Partitioned parallel computing method and system for dynamic flood risk map

Abstract

The application provides a kind of dynamic flood risk map partition parallel computing method and system, the method comprises: according to the real-time weather forecast information of target area, match target primary partition scheme, to divide the spatial computing domain of target area into multiple primary computing partitions, and parallelly execute the simulation calculation of flood risk factor in multiple primary computing partitions;If it is monitored that there is target primary computing partition with abnormal real-time computing load index value in multiple primary computing partitions, then match target secondary partition scheme, to decompose target primary computing partition into multiple secondary computing partitions, and switch the simulation calculation for target primary computing partition to the parallel simulation calculation of flood risk factor for multiple secondary computing partitions;According to the simulation calculation result of each primary computing partition and the simulation calculation result of each secondary computing partition, generate the flood risk map of target area.The application realizes the low-cost, fast, stable and reliable parallel generation of dynamic flood risk map.

Description

Technical Field

[0001] This invention relates to the field of water conservancy and hydropower technology, and in particular to a method and system for parallel calculation of dynamic flood risk maps by region. Background Technology

[0002] Flood dynamic risk maps are a core support for disaster prevention, mitigation, and emergency decision-making. To achieve accurate simulation of flood evolution, it is usually necessary to couple multiple models, such as one-dimensional river networks, two-dimensional surface and underground pipe networks. Since the coupled models involve massive numerical calculations, parallel computing technology must be used to improve simulation efficiency in order to meet the extremely high requirements for the timeliness of flood forecasting, so as to achieve dynamic generation of risk maps at the minute or even second level.

[0003] Existing technologies typically employ a parallel strategy based on geometric space partitioning. This approach primarily involves uniformly dividing the computational domain into multiple sub-regions according to geometric rectangles and assigning them to different computing nodes for synchronous execution, thereby leveraging multi-core computing power to accelerate the simulation process of flood evolution.

[0004] However, the aforementioned technologies sever the hydraulic connections between multiple model components, which can easily lead to huge cross-partition communication overhead. Frequent data exchanges offset the benefits of parallel acceleration, and the fixed partitioning pattern can easily cause the computing nodes to exhibit a severe load imbalance during the simulation process, resulting in low resource utilization efficiency and making it difficult to support the real-time generation speed of dynamic flood risk maps under complex working conditions and extreme scenarios. Summary of the Invention

[0005] This invention provides a partitioned parallel computing method and system for dynamic flood risk maps, which solves the problems of existing technologies where parallel partitioning severs the hydraulic connections between multiple model components, resulting in high communication overhead, and static partitioning is difficult to adapt to dynamic migration of computing load, leading to node load imbalance. This invention achieves low-cost, fast, stable, and reliable parallel generation of dynamic flood risk maps.

[0006] This invention provides a method for partitioned parallel computation of dynamic flood risk maps, comprising: Based on real-time weather forecast information for the target area, match the target first-level zoning plan from the first-level zoning plan database; According to the target first-level zoning scheme, the spatial computing domain of the target area is divided into multiple first-level computing zones, and the simulation calculation of flood risk factors is performed in parallel on the multiple first-level computing zones. If, during the simulation calculation process, an abnormal real-time computing load index value is detected in a target first-level computing partition among the multiple first-level computing partitions, then the target second-level partition scheme corresponding to the target first-level computing partition is matched from the second-level partition scheme library. According to the target secondary zoning scheme, the target primary calculation zone is decomposed into multiple secondary calculation zones, and the simulation calculation of flood risk factors for the target primary calculation zone is switched to parallel simulation calculation of flood risk factors for the multiple secondary calculation zones. Based on the simulation results of each of the first-level calculation zones and the simulation results of each of the second-level calculation zones, a flood risk map of the target area is generated; The primary partitioning plan library includes multiple primary partitioning schemes. Each primary partitioning scheme is obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios, while keeping the coupling nodes intact. The secondary partitioning plan library includes secondary partitioning schemes pre-constructed for the topology structure of each primary calculation partition corresponding to each primary partitioning scheme. The spatial topology map includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes. The coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

[0007] According to the method for parallel computation of dynamic flood risk maps by region provided by the present invention, the construction steps of the first-level regional contingency plan database include: In multiple flood scenarios, offline hydrodynamic simulations are performed on the target area to obtain the hydraulic interaction information of the coupling interfaces between the model components under each flood scenario, as well as the hydraulic element information of each model component under each flood scenario. Based on the hydraulic interaction information, the cumulative time steps of hydraulic interaction of each coupling interface under each flood scenario are counted to obtain the total activation time of each coupling interface under each flood scenario; Based on the total activation duration, obtain the coupling strength coefficient of each of the coupling interfaces; Based on the coupling strength coefficient, the spatial topology graph is constructed; Based on the hydraulic element information, the weight of each node in the spatial topology map under each flood scenario is determined, and based on the weight, the first-level zoning scheme of the target area under each flood scenario is obtained. Based on all the aforementioned first-level partitioning schemes, construct the first-level partitioning plan library.

[0008] According to the present invention, a method for partitioned parallel computation of a dynamic flood risk map is provided, wherein obtaining the coupling strength coefficient of each coupling interface based on the total activation duration includes: For each coupling interface, if the cumulative water volume of hydraulic interaction occurring at the coupling interface under any flood scenario is less than or equal to the target water volume, the activation duration percentage of the coupling interface under any flood scenario is determined according to a preset percentage value. If the cumulative water volume is greater than the target water volume, the activation time ratio of the coupling interface in any flood scenario is determined based on the ratio between the total activation time of the coupling interface in any flood scenario and the total simulation time of the any flood scenario. The activation duration percentage of each of the coupling interfaces in each of the flood scenarios is obtained through iteration. Based on the scenario weights corresponding to each flood scenario, the activation duration percentages of each of the coupling interfaces in all flood scenarios are weighted and fused to obtain the coupling strength coefficient of each of the coupling interfaces. The scenario weights corresponding to each flood scenario are determined by normalizing the reciprocal of the recurrence period of each flood scenario.

[0009] According to the present invention, a method for partitioned parallel computation of a dynamic flood risk map is provided, wherein constructing the spatial topology map based on the coupling strength coefficient includes: Each model unit in each of the aforementioned model components is aggregated in the same category to obtain multiple initial independent nodes; the multiple initial independent nodes include initial river segment nodes generated by aggregating multiple connected river segment units, initial surface grid nodes generated by aggregating multiple adjacent grid units, and initial pipeline nodes generated by aggregating multiple connected pipeline units. For each initial river segment node, among the model units of each initial surface grid node and each initial pipeline node, a first model unit is determined that has a coupling interface with any model unit in the initial river segment node that has a coupling strength coefficient greater than a preset coefficient. The initial river segment node is aggregated with all the first model units to form the coupling node, and each of the first model units is deleted from its original surface grid node or initial pipeline node. For each deleted initial surface grid node, in each model unit of each deleted initial pipeline node, a second model unit with a coupling interface having a coupling strength coefficient greater than the preset coefficient with any model unit in the deleted initial surface grid node is identified, and the deleted initial surface grid node and all the second model units are aggregated into the coupling node. Each of the second model units is removed from its corresponding deleted initial network node; Based on the coupled nodes generated by aggregation, as well as the deleted initial river segment nodes, deleted initial surface grid nodes, and deleted initial pipeline nodes, a vertex set is constructed, and an edge set is constructed based on the topological relationships between the nodes in the vertex set; The spatial topology graph is constructed based on the vertex set and the edge set.

[0010] According to the present invention, a method for partitioned parallel computation of a dynamic flood risk map is provided, wherein determining the weight of each node in the spatial topology map under each flood scenario based on the hydraulic element information includes: For each type of model unit, the basic weight of the model unit under each flood scenario is calculated based on the calculated strength coefficient of the model component corresponding to the model unit and the hydraulic element information of the model component corresponding to the model unit under each flood scenario. For each of the surface grid nodes, river segment nodes, and pipeline nodes, the weight of the node in each flood scenario is calculated based on the number of model units contained within the node and the basic weight of the same type of model units contained within the node in each flood scenario. For the coupled node, the weight of the coupled node in each flood scenario is calculated based on the sum of the basic weights of all types of model units contained in the coupled node under each flood scenario, and the coupling computation additional overhead coefficient of the coupled node; the coupling computation additional overhead coefficient is calculated based on the average coupling strength coefficient of all coupled interfaces contained in the coupled node.

[0011] According to the present invention, a method for parallel computation of dynamic flood risk maps by zoning, wherein obtaining a first-level zoning scheme for the target area under each flood scenario based on the weights includes: For each flood scenario, the nodes in the spatial topology graph are partitioned with the constraint objective of minimizing the number of partition cutting edges and maximizing the balance between the sum of the weights of all nodes in each first-level computational partition under the flood scenario. Based on the partitioning results, a first-level partitioning scheme under the flood scenario is generated.

[0012] According to the present invention, a method for partitioned parallel computation of dynamic flood risk maps is provided, the method further comprising: Real-time acquisition of the maximum hydraulic wave velocity and minimum spatial step size of each type of model unit within each primary computing partition; Based on the maximum hydraulic wave velocity, the minimum spatial step size, and the communication step size of the computing nodes corresponding to each first-level computing partition, calculate the number of overlapping layers of each type of model unit in each first-level computing partition, and construct an extended overlapping region for each first-level computing partition based on the number of overlapping layers. Upon detecting the completion of the simulation calculation in the core area of ​​any level computing partition, an asynchronous communication task is initiated for the aforementioned level computing partition, and the pending computing tasks within the aforementioned level computing partition continue to be executed synchronously; the pending computing tasks refer to auxiliary computing tasks or simulation computing tasks at the next time point that are not dependent on the asynchronous communication task. The asynchronous communication task includes a receiving task and / or a sending task; the receiving task is used to receive a first flood risk factor of the first extended overlapping area simulated in the main computing partition that has the ownership of the first extended overlapping area of ​​the arbitrary first-level computing partition, and to cover the second flood risk factor of the first extended overlapping area simulated in the arbitrary first-level computing partition with the first flood risk factor; the sending task is used to send the flood risk factors of the second extended overlapping areas of other adjacent computing partitions simulated in the arbitrary first-level computing partition.

[0013] According to the present invention, a method for parallel computation of dynamic flood risk maps by region, wherein the simulation calculation of flood risk factors is performed in parallel on the plurality of first-level computational regions, comprising: Multiple master computing nodes are invoked to perform parallel simulation calculations of flood risk factors on the multiple primary computing partitions; each primary computing partition corresponds one-to-one with a master computing node. The hot standby nodes of each of the main computing nodes are invoked to collect the heartbeat signals and checkpoint data generated by each of the main computing nodes in real time during the simulation calculation process; When the number of lost heartbeat signals of any master computing node reaches a preset threshold, the hot standby node of the master computing node is driven to load the latest checkpoint data of the master computing node and take over the simulation computing task of the master computing node. After the hot standby node of any primary computing node completes data loading and task takeover, the communication connection between the hot standby node of any primary computing node and other computing nodes other than the primary computing node is updated.

[0014] According to the partitioned parallel computing method for dynamic flood risk maps provided by the present invention, for each first-level computing partition, the step of obtaining the real-time computing load index value of the first-level computing partition includes: The real-time hydraulic element information of each model unit within the first-level calculation partition is statistically analyzed, and the real-time hydraulic element information includes real-time water depth and real-time flow velocity. Obtain hardware resource utilization information of the computing nodes that perform parallel simulation computing in the first-level computing partition, the hardware resource utilization information including processor utilization and memory bandwidth utilization; Based on the weighting coefficients corresponding to the real-time hydraulic element information and the weighting coefficients corresponding to the hardware resource utilization information, a multi-factor weighted fusion is performed on the real-time hydraulic element information and the hardware resource utilization information to obtain the real-time computing load index value of the first-level computing partition.

[0015] This invention also provides a partitioned parallel computing system for dynamic flood risk maps, comprising: The first matching unit is used to match the target first-level zoning plan from the first-level zoning plan database based on the real-time weather forecast information of the target area. The first computing unit is used to divide the spatial computing domain of the target area into multiple first-level computing partitions according to the target first-level zoning scheme, and to perform simulation calculations of flood risk factors in parallel on the multiple first-level computing partitions. The second matching unit is used to match the target secondary partition scheme corresponding to the target primary computing partition from the secondary partition scheme library if an abnormal real-time computing load index value is detected in the multiple primary computing partitions during the simulation calculation process. The second computing unit is used to decompose the target primary computing partition into multiple secondary computing partitions according to the target secondary zoning scheme, and to switch the simulation calculation of flood risk factors for the target primary computing partition to the parallel simulation calculation of flood risk factors for the multiple secondary computing partitions. The generation unit is used to generate a flood risk map of the target area based on the simulation calculation results of each of the first-level calculation partitions and the simulation calculation results of each of the second-level calculation partitions. The primary partitioning plan library includes multiple primary partitioning schemes. Each primary partitioning scheme is obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios, while keeping the coupling nodes intact. The secondary partitioning plan library includes secondary partitioning schemes pre-constructed for the topology structure of each primary calculation partition corresponding to each primary partitioning scheme. The spatial topology map includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes. The coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a partitioned parallel computing method for dynamic flood risk maps as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a partitioned parallel computing method for dynamic flood risk maps as described in any of the above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a partitioned parallel computing method for dynamic flood risk maps as described above.

[0019] The method and system for partitioned parallel computation of dynamic flood risk maps provided by this invention identifies strongly interactive coupling interfaces through multi-scenario physical pre-analysis, forcibly aggregating heterogeneous model units into indivisible coupled nodes to ensure the integrity of the physical process, and constructs a two-level partitioned contingency plan system. This achieves a deep integration of initial global scenario matching and dynamic decomposition of local hotspot areas during the computation process. Thus, while avoiding the interruption of hydraulic connections from the underlying topology, it fundamentally solves the problem of dynamic load imbalance caused by the movement of flood fronts, significantly improving the parallel acceleration efficiency, resource utilization, numerical accuracy, and system stability of dynamic flood risk map generation under multi-model coupling. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts illustrating the partitioned parallel calculation method for dynamic flood risk maps provided by this invention.

[0022] Figure 2 This is the second flowchart of the partitioned parallel calculation method for dynamic flood risk maps provided by the present invention.

[0023] Figure 3 This is a schematic diagram of the aggregate of river section nodes provided by the present invention.

[0024] Figure 4 This is a schematic diagram of the aggregate of mesh nodes provided by the present invention.

[0025] Figure 5 This is a schematic diagram of the aggregate of pipeline nodes provided by the present invention.

[0026] Figure 6 This is a schematic diagram of the aggregate of coupling nodes provided by the present invention.

[0027] Figure 7 This is a schematic diagram of the calculation area and the overlapping area provided by the present invention.

[0028] Figure 8This is the third flowchart of the partitioned parallel calculation method for dynamic flood risk maps provided by this invention.

[0029] Figure 9 This is a schematic diagram of the partitioned parallel computing system for dynamic flood risk maps provided by the present invention.

[0030] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0032] All actions involving the acquisition of signal information or data in this application are carried out in accordance with the relevant data protection laws and policies of the country where the application is located, and with the authorization granted by the owner of the relevant device.

[0033] Dynamic flood risk maps refer to a technology that rapidly calculates and dynamically updates risk factors such as flood inundation range, water depth, and flow velocity based on real-time or forecasted hydrological conditions. With the increasing impact of climate change and human activities, flood risks in watersheds and cities are becoming increasingly complex, placing extremely high demands on the generation speed of dynamic flood risk maps, typically requiring a simulation update within minutes or even seconds. Currently used flood evolution models often involve multi-model coupling, such as the coupling of a one-dimensional river network model with a two-dimensional surface water dynamics model, and may even include urban drainage network models. These coupled models involve enormous computational demands, especially in large-scale, high-resolution scenarios, where serial computation is far from meeting the dynamic update requirements. To improve generation speed, parallel computing techniques are typically employed. Existing technologies usually employ a parallel strategy based on geometric space partitioning. This approach mainly involves uniformly dividing the computational domain into multiple sub-regions according to geometric rectangles and assigning them to different computing nodes for synchronous execution, thereby utilizing multi-core computing power to accelerate the flood evolution simulation process. However, existing parallel computing methods have the following shortcomings: First, geometric partitioning severs hydraulic connections. Simple geometric rectangular partitioning often directly cuts off the hydraulic interaction between river networks and floodplains, and between the surface and pipe networks, leading to a surge in cross-partition communication overhead, and frequent data exchange offsetting the benefits of parallel acceleration.

[0034] Secondly, static partitioning leads to load imbalance and low resource utilization efficiency. Existing methods mostly use fixed partitioning, which cannot adapt to the dynamic load migration caused by the movement of the flood front during flood evolution. For example, computing nodes in dry areas not reached by the flood are idle, while computing nodes in active areas where the flood front is located are severely congested, resulting in resource waste of idle dry areas and congested flood areas; moreover, it does not distinguish which connections are physically frequently interacting (such as the main channel and the floodplain) and which connections are rarely activated (such as tributaries that are difficult to reach at high water levels), resulting in excessively large partitions, uneven load, and low resource utilization efficiency.

[0035] Therefore, there is an urgent need for a partitioned parallel computing method that can take into account the integrity of physical processes, load balancing, and fault tolerance and reliability, so as to meet the comprehensive requirements of dynamic flood risk maps with multiple coupled models (including one-dimensional river network, two-dimensional surface and drainage network and other hydrological and hydrodynamic models) for computing speed, computing cost, accuracy and stability.

[0036] To address this deficiency, this application provides a partitioned parallel computing method for dynamic flood risk maps. Figure 1 This is one of the flowcharts illustrating the partitioned parallel computation method for dynamic flood risk maps provided by this invention; for example... Figure 1 As shown, the method includes: Step 110: Based on the real-time meteorological forecast information of the target area, match the target first-level zoning scheme from the first-level zoning scheme library. The first-level zoning scheme library includes multiple first-level zoning schemes, each of which is obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios while keeping the coupling nodes intact. The second-level zoning scheme library includes second-level zoning schemes pre-constructed for the topology structure of each first-level calculation zone corresponding to each of the first-level zoning schemes. The spatial topology map includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes. The coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

[0037] Optionally, before performing step 110, an offline simulation is first performed on the target area to construct a primary-level zoning contingency plan library and a secondary-level zoning contingency plan library. The target area here refers to the geographic spatial range for which flood evolution simulation and risk mapping are required, which may include urban central areas, specific watershed basins, or drainage areas.

[0038] The primary zoning plan library here refers to a pre-stored set of optimal computational domain partitioning schemes for different flood scenarios. Each partitioning scheme is predetermined through offline physical analysis while ensuring the integrity of the hydraulic coupling process. The primary zoning plan library can be generated by offline simulation of multiple design flood scenarios, forcibly aggregating all units with actual hydraulic interactions into indivisible coupled nodes, fundamentally ensuring the integrity of the coupling, and performing graph partitioning for flood scenarios with different return periods.

[0039] For example, in some embodiments, the steps of constructing the primary partition plan library include: In multiple flood scenarios, offline hydrodynamic simulations are performed on the target area to obtain the hydraulic interaction information of the coupling interfaces between the model components under each flood scenario, as well as the hydraulic element information of each model component under each flood scenario. Based on the hydraulic interaction information, the cumulative time steps of hydraulic interaction of each coupling interface under each flood scenario are counted to obtain the total activation time of each coupling interface under each flood scenario; Based on the total activation duration, obtain the coupling strength coefficient of each of the coupling interfaces; Based on the coupling strength coefficient, the spatial topology graph is constructed; Based on the hydraulic element information, the weight of each node in the spatial topology map under each flood scenario is determined, and based on the weight, the first-level zoning scheme of the target area under each flood scenario is obtained. Based on all the aforementioned first-level partitioning schemes, construct the first-level partitioning plan library.

[0040] Optionally, when constructing the primary-level regional contingency plan database, multiple representative flood scenarios can be selected for the target area. Flood scenarios refer to pre-defined flood conditions with different rainfall intensities and return periods, such as Scenario A (frequent light rain, 2-year return period, 24-hour rainfall of 65 mm), Scenario B (frequent heavy rain, 10-year return period, 24-hour rainfall of 174 mm), and Scenario C (extreme torrential rain, 100-year return period, 24-hour rainfall of 298 mm), etc. The specific scenarios are selected based on the actual area, and this embodiment does not impose specific limitations on them. Furthermore, multiple model components are configured for the target area, specifically including one-dimensional river segments, two-dimensional surface grids, and drainage pipe networks (also known as pipe networks), etc., and this embodiment does not impose specific limitations on them.

[0041] Subsequently, under multiple flood scenarios, offline hydrodynamic simulations were performed on the target area using a multi-model component coupling approach. The hydraulic interaction information of the coupling interfaces between each model component under each flood scenario, as well as the hydraulic element information of each model component under each flood scenario, were recorded and stored. It should be noted that this offline simulation can run on a single machine or on a small-scale cluster with a scale smaller than a set value; this embodiment does not specifically limit this.

[0042] The coupling interface here refers to the boundary where water exchange occurs between different model components, such as the interface between a one-dimensional river segment and a two-dimensional surface grid, the connection point between a one-dimensional river segment and a drainage network, and the overflow point between a drainage network and a two-dimensional surface grid. This embodiment does not specifically limit this.

[0043] The hydraulic interaction information here includes the hydraulic exchange status information of the coupling interface at each time step under various flood scenarios, as well as the water volume information of the hydraulic interaction; the hydraulic exchange status information is used to identify whether water volume exchange exists.

[0044] The hydraulic information here includes, but is not limited to, the maximum water depth for each model element. Maximum flow rate and maximum wave speed , This represents gravitational acceleration. For a two-dimensional surface grid, This represents the depth of surface water; for a one-dimensional river segment, The average water depth across the cross section (i.e., water level minus riverbed elevation); for pipe networks, This refers to the water head (or water depth) in the pipeline.

[0045] Subsequently, for each coupling interface, based on its hydraulic exchange state information at each time step under each flood scenario, the cumulative number of time steps of hydraulic interaction under each flood scenario is counted to obtain the total activation duration of each coupling interface under each flood scenario. Based on the total activation duration of each coupling interface in each flood scenario. This is used to obtain the coupling strength coefficient of each coupling interface. The coupling strength coefficient here is the core indicator for measuring the tightness of coupling between model components.

[0046] Here, the coupling strength coefficient can be dynamically quantized based on a calculation formula or obtained by mapping the total activation time with the coupling strength coefficient, etc. This embodiment does not specifically limit this.

[0047] For example, in order to accurately characterize the actual hydraulic interaction activity of each model component during the evolution of floods of different magnitudes, and to effectively eliminate the interference of minor water exchange caused by numerical fluctuations in the early stage of model calculation on the integrity of the partition, in one possible large-scale embodiment, obtaining the coupling strength coefficient of each coupling interface based on the total activation duration includes: For each coupling interface, if the cumulative water volume of hydraulic interaction occurring at the coupling interface under any flood scenario is less than or equal to the target water volume, the activation duration percentage of the coupling interface under any flood scenario is determined according to a preset percentage value. If the cumulative water volume is greater than the target water volume, the activation time ratio of the coupling interface in any flood scenario is determined based on the ratio between the total activation time of the coupling interface in any flood scenario and the total simulation time of the any flood scenario. The activation duration percentage of each of the coupling interfaces in each of the flood scenarios is obtained through iteration. Based on the scenario weights corresponding to each flood scenario, the activation duration percentages of each of the coupling interfaces in all flood scenarios are weighted and fused to obtain the coupling strength coefficient of each of the coupling interfaces. The scenario weights corresponding to each flood scenario are determined by normalizing the reciprocal of the recurrence period of each flood scenario.

[0048] Optionally, for each coupling interface, the coupling strength coefficient is calculated using the following formula: ; ; in, This is the coupling strength coefficient of the coupling interface; This represents the percentage of activation time for the coupling interface under any flood scenario. This represents the total activation duration of the coupling interface under any flood scenario; The total simulation duration for any flood scenario; The target water volume (also known as the water volume threshold) can be dynamically adjusted according to the grid scale of the target area, such as being designed to be 0.1 m3. The cumulative water volume of hydraulic interaction occurring at this coupling interface under any flood scenario is expressed in m3. When the cumulative water volume is less than a certain value, the coupling interface can be ignored. The scenario weights for any flood scenario can be determined by normalizing the reciprocal of the recurrence period of the flood scenario, while also taking into account the importance of extreme events and optimizing the values ​​appropriately. For example, the weight range for high-frequency scenarios with a recurrence frequency higher than the first value can be 0.65~0.75, and the weight range for low-frequency scenarios with a recurrence frequency lower than the second value can be 0.25~0.35. For instance, the weight for a 5-year flood is w_5=0.7, and the weight for a 100-year flood is w_100=0.3. The first value is greater than the second value. The value range of is [0,1], and its physical meaning is: when , When ≈1.0, it indicates that the coupling interface is activated almost throughout all flood scenarios (such as the boundary between the main channel and the floodplain); when 0.3 < When <0.7, it indicates that the coupling interface is activated only in certain scenarios or during certain time periods (e.g., high beaches); when When ≈0, it indicates that the coupling interface is not activated in all scenarios (such as extremely high, flood-prone areas), which is a permanent decoupling edge.

[0049] This embodiment achieves accurate quantification of the physical activity of the coupled interface by deeply integrating water volume threshold filtering with time proportion weighting. This provides a physically consistent criterion for subsequently aggregating strong interaction units into indivisible coupled nodes, thereby maximizing the flexibility of partition design while ensuring the accuracy of parallel computing.

[0050] After obtaining the coupling strength coefficients of each coupling interface, a spatial topology graph can be constructed based on these coefficients. This process involves first aggregating the model units within the model components of the same type. Then, using the coupling strength coefficients of each coupling interface, coupling nodes are aggregated for the aggregated independent nodes to construct the vertex set of the spatial topology graph. Finally, edges between the nodes in the vertex set are established based on the topological relationships between them, thus obtaining the spatial topology graph.

[0051] For example, in order to efficiently identify and solidify hydraulically coupled structures in massive computing units, in some possible embodiments, constructing the spatial topology map based on the coupling strength coefficient includes: Each model unit in each of the aforementioned model components is aggregated in the same category to obtain multiple initial independent nodes; the multiple initial independent nodes include initial river segment nodes generated by aggregating multiple connected river segment units, initial surface grid nodes generated by aggregating multiple adjacent grid units, and initial pipeline nodes generated by aggregating multiple connected pipeline units. For each initial river segment node, among the model units of each initial surface grid node and each initial pipeline node, a first model unit is determined that has a coupling interface with any model unit in the initial river segment node that has a coupling strength coefficient greater than a preset coefficient. The initial river segment node is aggregated with all the first model units to form the coupling node, and each of the first model units is deleted from its original surface grid node or initial pipeline node. For each deleted initial surface grid node, in each model unit of each deleted initial pipeline node, a second model unit with a coupling interface having a coupling strength coefficient greater than the preset coefficient with any model unit in the deleted initial surface grid node is identified, and the deleted initial surface grid node and all the second model units are aggregated into the coupling node. Each of the second model units is removed from its corresponding deleted initial network node; Based on the coupled nodes generated by aggregation, as well as the deleted initial river segment nodes, deleted initial surface grid nodes, and deleted initial pipeline nodes, a vertex set is constructed, and an edge set is constructed based on the topological relationships between the nodes in the vertex set; The spatial topology graph is constructed based on the vertex set and the edge set.

[0052] Optionally, the model units in each model component are first aggregated to obtain multiple initial independent nodes. The model units here include river segment units, surface grid units (also called grid units), and pipeline units. Accordingly, when performing the aggregation, specifically, for all river segment units, multiple connected river segment units are aggregated into an initial river segment node according to a certain number (e.g., NR); for all grid units, multiple adjacent grid units are aggregated into an initial surface grid node according to a certain number (e.g., NG); for all pipeline units, multiple connected pipeline units are aggregated into an initial pipeline node according to a certain number (assumed to be NP).

[0053] Then, based on the coupling relationship between model units, coupling nodes (C-Nodes) are identified and constructed. Specifically, all coupling interfaces with S>0 are traversed, and the following operations are performed on each coupling interface: First traversal: Traverse each aggregated initial channel node (each node represents approximately NR segment units). In each model unit of each initial surface grid node and each model unit of each initial pipeline node, find the first model unit with a coupling strength coefficient greater than a preset coefficient (e.g., 0) that has a coupling interface with any segment unit within that node. If there exists a coupling relationship with any segment unit within that node and... The first model unit with a value greater than 0, i.e., the pipeline unit and / or grid unit, will have its initial channel node (also known as the river segment node) re-marked as a coupling node C-Node. All first model units (i.e., pipeline units and / or grid units) directly coupled to this node will be aggregated to this coupling node C-Node. Simultaneously, each first model unit will be deleted from its original surface grid node or initial pipeline node. It should be noted that if a grid unit has active coupling interfaces with multiple channel nodes simultaneously, the relevant channel nodes and grid units will be merged into a composite coupling node to ensure the logical indivisibility of the water flow exchange process.

[0054] Second traversal: Traverse all aggregated mesh nodes (i.e., all initial surface mesh nodes after the first traversal). Among the model elements of all deleted initial pipeline nodes (all initial pipeline nodes after the first traversal), identify second model elements that have a coupling strength coefficient greater than a preset coefficient (e.g., 0) with any mesh element in that node. If there exists a direct coupling relationship with any mesh element inside that node and... The second model unit (i.e., the pipeline unit) with a value greater than 0 is remarked as the coupling node C-Node. The second model units (i.e., the pipeline units) that are coupled with this node are aggregated to the coupling node C-Node, and these second model units are deleted from the initial pipeline node after deletion.

[0055] It should be noted that since the first traversal has already handled river-related couplings, this section only focuses on the overflow or inflow relationship between the surface and the pipeline network.

[0056] By traversing in the above order, it can be strictly ensured that objects involved by multiple coupled interfaces (such as a surface grid connecting both a pipeline network and a river segment) are grouped into the same coupled node. For a composite coupled node where a grid node is coupled to both a pipeline network node and a river segment node... The value is taken from the coupling interface involved. The average value.

[0057] Optionally, by traversing in the above order, various types of model units can be aggregated into four types of nodes: the coupled nodes generated by aggregation (also called C-Nodes), the deleted initial river segment nodes (i.e., the final river segment nodes after two traversals), also called river segment nodes or R-Nodes, the deleted initial surface grid nodes (i.e., the final surface grid nodes after two traversals), also called grid nodes or G-Nodes, and the deleted initial pipe network nodes (i.e., the final pipe network nodes after two traversals), also called pipe network nodes or P-Nodes. Among them, the coupled nodes forcibly aggregate all model units with actual hydraulic interactions.

[0058] Subsequently, P-Node, G-Node, R-Node, and the coupled node C-Node are combined to form a vertex set V, and an edge set E is constructed based on the physical topological relationships between the model units contained in these nodes. This results in the construction of a spatial topological graph G=(V, E), which fully represents the computational domain structure of the target region.

[0059] This embodiment uses the above two-level traversal aggregation mechanism to abstract the original massive and heterogeneous units into four types of nodes with physical consistency. By locking the strong hydraulic interaction units inside the coupled nodes, the massive cross-process communication requirements caused by cutting frequent interaction boundaries in parallel computing are eliminated, ensuring the accuracy and parallel efficiency of simulation computing from the topology level.

[0060] After establishing the spatial topology map, it is necessary to determine the weight of each node under different flood scenarios based on the hydraulic element information of each model component under various flood scenarios. This weight will serve as the core criterion for subsequent graph partitioning to achieve load balancing. The weight calculation here is dynamically performed using the hydraulic element information of the model components corresponding to the model units in each node under different flood scenarios, as well as the number of model units contained in each node.

[0061] For example, in one possible implementation, determining the weight of each node in the spatial topology map under each flood scenario based on the hydraulic element information includes: For each type of model unit, the basic weight of the model unit under each flood scenario is calculated based on the calculated strength coefficient of the model component corresponding to the model unit and the hydraulic element information of the model component corresponding to the model unit under each flood scenario. For each of the surface grid nodes, river segment nodes, and pipeline nodes, the weight of the node in each flood scenario is calculated based on the number of model units contained within the node and the basic weight of the same type of model units contained within the node in each flood scenario. For the coupled node, the weight of the coupled node in each flood scenario is calculated based on the sum of the basic weights of all types of model units contained in the coupled node under each flood scenario, and the coupling computation additional overhead coefficient of the coupled node; the coupling computation additional overhead coefficient is calculated based on the average coupling strength coefficient of all coupled interfaces contained in the coupled node.

[0062] Optionally, for each of the surface grid nodes, river segment nodes, and pipeline nodes, its weight under each flood scenario is... Specifically, it can be calculated using the following formula: ; in, The number of model units of a single type aggregated by this node; The computational intensity coefficient (also known as the model type coefficient) of the model components of the corresponding model type aggregated for this node reflects the difference in average computational intensity of each physical unit of different types of model components (e.g., the unit computation time is different for two-dimensional meshes, one-dimensional river segments, and pipeline nodes). It can be obtained through benchmark testing, such as selecting typical examples to run in the same hardware environment and calculating the average single-step time, with two-dimensional mesh as the benchmark. =1), then the one-dimensional river segment Approximately 0.2~0.4, pipeline network Approximately 0.1~0.3; and For the model components of the corresponding model type aggregated for this node, the water depth and regional reference water depth (such as the average maximum water depth of the study area) are determined from the hydraulic element information under the current flood scenario. For G-Node, the water depth of this node can be the average of the maximum water depths of all grids within the aggregate; for R-Node, the water depth of this node can be the average water depth of all cross sections; for P-Node, the water depth of this node is directly taken as the node water depth. This is the sensitivity coefficient, which can be set according to actual needs. For example, it can be set to... =1.0. This weight calculation reflects the sum of the static load and the hydraulic dynamic load of the region under the corresponding rainfall intensity.

[0063] For coupled nodes (C-Nodes), their weights under each flood scenario Specifically, it can be calculated using the following formula: ; The summation term iterates through all the model units contained in the C-Node; ω is the additional overhead coefficient for coupling computation, which is calculated based on the average coupling strength coefficient of all coupling interfaces contained in the coupling node; The basic overhead coefficient is used to reflect the basic computational overhead of the coupling interface. Its value ranges from 0.05 to 0.10, and can be calibrated through benchmark testing. For example, it can be set to 0.05. As a regulating factor, controlling The impact on additional overhead can typically be set to 1.0, making... =0 = , =1 = Therefore, by using this formula to calculate the weight of coupled nodes, the more frequently activated coupled interfaces can be assigned higher weights, thus more accurately reflecting their computational overhead.

[0064] This embodiment constructs a high-fidelity node weight model by introducing model type coefficients, dynamic water depth influence factors, and coupling computation additional overhead coefficients. This weight calculation scheme not only considers the number of model units but also deeply considers the difference in computational complexity between two-dimensional and one-dimensional models, as well as the dynamic contribution of water surface coverage to the computational load. This enables the graph partitioning algorithm to achieve true load balancing in a physical sense, shortening the synchronization waiting time in parallel computing.

[0065] After determining the weights of each node in the spatial topology graph, the flood scenarios can be partitioned according to these weights to generate corresponding first-level partitioning schemes for each flood scenario, ultimately constructing a first-level partitioning contingency plan library. It should be noted that when partitioning, while ensuring that coupled nodes are not split, the partitioning can be performed with the goal of balancing partition weights, or with the goal of minimizing the number of partition cutting edges while maintaining balanced partition weights. This embodiment does not impose specific limitations on these methods.

[0066] For example, in one possible implementation, obtaining the first-level zoning scheme of the target area under each of the flood scenarios according to the weights includes: For each flood scenario, the nodes in the spatial topology graph are partitioned with the constraint objective of minimizing the number of partition cutting edges and maximizing the balance between the sum of the weights of all nodes in each first-level computational partition under the flood scenario. Based on the partitioning results, a first-level partitioning scheme under the flood scenario is generated.

[0067] Optionally, for each flood scenario, the weights of each node are used as the vertex weights of the graph. The objective is to minimize the number of cut edges between partitions (i.e., minimize the communication boundary) and maximize the balance between the sum of the weights of all nodes within each first-level computational partition. The Graph Partitioning into M-parts (METIS) or Parallel Partitioning into M-parts (ParMETIS) tool is then used to perform N operations on the spatial topology graph G. zone Road partitioning is used to divide the computational space of the entire target region into N parts. zone One first-level computing partition (i.e., corresponding to N) zone (This involves multiple computational processes or nodes) to obtain the first-level zoning scheme for the flood scenario. After zoning, all first-level zoning schemes and their corresponding scenario identifiers are stored in the first-level zoning plan library. Since all model units with coupling relationships have been forcibly retained in the same node through aggregation, no additional constraints are required, and coupling integrity is naturally guaranteed.

[0068] It should be noted that when generating the first-level partitioning scheme, the partitioning results of the first-level computational partitions can be mapped back to the original model units. That is, the nodes in each partition directly correspond to their original model units; all the original model units contained in each coupled node are assigned to the same partition where the node is located.

[0069] In this embodiment, the partitioning is constrained by minimizing the number of partition cutting edges and maximizing weight balance. This ensures consistent computation time across all computing nodes while minimizing data exchange between nodes. This optimization objective ensures that the large-scale computing cluster can achieve higher parallel speedup and system throughput when processing ultra-high resolution flood models.

[0070] Additionally, it should be noted that the secondary partitioning plan library is constructed for areas within each primary computing partition corresponding to each primary partitioning scheme that may form computing hotspots (such as areas frequently traversed by flood fronts, floodplains, and waterlogged areas). Before the overall calculation begins, secondary partitioning schemes are pre-generated using a graph partitioning tool based on the internal topology of each primary computing partition corresponding to each primary partitioning scheme. In other words, secondary partitioning is a re-division within the primary partitions, used for rapid activation when the primary partition load exceeds the limit during calculation. The internal units of coupled nodes (C-Nodes) cannot be further divided; therefore, secondary partitioning only affects the recombination of independent unit nodes and coupled nodes, while the coupled nodes themselves remain intact. Each secondary partitioning scheme is associated with its corresponding primary partition and triggering conditions (i.e., the triggering conditions for activating the secondary partition, such as load thresholds), and all are stored in the secondary partitioning plan library. The generation of secondary partitioning schemes does not depend on specific rainstorm scenarios but is based on a general analysis of the internal structure of the primary partitions. Within each primary partition, its internal nodes are pre-divided into several secondary partitions using a graph partitioning tool to generate a secondary partitioning plan. In order to achieve the best balance between load balancing effect and communication overhead and to achieve simplicity and efficiency in engineering, it can be divided into 2 to 4 partitions, preferably 2 secondary partitions.

[0071] Furthermore, after obtaining the primary and secondary zoning contingency plan libraries through offline construction, the real-time application stage can be entered. In this stage, real-time weather forecast information of the target area, such as return period and rainfall distribution, can be obtained in real time. Based on the scenario characteristics corresponding to the real-time weather forecast information, the primary zoning scheme with the closest flood scenario can be dynamically matched from the primary zoning contingency plan library as the target primary zoning scheme.

[0072] Step 120: According to the target first-level zoning scheme, the spatial computing domain of the target area is divided into multiple first-level computing partitions, and the simulation calculation of flood risk factors is performed in parallel on the multiple first-level computing partitions.

[0073] After obtaining the target primary partitioning scheme, multiple master computing nodes (i.e., computing processes) can be launched. Each master computing node loads model data for one primary computing partition to simulate and calculate the flood risk factors of each primary computing partition under the current scenario in parallel. These flood risk factors include, but are not limited to, water depth, flow velocity, inundation extent, and flood risk level. It should be noted that during the simulation calculation, each master computing node can also interact with data at its respective partition boundaries to ensure the continuity of the physical process.

[0074] For example, in order to maximize the hiding of data exchange overhead in intervals while ensuring computational accuracy, in one possible embodiment, the method further includes: Real-time acquisition of the maximum hydraulic wave velocity and minimum spatial step size of each type of model unit within each primary computing partition; Based on the maximum hydraulic wave velocity, the minimum spatial step size, and the communication step size of the computing nodes corresponding to each first-level computing partition, the number of overlapping layers of each type of model unit in each first-level computing partition is calculated, and an extended overlapping region is constructed for each first-level computing partition based on the number of overlapping layers. The extended overlapping region is used to exchange boundary information between partitions to ensure that physical disturbances do not propagate beyond the boundary of the overlapping region within the communication interval.

[0075] Upon detecting the completion of the simulation calculation in the core area of ​​any level computing partition, an asynchronous communication task is initiated for the aforementioned level computing partition, and the pending computing tasks within the aforementioned level computing partition continue to be executed synchronously; the pending computing tasks refer to auxiliary computing tasks or simulation computing tasks at the next time point that are not dependent on the asynchronous communication task. The asynchronous communication task includes a receiving task and / or a sending task; the receiving task is used to receive a first flood risk factor of the first extended overlapping area simulated in the main computing partition that has the ownership of the first extended overlapping area of ​​the arbitrary first-level computing partition, and to cover the second flood risk factor of the first extended overlapping area simulated in the arbitrary first-level computing partition with the first flood risk factor; the sending task is used to send the flood risk factors of the second extended overlapping areas of other adjacent computing partitions simulated in the arbitrary first-level computing partition.

[0076] Optionally, to address the issue that traditional methods rely on empirically determined overlapping layers (e.g., 4 layers) and fail to consider differences in hydraulic wave velocities across different regions, leading to redundant calculations in slow-flow areas and insufficient accuracy in fast-flow areas, this embodiment can base calculations on the maximum hydraulic wave velocity of each type of model unit within each first-level calculation partition. The number of overlapping layers is determined by adaptation, in order to calculate the required number of overlapping layers for each type of model unit within each first-level computational partition. : ; in, Communication step size (total physical duration corresponding to 2-5 computation time steps); This refers to the minimum spatial step size (such as the size of a two-dimensional mesh, the length of a one-dimensional river segment, or the length of a pipe) for this type of model unit within the partition. For safety margin, its value is directly related to the numerical format of the model. For the common finite volume method, if the template width is m layers (that is, to calculate one element, information from the surrounding m layers of elements is required). =m (2 for second-order Godunov format, 3 for fifth-order Weno format). The physical meaning of this formula is to ensure that the communication interval... Within the core region, physical disturbances (i.e., disturbances propagating at wave speeds) will not propagate beyond the overlapping area, thus preventing boundary numerical errors from affecting the core region. Different partitions are defined according to the various types of model units within them. and Calculate different This allows for defining different numbers of overlapping layers for grid units, river segment units, and pipeline units within different partitions, respectively. , , .

[0077] After obtaining the number of overlapping layers for each type of model element within each first-level computational partition, extended overlapping regions can be constructed for each first-level computational partition. Specifically, the mesh elements within a first-level computational partition can be extended outwards. Adjacent grid layers can be extended outwards for river segment units within a first-level computational partition. Each connected section extends outward from the pipeline unit within the first-level calculation zone. Connected nodes are used to ensure that physical disturbances within the communication interval do not propagate beyond the overlap area boundary. It should be noted that during the expansion process, if coupled nodes belonging to different first-level partitions are encountered, all model units contained in that coupled node will be included in the same expansion overlap area.

[0078] During parallel computing execution, an asynchronous communication pipeline scheduling mechanism is employed, utilizing a non-blocking message passing interface (MPI) to achieve overlap between computation and communication. Specifically, after completing the simulation computation of its internal core area, each first-level computing partition immediately initiates asynchronous sending and receiving of data in the overlapping area. Simultaneously, it executes auxiliary computing tasks or the next time step computation directly without waiting for communication to complete, thus hiding communication latency. For the same entity (i.e., the same model unit) within the overlapping area, a master-slave update strategy is used for assimilation. The partition possessing ownership of the entity becomes the master partition, and its computation results overwrite the corresponding results of neighboring partitions (slave partitions), ensuring consistency of global variables.

[0079] This embodiment uses a wave speed-driven adaptive overlap region determination mechanism to accurately match the physical disturbance propagation speed with the discrete grid scale. While ensuring the continuity of physical topology across partitions, it effectively hides network transmission latency by combining asynchronous communication pipelines, significantly improving the speedup ratio of parallel computing.

[0080] Furthermore, existing parallel systems suffer from overall simulation interruption if a computing node fails, failing to meet the stability requirements of real-time forecasting. Therefore, to ensure high reliability of real-time flood forecasting tasks in complex hardware environments, in some possible implementations, the parallel execution of flood risk factor simulation calculations across multiple primary computing partitions includes: Multiple master computing nodes are invoked to perform parallel simulation calculations of flood risk factors on the multiple primary computing partitions; each primary computing partition corresponds one-to-one with a master computing node. The hot standby nodes of each of the main computing nodes are invoked to collect the heartbeat signals and checkpoint data generated by each of the main computing nodes in real time during the simulation calculation process; When the number of lost heartbeat signals of any master computing node reaches a preset threshold, the hot standby node of the master computing node is driven to load the latest checkpoint data of the master computing node and take over the simulation computing task of the master computing node. After the hot standby node of any primary computing node completes data loading and task takeover, the communication connection between the hot standby node of any primary computing node and other computing nodes other than the primary computing node is updated.

[0081] Optionally, at least one hot standby node can be configured for each primary computing node performing computational tasks. During the simulation, each primary computing node can send a heartbeat signal to the hot standby node at preset time intervals (e.g., once every 2 seconds) and continuously store multiple (e.g., 3) recent checkpoint data to shared storage or the hot standby node's memory at preset computational step sizes (e.g., every 60 time steps). When the monitoring module detects that the number of consecutive heartbeat signal losses of a primary computing node reaches a preset threshold (e.g., 3 times), it determines that the primary computing node has experienced a hardware failure or process crash. At this time, the hot standby node is immediately activated, loads the latest checkpoint data to restore the computational state, takes over the original partitioned simulation tasks, and notifies adjacent computing nodes to modify the communication topology of the message passing interface to achieve second-level fault switching and ensure uninterrupted simulation.

[0082] This embodiment achieves second-level fault self-healing of the parallel computing system through hot standby redundancy and heartbeat monitoring mechanisms, effectively solving the problem of overall simulation task interruption caused by single-point failure in large-scale distributed simulation, and ensuring the continuity and stability of flood risk map generation.

[0083] Step 130: If, during the simulation calculation process, an abnormal real-time computing load index value is detected in a target first-level computing partition among the multiple first-level computing partitions, then a target second-level partition scheme corresponding to the target first-level computing partition is matched from the second-level partition scheme library.

[0084] Optionally, as rainfall evolves over time, the flood front moves continuously within the computational domain, causing dynamic migration of real-time computational load index values ​​in different partitions. Therefore, a real-time load monitoring mechanism can capture the real-time computational load index values ​​in the primary computational partitions. When the real-time computational load index value of any primary computational partition exceeds the limit, the partition computation scheme can be adjusted in a timely manner to achieve adaptive scheduling of computing resources and rebalancing of internal resources.

[0085] For example, in order to accurately characterize the computational pressure by comprehensively considering both the physical hydraulic conditions and the underlying hardware resource status, in some possible embodiments, the step of obtaining the real-time computational load index value of each first-level computing partition includes: The real-time hydraulic element information of each model unit within the first-level calculation partition is statistically analyzed, and the real-time hydraulic element information includes real-time water depth and real-time flow velocity. Obtain hardware resource utilization information of the computing nodes that perform parallel simulation computing in the first-level computing partition, the hardware resource utilization information including processor utilization and memory bandwidth utilization; Based on the weighting coefficients corresponding to the real-time hydraulic element information and the weighting coefficients corresponding to the hardware resource utilization information, a multi-factor weighted fusion is performed on the real-time hydraulic element information and the hardware resource utilization information to obtain the real-time computing load index value of the first-level computing partition.

[0086] Optionally, during flood calculation, taking into account both hydraulic information and hardware resource consumption, the real-time computational load index value of each primary computational partition is statistically analyzed at certain time intervals (e.g., every 100 computation time steps or every 60 seconds of computation time). The specific calculation formula is as follows: ; in, , , These represent the real-time water depths for grid cells, river segment cells, and pipeline cells, respectively. , , These are the real-time flow velocities for grid cells, river segment cells, and pipeline cells, respectively. , , , The weighting coefficients corresponding to real-time hydraulic element information can be obtained through statistical calibration of the time consumption of multiple sets (e.g., 30 sets) of standard test cases. =0.45、 =0.35、 =0.15、 =0.25; ζ represents the weighting coefficient corresponding to the hardware resource utilization information, and its value range is usually 0.1~0.5. The specific value can be dynamically adjusted according to the model of the computing node. This refers to processor utilization, such as graphics processing unit (GPU) utilization. This represents the memory bandwidth utilization of the computing node.

[0087] In this embodiment, by performing multi-factor weighted fusion of real-time hydraulic element information and computing node hardware resource utilization, the accurate perception of computing bottlenecks is achieved. This enables the keen detection of computing hotspot migration caused by flood evolution, providing a reliable triggering criterion for accurate triggering of secondary partitions and avoiding unnecessary partition adjustment overhead.

[0088] After obtaining the real-time computing load metric values ​​for each primary computing partition, the average of the real-time computing load metric values ​​for all primary computing partitions can be calculated to obtain the global average load. avg It also calculates the real-time computing load index values ​​of each primary computing partition and the global average load. avg If the positive deviation between the two exceeds a preset threshold (e.g., 20%), and the positive deviation of any first-level computing partition exceeds the preset threshold, then the real-time computing load index value of that first-level computing partition is determined to be abnormal, and it is taken as the target first-level computing partition. The target second-level partition scheme corresponding to the target first-level computing partition is matched from the second-level partition scheme library to start the second-level partition adjustment for the target first-level computing partition.

[0089] Step 140: According to the target secondary zoning scheme, the target primary computing zoning is decomposed into multiple secondary computing zoning, and the simulation calculation of flood risk factors for the target primary computing zoning is switched to parallel simulation calculation of flood risk factors for the multiple secondary computing zoning.

[0090] Optionally, after obtaining the target secondary partitioning scheme, the overloaded target primary computing partition can be broken down into multiple secondary computing partitions based on the target secondary partitioning scheme. These secondary partitions can then be allocated to computing nodes with idle computing resources to share the computing tasks. This switches the simulation calculation of flood risk factors for the target primary computing partition to parallel simulation calculation of flood risk factors for multiple secondary computing partitions, achieving a response time within seconds. Here, the secondary partitioning only affects the primary partition itself and does not affect the boundaries or overlapping areas of other partitions. To avoid additional overhead caused by frequent adjustments to computing partitions, the partition adjustment interval needs to be greater than 60 computation time steps.

[0091] In addition, in order to save computing resources and in combination with model algorithm optimization, partitions in which the water depth of all model units within a partition is less than the dry water depth threshold (e.g., 0.001m) and the water depth in the overlapping area is also less than the dry water depth threshold are identified as dry unit partitions. When performing simulation calculations, the hydraulic state of the previous time step can be directly copied.

[0092] Step 150: Generate a flood risk map of the target area based on the simulation results of each of the first-level calculation zones and the simulation results of each of the second-level calculation zones.

[0093] Optionally, at each computation time step or a specified output interval, simulation results for each partition, such as water depth, flow velocity, inundation range, and risk level, can be collected and aggregated to the master node through parallel reduction operations or directly using distributed rendering technology to generate the following risk map elements in real time: flood inundation range boundaries determined based on a water depth threshold (e.g., 0.05m); water depth distribution map; flow velocity vector field; and risk level partitions (e.g., low, medium, and high risk zones). The generated flood risk map can be pushed to WebGIS or display terminals in real time, achieving second-level updates of the dynamic flood risk map.

[0094] The method provided in this embodiment identifies strongly interactive coupling interfaces through multi-scenario physical pre-analysis, forcibly aggregates heterogeneous model units into indivisible coupled nodes to ensure the integrity of the physical process, and constructs a two-level partitioned contingency plan system. This achieves a deep integration of initial global scenario matching and dynamic decomposition of local hotspot areas during the calculation process. Thus, while avoiding the interruption of hydraulic connections from the bottom-level topology, it fundamentally solves the problem of dynamic load imbalance caused by the movement of flood fronts, significantly improving the parallel acceleration efficiency, resource utilization, numerical accuracy, and system stability of dynamic flood risk map generation under multi-model coupling.

[0095] To further verify the effectiveness of the method provided in this application, the following explanation will be based on a typical city.

[0096] The city's central urban area has a highly undulating terrain, with multiple rivers flowing through it and a relatively complete drainage network system. This urban flood simulation employs a coupled hydrodynamic model of a one-dimensional river network, drainage network, and two-dimensional surface data, covering a modeling area of ​​approximately 600 km². 2 The grid is divided into approximately 6 million grids, and takes into account approximately 280 km of drainage channels (700 channel sections) and approximately 200,000 drainage pipe sections (corresponding to approximately 200,000 inspection wells).

[0097] Figure 2 This is the second flowchart of the partitioned parallel calculation method for dynamic flood risk maps provided by the present invention.

[0098] like Figure 2As shown, the method provided in this application can realize real-time dynamic simulation of rainstorm flooding risk and rapid generation of risk maps. The specific implementation steps are as follows: Step A: Multi-scenario offline physical pre-analysis. Three rainstorm scenarios were designed and selected: Scenario A (frequent light rain, 2-year return period, 24-hour rainfall of 65 mm), Scenario B (frequent rainstorm, 10-year return period, 24-hour rainfall of 174 mm), and Scenario C (extreme heavy rainstorm, 100-year return period, 24-hour rainfall of 298 mm). Using the constructed coupled model, the three scenarios were simulated (time step of 1 second). The activation status (whether water exchange occurred) of each coupled interface at each time step under each flood scenario was recorded, and the hydraulic element information of each model component under each flood scenario, including maximum water depth, was statistically analyzed. Maximum flow rate Calculate the maximum wave speed The total duration of the three scenario simulations is 24 hours, that is... =24×3600s = 43,200s.

[0099] Step B, based on the coupling strength coefficient of the total activation duration Calculation. Based on the above formula for calculating the coupling strength coefficient, calculate the coupling strength coefficient for each coupling interface. Value. Set the target water volume (also known as the water volume threshold). = 0.1 m 3 (That is, the water exchange volume in a single step is less than 0.1 m³) 3 (Considered as inactive). Scenario weights are normalized based on the inverse of the recurrence period and take into account the importance of extreme events, such as taking... = 0.5 (Scenario A, once every 2 years). = 0.3 (Scenario B, once in 10 years). = 0.2 (Scenario C, once in 100 years).

[0100] Simulation results show that approximately 78% of the inspection wells did not overflow under any of the three scenarios. =0), the remaining 22% of inspection wells overflowed in at least one scenario; approximately 85% of river sections experienced lateral inflow at least in extreme scenarios. Therefore The total number of coupling interfaces with a value greater than 0 is approximately 200,000 × 0.22 + 700 × 0.85 = 44,000 + 595 = 44,595.

[0101] Using the PG-2325 coupling interface as an example, this illustrates the activation duration percentage Rs and the coupling strength coefficient. Calculation method: Scenario A activation steps 0, =0; Scenario B is activated in 2400 steps. =2400 / 43200≈0.0556; Scenario C requires 15200 activation steps. =15200 / 43200≈0.3519. Therefore... = 0.5×0 + 0.3×0.0556 + 0.2×0.3519 =0.08706 (>0).

[0102] Step C: Constructing a primary partition and contingency plan library based on coupled node aggregation. First, perform similar aggregation on independent units (including grid units, pipeline units, and river segment units): every 5 connected river segment units are aggregated into one river segment node, totaling 700 / 5=140; every 100 adjacent grid units are aggregated into one grid node, totaling 6,000,000 / 100=60,000; every 50 connected pipeline units are aggregated into one pipeline node, totaling 200,000 / 50=4,000.

[0103] Secondly, identify and construct coupled nodes (C-Nodes): (1) Traverse all aggregated river segment nodes (i.e., 140 river segment nodes, each representing approximately 5 original river segment units), if there exists a coupled relationship with any river segment unit within that node and For grid cells and pipe network cells with a value >0, re-mark the river segment node as a coupling node C-Node, aggregate all grid cells and pipe network cells directly coupled to this node into this coupling node, and simultaneously delete all grid cells and pipe network cells directly coupled to this node from the original grid node and the original pipe network node's aggregate. (2) Traverse all aggregated grid nodes (i.e., 60,000 grid nodes, each node representing approximately 100 original grid cells), if there exists a grid cell directly coupled to any grid cell within the node and For network elements with a value >0, the node is re-marked as a coupled node (C-Node), and all network elements (or pipe elements) coupled to this node are aggregated into it. Simultaneously, the pipe elements coupled to this node are removed from the original network element's aggregate. For composite coupled nodes (i.e., nodes that simultaneously contain grid nodes, network elements, and river nodes), ... Values ​​are taken from the interfaces involved. The average value, for example, a G-Node simultaneously with a P-Node ( =0.09) and an R-Node ( If the coupling is 0.15, then the three elements merge into a C-Node. = (0.09+0.15) / 2 = 0.12).

[0104] Figure 3 This is a schematic diagram of the aggregate of river section nodes provided by the present invention. Figure 4 This is a schematic diagram of the aggregate of mesh nodes provided by the present invention. Figure 5 This is a schematic diagram of the aggregate of pipeline nodes provided by the present invention. Figure 6 This is a schematic diagram of the aggregate of coupling nodes provided by the present invention.

[0105] like Figure 3-6 As shown, by traversing in the above order, various types of model units can be aggregated into four types of nodes: C-Node, R-Node, G-Node, and P-Node.

[0106] Next, construct the aggregated undirected graph (also known as a spatial topological graph). The vertex set V of the undirected graph G consists of P-Node, R-Node, G-Node and all C-Nodes, and the edge set E represents the topological connections between adjacent nodes (without considering weights).

[0107] Secondly, node weights are calculated. Model type coefficients are calibrated through benchmark testing. Using a two-dimensional grid cell as the baseline, the single-step time is 0.15 ms for a grid cell, 0.04 ms for a river segment cell, and 0.025 ms for a pipeline cell. Therefore, the time consumption for the two-dimensional model, the river channel model (also known as the river segment model), and the pipeline model is... The values ​​are 1.0, 0.27, and 0.17, respectively. Regional reference water depth. =3.0 m, sensitivity coefficient =1.0. Basic additional overhead factor =0.06, adjustment factor =1.0. Taking scenario B (once in 10 years) as an example: The relevant parameters for any G-Node are, =0.45 m, =100, then its corresponding weight is: = 1.0×100×(1+0.45 / 3)= 115.0.

[0108] Any C-Node (containing 5 pipe elements and 100 mesh elements), its internal pipe elements... =1.5m, grid cell =1.2 m, =0.32. Basic component: Pipe element contribution 0.17×5×(1+1.5 / 3)=1.275, mesh element contribution 1.0×100×(1+1.2 / 3)=140, total 141.275. ω= 0.06×(1+1.0×0.32)=0.0792, = 141.275×1.0792≈152.46.

[0109] Subsequently, a primary-level zonal contingency plan library was generated for various design flood scenarios: for scenarios A, B, and C, the weights of all nodes were recalculated. Taking representative water depths for each scenario, ParMETIS is used to perform 16-way partitioning on the undirected graph G (corresponding to 16 GPUs), resulting in three first-level partitioning schemes (A_16, B_16, C_16), which are stored in the first-level partitioning scheme library, with B_16 as the default initial partition.

[0110] First-level partition result mapping: Map the partition results back to the original cells. Each independent cell node corresponds to its original cell, and all the original cells contained in each C-Node are assigned to the same partition where that node is located.

[0111] Step D: Determining the number of adaptive overlap layers based on wave velocity. Based on the wave velocity distribution recorded in Step A, calculate the required number of overlap layers for each type of model element within each first-level computational partition. Taking a section along a certain river channel as an example: Surface =6.0 m / s, communication step size =3 steps × 1s / step = 3s, minimum spatial step size =10 m, numerical format is second-order Godunov ( =2), then = ceil (6.0 × 3 / 10) + 2 = 4 layers. Using a similar method, the number of expansion layers for the river segment unit and the pipeline unit are calculated separately: for the river channel unit (i.e., the river segment unit), the minimum spatial step size is the minimum cross-sectional spacing. ≈400 m, the numerical scheme adopts the central scheme finite difference method ( =1), then =2; For pipe elements, the spatial step size is the pipe length. ≈10 m, the numerical scheme adopts the central scheme finite difference method ( =1), then =3.

[0112] Figure 7 This is a schematic diagram of the calculation area and the overlapping area provided by the present invention.

[0113] Step E: Construct a contingency plan library for the extended overlap area and secondary partitioning. Based on the calculated... , , For each primary partition, an extended overlap area is constructed: grid cells expand outwards. Layers, river segment units expand outwards Each cross-section, the pipeline unit extends outwards. Each node, specifically as follows Figure 7As shown. If the expansion encounters coupled nodes belonging to different first-level partitions, all units of that node are included in the overlapping area. For the 16 first-level partitions, potential hotspot areas (such as historically flood-prone areas) are identified. In advance, METIS is used to re-divide the independent unit nodes and coupled nodes (C-Node as a whole) within each first-level partition into two second-level sub-partitions, generating 16 sets of second-level partition plans (one set of two sub-partition plans for each first-level partition), which are stored in the second-level partition plan library.

[0114] Step F involves cross-partition computation based on asynchronous communication and hot standby redundancy. Before real-time computation, based on real-time weather forecast information (the rainfall is expected to reach 120mm, lasting 6 hours), the closest "Scenario B (10-year return period)" partitioning scheme is loaded from the primary partitioning scheme library. Sixteen GPU processes are started, each responsible for one primary partition. Non-blocking MPI is used to implement asynchronous communication, that is, after calculating the core area at each time step (1s), data exchange in the overlapping area is started immediately, and the computation continues to the next time step. Two hot standby nodes are configured, saving a checkpoint every 60 steps (60s), and sending a heartbeat signal every 2s. When a GPU (such as GPU 7) fails, the hot standby node loads the latest checkpoint after three consecutive heartbeat losses (about 6s), takes over partition 7, and updates the communication topology. The interruption time is about 8s, which meets the real-time requirements.

[0115] Figure 8 This is the third flowchart of the partitioned parallel calculation method for dynamic flood risk maps provided by this invention.

[0116] Step G involves dynamic secondary zoning adjustments based on hydraulic and hardware loads. For example... Figure 8 As shown, at time step 7200 (2 hours), heavy rainfall occurred in the southwest, causing a rapid increase in water depth in multiple grids within the 5th primary partition, resulting in a sharp rise in computational load. The load monitoring module calculates the load every 100 steps. The value (Formula 7) is calculated using the hydraulic weighting coefficient. =0.45、 =0.35、 =0.15、 =0.25, hardware coefficient =0.2, ζ=0.1 (the hardware weighting coefficients in this embodiment are based on the NVIDIA A100 GPU calibration; other models can be adjusted through benchmark testing), and the first-level partition is measured. Reaching 1.2 times the global average, that is / avg = 1.2, exceeding the preset threshold (positive deviation 20%, i.e., L>1.2×). (avg). The master node immediately retrieves the plan for partition number 5 from the secondary partition plan library, splits it into two secondary sub-partitions, and allocates them to two idle CPU cores for auxiliary computation. The load quickly returns to normal. During the splitting process, coupled nodes remain intact, with no loss of accuracy.

[0117] Step H: Real-time dynamic flood risk map generation. Every 300 time steps (300 seconds), water depth and flow velocity results for each zone are collected, summarized to the main node, and inundation range map, water depth distribution map, flow velocity vector field, and risk level map are generated in real time and pushed to the WebGIS platform. This simulation ran for a total of 21,600 time steps (6 hours of actual rainfall), with a total time of approximately 6.2 minutes (including dynamic adjustments), meeting the real-time forecast timeliness requirements.

[0118] Compared to traditional static geometric partitioning methods, the method provided in this application reduces load balancing from ±38% to ±10%, reduces cross-partition communication by 45%, and improves computation speed by approximately 1.6 times. The coupled node aggregation strategy ensures that all units with actual water exchange capabilities are always located in the same partition, guaranteeing model accuracy. The hot standby mechanism ensures the reliability of long-term simulations, preventing interruptions due to single-point failures.

[0119] In summary, the method provided in this application solves the problems of fragmented multi-model components, high communication overhead, mechanical partitioning constraints, lack of physical basis for the number of overlapping layers, difficulty in adaptive adjustment of uneven load, and lack of fault tolerance mechanism in existing parallel technologies. It achieves fast, stable, and reliable parallel generation of dynamic flood risk maps. The method provided in this application has the following specific advantages: (1) Physically driven intelligent partitioning ensures the integrity of coupling from the source. Through multi-scenario offline pre-analysis, all actual water exchange coupling interfaces (S>0) are identified, reducing the amount of coupling calculation, and the units involved are forcibly aggregated into indivisible coupling nodes (C-Node), fundamentally eliminating the possibility of coupling relationships being cut across partitions, and avoiding the accuracy loss caused by improper threshold or edge weight settings in traditional graph partitioning.

[0120] (2) Layered adaptive load balancing effectively improves computational efficiency. The node weight comprehensively considers the number of units, model type, water depth dynamics and coupling strength S, making the load prediction closer to reality; the combination of the first-level partition plan library (multi-scenario adaptation) and the second-level partition dynamic adjustment (multi-sub-region fast response) realizes global and local dual load balancing.

[0121] (3) Wave speed driven overlapping area and asynchronous hot standby, balancing accuracy and reliability. The number of overlapping layers is accurately calculated based on wave speed, communication step size and spatial step size, replacing the coarse method of fixed number of layers; non-blocking MPI is used to realize the overlap of calculation and communication, hiding the communication delay; hot standby node and checkpoint mechanism are configured to realize second-level fault switching, ensuring uninterrupted real-time simulation for a long time.

[0122] (4) Integrated dynamic risk map generation to support real-time forecasting and decision-making. The simulation results are converted into risk map elements such as inundation range, water depth cloud map, flow velocity field, and risk level in real time, and pushed to the WebGIS platform to form a complete closed loop from simulation to output, which meets the dual requirements of timeliness and visualization for urban flood control.

[0123] The following describes the partitioned parallel computing system for dynamic flood risk maps provided by the present invention. The partitioned parallel computing system for dynamic flood risk maps described below can be referred to in correspondence with the partitioned parallel computing method for dynamic flood risk maps described above.

[0124] Figure 9 This is a schematic diagram of the partitioned parallel computing system for dynamic flood risk maps provided by the present invention.

[0125] like Figure 9 As shown, the system includes: The first matching unit 910 is used to match the target first-level zoning scheme from the first-level zoning plan database based on the real-time weather forecast information of the target area. The first computing unit 920 is used to divide the spatial computing domain of the target area into multiple first-level computing partitions according to the target first-level zoning scheme, and to perform simulation calculations of flood risk factors in parallel on the multiple first-level computing partitions. The second matching unit 930 is used to match the target secondary partition scheme corresponding to the target primary computing partition from the secondary partition scheme library if an abnormal real-time computing load index value is detected in the multiple primary computing partitions during the simulation calculation process. The second computing unit 940 is used to decompose the target primary computing partition into multiple secondary computing partitions according to the target secondary zoning scheme, and switch the simulation calculation of flood risk factors for the target primary computing partition to the parallel simulation calculation of flood risk factors for the multiple secondary computing partitions. The generation unit 950 is used to generate a flood risk map of the target area based on the simulation calculation results of each of the first-level calculation partitions and the simulation calculation results of each of the second-level calculation partitions; The primary partitioning plan library includes multiple primary partitioning schemes. Each primary partitioning scheme is obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios, while keeping the coupling nodes intact. The secondary partitioning plan library includes secondary partitioning schemes pre-constructed for the topology structure of each primary calculation partition corresponding to each primary partitioning scheme. The spatial topology map includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes. The coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

[0126] The system provided in this embodiment identifies strongly interactive coupling interfaces through multi-scenario physical pre-analysis, forcibly aggregating heterogeneous model units into indivisible coupled nodes to ensure the integrity of the physical process, and constructs a two-level partitioned contingency plan system. This achieves a deep integration of initial global scenario matching and dynamic decomposition of local hotspot areas during the calculation process. Thus, while avoiding the interruption of hydraulic connections from the underlying topology, it fundamentally solves the problem of dynamic load imbalance caused by the movement of flood fronts, significantly improving the parallel acceleration efficiency, resource utilization, numerical accuracy, and system stability of dynamic flood risk map generation under multi-model coupling.

[0127] The apparatus provided by the present invention is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0128] Figure 10 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 10As shown, the electronic device may include: a processor 1010, a communication interface 1020, a memory 1030, and a communication bus 1040, wherein the processor 1010, the communication interface 1020, and the memory 1030 communicate with each other through the communication bus 1040. The processor 1010 can call logical instructions in the memory 1030 to execute a zonal parallel calculation method for a dynamic flood risk map. This method includes: matching a target first-level zoning scheme from a first-level zoning plan library based on real-time weather forecast information of the target area; dividing the spatial computing domain of the target area into multiple first-level computing zones according to the target first-level zoning scheme, and performing parallel simulation calculations of flood risk factors on the multiple first-level computing zones; if, during the simulation calculation, an abnormal real-time computing load index value is detected in a target first-level computing zone among the multiple first-level computing zones, matching a target second-level zoning scheme corresponding to the target first-level computing zone from a second-level zoning plan library; and decomposing the target first-level computing zone into multiple second-level computing zones according to the target second-level zoning scheme, and performing calculations for the target first-level computing zone... The simulation calculation of flood risk factors is switched to parallel simulation calculation of flood risk factors for the multiple secondary calculation partitions; based on the simulation calculation results of each primary calculation partition and each secondary calculation partition, a flood risk map of the target area is generated; the primary partition plan library includes multiple primary partition schemes, each of which is obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios while keeping the coupling nodes intact; the secondary partition plan library includes secondary partition schemes pre-constructed for the topology structure of each primary calculation partition corresponding to each primary partition scheme; the spatial topology map includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes, and the coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

[0129] Furthermore, the logical instructions in the aforementioned memory 1030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the partitioned parallel calculation method for dynamic flood risk maps provided by the above methods. This method includes: matching a target first-level partition scheme from a first-level partition plan library based on real-time weather forecast information of the target area; dividing the spatial computing domain of the target area into multiple first-level computing partitions according to the target first-level partition scheme, and performing parallel simulation calculations of flood risk factors on the multiple first-level computing partitions; if a target first-level computing partition with an abnormal real-time computing load index value is detected in the multiple first-level computing partitions during the simulation calculation process, then matching a target second-level partition scheme corresponding to the target first-level computing partition from a second-level partition plan library; and performing parallel calculations of flood risk factors on the target second-level partitions according to the target second-level partition scheme. The primary computational partition is decomposed into multiple secondary computational partitions, and the simulation calculation of flood risk factors for the target primary computational partition is switched to parallel simulation calculation of flood risk factors for the multiple secondary computational partitions. A flood risk map of the target area is generated based on the simulation results of each primary and secondary computational partition. The primary partition contingency plan library includes multiple primary partition schemes, each obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios while maintaining the integrity of the coupling nodes. The secondary partition contingency plan library includes pre-constructed secondary partition schemes based on the topological structure of each primary computational partition corresponding to each primary partition scheme. The spatial topology map includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes. The coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

[0131] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a partitioned parallel computing method for dynamic flood risk maps provided by the methods described above. This method includes: matching a target first-level zoning scheme from a first-level zoning plan library based on real-time weather forecast information of the target area; dividing the spatial computing domain of the target area into multiple first-level computing partitions according to the target first-level zoning scheme, and performing parallel simulation calculations of flood risk factors on the multiple first-level computing partitions; if, during the simulation calculation, an abnormal real-time computing load index value is detected in a target first-level computing partition among the multiple first-level computing partitions, matching a target second-level zoning scheme corresponding to the target first-level computing partition from a second-level zoning plan library; and decomposing the target first-level computing partition into multiple second-level computing partitions according to the target second-level zoning scheme. The system partitions the target area and switches the simulation calculation of flood risk factors for the target primary calculation area to a parallel simulation calculation of flood risk factors for the multiple secondary calculation areas. Based on the simulation calculation results of each primary calculation area and each secondary calculation area, a flood risk map of the target area is generated. The primary calculation area plan library includes multiple primary calculation area schemes, each of which is obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios while keeping the coupling nodes intact. The secondary calculation area plan library includes secondary calculation area schemes pre-constructed for the topology structure of each primary calculation area corresponding to each primary calculation area scheme. The spatial topology map includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes. The coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

[0132] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0133] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for partitioned parallel computation of dynamic flood risk maps, characterized in that, include: Based on real-time weather forecast information for the target area, match the target first-level zoning plan from the first-level zoning plan database; According to the target first-level zoning scheme, the spatial computing domain of the target area is divided into multiple first-level computing zones, and the simulation calculation of flood risk factors is performed in parallel on the multiple first-level computing zones. If, during the simulation calculation process, an abnormal real-time computing load index value is detected in a target first-level computing partition among the multiple first-level computing partitions, then the target second-level partition scheme corresponding to the target first-level computing partition is matched from the second-level partition scheme library. According to the target secondary zoning scheme, the target primary calculation zone is decomposed into multiple secondary calculation zones, and the simulation calculation of flood risk factors for the target primary calculation zone is switched to parallel simulation calculation of flood risk factors for the multiple secondary calculation zones. Based on the simulation results of each of the first-level calculation zones and the simulation results of each of the second-level calculation zones, a flood risk map of the target area is generated; The primary partitioning plan library includes multiple primary partitioning schemes. Each primary partitioning scheme is obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios, while keeping the coupling nodes intact. The secondary partitioning plan library includes secondary partitioning schemes pre-constructed for the topology structure of each primary calculation partition corresponding to each primary partitioning scheme. The spatial topology diagram includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes. The coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

2. The method for partitioned parallel computation of dynamic flood risk maps according to claim 1, characterized in that, The steps for constructing the primary partition contingency plan database include: In multiple flood scenarios, offline hydrodynamic simulations are performed on the target area to obtain the hydraulic interaction information of the coupling interfaces between the model components under each flood scenario, as well as the hydraulic element information of each model component under each flood scenario. Based on the hydraulic interaction information, the cumulative time steps of hydraulic interaction of each coupling interface under each flood scenario are counted to obtain the total activation time of each coupling interface under each flood scenario; Based on the total activation duration, obtain the coupling strength coefficient of each of the coupling interfaces; Based on the coupling strength coefficient, the spatial topology graph is constructed; Based on the hydraulic element information, the weight of each node in the spatial topology map under each flood scenario is determined, and based on the weight, the first-level zoning scheme of the target area under each flood scenario is obtained. Based on all the aforementioned first-level partitioning schemes, construct the first-level partitioning plan library.

3. The method for partitioned parallel computation of dynamic flood risk maps according to claim 2, characterized in that, The step of obtaining the coupling strength coefficient of each coupling interface based on the total activation duration includes: For each coupling interface, if the cumulative water volume of hydraulic interaction occurring at the coupling interface under any flood scenario is less than or equal to the target water volume, the activation duration percentage of the coupling interface under any flood scenario is determined according to a preset percentage value. If the cumulative water volume is greater than the target water volume, the activation time ratio of the coupling interface in any flood scenario is determined based on the ratio between the total activation time of the coupling interface in any flood scenario and the total simulation time of the any flood scenario. The activation duration percentage of each of the coupling interfaces in each of the flood scenarios is obtained through iteration. Based on the scenario weights corresponding to each flood scenario, the activation duration percentages of each of the coupling interfaces in all flood scenarios are weighted and fused to obtain the coupling strength coefficient of each of the coupling interfaces. The scenario weights corresponding to each flood scenario are determined by normalizing the reciprocal of the recurrence period of each flood scenario.

4. The method for partitioned parallel computation of dynamic flood risk maps according to claim 2, characterized in that, The step of constructing the spatial topology map based on the coupling strength coefficient includes: Each model unit in each of the aforementioned model components is aggregated in the same category to obtain multiple initial independent nodes; the multiple initial independent nodes include initial river segment nodes generated by aggregating multiple connected river segment units, initial surface grid nodes generated by aggregating multiple adjacent grid units, and initial pipeline nodes generated by aggregating multiple connected pipeline units. For each initial river segment node, among the model units of each initial surface grid node and each initial pipeline node, a first model unit is determined that has a coupling interface with any model unit in the initial river segment node that has a coupling strength coefficient greater than a preset coefficient. The initial river segment node is aggregated with all the first model units to form the coupling node, and each of the first model units is deleted from its original surface grid node or initial pipeline node. For each deleted initial surface grid node, in each model unit of each deleted initial pipeline node, a second model unit with a coupling interface having a coupling strength coefficient greater than the preset coefficient with any model unit in the deleted initial surface grid node is identified, and the deleted initial surface grid node and all the second model units are aggregated into the coupling node. Each of the second model units is removed from its corresponding deleted initial network node; Based on the coupled nodes generated by aggregation, as well as the deleted initial river segment nodes, deleted initial surface grid nodes, and deleted initial pipeline nodes, a vertex set is constructed, and an edge set is constructed based on the topological relationships between the nodes in the vertex set; The spatial topology graph is constructed based on the vertex set and the edge set.

5. The method for partitioned parallel computation of dynamic flood risk maps according to claim 2, characterized in that, The step of determining the weight of each node in the spatial topology map under each flood scenario based on the hydraulic element information includes: For each type of model unit, the basic weight of the model unit under each flood scenario is calculated based on the calculated strength coefficient of the model component corresponding to the model unit and the hydraulic element information of the model component corresponding to the model unit under each flood scenario. For each of the surface grid nodes, river segment nodes, and pipeline nodes, the weight of the node in each flood scenario is calculated based on the number of model units contained within the node and the basic weight of the same type of model units contained within the node in each flood scenario. For the coupled node, the weight of the coupled node in each flood scenario is calculated based on the sum of the basic weights of all types of model units contained in the coupled node under each flood scenario, and the coupling computation additional overhead coefficient of the coupled node; the coupling computation additional overhead coefficient is calculated based on the average coupling strength coefficient of all coupled interfaces contained in the coupled node.

6. The method for partitioned parallel computation of dynamic flood risk maps according to claim 2, characterized in that, The step of obtaining the first-level zoning scheme of the target area under each flood scenario based on the weights includes: For each flood scenario, the nodes in the spatial topology graph are partitioned with the constraint objective of minimizing the number of partition cutting edges and maximizing the balance between the sum of the weights of all nodes in each first-level computational partition under the flood scenario. Based on the partitioning results, a first-level partitioning scheme under the flood scenario is generated.

7. The method for partitioned parallel computation of dynamic flood risk maps according to any one of claims 1-6, characterized in that, The method further includes: Real-time acquisition of the maximum hydraulic wave velocity and minimum spatial step size of each type of model unit within each primary computing partition; Based on the maximum hydraulic wave velocity, the minimum spatial step size, and the communication step size of the computing nodes corresponding to each first-level computing partition, calculate the number of overlapping layers of each type of model unit in each first-level computing partition, and construct an extended overlapping region for each first-level computing partition based on the number of overlapping layers. Upon detecting the completion of the simulation calculation in the core area of ​​any level computing partition, an asynchronous communication task is initiated for the aforementioned level computing partition, and the pending computing tasks within the aforementioned level computing partition continue to be executed synchronously; the pending computing tasks refer to auxiliary computing tasks or simulation computing tasks at the next time point that are not dependent on the asynchronous communication task. The asynchronous communication task includes a receiving task and / or a sending task; the receiving task is used to receive a first flood risk factor of the first extended overlapping area simulated in the main computing partition that has the ownership of the first extended overlapping area of ​​the arbitrary first-level computing partition, and to cover the second flood risk factor of the first extended overlapping area simulated in the arbitrary first-level computing partition with the first flood risk factor; the sending task is used to send the flood risk factors of the second extended overlapping areas of other adjacent computing partitions simulated in the arbitrary first-level computing partition.

8. The method for partitioned parallel computation of dynamic flood risk maps according to any one of claims 1-6, characterized in that, The parallel simulation calculation of flood risk factors for the multiple first-level computational partitions includes: Multiple master computing nodes are invoked to perform parallel simulation calculations of flood risk factors on the multiple primary computing partitions; each primary computing partition corresponds one-to-one with a master computing node. The hot standby nodes of each of the main computing nodes are invoked to collect the heartbeat signals and checkpoint data generated by each of the main computing nodes in real time during the simulation calculation process; When the number of lost heartbeat signals of any master computing node reaches a preset threshold, the hot standby node of the master computing node is driven to load the latest checkpoint data of the master computing node and take over the simulation computing task of the master computing node. After the hot standby node of any primary computing node completes data loading and task takeover, the communication connection between the hot standby node of any primary computing node and other computing nodes other than the primary computing node is updated.

9. The method for partitioned parallel computation of dynamic flood risk maps according to any one of claims 1-6, characterized in that, For each of the first-level computing partitions, the steps for obtaining the real-time computing load metric value of the first-level computing partition include: The real-time hydraulic element information of each model unit within the first-level calculation partition is statistically analyzed, and the real-time hydraulic element information includes real-time water depth and real-time flow velocity. Obtain hardware resource utilization information of the computing nodes that perform parallel simulation computing in the first-level computing partition, the hardware resource utilization information including processor utilization and memory bandwidth utilization; Based on the weighting coefficients corresponding to the real-time hydraulic element information and the weighting coefficients corresponding to the hardware resource utilization information, a multi-factor weighted fusion is performed on the real-time hydraulic element information and the hardware resource utilization information to obtain the real-time computing load index value of the first-level computing partition.

10. A partitioned parallel computing system for dynamic flood risk maps, characterized in that, include: The first matching unit is used to match the target first-level zoning plan from the first-level zoning plan database based on the real-time weather forecast information of the target area. The first computing unit is used to divide the spatial computing domain of the target area into multiple first-level computing partitions according to the target first-level zoning scheme, and to perform simulation calculations of flood risk factors in parallel on the multiple first-level computing partitions. The second matching unit is used to match the target secondary partition scheme corresponding to the target primary computing partition from the secondary partition scheme library if an abnormal real-time computing load index value is detected in the multiple primary computing partitions during the simulation calculation process. The second computing unit is used to decompose the target primary computing partition into multiple secondary computing partitions according to the target secondary zoning scheme, and to switch the simulation calculation of flood risk factors for the target primary computing partition to the parallel simulation calculation of flood risk factors for the multiple secondary computing partitions. The generation unit is used to generate a flood risk map of the target area based on the simulation calculation results of each of the first-level calculation partitions and the simulation calculation results of each of the second-level calculation partitions. The primary partitioning plan library includes multiple primary partitioning schemes. Each primary partitioning scheme is obtained by dividing the nodes in the spatial topology map of the target area under different flood scenarios, while keeping the coupling nodes intact. The secondary partitioning plan library includes secondary partitioning schemes pre-constructed for the topology structure of each primary calculation partition corresponding to each primary partitioning scheme. The spatial topology diagram includes the coupling nodes, surface grid nodes, river segment nodes, and pipeline nodes. The coupling nodes are obtained by aggregating multiple model units with hydraulic interaction relationships.

Images