A city waterlogging simulation and prediction method and device based on multi-source data fusion

Abstract

The application discloses a kind of urban waterlogging simulation prediction methods and equipment based on multi-source data fusion.The method comprises the following steps: receiving weather forecast data and underground drainage pipe network telemetry data, performing data assimilation algorithm to correct state vector, completing data access and initial field updating (S101);Run one-dimensional and two-dimensional coupled hydrodynamic calculation model, dynamically calculate allowable time step, trigger adaptive time step control, complete coupling calculation and adaptive step dynamic calculation (S102);When grid water depth is lower than threshold in recession period, cut off convection term and keep static hydrodynamic term, complete recession cut-off processing and waterlogging duration statistics (S103).The application can build smooth initial field, suppress interaction calculation solving divergence, and ensure the continuity of computing power and high-precision prediction.

Description

Technical Field

[0001] This application relates to the field of computer digital computing processing and simulation technology, and in particular to a method and device for simulating and predicting urban flooding based on multi-source data fusion. Background Technology

[0002] In urban flooding simulation and flood control scheduling scenarios, water management planning and emergency dispatch systems need to predict the flood discharge process of underground drainage networks and the surface inundation process over a long period of time. Existing simulation schemes based on computer numerical solvers often face bottlenecks when dealing with complex rainstorm conditions, such as long computation time and large deviations between predicted flood area and receding time and actual occurrences.

[0003] Currently, conventional urban flooding simulation systems suffer from significant technical deficiencies. Traditional simulation systems rely on relatively singular input boundary data sources. There is a lack of effective data assimilation and multi-source data fusion mechanisms between gridded rainfall forecast data from meteorological radar and telemetry data from underground drainage networks. This leads to substantial deviations in the initial computational field of the hydrodynamic calculation model, and this initial high-frequency noise is amplified dramatically with nonlinear iteration. Furthermore, the two-dimensional shallow surface water model used for surface flood evolution and the one-dimensional pipe network hydraulic model used for underground pipe networks exhibit inconsistent stability time steps during iterative solutions at the computer level due to differences in spatial grid scale and governing equation properties. Existing coarse-grained forced-coupled grid data interaction algorithms are prone to causing severe oscillations in calculated values ​​at one-dimensional and two-dimensional interaction nodes when the pipe network is under full-pipe pressure flow and surface water changes rapidly due to heavy rainfall, leading to divergence in the mass and momentum conservation equations. Furthermore, in the simulation of the extremely long process of receding and accumulating water after the end of rainfall, the surface grid nodes frequently undergo alternating wet and dry states. The existing computational logic cannot effectively handle the convection term in the momentum update equation when the water depth is extremely shallow, often causing floating-point overflow errors of dividing by zero or generating extremely high flow velocity anomalies, resulting in severe consumption of computing resources and even system deadlock. Summary of the Invention

[0004] To address the technical problems in existing urban flooding simulation calculations, such as heterogeneous data input bias, easy divergence in one-dimensional and two-dimensional coupled numerical calculations, and system deadlock caused by the switching between dry and wet boundaries during the receding water stage, this application provides an urban flooding simulation and prediction method, electronic device, and storage medium based on multi-source data fusion.

[0005] This application provides a method for urban flooding simulation and prediction based on multi-source data fusion, comprising: receiving meteorological forecast data and underground drainage network telemetry data, wherein the meteorological forecast data includes a gridded rainfall forecast data stream output by a weather radar system, and the underground drainage network telemetry data includes real-time telemetry time series data transmitted from IoT level sensors and ultrasonic flow meters; spatially aligning the meteorological forecast data and the underground drainage network telemetry data, converting physical coordinates into index identifiers of simulation calculation grid cells and one-dimensional pipeline nodes through coordinate analysis operators to obtain spatially aligned multi-source observation data; and performing an extended Kalman filter assimilation algorithm to estimate and correct the hydrodynamic state vector containing water depth and flow velocity, constructing and updating the simulation initial field matrix. Based on the simulation initial field matrix, a coupled calculation engine of a one-dimensional pipeline hydraulic model and a two-dimensional shallow surface water model is run to detect the water level gradient difference of interactive nodes to calculate bidirectional flux, and based on CFL conditions, scanning the global maximum characteristic wave velocity before the macroscopic time step, dynamically calculating the maximum allowable time step size of the next micro-cycle, and triggering adaptive step size reduction in response to local flow velocity abrupt changes. During the receding water calculation phase, in response to the detection that the calculated water depth of the two-dimensional grid is lower than the critical wet-dry threshold, the update of the convection term of the momentum equation of the two-dimensional grid is stopped and the normal velocity is cleared to zero. The hydrostatic term is retained to perform micro-perturbation calculation, and the time steps when the calculated water depth is greater than the preset water accumulation threshold are counted. The scalar result of the water accumulation duration at the node is output.

[0006] Furthermore, the execution of the extended Kalman filter assimilation algorithm to estimate and correct the hydrodynamic state vector containing water depth and flow velocity, and to construct and update the simulation initial field matrix, includes: using the gridded rainfall data in the meteorological forecast data and the measured liquid level and flow velocity data in the underground drainage network telemetry data as observation variables; using the hydrodynamic state vector and system state transition matrix of the previous time step to calculate the prior state estimate and the corresponding prior error covariance matrix of the current time step; combining the observation variables to calculate the Kalman gain matrix, and using the Kalman gain matrix to update and correct the prior state estimate, generating the posterior state estimate as the simulation initial field matrix.

[0007] Furthermore, the step of calculating the Kalman gain matrix based on the observed variables, and using the Kalman gain matrix to update and correct the prior state estimate to generate the posterior state estimate as the initial field matrix for simulation, is configured to execute the following system matrix update logic:

[0008]

[0009]

[0010] Among them, K kP represents the Kalman gain matrix at the k-th time step; k - H represents the prior error covariance matrix at the k-th time step; H represents the observation matrix, H T R represents the transpose of the observation matrix; R represents the observation noise covariance matrix; x k x represents the posterior state estimate at the k-th time step; k - z represents the prior state estimate at the k-th time step; k The observed variable represents the variable at the k-th time step; based on the calculated Kalman gain matrix K... k The model dynamically adjusts the weight allocation between predictions and measured data, filters out high-frequency noise in the observed variables, and outputs the smoothed posterior state estimate x. k Update to the optimal state space of the initial field matrix of the simulation.

[0011] Furthermore, the detection of the water level gradient difference at the interaction nodes is used to calculate the bidirectional flux. Based on the CFL condition, the global maximum characteristic wave velocity is scanned before the macroscopic time step, and the maximum allowable time step of the next microscopic cycle is dynamically calculated. This includes: extracting the pipe head of the one-dimensional pipe hydraulic model at the inspection well node, and the surface water depth of the two-dimensional shallow surface water model at the corresponding surface grid; calculating the water level gradient difference between the pipe head and the surface water depth; when the absolute value of the water level gradient difference exceeds the set interaction trigger threshold, calculating the water exchange data occurring at the interaction node as the bidirectional flux according to the outflow formula; traversing the gravity wave velocity and convection velocity components of the entire domain calculation grid, superimposing them to obtain the local characteristic wave velocity of each grid cell, and extracting the maximum value as the global maximum characteristic wave velocity.

[0012] Furthermore, the maximum allowable time step for dynamically calculating the next micro-cycle, in response to adaptive step size reduction triggered by a sudden change in local flow velocity, is configured to: determine the basic micro-step size based on the extracted global maximum characteristic wave velocity using the following stability condition relationship:

[0013]

[0014] Where, Δt base C represents the basic microstep size; CFL This indicates the set threshold for the number of Courants, and ;Δx min V represents the minimum spatial step size in the global two-dimensional computational grid; maxThis represents the globally maximum characteristic wave velocity extracted by the scan; during the coupled computation process, the local velocity gradient is continuously monitored, and in response to a sudden increase in local velocity caused by full pipe overflow at the interaction node and the local velocity gradient exceeding a preset gradient abrupt change threshold, the basic microstep size Δt is adjusted according to a preset attenuation coefficient. base The frequency and order are reduced and pruned, and the time step value obtained after pruning is used as the actual time step for the current micro-period iterative solution.

[0015] Furthermore, the step of responding to the detection that the calculated water depth of the two-dimensional grid is lower than the critical wet-dry threshold, stopping the update of the convection term of the momentum equation of the two-dimensional grid and clearing the normal velocity to zero, while retaining the hydrostatic term to perform micro-perturbation calculations, includes: at the end of each micro-calculation cycle, traversing and extracting the calculated water depth of all the two-dimensional grids in the two-dimensional computational domain; comparing the extracted calculated water depth with the set critical wet-dry threshold, wherein the critical wet-dry threshold is configured as a constant greater than zero; when the comparison result determines that the calculated water depth is less than the critical wet-dry threshold, marking the corresponding grid as an interference interruption grid in the shallow water equation solver, and performing flux truncation operation in the numerical discretization format of the interference interruption grid.

[0016] Furthermore, the flux truncation operation performed in the numerical discretization scheme of the interferometric flow-cutting grid is configured as follows: for cells marked as belonging to the interferometric flow-cutting grid, in the momentum conservation equation update operation of the control volume, the discrete flux coefficient of the convection term of its normal velocity vector is reset to zero, cutting off the momentum exchange data path between the interferometric flow-cutting grid and the adjacent normal wet grid; in the water depth state update equation, the hydrostatic term affected by the bottom elevation gradient and the rainfall source term are retained to participate in the iterative summation calculation of the next time step; the water level state variables of the next time step are calculated based on the mass conservation equation after the momentum exchange is truncated.

[0017] Furthermore, the step of statistically analyzing the time steps in which the calculated water depth exceeds the preset water accumulation threshold and outputting a scalar result of the node water accumulation duration includes: configuring an independent time accumulation register in the system; in response to determining that the calculated water depth exceeds the preset water accumulation threshold, performing an incrementing count operation on the time accumulation register within each simulation microstep period; and upon completion of the overall simulation task, extracting the count values ​​stored in all time accumulation registers and converting them into an absolute time dimension in combination with the actual microstep data, outputting a spatial water accumulation distribution map containing node time-series data.

[0018] This application provides an electronic device, including: a memory for storing computer instructions; and a processor connected to the memory for executing the computer instructions stored in the memory to implement the aforementioned urban flooding simulation and prediction method based on multi-source data fusion.

[0019] This application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned urban flooding simulation and prediction method based on multi-source data fusion. Attached Figure Description

[0020] Figure 1 This is a flowchart of the urban flooding simulation and prediction method based on multi-source data fusion provided in the embodiments of this application.

[0021] Figure 2 This is a logical flowchart of the multi-source data assimilation and initial field construction process provided in the embodiments of this application.

[0022] Figure 3 This is a block diagram of the logical structure of the one-dimensional and two-dimensional coupled computation control and adaptive time step computation engine provided in the embodiments of this application.

[0023] Figure 4 This is a flowchart of the dry / wet boundary smoothing operator and truncation processing provided in the embodiments of this application.

[0024] Figure 5 This is a system architecture diagram of an electronic device for performing urban flooding simulation and prediction tasks, provided in an embodiment of this application.

[0025] Explanation of reference numerals in the attached figures

[0026] In the diagram: 201 - Task Scheduling Center, 202 - Kalman Filter, 301 - One-Dimensional Pipeline Network Solution Area, 302 - Two-Dimensional Surface Solution Area, 303 - Coupled Interactive Boundary Control Area, 401 - Smoothing Decision Operator, 402 - Flux Truncation Controller, 501 - Processor, 502 - Memory, 503 - Communication Bus Interface, S101 - Data Access and Initial Field Update, S102 - Coupled Calculation and Adaptive Step Size Dynamic Calculation, S103 - Drainage Cutoff Processing and Water Accumulation Duration Statistics, V 101 -Weather forecast data, V 102 - Remote sensing data of underground drainage pipe network, V 201 -Hydrodynamic state vector, V 202 -Simulation initial field matrix, V 302 -Graphics processor memory, V 501 - Electronic devices. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application.

[0028] To support large-scale urban hydrodynamic spatial hydrological calculations, the hardware foundation and deployment environment upon which this application's embodiments rely need to meet specific configuration requirements. During system initialization, the data access and cleaning server needs to pre-load local urban 2D digital elevation model files and 1D geographic information system (GIS) base vector data. This foundational data provides a reference for subsequent construction of the computational grid and confirmation of the spatial coordinates of interactive nodes. Considering the massive number of sensing devices deployed within the city, the computing node cluster needs to be configured with a corresponding high-concurrency data stream bus platform. A message broker cluster based on a distributed publish-subscribe model or a lightweight IoT message broker is deployed to handle high-concurrency time-series push signals from field IoT devices. This bus mechanism ensures that transient high-frequency monitoring data during rainfall periods can be delivered to the computing engine without obstruction.

[0029] Figure 1 This is a flowchart of an urban flooding simulation and prediction method based on multi-source data fusion provided in this application embodiment. During execution, the method mainly involves the collaborative operation of a data access and cleaning server, a computing node cluster, and a result rendering and output gateway. The method includes the following core steps:

[0030] Data access and initial field update S101, receiving weather forecast data V 101 Remote sensing data of underground drainage pipe network V 102 The meteorological forecast data V101 includes a gridded rainfall forecast data stream output by a weather radar system, and the underground drainage network telemetry data V102 includes real-time telemetry time series data transmitted from IoT level sensors and ultrasonic flow meters. Spatial alignment processing is performed on the meteorological forecast data V101 and the underground drainage network telemetry data V102. Physical coordinates are converted into index identifiers for simulation calculation grid cells and one-dimensional pipeline nodes using coordinate analysis operators to obtain spatially aligned multi-source observation data. An extended Kalman filter assimilation algorithm is then executed to estimate and correct the hydrodynamic state vector containing water depth and flow velocity, constructing and updating the initial simulation field matrix.

[0031] In this embodiment, the data access and cleaning server receives gridded rainfall forecast data streams from the meteorological bureau's weather radar system via a unified encapsulated application programming interface and industrial control communication protocol. Simultaneously, the server receives real-time telemetry time-series data from IoT level sensors and ultrasonic flow meters deployed at inspection wells and drainage zone boundaries in the urban underground drainage network. Since the weather forecast data is in macroscopic grid form, while the sensor data is in physical point form, the data access module embeds a spatial topology mapping table. Using this topology mapping table, the system uses coordinate analysis operators to transform physical coordinates in the actual geographic space into index identifiers for two-dimensional computational grid cells and one-dimensional pipeline nodes within the computer simulation domain, completing the spatial alignment of the two types of heterogeneous data.

[0032] After receiving and spatially aligning the meteorological forecast data V101 and the underground drainage network telemetry data V102, this embodiment further performs unified preprocessing on the aforementioned multi-source data. Specifically, the data access and cleaning server synchronizes the gridded rainfall forecast data stream and the real-time telemetry time series data according to a unified simulation calculation time step. For measured data in missing sensor periods, adjacent time intervals are interpolated to fill in the gaps. Abnormal measurement values ​​exceeding the physical range or exhibiting significant distortion are removed. The two types of data are then normalized to a unified dimension and numerical range. After the above preprocessing, the system obtains a spatially aligned and time-synchronized multi-source observation dataset. This multi-source observation dataset is the observation data input used in the extended Kalman filter assimilation algorithm in step S101. It serves as the data foundation for subsequent calculations of prior state estimates, the Kalman gain matrix, and the generation of posterior state estimates, thereby constructing the initial field matrix for simulation.

[0033] Figure 2 This is a logical flowchart of the multi-source data assimilation and initial field construction process provided in this application embodiment. The task scheduling center 201 drives the Kalman filter 202 to perform system matrix updates. Due to the highly nonlinear characteristics of the hydrodynamic evolution process, the system executes the extended Kalman filter assimilation algorithm to optimally estimate and update the hydrodynamic state vector. To quantify this process, the system first defines the hydrodynamic state vector V. 201 For X t This vector contains the water depth scalar and current velocity vector at the center of each discrete grid. Its nonlinear prediction state equation is expressed as:

[0034]

[0035] Among them, X t V represents the hydrodynamic state vector at time step t. 201 ; f represents the discrete state transition function of hydrodynamics based on the shallow water equation and the Saint-Venant equation; X t-1 u represents the state vector of the previous time step; t The input for the current time step is the weather forecast rainfall source item; W t This represents the process noise term that accompanies the model during its evolution.

[0036] Because the spatiotemporal resolution of the radar rainfall forecast data often differs from that of the measured data from underground pipe networks, the system needs to calculate and correct the Kalman gain. The system uses gridded rainfall data from meteorological forecasts and measured liquid level and flow velocity data from underground drainage pipe network telemetry data as the set of observation variables Z. tThe system processor performs forward inversion operations on the measured data vector and the measurement noise covariance matrix each cycle to correct the initial field. The system recursively calculates the prior state estimate and its prior error covariance matrix for the current time step using the hydrodynamic state vector and system state transition matrix from the previous time step. It then calculates the Kalman gain matrix using the observed variable set Zt, and updates and corrects the prior state estimate based on this Kalman gain matrix, generating a posterior state estimate as the assimilated and corrected data result, which is used to update the simulation initial field matrix. Since the data progresses in discrete time steps during the hydrodynamic evolution process, the specific logic for updating the system matrix by calculating the Kalman gain matrix using observed variables and updating and correcting the prior state estimate is as follows:

[0037]

[0038]

[0039] In this matrix system, K k P represents the Kalman gain matrix at the k-th time step; k - H represents the prior error covariance matrix generated by the system evolution at the k-th time step; H represents the observation matrix used to map the state space to the observation space; H T R represents the transpose of the observation matrix; R represents the observation noise covariance matrix related to the sensor hardware characteristics; x k x represents the posterior state estimate output after assimilation at the k-th time step; k - z represents the prior state estimate independently predicted by the model at the k-th time step; k This represents the observed variable entered at the k-th time step. The system can use the Kalman gain matrix K calculated in real time. k The system dynamically adjusts the confidence weight allocation between mathematical model predictions and physical measurement data. In system operations, this assimilation filtering mechanism effectively filters out high-frequency noise from observed variables caused by sensor fluctuations or intermittent rainfall. Subsequently, the task scheduling center 201 processes the smoothed posterior state estimate x... k Update to the initial field matrix V of the simulation 202 In the optimal state space, these serve as the initial boundary conditions for subsequent large-scale hydrodynamic calculations. Through this multi-source data assimilation and initial field construction process, the discrete initial error of the forecast data is significantly reduced, ensuring the convergence of the iterative solution of the nonlinear equation.

[0040] Coupled computation and adaptive step-size dynamic computation S102, based on the simulation initial field matrix V 202A coupled computational engine for a one-dimensional pipe network hydraulic model and a two-dimensional surface shallow water model is run to detect the water level gradient difference of interactive nodes to calculate bidirectional flux. Based on the CFL stability condition, the global maximum characteristic wave velocity is scanned before the macroscopic time step, and the maximum allowable time step of the next micro cycle is dynamically calculated. Adaptive step size reduction is triggered in response to local velocity mutations.

[0041] Figure 3 This is a logical structure block diagram of the one-dimensional and two-dimensional coupled computation control and adaptive time-step computation engine provided in this application embodiment. The computing node cluster is configured with a central processing unit (CPU) and a graphics processing unit (GPU). The CPU mainly undertakes the aforementioned data scheduling and assimilation work, while the GPU undertakes large-scale numerical parallel computation tasks. The GPU's video memory V... 302 The system is statically divided into a one-dimensional pipe network solution area 301, a two-dimensional surface solution area 302, and a coupled interactive boundary control area 303. The one-dimensional pipe network solution area 301 runs a pipe model based on the Saint-Venant equations; the two-dimensional surface solution area 302 runs a surface model based on the complete shallow water equations. The coupled interactive boundary control area 303 continuously updates mass and momentum flux at mapped and aligned interactive nodes through a bidirectional memory synchronization mechanism.

[0042] The computational unit periodically extracts the piezometric head at the manhole nodes of the one-dimensional pipe network hydraulic model, and the surface water depth at the corresponding surface grid nodes of the two-dimensional shallow surface water model. It then calculates the pipe network head h. 1d With surface water depth h 2d The difference in water level gradient between them. Based on intelligent judgment logic, the direction of water flow is determined and the matching weir outflow algorithm or orifice outflow algorithm is called to solve for the bidirectional interactive flux Q. ex Under typical orifice outflow conditions, the mathematical model for calculating water exchange data is expressed as follows:

[0043]

[0044] Among them, Q ex This represents the water exchange data occurring at the interaction node; c represents the flow coefficient associated with the geometry of the inspection well or storm drain; A o The effective physical cross-sectional area of ​​the exchange orifice is represented by g; g represents the gravitational acceleration constant; h 2d h represents the surface water depth calculated from a two-dimensional surface grid region. 1d This represents the network head obtained from the calculation of a one-dimensional underground pipeline system.

[0045] During periods of heavy rainfall, underground pipe networks transition from full-load to pressure flow, causing drastic fluctuations in water level gradients. This is to prevent the interactive flow Q generated during explicit differential solutions. exExcessive mass can cause the local mass conservation equations to diverge. This application embeds an adaptive time-step controller based on the CFL stability condition within the coupled interactive boundary control region 303. This controller operates at each macroscopic time step T. step Before starting, force a scan of the global computational grid to extract the global maximum characteristic wave velocity V. max The gravity wave velocity and convection velocity components of the global computational grid are traversed, and the two are superimposed to obtain the local characteristic wave velocity of each grid cell at this moment. The term with the largest absolute value is extracted as V. max .

[0046] Based on the globally maximum characteristic wave velocity extracted by the scan, the controller uses the stability conditional relationship to determine the fundamental micro step size Δt at the macroscopic level. base :

[0047]

[0048] Where, Δt base This represents the basic microstep size assigned by the computing engine; C CFL This represents the set Courant number safety threshold, which satisfies... ;Δx min V represents the minimum spatial step size in the global two-dimensional computational grid distribution; max V represents the global maximum characteristic wave velocity. max .

[0049] To accommodate the anisotropic water flow evolution in both the x and y dimensions, specifically within each two-dimensional interactive grid, the maximum allowable microstep size is determined. The refined restrictions are expressed in the form of the following dynamic calculation formula:

[0050]

[0051] in, Indicates the maximum allowed time step at the grid level; C r Corresponding to the aforementioned Courant constant threshold; and These represent the spatial step size of the grid in the horizontal and vertical directions, respectively; v x and v y represents the convective velocity components of the water flow in the x and y directions, respectively; h represents the water depth of the current grid. The time step ensures that within one period, gravity waves and convective materials in shallow water cannot penetrate and cross a distance exceeding one computational grid.

[0052] On this infrastructure, the adaptive controller implements frequency and order reduction pruning. During coupled computation, the system continuously monitors the local velocity gradient of the interaction nodes. When it detects drastic fluctuations in the difference between the surface water depth and the one-dimensional piezometric head, causing a sudden increase in local velocity and exceeding the preset gradient abrupt change threshold, it determines that the system faces the risk of numerical oscillation. The adaptive controller adjusts the basic microstep Δt according to a preset attenuation coefficient. base Forced frequency reduction and computation are performed, and the time step value obtained after pruning is used as the actual time step for the current micro-cycle iteration solution. Based on the CFL stability condition, the maximum allowable micro-step size and frequency reduction pruning logic are output, suppressing and inhibiting the oscillation divergence problem caused by coarse-grained forced coupling from the underlying computing power allocation.

[0053] S103: During the water drainage cutoff process and water accumulation duration statistics, in response to the detection that the calculated water depth of the two-dimensional grid is lower than the critical wet-dry threshold, the update of the convection term of the momentum equation of the two-dimensional grid is stopped and the normal velocity is cleared to zero. The hydrostatic term is retained to perform micro-perturbation calculation, and the time steps when the calculated water depth is greater than the preset water accumulation threshold are counted. The scalar result of the water accumulation duration of the node is output.

[0054] Figure 4 This is a flowchart of the smoothing operator and truncation process for the wet-dry boundary provided in this embodiment. During the receding water simulation phase after rainfall ends, the water flow gradually dissipates, and the water depth approaches zero. In traditional shallow water equation solvers, the denominator of the convection term in the momentum equation includes a water depth parameter, which will generate an extremely high, non-physical flow velocity when the water depth is extremely shallow. The system mounts a smoothing decision operator 401 and a flux truncation operator 402 in the memory computing area.

[0055] At the end of each micro-computation cycle, the smoothing decision operator 401 iterates through and extracts the computational water depth h of all two-dimensional grids within the two-dimensional computational domain. The extracted computational water depth h is then compared with a set critical wet / dry threshold H. dry Numerical comparison was performed. The critical wet / dry threshold H... dry H is configured as a constant greater than zero. dry The value is 0.005m.

[0056] The simulation was identified as being in a long-term recession simulation phase, and the calculated water depth h of a certain two-dimensional grid was determined. <H dryAt this time, flux cutoff 402 marks the corresponding grid as an interferometric flow-cut grid. For the marked interferometric flow-cut grid, in the discrete operation of updating the momentum conservation equation for the control volume, flux cutoff 402 performs a flux truncation operation: resetting the discrete flux coefficient of the convection term of its normal velocity vector to zero, and simultaneously clearing its normal velocity to zero. This operation cuts off the momentum exchange data path between the interferometric flow-cut grid and the adjacent normal wet grid, blocking the nonlinear convection iteration. After cutting off momentum exchange, in the mass conservation equation for water depth state update, the hydrostatic term affected by the bottom elevation gradient and the residual rainfall source term are retained to participate in the iterative summation calculation of the next time step, performing a small physical perturbation calculation. Based on this hybrid equation of cutting off momentum and retaining mass conservation, the water level state variables of the next time step are calculated, avoiding the floating-point overflow error caused by division by zero in the convection term at the dry boundary, and maintaining the non-negative water depth property in the global matrix solution process.

[0057] During this long-term simulation, the system configures an independent time accumulation register for each monitoring node in memory. In each calculation frame, in response to the determination that the calculated water depth exceeds a preset water accumulation threshold, the system increments the corresponding time accumulation register within each simulation micro-step period. When the overall simulation task is completed, the result rendering output gateway extracts the count values ​​stored in all time accumulation registers from the graphics processor's video memory, multiplies them by the actual dynamic fluctuation micro-step data, and converts them into a scalar result in the absolute time dimension. The output includes a vector spatial distribution layer containing the submerged area polygon and the time-series data of water accumulation duration for each node, and pushes it to the terminal platform. This wet-dry boundary processing mechanism ensures that the system maintains consistent matrix calculations during the ultra-long receding water cycle, achieving accurate statistics on water accumulation duration.

[0058] The threshold parameters such as 0.005m, 0.15m, and 0.4 given in the above embodiments are examples adapted to typical urban pipeline network physical environments. Those skilled in the art can use other equivalent numerical calibration logic to achieve the technical objectives of this application in terms of boundary control and step size convergence, based on specific terrain roughness or business requirements.

[0059] Based on the aforementioned embodiments, a multi-level frequency reduction buffer logic is configured in the coupled interactive boundary control region. When a sudden increase in interactive flow is detected by the water level gradient, the step size is trimmed according to a single fixed attenuation coefficient, and a proportional-integral-differential dynamic frequency reduction damper containing a hysteresis interval is introduced. Through dynamic damping adjustment, the macroscopic computing speed is maintained without divergence, improving the local computing power idleness defect caused by conventional single-stage halving and order reduction.

[0060] Based on the same inventive concept, this application provides an electronic device for performing the above-described method. Figure 5This is a system architecture diagram of an electronic device for performing urban flooding simulation and prediction tasks, provided in an embodiment of this application. The electronic device V 501 It includes a processor 501, a memory 502, and a communication bus interface 503. The memory 502 is configured to store the operating system and computer instruction programs for implementing the methods described above. The processor 501 and the memory 502 are electrically connected via an internal high-speed communication bus interface 503. The processor 501 calls and executes the computer instructions stored in the memory, and in the running state, implements the aforementioned logic steps for waterlogging simulation and prediction, including multi-source data Kalman assimilation, bidirectional adaptive time step calculation, and dry-wet boundary blocking based on critical water depth. This electronic device V 501 It can take the form of a centralized blade server cluster or a local high-performance workstation with a heterogeneous architecture including a central processing unit and a graphics processing unit.

[0061] This application provides a computer-readable storage medium storing a computer program that implements the aforementioned urban flooding simulation and prediction method based on multi-source data fusion. When the computer program is read and executed by a processor of an electronic device, it completes the urban flooding simulation and prediction method based on multi-source data fusion involved in all the aforementioned embodiments. The storage medium may be in the form of non-volatile flash memory, solid-state drive, magnetic storage medium, or optical disc, etc.

[0062] The solution provided in this application reconstructs the initial boundary by constructing a low-level data assimilation algorithm, eliminating discrete abrupt errors from multi-source heterogeneous input sources, and providing a convergent and stable foundational field for nonlinear hydrodynamic solutions. The synergistically introduced adaptive time-step constraint and frequency reduction scheduling scheme based on CFL stability conditions address the risks of mass non-conservation and divergence collapse during one-dimensional and two-dimensional grid interaction operations under heavy rainfall full-load conditions. Combined with a dry-wet smoothing interruptor for receding water conditions, it reduces the ineffective computational power consumption when the water depth approaches zero. These three elements generate a system-level synergistic effect at the temporal control and low-level matrix optimization levels, enabling the system to output prediction results of the inundation evolution process and node water accumulation duration with high computational efficiency and robustness.

[0063] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for simulating and predicting urban flooding based on multi-source data fusion, characterized in that, include: The system receives meteorological forecast data and underground drainage network telemetry data. The meteorological forecast data includes a gridded rainfall forecast data stream output by a weather radar system, and the underground drainage network telemetry data includes real-time telemetry time series data transmitted from IoT level sensors and ultrasonic flow meters. Spatial alignment processing is performed on the meteorological forecast data and the underground drainage network telemetry data. A coordinate analysis operator is used to transform physical coordinates into indexes of simulation calculation grid cells and one-dimensional pipeline nodes, resulting in spatially aligned multi-source observation data. An extended Kalman filter assimilation algorithm is then executed to estimate and correct the hydrodynamic state vector containing water depth and flow velocity, constructing and updating the initial simulation field matrix. The simulation initial field matrix is ​​used to run a coupled calculation engine of a one-dimensional pipe network hydraulic model and a two-dimensional surface shallow water model. The water level gradient difference of the interactive nodes is detected to calculate the bidirectional flux. Based on the CFL condition, the global maximum characteristic wave velocity is scanned before the macroscopic time step, and the maximum allowable time step of the next micro cycle is dynamically calculated. In response to local velocity mutations, an adaptive step size reduction is triggered. In the receding water calculation stage, in response to the detection that the calculated water depth of the two-dimensional grid is lower than the critical wet-dry threshold, the update of the convection term of the momentum equation of the two-dimensional grid is stopped and the normal velocity is cleared to zero. The hydrostatic term is retained to perform micro-perturbation calculation, and the time steps when the calculated water depth is greater than the preset water accumulation threshold are counted. The scalar result of the node water accumulation duration is output.

2. The urban flooding simulation and prediction method based on multi-source data fusion as described in claim 1, characterized in that, The execution of the extended Kalman filter assimilation algorithm estimates and corrects the hydrodynamic state vector containing water depth and flow velocity, constructs and updates the simulation initial field matrix, including: using gridded rainfall data from the meteorological forecast data and measured liquid level and flow velocity data from the underground drainage network telemetry data as observation variables; using the hydrodynamic state vector and system state transition matrix of the previous time step to calculate the prior state estimate and the corresponding prior error covariance matrix of the current time step; combining the observation variables to calculate the Kalman gain matrix, and using the Kalman gain matrix to update and correct the prior state estimate, generating the posterior state estimate as the simulation initial field matrix.

3. The urban flooding simulation and prediction method based on multi-source data fusion as described in claim 2, characterized in that, The calculation of the Kalman gain matrix based on the observed variables, and the use of the Kalman gain matrix to update and correct the prior state estimate, generating the posterior state estimate as the initial field matrix for simulation, are configured to execute the following system matrix update logic: Among them, K k P represents the Kalman gain matrix at the k-th time step; k - H represents the prior error covariance matrix at the k-th time step; H represents the observation matrix, H T R represents the transpose of the observation matrix; R represents the observation noise covariance matrix; x k x represents the posterior state estimate at the k-th time step; k - z represents the prior state estimate at the k-th time step; k The observed variable represents the variable at the k-th time step; based on the calculated Kalman gain matrix K... k The model dynamically adjusts the weight allocation between predictions and measured data, filters out high-frequency noise in the observed variables, and outputs the smoothed posterior state estimate x. k Update to the optimal state space of the initial field matrix of the simulation.

4. The urban flooding simulation and prediction method based on multi-source data fusion as described in claim 1, characterized in that, The detection of the water level gradient difference at the interaction nodes is used to calculate the bidirectional flux. Based on the CFL condition, the global maximum characteristic wave velocity is scanned before the macroscopic time step, and the maximum allowable time step of the next microscopic cycle is dynamically calculated. This includes: extracting the pipe head of the one-dimensional pipe hydraulic model at the inspection well node, and the surface water depth of the two-dimensional shallow surface water model at the corresponding surface grid; calculating the water level gradient difference between the pipe head and the surface water depth; when the absolute value of the water level gradient difference exceeds the set interaction trigger threshold, calculating the water exchange data occurring at the interaction node according to the outflow formula as the bidirectional flux; traversing the gravity wave velocity and convection velocity components of the global calculation grid, superimposing them to obtain the local characteristic wave velocity of each grid cell, and extracting the maximum value as the global maximum characteristic wave velocity.

5. The urban flooding simulation and prediction method based on multi-source data fusion as described in claim 4, characterized in that, The maximum allowable time step for dynamically calculating the next micro-cycle, in response to adaptive step size reduction triggered by a sudden change in local flow velocity, is configured to determine the basic micro-step size based on the extracted global maximum characteristic wave velocity using the following stability condition relationship: Where, Δt base C represents the basic microstep size; CFL This indicates the set threshold for the number of Courants, and ;Δx min V represents the minimum spatial step size in the global two-dimensional computational grid; max This represents the globally maximum characteristic wave velocity extracted by scanning; during the coupled computation process, the local velocity gradient is continuously monitored, and in response to a sudden increase in local velocity caused by full pipe overflow at the interaction node and the local velocity gradient exceeding a preset gradient abrupt change threshold, the basic microstep size Δt is adjusted according to a preset attenuation coefficient. base The frequency and order are reduced and pruned, and the time step value obtained after pruning is used as the actual time step used for the current micro-period iterative solution.

6. The urban flooding simulation and prediction method based on multi-source data fusion as described in claim 1, characterized in that, The response to detecting that the calculated water depth of the two-dimensional grid is lower than the critical wet-dry threshold, stopping the update of the convection term of the momentum equation of the two-dimensional grid and clearing the normal velocity to zero, while retaining the hydrostatic term to perform micro-perturbation calculations, includes: at the end of each micro-calculation cycle, traversing and extracting the calculated water depth of all the two-dimensional grids in the two-dimensional computational domain; numerically comparing the extracted calculated water depth with the set critical wet-dry threshold, which is configured as a constant greater than zero; when the comparison result determines that the calculated water depth is less than the critical wet-dry threshold, marking the corresponding grid as an interference-interrupted flow grid in the shallow water equation solver, and performing flux truncation operation in the numerical discretization format of the interference-interrupted flow grid.

7. The urban flooding simulation and prediction method based on multi-source data fusion as described in claim 6, characterized in that, The flux truncation operation performed in the numerical discretization scheme of the interferometric flow-cutting grid is configured as follows: for a cell marked as the interferometric flow-cutting grid, in the momentum conservation equation update operation of the control volume, the convection term discrete flux coefficient of its normal velocity vector is reset to zero, thereby cutting off the momentum exchange data path between the interferometric flow-cutting grid and the adjacent normal wet grid. In the water depth state update equation, the hydrostatic term affected by the bottom elevation gradient and the rainfall source term are retained to participate in the iterative summation calculation of the next time step; the water level state variables of the next time step are calculated based on the mass conservation equation after truncation of momentum exchange.

8. The urban flooding simulation and prediction method based on multi-source data fusion as described in claim 1, characterized in that, The process of calculating the time steps in which the calculated water depth exceeds the preset water accumulation threshold and outputting a scalar result of the node water accumulation duration includes: configuring an independent time accumulation register in the system; in response to determining that the calculated water depth exceeds the preset water accumulation threshold, performing an incrementing count operation on the time accumulation register within each simulation microstep period; and upon completion of the overall simulation task, extracting the count values ​​stored in all time accumulation registers and converting them into an absolute time dimension in combination with the actual microstep data, outputting a spatial water accumulation distribution map containing node time-series data.

9. An electronic device, characterized in that, include: A memory for storing computer instructions; a processor connected to the memory for executing the computer instructions stored in the memory to implement the urban flooding simulation and prediction method based on multi-source data fusion as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the urban flooding simulation and prediction method based on multi-source data fusion as described in any one of claims 1 to 8.

Images