoirs (such as the Three Gorges Reservoir and the Geheyan Reservoir) at time t; ∑Q trib_k (t) represents the sum of the inflows of all tributaries (such as the Xiangjiang River, Zishui River, Yuanjiang River, and Lishui River) within the convergence boundary at time t; Q int (t) represents the lateral runoff generated by rainfall in the aggregated area that is not controlled by the hydrological station.

[0068] In this way, the originally complex multi-source inflow is integrated into a single driving variable.

[0069] Step 302: Based on the historical observation data in the model input dataset, fit and generate an adaptive static structure of water level-reservoir capacity; perform signal decomposition on the total inflow sequence of the aggregated reservoir to obtain a trend term reflecting the slow change characteristics of inflow and a disturbance term reflecting the abrupt change characteristics of inflow.

[0070] This step establishes the static baseline and dynamic driving force of the storage-discharge relationship. Using historically measured water level and flow data, combined with river topography data, historical reservoir capacity sequences are inferred through water balance analysis. The least squares method is used to perform regression analysis on the historical water level-reservoir capacity data pairs to obtain the basic water level-reservoir capacity curve, i.e., the water level-reservoir capacity adaptive static structure V. s (Z). This structure represents the average storage capacity under steady-state flow conditions.

[0071] Furthermore, in order to capture the dynamic impact of flood rise and fall on storage capacity, the system analyzes the aggregated reservoir inflow sequence Q obtained in step 301. in (t) Signal decomposition is performed. Specifically, methods such as moving average filtering, wavelet transform, or Empirical Mode Decomposition (EMD) can be used. The decomposition process can be represented as:

[0072] Q in (t)=Q bar_in (t)+Q' in (t);

[0073] Among them, Q bar_in (t) represents the trend term, indicating the low-frequency gradual change component in the inflow process, reflecting the overall magnitude and background flow of the flood; Q' in (t) represents the disturbance term, indicating high-frequency abrupt changes in the inflow process, reflecting the rapid rise and fall rate of the flood wave. For example, when the flood peak arrives rapidly, Q' in (t) will show a significant positive value; while in the early stage of receding water, Q' in (t) may then turn negative. These two components will be used to control the overall rate and local shape of the reservoir capacity curve, respectively.

[0074] Step 303: Construct a time-varying driving factor to control the rate of change of reservoir capacity using the trend term, and construct a nonlinear enhancement operator to control the shape correction of the control curve using the disturbance term; use the time-varying driving factor and the nonlinear enhancement operator to adjust the parameter weights of the water level-reservoir capacity adaptive static structure in real time, and generate a dynamic water level-reservoir capacity relationship that evolves dynamically with the inflow process.

[0075] Traditional static curves cannot reflect the increase in reservoir capacity caused by wedge-shaped storage or additional gradient during flood rise. This embodiment solves this problem through a dynamic parameter adjustment mechanism. Specifically, the time-varying driving factor α(t) is mainly used to adjust the overall rate of change of reservoir capacity with water level, and its preferred formula is:

[0076] α(t)=Q barin (t) / Q max ;

[0077] Among them, Q max The maximum inflow value in the historical observation sequence is used for normalization.

[0078] Nonlinear enhancement operator B i Primarily used to control the degree of nonlinear curvature of curves, i.e., shape correction, its preferred construction formula is:

[0079] B i =b i *Q' in (t);

[0080] Among them, b i These are the enhancement operator coefficients to be calibrated, typically ranging from 0 to 1, used to control the influence weight of the disturbance term.

[0081] The adaptive static structure of water level-reservoir capacity is represented by a third-order polynomial. The specific calculation expression for the dynamic water level-reservoir capacity relationship V(Z) is as follows:

[0082] ;

[0083] Where Z is the water level at the current time t; V(Z) is the calculated dynamic reservoir capacity; a0, a1, a2, and a3 are the static polynomial coefficients, determined by fitting historical data. B1, B2, B3, and B4 are the nonlinear enhancement operators for each order of the corresponding polynomial, defined as the product of the disturbance term and the preset enhancement operator coefficients.

[0084] This formula shows that when the flood is in a stable period (Q') in When (t) approaches 0, the exponential term B i Approaching 0, When the value approaches 1, the dynamic relation V(Z) degenerates into the static relation V. s (Z), which conforms to the laws of physics. However, when the floodwaters rise rapidly (Q'),... in When (t) is large, B i As the value increases, the coefficients are nonlinearly amplified or reduced by α(t) (depending on whether α(t) is less than or greater than 1 and the characteristics of the base), causing the calculated reservoir capacity V(Z) to shift relative to the static value. For example, during the rising water stage, due to the increased water surface gradient, the same water level in front of the dam corresponds to a larger actual water volume. This formula can accurately simulate the physical phenomenon of reservoir capacity expansion through parameter adjustments.

[0085] Optionally, establishing a parameterized structural model of the aggregated reservoir that characterizes the overall storage and discharge characteristics of the main stream and the lake may also include the following processes, such as... Figure 3 As shown:

[0086] The multi-level equivalent unit method is used to dynamically aggregate the main river channel and lake flood storage and detention areas, determine the adaptive aggregation boundary and equivalent dam site section, and generalize the water body within the adaptive aggregation boundary into aggregated reservoir units.

[0087] Based on the model input dataset, the outflow from the upstream reservoir, the inflow from the tributaries, and the confluence flow between the sections are extracted and merged into a total inflow sequence of the reservoir.

[0088] Using the total inflow sequence of the aggregated reservoir as the driving variable, the storage and release relationship of the aggregated reservoir unit is constructed.

[0089] Compared with the traditional method of using a fixed reservoir capacity curve, the dynamic water level-reservoir capacity relationship of the present invention can adapt to the flood rise and fall rate, accurately reflecting the changes in additional gradient and wedge-shaped storage.

[0090] The following is a typical calculation example:

[0091] Suppose the trend term Q at a certain moment bar_in (t) = 15000 m³ / s, historical maximum inflow Q max =50000m³ / s, then the time-varying driving factor α(t) = 15000 / 50000 = 0.3; assuming the disturbance term Q' in (t)=2000m 3 / s, enhancement operator coefficient b1=0.5, then B1=0.5×2000=1000. Let the static coefficient a0=100×10 6 m 3 a1 = 50 × 10 6 m 3 / m、a2=10×10 6 m 3 / m 2 a3 = 1 × 10 6 m 3 / m 3 When the water level Z = 30m, substituting into the formula, the dynamic reservoir capacity V(Z) is calculated to be approximately 275 × 10⁻⁶ m. 8 m 3 Compared with the static reservoir capacity, it reflects the outward expansion effect of reservoir capacity during the flood season.

[0092] Example 4 details the construction process of the composite water level-flow relationship, focusing on solving the common problem of non-phase response in river and lake systems, namely, the complex rope curve characteristics of the water level-flow relationship due to multiple influences of lag time, backwater, and scouring and silting.

[0093] Step 401: Extract water level and flow characteristics from the observation subset of the model input dataset, and construct the water level-flow master relationship based on the power law function through multi-scale nonlinear mapping.

[0094] This step establishes the baseline framework for discharge capacity. The system filters historical data to identify periods of stable flow where there is no significant backwater pressure, such as the dry season or the end of the receding water period. Regression analysis is then performed on these data to construct the primary water level-discharge relationship Q. b (Z). This relationship is preferably expressed in the power-law function form of the Manning formula or the simplified Saint-Venant equations:

[0095] Q b (Z) = c*(Z - Z0) p ;

[0096] Or simplified to: Q b (Z)=c*Z' p ;

[0097] Where c is the flow coefficient, p is the flow index, typically close to 1.5 to 2.0, Z0 is the riverbed elevation or zero-flow weighted water level, and Z' is the effective water depth. This master control relationship represents the single mapping capability of the control section water level to the flow rate under ideal, undisturbed conditions.

[0098] Step 402: Based on the water level-discharge master control relationship, a time lag correction model representing the flood propagation time delay, a backwater feedback model representing the downstream water level's inhibitory effect on outflow, and a scouring and deposition compensation model representing changes in riverbed topography are constructed respectively. Through a hierarchical parameterized coupling mechanism, the time lag correction model, the backwater feedback model, and the scouring and deposition compensation model are superimposed on the water level-discharge master control relationship to establish a composite water level-discharge relationship driven by the interaction of the time domain, spatial domain, and morphological domain.

[0099] This step expands the single master control relationship into a multi-dimensional coupled dynamic relationship. The hierarchical parameterized coupling mechanism refers to parameterizing the three main factors affecting flow—delay (time domain), backwater (spatial domain), and scouring and sedimentation (morphological domain)—as independent correction terms, and then integrating them into the master control equation in a superimposed or nested manner.

[0100] Specifically, the general expression of the composite water level-discharge relationship Q(Z) can be constructed as follows:

[0101] Z real (t)=[Q(t-τ(t)) / c] (1 / p) +Z bk (t)+Δh bed ;

[0102] Among them, Z real (t) represents the measured water level. The first term is the theoretical water level derived from the flow rate (including lag time), and the second term is Z. bk (t) represents the water level rise caused by the backwater jacking, and the third term Δh bed This refers to the water level shift caused by riverbed erosion and siltation.

[0103] Alternatively, it can be expressed in the form of inverse flow rate: Q(t) = c * [Z(t + τ'(t)) - Z bk (t+τ'(t))-Δh bed ] p ;

[0104] Where Q(t) is the actual flow rate at time t, Z(t+τ'(t)) is the water level observation value at future time t+τ'(t) needed to obtain the flow rate at time t, τ'(t) is the reverse lag used to compensate for the time delay of flood wave propagation, and Z... bk (t+τ'(t)) is the top water level correction term at the corresponding time, which needs to be used synchronously with the water level, and p is the flow rate index.

[0105] Step 403, construct the lag correction model, such as Figure 4 As shown, the specific steps include: calculating the time rate of change of the total inflow sequence of the aggregated reservoir to obtain the inflow lag gradient; nonlinearly and dynamically amplifying the preset baseline lag parameters based on the inflow lag gradient to generate a time-varying lag factor; and using the time-varying lag factor to construct a time-series residual feedback channel in the water level-flow control relationship to establish a correction structure that drives the water level-flow control relationship to dynamically delay mapping with changes in inflow.

[0106] This step details the implementation of the time-delay correction model (time domain). In long-channel reservoirs or river-lake systems, the propagation time from upstream to downstream control sections is time-consuming, and this propagation time (delay) is not constant but varies with the flow rate and the rate of rise and fall. The specific formula for calculating the time-varying delay factor τ(t) is as follows:

[0107] τ(t)=τ0*(1+k1*dQ in );

[0108] Where τ(t) is the time-varying lag factor, τ0 is the baseline lag parameter to be calibrated, representing the propagation time under the average flow velocity; k1 is the lag amplification factor to be calibrated, which is a positive real number to be calibrated; dQ in The inflow lag gradient is represented by the dynamic time delay mapping, which is achieved by introducing a time offset term Q(t-τ(t)) into the water level-flow control relationship.

[0109] dQ in The calculation can be performed using the finite difference method:

[0110] dQ in (t)=[Q in (t)-Q in [(t-Δt)] / Q in (t-Δt);

[0111] When the flood is in a rapid rising phase (dQ) in When the flood peak is greater than 0, the flood wave becomes steeper, and although the wave velocity increases, the overall lag effect caused by the flattening of the flood peak may be amplified due to the enhanced regulation and storage capacity of the river channel, or it may manifest as accelerated propagation in some river types. By adjusting the sign and magnitude of the parameter k1, this formula can flexibly adapt to different watershed characteristics, such as rapid rise and slow receding or slow rise and rapid receding. Using the calculated τ(t), the system introduces the Q(t-τ(t)) term into the master control relationship to achieve dynamic time delay mapping.

[0112] The value of parameter k1 has a significant impact on the time delay response: increasing k1 makes the time delay factor more sensitive to the inflow gradient, which is suitable for steep river sections with drastic changes in water surface gradient; decreasing k1 makes the time delay change more gradual, which is suitable for gentle river sections. The typical value range is 0.1 to 1.0, and it can be determined based on historical flood data.

[0113] Step 404: Obtain real-time water level data of the equivalent dam site section and calculate the rate of change of water level over time; use the rate of change of water level as the driving variable to construct a backwater feedback function that characterizes the degree of backwater in the downstream river channel; superimpose the backwater feedback function as a negative feedback term onto the water level variable in the composite water level-discharge relationship to compensate for the backwater effect of the downstream water level rise on the outflow process.

[0114] This step details the implementation of the backwater feedback model (spatial domain). In the middle reaches of the Yangtze River, the outflow at the Luoshan section is not only controlled by the upstream inflow but also significantly affected by the downstream Hankou water level and the backwater effect of Dongting Lake. This backwater effect manifests as follows: at the same flow rate, the water level is raised; or at the same water level, the discharge capacity is reduced.

[0115] To quantify this effect, this embodiment constructs a return water feedback function Z. bk (t). Preferably, the function uses the normalized acceleration of the rate of change of water level as the driving variable:

[0116] dZ=[Z(t)-Z(t-Δt)] / [Z(t-Δt)-Z(t-2Δt)+ε];

[0117] Where ε is a small constant to prevent the denominator from being zero, for example, 10. -6 The return water feedback function can be specifically expressed as:

[0118] Z bk (t)=Z bk0 *(1+k2*dZ);

[0119] Among them, Z bk0 The reference top support water level is k2, and the top support amplification factor is k2.

[0120] The physical meaning of this formula is that when the acceleration of water level rise suddenly increases (dZ increases significantly), downstream obstruction or backlog occurs, for example, the downstream water level rises faster than the upstream water level, causing the water level at this section to rise passively and rapidly. At this time, Z... bk As (t) increases, in equation Z real =Z theory +Z bk (Z) theory Z is the theoretical water level. bk This accounts for a larger proportion of the increase in water level (due to the backwater effect), explaining why the measured water level Z... real The data is very high, but the actual traffic volume has not increased accordingly.

[0121] Step 405: Obtain the topographic evolution characteristics and historical sedimentation change rate of the river channel cross section; construct a scour-deposition offset correction term that can reflect the change of riverbed elevation over time based on the topographic evolution characteristics; use the scour-deposition offset correction term to compensate for the morphological shift of the water level variable in the composite water level-discharge relationship, so as to eliminate the long-term drift effect of topographic changes on the water level-discharge response relationship.

[0122] This step details the implementation of the scour-deposition compensation model (morphological domain). Riverbed topography is not static; long-term scour causes a leftward shift in the water level-discharge relationship (lower water level for the same discharge), while deposition causes a rightward shift. Specifically, the system periodically imports riverbed topographic measurement data, or performs water level analysis based on the same discharge from long-series hydrological data, to calculate the offset Δh of the riverbed elevation relative to the baseline year. bed The scouring and silting offset correction term can be constructed as a piecewise linear function or a step function at time t: ΔZ bed (t)=Δh bed_Year (y), ΔZ bed (t) represents the water level shift caused by riverbed scouring and deposition, Δh bed_Year (y) represents the change in riverbed elevation in year y relative to the baseline year (such as the initial year of station establishment or a stable year). In real-time forecasting, this term is added as a slowly changing constant to the composite relation. Although its short-term fluctuations are small, it is crucial for maintaining the accuracy of the model across multiple years, avoiding systematic overestimation errors in water level forecasts caused by riverbed incision.

[0123] As an improvement to the above scheme, in some embodiments, a machine learning residual compensation module can also be introduced. For the residual part that the above physical parameterization model (main control + lag + top support + scouring and silting) still cannot explain, a lightweight multilayer perceptron (MLP) can be used for fitting, and the output of the MLP can be superimposed on the composite relationship as a fifth correction term to form a hybrid modeling scheme of physical mechanism + data-driven approach.

[0124] Example 5 details the construction and execution process of the adaptive solution mechanism, focusing on solving the problem of computational divergence or water imbalance caused by the inability of a single solution algorithm to adapt to rapidly changing boundary conditions when traditional hydrological models face strong nonlinear hysteresis effects.

[0125] Step 501: Construct a time-delay state discriminator to monitor the time-varying time-delay factor in the composite water level-flow relationship in real time and output the real-time time-delay state that reflects the current flood response characteristics.

[0126] In this embodiment, the time-delay state discriminator is a logic control unit, similar to the control logic of a vehicle's automatic transmission. It does not directly participate in numerical calculations but is responsible for monitoring the system's operating conditions. The monitored metric is the time-varying time-delay factor τ(t). This factor quantifies the degree of propagation delay of the flood wave at the current moment. The discriminator reads the τ(t) value at each moment to determine whether the current river and lake system is in a quasi-steady-state operating mode or a violent dynamic transition mode.

[0127] Specifically, the discriminator internally presets a set of threshold parameters that can distinguish different hydraulic response stages. The discriminator's output—the real-time lag state—is an enumerated signal or flag, for example, marked as state 0 (static / synchronous) or state 1 (dynamic / out-of-phase). This state signal serves as the control command for the downstream algorithm selector, ensuring that the solver always operates in the mode most suitable for the current hydraulic characteristics.

[0128] Step 502: In parallel, a dynamic reservoir capacity synchronous iterative algorithm suitable for the synchronous response stage of water level and flow rate, and a prediction-correction hybrid algorithm suitable for the heterogeneous response stage of water level and flow rate are established to form a heterogeneous and complementary dual algorithm channel.

[0129] This step constructs two parallel computing engines. Heterogeneous complementarity means that these two algorithms have different mathematical principles and applicable scenarios, but their functions are complementary.

[0130] Channel A: Dynamic Reservoir Capacity Synchronous Iterative Algorithm. This algorithm is based on the synchronous response assumption, which assumes that changes in water level and flow rate are quasi-synchronous within a short time step, such as 1 hour. Its logic involves simultaneously solving the storage and release equations and the water balance equations, and then using a fixed-point iteration method or the Newton-Raphson method to simultaneously solve for the current water level Z(t+1) and flow rate Q(t+1). This method offers fast convergence, high accuracy, and strict water balance assurance when the lag time is small and the water flow is stable.

[0131] Channel B: Prediction-Correction Hybrid Algorithm. This algorithm is designed to handle cases with significant anomalies. When large time delays exist, i.e., τ(t) is large, the outflow Q(t+1) is essentially determined by past inflow conditions, weakening its instantaneous correlation with the current water level. Forcing synchronous iteration may result in the equations failing to converge due to Jacobian matrix singularities or iterative oscillations. This channel employs a step-by-step strategy: predicting approximate water level or flow rate based on time delay relationships, and then correcting using the water balance equation. This method sacrifices some synchronization accuracy but gains higher numerical stability, ensuring the program does not crash during periods of dramatic flood peak evolution.

[0132] Step 503: Based on the real-time delay state, automatically switch between the two algorithm channels and select the algorithm path that matches the current response characteristics to solve the flood control calculation equations.

[0133] This step implements dynamic routing of the solution path. At each simulation time step, for example, t=k, the system first calls the discriminator in step 501 to obtain the state. If the state is static, the data stream is directed to channel A; if the state is dynamic, the data stream is directed to channel B. After the calculation is completed, regardless of which channel it passes through, the calculated state variables (water level, flow rate, reservoir capacity) in a uniform format are output and written back to the system state memory as the initial conditions at time t=k+1. This mechanism avoids the rigid use of a single algorithm by traditional models throughout the flood process, balancing computational efficiency during the normal water period and computational stability during the flood period.

[0134] Step 504: Use the time-varying lag factor to compare with the preset stable threshold and fluctuation threshold using the lag state discriminator. When the time-varying lag factor is lower than the stable threshold, the real-time lag state is determined to be in the static range. When it is higher than the fluctuation threshold, it is determined to be in the dynamic range. When it is between the stable threshold and the fluctuation threshold, the determination state of the previous moment is maintained.

[0135] This step elaborates on the quantification criteria for state determination. To avoid frequent jumps in the algorithm at critical points, this embodiment preferably introduces hysteresis intervals or double threshold logic.

[0136] Specifically, a stability threshold τ is set. s and fluctuation threshold τ d The preferred empirical value is τ. s =1, corresponding to 1 computation time step, and τ d =2, corresponding to 2 computation time steps. The decision logic is as follows:

[0137] If τ(t) < τ s This is determined to be a static interval. At this time, the flood spreads rapidly or it is during the dry season, and the water level and flow rate are basically synchronized.

[0138] If τ(t) > τ dThis is determined to be a dynamic range. At this point, the flood lag effect is significant, and the waveform deformation is large.

[0139] If τ s ≤τ(t)≤τ d To maintain system stability, the judgment state of the previous moment can be kept unchanged, or it can be defined as a transition interval (the more stable prediction-correction method is used by default).

[0140] Step 505: When in the static interval, activate the dynamic reservoir capacity synchronous iterative algorithm, and solve the reservoir capacity, water level and flow variables simultaneously until the convergence condition is met.

[0141] In this step, the specific execution flow of the dynamic storage capacity synchronization iterative algorithm is as follows:

[0142] Initialization: Assume the initial water level Z at time t+1. (0) (t+1) equals the water level Z at the previous moment. t .

[0143] Iterative calculation (kth iteration):

[0144] a) Based on the current assumed water level Z (k) (t+1) and the composite water level-discharge relationship Q(Z), calculate the trial discharge Q. (k) (t+1).

[0145] b) According to the water balance equation: V (k) (t+1)=V t +[Q in (t)+Q in (t+1)-Q t -Q (k) (t+1)]*Δt / 2, calculate the trial storage capacity.

[0146] c) Based on the inverse function of the dynamic water level-reservoir capacity relationship V(Z), from V (k) (t+1) Back-calculate the new water level Z (k+1) (t+1).

[0147] Convergence criterion: Check | Z (k+1) (t+1)-Z (k) If (t+1)| is less than the preset tolerance ε, for example, 0.001 meters, output the result; otherwise, set k=k+1 and return to continue the iteration.

[0148] Step 506: When in the dynamic range, the prediction-correction hybrid algorithm is activated, which uses the historical flow sequence containing lag information to predict the future water level response trend, and performs correction and solution based on the dynamic water level-storage capacity relationship and water balance equation.

[0149] In this step, the specific execution flow of the prediction-correction hybrid algorithm is as follows, performing logical closed-loop processing on the data source for the lag time window:

[0150] Prediction step: Directly calculate the predicted water level Z at time t+1 using the master control lag relationship of the flow rate. pred The formula is:

[0151] Z pred (t+1)=g -1 (Q base );

[0152] Among them, g -1 (.) is the inverse function of the main control water level-flow rate relationship, and the reference flow rate Q is... base This depends on the lag factor τ(t). Here, we need to refer to the flow rate information at time t+1-τ(t). If t+1-τ(t) points to a future time, i.e., the lag is very short, less than the forecast step size, then Q... base The predicted flow sequence from the upstream hydrological station is used and calculated to be transmitted to this section; if t+1-τ(t) points to the past or present time, i.e., the time lag is relatively long, then Q base Use known measured flow sequences or historical calculated values. This process ensures the integrity of the logic chain under all time delay conditions.

[0153] Correction step: a) Using the predicted water level Z pred Query the dynamic storage capacity curve V(Z) to obtain the predicted storage capacity V. pred .

[0154] b) V pred Substituting into the transformed form of the water balance equation, we can solve for the final flow rate Q that satisfies the water balance. final The formula is:

[0155] Q final =Q in (t)+Q in (t+1)-Q t -2*(V pred -V t ) / Δt.

[0156] By employing the non-iterative strategy of determining the water level first and then the flow rate, the risk of divergence of high-dimensional nonlinear equations in the lag interval is avoided.

[0157] Example 6 details the parameter calibration and correction process, focusing on solving the common problem of different parameters having the same effect in hydrological models, that is, different parameter combinations may lead to similar simulation results, reducing the physical interpretability of parameters and the generalization ability of the model.

[0158] Step 601: Obtain the simulated water level value and the measured water level value, as well as the simulated flow rate value and the measured flow rate value at the same time.

[0159] This step is fundamental to error calculation. During model training or real-time calibration, the system obtains the measured water level Z of the control section (such as Luoshan Station) from the historical database or real-time telemetry network. obs (t) and measured flow rate Q obs (t). Simultaneously, the model runs based on the current parameter set Θ, outputting the corresponding simulated water level Z. sim (t) and simulated flow Q sim (t). To eliminate the influence of dimensions, these data are usually normalized or balanced by weights in the subsequent objective function.

[0160] Step 602: Construct a dynamic weight allocation mechanism, and assign dynamic weights to the water level fitting error and the flow rate fitting error respectively based on the differences in the importance of water level and flow rate at different forecast stages.

[0161] Traditional calibration methods often use fixed weights, such as 50% for water level and 50% for flow rate. However, the focus of flood control scheduling changes with water conditions. This embodiment introduces a dynamic weight allocation mechanism.

[0162] Specifically, the water level weight ω is defined. Z (t) and flow weight ω Q (t), satisfying ω Z (t)+ω Q (t)=1. The allocation strategy can be adaptively adjusted based on the inflow size:

[0163] When the inflow is small (dry season or normal water season), the flow measurement error is relatively large and the focus on flood control is low. At this time, the system automatically increases the water level weight (e.g., ω). Z =0.8), with a focus on ensuring the accuracy of water level simulation to serve shipping or ecological scheduling.

[0164] When the inflow is large (during flood season), accurate prediction of peak flow is crucial for dam peak reduction and staggering. In this case, the system automatically increases the flow weight (e.g., ω). Q =0.7), focusing on capturing the flood peak process.

[0165] Step 603: Based on dynamic weights, the water level fitting error and the flow rate fitting error are weighted and summed to generate a multi-index comprehensive objective function for evaluating the model fitting performance, and the model parameters are corrected with the goal of minimizing the multi-index comprehensive objective function.

[0166] The mathematical expression for the multi-index comprehensive objective function J can be constructed as follows:

[0167] J=∑t=1 T [ω Z (t)*((Z sim (t)-Z obs (t)) / σ Z ) 2 +ω Q (t)*((Q sim (t)-Q obs (t)) / σ Q ) 2 ];

[0168] Where T is the length of the calibration time window; σ Z and σ Q These are the standard deviations of the measured water level and flow rate series, respectively, used to eliminate dimensional differences.

[0169] This objective function not only considers numerical approximation, but also implicitly emphasizes key flood control characteristics (such as flood peaks) through the introduction of dynamic weights. The process of model correction involves finding the parameter combination that minimizes J in the parameter space.

[0170] Step 604: Using a parameter decoupling strategy, the high-dimensional parameter set of the aggregated reservoir parameterized structure model is mapped into mutually independent subsets of reservoir capacity parameters and flow parameters. The reservoir capacity parameter subset and flow parameter subset are independently optimized in the wide-domain parameter space to identify the physical effective range of each parameter subset.

[0171] This step is the first stage of the hybrid calibration strategy – dimensionality reduction and decoupling. The aggregated reservoir model contains two classes of parameters with distinctly different properties: a subset of reservoir capacity parameters Θ that controls water storage capacity. v (e.g., polynomial coefficient a) i Enhancement coefficient b i and the subset of flow parameters Θ for controlling the discharge capacity. Q (e.g., flow coefficient c, exponent p, delay coefficient k1, etc.). If directly calibrated simultaneously in a high-dimensional space, it is easy to get trapped in local optima. This embodiment uses a physical mechanism decoupling method:

[0172] Lock Θ Q Only the rate is fixed. v Using measured inflow and outflow, the theoretical reservoir capacity at each moment is directly inferred from the water balance equation. Then, using the measured water level, the VZ relationship is fitted. This step does not involve flow rate calculation; it is a one-to-one function fitting, which can quickly determine Θ. v The physical effective range R v .

[0173] Lock Θ v Only the rate is fixed. QUsing the measured water level as input, the composite water level-flow rate formula is directly used to calculate the outflow, which is then compared with the measured flow rate. This step does not involve reservoir capacity calculation and can quickly determine Θ. Q The physical effective range R Q A wide-range parameter space refers to setting the initial search range of parameters to be very wide, such as [0,+∞] or [-10,10], to ensure that no physical possible solutions are missed.

[0174] Furthermore, a high-dimensional parameter search space is reconstructed based on the physical effective interval, and global collaborative optimization is performed in this space to obtain the optimal parameter combination that takes into account both storage and leakage coupling characteristics.

[0175] Step 605: Statistically analyze the independent optimization results of the subset of storage capacity parameters and the subset of flow parameters. By applying a preset proportion of numerical perturbation, construct a refined parameter boundary that covers the range of real physical changes and is significantly narrower than the original wide-area space.

[0176] This step is the second stage – space contraction. After the independent optimization in step 604, anchor values ​​for each parameter under ideal decoupling conditions are obtained. However, in actual operation, storage and leakage are coupled, and the optimal solution may fluctuate around the anchor points. The system uses the parameter values ​​obtained from independent optimization as the center and applies a small percentage numerical perturbation (e.g., ±10% or ±15%) to construct a new search boundary. For example, if the lag amplification factor k1 obtained from independent calibration is 0.5, then the refined parameter boundary is constructed as [0.45, 0.55]. The new boundary range is typically several orders of magnitude smaller than the original wide-area space, reducing the difficulty and uncertainty of subsequent searches.

[0177] Step 606: Using the refined parameter boundary as a constraint, a joint search is performed on the subset of reservoir capacity parameters and the subset of flow parameters within the same solution space; a dual convergence criterion including the objective function improvement rate and the stability of the population parameters is set, and the optimal parameter combination is output when the dual convergence criterion is satisfied.

[0178] This step is the third stage – global coordination. Within the contracted, fine-grained space, the system releases all parameters, allowing them to vary simultaneously to capture the weak coupling effects of the water storage and release processes. Cooperative optimization can employ evolutionary algorithms, such as Differential Evolution (DE) or Particle Swarm Optimization (PSO). To prevent overfitting or premature convergence, a strict dual convergence criterion is set:

[0179] Criterion for the improvement rate of the objective function: The optimal objective function value J of a population spanning N consecutive generations (e.g., 20 generations). best The relative rate of change is less than the threshold ε J (Preferred to be 10) -3 ).

[0180] Population parameter stability criterion: The variance of the distribution of all individuals in the current population in the parameter space is less than the threshold ε. Θ (Preferred to be 10) -2 ).

[0181] The algorithm stops iterating and outputs the final optimal parameter combination only when both conditions are met simultaneously.

[0182] Compared to direct global optimization without this strategy, the method in this embodiment not only improves the convergence speed, but also improves the Nash efficiency coefficient (NSE) of the obtained parameters during the validation period, thereby enhancing the robustness of the forecast.

[0183] As an alternative, on embedded devices with limited computing resources, steps 604 and 605 can be omitted, and the system can directly run using a pre-trained set of fixed parameters. Alternatively, only the 2-3 most sensitive parameters (such as the flow coefficient c and baseline lag τ0) can be fine-tuned online, while the other parameters are frozen. Although this simplified approach sacrifices some accuracy, it can meet the application requirements of low power consumption and low computing power.

[0184] Example 7: An optional scheme for an intelligent flood evolution forecasting method based on dynamic storage-discharge relationships in river and lake regulation systems is provided, comprising the following steps:

[0185] Step 701: Construct a fusion method of graph embedding and dynamic structure reconstruction. Align multi-source heterogeneous hydrological data such as control section flow and reservoir outflow according to time series, and establish an initial spatiotemporal correlation graph structure based on spatial location, physical connectivity and hydrodynamic influence range. Use node feature aggregation mechanism to extract the correlation patterns between multi-source information, and dynamically adjust the hydraulic connectivity intensity and interaction relationship through edge weight self-learning mechanism to form a spatiotemporally enhanced graph structure that can reflect the evolution law of flood. Project the correlation features of each node and its edge to the same spatiotemporal scale to generate a model input dataset with a unified graph structure.

[0186] Step 702: Treat the main river channel and lake flood storage and detention areas as a unified unit with a common storage and discharge mechanism, generalize it as a clustered reservoir, and construct a clustered reservoir representation system based on the model input dataset: By introducing time-varying factors to drive the generation of an adaptive polynomial framework, a multi-dimensional dynamic relationship of water level-storage capacity is formed, which changes with time, water level and hydraulic boundary, etc., and a multi-factor interactive coupling model is constructed to represent the influence of factors such as flood rise and fall, downstream backwater and scouring and deposition, and a composite water level-discharge relationship with time-varying feedback characteristics is generated, which constitutes a parameterized structural model of the clustered reservoir that can accurately characterize the storage and discharge response features of the clustered reservoir.

[0187] Step 703: Based on the parameterized structural model of the aggregated reservoir, a flood regulation calculation equation set is constructed by combining the storage and discharge balance and continuity equations. An adaptive solution mechanism is designed based on the time lag characteristics in the composite water level-discharge relationship, so that the model can automatically match the appropriate calculation path to update the boundary conditions, forming an aggregated reservoir flood regulation calculation framework suitable for complex river and lake systems.

[0188] Step 704: Establish a hybrid global-local intelligent optimization method. Using the aggregated reservoir flood control calculation framework as the core execution environment, automatically calibrate the key storage and discharge parameters in the parameterized structural model of the aggregated reservoir. Driven by real-time observation data, dynamically correct the model parameters and state variables so that the model can reflect the current hydrological state changes in real time. Use the corrected model to perform rolling prediction and output the water level process, flow process and flood peak characteristics information in the future forecast period.

[0189] According to one aspect of this application, a model input dataset with a unified graph structure is generated, further comprising:

[0190] Multi-source heterogeneous hydrological data from reservoirs, rivers, and lakes are used to generate an initial graph structure G0 through dynamic topological mapping. A multimodal hydrological time series matrix M(t) is constructed to drive the time-varying evolution of the initial graph structure G0, resulting in a spatiotemporal dynamic correlation graph structure G that can describe the changing laws of hydraulic connections and time-dependent variations. t ;

[0191] Extracting the spatiotemporal dynamic correlation graph structure G t Node embedding features h v (t), the process of node completion, anomaly detection, residual feedback and adaptive fusion is performed to realize multi-stage feature reconstruction and generate the dynamically completed optimized graph structure G. t *;

[0192] The optimized spatiotemporal graph structure G t The model is transformed into a high-dimensional embedding matrix E(t), and then subjected to tensor expansion and spatiotemporal normalization to generate a model input dataset T with a uniform spatiotemporal scale. t *

[0193] E(t) = [h1(t);h2(t);…;h N (t)]∈R N×d ;

[0194] T t *=Norm(Unfold(E(t)));

[0195] Where N is the number of nodes (spatial dimension), d is the node embedding dimension (feature dimension), Unfold is the tensor expansion function, and Norm is the spatiotemporal normalization function.

[0196] In this embodiment, G t * A node embedding E(t) of d=64 dimensionality is generated through graph embedding and expanded into a three-dimensional tensor structure with K=20 (5-day forecast period), N=17, and d=64. Spatiotemporal normalization is performed on each dimension to ensure consistency across different feature scales. The final obtained T t *Serves as a direct input to the parameterized model of the storage and release response.

[0197] According to one aspect of this application, a spatiotemporal dynamic correlation graph structure G is obtained that can describe the spatiotemporal dynamic relationship structure and the temporal dependence variation law of hydraulic connection. t Furthermore:

[0198] Data collected from surface hydrological stations (Sg), radar precipitation (R(x,t), remote sensing inversion (A(t)), and numerical weather prediction (F)). num The data (t) and the observation data are synchronized in time using a uniform sampling rate Δt, and spatial registration is performed through coordinate transformation and projection methods to form a multimodal hydrological time series matrix M(t);

[0199] M(t) = [Sg(t), R(x,t), A(t), F num (t)].

[0200] In this embodiment, hourly data from 17 surface hydrological stations in the Three Gorges-Luoshan section are selected as Sg(t). The radar quantitative precipitation estimation product is interpolated to a unified grid in the watershed at a resolution of 1km×1km as R(x,t). This is combined with the 500m resolution remote sensing water body inversion result A(t) and the global / regional numerical weather prediction F. num The rainfall forecast field (t) was used to construct a multimodal hydrological time series matrix M(t). Resampling was performed with a uniform step size of Δt=6h, and coordinate system unification was achieved using the Lambert projection coordinate system. Finally, a multimodal hydrological time series matrix M(t) covering the entire Three Gorges-Jingjiang-Dongting Lake area was formed, which solved the problem of inconsistent scales between different data sources.

[0201] An initial graph structure G0 is established based on the river connectivity, where node v represents a hydrological station or cross-section, and node feature vector x v (t) includes elements such as water level, flow rate, and rainfall, e vu This represents the hydraulic connection and spatial distance, based on the river channel distance d. vu Correlation with historical flow ρ vu (i.e., the Pearson correlation coefficient between the historical flow sequences of nodes v and u) Calculate the initial value w of the edge weight. vu .

[0202] w vu =exp(-d vu / σ d )·(1+ρ vu ) / 2.

[0203] In this embodiment, 17 key nodes, including the Three Gorges, Zhicheng, Shashi, Luoshan, and the inlets of each tributary, are incorporated into the graph structure, and the river distance d between any two nodes is calculated. vu Take the distance attenuation coefficient σ d A distance attenuation term is constructed at 80km. The Pearson correlation coefficient ρ is calculated using hourly flow data from 2013 to 2022. vu The edge weight w is generated by the formula. vu This forms the initial graph structure G0, in which the paths from Three Gorges to Zhicheng, Zhicheng to Shashi, and Dongting Lake to Luoshan are given higher weights due to their close hydraulic connections, which can more realistically reflect the flood propagation paths in the study area.

[0204] The graph structure is dynamically evolved using the multimodal hydrological time series matrix M(t) as a driving factor, and the node features x are updated. v (t) and edge weight w vu (t), thus obtaining the spatiotemporal dynamic correlation graph structure G. t .

[0205] w vu (t)=Softmax u (α·sim(x v (t),x u (t)));

[0206] Where sim is the similarity function and α is the self-attention coefficient.

[0207] In this embodiment, an attention mechanism is used to update edge weights based on node flow and water level changes, automatically increasing the attention of downstream nodes to upstream nodes during the intensified flood propagation phase. For example, during rising water levels, the similarity between the Three Gorges and Zhicheng nodes increases, causing w vu (t) Automatic enhancement, and when the high water level of Dongting Lake leads to an increased backwater effect, the influence of the Dongting Lake node on the Luoshan node also increases, ultimately yielding G that better reflects actual hydrodynamic conditions. t .

[0208] According to one aspect of this application, an optimized graph structure G with dynamic completion is generated. t *, further:

[0209] Extracting the spatiotemporal embedding features h of nodes v (t), utilizing graph message passing and neighborhood aggregation mechanisms to achieve G t Interpolation completion of missing nodes yields the node completion matrix X. v_hat (t);

[0210] h v (t)=Σ u∈N (v)wvu (t)·x u (t).

[0211] In this embodiment, the Qingjiang Gaobazhou station experienced short-term data gaps during some heavy rain periods. h was obtained through feature aggregation from neighboring stations such as Zhicheng and Yichang. v (t) enables smooth completion of missing test segments.

[0212] Complete the matrix X for nodes v_hat An outlier was identified by applying a sliding window residual threshold detection algorithm to the observed subset of (t), and error correction was performed using a weighted reconstruction method to obtain the outlier correction matrix X. v_tilde (t);

[0213] r v (t)=|X v_hat (t)-x v (t)|;

[0214] Where, r v (t) represents the residual between the node completion value and the original observation.

[0215] In this embodiment, residual detection with a window width of 12 hours is applied to Q(t) at Zhicheng Station. When the residual exceeds 3σ, it is marked as an anomaly. Neighborhood weighted reconstruction is used to effectively eliminate false spikes caused by radar rainfall false alarms. The corrected sequence is smoother and more in line with the laws of hydraulics.

[0216] A residual-gated feedback unit is introduced into the predicted subset of the node completion matrix to calculate the node prediction error e. v (t), and feed it back to the neighborhood to adjust the edge weights, resulting in the feedback correction matrix X. f v (t).

[0217] e v (t)=g(t)*(X v_hat (t)-X v (t-1));

[0218] Where g(t) is the gating coefficient.

[0219] In this embodiment, the residual is fed back to the relevant nodes to address the water level forecast deviation in the Dongting Lake area. This increases the weight of nodes closely related to the lake area (such as Chenglingji), thereby improving the ability to represent the slow response characteristics of the lake area in the graph.

[0220] Fusion Anomaly Correction Matrix X v_tilde (t) and feedback correction matrix X f v (t), and utilize edge weight w vuThe self-regulating mechanism of (t) achieves synergistic optimization of structure and features, forming a dynamically completed optimized graph structure G. t *

[0221] x v (t)=β·X v_tilde (t)+(1-β)·X f v (t);

[0222] Where β is the fusion weight coefficient.

[0223] In this embodiment, the fused diagram structure G t * It exhibits more stable timing characteristics on the critical transmission link of the Three Gorges-Luoshan, with reduced node noise and edge weight connectivity that better matches the hydrodynamic response.

[0224] According to one aspect of this application, a parameterized structural model of a condensed reservoir that can accurately characterize the storage and release response features of a condensed reservoir is constructed, further comprising:

[0225] The multi-level equivalent unit method, combined with watershed zoning clustering results, is used to dynamically aggregate the main stream and lake flood storage and detention areas, determine the adaptive aggregation boundary and equivalent dam site cross-section, and incorporate the upstream reservoir outflow sequence Q. out_u (t), Tributary inflow Q trib_k (t) and interval convergence Q int (t) is combined to obtain the total inflow sequence Q of the aggregated reservoir. in (t);

[0226] Q in (t)=∑Q out_u (t)+∑Q trib_k (t)+Q int (t).

[0227] In this embodiment, a multi-level equivalent unit method is used to dynamically aggregate the river channels and lake areas in the study area, and a hierarchical clustering method is used to determine the aggregation domain, so that the outflow from the Three Gorges Reservoir, the outflow from the Geheyan Reservoir, the inflow from tributaries, and the lake water are uniformly mapped into a whole. At the tail end of the aggregation domain, the equivalent dam site cross-section is automatically identified, its location corresponding to the Luoshan control cross-section, which is used as the unified outflow outlet of the aggregation reservoir. Total inflow Q in (t) is the outflow sequence Q from the Three Gorges and Geheyan Reservoirs out_u (t), the water inflow sequence of tributaries such as the Xiangjiang River, Zishui River, Yuanjiang River, and Lishui River Q trib_k (t) and the flow rate Q directly inflowed from rainfall in the Zhicheng-Luoshan section. int (t) is constructed, and time alignment is performed and summed with a uniform time step of Δt=6h.

[0228] Based on the model input dataset, an adaptive static structure of water level-storage capacity of the aggregated reservoir is established, and the response amplitude of the static framework is adjusted by the total inflow sequence Qin(t) as the driving force, forming a multidimensional dynamic evolution water level-storage capacity function V(Z) that couples water level Z(t), inflow Qin(t), time t and storage capacity V(t).

[0229] Based on the interactive coupling mechanism, multi-dimensional influencing factors such as time delay effect, backwater backwater and sedimentation evolution are integrated and coupled in a hierarchical parameterization manner to form a dynamic feedback system covering the time domain, spatial domain and morphological domain, and a water level-flow composite function Q(Z) with time-varying feedback characteristics is obtained.

[0230] By coupling the dynamic water level-storage capacity function and the composite water level-flow function in the parameter space, the overall expression of storage and release behavior is realized, forming a parameterized structural model M of the aggregated reservoir that characterizes the storage and release features of the aggregated reservoir.

[0231] M={V(Z),Q(Z)}.

[0232] In this embodiment, the dynamic water level-storage capacity expression V(Z) and the composite water level-flow function Q(Z) are parameterized to maintain a continuous relationship between the two within the same time step and water level range, thus forming a complete aggregated reservoir parameterized structure model for flood control calculation and parameter calibration.

[0233] According to one aspect of this application, a water level-flow composite function Q(Z) with time-varying feedback characteristics is obtained, further as follows:

[0234] Water level and flow rate features are extracted from the observation subset of the model input dataset. A power-law function-based water level-flow rate master relationship Q is constructed using a multi-scale nonlinear mapping method. b (Z);

[0235] Q b (Z)=c(Z') p ;

[0236] Where c and p are the scale and exponent parameters of the power-law master function to be calibrated, and Z'(t) is the water level (effective water depth) in the master single water level-flow relationship.

[0237] In this embodiment, water level and flow data for the Luoshan section from 2013 to 2022 were first collected to form the master control sample set. A double logarithmic linearization method was used to perform power-law regression on (Z,Q) to initially obtain the value ranges of parameters c and p. The power-law function constructed based on this can serve as the benchmark expression for subsequent lag correction, backwater feedback, and scour-siltation compensation, providing a consistent master control framework for the composite water level-flow relationship.

[0238] A time-series residual feedback channel is constructed in the water level-flow rate master control relationship, and the flow rate hysteresis gradient dQ is calculated. in A time-varying delay factor τ(t) is formed, and a time delay mapping is established to drive the dynamic update of the master control function, thus obtaining the delay correction model;

[0239] dQ in (t)=[Q in (t)-Q in [(t-Δt)] / Q in (t-Δt);

[0240] τ(t)=τ0(1+k1·dQ in );

[0241] Q(t-τ(t))=Qb(Z'(t));

[0242] Z'(t)=[Q(t-τ(t)) / c] 1 / p ;

[0243] Where τ0 is the baseline lag parameter to be calibrated, and k1 is the lag amplification factor to be calibrated.

[0244] Constructing a return water feedback function Z using the rate of change of water level at the control section dZ bk (t) characterizes the backwater effect of the downstream river level rise on the outflow process of the aggregate reservoir, and obtains the backwater feedback model.

[0245] dZ=[Z(t)-Z(t-Δt)] / [Z(t-Δt)-Z(t-2Δt)+ε];

[0246] Z bk (t)=Z bk0 (1+k2·dZ);

[0247] Where ε is a small constant to prevent the denominator from being zero (e.g., 10). -6 ), Z bk0 Z represents the reference strength parameter of the top support (reference top support water level) to be calibrated, and k2 is the amplification factor of the top support to be calibrated. For initial times such as t=0 and t=1, Z bk (t)=Z bk0 .

[0248] In this embodiment, the water level change rate of Luoshan itself is used to reflect the backwater effect of upstream river backwater and downstream backwater. A backwater feedback term based on the water level change rate is constructed, and the backwater intensity is controlled by parameter k2 to match the differentiated hydrodynamics of drought and flood stages.

[0249] Based on the characteristics of topographic evolution and sedimentation changes, a correction term ΔZ for scour and sedimentation migration is constructed. bedAdjust the water level-discharge pattern shift caused by topographic uplift or scouring to generate a scouring and deposition compensation model.

[0250] In this embodiment, it is assumed that ΔZ within the forecast period bed It is approximately unchanging over time and is used as a compensation term to correct long-term morphological shifts in the master control function.

[0251] By establishing a hierarchical parameterized coupled lag correction, top backing feedback and scouring and sedimentation compensation model, a dynamic feedback system driven by the interaction of the time domain, spatial domain and morphological domain is established, and a composite water level-flow function Q(Z) with time-varying feedback characteristics is obtained.

[0252] Z(t) = Z'(t) + Z bk (t)+Δh bed (t);

[0253] Q(Z)=c[Z(t)-Z bk (t)+Δh bed (t)] p ;

[0254] Z(t) = [Q(t - τ(t)) / c] 1 / p +Z bk (t)+Δh bed (t);

[0255] Z(t) is the measured water level, which consists of three parts: the single main control relationship, the water level affected by the backwater, and the water level compensated for scouring and silting.

[0256] In this embodiment, the main control water level and flow function is first constructed and a dynamic feedback term is superimposed so that the outflow at the Luoshan control section can synchronously respond to the time lag effect, the backwater effect and the scouring and sedimentation effect, thereby realizing the complete expression of the dynamic structure of the reservoir's storage and discharge response.

[0257] According to one aspect of this application, a computational framework for aggregated reservoir flood control applicable to complex river-lake systems is generated, further comprising:

[0258] Using the control section as a node, the storage and discharge equations and the water balance equations of the parameterized structural model M of the aggregated reservoir are combined to form a set of flood regulation calculation equations for the aggregated reservoir that can simultaneously describe storage and stagnation changes and hydrodynamic feedback.

[0259] V(t+1)-V(t)=(Q in (t+1)+Q in (t)-Q(t+1)-Q(t)) / 2·Δt;

[0260] Q(t+1) = Q(Z(t+1));

[0261] V(t+1)=V(Z(t+1)).

[0262] In this embodiment, the Luoshan control section is used as the solution node. The reservoir capacity-outflow correlation is established based on the dynamic reservoir capacity function V(Z) and the composite water level-flow function Q(Z) to characterize its storage and discharge equation. Combined with the water balance equation, a flood regulation calculation equation set is formed, providing a unified calculation framework for subsequent adaptive solution.

[0263] A time-delay state discriminator is constructed to dynamically monitor the time-delay factor τ(t) in the composite water level-flow function. Simultaneously, a dynamic reservoir capacity synchronous iterative algorithm and a prediction-correction hybrid algorithm are constructed. The solution algorithm is automatically switched based on the discriminator feedback information to form an adaptive solution mechanism.

[0264] Based on the adaptive solution mechanism, the information such as water level, flow rate, and reservoir capacity obtained at each moment is written back to the parameterized structural model of the aggregated reservoir, dynamically updating the internal parameters of the water level-reservoir capacity relationship and the water level-flow rate relationship, and simultaneously refreshing the boundary conditions for the next calculation, realizing the closed-loop update of flood control calculation, and forming the aggregated reservoir flood control calculation framework.

[0265] In this embodiment, the flood control calculation results Z(t+1), Q(t+1), and V(t+1) at each time step are written back to the aggregated reservoir parameterized structural model M. The model adaptively updates the parameters in the dynamic reservoir capacity function V(Z,t) and the composite water level-discharge function Q(Z,t) based on the new state variables, ensuring that the model continuously reflects the temporal changes in storage and outflow capacity throughout the entire flood process. The updated functions then serve as the boundary conditions and storage / discharge structure inputs for the next calculation step, achieving a rolling closed-loop update of the flood control calculation.

[0266] According to one aspect of this application, an adaptive solution mechanism is further formed as follows:

[0267] Real-time calculation of the lag factor τ(t) in the composite water level-flow function, and construction of a state discriminator. When τ(t) is below the stability threshold τ... s When τ(t) is determined to be in a static interval, it is considered to be in a static interval when τ(t) is higher than the fluctuation threshold τ. d When the time is determined to be a dynamic range, a time-delayed state signal is output.

[0268] In this embodiment, by monitoring the gradient of inflow changes, the time lag term in the composite water level-flow function is calculated, and based on empirical analysis of typical flood processes in the Luoshan River section, a static threshold τ is selected. s =1, dynamic threshold τ d =2. When τ(t) < τ s It is assumed that the water level and flow rate responses are approximately synchronized; when τ(t) > τ d It is assumed that there is a significant lag; for τ s ≤τ(t)≤τ d In cases where a time-delay state maintenance rule is used, frequent switching can be avoided.

[0269] A dynamic reservoir capacity iterative algorithm is constructed to simultaneously solve for water level, flow rate, and reservoir capacity variables in the static interval, ensuring the convergence of the storage and release equation. A prediction-correction hybrid algorithm is also constructed to predict future water level trends using the lagging flow rate in the significant lag interval. Correction is performed based on the water level-storage capacity function and the water balance equation, and the real-time flow rate is solved. The two form a heterogeneous and complementary dual algorithm channel.

[0270] In this embodiment, the synchronous iterative algorithm first assumes that the initial water level at time t+1 is equal to the water level at time t: Z (k=0) (t+1)=Z(t), calculate the flow rate Q at time t+1 based on the water level-flow equation. (k) (t+1)=Q(Z (k) Substituting (t+1) into the water balance equation, we can calculate the reservoir capacity V at time t+1. (k) (t+1), the water level Z at time t+1 is calculated by inverse calculation using the dynamic reservoir capacity function. (k+1) (t+1), assuming water level Z (k) With reverse calculation of water level Z (k+1) When the accuracy requirements are met, the flood control calculation is completed at time t+1, and the calculation continues along the time scale until the end of the forecast period; otherwise, the water level Z is calculated in reverse. (k+1) Recalculate the flow rate Q (k+1) (t+1), continue iteratively to solve for the water level and flow rate at time t+1. The prediction-correction rule, based on the measured or predicted flow rate at time t+1-τ, derives the water level Z(t+1)=g at time t+1. -1 (Q(t+1-τ)) is used to derive the flow rate Q(t+1) at time t+1 based on the water balance equation, avoiding iterative solutions and exhibiting greater stability in the significant lag phase.

[0271] Based on heterogeneous complementary dual-algorithm channels, when the lag state is in a static interval, the dynamic capacity iteration algorithm channel is activated, and when it is in a dynamic interval, the prediction-correction hybrid algorithm channel is enabled. The two are dynamically switched in the time series, forming a dual-channel adaptive solution mechanism that is sensitive to lag changes.

[0272] According to one aspect of this application, generating information on water level processes, flow processes, and flood peak characteristics over a future forecast period, further includes:

[0273] Based on the principle of error minimization, a dynamic weight allocation mechanism is designed, taking into account the simulated water level value Z. sim (t) and measured value Z obs (t), simulated flow value Q sim (t) and measured value Q obs The fitting effect of (t) is used to form a multi-index comprehensive objective function J that simultaneously measures the fitting performance of water level and flow rate.

[0274] J=∑t=1 T [ω Z (t)(Z sim (t)-Z obs (t)) 2 +ω Q (t)(Q sim (t)-Q obs (t)) 2 ];

[0275] Where, ω Z (t) represents the weight of the water level fitting effect; ω Q (t) represents the weight of the flow fitting effect; the two satisfy ω Z (t)+ω Q (t)=1.

[0276] In this embodiment, a dynamic weighting mechanism is used to adjust the weights of the water level and flow rate fitting effects. When calibrating the water level-reservoir capacity function alone, the flow rate weight ω... Q (t) is set to 0, and the water level weight ω Z (t) is set to 1; water level weight ω is used when calibrating the water level-flow function alone. Z (t) is set to 0, and the flow weight ω Q (t) is 1;

[0277] Based on a multi-index comprehensive objective function J, a hybrid global-local intelligent optimization algorithm is used to automatically calibrate the parameterized structural model of the aggregated reservoir. The optimal parameter combination Θ is obtained through a phased parameter decoupling and global collaborative convergence strategy. * ;

[0278] Based on the optimal parameter combination Θ * By using the latest observational data to assimilate and correct the internal parameters and state variables of the model, the parameterized structural model of the aggregated reservoir can reflect the latest state of flood evolution in real time. With the updated model as the core execution environment, the storage and discharge equations and flood regulation calculations are advanced hourly to achieve rolling forecasts of water level processes, flow processes and flood peak characteristics in the future forecast period.

[0279] According to one aspect of this application, the optimal parameter combination Θ is obtained. * Further:

[0280] The high-dimensional parameter set Θ of the aggregated reservoir parameterized model structure is mapped in stages into two independent subsets of reservoir capacity curve parameters Θ. V and the subset of flow curve parameters Θ Q After decoupling, optimization is performed separately in the wide-domain boundary of the low-dimensional space to obtain the stable parameter range dominated by a single physical mechanism.

[0281] In this embodiment, the high-dimensional parameter space is first split into a subset Θ of parameters related to the storage capacity curve. V A subset of parameters related to water level-flow rate Θ Q The parameters of the time-period measured water level-driven water level-flow sub-model were optimized and calibrated. Q ;Parameter optimization and calibration of the reservoir capacity sub-model driven by measured flow rate Θ V This low-dimensional independent calibration avoids the curse of dimensionality in high-dimensional spaces, significantly stabilizes the optimization process, and provides reliable initial values ​​and a reasonable range of physical parameters for subsequent co-optimization.

[0282] Based on the statistical analysis of the value ranges of each parameter subset using multiple independent calibration results, and through a perturbation expansion strategy, the original wide-domain boundary is replaced with a parameter space that is significantly narrowed and covers the range of real physical changes. Global collaborative optimization is then performed on the high-dimensional parameter set Θ in the narrow-domain solution space.

[0283] In the global collaborative optimization process, a dual convergence criterion is set, and the improvement rate of the objective function and the stability of the population parameters are monitored at the same time. When the dual convergence condition is met, it indicates that the multi-parameter collaborative convergence is achieved, and the global optimal parameter combination is output.

[0284] In this embodiment, a convergence threshold ε for the objective function is set. J =10 -1 With parameter change threshold ε Θ =10 -2 In actual operation, the coordinated calibration typically converges stably after 120-300 iterations. The final parameters remain consistent in independent test floods, and the overall forecast error of the model is significantly improved in both time and accuracy compared to the initial wide-range coordinated calibration.

[0285] This application employs a parametric dynamic modeling approach driven by inflow characteristics. By decomposing the total inflow sequence of a reservoir into trend and disturbance terms, time-varying driving factors and nonlinear enhancement operators are constructed to dynamically modulate the parameter weights of the static water level-storage capacity polynomial in real time. This approach transforms the storage capacity function from a rigid single-valued curve into one that adaptively deforms with the flood rise and fall rate, accurately capturing the additional gradient and wedge-shaped storage changes that traditional static models cannot describe, effectively solving the problem of inaccurate quantification of dynamic storage capacity. It also addresses the systematic bias in dynamic storage capacity calculations caused by the rigid representation of storage and discharge structures (i.e., the fixed water level-storage capacity curve cannot reflect the dynamic effects of floods).

[0286] This application employs a dual strategy of physical structure parameterization and adaptive algorithm switching. At the physical level, a composite water level-discharge relationship is constructed, integrating a time-varying lag factor (based on inflow gradient) and a backwater backwater feedback term (based on water level acceleration), explicitly characterizing the heterogeneous hysteresis loop features of water level and discharge. At the computational level, an adaptive solution mechanism incorporating a lag state discriminator is established. This mechanism can automatically switch between a dynamic reservoir capacity synchronous iterative algorithm (suitable for steady state) and a prediction-correction hybrid algorithm (suitable for dynamic conditions) based on the degree of lag in real-time monitoring. This approach effectively avoids the oscillation risk of iterative methods in the strongly nonlinear lag interval, ensuring the convergence and water balance of the flood control calculation equations under drastically changing boundaries, and improving the numerical stability of the forecast. It solves the problem of unstable solution for non-co-phase evolution processes under complex boundaries (i.e., the tendency of traditional single algorithms to diverge in significant lag intervals).

[0287] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for intelligent forecasting of flood evolution in river and lake regulation systems based on dynamic storage-discharge relationships, characterized in that, include: A spatiotemporal dynamic correlation graph structure reflecting the hydraulic connections of the watershed is constructed. The collected multi-source hydrological data is mapped to the spatiotemporal dynamic correlation graph structure. Through node feature aggregation and edge weight update, a model input dataset with a unified spatiotemporal scale is generated. Based on the inflow change features extracted from the model input dataset, a parameterized structural model of the aggregated reservoir is established to characterize the overall storage and discharge features of the main stream and the lake. It includes a dynamic water level-storage capacity relationship driven by the inflow change features in the model input dataset, as well as a composite water level-discharge relationship integrating the lag effect and backwater backwater feedback. A flood control calculation equation set is constructed based on the parameterized structural model of the aggregated reservoir. Based on the real-time lag state identified from the composite water level-discharge relationship, an adaptive solution mechanism is used to solve the flood control calculation equation set to obtain the calculation state variables. Based on real-time observation data, the parameterized structural model of the aggregated reservoir is corrected. Rolling prediction is performed using the corrected model and calculated state variables to output water level and flow process information for future periods. Establish a parameterized structural model of the aggregated reservoir to characterize the overall storage and discharge characteristics of the main stream and lake, including: The multi-level equivalent unit method is used to dynamically aggregate the main river channel and lake flood storage and detention areas, determine the adaptive aggregation boundary and equivalent dam site section, and generalize the water body within the adaptive aggregation boundary into aggregated reservoir units. Based on the model input dataset, the outflow from the upstream reservoir, the inflow from the tributaries, and the confluence flow between the sections are extracted and merged into a total inflow sequence of the reservoir. Using the total inflow sequence of the aggregated reservoir as the driving variable, the storage and release relationship of the aggregated reservoir unit is constructed; The dynamic water level-reservoir capacity relationship is constructed through the following steps: Based on historical observation data in the model input dataset, an adaptive static structure of water level and reservoir capacity is generated by fitting. Signal decomposition was performed on the total inflow sequence of the Juhe Reservoir to obtain a trend term reflecting the gradual change in inflow and a disturbance term reflecting the abrupt change in inflow. A time-varying driving factor controlling the rate of change of storage capacity is constructed using a trend term, and a nonlinear enhancement operator controlling the shape correction of the control curve is constructed using a disturbance term. By using time-varying driving factors and nonlinear enhancement operators, the parameter weights of the water level-storage capacity adaptive static structure are adjusted in real time to generate a dynamic water level-storage capacity relationship that evolves dynamically with the inflow process. The composite water level-flow rate relationship is constructed through the following steps: Water level and flow characteristics are extracted from the observation subset of the model input dataset, and a water level-flow control relationship based on a power law function is constructed through multi-scale nonlinear mapping. Based on the water level-discharge control relationship, a time lag correction model representing the flood propagation time delay, a top-feedback feedback model representing the downstream water level's inhibitory effect on outflow, and a scouring and deposition compensation model representing changes in riverbed topography are constructed respectively. By using a hierarchical parameterized coupling mechanism, the time-delay correction model, the top-support feedback model, and the scouring and sedimentation compensation model are superimposed on the water level-flow master control relationship, thus establishing a composite water level-flow relationship driven by the interaction of the time domain, spatial domain, and morphological domain.

2. The method according to claim 1, characterized in that, A spatiotemporal dynamic correlation graph structure reflecting the hydraulic connections of the watershed is constructed, and a model input dataset with a unified spatiotemporal scale is generated, including: The collected multi-source hydrological data were synchronized in time and registered in space to construct a multimodal hydrological time series matrix and establish an initial graph structure based on the physical connectivity of the river channels. The initial graph structure is driven by a multimodal hydrological time series matrix to undergo time-varying evolution. The edge weights are updated by calculating the dynamic similarity between nodes to obtain a spatiotemporally dynamic relational graph structure. Extract the node embedding features of the spatiotemporal dynamic relational graph structure, perform node missing completion, anomaly detection and residual feedback correction to obtain an optimized graph structure containing reconstruction features; The optimized graph structure is transformed into a high-dimensional embedding matrix, and tensor expansion and spatiotemporal normalization are performed to generate the model input dataset.

3. The method according to claim 1, characterized in that, The adaptive static structure of water level-reservoir capacity is represented by a third-order polynomial; the specific calculation expression for the dynamic water level-reservoir capacity relationship is as follows: ； Where Z is the water level at the current time t, V(Z) is the dynamic reservoir capacity, and a0, a1, a2, a3 are the coefficients of the static polynomial; α(t) is a time-varying driving factor, defined as the ratio of the trend term to the historical maximum inflow; B1, B2, B3, and B4 are the nonlinear enhancement operators corresponding to each order of terms, defined as the product of the perturbation term and the preset enhancement operator coefficients.

4. The method according to claim 1, characterized in that, The time-delay correction model is constructed through the following steps: Calculate the time rate of change of the total inflow sequence of the aggregated reservoir to obtain the inflow lag gradient; The preset baseline lag parameters are nonlinearly and dynamically amplified based on the inflow lag gradient to generate a time-varying lag factor. By utilizing time-varying lag factors, a time-series residual feedback channel is constructed in the water level-flow control relationship, and a correction structure is established to drive the water level-flow control relationship to dynamically map the time delay as the inflow changes. The specific formula for calculating the time-varying lag factor is: τ(t) = τ0 * (1 + k1 * dQ) in ); Where τ(t) is the time-varying lag factor, τ0 is the baseline lag parameter to be calibrated, k1 is the lag amplification factor to be calibrated, and dQ in The inflow hysteresis gradient; Dynamic time delay mapping is achieved by introducing a time offset term Q(t-τ(t)) into the water level-flow master relationship.

5. The method according to claim 1, characterized in that, The top-support feedback model is constructed through the following steps: Obtain real-time water level data of the equivalent dam site cross section and calculate its rate of change over time; Using the rate of change in water level as the driving variable, a backwater feedback function is constructed to characterize the degree of backwater in the downstream river channel; The return water feedback function is superimposed as a negative feedback term onto the water level variable in the composite water level-flow relationship to compensate for the backwater effect of downstream water level rise on the outflow process.

6. The method according to claim 4, characterized in that, The adaptive solution mechanism is constructed through the following steps: Construct a time-delay state discriminator to monitor the time-varying time-delay factor in the composite water level-discharge relationship in real time and output the real-time time-delay state that reflects the current flood response characteristics; A dynamic reservoir capacity synchronous iterative algorithm suitable for the synchronous response stage of water level and flow rate is established in parallel, and a prediction-correction hybrid algorithm suitable for the heterogeneous response stage of water level and flow rate is established, forming a heterogeneous and complementary dual algorithm channel. Based on the real-time delay state, the system automatically switches between the two algorithm channels and selects the algorithm path that matches the current response characteristics to solve the flood control calculation equations.

7. The method according to claim 6, characterized in that, The specific execution process of the adaptive solution mechanism includes: The time-varying lag factor is compared with the stability threshold and the fluctuation threshold using a lag state discriminator. When the time-varying lag factor is lower than the stability threshold, the real-time lag state is determined to be in the static range. When it is higher than the fluctuation threshold, it is determined to be in the dynamic range. When in the static interval, activate the dynamic reservoir capacity synchronous iterative algorithm, and solve the reservoir capacity, water level and flow variables simultaneously until the convergence condition is met. When in the dynamic range, the prediction-correction hybrid algorithm is activated, which uses historical flow sequences containing lag information to predict future water level response trends, and performs correction solutions based on the dynamic water level-storage capacity relationship and water balance equation.

Images