Adaptive neural network flood routing simulation and risk assessment system and method

Abstract

The invention discloses an adaptive neural network flood routing simulation and risk assessment method. The method comprises the following steps: step 1, generating multi-source hydrometeorological preprocessing data; 2, constructing a river network topological graph, generating node feature vectors and edge feature vectors, and forming time-space diagram input data; 3, inputting the liquid state time constant neural network to execute continuous time state updating, and generating a node hydrological state prediction vector set; step 4, injecting hydrodynamic physical consistency residual errors to generate a physical constraint hydrological state prediction vector set; 5, executing adaptive lag compensation, and generating a lag compensation hydrological state prediction vector set; 6, mapping to generate flow, water level and flood peak arrival time prediction data; and step 7, outputting a risk grading graph and early warning threshold list data. According to the invention, accurate prediction of the flood process and dynamic generation of the risk map are realized, and early warning precision and response efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of flood risk assessment, and in particular to an adaptive neural network flood evolution simulation and risk assessment system and method. Background Art

[0002] Existing flood evolution prediction and risk assessment methods mostly rely on traditional hydrological models and hydrodynamic models, and are often driven by a single data source. For example, only data from ground rain gauges and water level stations are used for simulation. In basins with complex river networks and large differences in the temporal and spatial distribution of rainfall, such methods are difficult to accurately reflect the impact of upstream water inflow on downstream flood peaks. The parameters of traditional models are mostly fixed values, and they are not adaptable enough to changes in basin conditions. The time lag effect of the rainfall-runoff relationship and the response ability to the rapid change stage of the flood peak are weak, which can easily lead to large deviations in flood peak flow predictions and delayed peak time.

[0003] In terms of spatial expression, existing methods have limited structural modeling capabilities for basin observation sections and river sections, the construction of node and edge features is incomplete, and it is difficult to effectively map multi-source hydrological and meteorological data to spatiotemporal feature expressions reflecting hydrological processes. Rule-based risk assessment methods rely on fixed thresholds and lack dynamic coupling with real-time prediction results, resulting in flood risk classification that lacks specificity and timeliness. The application of existing deep learning models in flood evolution prediction mostly focuses on sequence prediction, lacks integration with hydrodynamic constraints, and the prediction results have deviations in physical consistency. The model's adaptability to rainfall mutations and stability handling of receding water sections are insufficient.

[0004] In dealing with hysteresis effects, conventional methods mostly rely on static corrections and are unable to make adaptive adjustments based on the real-time status of the basin, resulting in difficulty in controlling prediction errors during the flood peak formation and propagation stages. At the risk output level, most methods fail to spatially superimpose flood evolution prediction results with the exposure of basin infrastructure, lack a refined assessment of the impact range of key facilities, and lack the ability to dynamically generate warning threshold lists. These problems limit the accuracy of flood forecasts, response speed, and the scientific nature of risk assessments when dealing with complex hydrological scenarios.

[0005] Therefore, how to provide an adaptive neural network flood evolution simulation and risk assessment system and method is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One purpose of the present invention is to propose an adaptive neural network flood evolution simulation and risk assessment system and method. The present invention integrates multi-source hydrological and meteorological data, river network topology modeling and hydrodynamic physical consistency constraints, combines adaptive lag compensation and dynamic risk classification, and realizes accurate prediction of flood processes and generation of risk maps. It has the advantages of high prediction accuracy, fast response speed and strong interpretability of risk assessment results.

[0007] The adaptive neural network flood evolution simulation and risk assessment method according to an embodiment of the present invention includes the following steps: Step 1: Collect basin rainfall observation data and river water level observation data, and pre-process them to obtain multi-source hydrological and meteorological pre-processed data; Step 2: Construct a river network topology map based on multi-source hydrological and meteorological preprocessed data, construct the observation section as a node feature vector, and construct the river section as an edge feature vector to form the spatiotemporal graph input data; Step 3: Input the spatiotemporal graph input data into the liquid time constant neural network embedded with the graph attention mechanism, perform continuous time state update, and generate a set of node hydrological state prediction vectors; Step 4: inject the hydrodynamic physical consistency residual into the node hydrological state prediction vector set to generate the physically constrained hydrological state prediction vector set; Step 5: performing adaptive hysteresis compensation on the set of physically constrained hydrological state prediction vectors to generate a set of hysteresis-compensated hydrological state prediction vectors; Step 6: Map the hysteresis-compensated hydrological state prediction vector set into flow prediction data, water level prediction data, and flood peak arrival time prediction data to form flood evolution prediction result data; Step 7. Based on the flood evolution prediction results data, calculate the inundation depth data, flow velocity data and arrival time data under different annual exceedance probabilities, combine the spatial data of the basin infrastructure to generate flood risk level data and key facility exposure data, and output flood risk classification map data and flood warning threshold list data.

[0008] Optionally, step one includes: performing time series missing value filling, outlier removal and time granularity unified processing on the basin rainfall observation data to generate rainfall preprocessing data; performing time series missing value filling, outlier removal and station benchmark unified processing on the river water level observation data to generate water level preprocessing data; synchronously aligning the rainfall preprocessing data and the water level preprocessing data according to a unified time benchmark to construct multi-source hydrological and meteorological preprocessing data.

[0009] Optionally, step two includes: extracting water level preprocessing data and time index corresponding to each observation section from multi-source hydrological and meteorological preprocessing data, combining the water level preprocessing data of the same observation section under continuous time index in sequence to construct a node feature vector; extracting rainfall preprocessing data, roughness data and slope data corresponding to each river section from multi-source hydrological and meteorological preprocessing data, combining the data of the same river section under continuous time index in sequence to construct an edge feature vector; combining the node feature vector and the edge feature vector according to the upstream and downstream relationship of the river section connection to generate a river network topology map; arranging all node feature vectors and edge feature vectors in the river network topology map according to the time index to form a space-time map input data.

[0010] Optionally, the processing of the liquid time constant neural network of the embedded graph attention mechanism in step 3 includes: Arrange the spatiotemporal graph input data in ascending order by time index, extract the node feature vectors and edge feature vectors at each time index, initialize the learnable time constant parameters for each node, perform normalization based on the range of the node feature vector to obtain the initial value of the time constant, initialize the graph attention parameters for each edge, bind them to the edge feature vector corresponding to the edge, and use the node feature vector at the first time index as the initial node hydrological state prediction vector; Under the current time index, the current node feature vector is connected with the node hydrological state prediction vector of the previous time index to form a query vector. For each incoming edge connected to the node, the node hydrological state prediction vector of the previous time index of the adjacent node is connected with the edge feature vector of the incoming edge to form a key vector and a corresponding value vector. Perform a dot product operation on the query vector and the key vector along the corresponding dimensions to obtain a relevance score. Multiply the relevance score by the graph attention parameter of the incoming edge element-by-element, and add the result after linear transformation of the edge feature vector to obtain an unnormalized attention score. Apply the graph attention parameter of the corresponding edge to the attention score, normalize the scores of all incoming edges of the same node to obtain a set of attention weights, and perform a weighted sum of the corresponding value vectors with the attention weights to generate a graph attention aggregation vector. The graph attention aggregation vector and the current node feature vector are connected and input into the liquid time constant unit. The update coefficient is calculated based on the time constant parameter and the time index interval. The hydrological state prediction vector of the node at the previous time index and the candidate node state vector are weighted updated to obtain the current node hydrological state prediction vector. The calculation is repeated on all time indexes, and the node hydrological state prediction vectors of each time index are collected to form a node hydrological state prediction vector set.

[0011] Optionally, the step 4 includes: At each time index, the hydrological state prediction vector of the current node and the hydrological state prediction vector of the adjacent nodes are read from the node hydrological state prediction vector set. Combined with the corresponding edge feature vectors, the inflow and outflow of the current node are summed up within the same time index, and the difference between the sum of the inflow and the sum of the outflow is calculated to obtain the continuity equation residual. At each time index, the time index interval is combined based on the difference between the hydrological state prediction vector of the current node and the hydrological state prediction vector of the previous time index node, and the momentum equation residual is calculated by combining the water level gradient along the river channel and the friction slope determined by the roughness and flow velocity. At each time index, the residuals of the continuity equation and the momentum equation are weighted according to the adaptive weights to synthesize the hydrodynamic physical consistency residual vector. The adaptive weights are determined by the monotonic mapping of the residual amplitude, the time index interval and the time constant parameter. At each time index, a linear correction is performed on the hydrological state prediction vector of the current node based on the hydrodynamic physical consistency residual vector and the residual injection gain coefficient to generate a physically constrained hydrological state prediction vector. The residual injection gain coefficient is determined jointly by the time constant parameter and the time index interval and is limited to a preset closed interval. Repeat the process for all time indexes, collect the physically constrained hydrological state prediction vectors according to the time index, and form a set of physically constrained hydrological state prediction vectors.

[0012] Optionally, the step five includes: At each time index, the physical constraint hydrological state prediction vector of the current node and the physical constraint hydrological state prediction vectors of several historical time indexes are extracted from the set of physical constraint hydrological state prediction vectors. The cumulative curve of the corresponding rainfall process and the cumulative curve of the flow process are calculated. The time difference between the rainfall peak and the flow peak is determined based on the position difference of the characteristic inflection points of the two cumulative curves, and the time lag estimate of the current node is generated. At each time index, the time lag estimate and the corresponding time constant parameter of the current node are extracted, and the time constant parameter of the previous time index is extracted. The time lag estimate and the time constant parameter are normalized to obtain a standardized time lag amount and a standardized time constant. The change rate of the time constant parameter between adjacent time indexes is calculated to obtain a time constant change amount. A weight pair is determined based on the standardized time lag amount and the time constant change amount so that the sum of the weight pair is one. The standardized time lag amount and the standardized time constant are linearly fused according to the weight pair to generate an unrestricted correction coefficient. The unrestricted correction coefficient is monotonically normalized and clipped with upper and lower bounds, and the result is restricted to a preset closed interval to obtain an adaptive time lag correction coefficient. At each time index, the alignment position of the current node's physical constraint hydrological state prediction vector on the time axis is adjusted according to the adaptive time lag correction coefficient. Interpolation updates are performed on the hydrological state prediction vector of the node that needs to be moved forward, and delayed smoothing updates are performed on the hydrological state prediction vector of the node that needs to be moved backward. At each time index, the node hydrological state prediction vector after adaptive lag compensation is output, and the output vectors of all nodes are collected according to the time index to form a lag-compensated hydrological state prediction vector set.

[0013] Optionally, the flood risk classification map data and flood warning threshold list data specifically include: Extract flow prediction data, water level prediction data, and flood peak arrival time prediction data from the flood evolution prediction result data obtained by mapping the hysteresis-compensated hydrological state prediction vector set. Combined with the corresponding rainfall and basin infrastructure spatial data, calculate the inundation depth data, flow velocity data, and arrival time data of each grid cell. The flooding depth data, flow velocity data, and arrival time data of each grid cell are matched with a preset risk classification condition set. The risk classification condition set consists of a flooding depth threshold, a flow velocity threshold, and an arrival time threshold, and is divided into four levels: When the inundation depth is greater than or equal to three meters and the flow velocity is greater than or equal to two meters per second and the arrival time is less than or equal to one hour, the grid cell is marked as a level one risk; When the inundation depth is between two and three meters, the flow velocity is between one and two meters per second, or the arrival time is between one and three hours, and the first-level risk conditions are not met, the grid cell is marked as a second-level risk; When the inundation depth is between 0.5 and 2 meters, or the flow velocity is between 0.5 and 1 meter per second, or the arrival time is between 3 and 6 hours, and the risk conditions of level 1 and level 2 are not met, the grid cell is marked as level 3 risk; When the inundation depth is less than 0.5 meters and the flow velocity is less than 0.5 meters per second and the arrival time is greater than six hours, the grid cell is marked as low risk; At each time index, the risk level labels of all grid cells are synthesized into flood risk level data, the flood risk level data and key facility exposure data are spatially superimposed, the types and quantities of key facilities covered under each risk level are extracted, and the flood warning threshold list data is generated and output.

[0014] The adaptive neural network flood evolution simulation and risk assessment system according to an embodiment of the present invention includes the following modules: Data acquisition and preprocessing module, used to collect basin rainfall observation data and river water level observation data and generate multi-source hydrological and meteorological preprocessed data; The river network topology construction module is used to convert multi-source hydrological and meteorological pre-processed data into node feature vectors and edge feature vectors and form spatiotemporal graph input data; Liquid time constant neural network processing module, used to input spatiotemporal graph data and perform continuous time state update to generate a set of node hydrological state prediction vectors; A physical consistency residual injection module is used to inject hydrodynamic physical consistency residuals into the node hydrological state prediction vector set to generate a physical constraint hydrological state prediction vector set; An adaptive hysteresis compensation module, configured to perform adaptive hysteresis compensation on a set of physically constrained hydrological state prediction vectors to generate a set of hysteresis-compensated hydrological state prediction vectors; The flood evolution result generation module is used to map the hysteresis-compensated hydrological state prediction vector set into flow prediction data, water level prediction data and flood peak arrival time prediction data; The risk assessment and warning generation module is used to generate flood risk classification map data and flood warning threshold list data.

[0015] The beneficial effects of the present invention are: (1) By constructing multi-source hydrological and meteorological preprocessing data and river network topology maps, the observed sections and river segments are converted into node feature vectors and edge feature vectors. A liquid time constant neural network with an embedded graph attention mechanism is introduced to achieve continuous time update prediction of the basin's hydrological status, improve the prediction accuracy of flood peak flow and peak appearance time, and reduce calculation deviations under complex river network conditions.

[0016] (2) The hydrodynamic physical consistency residual is injected into the node hydrological state prediction vector set, and combined with the adaptive lag compensation mechanism, the rainfall-runoff time lag is dynamically corrected to improve the response speed and stability of flood evolution prediction during the rainfall mutation and rapid flood peak formation stages, and to enhance the smoothness and physical rationality of the water receding section prediction.

[0017] (3) The prediction results after lag compensation are converted into flow, water level and flood peak arrival time data. Combined with the spatial data of the basin infrastructure, the inundation depth, flow velocity and arrival time are calculated based on the exceedance probability in different years. The flood risk is graded according to multiple condition thresholds, and a flood warning threshold list is dynamically generated to improve the pertinence of risk assessment and the feasibility of warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is the overall flow chart of the adaptive neural network flood evolution simulation and risk assessment method proposed in the present invention; Figure 2 This is a schematic diagram of the hydrodynamic physical consistency residual injection and adaptive hysteresis compensation process proposed in the present invention. DETAILED DESCRIPTION

[0019] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0020] refer to Figure 1 and Figure 2 , an adaptive neural network flood evolution simulation and risk assessment system and method, comprising the following steps: Step 1: Collect basin rainfall observation data and river water level observation data, and pre-process them to obtain multi-source hydrological and meteorological pre-processed data; Step 2: Construct a river network topology map based on multi-source hydrological and meteorological preprocessed data, construct the observation section as a node feature vector, and construct the river section as an edge feature vector to form the spatiotemporal graph input data; Step 3: Input the spatiotemporal graph input data into the liquid time constant neural network embedded with the graph attention mechanism, perform continuous time state update, and generate a set of node hydrological state prediction vectors; Step 4: inject the hydrodynamic physical consistency residual into the node hydrological state prediction vector set to generate the physically constrained hydrological state prediction vector set; Step 5: performing adaptive hysteresis compensation on the set of physically constrained hydrological state prediction vectors to generate a set of hysteresis-compensated hydrological state prediction vectors; Step 6: Map the hysteresis-compensated hydrological state prediction vector set into flow prediction data, water level prediction data, and flood peak arrival time prediction data to form flood evolution prediction result data; Step 7. Based on the flood evolution prediction results data, calculate the inundation depth data, flow velocity data and arrival time data under different annual exceedance probabilities, combine the spatial data of the basin infrastructure to generate flood risk level data and key facility exposure data, and output flood risk classification map data and flood warning threshold list data.

[0021] In this embodiment, step one includes: performing time series missing value filling, outlier removal and time granularity unified processing on the basin rainfall observation data to generate rainfall preprocessing data; performing time series missing value filling, outlier removal and station benchmark unified processing on the river water level observation data to generate water level preprocessing data; synchronously aligning the rainfall preprocessing data and the water level preprocessing data according to a unified time benchmark to construct multi-source hydrological and meteorological preprocessing data.

[0022] In this embodiment, step two includes: extracting water level preprocessing data and time index corresponding to each observation section from multi-source hydrological and meteorological preprocessing data, combining the water level preprocessing data of the same observation section under continuous time index in sequence to construct a node feature vector; extracting rainfall preprocessing data, roughness data and slope data corresponding to each river section from multi-source hydrological and meteorological preprocessing data, combining the data of the same river section under continuous time index in sequence to construct an edge feature vector; combining the node feature vector and the edge feature vector according to the upstream and downstream relationship of the river section to generate a river network topology map; arranging all node feature vectors and edge feature vectors in the river network topology map according to the time index to form a spatiotemporal graph input data.

[0023] In this embodiment, the processing of the liquid time constant neural network embedded in the graph attention mechanism in step 3 includes: Arrange the spatiotemporal graph input data in ascending order by time index, extract the node feature vectors and edge feature vectors at each time index, initialize the learnable time constant parameters for each node, perform normalization based on the range of the node feature vector to obtain the initial value of the time constant, initialize the graph attention parameters for each edge, bind them to the edge feature vector corresponding to the edge, and use the node feature vector at the first time index as the initial node hydrological state prediction vector; Under the current time index, the current node feature vector is connected with the node hydrological state prediction vector of the previous time index to form a query vector. For each incoming edge connected to the node, the node hydrological state prediction vector of the previous time index of the adjacent node is connected with the edge feature vector of the incoming edge to form a key vector and a corresponding value vector. Perform a dot product operation on the query vector and the key vector along the corresponding dimensions to obtain a relevance score. Multiply the relevance score by the graph attention parameter of the incoming edge element-by-element, and add the result after linear transformation of the edge feature vector to obtain an unnormalized attention score. Apply the graph attention parameter of the corresponding edge to the attention score, normalize the scores of all incoming edges of the same node to obtain a set of attention weights, and perform a weighted sum of the corresponding value vectors with the attention weights to generate a graph attention aggregation vector. Connect the graph attention aggregation vector and the current node feature vector to the liquid time constant unit, calculate the update coefficient based on the time constant parameter and the time index interval, perform weighted update on the previous time index node hydrological state prediction vector and the candidate node state vector, and obtain the current node hydrological state prediction vector. Repeat the calculation on all time indexes, and collect the node hydrological state prediction vectors of each time index to form a node hydrological state prediction vector set; specifically including: Connect the graph attention aggregation vector and the current node feature vector into the liquid time constant unit, and for each node at the time index Based on: ; in, Represents the node index, corresponding to the observation section in the river network topology diagram, Represents the time index, Represents the node feature vector at the current time index, represents the graph attention aggregation vector, Represents a vector concatenation operation, represents the learnable weight vector, represents the learnable bias scalar, represents a smooth non-negative function, represents a small positive constant used for numerical stability; Compute the time constant parameter using the attention-weighted incoming edge feature vector: ; in, Representation node The set of incoming adjacent nodes, Indicates the upstream adjacent node index, Represents a slave node Pointing to a node The attention weights are non-negative and sum to 1; : Slave node Pointing to a node The edge eigenvector of ; Get the modulation factor: ; in, represents the linear rectification activation function, represents the learnable weight vector, represents the learnable bias scalar, represents the modulation factor calculated from the edge feature aggregation result; This forms the effective time constant: ; in, Represents the effective time constant after modulation; Based on the interval between adjacent time indexes Calculate the continuous-time update coefficients: ; in, Indicates the interval between the current and previous time indexes. represents the original continuous-time update coefficient; Through bounded mapping: ; in, represents the lower and upper bounds of the update coefficients, represents the reference center value, Represents the temperature coefficient, a positive real number, represents the Sigmoid function, Update coefficient indicating numerical stability; The hydrological state prediction vector of the index node at the previous time and the candidate node state vector renew:

[0024] in, Represents the node hydrological state prediction vector of the previous time index, represents the candidate node state vector generated by the liquid time constant unit, The node hydrological state prediction vector representing the current time index; Perform weighted update to obtain the current node hydrological state prediction vector, repeat the calculation on all time indexes, and collect the node hydrological state prediction vectors of each time index to form a node hydrological state prediction vector set.

[0025] In this embodiment, the step 4 includes: At each time index, the hydrological state prediction vector of the current node and the hydrological state prediction vector of the adjacent nodes are read from the node hydrological state prediction vector set. Combined with the corresponding edge feature vectors, the inflow and outflow of the current node are summed up within the same time index, and the difference between the sum of the inflow and the sum of the outflow is calculated to obtain the continuity equation residual. At each time index, the time index interval is combined based on the difference between the hydrological state prediction vector of the current node and the hydrological state prediction vector of the previous time index node, and the momentum equation residual is calculated by combining the water level gradient along the river channel and the friction slope determined by the roughness and flow velocity. At each time index, the residuals of the continuity equation and the momentum equation are weighted according to the adaptive weights to synthesize the hydrodynamic physical consistency residual vector. The adaptive weights are determined by the monotonic mapping of the residual amplitude, the time index interval and the time constant parameter. At each time index, a linear correction is performed on the hydrological state prediction vector of the current node based on the hydrodynamic physical consistency residual vector and the residual injection gain coefficient to generate a physically constrained hydrological state prediction vector. The residual injection gain coefficient is determined jointly by the time constant parameter and the time index interval and is limited to a preset closed interval. Repeat the process for all time indexes, collect the physically constrained hydrological state prediction vectors according to the time index, and form a set of physically constrained hydrological state prediction vectors.

[0026] Specifically include: At each time index, the current node hydrological state prediction vector and the hydrological state prediction vectors of the connected adjacent nodes are obtained from the node hydrological state prediction vector set, and the continuity equation residual is calculated in combination with the corresponding edge eigenvectors. The continuity equation residual is: ; in, Representation node The set of incoming adjacent nodes, Representation node The set of outgoing adjacent nodes, Represents a time index Next node Flow Node The flow rate is calculated based on the node hydrological state prediction vector of the adjacent node and the corresponding edge feature vector. Represents a time index Next node Flow Node Traffic volume; Calculate the momentum equation residual, which is: ; in, Represents a time index Next node The total flow, : time index interval, : gravitational acceleration constant, : Time index Next node The water level height, : The water level gradient along the river channel is calculated based on the water level difference between adjacent nodes and the length of the river section. : Time index The friction slope under the flow is calculated based on the roughness and flow velocity; The continuity equation residual and the momentum equation residual are combined according to the weighted coefficient to obtain the physically consistent residual vector: ; in, is the residual weight coefficient of the continuity equation, is the residual weight coefficient of the momentum equation; The physical consistency residual vector is used as a correction factor to act on the hydrological state prediction vector of the current node to generate the physically constrained hydrological state prediction vector: ; in, is the residual correction step coefficient, is the corrected node hydrological state prediction vector; The calculation is repeated on all time indexes, and the physical constraint hydrological state prediction vectors of each time index are collected to form a set of physical constraint hydrological state prediction vectors.

[0027] In this embodiment, the step five includes: At each time index, the physical constraint hydrological state prediction vector of the current node and the physical constraint hydrological state prediction vectors of several historical time indexes are extracted from the set of physical constraint hydrological state prediction vectors. The cumulative curve of the corresponding rainfall process and the cumulative curve of the flow process are calculated. The time difference between the rainfall peak and the flow peak is determined based on the position difference of the characteristic inflection points of the two cumulative curves, and the time lag estimate of the current node is generated. Under each time index, the time lag estimate and the corresponding time constant parameter of the current node are extracted, and the time constant parameter of the previous time index is extracted. The time lag estimate and the time constant parameter are normalized to obtain the standardized time lag and the standardized time constant. The change rate of the time constant parameter between adjacent time indexes is calculated to obtain the time constant change. The weight pair is determined according to the standardized time lag and the time constant change so that the sum of the weight pair is one. The standardized time lag and the standardized time constant are linearly fused according to the weight pair to generate an unrestricted correction coefficient. The unrestricted correction coefficient is monotonically normalized and clipped with upper and lower bounds to limit the result to a preset closed interval to obtain an adaptive Adaptive time lag correction coefficient; wherein, performing normalization and upper and lower bound clipping on the unrestricted correction coefficient includes: reading the unrestricted correction coefficient of the current node and the preset normalized target range at each time index, subtracting the minimum value of the unrestricted correction coefficients of all nodes in the time index from the unrestricted correction coefficient, and dividing it by the difference between the maximum and minimum values ​​of the unrestricted correction coefficients of all nodes in the time index to obtain a normalized result; comparing the normalized result with a preset lower bound value, and replacing it with the lower bound value when it is less than the lower bound value, and comparing the normalized result with a preset upper bound value, and replacing it with the upper bound value when it is greater than the upper bound value, and outputting the adaptive time lag correction coefficient limited to the preset closed interval; At each time index, the alignment position of the current node's physical constraint hydrological state prediction vector on the time axis is adjusted according to the adaptive time lag correction coefficient. Interpolation updates are performed on the hydrological state prediction vector of the node that needs to be moved forward, and delayed smoothing updates are performed on the hydrological state prediction vector of the node that needs to be moved backward. The specific processing of "performing interpolation update and performing delayed smoothing update" includes: Perform interpolation update on the hydrological state prediction vector of the node that needs to be moved forward: at each time index, calculate the forward step size based on the adaptive time lag correction coefficient, and record the current time index as t; when the forward step size is less than one time step, select the prediction vectors at time indexes t and t+1, generate the interpolation vector according to the linear weight corresponding to the forward step size, and write the interpolation vector to t; when the forward step size is not less than one time step, repeat the linear interpolation and writing per unit time step until all forward steps are completed; when there is no t+1 available at the end, use the end vector copy as the interpolation input; Perform delayed smoothing update on the hydrological state prediction vector of the node that needs to be shifted back: at each time index, calculate the shift step length based on the adaptive time lag correction coefficient, and the current time index is recorded as t; when the shift step length is less than one time step, the prediction vector at t and the prediction vector at t+1 are distributed and written at t and t+1 according to the linear weight corresponding to the shift step length; when the shift step length is not less than one time step, first write the prediction vector at t to the position of t+ the full step back, and the remaining part less than one full step is proportionally distributed between the two adjacent target indices; perform fixed window weighted average smoothing on the written position and the adjacent positions, and the weight decreases monotonically with the distance from the target index; when the beginning or end crosses the boundary, the beginning or end vector is used to fill the boundary; At each time index, the node hydrological state prediction vector after adaptive lag compensation is output, and the output vectors of all nodes are collected according to the time index to form a lag-compensated hydrological state prediction vector set.

[0028] In this embodiment, the flood risk classification map data and flood warning threshold list data specifically include: Extract flow prediction data, water level prediction data, and flood peak arrival time prediction data from the flood evolution prediction result data obtained by mapping the hysteresis-compensated hydrological state prediction vector set. Combined with the corresponding rainfall and basin infrastructure spatial data, calculate the inundation depth data, flow velocity data, and arrival time data of each grid cell. The flooding depth data, flow velocity data, and arrival time data of each grid cell are matched with a preset risk classification condition set. The risk classification condition set consists of a flooding depth threshold, a flow velocity threshold, and an arrival time threshold, and is divided into four levels: When the inundation depth is greater than or equal to three meters and the flow velocity is greater than or equal to two meters per second and the arrival time is less than or equal to one hour, the grid cell is marked as a level one risk; When the inundation depth is between two and three meters, the flow velocity is between one and two meters per second, or the arrival time is between one and three hours, and the first-level risk conditions are not met, the grid cell is marked as a second-level risk; When the inundation depth is between 0.5 and 2 meters, or the flow velocity is between 0.5 and 1 meter per second, or the arrival time is between 3 and 6 hours, and the risk conditions of level 1 and level 2 are not met, the grid cell is marked as level 3 risk; When the inundation depth is less than 0.5 meters and the flow velocity is less than 0.5 meters per second and the arrival time is greater than six hours, the grid cell is marked as low risk; At each time index, the risk level labels of all grid cells are synthesized into flood risk level data, the flood risk level data and key facility exposure data are spatially superimposed, the types and quantities of key facilities covered under each risk level are extracted, and the flood warning threshold list data is generated and output.

[0029] The adaptive neural network flood evolution simulation and risk assessment system according to an embodiment of the present invention includes the following modules: Data acquisition and preprocessing module, used to collect basin rainfall observation data and river water level observation data and generate multi-source hydrological and meteorological preprocessed data; The river network topology construction module is used to convert multi-source hydrological and meteorological pre-processed data into node feature vectors and edge feature vectors and form spatiotemporal graph input data; Liquid time constant neural network processing module, used to input spatiotemporal graph data and perform continuous time state update to generate a set of node hydrological state prediction vectors; A physical consistency residual injection module is used to inject hydrodynamic physical consistency residuals into the node hydrological state prediction vector set to generate a physical constraint hydrological state prediction vector set; An adaptive hysteresis compensation module, configured to perform adaptive hysteresis compensation on a set of physically constrained hydrological state prediction vectors to generate a set of hysteresis-compensated hydrological state prediction vectors; The flood evolution result generation module is used to map the hysteresis-compensated hydrological state prediction vector set into flow prediction data, water level prediction data and flood peak arrival time prediction data; The risk assessment and warning generation module is used to generate flood risk classification map data and flood warning threshold list data.

[0030] Example 1: In order to verify the feasibility of the present invention in implementation, the present invention was applied to the flood evolution prediction and risk assessment work in a medium-sized river basin. The river network structure of this river basin is relatively complex, with many tributaries, and the spatial distribution of rainfall processes varies greatly. The traditional method that relies on a single hydrological model has problems in this area, such as prediction lag, large peak error, and inaccurate risk classification. In the past, in similar scenarios, the arrival time of the flood peak was often delayed, and the flooding risks of some key infrastructure were not identified in advance, resulting in passive flood control scheduling.

[0031] In practical applications, rainfall observation stations and river water level observation stations are first deployed within the basin, and combined with satellite and radar rainfall inversion data to form multi-source hydrological and meteorological preprocessed data of rainfall and water level. After time alignment and missing measurement filling, these data are constructed to construct a river network topology map containing the connection relationship between observation sections and river sections. The observation section information is converted into node feature vectors, and the river section information is converted into edge feature vectors to form a complete spatiotemporal graph input data.

[0032] The above-mentioned spatiotemporal graph is input into a liquid time constant neural network embedded with a graph attention mechanism, and a set of node hydrological state prediction vectors is obtained through continuous-time state updates. During the prediction process, the network uses graph attention to calculate the weighted information of adjacent nodes and adjusts the state update speed in combination with the time constant parameter. It can automatically shorten the response time when heavy rainfall occurs and maintain stable predictions in the stable stage. Subsequently, the physically consistent residual is injected into the node hydrological state prediction vector to ensure that the changes in flow and water level meet the constraints of the continuity equation and momentum equation, thereby improving the physical rationality of the prediction results.

[0033] When addressing rainfall-runoff time lag effects, an adaptive lag compensation mechanism adjusts the time axis position and smoothes the state values ​​of the physically constrained hydrological state prediction vector to ensure alignment between flood peak predictions and actual measurements. During this process, the estimated lag value is combined with a time constant parameter to generate an adaptive lag correction coefficient. This allows for targeted compensation at different nodes during flood peak formation and water recede, reducing deviations in peak timing predictions.

[0034] The compensated prediction results are mapped into flow forecast data, water level forecast data and flood peak arrival time forecast data for future periods. Combined with the spatial distribution information of the basin infrastructure, the rasterized inundation depth, flow velocity and arrival time are calculated based on the exceedance probability in different years. Flood risk level data are generated according to the classification threshold, and the exposure data of key facilities are superimposed to form a flood risk classification map and a flood warning threshold list. In actual operation, the risk classification map provides an intuitive spatial reference for flood control command, and the warning threshold list clarifies the triggering conditions under each level of risk.

[0035] To verify its effectiveness, the method of the present invention was compared with the prediction results of a traditional single hydrological model during the same flood process. The maximum relative error of the present invention in flood peak flow prediction was controlled within 2%, the peak time prediction error was 0 hours, and the root mean square error of flow changes in the receding water section was approximately 45% lower than that of traditional methods. In terms of risk classification accuracy, the consistency rate between the flood risk level generated by the present invention and the actual inundation situation reached over 92%, and the early identification rate of risk areas at level 2 and above increased by nearly 30%. For impact assessment of critical infrastructure, the present invention can provide an early warning 6 hours before the flood peak arrives, making the deployment of defense measures more proactive.

[0036] The introduction of the method of the present invention improves the accuracy and timeliness of flood evolution prediction and the scientific nature of risk assessment. In particular, in scenarios involving complex river basins and multi-source data fusion, it demonstrates superior stability and reliability over traditional technologies. Table 1 shows the comparative data of the method of the present invention and traditional methods in practical applications on multiple key indicators.

[0037] Table 1: Comparison of flood evolution prediction and risk assessment results

[0038] From the comparison results in Table 1, it can be seen that the proposed method outperforms traditional methods in multiple core indicators of flood evolution prediction and risk assessment. In terms of hydrological prediction accuracy, the maximum relative error of peak flow prediction is reduced from 6.8% of the traditional method to 1.9%, with an error reduction of 72%. The peak time prediction error is reduced from 2 hours to 0.5 hours, improving the alignment accuracy of the peak time. The root mean square error of the flow in the receding section is also reduced from 184 m³ / s to 101 m³ / s, a decrease of 45%, effectively improving the stability of the water receding process; in terms of risk grading accuracy, the consistency rate between risk level and actual measurement situation increased from 78% to 92%, an increase of 14 percentage points, and the early identification rate of risk areas at level 2 and above increased from 61% to 79%, an increase of 30%, making the deployment of defense measures more targeted and timely; in terms of early warning response capability, the method of the present invention can provide effective early warning 6 hours before the arrival of the flood peak, an increase of 4 hours in advance compared with traditional methods. At the same time, the accuracy of the key facility coverage list increased from 80% to 95%, an increase of 15 percentage points, ensuring that key protection targets can be accurately identified; in terms of computational efficiency, the single prediction calculation time was shortened from 95 seconds to 54 seconds, an efficiency improvement of 43%, and the utilization rate of multi-source data fusion also increased from 65% to 96%, an increase of 31%, indicating that the present invention can make more full use of multi-source observation information, thereby achieving comprehensive optimization in terms of accuracy, timeliness and risk identification.

[0039] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. Adaptive neural network flood evolution simulation and risk assessment method, characterized by: The steps include: Step 1: Collect basin rainfall observation data and river water level observation data, and pre-process them to obtain multi-source hydrological and meteorological pre-processed data; Step 2: Construct a river network topology map based on multi-source hydrological and meteorological preprocessed data, construct the observation section as a node feature vector, and construct the river section as an edge feature vector to form the spatiotemporal graph input data; Step 3: Input the spatiotemporal graph input data into the liquid time constant neural network embedded with the graph attention mechanism, perform continuous time state update, and generate a set of node hydrological state prediction vectors; Step 4: inject the hydrodynamic physical consistency residual into the node hydrological state prediction vector set to generate the physically constrained hydrological state prediction vector set; Step 5: performing adaptive hysteresis compensation on the set of physically constrained hydrological state prediction vectors to generate a set of hysteresis-compensated hydrological state prediction vectors; Step 6: Map the hysteresis-compensated hydrological state prediction vector set into flow prediction data, water level prediction data, and flood peak arrival time prediction data to form flood evolution prediction result data; Step 7. Based on the flood evolution prediction results data, calculate the inundation depth data, flow velocity data and arrival time data under different annual exceedance probabilities, combine the spatial data of the basin infrastructure to generate flood risk level data and key facility exposure data, and output flood risk classification map data and flood warning threshold list data.

2. The adaptive neural network flood evolution simulation and risk assessment method according to claim 1 is characterized in that: The step 1 includes: performing time series missing value filling, outlier removal and time granularity unified processing on the basin rainfall observation data to generate rainfall preprocessed data; performing time series missing value filling, outlier removal and station benchmark unified processing on the river water level observation data to generate water level preprocessed data; synchronously aligning the rainfall preprocessed data and the water level preprocessed data according to a unified time benchmark to construct multi-source hydrological and meteorological preprocessed data.

3. The adaptive neural network flood evolution simulation and risk assessment method according to claim 2 is characterized in that: The second step includes: extracting water level preprocessing data and time index corresponding to each observation section from multi-source hydrological and meteorological preprocessing data, combining the water level preprocessing data of the same observation section under continuous time index in sequence to construct a node feature vector; extracting rainfall preprocessing data, roughness data and slope data corresponding to each river section from multi-source hydrological and meteorological preprocessing data, combining the data of the same river section under continuous time index in sequence to construct an edge feature vector; combining the node feature vector and the edge feature vector according to the upstream and downstream relationship of the river section to generate a river network topology map; arranging all node feature vectors and edge feature vectors in the river network topology map according to the time index to form a spatiotemporal map input data.

4. The adaptive neural network flood evolution simulation and risk assessment method according to claim 3 is characterized in that: The processing of the liquid time constant neural network embedded in the graph attention mechanism in step 3 includes: Arrange the spatiotemporal graph input data in ascending order by time index, extract the node feature vectors and edge feature vectors at each time index, initialize the learnable time constant parameters for each node, perform normalization based on the range of the node feature vector to obtain the initial value of the time constant, initialize the graph attention parameters for each edge, bind them to the edge feature vector corresponding to the edge, and use the node feature vector at the first time index as the initial node hydrological state prediction vector; Under the current time index, the current node feature vector is connected with the node hydrological state prediction vector of the previous time index to form a query vector. For each incoming edge connected to the node, the node hydrological state prediction vector of the previous time index of the adjacent node is connected with the edge feature vector of the incoming edge to form a key vector and a corresponding value vector. Perform a dot product operation on the query vector and the key vector along the corresponding dimensions to obtain a relevance score. Multiply the relevance score by the graph attention parameter of the incoming edge element-by-element, and add the result after linear transformation of the edge feature vector to obtain an unnormalized attention score. Apply the graph attention parameter of the corresponding edge to the attention score, normalize the scores of all incoming edges of the same node to obtain a set of attention weights, and perform a weighted sum of the corresponding value vectors with the attention weights to generate a graph attention aggregation vector. The graph attention aggregation vector and the current node feature vector are connected and input into the liquid time constant unit. The update coefficient is calculated based on the time constant parameter and the time index interval. The hydrological state prediction vector of the node at the previous time index and the candidate node state vector are weighted updated to obtain the current node hydrological state prediction vector. The calculation is repeated on all time indexes, and the node hydrological state prediction vectors of each time index are collected to form a node hydrological state prediction vector set.

5. The adaptive neural network flood evolution simulation and risk assessment method according to claim 4 is characterized in that: The fourth step includes: At each time index, the hydrological state prediction vector of the current node and the hydrological state prediction vector of the adjacent nodes are read from the node hydrological state prediction vector set. Combined with the corresponding edge feature vectors, the inflow and outflow of the current node are summed up within the same time index, and the difference between the sum of the inflow and the sum of the outflow is calculated to obtain the continuity equation residual. At each time index, the time index interval is combined based on the difference between the hydrological state prediction vector of the current node and the hydrological state prediction vector of the previous time index node, and the momentum equation residual is calculated by combining the water level gradient along the river channel and the friction slope determined by the roughness and flow velocity. At each time index, the residuals of the continuity equation and the momentum equation are weighted according to the adaptive weights to synthesize the hydrodynamic physical consistency residual vector. The adaptive weights are determined by the monotonic mapping of the residual amplitude, the time index interval and the time constant parameter. At each time index, a linear correction is performed on the hydrological state prediction vector of the current node based on the hydrodynamic physical consistency residual vector and the residual injection gain coefficient to generate a physically constrained hydrological state prediction vector. The residual injection gain coefficient is determined jointly by the time constant parameter and the time index interval and is limited to a preset closed interval. Repeat the process for all time indexes, collect the physically constrained hydrological state prediction vectors according to the time index, and form a set of physically constrained hydrological state prediction vectors.

6. The adaptive neural network flood evolution simulation and risk assessment method according to claim 5 is characterized in that: The step five includes: At each time index, the physical constraint hydrological state prediction vector of the current node and the physical constraint hydrological state prediction vectors of several historical time indexes are extracted from the set of physical constraint hydrological state prediction vectors. The cumulative curve of the corresponding rainfall process and the cumulative curve of the flow process are calculated. The time difference between the rainfall peak and the flow peak is determined based on the position difference of the characteristic inflection points of the two cumulative curves, and the time lag estimate of the current node is generated. At each time index, the time lag estimate and the corresponding time constant parameter of the current node are extracted, and the time constant parameter of the previous time index is extracted. The time lag estimate and the time constant parameter are normalized to obtain a standardized time lag amount and a standardized time constant. The change rate of the time constant parameter between adjacent time indexes is calculated to obtain a time constant change amount. A weight pair is determined based on the standardized time lag amount and the time constant change amount so that the sum of the weight pair is one. The standardized time lag amount and the standardized time constant are linearly fused according to the weight pair to generate an unrestricted correction coefficient. The unrestricted correction coefficient is monotonically normalized and clipped with upper and lower bounds, and the result is restricted to a preset closed interval to obtain an adaptive time lag correction coefficient. At each time index, the alignment position of the current node's physical constraint hydrological state prediction vector on the time axis is adjusted according to the adaptive time lag correction coefficient. Interpolation updates are performed on the hydrological state prediction vector of the node that needs to be moved forward, and delayed smoothing updates are performed on the hydrological state prediction vector of the node that needs to be moved backward. At each time index, the node hydrological state prediction vector after adaptive lag compensation is output, and the output vectors of all nodes are collected according to the time index to form a lag-compensated hydrological state prediction vector set.

7. The adaptive neural network flood evolution simulation and risk assessment method according to claim 6, characterized in that: The flood risk classification map data and flood warning threshold list data specifically include: Extract flow prediction data, water level prediction data, and flood peak arrival time prediction data from the flood evolution prediction result data obtained by mapping the hysteresis-compensated hydrological state prediction vector set. Combined with the corresponding rainfall and basin infrastructure spatial data, calculate the inundation depth data, flow velocity data, and arrival time data of each grid cell. The flooding depth data, flow velocity data, and arrival time data of each grid cell are matched with a preset risk classification condition set. The risk classification condition set consists of a flooding depth threshold, a flow velocity threshold, and an arrival time threshold, and is divided into four levels: When the inundation depth is greater than or equal to three meters and the flow velocity is greater than or equal to two meters per second and the arrival time is less than or equal to one hour, the grid cell is marked as a level one risk; When the inundation depth is between two and three meters, the flow velocity is between one and two meters per second, or the arrival time is between one and three hours, and the first-level risk conditions are not met, the grid cell is marked as a second-level risk; When the inundation depth is between 0.5 and 2 meters, or the flow velocity is between 0.5 and 1 meter per second, or the arrival time is between 3 and 6 hours, and the risk conditions of level 1 and level 2 are not met, the grid cell is marked as level 3 risk; When the inundation depth is less than 0.5 meters and the flow velocity is less than 0.5 meters per second and the arrival time is greater than six hours, the grid cell is marked as low risk; At each time index, the risk level labels of all grid cells are synthesized into flood risk level data, the flood risk level data and key facility exposure data are spatially superimposed, the types and quantities of key facilities covered under each risk level are extracted, and the flood warning threshold list data is generated and output.

8. An adaptive neural network flood evolution simulation and risk assessment system, applied to the adaptive neural network flood evolution simulation and risk assessment method according to any one of claims 1 to 7, characterized in that: Includes the following modules: Data acquisition and preprocessing module, used to collect basin rainfall observation data and river water level observation data and generate multi-source hydrological and meteorological preprocessed data; The river network topology construction module is used to convert multi-source hydrological and meteorological pre-processed data into node feature vectors and edge feature vectors and form spatiotemporal graph input data; Liquid time constant neural network processing module, used to input spatiotemporal graph data and perform continuous time state update to generate a set of node hydrological state prediction vectors; A physical consistency residual injection module is used to inject hydrodynamic physical consistency residuals into the node hydrological state prediction vector set to generate a physical constraint hydrological state prediction vector set; An adaptive hysteresis compensation module, configured to perform adaptive hysteresis compensation on a set of physically constrained hydrological state prediction vectors to generate a set of hysteresis-compensated hydrological state prediction vectors; The flood evolution result generation module is used to map the hysteresis-compensated hydrological state prediction vector set into flow prediction data, water level prediction data and flood peak arrival time prediction data; The risk assessment and warning generation module is used to generate flood risk classification map data and flood warning threshold list data.

Images