Water depth intelligent simulation assimilation method and device based on diffusion model

Abstract

The invention relates to an intelligent simulation assimilation method and device for ponding depth based on a diffusion model, and the method comprises the steps: giving full play to the advantages of the diffusion model based on a Bayesian theory and a data assimilation thought, and achieving the efficient correction of a flood model state through the limited real-time ponding monitoring data; by solving the logarithmic gradient of a posterior distribution function and guiding the sampling process, prior distribution and real-time observation information are effectively fused, so that the regional ponding state is stably estimated under the condition that monitoring points are sparse, error accumulation in model rolling calculation is remarkably inhibited, and the business precision and practicability of urban flood deduction are improved. Therefore, the problems that in the prior art, a systematic intelligent simulation-assimilation fusion system is not formed yet, and error accumulation in model rolling calculation is difficult to restrain are solved.

Description

Technical Field

[0001] This application relates to the field of data assimilation technology, and in particular to a method and apparatus for intelligent simulation assimilation of water depth based on a diffusion model. Background Technology

[0002] With the acceleration of climate change and urbanization, urban flooding disasters are becoming more frequent, placing higher demands on refined simulation and dynamic early warning. Currently, flood models are susceptible to multi-source uncertainties such as initial conditions, parameter and structural errors during long-term simulations, leading to error accumulation and decreased accuracy. Therefore, it is urgent to integrate real-time monitoring data to dynamically correct the model's state.

[0003] Data assimilation, as an effective theoretical framework for fusing observations and models, has demonstrated good correction capabilities in flood simulation. However, traditional methods such as variational assimilation and aggregate assimilation typically rely on extensive stochastic simulations or iterative optimizations, resulting in high computational costs and difficulty in adapting to the real-time assimilation requirements of high-dimensional, multi-source monitoring data. In recent years, artificial intelligence methods, represented by deep learning, have provided new paths for high-dimensional data modeling and have been gradually introduced into applications such as model error correction, parameter estimation, and end-to-end learning, significantly improving computational efficiency. However, existing intelligent assimilation methods mostly rely on high-coverage monitoring data as input, making it difficult to directly transfer to real-world scenarios like urban flooding where monitoring coverage is generally low. Therefore, existing technologies explore intelligent reconstruction methods for sparse monitoring conditions, such as image inpainting generative models and spatial coding networks, which can still achieve high-quality state reconstruction under extremely low monitoring coverage, providing valuable insights for sparse data assimilation in urban flood simulation.

[0004] However, the application of existing technologies in the field of flood control is still in its infancy. A systematic intelligent simulation-assimilation fusion system has not yet been formed, and it is difficult to suppress the accumulation of errors in the rolling calculation of the model, which urgently needs to be solved. Summary of the Invention

[0005] This application provides a method and apparatus for intelligent simulation and assimilation of water depth based on a diffusion model, in order to solve the problems that existing technologies have not yet formed a systematic intelligent simulation-assimilation fusion system and are difficult to suppress the accumulation of errors in the rolling calculation of the model.

[0006] The first aspect of this application provides a method for intelligent simulation and assimilation of water depth based on a diffusion model, comprising the following steps: performing unstructured grid discretization and hydro-hydrodynamic coupling calculations on a target urban study area to construct a flood numerical model of the target urban study area, and acquiring multiple random rainfall scenarios, and inputting the multiple random rainfall scenarios into the flood numerical model to construct an urban flood scenario dataset; inputting the urban flood scenario dataset into a pre-constructed unconditional diffusion model, and adding different levels of noise to the training samples in the urban flood scenario dataset through a denoiser in the unconditional diffusion model to generate corresponding noisy states, and reconstructing the original noise-free states corresponding to the urban flood scenario dataset based on the noisy states and a preset diffusion time step to obtain a trained unconditional diffusion model; and acquiring sparse water accumulation observation data corresponding to the target urban study area. The denoising process of the sparse water accumulation observation data is performed using the denoiser of the trained unconditional diffusion model to obtain a noise-free estimated state at the current moment. The noise-free estimated state is then mapped to the model prediction value of the corresponding observation point through a preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, the likelihood fraction function used to constrain the analysis variables to satisfy the observation constraints is calculated based on the actual water accumulation observation data of the target city study area. The prior distribution score of the urban flood distribution in the target city study area is estimated using the denoiser of the trained unconditional diffusion model, and the prior distribution score and the likelihood fraction function are added to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until the preset iteration end requirement is met, so as to finally obtain the final flood simulation result that meets the observation constraints.

[0007] Optionally, in one embodiment of this application, the step of performing unstructured grid discretization and hydro-hydrodynamic coupling calculation operations on the target urban study area to construct a flood numerical model of the target urban study area, and obtaining multiple random rainfall scenarios, and inputting the multiple random rainfall scenarios into the flood numerical model to construct an urban flood scenario dataset, includes: performing unstructured grid discretization on the target urban study area to obtain the river network boundary line and road network boundary line of the target urban study area, and using the river network boundary line and the road network boundary line as corresponding discrete control lines; constructing a flood numerical model of the target urban study area based on the discrete control lines and a preset hydro-hydrodynamic coupling calculation strategy; collecting preset historical precipitation data, and generating the multiple random rainfall scenarios based on a preset random rainstorm displacement strategy and the historical precipitation data, and inputting the multiple random rainfall scenarios into the flood numerical model to output corresponding flood process data, and constructing the urban flood scenario dataset based on the flood process data.

[0008] Optionally, in one embodiment of this application, the step of inputting the urban flood scenario dataset into a pre-constructed unconditional diffusion model to add different levels of noise to the training samples in the urban flood scenario dataset through a denoiser in the unconditional diffusion model to generate corresponding noisy states, and reconstructing the original noise-free states corresponding to the urban flood scenario dataset based on the noisy states and a preset diffusion time step to obtain a trained unconditional diffusion model, includes: inputting the urban flood scenario dataset into the pre-constructed unconditional diffusion model to determine a preset diffusion model training framework and a forward process based on a score assimilation framework according to the urban flood scenario dataset and the different levels of noise, and determining the corresponding training objective through the forward process; Based on the forward process and the training objective, a scoring function based on the fractional assimilation framework is constructed, and the correlation between the scoring function based on the fractional assimilation framework and the denoiser is determined. Based on the correlation, a state relationship is established between the diffusion model training framework and the states corresponding to the diffusion time steps in the forward process of the fractional assimilation framework. According to the state relationship, the denoiser in the unconditional diffusion model framework is converted into the corresponding scoring function form. Based on the scoring function form, different levels of noise are added to the training data of the urban flood scenario dataset to obtain the noisy state. The original noise-free state corresponding to the urban flood scenario dataset is reconstructed through the noisy state and the diffusion time step to obtain the trained unconditional diffusion model.

[0009] Optionally, in one embodiment of this application, the step of estimating the prior distribution score of urban flood distribution in the target city study area using the denoiser of the trained unconditional diffusion model, adding the prior distribution score to the likelihood function to obtain the posterior distribution function, and sampling the posterior distribution function to iteratively update the noisy state until a preset iteration end requirement is met, so as to finally obtain the final flood simulation result that meets the observation constraints, includes: estimating the prior distribution score of urban flood distribution in the target city study area using the denoiser of the trained unconditional diffusion model, so as to determine the probability of water accumulation at each spatial point in the target city study area based on the prior distribution score; mapping the simulated global water accumulation distribution corresponding to the target city study area to the water accumulation observation points through the preset observation operator, so as to solve the urban flood distribution. The likelihood function score of flood distribution is obtained; during the data assimilation process in the inference phase, the denoising device is used to convert the noisy state into the corresponding denoised state, and the denoised state is mapped to the corresponding observed value through the observation operator, and the actual observed water accumulation data corresponding to the target city study area is obtained; the observed value and the actual observed water accumulation data are compared to obtain the difference data between the observed value and the actual observed water accumulation data, and the fractional function of urban flood distribution is calculated based on the difference data, the likelihood function score, and the prior distribution score; based on the fractional function, the corresponding flood simulation information and noise state update direction are obtained, and based on the noise state update direction, the preset denoising operation is iteratively performed on the flood simulation information until the flood simulation data meets the preset denoising requirements, so as to generate the final flood simulation result of the target city study area.

[0010] A second aspect of this application provides an intelligent simulation and assimilation device for water depth based on a diffusion model, comprising: a construction module for performing unstructured grid discretization and hydro-hydrodynamic coupling calculations on a target urban study area to construct a flood numerical model of the target urban study area, and acquiring multiple random rainfall scenarios, and inputting the multiple random rainfall scenarios into the flood numerical model to construct an urban flood scenario dataset; a priori solution module for inputting the urban flood scenario dataset into a pre-constructed unconditional diffusion model, adding different levels of noise to the training samples in the urban flood scenario dataset through a denoiser in the unconditional diffusion model to generate corresponding noisy states, and reconstructing the original noise-free states corresponding to the urban flood scenario dataset based on the noisy states and a preset diffusion time step to obtain a trained unconditional diffusion model; and a likelihood solution module for obtaining the sparseness corresponding to the target urban study area. The system uses water accumulation observation data and the denoiser of the trained unconditional diffusion model to denoise the sparse water accumulation observation data to obtain a noise-free estimated state at the current time. The noise-free estimated state is then mapped to the model prediction value of the corresponding observation point using a preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, a likelihood fraction function is calculated based on the actual water accumulation observation data of the target city study area to constrain the analysis variables to meet the observation constraints. A posterior sampling module is used to estimate the prior distribution score of the urban flood distribution in the target city study area using the denoiser of the trained unconditional diffusion model. The prior distribution score is added to the likelihood fraction function to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until a preset iteration end requirement is met, ultimately obtaining the final flood simulation result that meets the observation constraints.

[0011] Optionally, in one embodiment of this application, the construction module includes: a discrete unit, used to perform unstructured grid discretization operations on the target city study area to obtain the river network boundary line and road network boundary line of the target city study area, and use the river network boundary line and the road network boundary line as the corresponding discrete control lines; a first establishment unit, used to construct a flood numerical model of the target city study area based on the discrete control lines and a preset hydrological-hydrodynamic coupling calculation strategy; and a data acquisition unit, used to acquire preset historical precipitation data, and generate the multiple random rainfall scenarios based on a preset random rainstorm displacement strategy and the historical precipitation data, and input the multiple random rainfall scenarios into the flood numerical model to output corresponding flood process data, and construct the urban flood scenario dataset based on the flood process data.

[0012] Optionally, in one embodiment of this application, the prior solution module includes: a determination unit, configured to input the urban flood scenario dataset into a pre-constructed unconditional diffusion model to determine a preset diffusion model training framework and a forward process based on a fractional assimilation framework according to the urban flood scenario dataset and the different levels of noise, and to determine the corresponding training objective through the forward process; a second establishment unit, configured to construct the score function based on the fractional assimilation framework based on the forward process and the training objective, and to determine the correlation between the score function based on the fractional assimilation framework and the denoiser, so as to establish a state relationship between the diffusion model training framework and the states corresponding to the diffusion time steps in the forward process of the fractional assimilation framework based on the correlation; a conversion unit, configured to convert the denoiser in the unconditional diffusion model framework into the corresponding score function form according to the state relationship; and a reconstruction unit, configured to add different levels of noise to the training data of the urban flood scenario dataset based on the score function form to obtain the noisy state, and to reconstruct the original noise-free state corresponding to the urban flood scenario dataset through the noisy state and the diffusion time step to obtain the trained unconditional diffusion model.

[0013] Optionally, in one embodiment of this application, the posterior sampling module includes: an estimation unit, used to estimate, using the denoiser of the trained unconditional diffusion model, a prior distribution score of the urban flood distribution within the target city study area, so as to determine the probability of water accumulation at each spatial point within the target city study area based on the prior distribution score; a mapping unit, used to map the simulated global water accumulation distribution corresponding to the target city study area to water accumulation observation points using the preset observation operator, so as to solve for the likelihood function score of the urban flood distribution; and a denoising unit, used to convert the noisy state into a corresponding denoised state using the denoiser during the data assimilation process in the inference stage, and to convert the denoised state into a denoised state using the observation operator. The system maps states to corresponding observed values ​​and acquires actual water accumulation observation data for the target city study area. A comparison unit compares the observed values ​​with the actual water accumulation observation data to obtain the difference data between them. Based on the difference data, the likelihood function score, and the prior distribution score, a fractional function of the city's flood distribution is calculated. An iteration unit acquires the corresponding flood simulation information and noise state update direction based on the fractional function. Based on the noise state update direction, iteratively performs a preset denoising operation on the flood simulation information until the flood simulation data meets the preset denoising requirements, thereby generating the final flood simulation result for the target city study area.

[0014] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intelligent simulation assimilation method for water depth based on a diffusion model as described in the above embodiments.

[0015] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described intelligent simulation assimilation method for water depth based on a diffusion model.

[0016] A fifth aspect of this application provides a computer program product, including a computer program that is executed to implement the above-described intelligent simulation assimilation method for water depth based on a diffusion model.

[0017] Therefore, the embodiments of this application have the following beneficial effects: The embodiments of this application construct a flood numerical model of the target city study area by performing unstructured grid discretization and hydro-hydraulic coupled calculation operations on the target city study area, and obtain various random rainfall scenarios. These scenarios are then input into the flood numerical model to construct an urban flood scenario dataset. The urban flood scenario dataset is input into a pre-constructed unconditional diffusion model. Different levels of noise are added to the training samples in the urban flood scenario dataset through a denoiser in the unconditional diffusion model to generate corresponding noisy states. Based on the noisy states and a preset diffusion time step, the original noise-free states corresponding to the urban flood scenario dataset are reconstructed to obtain the trained unconditional diffusion model. Sparse water accumulation observation data corresponding to the target city study area are obtained, and the trained unconditional diffusion model is then used... The model's denoiser processes sparse water accumulation observation data to obtain a noise-free estimate of the current state. It then maps this noise-free estimate to the corresponding observation point model prediction value using a preset observation operator for the target city study area. When the observation error follows a Gaussian distribution, it calculates the likelihood fraction function to constrain the analysis variables to meet the observation constraints based on the actual water accumulation observation data of the target city study area. The denoiser of the trained unconditional diffusion model estimates the prior distribution score of urban flood distribution in the target city study area. The prior distribution score and the likelihood fraction function are added to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until a preset iteration end requirement is met, ultimately obtaining the final flood simulation result that meets the observation constraints. This application effectively integrates the prior distribution with real-time observation information by solving the logarithmic gradient of the posterior distribution function and using it to guide the sampling process. This enables stable estimation of regional water accumulation under sparse monitoring points, significantly suppressing error accumulation in the model's rolling calculations and improving the operational accuracy and practicality of urban flood simulation. This solves the problems of existing technologies not yet forming a systematic intelligent simulation-assimilation fusion system, and the difficulty in suppressing error accumulation in model rolling calculations.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a smart simulation assimilation method for water depth based on a diffusion model, provided according to an embodiment of this application. Figure 2 A schematic diagram of urban flooding scenario data provided as an embodiment of this application; Figure 3 A schematic diagram of a model diffusion process and a reverse process is provided for one embodiment of this application; Figure 4 A schematic diagram of an unconditional diffusion model training process provided for one embodiment of this application; Figure 5 A schematic diagram of a data assimilation process provided for one embodiment of this application; Figure 6 This is an example diagram of a diffusion-based intelligent simulation assimilation device for water depth according to an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0020] Among them, 10-Intelligent simulation and assimilation device for water depth based on diffusion model; 100-Construction module, 200-Priority solution module, 300-Posterior sampling module, 400-Likelihood solution module; 701-Memory, 702-Processor, 703-Communication interface. Detailed Implementation

[0021] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0022] The following describes, with reference to the accompanying drawings, an intelligent simulation and assimilation method and apparatus for water depth based on a diffusion model, according to embodiments of this application. Addressing the problems mentioned in the background art, this application provides an intelligent simulation and assimilation method for water depth based on a diffusion model. In this method, an unstructured grid discretization and hydrological-hydraulic coupled calculation operation are performed on the target urban study area to construct a flood numerical model of the target urban study area. Multiple random rainfall scenarios are obtained and input into the flood numerical model to construct an urban flood scenario dataset. The urban flood scenario dataset is then input into a pre-constructed unconditional diffusion model. A denoiser in the unconditional diffusion model adds different levels of noise to the training samples in the urban flood scenario dataset to generate corresponding noisy states. Based on the noisy states and a preset diffusion time step, the original noise-free states corresponding to the urban flood scenario dataset are reconstructed to obtain the trained unconditional diffusion model. The method also obtains the sparse water depth corresponding to the target urban study area. The sparse water accumulation observation data is processed using a denoiser of a trained unconditional diffusion model to obtain a noise-free estimate of the current state. The noise-free estimate is then mapped to the model prediction value of the corresponding observation point using a preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, the likelihood fraction function used to constrain the analysis variables to meet the observation constraints is calculated based on the actual water accumulation observation data of the target city study area. The prior distribution score of the urban flood distribution in the target city study area is estimated using the denoiser of the trained unconditional diffusion model. The prior distribution score and the likelihood fraction function are added to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until a preset iteration end requirement is met, ultimately obtaining the final flood simulation result that meets the observation constraints. This application solves for the logarithmic gradient of the posterior distribution function and uses it to guide the sampling process, thereby effectively fusing prior distribution with real-time observation information. This enables stable estimation of regional waterlogging conditions even under sparse monitoring points, significantly suppressing error accumulation in rolling model calculations and improving the operational accuracy and practicality of urban flood simulations. Thus, it solves the problems of existing technologies lacking a systematic intelligent simulation-assimilation fusion system and struggling to suppress error accumulation in rolling model calculations.

[0023] Specifically, Figure 1 This is a flowchart of a method for intelligent simulation and assimilation of water depth based on a diffusion model, provided in an embodiment of this application.

[0024] like Figure 1 As shown, the intelligent simulation and assimilation method for water depth based on the diffusion model includes the following steps: In step S101, unstructured grid discretization and hydro-hydrodynamic coupling calculation operations are performed on the target city study area to construct a flood numerical model of the target city study area, and various random rainfall scenarios are obtained and input into the flood numerical model to construct an urban flood scenario dataset.

[0025] It should be noted that, based on Bayesian statistical theory, the problem of "estimating the state of a waterlogged field under given sparse water depth monitoring values" can be considered as a posterior distribution estimation problem. To solve for this posterior distribution, the embodiments of this application can take the logarithmic gradient of the Bayesian formula to obtain the following equation:

[0026] in, Given a water depth monitoring value, for The model state estimate at time step, The posterior distribution score is the value of the posterior distribution. For the prior distribution score, The value is the score of the likelihood function.

[0027] As can be seen from the above formula, the posterior distribution score to be solved is equal to the sum of the prior distribution score and the likelihood function distribution value. Therefore, the aforementioned problem of solving the posterior distribution can be further transformed into solving the prior distribution function score and the likelihood function score.

[0028] Therefore, as a feasible approach, the embodiments of this application first perform unstructured grid discretization on the entire study area (i.e., the target city study area) and conduct coupled hydrological and hydrodynamic calculations to construct a high-resolution flood numerical model for the region. Simultaneously, multiple random rainfall scenarios are acquired and input into the flood numerical model to construct an urban flood scenario dataset, providing reliable data support for subsequent urban flood projections.

[0029] Optionally, in one embodiment of this application, unstructured grid discretization and hydro-hydrodynamic coupling calculation operations are performed on the target urban study area to construct a flood numerical model of the target urban study area, and multiple random rainfall scenarios are obtained. These random rainfall scenarios are then input into the flood numerical model to construct an urban flood scenario dataset. This includes: performing unstructured grid discretization on the target urban study area to obtain the river network boundary lines and road network boundary lines of the target urban study area, and using these river network boundary lines and road network boundary lines as corresponding discrete control lines; constructing a flood numerical model of the target urban study area based on the discrete control lines and a preset hydro-hydrodynamic coupling calculation strategy; collecting preset historical precipitation data, and generating multiple random rainfall scenarios based on a preset random rainstorm displacement strategy and historical precipitation data, and inputting these multiple random rainfall scenarios into the flood numerical model to output corresponding flood process data, and constructing an urban flood scenario dataset based on the flood process data.

[0030] In its implementation, this application constructs a high-resolution flood numerical model based on unstructured grid discretization and hydro-hydrodynamic coupling calculations for the entire study area. During the unstructured grid discretization process, river network boundaries and road network boundaries are used as discretization control lines to ensure that the generated grid closely matches the actual terrain features, thereby ensuring that subsequent flood numerical calculations effectively reflect key confluence boundary information. For key areas covered by high-resolution elevation data, this application further reduces the maximum grid area constraint of the unstructured grid discretization to improve the accuracy of flood flow field simulation in key areas, thus providing reliable data support for flood control emergency assessment.

[0031] Furthermore, to fully reflect the spatiotemporal variability of rainfall fields, this application embodiment can employ statistical downscaling products (target spatiotemporal resolution of 1 hour and 1 kilometer) based on precipitation data from the US Climate Prediction Center's CMORPH satellite, combined with historical precipitation data, to generate no fewer than 40 random rainfall scenarios using a random storm displacement method. These scenarios are then used as the driving input for a flood numerical model to generate a high-resolution two-dimensional snapshot of urban waterlogging depth, thereby constructing a corresponding urban flood scenario dataset. This dataset is as follows: Figure 2 As shown.

[0032] Therefore, the embodiments of this application generate high-resolution two-dimensional snapshots of urban water depth using a fraction-based unconditional diffusion model, and use these snapshots as training data to drive the model to learn the high-dimensional distribution of water field simulation results of a high-precision numerical model, thereby effectively improving the generalization and reliability of model training.

[0033] In step S102, the urban flood scenario dataset is input into the pre-built unconditional diffusion model. Different levels of noise are added to the training samples in the urban flood scenario dataset through the denoiser in the unconditional diffusion model to generate the corresponding noisy state. Based on the noisy state and the preset diffusion time step, the original noiseless state corresponding to the urban flood scenario dataset is reconstructed to obtain the trained unconditional diffusion model.

[0034] It should be noted that, in the embodiments of this application, the score-based diffusion model used is a deep learning method capable of generating samples from highly complex data distributions. Its forward process involves the original data... Gradually add noise to transform it into near-random noise. Among them, diffusion time step The distribution changes from 0 to T. This distribution can be represented as:

[0035] Theoretically, the embodiments of this application are based on fractional functions. The above process can be repeated step by step in reverse. For example... Figure 3 As shown, in practice, the embodiments of this application can utilize a convolutional neural network UNet to approximate this fractional function. The training process involves providing noisy samples to the network. and its corresponding noise-free samples This allows the model to learn the denoising mapping. Once the denoiser is trained, the model can start from Gaussian white noise samples and gradually perform the reverse denoising process, transforming the data into structured data through iterative discrete-time steps.

[0036] Therefore, the embodiments of this application can effectively improve the model's fit to real flood scenarios, thereby providing more reliable model support for subsequent flood simulation, prediction and other tasks.

[0037] Optionally, in one embodiment of this application, an urban flood scenario dataset is input into a pre-built unconditional diffusion model to add different levels of noise to the training samples in the urban flood scenario dataset through a denoiser in the unconditional diffusion model, thereby generating corresponding noisy states. Based on the noisy states and a preset diffusion time step, the original noise-free states corresponding to the urban flood scenario dataset are reconstructed to obtain the trained unconditional diffusion model. This includes: inputting the urban flood scenario dataset into the pre-built unconditional diffusion model to determine a preset diffusion model training framework and a forward process based on a score assimilation framework according to the urban flood scenario dataset and different levels of noise, and determining the corresponding... Training objective: Based on the forward process and training objective, construct a scoring function based on a score assimilation framework, and determine the correlation between the scoring function based on the score assimilation framework and the denoiser. Based on the correlation, establish the state relationship between the diffusion model training framework and the states corresponding to the diffusion time steps in the forward process based on the score assimilation framework. According to the state relationship, convert the denoiser in the unconditional diffusion model framework into the corresponding scoring function form. Based on the scoring function form, add different levels of noise to the training data of the urban flood scenario dataset to obtain noisy states. Reconstruct the original noise-free states corresponding to the urban flood scenario dataset through the noisy states and diffusion time steps to obtain the trained unconditional diffusion model.

[0038] In practical implementation, embodiments of this application may use the elucidated diffusion model (EDM) training framework, which trains a neural network to process noisy states. and noise level Mapped to the denoised state In a score-based assimilation framework, the network's objective is not to directly predict the denoised state, but rather to predict the noise term. That is, given the noise state and the diffusion time The output noise is estimated. Therefore, embodiments of this application require first optimizing the noise denoiser in the EDM. Convert to common scoring function form The relationship is as follows (z represents):

[0039] To derive the above formula, embodiments of this application may be provided with... and Representing the forward processes of EDM and Score-based Data Assimilation (SDA) at time steps respectively. state, This is the Wiener process. The derivation consists of the following three steps: 1. Define the forward processes of EDM and SDA and their respective training objectives; 2. Deriving the noise denoiser With the scoring function in SDA The relationship between them; 3. Establish and The correspondence.

[0040] It should be noted that in EDM, the forward process is as follows:

[0041] noise reducer The training objective is:

[0042] Under the SDA framework, the forward process is as follows:

[0043] By training the fractional function approximate Therefore, we can conclude that: =

[0044]

[0045] Subsequently, embodiments of this application can be utilized express We can obtain the following formula:

[0046] The embodiments of this application are acceptable. To eliminate Thus, we obtain the following formula:

[0047] Subsequently, embodiments of this application may utilize z instead of The derivation can be completed by combining the last two equations.

[0048] It is understandable that, such as Figure 4 As shown, embodiments of this application can use a UNet model with an attention mechanism as a denoiser. During the training of the denoiser, noise is added to the training data at different levels and parameterized by the diffusion time step; the training objective of the denoiser is to reconstruct the training data given the noise state and time step.

[0049] Therefore, the embodiments of this application, by leveraging the attention mechanism UNet, enable the model to accurately learn the relationship between noise and state, thereby improving the ability to capture the characteristics of urban flood data, enhancing the accuracy of reconstructing the original state, and providing a reliable model foundation for subsequent flood scenario simulation and prediction.

[0050] In step S103, sparse water accumulation observation data corresponding to the target city study area is obtained, and the denoising data of the sparse water accumulation observation data is processed by the denoiser of the trained unconditional diffusion model to obtain the noise-free estimated state at the current time. The noise-free estimated state is mapped to the model prediction value of the corresponding observation point through the preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, the likelihood fraction function used to constrain the analysis variables to satisfy the observation constraints is calculated based on the actual water accumulation observation data of the target city study area.

[0051] In step S104, the prior distribution score of urban flood distribution in the target city study area is estimated using the denoiser of the trained unconditional diffusion model. The prior distribution score is then added to the likelihood function to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until the preset iteration end requirement is met, so as to finally obtain the final flood simulation result that meets the observation constraints.

[0052] In the embodiments of this application, the likelihood function fraction of urban flood distribution can be solved by a preset observation operator corresponding to the target city study area. It can be assumed that the observation error follows a Gaussian distribution, and the observation operator for the study area is set as follows. This is used to map the overall simulated water distribution to the water observation points, from which we can obtain:

[0053] in, It follows a Gaussian distribution; To simulate the distribution of water depth; Additional covariance of observation error Used to compensate for diffusion steps hour The model approximates the true state using a single Gaussian covariance to obtain an analytical likelihood, encompassing both observation and diffusion approximation errors. The logarithmic form of the likelihood probability function is:

[0054] in, This is similar to the second term of the three-dimensional variational assimilation loss function.

[0055] Therefore, the gradient of the likelihood term is:

[0056] This term reflects the direction of the constraint imposed by the observations on the diffusion variable. Its general formula, similar to the Gaussian likelihood, is:

[0057] Structurally identical, only the simple noise variance in the SDA framework is removed. This is further generalized to a covariance matrix V, and allows the observation operator H to be an arbitrarily differentiable mapping. On the other hand, the diffusion model provides prior scores. According to Bayes' rule, the gradient of the posterior distribution can be written as the sum of the prior score and the likelihood function, thus yielding the form of the posterior score:

[0058] Subsequently, in this embodiment, the observation information can be injected using a fraction-driven reverse process. Starting from random noise, the state is gradually guided to approach the observation constraints through the denoising process of the diffusion model, while maintaining the physical consistency of the generated distribution.

[0059] It is understood that the embodiments of this application, based on Bayesian theory and the idea of ​​data assimilation, fully leverage the advantages of diffusion models to achieve efficient correction of flood model states using limited real-time water accumulation monitoring data. Furthermore, the embodiments of this application, by solving the logarithmic gradient of the posterior distribution function and using it to guide the sampling process, effectively fuse prior distribution with real-time observation information. This enables stable estimation of regional water accumulation states even under sparse monitoring points, significantly suppressing error accumulation in model rolling calculations and improving the operational accuracy and practicality of urban flood simulation.

[0060] Optionally, in one embodiment of this application, the prior distribution score of urban flood distribution in the target city study area is estimated using the denoiser of the trained unconditional diffusion model. The prior distribution score is then added to the likelihood function to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until a preset iteration end requirement is met, so as to finally obtain the final flood simulation result that meets the observation constraints. This includes: estimating the prior distribution score of urban flood distribution in the target city study area using the denoiser of the trained unconditional diffusion model, and determining the probability of water accumulation at each spatial point in the target city study area based on the prior distribution score; mapping the simulated global water accumulation distribution corresponding to the target city study area to the water accumulation observation points through a preset observation operator to solve the problem. The likelihood function of urban flood distribution is calculated. During the data assimilation process in the inference phase, a denoiser is used to convert noisy states into corresponding denoised states, and the denoised states are mapped to corresponding observation values ​​through observation operators. Actual water accumulation observation data for the target city study area is also obtained. The observed values ​​and actual water accumulation observation data are compared to obtain the difference data between them. Based on the difference data, the likelihood function score, and the prior distribution score, a fractional function of urban flood distribution is calculated. Based on the fractional function, the corresponding flood simulation information and noise state update direction are obtained. Based on the noise state update direction, a preset denoising operation is iteratively performed on the flood simulation information until the flood simulation data meets the preset denoising requirements, thus generating the final flood simulation results for the target city study area.

[0061] In the specific implementation process, for solving the prior distribution score, this embodiment of the application can estimate the corresponding score value by training the diffusion model described above using the flood dataset and then using a denoiser, thereby determining the probability of water accumulation at each spatial location. (This can be approximately equivalent to the flood susceptibility at each spatial point).

[0062] For the fractional solution of the likelihood term, this embodiment of the application can assume that the observation error follows a Gaussian distribution and derive it based on step S103. In the inference stage, such as... Figure 5 As shown, in this embodiment of the application, the trained denoiser can first be utilized. Noisy state Transform into its noise-free estimate Subsequently, the state is estimated using the observation operator H. The model predictions are mapped to the corresponding observed values ​​and compared with the actual measured observation data y (e.g., waterlogging monitoring results). The difference between the model predictions and the actual observations is used to construct the gradient of the likelihood term, thereby calculating the posterior fractional function. This provides directional information for subsequent status updates.

[0063] To progressively sample from the posterior distribution and avoid error accumulation during multi-step updates, this embodiment employs the Langevin Monte Carlo (LMC) method for progressive denoising and state updates. During this iterative process, the model simultaneously utilizes learned prior information and likelihood constraints from actual observations to continuously correct the state, ultimately obtaining a fully denoised and observation-compliant data assimilation result when the time step approaches zero.

[0064] Therefore, the embodiments of this application can address the problem of error accumulation in the rolling calculation of flood models by using sparse monitoring data to achieve dynamic correction of the initial state of the model, significantly improving the accuracy and robustness of the model in operational simulations, promoting the improvement of the flood digital twin system's capabilities in virtual-real interaction, real-time simulations and accurate decision-making, and strongly supporting the construction of resilient cities and efficient disaster prevention and mitigation.

[0065] According to the intelligent simulation and assimilation method for water accumulation depth based on a diffusion model proposed in this application, a flood numerical model of the target city study area is constructed by performing unstructured grid discretization and hydrological-hydraulic coupled calculation operations on the target city study area. Multiple random rainfall scenarios are obtained and input into the flood numerical model to construct an urban flood scenario dataset. The urban flood scenario dataset is then input into a pre-constructed unconditional diffusion model. Different levels of noise are added to the training samples in the urban flood scenario dataset through a denoiser in the unconditional diffusion model to generate corresponding noisy states. Based on the noisy states and a preset diffusion time step, the original noise-free states corresponding to the urban flood scenario dataset are reconstructed to obtain the trained unconditional diffusion model. Sparse water accumulation observation data corresponding to the target city study area are obtained, and... The denoising function of the trained unconditional diffusion model is used to denoise the sparse water accumulation observation data to obtain the noise-free estimated state at the current moment. The noise-free estimated state is then mapped to the model prediction value of the corresponding observation point through the preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, the likelihood fraction function used to constrain the analysis variables to satisfy the observation constraints is calculated based on the actual water accumulation observation data of the target city study area. The prior distribution fraction of the urban flood distribution in the target city study area is estimated using the denoising function of the trained unconditional diffusion model. The prior distribution fraction and the likelihood fraction function are added to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until the preset iteration end requirement is met, so as to finally obtain the final flood simulation result that meets the observation constraints. This application solves the logarithmic gradient of the posterior distribution function and uses it to guide the sampling process, thereby effectively integrating the prior distribution with real-time observation information. It can stably estimate the regional water accumulation status under the condition of sparse monitoring points, significantly suppress the accumulation of errors in the rolling calculation of the model, and improve the operational accuracy and practicality of urban flood simulation.

[0066] Secondly, with reference to the accompanying drawings, a water depth-based intelligent simulation assimilation device based on a diffusion model is described according to an embodiment of this application.

[0067] Figure 6 This is a block diagram of an intelligent simulation assimilation device for water depth based on a diffusion model, according to an embodiment of this application.

[0068] like Figure 6 As shown, the intelligent simulation and assimilation device 10 based on the diffusion model for water depth includes: a construction module 100, a priori solution module 200, a posterior sampling module 300, and a likelihood solution module 400.

[0069] The construction module 100 is used to perform unstructured grid discretization and hydro-hydrodynamic coupling calculation operations on the target city study area to construct a flood numerical model of the target city study area, obtain multiple random rainfall scenarios, and input multiple random rainfall scenarios into the flood numerical model to construct an urban flood scenario dataset.

[0070] The prior solution module 200 is used to input the urban flood scenario dataset into the pre-built unconditional diffusion model, so as to add different levels of noise to the training samples in the urban flood scenario dataset through the denoiser in the unconditional diffusion model to generate the corresponding noisy state. Based on the noisy state and the preset diffusion time step, the original noiseless state corresponding to the urban flood scenario dataset is reconstructed to obtain the trained unconditional diffusion model.

[0071] The likelihood calculation module 300 is used to acquire sparse water accumulation observation data corresponding to the target city study area, and to denoise the sparse water accumulation observation data using the denoiser of the trained unconditional diffusion model to obtain the noise-free estimated state at the current time. The noise-free estimated state is mapped to the model prediction value of the corresponding observation point through the preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, the likelihood fraction function used to constrain the analysis variables to satisfy the observation constraints is calculated based on the actual water accumulation observation data of the target city study area.

[0072] The posterior sampling module 400 is used to estimate the prior distribution score of urban flood distribution in the target city study area using the denoiser of the trained unconditional diffusion model. The prior distribution score is added to the likelihood function to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until the preset iteration end requirement is met, so as to finally obtain the final flood simulation result that meets the observation constraints.

[0073] Optionally, in one embodiment of this application, the construction module 100 includes: a discrete unit, a first establishment unit, and a acquisition unit.

[0074] The discrete element is used to perform unstructured grid discretization on the target city study area to obtain the river network boundary line and road network boundary line of the target city study area, and use the river network boundary line and road network boundary line as the corresponding discrete control line.

[0075] The first unit is used to construct a numerical model of flooding in the target city study area based on discrete control lines and a preset hydrological-hydrodynamic coupling calculation strategy.

[0076] The data acquisition unit is used to collect preset historical precipitation data, and based on the preset random rainstorm displacement strategy and historical precipitation data, generate a variety of random rainfall scenarios. The various random rainfall scenarios are then input into the flood numerical model to output the corresponding flood process data, and an urban flood scenario dataset is constructed based on the flood process data.

[0077] Optionally, in one embodiment of this application, the priori solution module 200 includes: a determination unit, a second establishment unit, a transformation unit, and a reconstruction unit.

[0078] The determining unit is used to input the urban flood scenario dataset into the pre-built unconditional diffusion model to determine the preset diffusion model training framework and the forward process based on the score assimilation framework according to the urban flood scenario dataset and different levels of noise, and to determine the corresponding training target through the forward process.

[0079] The second establishment unit is used to construct a score function based on the score assimilation framework based on the forward process and the training objective, and to determine the correlation between the score function based on the score assimilation framework and the denoiser, so as to establish the state relationship between the diffusion model training framework and the state corresponding to the diffusion time step in the forward process based on the correlation.

[0080] The transformation unit is used to convert the denoiser in the unconditional diffusion model framework into the corresponding score function form according to the state relationship.

[0081] The reconstruction unit is used to add different levels of noise to the training data of the urban flood scenario dataset based on the scoring function form to obtain noisy states, and then reconstruct the original noise-free states corresponding to the urban flood scenario dataset through the noisy states and diffusion time steps to obtain the trained unconditional diffusion model.

[0082] Optionally, in one embodiment of this application, the posterior sampling module 300 includes: an estimation unit, a mapping unit, a denoising unit, a comparison unit, and an iteration unit.

[0083] The estimation unit is used to estimate the prior distribution score of urban flood distribution within the target city study area using the denoiser of the trained unconditional diffusion model, so as to determine the probability of water accumulation at each spatial point within the target city study area based on the prior distribution score.

[0084] The mapping unit is used to map the simulated global water accumulation distribution corresponding to the target city study area to the water accumulation observation point through a preset observation operator, so as to solve the likelihood function value of the urban flood distribution.

[0085] The denoising unit is used to convert noisy states into corresponding denoised states during the data assimilation process in the inference phase using a denoiser, and to map the denoised states to corresponding observation values ​​through observation operators, and to obtain the actual observation data of water accumulation in the target city study area.

[0086] The comparison unit is used to compare the observed values ​​with the actual observed water accumulation data to obtain the difference data between the observed values ​​and the actual observed water accumulation data, and to calculate the fractional function of urban flood distribution based on the difference data, the likelihood function score and the prior distribution score.

[0087] The iterative unit is used to obtain the corresponding flood simulation information and noise state update direction based on the fractional function, and to perform preset denoising operations on the flood simulation information iteratively based on the noise state update direction until the flood simulation data meets the preset denoising requirements, so as to generate the final flood simulation results of the target city study area.

[0088] It should be noted that the foregoing explanation of the embodiment of the intelligent simulation assimilation method for water depth based on the diffusion model also applies to the intelligent simulation assimilation device for water depth based on the diffusion model in this embodiment, and will not be repeated here.

[0089] The intelligent simulation and assimilation device for water depth based on a diffusion model proposed in this application includes a construction module 100, used to perform unstructured grid discretization and hydrological-hydraulic coupled calculation operations on the target urban study area to construct a flood numerical model of the target urban study area, and to acquire multiple random rainfall scenarios, and input the multiple random rainfall scenarios into the flood numerical model to construct an urban flood scenario dataset; a priori solution module 200, used to input the urban flood scenario dataset into a pre-constructed unconditional diffusion model, to add different levels of noise to the training samples in the urban flood scenario dataset through a denoiser in the unconditional diffusion model to generate corresponding noisy states, and to reconstruct the original noiseless states corresponding to the urban flood scenario dataset based on the noisy states and a preset diffusion time step, so as to obtain the trained unconditional diffusion model; and a likelihood solution module 300, used to acquire the corresponding values ​​of the target urban study area. The system uses sparse water accumulation observation data and a denoiser of a trained unconditional diffusion model to denoise the data, obtaining a noise-free estimate of the current state. The noise-free estimate is then mapped to the model prediction value of the corresponding observation point using a preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, the likelihood fraction function used to constrain the analysis variables to meet the observation constraints is calculated based on the actual water accumulation observation data of the target city study area. A posterior sampling module 400 estimates the prior distribution score of the urban flood distribution in the target city study area using the denoiser of the trained unconditional diffusion model. The prior distribution score is added to the likelihood fraction function to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until a preset iteration end requirement is met, ultimately obtaining the final flood simulation result that meets the observation constraints. This application solves the logarithmic gradient of the posterior distribution function and uses it to guide the sampling process, thereby effectively integrating the prior distribution with real-time observation information. It can stably estimate the regional water accumulation status under the condition of sparse monitoring points, significantly suppress the accumulation of errors in the rolling calculation of the model, and improve the operational accuracy and practicality of urban flood simulation.

[0090] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 701, the processor 702, and the computer program stored on the memory 701 and executable on the processor 702.

[0091] When the processor 702 executes the program, it implements the intelligent simulation assimilation method for water depth based on the diffusion model provided in the above embodiments.

[0092] Furthermore, electronic devices also include: Communication interface 703 is used for communication between memory 701 and processor 702.

[0093] The memory 701 is used to store computer programs that can run on the processor 702.

[0094] The memory 701 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0095] If the memory 701, processor 702, and communication interface 703 are implemented independently, then the communication interface 703, memory 701, and processor 702 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0096] Optionally, in a specific implementation, if the memory 701, processor 702, and communication interface 703 are integrated on a single chip, then the memory 701, processor 702, and communication interface 703 can communicate with each other through an internal interface.

[0097] The processor 702 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0098] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described intelligent simulation assimilation method for water depth based on a diffusion model.

[0099] This application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-described intelligent simulation assimilation method for water depth based on a diffusion model.

[0100] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0101] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0102] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0103] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0104] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0105] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0106] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0107] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for intelligent simulation and assimilation of water depth based on a diffusion model, characterized in that, Includes the following steps: Unstructured grid discretization and hydro-hydraulic coupled calculation operations are performed on the target city study area to construct a flood numerical model of the target city study area, and various random rainfall scenarios are obtained. The various random rainfall scenarios are then input into the flood numerical model to construct an urban flood scenario dataset. The urban flood scenario dataset is input into a pre-built unconditional diffusion model. Different levels of noise are added to the training samples in the urban flood scenario dataset through the denoiser in the unconditional diffusion model to generate corresponding noisy states. Based on the noisy states and a preset diffusion time step, the original noise-free states corresponding to the urban flood scenario dataset are reconstructed to obtain the trained unconditional diffusion model. Acquire sparse water accumulation observation data corresponding to the target city study area, and use the denoiser of the trained unconditional diffusion model to denoise the sparse water accumulation observation data to obtain the noise-free estimated state at the current time. The noise-free estimated state is mapped to the model prediction value of the corresponding observation point through the preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, the likelihood fraction function used to constrain the analysis variables to satisfy the observation constraints is calculated based on the actual water accumulation observation data of the target city study area. The prior distribution score of urban flood distribution in the target city study area is estimated using the denoiser of the trained unconditional diffusion model. The prior distribution score is then added to the likelihood function to obtain the posterior distribution function. The posterior distribution function is sampled to iteratively update the noisy state until the preset iteration end requirement is met, so as to finally obtain the final flood simulation result that meets the observation constraints.

2. The method according to claim 1, characterized in that, The process involves performing unstructured grid discretization and hydro-hydraulic coupled computation on the target urban study area to construct a flood numerical model for the area. Multiple random rainfall scenarios are then acquired and input into the flood numerical model to construct an urban flood scenario dataset, including: Unstructured grid discretization is performed on the target city study area to obtain the river network boundary line and road network boundary line of the target city study area, and the river network boundary line and road network boundary line are used as the corresponding discrete control lines; Based on the discrete control lines and the preset hydrological-hydraulic coupling calculation strategy, a flood numerical model of the target city study area is constructed. Collect preset historical precipitation data, and generate multiple random rainfall scenarios based on preset random rainstorm displacement strategies and the historical precipitation data. Input the multiple random rainfall scenarios into the flood numerical model to output corresponding flood process data, and construct the urban flood scenario dataset based on the flood process data.

3. The method according to claim 1, characterized in that, The process of inputting the urban flood scenario dataset into a pre-constructed unconditional diffusion model, adding different levels of noise to the training samples in the urban flood scenario dataset through a denoiser in the unconditional diffusion model to generate corresponding noisy states, and reconstructing the original noise-free states corresponding to the urban flood scenario dataset based on the noisy states and a preset diffusion time step to obtain the trained unconditional diffusion model includes: The urban flood scenario dataset is input into a pre-built unconditional diffusion model to determine a preset diffusion model training framework and a forward process based on a score assimilation framework, based on the urban flood scenario dataset and the different levels of noise, and to determine the corresponding training objective through the forward process. Based on the forward process and the training objective, the scoring function based on the fraction assimilation framework is constructed, and the correlation between the scoring function based on the fraction assimilation framework and the denoiser is determined. Based on the correlation, the state relationship between the diffusion model training framework and the states corresponding to the diffusion time steps in the forward process of the fraction assimilation framework is established. Based on the state relationship, the denoiser in the unconditional diffusion model framework is converted into the corresponding score function form; Based on the scoring function form, different levels of noise are added to the training data of the urban flood scenario dataset to obtain the noisy state. The original noise-free state corresponding to the urban flood scenario dataset is then reconstructed using the noisy state and the diffusion time step to obtain the trained unconditional diffusion model.

4. The method according to claim 1, characterized in that, The process involves estimating the prior distribution score of urban flood distribution in the target city study area using the denoiser of the trained unconditional diffusion model, adding the prior distribution score to the likelihood function to obtain the posterior distribution function, and sampling the posterior distribution function to iteratively update the noisy state until a preset iteration end requirement is met, ultimately obtaining the final flood simulation result that meets the observation constraints. This includes: Using the denoiser of the trained unconditional diffusion model, the prior distribution score of the urban flood distribution within the target city study area is estimated, so as to determine the probability of water accumulation at each spatial point within the target city study area based on the prior distribution score. The simulated global water accumulation distribution corresponding to the target city study area is mapped to the water accumulation observation point by the preset observation operator in order to solve the likelihood function value of the urban flood distribution. During the data assimilation process in the inference phase, the denoising device is used to convert the noisy state into a corresponding denoised state, and the observation operator maps the denoised state to the corresponding observation value. Obtain actual observation data of water accumulation in the target city's study area; The observed values ​​and the actual observed water accumulation data are compared to obtain the difference data between the observed values ​​and the actual observed water accumulation values. Based on the difference data, the likelihood function score and the prior distribution score, the fractional function of the urban flood distribution is calculated. Based on the fractional function, the corresponding flood simulation information and noise state update direction are obtained. Based on the noise state update direction, a preset denoising operation is iteratively performed on the flood simulation information until the flood simulation data meets the preset denoising requirements, so as to generate the final flood simulation result of the target city study area.

5. A smart simulation assimilation device for water depth based on a diffusion model, characterized in that, include: The module is used to perform unstructured grid discretization and hydro-hydrodynamic coupling calculation operations on the target city study area to construct a flood numerical model of the target city study area, and to obtain multiple random rainfall scenarios. The multiple random rainfall scenarios are then input into the flood numerical model to construct an urban flood scenario dataset. The prior solution module is used to input the urban flood scenario dataset into a pre-built unconditional diffusion model, so as to add different levels of noise to the training samples in the urban flood scenario dataset through the denoiser in the unconditional diffusion model to generate corresponding noisy states, and reconstruct the original noiseless state corresponding to the urban flood scenario dataset based on the noisy states and the preset diffusion time step to obtain the trained unconditional diffusion model. The likelihood calculation module is used to obtain sparse water accumulation observation data corresponding to the target city study area, and to denoise the sparse water accumulation observation data using the denoiser of the trained unconditional diffusion model to obtain the noise-free estimated state at the current time. The noise-free estimated state is mapped to the model prediction value of the corresponding observation point through the preset observation operator corresponding to the target city study area. When the observation error follows a Gaussian distribution, the likelihood fraction function used to constrain the analysis variables to satisfy the observation constraints is calculated based on the actual water accumulation observation data of the target city study area. The posterior sampling module is used to estimate the prior distribution score of urban flood distribution in the target city study area using the denoiser of the trained unconditional diffusion model, and to add the prior distribution score to the likelihood function to obtain the posterior distribution function. The posterior distribution function is then sampled to iteratively update the noisy state until a preset iteration end requirement is met, so as to finally obtain the final flood simulation result that meets the observation constraints.

6. The apparatus according to claim 5, characterized in that, The building module includes: Discrete cells are used to perform unstructured grid discretization operations on the target city study area to obtain the river network boundary line and road network boundary line of the target city study area, and use the river network boundary line and road network boundary line as the corresponding discrete control lines; The first establishment unit is used to construct a flood numerical model of the target city study area based on the discrete control lines and the preset hydrological-hydrodynamic coupling calculation strategy. The data acquisition unit is used to acquire preset historical precipitation data, and based on the preset random rainstorm displacement strategy and the historical precipitation data, generate the various random rainfall scenarios, and input the various random rainfall scenarios into the flood numerical model to output the corresponding flood process data, and construct the urban flood scenario dataset based on the flood process data.

7. The apparatus according to claim 5, characterized in that, The prior solution module includes: The determining unit is used to input the urban flood scenario dataset into a pre-constructed unconditional diffusion model to determine a preset diffusion model training framework and a forward process based on a score assimilation framework according to the urban flood scenario dataset and the different levels of noise, and to determine the corresponding training target through the forward process. The second establishment unit is used to construct the score function based on the score assimilation framework based on the forward process and the training objective, and determine the correlation between the score function based on the score assimilation framework and the denoiser, so as to establish the state relationship between the diffusion model training framework and the state corresponding to the diffusion time step in the forward process of the score assimilation framework based on the correlation. A conversion unit is used to convert the denoiser in the unconditional diffusion model framework into a corresponding score function form according to the state relationship; The reconstruction unit is used to add different levels of noise to the training data of the urban flood scenario dataset based on the scoring function form to obtain the noisy state, and reconstruct the original noise-free state corresponding to the urban flood scenario dataset through the noisy state and the diffusion time step to obtain the trained unconditional diffusion model.

8. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, the processor executing the program to implement the intelligent simulation assimilation method for water depth based on a diffusion model as described in any one of claims 1-4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the diffusion-based intelligent simulation assimilation method for water depth as described in any one of claims 1-4.

10. A computer program product, comprising a computer program, characterized in that, The computer program is executed to implement the intelligent simulation assimilation method for water depth based on a diffusion model as described in any one of claims 1-4.

Images