A flood rapid evolution and inundation simulation method based on 1D-CNN algorithm

Abstract

A flood rapid evolution and inundation simulation method based on a 1D-CNN algorithm, first, a two-dimensional hydrodynamic model of the research area is constructed, and the model roughness parameters are calibrated and verified; second, the two-dimensional hydrodynamic model is run to simulate the flood evolution and inundation simulation process under different historical working conditions, and the inundation water depth simulation results output by the model are extracted; then, the input and output structure of the 1D-CNN model is determined, and the training, verification and test sample sets are generated, the flood rapid evolution and inundation simulation model based on the 1D-CNN algorithm is established, and the internal weight parameters of the model are trained and optimized; finally, the application effect of the 1D-CNN model is tested, the simulation results of the two-dimensional hydrodynamic model are taken as the benchmark, the flood inundation simulation and prediction effect of the 1D-CNN model established and trained are compared and analyzed, and the accuracy and timeliness of the 1D-CNN model are evaluated. The present application can effectively solve the problems of long calculation time, large consumption and poor stability of the traditional two-dimensional hydrodynamic model, quickly and accurately simulate the flood evolution and inundation situation, meet the application requirements of real-time flood forecasting, and provide technical support for flood early warning and simulation work.

Description

Technical Field

[0001] This invention belongs to the field of flood forecasting technology and relates to a method for rapid flood evolution and inundation simulation based on the 1D-CNN algorithm. Background Technology

[0002] Rapidly and accurately simulating or predicting the evolution of floods and the spatiotemporal distribution of inundation information is an important means to accurately grasp the impact of floods and mitigate their losses. It is also one of the key technologies for flood early warning and simulation.

[0003] Two-dimensional hydrodynamic models possess powerful capabilities in simulating surface runoff. By establishing a two-dimensional hydrodynamic numerical model of the study area and using numerical solving algorithms to solve the continuity and motion equations of the flow, the spatiotemporal variations of flood elements such as water level, depth, velocity, and direction under given inflow conditions can be calculated, thereby enabling the simulation and prediction of flood evolution and inundation. However, due to the complexity and computational burden of solving the relevant physical equations of flow motion, the solution process for two-dimensional hydrodynamic numerical models is typically time-consuming, and the model's computational stability is poor, making it difficult to meet the needs of real-time flood forecasting.

[0004] Currently, deep learning technology is increasingly being applied to the water conservancy field due to its excellent data analysis and mining capabilities and good applicability. In particular, in recent years, models such as Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Long Short-Term Memory Networks (LSTM) have received widespread attention and research in hydrological forecasting, flood inundation prediction, and rainfall forecasting. Compared with traditional flood forecasting models, deep learning models are "black box" models. They achieve the purpose of simulation and prediction by mining the complex nonlinear relationships between the model's input and output data and applying them to unknown samples. They have the advantages of rapid construction, relatively simple model structure, ease of mastery, and ease of generalization.

[0005] One-dimensional convolutional neural networks (1D-CNNs) are artificial neural networks commonly used in time series models and natural language processing, possessing good capabilities in temporal capture and feature extraction analysis. Flood evolution can be understood as the change in water depth over time within a given inflow area. Therefore, this invention aims to improve the computational efficiency and stability of flood evolution and inundation simulation, proposing a rapid flood evolution and inundation simulation method based on the 1D-CNN algorithm. A one-dimensional convolutional neural network (1D-CNN) deep learning model is constructed. A training sample set is generated using the flow sequence input data from a two-dimensional hydrodynamic model and the spatially distributed simulated water depth output data. Through model training, the nonlinear relationship between the spatiotemporal changes of upstream inflow and inundation depth within the study area is established, enabling rapid flood evolution and inundation simulation analysis. Summary of the Invention

[0006] This invention addresses the problems existing in the application of existing technologies by providing a method for rapid flood evolution and inundation simulation based on the 1D-CNN algorithm.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for rapid flood evolution and inundation simulation based on 1D-CNN algorithm includes the following steps:

[0009] The first step is to construct a two-dimensional hydrodynamic model for the study area, and to calibrate and validate the parameters of the two-dimensional hydrodynamic model using historical floods. Specifically:

[0010] A two-dimensional hydrodynamic simulation model of the study area was constructed. The terrain mesh adopted was an unstructured triangular mesh. After terrain interpolation, a terrain file for the study area was generated for model identification and calculation. The roughness parameters of the two-dimensional hydrodynamic model were calibrated by selecting typical historical floods, and the model parameters and simulation accuracy were verified by selecting other historical floods.

[0011] The second step involves running the calibrated and validated two-dimensional hydrodynamic model to simulate flood evolution and inundation processes under different historical flood conditions. The inundation depth simulation results output by the two-dimensional hydrodynamic model are then extracted. Specifically:

[0012] Historical flood data for the study area were collected and compiled. The flow processes of these historical floods were used as inflow boundary conditions for a two-dimensional hydrodynamic model, driving the model constructed in the first step and outputting simulation results of flood evolution and inundation corresponding to the historical floods. This included the time-varying processes of hydraulic elements such as water level, depth, flow direction, and velocity across all triangular grids. The inflow boundary flow process data from historical floods and the simulated depth change data from the triangular grids will be used in the third step to determine the input-output structure of the 1D-CNN model and generate training, validation, and test datasets.

[0013] The third step is to determine the input and output structure of the 1D-CNN model and generate training, validation, and test sample sets. A rapid flood evolution and inundation simulation model based on the 1D-CNN algorithm is then established. The model's internal weight parameters are trained and optimized using the training and validation sample set data. Specifically:

[0014] 3.1) Determining the input and output structure of the 1D-CNN model

[0015] The input and output data of the 1D-CNN model are fixed-length vector sequences. Based on the inflow boundary flow sequence data from the two-dimensional hydrodynamic model in step two and the extracted triangular mesh simulation of water depth changes, the input and output structure of the 1D-CNN model is defined. According to the input and output settings of the two-dimensional hydrodynamic model, the input of the 1D-CNN model is designed as follows: Let P be the flow rate at the inflow boundary M at time t, and P be the number of preceding time periods at time t; the output is Output. t =[H t1 H t2 ,…,H ti ,…,H tW ], H ti Let be the water depth value of grid i at time t.

[0016] 3.2) Generate training, validation, and test sample sets for the 1D-CNN model.

[0017] Based on the input-output structure of the 1D-CNN model determined in step 3.1), a sample set is generated using the inflow boundary flow sequence data and the simulated water depth change process sequence data of the two-dimensional hydrodynamic model under different historical flood conditions in step 2, by adopting a time-by-time sliding segmentation method.

[0018] Historical flood events were divided into training and testing periods. During the training period, flood events generated training and validation sample sets in a 7:3 ratio. Training samples were used to train the weight parameters of the 1D-CNN model, while validation samples were used to test the training performance of the 1D-CNN model. Test set samples generated during the testing period were used to test the trained 1D-CNN model, verifying and evaluating its predictive performance.

[0019] 3.3) Construction and Training of 1D-CNN Flood Evolution and Inundation Simulation Model

[0020] Construct a 1D-CNN model consisting of one input layer, six hidden layers, and one output layer. The six hidden layers are: two convolutional layers, one flattening layer, and three dense fully connected layers. The structures of the input and output layers correspond to the input-output structure of the 1D-CNN model defined in step 3.1).

[0021] The 1D-CNN deep learning model uses an optimization method to determine its internal weight parameters based on historical sample data. Hyperparameters such as learning rate, loss function, activation function, optimization algorithm, batch size, and epoch are set, and the model weight parameters are trained using the training and validation sample sets from step 3.2).

[0022] The fourth step is to test the 1D-CNN model, analyze the accuracy and application effect of the trained 1D-CNN model in flood evolution and inundation simulation, and apply the 1D-CNN model to flood forecasting.

[0023] The test sample set in step 3.2) is used to drive the 1D-CNN model trained in step 3.3) to output flood evolution and inundation simulation results. The simulation results of the 1D-CNN model are then compared with the simulation results of the corresponding two-dimensional hydrodynamic model. Using the simulation results of the two-dimensional hydrodynamic model as a benchmark, the simulation accuracy and application effect of the 1D-CNN flood evolution and inundation simulation model are evaluated and analyzed.

[0024] If the simulation accuracy and application effect meet the requirements, then the 1D-CNN model will be applied to flood forecasting.

[0025] If the simulation accuracy and application effect do not meet the requirements, return to step two, collect more historical flood data, and repeat steps two, three, and four. Use more historical flood data to generate more 1D-CNN model training samples, improve the training degree of the 1D-CNN model weight parameters, and improve the simulation and prediction accuracy of the 1D-CNN model until it meets the requirements. Then apply the 1D-CNN model to flood forecasting in the study area.

[0026] The aforementioned method for rapid flood evolution and inundation simulation based on the 1D-CNN algorithm can be applied to flood forecasting to quickly and accurately simulate or predict the evolution and inundation of floods, meeting the needs of real-time forecasting.

[0027] The beneficial effects of this invention are as follows:

[0028] Traditional two-dimensional hydrodynamic models for flood evolution and inundation simulation and prediction are often computationally time-consuming and unstable, making it difficult to meet the needs of real-time flood forecasting. This invention constructs a one-dimensional convolutional neural network (1D-CNN) deep learning model, which uses the input and output data of the two-dimensional hydrodynamic model to generate training sample data for the 1D-CNN deep learning model. By learning the complex nonlinear relationship between the input flow data and the output triangular grid water depth change process data of the two-dimensional hydrodynamic model, it realizes the proxy of the traditional two-dimensional hydrodynamic model, and quickly and accurately simulates or predicts the spatiotemporal distribution information of flood inundation, thus meeting the needs of real-time forecasting. Attached Figure Description

[0029] Figure 1 This is a topographic map of the Nenjiang River main stream from Fularji to Jiangqiao section used in the application of this invention.

[0030] Figure 2 This is a schematic diagram of the terrain mesh of the two-dimensional hydrodynamic model constructed in this invention.

[0031] Figure 3 This is a schematic diagram of the input-output structure of the 1D-CNN model as defined in this invention.

[0032] Figure 4This is a structural diagram of the 1D-CNN deep learning model built in this invention.

[0033] Figure 5 This is a distribution map of the 12 control points selected in this invention for analyzing the accuracy of water depth simulation prediction results.

[0034] Figure 6 These are comparison diagrams of the water depth changes at control points Watermark 1 to 6 during the 2021 flood, simulated by the 1D-CNN model and the 2D hydrodynamic model in this invention. Specifically, Figure (a) shows the water depth change at Watermark 1, Figure (b) shows the water depth change at Watermark 2, Figure (c) shows the water depth change at Watermark 3, Figure (d) shows the water depth change at Watermark 4, Figure (e) shows the water depth change at Watermark 5, and Figure (f) shows the water depth change at Watermark 6.

[0035] Figure 7 These are comparative diagrams showing the water depth changes at control points Watermark 7–12 during a flood event in 2021, simulated using the 1D-CNN model and the 2D hydrodynamic model in this invention. Specifically, Figure (a) shows the water depth changes at Watermark 7, Figure (b) at Watermark 8, Figure (c) at Watermark 9, Figure (d) at Watermark 10, Figure (e) at Watermark 11, and Figure (f) at Watermark 12.

[0036] Figure 8 This is a graph showing the changes in three evaluation indicators used by the 1D-CNN model in this invention to predict the 2021 flood inundation process. Detailed Implementation

[0037] The present invention will be further illustrated below with specific examples.

[0038] This invention proposes a flood evolution and inundation simulation method based on the 1D-CNN algorithm. The results of flood events in the training and test sets, respectively, in a watershed example application represent the training and prediction performance of the deep learning model. The invention is further illustrated below with implementation examples and accompanying figures.

[0039] The Nenjiang River, located in Northeast my country, is a tributary of the Songhua River and one of the main sources of floodwaters for the Songhua River's main stream. The flood control dikes along the Nenjiang River from Fularji to Jiangqiao can reach a maximum distance of 16 km. The river channel is dotted with villages, towns, and farmland, and the dike situation is complex, posing a significant risk of flooding during the flood season. Flood control efforts along the Nenjiang River pay close attention to the evolution of floodwaters and the extent of inundation within the river channel. The topography of this section of the river is shown in the attached diagram. Figure 1 This river section was selected as a research example. A two-dimensional hydrodynamic model of the study area was constructed, and the water depth results output by the model were extracted to generate a sample set for training a deep learning model. This enabled rapid flood evolution and inundation simulation analysis based on the 1D-CNN deep learning algorithm. The main steps are as follows:

[0040] The first step is to construct a two-dimensional hydrodynamic model of the study area and calibrate and verify the parameters of the two-dimensional hydrodynamic model using historical floods.

[0041] A two-dimensional hydrodynamic model of the Nenjiang River main stream from Fularji to Jiangqiao was constructed using the MIKE 21FM model. The study area uses Fularji station on the main stream as the upstream boundary station and Jiangqiao station as the downstream boundary station. The main stream is fed by two tributaries, the Yalu River and the Chuoer River, with the Nianzishan and Liangjiazi hydrological stations on these tributaries, respectively. The model uses Fularji station on the main stream, Nianzishan station, and Liangjiazi station on the tributaries as input boundary stations, and the water level-discharge relationship at Jiangqiao station as the outflow boundary. The constructed topographic mesh of the study area is attached. Figure 2 In addition, the model incorporates Dike hydraulic structures to reflect the water-blocking effects of water-related structures such as dikes, embankments, roads, and railways within the region. The channel roughness parameters of the two-dimensional hydrodynamic model were calibrated using floods during the 2019 and 2020 flood seasons, and validated using the 2021 flood season flood.

[0042] The second step is to run the calibrated and verified two-dimensional hydrodynamic model to simulate the flood evolution and inundation process under different historical flood conditions, and extract the inundation depth simulation results output by the two-dimensional hydrodynamic model.

[0043] Historical hydrological data from 2010 to 2021 for the study area were collected and compiled, from which 12 historical flood events were selected. The discharge data of these 12 historical flood events were used as inflow boundary conditions to drive the two-dimensional hydrodynamic model constructed in the first step, outputting corresponding flood evolution and inundation simulation results. The inflow boundary discharge process data of the historical flood events and the simulated water depth change process data with extracted triangular meshes will be used in the third step to determine the input-output structure of the 1D-CNN model and generate training, validation, and test sample datasets.

[0044] The third step is to determine the input and output structure of the 1D-CNN model and generate training, validation, and test sample sets. A rapid flood evolution and inundation simulation model based on the 1D-CNN algorithm is then established. The model's internal weight parameters are trained and optimized using the training and validation sample set data. Specifically:

[0045] 3.1) Determining the input and output structure of the 1D-CNN model

[0046] The two-dimensional hydrodynamic model of the Nenjiang River main stream includes three inflow boundaries. The model output contains simulated water depth data of 73,202 triangular meshes within the modeling region. The specified number of preceding time periods is 64. The corresponding input and output structure of the 1D-CNN model is shown in the appendix. Figure 3 .

[0047] 3.2) Generate training, validation, and test sample sets for the 1D-CNN model.

[0048] The second step outputs simulation results of 12 two-dimensional hydrodynamic models representing different historical flood scenarios. Based on the input-output structure of the 1D-CNN model determined in step 3.1), a sample set is generated using a time-by-time sliding segmentation method. Among them, 1425 samples were generated for the 2021 flood as the test set; 35547 samples were generated for the other floods. These samples were randomly divided into training and validation sets in a 7:3 ratio.

[0049] 3.3) Construction and Training of 1D-CNN Flood Evolution and Inundation Simulation Model

[0050] A 1D-CNN model is constructed, consisting of one input layer, six hidden layers, and one output layer. The six hidden layers are: two convolutional layers, one flattened layer, and three dense fully connected layers (hidden layers). The two convolutional layers have 32 and 128 neurons respectively, and the three dense fully connected layers have 32, 256, and 512 neurons respectively. The structures of the input and output layers are consistent with the model input-output structure determined in step 3.1). A schematic diagram of the 1D-CNN model structure is shown below. Figure 4 .

[0051] The deep learning model uses an optimization method to determine its parameters based on historical sample data. The model learning rate is set to 0.001, the loss function to MSE, the activation function to ReLU, the optimization algorithm to Adam, the batch size to 10, and the epoch to 100. The model weight parameters are trained using the training and validation sample sets from step 3.2).

[0052] The fourth step is to test the 1D-CNN model, analyze the accuracy and application effect of the trained 1D-CNN model in flood evolution and inundation simulation, and apply the 1D-CNN model to flood forecasting.

[0053] The test set generated in step 3.2) is used to drive the 1D-CNN model trained in step 3.3) to generate flood evolution and inundation simulation results. The simulation results of the 1D-CNN model are compared with the simulation results of the corresponding two-dimensional hydrodynamic model. Based on the simulation results of the two-dimensional hydrodynamic model, the prediction accuracy and application effect of the 1D-CNN flood evolution model are analyzed.

[0054] A total of 12 control points were selected in the upstream and downstream areas and the main channel beach of the study area. The distribution of the control points is shown in the appendix. Figure 5 The simulated water depth results of the two models at the control points were compared, and the prediction error was quantified using two commonly used evaluation metrics, with the simulated water depth results of MIKE 21FM as the benchmark:

[0055] 1. Nash efficiency coefficient (NSE)

[0056]

[0057] In equation (1), N is the sample size, and Q is the sample size. i and P i These are the observed value and the predicted value at time i, respectively. In this case, the observed value refers to the simulated value of the two-dimensional hydrodynamic model, and the predicted value refers to the predicted value of the deep learning model. This represents the mean of the observed sequence. The value of NSE generally varies between 0 and 1, with NSE = 1 indicating a perfect fit between the two sets of data.

[0058] 2. Root Mean Square Error (RMSE)

[0059]

[0060] The meanings of the parameters in equation (2) are the same as those in the NSE calculation formula. RMSE = 0 indicates that the two sets of data are perfectly fitted.

[0061] A comparison chart of the water depth changes at 12 control points predicted by the two-dimensional hydrodynamic model and the 1D-CNN model for the 2021 flood events is attached. Figure 6 and attached Figure 7 The statistical results are shown in Table 1. As can be seen from Table 1, the NSE of the water depth at the 12 control points is almost always above 0.9, and the RMSE is also generally low. This is consistent with the water depth variation process. Figure 6 This demonstrates that the trained 1D-CNN model has learned well the process of water depth variation in the grid simulated by the two-dimensional hydrodynamic model, and can accurately predict the water depth variation process of grid cells at different locations within the study area.

[0062] Table 1 shows the NSE and RMSE of water depth results at 112 control points.

[0063]

[0064] The model's performance in simulating and predicting flood inundation extent across the entire study area was evaluated using Precision, Recall, and F1 indices.

[0065]

[0066]

[0067]

[0068] In the above formula, grids with a water depth greater than 0m are considered as submerged grids. Truepositive represents the number of submerged grids correctly predicted by the 1D-CNN model (i.e., also submerged in the 2D hydrodynamic model simulation); Totalpredictedpositive represents the total number of submerged grids predicted by the 1D-CNN model; Totalactualpositive represents the total number of submerged grids simulated by the 2D hydrodynamic model. Precision represents the accuracy of the 1D-CNN model's prediction results, and Recall represents the completeness of the 1D-CNN model's prediction results; F1 is the harmonic mean of Precision and Recall, defined as the degree of matching between the 1D-CNN model's prediction results and the 2D hydrodynamic model simulation results, with a value of 1 indicating a perfect match.

[0069] Appendix Figure 8 This study describes the changes in three assessment metrics for predicting the 2021 flood inundation process using a 1D-CNN model. From... Figure 8 As can be seen, the values ​​of Precision, Recall, and F1 are close to 1 at most times, indicating that the grid flooding situation simulated by the 1D-CNN model in the entire study area is consistent with the simulation results of the two-dimensional hydrodynamic model. In particular, the simulation water depth sequences of the two models are more consistent during periods of greater water depth.

[0070] Analysis of the water depth prediction results at 12 control points and the flood inundation range prediction results for the entire study area demonstrates that the trained 1D-CNN model can accurately predict the spatiotemporal variation of water depth within the study area, and the model exhibits high stability.

[0071] Furthermore, regarding computation time, Table 2 shows the computation time for simulating the 2021 flood process using the 1D-CNN model and the 2D hydrodynamic model. It can be seen that the 2D hydrodynamic model took 2 hours and 20 minutes to compute, while the 1D-CNN model only took 5 seconds to predict the flood's evolution and inundation. The deep learning model effectively reduced the time required for flood evolution and inundation prediction from hours to seconds.

[0072] Table 2 Calculation time for the two models

[0073]

[0074] The above results demonstrate that the proposed method for rapid flood evolution and inundation simulation based on the 1D-CNN algorithm can act as a proxy for traditional two-dimensional hydrodynamic models, achieve better simulation and prediction accuracy, and significantly improve the stability and timeliness of flood evolution and inundation simulation analysis and calculation, thus meeting the needs of real-time flood forecasting.

[0075] The above-described embodiments are merely illustrative of the implementation methods of the present invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for rapid flood evolution and inundation simulation based on a 1D-CNN algorithm, characterized in that, Includes the following steps: The first step is to construct a two-dimensional hydrodynamic model of the study area, and to calibrate and verify the parameters of the two-dimensional hydrodynamic model using historical floods. A two-dimensional hydrodynamic simulation model of the study area is then constructed, with an unstructured triangular mesh used for the terrain. After terrain interpolation, a terrain file for the study area is generated for model identification and calculation. The roughness parameters of the two-dimensional hydrodynamic model were calibrated by selecting typical historical floods, and the model parameters and simulation accuracy were verified by selecting additional historical floods. The second step is to run the calibrated and verified two-dimensional hydrodynamic model to simulate the flood evolution and inundation process under different historical flood conditions, and extract the inundation depth simulation results output by the two-dimensional hydrodynamic model. Historical flood data for the study area were collected and compiled. The flow processes of historical floods were used as inflow boundary conditions for a two-dimensional hydrodynamic model. This model was used to drive the two-dimensional hydrodynamic model constructed in the first step and output the flood evolution and inundation simulation results corresponding to the historical floods. This included the time-varying processes of water level, water depth, flow direction, flow velocity, and other hydraulic elements for all triangular grids. The inflow boundary flow process data of historical floods and the simulated water depth change process data of the triangular grids will be used in the third step to determine the input and output structure of the 1D-CNN model and generate training, validation, and test sample datasets. The third step is to determine the input and output structure of the 1D-CNN model and generate training, validation and test sample sets, establish a flood rapid evolution and inundation simulation model based on the 1D-CNN algorithm, and use the training and validation sample set data to train and optimize the internal weight parameters of the model. 3.1) Determining the input and output structure of the 1D-CNN model The input and output data of the 1D-CNN model are fixed-length vector sequences. Based on the inflow boundary flow sequence data from the two-dimensional hydrodynamic model in step two and the extracted triangular mesh simulation of water depth changes, the input and output structure of the 1D-CNN model is defined. Based on the input and output settings of the two-dimensional hydrodynamic model, the input of the 1D-CNN model is designed as follows: , for t Inflow boundary at any moment M Traffic value, P for t Number of time intervals preceding the current time; The output is , for t Time Grid i The water depth value; 3.2) Generate training, validation, and test sample sets for the 1D-CNN model. Based on the 1D-CNN model input-output structure determined in step 3.1), a sample set is generated using the inflow boundary flow sequence data and the simulated water depth change process sequence data of the two-dimensional hydrodynamic model under different historical flood conditions in step 2, by adopting a time-by-time sliding segmentation method. Historical flood events are divided into training and testing periods. During the training period, flood events are used to generate training and validation sample sets in a 7:3 ratio. Training samples are used to train the weight parameters of the 1D-CNN model, and validation samples are used to test the training effect of the 1D-CNN model during training. The test set samples generated during the flood process during the test period were used to test the trained 1D-CNN model, and to verify and evaluate the predictive performance of the 1D-CNN model. 3.3) Construction and Training of 1D-CNN Flood Evolution and Inundation Simulation Model Construct a 1D-CNN model whose input and output layer structures correspond to the 1D-CNN model input-output structure defined in step 3.1); The 1D-CNN deep learning model uses an optimization method based on historical sample data to determine the internal weight parameters; The model weight parameters are trained using the training and validation sample sets from step 3.2); The fourth step involves testing the 1D-CNN model, analyzing its accuracy and application effectiveness in simulating flood evolution and inundation, and applying the 1D-CNN model to flood forecasting. The test sample set from step 3.2) is used to drive the 1D-CNN model trained in step 3.3) to output flood evolution and inundation simulation results. These results are then compared with the corresponding two-dimensional hydrodynamic model simulation results. Using the two-dimensional hydrodynamic model simulation results as a benchmark, the simulation accuracy and application effectiveness of the 1D-CNN flood evolution and inundation simulation model are evaluated and analyzed. If the simulation accuracy and application effect meet the requirements, then the 1D-CNN model will be applied to flood forecasting. If the simulation accuracy and application effect do not meet the requirements, return to step two, collect more historical flood data, and repeat steps two, three, and four. Use more historical flood data to generate more 1D-CNN model training samples, improve the training degree of the 1D-CNN model weight parameters, and improve the simulation and prediction accuracy of the 1D-CNN model until it meets the requirements. Then apply the 1D-CNN model to flood forecasting in the study area.

2. The method for rapid flood evolution and inundation simulation based on 1D-CNN algorithm according to claim 1, characterized in that, In step 3.1, the 1D-CNN model includes one input layer, six hidden layers, and one output layer; the six hidden layers are: two convolutional layers, one flattening layer, and three dense fully connected layers.

3. An application of the flood rapid evolution and inundation simulation method based on the 1D-CNN algorithm as described in claim 1 or 2, characterized in that, The aforementioned method for rapid flood evolution and inundation simulation is applied to flood forecasting, specifically for the rapid evolution and inundation simulation prediction of river floods.

Images