A physical and data fusion driven flood simulation method

Abstract

The application provides a kind of physical and data fusion driven flood simulation method, flood simulation technical field, method includes: pre-training obtains the city flood simulation model of training qualified, wherein the target loss function includes data supervision loss term and physical constraint loss term, physical constraint loss term is by mass conservation residual loss term, momentum conservation residual loss term and energy conservation residual loss term constitute;Ground elevation data and historical meteorological data of target area in city are input to hydrological model, water depth grid data and flow velocity grid data are obtained, and they are input to city flood simulation model for prediction processing, obtain the water depth grid data and flow velocity grid data of target area at different time points in the future, based on the data of different time points, generate the dynamic evolution map of city flood waterlogging of target area.The purpose is to improve the efficiency and accuracy of city flood scenario prediction.

Description

Technical Field

[0001] This application relates to the field of flood simulation technology, and in particular to a flood simulation method driven by the fusion of physics and data. Background Technology

[0002] Urban flooding can have severe impacts on cities and residents, including economic losses, infrastructure damage, and disruption of public services. Therefore, urban flood simulation is crucial. Traditional urban flood simulation methods are primarily based on physical mechanisms, which simulate hydrological and hydrodynamic processes to reveal the physical mechanisms of flood formation. This was the core of early research; however, running physical models requires various types of hydrological and geomorphological monitoring datasets and intensive computation. Summary of the Invention

[0003] In view of this, this application provides a physics and data fusion-driven flood simulation method, aiming to solve or partially solve the problems existing in the background art.

[0004] In a first aspect of this application, a physics and data fusion-driven flood simulation method is provided, the method comprising:

[0005] A qualified urban flood simulation model is obtained through pre-training. The objective loss function in the urban flood simulation model includes a data supervision loss term and a physical constraint loss term. The physical constraint loss term consists of a mass conservation residual loss term, a momentum conservation residual loss term, and an energy conservation residual loss term.

[0006] Historical meteorological data of a target area in the city over a period of time under the current rainfall event are collected, and the historical meteorological data and the elevation data of the target area are input into the LISFLOOD-FP hydrological model for hydrological simulation to obtain water depth raster data and flow velocity raster data over a period of time. The meteorological data includes at least rainfall intensity data and flow rate data.

[0007] The water depth raster data and the flow velocity raster data over a past period are input into the urban flood simulation model for prediction processing to obtain water depth raster data and flow velocity raster data of the target area at different time points in the future.

[0008] Based on the obtained water depth raster data and flow velocity raster data of the target area at different time points in the future, a dynamic evolution map of urban flooding in the target area is generated.

[0009] Compared with prior art, this application has the following advantages:

[0010] This application provides a physics and data fusion-driven flood simulation method. First, a qualified urban flood simulation model is pre-trained. The target loss function in the urban flood simulation model includes a data supervision loss term and a physical constraint loss term. The physical constraint loss term consists of a mass conservation residual loss term, a momentum conservation residual loss term, and an energy conservation residual loss term. Ground elevation data and meteorological data of the target area in the current city are collected and input into the LISFLOOD-FP hydrological model for hydrological simulation to obtain current water depth raster data and flow velocity raster data. The meteorological data includes at least rainfall intensity data and flow rate data. The current water depth raster data and flow velocity raster data are then input into the urban flood simulation model for prediction processing to obtain water depth raster data and flow velocity raster data of the target area at different time points in the future. Based on the obtained water depth raster data and flow velocity raster data of the target area at different time points in the future, a dynamic evolution map of urban flood accumulation in the target area is generated.

[0011] Therefore, this application embeds the conservation of mass, momentum, and energy of water flow as physical constraints into the urban flood simulation model. This allows the model to directly integrate physical constraints during training, ensuring that the model's output not only matches observational data but also follows fundamental laws of fluid dynamics. This results in a more accurate dynamic simulation of urban flooding evolution over different time periods. Furthermore, the simulation results from the LISFLOOD-FP hydrological model are used to train the urban flood simulation model, achieving more precise and efficient predictions of urban flood scenarios.

[0012] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0014] Figure 1 A flowchart illustrating a physics and data fusion-driven flood simulation method provided in this application embodiment;

[0015] Figure 2 The image provided in this application is a dynamic evolution diagram of urban flooding in a certain urban area under a certain rainfall scenario, obtained by a flood simulation method driven by physical and data fusion. Detailed Implementation

[0016] Exemplary embodiments of this application will now be described in more detail with reference to the accompanying drawings.

[0017] Figure 1 A flowchart illustrating a physics and data fusion-driven flood simulation method provided in this application embodiment is shown below. Figure 1 As shown, the method includes:

[0018] Step S1: Pre-train to obtain a qualified urban flood simulation model. The target loss function in the urban flood simulation model includes a data supervision loss term and a physical constraint loss term. The physical constraint loss term consists of a mass conservation residual loss term, a momentum conservation residual loss term, and an energy conservation residual loss term.

[0019] In this embodiment, the present application pre-trains a qualified urban flood simulation model. The target loss function in the urban flood simulation model includes two loss terms: a data supervision loss term and a physical constraint loss term.

[0020] In this embodiment, this application provides a method for simulating urban flooding. Urban flooding simulation in urban scenarios is quite complex. Therefore, to ensure the accuracy of the urban flooding simulation as much as possible, this application proposes a physical constraint loss term for the physical constraint loss term in the target loss function. This physical constraint loss term consists of a mass conservation residual loss term, a momentum conservation residual loss term, and an energy conservation residual loss term. The reason for proposing this physical constraint loss term is that, for urban flooding, due to the large number of man-made facilities such as buildings, roads, and pipe networks in urban areas, kinetic energy and potential energy conversion frequently occur during water flow (e.g., the flow velocity decreases but the water level rises after water hits a building, or potential energy is converted into kinetic energy when falling from a height). Therefore, the energy conservation loss term, by constraining the sum of kinetic energy and gravitational potential energy to be conserved in space and time, can avoid abnormally increased flow velocity without a decrease in water level in the model prediction, and avoid non-physical phenomena such as energy non-conservation when water flows over obstacles, thereby ensuring that the simulation results conform to the actual energy transfer laws. Therefore, this application simultaneously defines the mass conservation residual loss term, the momentum conservation residual loss term, and the energy conservation residual loss term as the physical constraint loss term.

[0021] Step S2: Collect historical meteorological data of the target area in the city over a period of time under the current rainfall event, and input the historical meteorological data and the elevation data of the target area into the LISFLOOD-FP hydrological model for hydrological simulation to obtain water depth raster data and flow velocity raster data over a period of time. The meteorological data includes at least rainfall intensity data and flow rate data.

[0022] In this embodiment, under the current actual rainfall event, historical meteorological data of the target area in the city over a past period from the start of the rainfall event to the present is collected, and ground elevation data of the target area in the city recorded in the database is also obtained. Then, the obtained ground elevation data of the target area and the collected historical meteorological data are input into the LISFLOOD-FP hydrological model for hydrological simulation to predict and obtain water depth raster data and flow velocity raster data of the target area at various time points during the past period. The target area in the city can be any part of the city, such as a 50 square kilometer area or a 100 square kilometer area. The meteorological data includes at least rainfall intensity data and flow rate data.

[0023] Step S3: Input the water depth raster data and the flow velocity raster data from the past period into the urban flood simulation model for prediction processing to obtain water depth raster data and flow velocity raster data of the target area at different time points in the future.

[0024] In this embodiment, after obtaining a qualified urban flood simulation model through pre-training in step S1, under actual rainfall events, the water depth raster data and current velocity raster data of the target area obtained through hydrological simulation using the LISFLOOD-FP hydrological model at various time points over a past period are input into the qualified urban flood simulation model for prediction processing. The urban flood simulation model will output water depth raster data and current velocity raster data of the target area at different time points in the future, starting from the current moment. The future period can be preset according to the actual application scenario, such as within 5 hours, 10 hours, 24 hours, etc. The time intervals between different time points can also be preset according to the actual application scenario, such as water depth raster data results every 10 minutes, water depth raster data results every 5 minutes, etc.

[0025] Step S4: Based on the obtained water depth raster data and flow velocity raster data of the target area at different time points in the future, generate a dynamic evolution map of urban flooding in the target area.

[0026] In this embodiment, after predicting and obtaining water depth raster data and flow velocity raster data of the target area at different time points in the future starting from the current moment through step S3, the obtained water depth raster data and flow velocity raster data at different time points are processed by corresponding hydrological analysis tools (such as ArcGIS) to obtain a dynamic evolution map of urban flooding in the target area. Figure 2 As shown, Figure 2The figure shows the dynamic evolution of floodwater accumulation in a designated area of ​​a city under a certain rainfall scenario, 60 minutes, 120 minutes, 180 minutes and 240 minutes after the rainfall event, using the physics and data fusion-driven flood simulation method provided in this application. The water depth in the figure is in meters.

[0027] This application provides a physics and data fusion-driven flood simulation method. First, a qualified urban flood simulation model is pre-trained. The target loss function in the urban flood simulation model includes a data supervision loss term and a physical constraint loss term. The physical constraint loss term consists of a mass conservation residual loss term, a momentum conservation residual loss term, and an energy conservation residual loss term. Ground elevation data and meteorological data of the target area in the current city are collected and input into the LISFLOOD-FP hydrological model for hydrological simulation to obtain current water depth raster data and flow velocity raster data. The meteorological data includes at least rainfall intensity data and flow rate data. The current water depth raster data and flow velocity raster data are then input into the urban flood simulation model for prediction processing to obtain water depth raster data and flow velocity raster data of the target area at different time points in the future. Based on the obtained water depth raster data and flow velocity raster data of the target area at different time points in the future, a dynamic evolution map of urban flood accumulation in the target area is generated.

[0028] Therefore, this application embeds the conservation of mass, momentum, and energy of water flow as physical constraints into the urban flood simulation model. This allows the model to directly integrate physical constraints during training, ensuring that the model's output not only matches observational data but also follows fundamental laws of fluid dynamics. This results in a more accurate dynamic simulation of urban flooding evolution over different time periods. Furthermore, the simulation results from the LISFLOOD-FP hydrological model are used to train the urban flood simulation model, achieving more precise and efficient predictions of urban flood scenarios.

[0029] In conjunction with the above embodiments, in one implementation, this application also provides a physical and data fusion-driven flood simulation method. In this physical and data fusion-driven flood simulation method, step S1 may include:

[0030] Step S11: Construct the training dataset and test dataset for model training. The sample data in the dataset includes water depth raster data and current velocity raster data of a local area in the city at the target time, as well as water depth raster data and current velocity raster data of the local area at different time points after the target time.

[0031] In this embodiment, the application first constructs the dataset required for model training and divides the dataset into a training dataset and a test dataset according to a preset ratio. Preferably, the preset ratio is set so that the training dataset comprises 80% of the sample data, and the test dataset comprises 20%. A sample data set in the dataset includes: water depth raster data and current velocity raster data of a local area in the city at a target time, and water depth raster data and current velocity raster data of the same local area at different time points within a period after the target time. The target time is a moment after a rainfall event, and the duration of this period is the same as the aforementioned future period, with the same time interval between the two. The water depth raster data and current velocity raster data of the local area at different time points within a period after the target time are the labels of the sample data. The dataset of this application includes sample data of the same local area in the city under the influence of different meteorological data, and also includes sample data of different local areas in the city under the influence of the same meteorological data.

[0032] Step S12: Optimize the model parameters of the constructed initial urban flood simulation model using sample data from the training dataset.

[0033] In this embodiment, after obtaining the training dataset and test dataset through step S11, the initial urban flood simulation model is trained to optimize the model parameters using sample data from the training dataset.

[0034] Step S13: Determine the training result of the initial urban flood simulation model through the target loss function.

[0035] In this embodiment, during the training process of the initial urban flood simulation model, the training result of the model is determined by the model's objective loss function, and if the training result does not meet the set conditions, the model parameters are updated and the model training continues.

[0036] Step S14: If the training results meet the set conditions, evaluate the initial urban flood simulation model after training using sample data in the test dataset to obtain the corresponding evaluation results.

[0037] In this embodiment, when the training result obtained through the model's target loss function meets the set conditions, the initial urban flood simulation model after training is further evaluated using sample data from the constructed test dataset to obtain the corresponding evaluation results. The set conditions can be selected to verify whether the loss is stable; when the loss no longer decreases for several consecutive rounds (e.g., 3-5 rounds) or begins to rise, it is determined that the model has converged or is overfitting, and training is stopped at this point. Alternatively, early stopping can be selected, setting a patience value (e.g., patience=5); if the loss does not improve within 5 consecutive rounds, training is terminated.

[0038] Step S15: If the evaluation results meet the application requirements, obtain a qualified urban flood simulation model.

[0039] In this embodiment, when the evaluation results obtained through step S14 meet the corresponding application requirements, a qualified urban flood simulation model is determined to be obtained. These application requirements may include the model's accuracy, F1 score, MSE, etc., reaching their respective preset thresholds.

[0040] In conjunction with the above embodiments, in one implementation, this application also provides a physics and data fusion-driven flood simulation method. In this physics and data fusion-driven flood simulation method, constructing an initial urban flood simulation model includes: constructing a physical constraint loss term based on mass conservation residual loss terms, momentum conservation residual loss terms, and energy conservation residual loss terms, and the weights corresponding to each residual loss term; constructing a data supervision loss term based on the mean square error between the model's prediction results and the actual results; constructing a target loss function based on the constructed physical constraint loss term and the data supervision loss term, and the weights corresponding to each loss term; constructing a total residual vector based on the mass, momentum, and energy conservation of water flow; and constructing an initial urban flood simulation model based on the target loss function and the total residual vector.

[0041] In this embodiment, one method for constructing the initial urban flood simulation model is as follows: First, construct mass conservation residual loss terms, momentum conservation residual loss terms, and energy conservation residual loss terms. Then, assign a corresponding weight term to each residual loss term; that is, each residual loss term has a corresponding weight term. Multiply the mass conservation residual loss term by its corresponding weight term, and multiply the momentum conservation residual loss term by its corresponding weight term, and multiply the energy conservation residual loss term by its corresponding weight term. Finally, add these three products to obtain the physical constraint loss term.

[0042] In this embodiment, the mean square error between the prediction result obtained by using an urban flood simulation model to predict the input data and the actual result corresponding to the input data is... Construct a data supervision loss term. Among them, The prediction result obtained by the model using the input data. To predict the obtained water depth raster data, To predict the obtained flow velocity raster data, This represents the actual result corresponding to the input data. For real water depth raster data, This is real flow velocity raster data.

[0043] In this embodiment, each of the constructed physical constraint loss term and data supervision loss term has its own corresponding weight term. The physical constraint loss term is multiplied by its corresponding weight term, and the data supervision loss term is multiplied by its corresponding weight term. Finally, the two products are added together to obtain the target loss function.

[0044] In this embodiment, this application constructs the total residual vector in the urban flood simulation model based on the conservation of mass, momentum, and energy of water flow. Finally, based on the constructed objective loss function and the total residual vector, the final initial urban flood simulation model is obtained.

[0045] In conjunction with the above embodiments, in one implementation, this application also provides a physics and data fusion-driven flood simulation method. In this physics and data fusion-driven flood simulation method, a physical constraint loss term is constructed based on the mass conservation residual loss term, momentum conservation residual loss term, and energy conservation residual loss term, as well as the weights corresponding to each residual loss term. This includes constructing the mass conservation equation for water flow. Where h is the water depth, u is the longitudinal velocity in the x-direction, v is the transverse velocity in the y-direction, and t is time; based on the mass conservation equation of the water flow, the mass conservation residual loss term is determined. ,in, The residual loss term is determined by mass conservation; the momentum conservation equation for water flow is constructed. Based on the momentum conservation equation of water flow, determine the momentum conservation residual loss term. ,in, The residual loss term is for momentum conservation; the energy conservation equation for water flow is constructed. Based on the energy conservation equation for water flow, determine the energy conservation residual loss term. ,in, The energy conservation residual loss term is defined; based on the determined mass conservation residual loss term, momentum conservation residual loss term, and energy conservation residual loss term, and the weights corresponding to each residual loss term, a physical constraint loss term is constructed. ,in, , and These are the weighting coefficients.

[0046] In this embodiment, the present application first constructs the mass conservation equation for water flow. Where h is the water depth, u is the longitudinal velocity in the x-direction, v is the transverse velocity in the y-direction, and t is time. The momentum conservation equation for the water flow is also constructed. And, constructing the energy conservation equation for water flow. .

[0047] In this embodiment, after constructing the mass conservation equation, momentum conservation equation, and energy conservation equation for the water flow, the mass conservation residual loss term is determined based on the mass conservation equation for the water flow. ,in, This represents the mass conservation residual loss term. Also, based on the momentum conservation equation for water flow, the momentum conservation residual loss term is determined. ,in, This represents the momentum conservation residual loss term. Furthermore, based on the energy conservation equation for water flow, the energy conservation residual loss term is determined. ,in, This is the energy conservation residual loss term.

[0048] In this embodiment, after constructing the mass conservation residual loss term, momentum conservation residual loss term, and energy conservation residual loss term, a physical constraint loss term is constructed based on these three residual loss terms and their respective weights. ,in, , and These are the weighting coefficients.

[0049] In conjunction with the above embodiments, in one implementation, this application also provides a physics and data fusion-driven flood simulation method. In this physics and data fusion-driven flood simulation method, an optional implementation for constructing the total residual vector based on the conservation of mass, momentum, and energy of water flow is: determining the mass conservation residual term based on the mass conservation equation of water flow. Based on the momentum conservation equation of water flow, determine the momentum conservation residual term. Based on the energy conservation equation for water flow, determine the energy conservation residual term. Based on the determined mass conservation residual term, momentum conservation residual term, and energy conservation residual term, and the weight corresponding to each residual term, a total residual vector is constructed. ,in, , , These are the weighting coefficients.

[0050] In conjunction with the above embodiments, in one implementation, this application also provides a physics and data fusion-driven flood simulation method. In this physics and data fusion-driven flood simulation method, an initial urban flood simulation model is constructed based on the target loss function and the total residual vector, including: constructing a basic urban flood simulation model composed of an input layer with a first number of neurons, a hidden layer with a second number of neurons, and an output layer with a third number of neurons; setting each neuron of the first hidden layer as... ,in, For the i-th neuron in the first hidden layer, Let be the temporal weight matrix of the i-th neuron in the first hidden layer. Let be the weight matrix of the i-th neuron in the first hidden layer along the x-direction. Let be the weight matrix of the i-th neuron in the first hidden layer along the y-direction. Let be the bias vector of the i-th neuron in the first hidden layer. The hyperbolic tangent activation function is used; the neurons in other hidden layers are set to... ,in, Let j be the j-th neuron in all hidden layers except the first hidden layer. Let represent the connection weights from the i-th neuron in the first hidden layer to the j-th neuron in other hidden layers. Let be the bias vector of the j-th neuron in all hidden layers except the first hidden layer; based on the basic urban flood simulation model, the target loss function, and the total residual vector, an initial urban flood simulation model is constructed.

[0051] In this embodiment, the present application first constructs a basic urban flood simulation model consisting of an input layer with a first number of neurons, a hidden layer with a second number of neurons, and an output layer with a third number of neurons. Preferably, the first number of neurons is 3, the second number of neurons is 4 hidden layers, each hidden layer preferably has 50 neurons, and the third number of neurons is preferably 3.

[0052] In this embodiment, each neuron in the first hidden layer is set as... ,in, For the i-th neuron in the first hidden layer, Let be the temporal weight matrix of the i-th neuron in the first hidden layer. Let be the weight matrix of the i-th neuron in the first hidden layer along the x-direction. Let be the weight matrix of the i-th neuron in the first hidden layer along the y-direction. Let be the bias vector of the i-th neuron in the first hidden layer. The hyperbolic tangent activation function is used. Meanwhile, the neurons in other hidden layers are set to... ,in, Let j be the j-th neuron in all hidden layers except the first hidden layer. Let represent the connection weights from the i-th neuron in the first hidden layer to the j-th neuron in other hidden layers. is the bias vector of the j-th neuron in all hidden layers except the first hidden layer.

[0053] In this embodiment, the final initial urban flood simulation model is constructed based on the basic urban flood simulation model, the objective loss function, and the total residual vector.

[0054] In conjunction with the above embodiments, in one implementation, this application also provides a physical and data fusion-driven flood simulation method. In this physical and data fusion-driven flood simulation method, a target loss function is constructed based on the constructed physical constraint loss term and the data supervision loss term, and the weights corresponding to each loss term. This includes: determining a first weight of the physical constraint loss term as one, and determining a second weight of the data supervision loss term as one, to balance data fitting and physical constraints; and using the determined first weight and second weight to perform a weighted summation of the physical constraint loss term and the data supervision loss term to construct the target loss function.

[0055] In this embodiment, a simplified fixed-value strategy is used to balance the importance of data fitting and physical constraints in the model prediction process. Specifically, the first weight of the physical constraint loss term is set to one, and the second weight of the data supervision loss term is also set to one. Then, the physical constraint loss term is multiplied by its first weight, and the data supervision loss term is multiplied by its second weight. The results of these two multiplications are added together to obtain the final target loss function.

[0056] In conjunction with the above embodiments, in one implementation, this application also provides a physics and data fusion-driven flood simulation method. In this physics and data fusion-driven flood simulation method, constructing sample data includes: inputting ground elevation data and meteorological data of a designated area in a city under historical rainfall scenarios into the LISFLOOD-FP hydrological model for hydrological simulation to obtain water depth raster data and flow velocity raster data at different time points within a preset time period; determining the water depth raster data and flow velocity raster data at time points other than the starting time point within the obtained preset time period as sample labels, and using them together with the water depth raster data and flow velocity raster data at the starting time point to constitute the sample data.

[0057] In this embodiment, the water depth raster data and current velocity raster data at each time point in the sample data are obtained by hydrological simulation using the LISFLOOD-FP hydrological model.

[0058] Specifically, ground elevation data and meteorological data of a designated area in a city at a specific time under historical rainfall scenarios are input into the LISFLOOD-FP hydrological model for hydrological simulation to obtain water depth raster data and current velocity raster data for that designated area at that specific time. Using the same implementation method, water depth raster data and current velocity raster data at different time points within a continuous preset time period can be calculated. Then, the water depth raster data and current velocity raster data at other time points within the preset time period (excluding the starting time point) are used as sample labels, while the water depth raster data and current velocity raster data at the starting time point within the preset time period are used as input data to the model. The sample labels and the input data together form a sample dataset for subsequent model training. Using the same implementation method, a large amount of sample data is constructed to form a dataset for model training. This dataset is then divided to obtain the training dataset and test dataset required for training the model.

[0059] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily necessary for the embodiments of this application.

[0060] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0061] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, embodiments of this application can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of this application can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0063] The above provides a detailed description of a physical and data fusion-driven flood simulation method provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and its core ideas. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A physics and data-driven flood simulation method, characterized in that, The method includes: A qualified urban flood simulation model is obtained through pre-training. The objective loss function in the urban flood simulation model includes a data supervision loss term and a physical constraint loss term. The physical constraint loss term consists of a mass conservation residual loss term, a momentum conservation residual loss term, and an energy conservation residual loss term. Historical meteorological data of a target area in the city over a period of time under the current rainfall event are collected, and the historical meteorological data and the elevation data of the target area are input into the LISFLOOD-FP hydrological model for hydrological simulation to obtain water depth raster data and flow velocity raster data over a period of time. The meteorological data includes at least rainfall intensity data and flow rate data. The water depth raster data and the flow velocity raster data over a past period are input into the urban flood simulation model for prediction processing to obtain water depth raster data and flow velocity raster data of the target area at different time points in the future. Based on the obtained water depth raster data and flow velocity raster data of the target area at different time points in the future, a dynamic evolution map of urban flooding and waterlogging in the target area is generated. The initial urban flood simulation model corresponding to the aforementioned urban flood simulation model includes: Constructing the mass conservation equation for water flow Where h is the water depth, u is the longitudinal velocity in the x direction, v is the transverse velocity in the y direction, and t is time; Based on the mass conservation equation of water flow, determine the mass conservation residual loss term. ,in, This is the residual loss term for quality conservation; Constructing the momentum conservation equation for water flow ; Based on the momentum conservation equation of water flow, determine the momentum conservation residual loss term. ,in, This is the residual loss term due to momentum conservation; Constructing the energy conservation equation for water flow ; Based on the energy conservation equation for water flow, the energy conservation residual loss term is determined. ,in, This is the energy conservation residual loss term; Based on the determined mass conservation residual loss term, momentum conservation residual loss term, and energy conservation residual loss term, and the weights corresponding to each residual loss term, a physical constraint loss term is constructed. ,in, , and These are the weighting coefficients; Based on the mean squared error between the model's predictions and the actual results, a data supervision loss term is constructed. Based on the constructed physical constraint loss term and the data supervision loss term, as well as the weights corresponding to each loss term, a target loss function is constructed. Based on the conservation of mass, momentum, and energy of water flow, a total residual vector is constructed; An initial urban flood simulation model is constructed based on the target loss function and the total residual vector.

2. The flood simulation method driven by physics and data fusion according to claim 1, characterized in that, Pre-training to obtain a qualified urban flood simulation model includes: Construct training and testing datasets for model training. The sample data in the datasets include water depth raster data and current velocity raster data of a local area in the city at the target time, as well as water depth raster data and current velocity raster data of the local area at different time points after the target time. The initial urban flood simulation model was trained and its parameters optimized using sample data from the training dataset. The training results of the initial urban flood simulation model are determined by the objective loss function. If the training results meet the set conditions, the initial urban flood simulation model after training is evaluated using sample data in the test dataset to obtain the corresponding evaluation results; If the evaluation results meet the application requirements, a qualified urban flood simulation model is obtained.

3. The physics and data fusion-driven flood simulation method according to claim 1, characterized in that, Based on the conservation of mass, momentum, and energy of water flow, a total residual vector is constructed, including: Determine the mass conservation residual term based on the mass conservation equation of water flow. ; Based on the momentum conservation equation of water flow, determine the momentum conservation residual term. ; Based on the energy conservation equation of water flow, determine the energy conservation residual term. ; Based on the determined mass conservation residual term, momentum conservation residual term, and energy conservation residual term, and the weight corresponding to each residual term, a total residual vector is constructed. ,in, , , These are the weighting coefficients.

4. The flood simulation method driven by physics and data fusion according to claim 1, characterized in that, Based on the objective loss function and the total residual vector, an initial urban flood simulation model is constructed, including: A basic urban flood simulation model is constructed, consisting of an input layer with a first number of neurons, a hidden layer with a second number of neurons, and an output layer with a third number of neurons. Let the neurons in the first hidden layer be... ,in, For the i-th neuron in the first hidden layer, Let be the temporal weight matrix of the i-th neuron in the first hidden layer. Let be the weight matrix of the i-th neuron in the first hidden layer along the x-direction. Let be the weight matrix of the i-th neuron in the first hidden layer along the y-direction. Let be the bias vector of the i-th neuron in the first hidden layer. It is the hyperbolic tangent activation function; Configure each neuron in the other hidden layers as ,in, Let j be the j-th neuron in all hidden layers except the first hidden layer. Let represent the connection weights from the i-th neuron in the first hidden layer to the j-th neuron in other hidden layers. Let be the bias vector of the j-th neuron in all hidden layers except the first hidden layer; Based on the aforementioned basic urban flood simulation model, the aforementioned objective loss function, and the aforementioned total residual vector, an initial urban flood simulation model is constructed.

5. The physics and data fusion-driven flood simulation method according to claim 1, characterized in that, Based on the constructed physical constraint loss term and the data supervision loss term, and the weights corresponding to each loss term, a target loss function is constructed, including: A first weight of the physical constraint loss term is determined to be one, and a second weight of the data supervision loss term is determined to be one, in order to balance data fitting and physical constraints; Using the determined first and second weights, the physical constraint loss term and the data supervision loss term are weighted and summed to construct the target loss function.

6. The physics and data fusion-driven flood simulation method according to claim 2, characterized in that, Constructing sample data includes: The ground elevation data and meteorological data of a designated area in the city under historical rainfall scenarios are input into the LISFLOOD-FP hydrological model to perform hydrological simulation and obtain water depth raster data and flow velocity raster data at different time points within a preset time period. The water depth raster data and flow velocity raster data at other time points besides the starting time point within the preset time period are determined as sample labels, and together with the water depth raster data and flow velocity raster data at the starting time point, they constitute sample data.

Images