Digital twin water conservancy super-converged algorithm model and hardware system performance tuning method

Abstract

The application discloses a digital twin water conservancy super-fusion algorithm model and a hardware system performance optimization method, and belongs to the technical field of a digital twin water conservancy system. Real-time interaction data of a model is extracted, and a docking configuration system is built. The docking configuration system has an adaptive interface for external service and an interaction mechanism for internal processing. The adaptive interface is adapted to a heterogeneous platform, realizes dynamic acquisition and analysis of external data of the model to obtain input data, and the interaction mechanism is embedded with a hydrology-hydrodynamic river network multi-scale hierarchical model to obtain real-time model parameter values and broadcast focus point information according to the input data. Forecasting and early warning parameter calculation is performed based on the real-time interaction data of the model to obtain current output data, and the current output data is converted into display data for rolling display. The docking configuration system with the adaptive interface is built to adapt to data formats of the heterogeneous platform and break data islands, and a parameter self-adaptive adjustment mechanism combining particle swarm optimization and reinforcement learning is adopted to solve the problem of insufficient adaptability of static parameters.

Description

Technical Field

[0001] This invention belongs to the technical field of digital twin water conservancy systems, and in particular relates to a digital twin water conservancy hyper-converged algorithm model and a hardware system performance optimization method. Background Technology

[0002] In the field of digital twin water conservancy, existing technologies have shortcomings in model operation, parameter correction, multi-scale simulation, and flood risk calculation.

[0003] In terms of data interaction between the model and the system, significant differences in data formats between the model service interface and the system lead to difficulties in interaction and affect the efficiency of model application. Regarding real-time forecasting and early warning, data integration between different platforms is hampered by variations in hardware, software, and communication protocols, making it difficult to efficiently acquire and process real-time rainfall and water level data for accurate rolling calculations.

[0004] Traditional methods for model parameter calibration suffer from poor accuracy, convergence speed, and robustness, making it difficult to accurately invert parameters and affecting model simulation results. When multi-scale models are nested, the lack of a unified adaptation interface and effective management mechanism makes data interaction and joint computation between models of different scales difficult, failing to accurately reflect water cycle and hydrodynamic processes.

[0005] In urban flood risk simulation, hydrological-hydrodynamic models converge slowly and take a long time under complex boundary conditions, making it difficult to accurately calculate flood depth in real time under rolling weather forecasts, thus affecting the efficiency of flood risk prediction and management. Summary of the Invention

[0006] To address the technical problems existing in the background art, this invention provides a digital twin water conservancy hyper-fusion algorithm model and a hardware system performance optimization method.

[0007] This invention adopts the following technical solution: a digital twin water conservancy hyper-converged algorithm model and hardware system performance optimization method, including the following steps:

[0008] Construct a model database, and extract real-time interaction data of the model based on the model database;

[0009] A connection configuration system is established, which has an adaptation interface for external services and an interaction mechanism for internal processing; wherein, the adaptation interface is adapted to heterogeneous platforms through a multi-level collaborative architecture, realizing the dynamic acquisition and parsing of external model data to generate input data;

[0010] The interaction mechanism embeds a multi-scale hierarchical model of hydrological-hydrodynamic river network, and uses a parameter adaptive adjustment mechanism to dynamically set the real-time model parameter values ​​of the multi-scale hierarchical model of hydrological-hydrodynamic river network; based on the input data, it uses the multi-scale hierarchical model of hydrological-hydrodynamic river network to generate broadcast information on points of interest.

[0011] Meanwhile, based on the real-time interactive data of the model, the forecast and early warning parameters are configured, and combined with rolling weather forecast data, the hydrological-hydrodynamic river network multi-scale hierarchical model is driven to generate dynamic current output data based on the broadcast information of the focus;

[0012] The current output data is converted into visual display data via an adapter interface. WEB The server results display module displays results in a scrolling manner.

[0013] In a further embodiment, the real-time interaction data of the model includes at least: model result library data, model network library data, model event library data, and model logic library data;

[0014] The external data for the model includes at least: user request data and real-time rainfall and water level data.

[0015] In a further embodiment, the adapter interface is built collaboratively through multiple levels of modules, as follows:

[0016] exist WEB Server deployment is based on JAVA The developed user request module reads external data from the model, performs preliminary format verification and conversion according to preset data parsing rules, and generates intermediate data that conforms to the interface transmission standard.

[0017] Deploy a protocol processing module in the interface adaptation layer, using a method based on... XML of SOAP The protocol encapsulates intermediate data to form standardized call requests and data messages;

[0018] Relying on HTTP The protocol transmits the encapsulated data packets to the internal interaction mechanism of the interface configuration system, enabling data communication with heterogeneous platforms.

[0019] In a further embodiment, the parameter adaptive adjustment mechanism includes: a basic iteration phase and a reinforcement learning phase;

[0020] The basic iteration phase optimizes parameters iteratively using the particle velocity-position update formula, while the reinforcement learning phase breaks local stagnation by adjusting the optimization strategy and triggering a particle regeneration mechanism.

[0021] In a further embodiment, the construction process of the hydrological-hydrodynamic river network multi-scale hierarchical model is as follows:

[0022] Create a multi-level, refined grid system according to requirements, including: a weather forecast grid layer. Catchment area grid layer Calculation model mesh layer Spatial overlay relationship; the spatial overlay relationship includes: weather forecast grid layer. and water catchment area grid layer Spatial overlay relationship between them, water catchment area grid layer and computational model mesh layer Spatial overlapping relationship between them;

[0023] Interpolation sampling points are formed by embedding multiple sets of floodwater depth schemes under different rainfall conditions into the multi-level refined grid system.

[0024] In a further embodiment, the process of generating the broadcast attention information is as follows:

[0025] Static variable data and dynamic monitoring data are extracted from the input data, and a dimension is constructed based on the static variable data. static feature matrix , For the number of catchment areas, It is a static variable type;

[0026] Organize dynamic monitoring data to form a dynamic monitoring matrix Its dimensions are ,in, For the types of dynamic monitoring data, For time step;

[0027] By spatial matching and timestamp alignment, the static feature matrix is... and dynamic monitoring matrix To achieve spatiotemporal correlation through fusion;

[0028] The analytic hierarchy process (AHP) was used to determine the weighting of static and dynamic variables, and the priority score for each catchment area was calculated. Prioritize catchment areas as the focus;

[0029] The basic unit for calculating the target of interest calculates and allocates rainfall and corresponding maximum flooding depth, and generates broadcast information on the target of interest by combining the basic information of the priority catchment area.

[0030] In a further embodiment, the calculation process for the allocated rainfall of the object of interest is as follows:

[0031] Retrieve the catchment area grid layer from the hydrological-hydrodynamic river network multi-scale hierarchical model. and weather forecast grid layer Priority catchment areas were identified. and the weather forecast grid units superimposed on it , forming an association set : , , ;

[0032] The priority catchment area is calculated using the following formula. Rainfall distribution :

[0033] ;

[0034] in, Priority catchment area and grid cells The area of ​​intersection, For grid cells The corresponding weather forecast rainfall, Priority catchment area area, In order to be in the priority catchment area The combination of all intersecting weather forecast grid cells.

[0035] In a further embodiment, the calculation process for the maximum flooding depth of the object of interest is as follows:

[0036] Catchment grid layer based on multi-scale hierarchical model of hydrological-hydrodynamic river network and computational model mesh layer The spatial overlapping relationship between them will prioritize the catchment areas. Rainfall distribution Mapped to the corresponding computational model grid cell The computational model mesh element is obtained. Input rainfall ;

[0037] Determine the maximum rainfall value within the rainfall range. and minimum rainfall Obtain the grid cells of the computational model respectively. With rainfall of Maximum flood depth and rainfall Maximum flood depth at that time ;

[0038] The interpolation method is used to calculate the model mesh elements. Enter rainfall Maximum flood depth:

[0039] In the formula, These are the interpolation coefficients.

[0040] The present invention has the following beneficial effects:

[0041] This invention establishes a configuration system with adaptable interfaces to accommodate heterogeneous platform data formats, enabling dynamic acquisition and parsing of multi-source information such as user-requested data and real-time rainfall and water level data. This breaks down "data silos" and provides comprehensive data support for model calculations.

[0042] A multi-scale hydrological-hydrodynamic model is constructed based on a fine-grained grid. By decoupling dynamic variables through catchment units, the model achieves collaborative simulation of macro-basin trends and micro-regional details, thereby improving operational efficiency while ensuring computational accuracy.

[0043] By optimizing the interface adaptation layer and interaction mechanism, the resource consumption of data transmission and model calculation is reduced, enabling efficient collaboration between the algorithm model and the hardware system, and meeting the timeliness requirements of high-frequency rolling calculation and real-time forecasting.

[0044] The modular structure of the model database and interface system supports the rapid access of new models and data types, and can be widely used in various scenarios such as watershed management, urban flood control, and reservoir scheduling, providing technical support for the large-scale promotion of digital twin water conservancy systems. Attached Figure Description

[0045] Figure 1 This is a flowchart of the digital twin water conservancy hyper-converged algorithm model and hardware system performance optimization method. Detailed Implementation

[0046] The present invention will now be further described with reference to the accompanying drawings.

[0047] Example 1

[0048] like Figure 1 As shown, the performance optimization method of digital twin water conservancy hyper-converged algorithm model and hardware system includes the following steps: constructing a model database and extracting real-time interaction data of the model based on the model database;

[0049] A connection configuration system is established, which has an adaptation interface for external services and an interaction mechanism for internal processing; wherein, the adaptation interface is adapted to heterogeneous platforms to realize the dynamic acquisition and parsing of external model data, thereby obtaining input data; the external model data includes at least: user request data and real-time rainfall and water level data;

[0050] The interaction mechanism embeds a multi-scale hierarchical model of hydrological-hydrodynamic river network, and uses a parameter adaptive adjustment mechanism to dynamically set the real-time model parameter values ​​of the multi-scale hierarchical model of hydrological-hydrodynamic river network; based on the input data, the multi-scale hierarchical model of hydrological-hydrodynamic river network is used to generate broadcast information on points of interest.

[0051] Meanwhile, based on the real-time interactive data of the model, the forecast and early warning parameters are configured, and combined with the rolling weather forecast data, the hydrological-hydrodynamic river network multi-scale hierarchical model generates dynamic current output data based on the broadcast information of the focus.

[0052] The information to be broadcast includes at least the following: reporting stations, one-dimensional river cross-sections, nodes, allocated rainfall, and maximum flood depth; correspondingly, the current output data includes: allocated rainfall, maximum flood depth, and the corresponding warning threshold, warning information, and warning plan.

[0053] The current output data is converted into visual display data via an adapter interface. WEB The server results display module displays results in a scrolling manner.

[0054] In a further embodiment, the model database is used to store application models, such as flood evolution models, water resource scheduling models, water environment simulation models, and other water-related professional models. Based on the above application models, real-time interaction data of the models is extracted from the model database, such as model result database data, model network database data, model event database data, and model logic database data.

[0055] There are numerous water conservancy-related application models, such as flood evolution models, water resource scheduling models, and water environment simulation models. Each model generates and relies on specific data during operation. This data is scattered across different modules of various models, presenting a fragmented state. By extracting and integrating real-time interactive data from these models, the data between different models is no longer isolated, breaking down data silos and thereby improving the overall integrity of the entire water conservancy system. This allows for comprehensive analysis of water-related situations from multiple dimensions.

[0056] Therefore, the system will encounter several problems during use, including heterogeneous data formats, inconsistent interface standards, and difficulties in cross-platform data communication. Interface adaptation is achieved through multi-level module collaboration, as follows:

[0057] exist WEB Server deployment is based on JAVA The developed user request module is used to read external data from the model. This module performs preliminary format verification and conversion on the external data using preset data parsing rules, generating intermediate data that conforms to the interface transmission standard, such as real-time rainfall and water level data from a watershed monitoring station. CSV Historical engineering parameters. XML User-uploaded scheduling rules. TXT .

[0058] Deploy based on the interface adaptation layer WebService The technology's protocol processing module processes intermediate data via a protocol based on... XML ofSOAP The protocol is encapsulated to form standardized call requests and data messages; relying on WebService The technology deployment protocol processing module adopts... XML of SOAP The protocol standardizes and encapsulates intermediate data to form unified format call requests and data messages, thereby standardizing data interaction.

[0059] Relying on HTTP The protocol will encapsulate XML The internal interaction mechanism transmits formatted data messages to the interface configuration system, enabling data communication with heterogeneous platforms. For example, it converts model input files to include watershed boundary conditions and engineering attributes. CSV The file records structured data such as river cross-section parameters and sluice gate scheduling rules. For example, it records... Java Output from the user request module. CSV Intermediate data (such as "river channel cross-section parameters (node ​​coordinates, roughness), sluice gate scheduling rules (time period-opening correspondence table)"), according to SOAP Protocol format conversion XML Message.

[0060] Furthermore, to facilitate the presentation of output data, the setup of the adapter interface also includes the following processes:

[0061] The WEB The server is also deployed based on JAVA The developed results display module is used to receive the current output data after format conversion and to present it visually; receiving XML The message is parsed into input that the computing engine can recognize (such as converting it into...). Ruby The parameter object of the calculation script is used to generate the flood evolution calculation. PNG, . CSV Results documents.

[0062] Correspondingly, the interface adaptation layer also deploys a response processing module, which receives the current output data returned by the internal interaction mechanism, converts it according to the front-end display requirements, and then feeds it back to the front-end. WEB The server's user request module completes the data interaction loop.

[0063] use. PNG 、 CSV Format the output or store the current output data, such as... PNG Output "Animation frames of the flood evolution process and a comparison chart of the inundation area"; CSV Store structured results such as "cross-sectional water level process, engineering facility flow capacity statistics, and flood loss assessment (inundated area and population)," convert the format according to front-end requirements, and then send them back. WEB server.

[0064] In another embodiment, the multi-scale hierarchical model of the hydrological-hydrodynamic river network involves numerous complex parameters that interact with each other. These parameters include fundamental parameters of the hydrological model (watershed characteristic parameters, underlying surface parameters), core parameters of the hydrodynamic model (flow motion parameters, boundary condition parameters), and model calculation control parameters (time step parameters, convergence control parameters). Traditional manual settings or simple optimization algorithms struggle to accurately find the optimal parameter combination under multi-dimensional and multi-constraint conditions, leading to deviations between the model simulation results and actual conditions, and failing to accurately reflect the hydrological-hydrodynamic processes of the river network.

[0065] Therefore, this embodiment employs a parameter adaptive adjustment mechanism to dynamically set the real-time model parameter values ​​of the hydrological-hydrodynamic river network multi-scale hierarchical model. Specifically, the parameter adaptive adjustment mechanism includes: a basic iteration phase and a reinforcement learning phase;

[0066] The basic iteration phase optimizes parameters iteratively using the particle velocity-position update formula, while the reinforcement learning phase breaks local stagnation by adjusting the optimization strategy and triggering a particle regeneration mechanism.

[0067] Furthermore, the iterative process of the basic iteration phase is as follows: Step 001: Define the core operating parameters, perform initialization operations on the particle population, and randomly assign initial positions to each dimension of each particle. and initial velocity ,in, i For particles, d For dimensions;

[0068] Step 002: Create an objective function based on the input data, calculate the fitness of each particle after initialization based on the objective function, and traverse the particle population to determine the optimal position of each particle. and individual fitness value ; Filter out the globally optimal position and global fitness value Simultaneously, a dimensional learning leader is determined for each particle in each dimension. and corresponding position ;

[0069] Step 003: Iteratively update the velocity and position of the particles using the update formula to obtain the updated fitness and optimal position of each particle. Individual fitness value and the global optimal position Global fitness value And record the number of times the particles stopped. Flag and number of iterations k ;

[0070] In this embodiment, the update formula is as follows:

[0071]

[0072] In the formula, and For iteration The particle velocity and position at this time, For inertial weights, and For iteration The particle velocity and position at this time, For particles i Learning factors and The learning speedup constant is generally 2. and All are particles i In dimensions d Iteration k The randomness of the search is determined by the number of random numbers generated. This is the global learning factor.

[0073] Step 004, if the number of pauses... Flag Less than or equal to learning leaders If the update interval is less than 0, then proceed to step 006; otherwise, proceed to step 005.

[0074] Step 005: Update the dimensions of each particle to determine the dimensional learning leader and its corresponding position, and proceed to step 006;

[0075] Step 006: Determine the global fitness value If the convergence tolerance is less than or equal to the value, output the global optimum position. This is the real-time model parameter value; if not, then execute 007.

[0076] Step 007, if the number of iterations... k Less than or equal to the preset number of iterations Then repeat steps 003 to 004; if the number of iterations... k Greater than the number of iterations Then, it enters the reinforcement learning stage, through which real-time model parameter values ​​are obtained.

[0077] During the iteration process, the particle swarm is prone to getting stuck in local optima and the iteration stagnates. That is, as the iteration progresses, the fitness improves slowly or even stops improving, making it impossible to effectively explore the global optimal solution and affecting the efficiency and effectiveness of model parameter optimization.

[0078] Therefore, the reinforcement learning phase of this embodiment includes the following steps:

[0079] Step 101: Recalculate the updated fitness and optimal position of each particle using the update formula. Individual fitness value and the global optimal position Global fitness value And record the number of times the particle stagnates during the strengthening phase. and number of iterations ;

[0080] Step 102, if the number of pauses during the strengthening phase... If the interval is less than or equal to the reinforcement phase update interval, proceed to step 104; otherwise, proceed to step 103.

[0081] Step 103: Update each dimension of each particle, determine the dimensional learning leader and its corresponding position, and proceed to step 104.

[0082] Step 104: Determine the global fitness value If the convergence tolerance is less than or equal to the value, output the global optimum position. This is the real-time model parameter value; if not, then execute 105.

[0083] Step 105, if the number of iterations... Less than or equal to the preset number of iterations Then repeat steps 101 to 102 to continue the iteration; if the number of iterations... Greater than the preset number of iterations Then, it is further determined whether the adaptive improvement value of the particle is less than or equal to the adaptive improvement threshold. If so, step 106 is executed; if the adaptive improvement value is greater than the adaptive improvement threshold, step 107 is executed.

[0084] Step 106: Trigger the particle regeneration mechanism to regenerate and initialize the velocity of non-global optimal particles. and initialization position And record the number of rebirths. The initialization speed and initialization position Satisfy the velocity rebirth constraint and the position rebirth constraint respectively; execute step 107.

[0085] It is worth mentioning that the velocity regeneration constraint is expressed as follows: ,in, The total number of particles, The globally optimal particle is excluded.

[0086] Location rebirth constraints include distance constraints and probability constraints, where the distance constraint is expressed as:

[0087] ,in, Let be the distance between the particle and the current best particle. This represents the maximum distance within the entire particle swarm. This is the distance threshold.

[0088] The probability condition is expressed as: ,in, The number is a random number between 0 and 1 for the particle. This is the probability threshold.

[0089] When both of the above conditions are met, position initialization is performed.

[0090] Step 107, until the number of rebirths is reached. If exhausted, output the globally optimal position. .

[0091] Mechanisms such as stagnation count judgment, dimensional learning leader update, and reinforcement learning phase are introduced to effectively identify and break the local optimum stagnation state in the particle swarm iteration process. When iteration stagnates, the learning leader and particle state are updated in a timely manner, and the reinforcement learning phase further strengthens the optimization strategy, prompting the algorithm to continuously explore better solutions and improving the efficiency of parameter optimization and global search capability.

[0092] Under the complex boundary conditions of urban river networks, the calculation of multi-scale hierarchical models of hydrological and hydrodynamic river networks often faces problems such as slow convergence and time-consuming and labor-intensive processes. Therefore, under rolling weather forecast conditions, it is extremely challenging to calculate the flood depth in real time and accurately.

[0093] To address the dual requirements of speed and accuracy in real-time computing, the construction process of the multi-scale hierarchical model of the hydrological-hydrodynamic river network in this embodiment is as follows:

[0094] In terms of accuracy, a multi-level refined grid system is created according to requirements. This multi-level refined grid system includes: a weather forecast grid layer. Catchment area grid layer Calculation model mesh layer Spatial overlay relationship; the spatial overlay relationship includes: weather forecast grid layer. and water catchment area grid layer Spatial overlay relationship between them, water catchment area grid layer and computational model mesh layer Spatial overlapping relationship between them;

[0095] The set of spatial overlay relationships not only reflects the geometric intersection between different layers, but also provides important basic data support for subsequent flooding depth calculations. It should be noted that, in order to obtain the intersection relationships between layers, a method is used... GISThe spatial data overlay analysis method was used to calculate the spatial overlay relationship between the meteorological forecast grid and the catchment unit, as well as the spatial overlay relationship between the catchment unit and the model grid.

[0096] Interpolation sampling points are formed by embedding multiple sets of floodwater depth schemes under different rainfall conditions into the multi-level refined grid system.

[0097] Regarding speed, the process of generating broadcast information of interest is as follows:

[0098] Static variable data and dynamic monitoring data are extracted from the input data, and a dimension is constructed based on the static variable data. static feature matrix , For the number of catchment areas, It is a static variable type; the static feature matrix described in this embodiment The element is the one that is marked. and Static variables, such as the area of ​​the catchment area, the slope of the terrain, the density of the drainage network, and the land use type.

[0099] Organize dynamic monitoring data to form a dynamic monitoring matrix Its dimensions are ,in, For the types of dynamic monitoring data, For the time step. Further, a dynamic monitoring matrix. For containing Monitoring values, such as rainfall, real-time water level, water depth, rainfall intensity, etc.

[0100] By spatial matching and timestamp alignment, the static feature matrix is... and dynamic monitoring matrix By integrating the two types of data, the spatiotemporal correlation between them can be achieved, specifically through the number of water areas identified by keywords. To integrate.

[0101] Based on this, the analytic hierarchy process (AHP) is used to determine the weight proportions of static and dynamic variables, which are expressed as follows: and , .

[0102] Calculate the priority score for each catchment area. Select priority catchment areas as the focus, and assign priority scores to them. The calculation formula is as follows:

[0103] In the formula, water catchment area The average value of static variable data, water catchment area In time step The average value of dynamic data.

[0104] Based on a pre-set scoring threshold, priority scores will be given attention. Catchment areas exceeding the scoring threshold If a data point is designated as a focus, the broadcast of information about that focus will prioritize generating information about it. Specifically, this includes the basic information of the focus and the corresponding basic unit for calculating and allocating rainfall and the corresponding maximum flooding depth. In this embodiment, the basic information refers to the flood reporting stations, one-dimensional river cross-sections, nodes, etc., mentioned above.

[0105] Furthermore, the calculation process for the allocated rainfall in priority catchment areas is as follows:

[0106] Retrieve the catchment area grid layer from the hydrological-hydrodynamic river network multi-scale hierarchical model. and weather forecast grid layer Priority catchment areas were identified. and the weather forecast grid units superimposed on it , forming an association set : , , ;

[0107] The priority catchment area is calculated using the following formula. Rainfall distribution :

[0108] ;

[0109] in, Priority catchment area and grid cells The area of ​​intersection, For grid cells The corresponding weather forecast rainfall, Priority catchment area area, In order to be in the priority catchment area The combination of all intersecting weather forecast grid cells.

[0110] The corresponding calculation process for the maximum flood depth is as follows:

[0111] Catchment grid layer based on multi-scale hierarchical model of hydrological-hydrodynamic river network and computational model mesh layer The spatial overlapping relationship between them will prioritize the catchment areas. Rainfall distribution Mapped to the corresponding computational model grid cell The computational model mesh element is obtained. Input rainfall ;

[0112] Determine the maximum rainfall value within the rainfall range. and minimum rainfall Obtain the grid cells of the computational model respectively. With rainfall of Maximum flood depth at that time and rainfall Maximum flood depth ;

[0113] The interpolation method is used to calculate the model mesh elements. Enter rainfall Maximum flood depth:

[0114] In the formula, These are the interpolation coefficients.

[0115] Because the pre-calculation stage uses a dense distribution of sampling points, linear interpolation is employed here to ensure both simplicity and accuracy in the calculation. In practical applications, the study area typically contains a large number of computational grid cells. To improve the efficiency of real-time calculation, this work adopts a parallel computing method, performing parallel cyclic processing on rainfall allocation and water depth interpolation for grid cells. This significantly improves the calculation speed and ensures that the system can respond quickly to complex rainfall scenarios, providing a solid technical guarantee for flood risk prediction and management.

[0116] Example 2

[0117] The method described in Example 1 is applied to a digital twin water conservancy hyper-converged integrated machine. The digital twin water conservancy hyper-converged integrated machine includes a hardware architecture and a software module, wherein the hardware architecture has a display and operation screen and several interfaces.

[0118] The software module includes: a first module, configured to build a model database and extract real-time interactive data of the model based on the model database;

[0119] The second module is configured to build a docking configuration system, which has an external service adaptation interface and an internal processing interaction mechanism. The adaptation interface is adapted to heterogeneous platforms to realize the dynamic acquisition and parsing of external model data, thereby obtaining input data. The external model data includes at least: user request data and real-time rainfall and water level data.

[0120] The third module is configured to embed a hydrological-hydrodynamic river network multi-scale hierarchical model within the interaction mechanism. Based on the input data, it sets the real-time model parameter values ​​and broadcast information of concern to the hydrological-hydrodynamic river network multi-scale hierarchical model. Simultaneously, it configures forecast and early warning parameters based on the real-time interaction data of the model, and drives the hydrological-hydrodynamic river network multi-scale hierarchical model to calculate the current output data based on the real-time model parameter values ​​in conjunction with rolling weather forecast data.

[0121] The current output data is converted into display data through an adaptation interface and then displayed in a scrolling manner.

Claims

1. A digital twin hydraulic hyper-converged algorithm model and hardware system performance optimization method, characterized in that, Includes the following steps: Construct a model database, and extract real-time interaction data of the model based on the model database; A connection configuration system is established, which has an adaptation interface for external services and an interaction mechanism for internal processing; wherein, the adaptation interface is adapted to heterogeneous platforms through a multi-level collaborative architecture, realizing the dynamic acquisition and parsing of external model data to generate input data; The interaction mechanism embeds a multi-scale hierarchical model of hydrological-hydrodynamic river network, and uses a parameter adaptive adjustment mechanism to dynamically set the real-time model parameter values ​​of the multi-scale hierarchical model of hydrological-hydrodynamic river network; based on the input data, it uses the multi-scale hierarchical model of hydrological-hydrodynamic river network to generate broadcast information on points of interest. Meanwhile, based on the real-time interactive data of the model, the forecast and early warning parameters are configured, and combined with rolling weather forecast data, the hydrological-hydrodynamic river network multi-scale hierarchical model is driven to generate dynamic current output data based on the broadcast information of the focus; The current output data is converted into visual display data via an adapter interface. WEB The server results display module displays the results in a scrolling manner; The process of generating the broadcast attention information is as follows: Static variable data and dynamic monitoring data are extracted from the input data, and a dimension is constructed based on the static variable data. static feature matrix , For the number of catchment areas, It is a static variable type; Organize dynamic monitoring data to form a dynamic monitoring matrix Its dimensions are ,in, For the type of dynamic monitoring data, For time step; By spatial matching and timestamp alignment, the static feature matrix is... and dynamic monitoring matrix To achieve spatiotemporal correlation through fusion; The analytic hierarchy process (AHP) was used to determine the weighting of static and dynamic variables, and the priority score for each catchment area was calculated. Prioritize catchment areas as the focus; The basic unit for calculating the target of interest calculates and allocates rainfall and corresponding maximum flooding depth, and generates broadcast information on the target of interest by combining the basic information of the priority catchment area.

2. The digital twin water conservancy hyper-converged algorithm model and hardware system performance optimization method according to claim 1, characterized in that, The real-time interactive data of the model includes at least: model result library data, model network library data, model event library data, and model logic library data; The external data for the model includes at least: user request data and real-time rainfall and water level data.

3. The digital twin water conservancy hyper-converged algorithm model and hardware system performance optimization method according to claim 1, characterized in that, The adaptation interface is built collaboratively through multiple levels of modules, and the process is as follows: exist WEB Server deployment is based on JAVA The developed user request module reads external data from the model, performs preliminary format verification and conversion according to preset data parsing rules, and generates intermediate data that conforms to the interface transmission standard. Deploy a protocol processing module in the interface adaptation layer, using a method based on... XML of SOAP The protocol encapsulates intermediate data to form standardized call requests and data messages; Relying on HTTP The protocol transmits the encapsulated data packets to the internal interaction mechanism of the interface configuration system, enabling data communication with heterogeneous platforms.

4. The digital twin water conservancy hyper-converged algorithm model and hardware system performance optimization method according to claim 1, characterized in that, The parameter adaptive adjustment mechanism includes: a basic iteration phase and a reinforcement learning phase; The basic iteration phase optimizes parameters iteratively using the particle velocity-position update formula, while the reinforcement learning phase breaks local stagnation by adjusting the optimization strategy and triggering a particle regeneration mechanism.

5. The digital twin water conservancy hyper-converged algorithm model and hardware system performance optimization method according to claim 1, characterized in that, The construction process of the hydrological-hydrodynamic river network multi-scale hierarchical model is as follows: Create a multi-level, refined grid system according to requirements, including: a weather forecast grid layer. Catchment area grid layer Calculation model mesh layer Spatial overlay relationship; the spatial overlay relationship includes: weather forecast grid layer. and water catchment area grid layer Spatial overlay relationship between them, water catchment area grid layer and computational model mesh layer Spatial overlapping relationship between them; Interpolation sampling points are formed by embedding multiple sets of floodwater depth schemes under different rainfall conditions into the multi-level refined grid system.

6. The digital twin water conservancy hyper-converged algorithm model and hardware system performance optimization method according to claim 1, characterized in that, The calculation process for the allocated rainfall of the object of interest is as follows: Retrieve the catchment area grid layer from the hydrological-hydrodynamic river network multi-scale hierarchical model. and weather forecast grid layer Priority catchment areas were identified. and the weather forecast grid units superimposed on it , forming an association set : , , ; The priority catchment area is calculated using the following formula. Rainfall distribution : ； in, Priority catchment area and grid cells The area of ​​intersection, For grid cells The corresponding weather forecast rainfall, Priority catchment area area, In order to be in the priority catchment area The combination of all intersecting weather forecast grid cells.

7. The digital twin water conservancy hyper-converged algorithm model and hardware system performance optimization method according to claim 1, characterized in that, The calculation process for the maximum flood depth of the object of interest is as follows: Catchment grid layer based on multi-scale hierarchical model of hydrological-hydrodynamic river network and computational model mesh layer The spatial overlapping relationship between them will prioritize the catchment areas. Rainfall distribution Mapped to the corresponding computational model grid cell The computational model mesh element is obtained. Input rainfall ; Determine the maximum rainfall value within the rainfall range. and minimum rainfall Obtain the grid cells of the computational model respectively. With rainfall of Maximum flood depth and rainfall Maximum flood depth at that time ; The interpolation method is used to calculate the model mesh elements. Enter rainfall Maximum flood depth: In the formula, These are the interpolation coefficients.

Images