A city waterlogging prediction method based on visual recognition

Abstract

The application discloses a kind of urban waterlogging prediction methods based on visual identification, belong to urban waterlogging prediction technical field, can solve the problem of lower prediction accuracy of existing prediction method, and prediction space and time are limited.Said method includes: S1, according to the waterlogging image of multiple waterlogging positions in target area, determine the real-time waterlogging depth of each waterlogging position;S2, according to the weather data of target area, historical waterlogging depth and real-time waterlogging depth, determine the predicted water depth time sequence of each waterlogging position in prediction period;S3, according to the predicted water depth time sequence of all waterlogging positions in prediction period, determine the prediction waterlogging information of target area in prediction period.The application is used to predict urban waterlogging information.

Description

TECHNICAL FIELD

[0001] The present application relates to a visual recognition-based urban waterlogging prediction method, belonging to the technical field of urban waterlogging prediction. BACKGROUND

[0002] Urban waterlogging prediction is a method of predicting future urban waterlogging disasters by monitoring the waterlogging conditions in urban areas, which is of great significance to ensuring urban functions and resident travel safety.

[0003] Currently, urban waterlogging prediction methods usually simulate waterlogging data of a target area according to meteorological data, and then predict waterlogging according to the simulated waterlogging data. Since there is a deviation between the simulated waterlogging data and the actual waterlogging data, the prediction accuracy of the existing prediction method is low. At the same time, the existing prediction method can only predict the waterlogging conditions of a few specific waterlogging positions in the target area at a plurality of discrete prediction times, and cannot predict the overall waterlogging state of the target area or the waterlogging conditions of the target area in future continuous time periods. Therefore, the prediction space and time of the existing prediction method are limited. SUMMARY

[0004] The present application provides a visual recognition-based urban waterlogging prediction method, which can solve the problem of low prediction accuracy and limited prediction space and time of the existing prediction method.

[0005] The present application provides a visual recognition-based urban waterlogging prediction method, which comprises:

[0006] S1, determining the real-time waterlogging depth of each waterlogging position according to the waterlogging images of a plurality of waterlogging positions in a target area;

[0007] S2, determining the predicted water depth time sequence of each waterlogging position in a prediction period according to the meteorological data, historical waterlogging depth and real-time waterlogging depth of the target area;

[0008] S3, determining the predicted waterlogging information of the target area in the prediction period according to the predicted water depth time sequence of all waterlogging positions in the prediction period.

[0009] Optionally, the S1 is specifically:

[0010] S11, determining the real-time flooding level of each waterlogging position according to the waterlogging images of a plurality of waterlogging positions in a target area;

[0011] S12, determining the real-time waterlogging depth of each waterlogging position according to the real-time flooding level.

[0012] Optionally, the S11 is specifically:

[0013] S111, identifying a preset reference from the waterlogging image of each waterlogging position in the target area, and determining real-time boundary information of the preset reference in the waterlogging image;

[0014] S112, determining a real-time flooding level of each waterlogging position according to the real-time boundary information.

[0015] Optionally, the S112 is specifically:

[0016] constructing a first correspondence relationship between the flooding level and the boundary information of the preset reference in the waterlogging image;

[0017] determining a real-time flooding level of each waterlogging position according to the first correspondence relationship and the real-time boundary information.

[0018] Optionally, the S12 is specifically:

[0019] constructing a second correspondence relationship between the flooding level and the waterlogging depth;

[0020] determining a real-time waterlogging depth of each waterlogging position according to the second correspondence relationship and the real-time flooding level.

[0021] Optionally, the meteorological data includes historical meteorological data and short-impending meteorological data; and the S2 is specifically:

[0022] S21, constructing a water depth time series prediction model by using a spatio-temporal graph neural network according to the historical meteorological data and the historical waterlogging depth of the target area;

[0023] S22, determining a predicted water depth time series of each waterlogging position in a prediction period by using the water depth time series prediction model according to the short-impending meteorological data of the target area and the real-time waterlogging depth.

[0024] Optionally, the water depth time series prediction model includes a prediction unit and a correction unit; and the S22 is specifically:

[0025] determining a predicted water depth of each waterlogging position at a current prediction time in a prediction period by using the prediction unit according to the short-impending meteorological data of the target area and the real-time waterlogging depth;

[0026] correcting the predicted water depth at the current prediction time by using the correction unit, taking the corrected predicted water depth as an input of the prediction unit to perform water depth prediction at a next prediction time, until predicted water depths at all prediction times in the prediction period are determined, to obtain a predicted water depth time series of each waterlogging position in the prediction period.

[0027] Optionally, the S3 is specifically:

[0028] determine the predicted water depth of any position in the target area corresponding to any time in the predicted period according to the predicted water depth time sequence of all water accumulation positions;

[0029] determine the predicted waterlogging information of the target area in the predicted period according to the predicted water depth of each position in the target area corresponding to each time in the predicted period.

[0030] Optionally, before the S1, the method further comprises:

[0031] According to the image information of the plurality of monitoring points in the target area, a large language model is used to screen a plurality of water accumulation positions from the plurality of monitoring points.

[0032] Optionally, the S111 specifically comprises:

[0033] A semantic segmentation model is used to identify a preset reference object from the water accumulation image of each water accumulation position in the target area, and real-time boundary information of the preset reference object in the water accumulation image is determined.

[0034] The present application can produce beneficial effects, including:

[0035] The present application identifies the real-time water accumulation depth of each water accumulation position according to the water accumulation image of the plurality of water accumulation positions, and then combines the meteorological data and the historical water accumulation depth of the target area to determine the predicted water depth time sequence of the plurality of water accumulation positions in the predicted period, and further predicts the waterlogging information of the entire target area in the predicted period according to the predicted water depth time sequence of the plurality of water accumulation positions. Since the real-time water accumulation depth on which the present application is based is identified according to the water accumulation image, it is real and real-time, thus improving the prediction accuracy of the waterlogging information. At the same time, the present application can predict the predicted water depth of any position in the target area at any time in the predicted period according to the predicted water depth time sequence of each water accumulation position in the predicted period, and can predict the predicted water depth of any position in the target area according to the predicted water depth of the plurality of water accumulation positions in the target area, that is, the present application can not only predict the water accumulation condition in a continuous period according to the water accumulation condition in a plurality of discrete time points, but also predict the water accumulation condition of the target area as a whole according to the water accumulation condition of a plurality of discrete positions, thus expanding the prediction space and the prediction time, and being able to more comprehensively predict the urban waterlogging information. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 A flowchart of a city waterlogging prediction method based on visual recognition provided for an embodiment of the present application;

[0037] Figure 2 A determination diagram of real-time water accumulation depth provided for an embodiment of the present application;

[0038] Figure 3 A technical roadmap of the prediction method provided for an embodiment of the present application. Detailed Implementation

[0039] The present invention will now be described in detail with reference to the embodiments, but the present invention is not limited to these embodiments.

[0040] This invention provides a method for predicting urban flooding based on visual recognition, such as... Figure 1 As shown, the method includes:

[0041] S1. Determine the real-time water depth at each water accumulation location based on water accumulation images of multiple water accumulation locations within the target area;

[0042] S2. Based on the meteorological data, historical water depth and real-time water depth of the target area, determine the predicted water depth sequence for each water accumulation location during the prediction period.

[0043] S3. Based on the predicted water depth time series of all water accumulation locations during the prediction period, determine the predicted waterlogging information of the target area during the prediction period.

[0044] Before S1, the method may also include:

[0045] Based on image information from multiple monitoring points within the target area, a large language model is used to filter out multiple locations of water accumulation from these monitoring points.

[0046] Specifically, this embodiment sets up multiple monitoring points within the target area. At each monitoring point, a fixed camera performs static monitoring, while a moving camera performs dynamic patrols, thereby obtaining video data for each monitoring point. The video data is then input into a large language model for visual recognition, which can filter out multiple monitoring points where water accumulation has occurred. The location of these water accumulation monitoring points is the location of the water accumulation. The large language model used in this embodiment is the existing open-source Large Language Model MetaAI 3 (LLaMA 3 for short).

[0047] Specifically, after identifying multiple locations of water accumulation, this embodiment performs data preprocessing on the video data from these locations. The data preprocessing includes:

[0048] 1. Data annotation: Keyframes of video data are extracted to obtain images containing flooding information. The images are then annotated in detail, including the boundaries of the flooded area, water depth, and severity level, thus obtaining annotated data.

[0049] 2. Data Augmentation: In order to obtain high-quality labeled data, methods such as image rotation, scaling, translation, and color transformation can be used to augment the labeled data, thereby improving the generalization ability and robustness of subsequent semantic segmentation models.

[0050] The waterlogging image of the waterlogging position is obtained by data preprocessing the video data of the waterlogging position.

[0051] S1 specifically can be:

[0052] S11, determining a real-time submergence level of each waterlogging position according to the waterlogging images of the plurality of waterlogging positions in the target area.

[0053] S12, determining a real-time waterlogging depth of each waterlogging position according to the real-time submergence level.

[0054] In this embodiment, S11 specifically can be:

[0055] S111, identifying a preset reference object from the waterlogging image of each waterlogging position in the target area, and determining real-time boundary information of the preset reference object in the waterlogging image.

[0056] Specifically, the preset reference object can be a common and easily estimated height reference object such as a human body or a vehicle in waterlogging. The boundary information of the preset reference object in the waterlogging image refers to the outline of the part not submerged by waterlogging. According to the real-time boundary information of the preset reference object, the submergence degree of the current waterlogging can be determined.

[0057] Further, as shown in Figure 2 S111 specifically can be:

[0058] The semantic segmentation model is used to identify the preset reference object from the waterlogging image of each waterlogging position in the target area, and determine the real-time boundary information of the preset reference object in the waterlogging image.

[0059] Semantic segmentation is an important task in the field of computer vision, aiming to make fine semantic division for each pixel in the image, thereby generating a more fine-grained understanding, which is usually used for scene segmentation and object contour recognition. Semantic segmentation models include the first model Fully Convolutional Networks (FCN) that realizes end-to-end semantic segmentation, U-Net with excellent performance in medical image processing through encoder-decoder architecture and skip connection, DeepLab series that expands the receptive field and improves the resolution through hollow convolution, and MaskR-CNN that can simultaneously realize target detection and instance segmentation and is widely used in tasks that require accurate contours. These semantic segmentation models have promoted the wide application of semantic segmentation technology in the fields of autonomous driving, remote sensing analysis and image editing.

[0060] The semantic segmentation model can assign semantic categories to each waterlogging image at a pixel level to extract rich semantic information from the image, achieve fine-grained understanding and analysis of the image, and accurately identify the position and boundary information of the preset reference object in the waterlogging image and perform fine analysis on the boundary information.

[0061] S112, determining a real-time flooding level of each waterlogging position according to the real-time boundary information.

[0062] Further, S112 can be specifically:

[0063] constructing a first correspondence relationship between the flooding level and the boundary information of the preset reference object in the waterlogging image;

[0064] determining a real-time flooding level of each waterlogging position according to the first correspondence relationship and the real-time boundary information.

[0065] Specifically, according to the boundary information of the preset reference object under different flooding levels, the flooding level can be divided into multiple flooding levels, so that the boundary information under different flooding levels has a corresponding flooding level, forming the first correspondence relationship.

[0066] In the embodiment, S12 can be specifically:

[0067] constructing a second correspondence relationship between the flooding level and the waterlogging depth;

[0068] determining a real-time waterlogging depth of each waterlogging position according to the second correspondence relationship and the real-time flooding level.

[0069] In practice, the second correspondence relationship between the flooding level and the waterlogging depth can be flexibly determined according to actual conditions. For example, the second correspondence relationship can be as shown in Figure 2 , taking a truck as the preset reference object, when the flooding level is 0 level, the waterlogging depth is about 0 cm, when the flooding level is 1 level, the waterlogging depth is about 80 cm, when the flooding level is 2 level, the waterlogging depth is about 200 cm; when the flooding level is 3 level, the waterlogging depth is about 300 cm; taking a human body as the preset reference object, when the flooding level is 0 level, the waterlogging depth is about 0 cm, when the flooding level is 1 level, the waterlogging depth is about 40 cm, when the flooding level is 2 level, the waterlogging depth is about 80 cm; when the flooding level is 3 level, the waterlogging depth is about 120 cm, and when the flooding level is 4 level, the waterlogging depth is about 150 cm.

[0070] Specifically, the meteorological data can include historical meteorological data and short-impending meteorological data. The historical meteorological data can include historical rainfall, historical cumulative rainfall, etc.; the short-impending meteorological data can include short-impending rainfall, short-impending cumulative rainfall, etc.

[0071] S2 specifically can be:

[0072] S21, according to the historical meteorological data and historical water depth of the target area, a water depth time series prediction model is constructed by using a spatio-temporal graph neural network;

[0073] S22, according to the short-range meteorological data of the target area and the real-time water depth of each water accumulation position, the water depth time series of each water accumulation position in the prediction period is determined by using the water depth time series prediction model.

[0074] Specifically, the historical water depth can be obtained through the monitoring data of the water accumulation monitoring station.

[0075] Since the water depth of the water accumulation position is a typical time series data, that is, the rainfall conditions and water depth of the previous time have a great influence on the water depth of the later time, the water depth prediction problem of the water accumulation position can be converted into a supervised learning problem. For example, the rainfall, cumulative rainfall and water depth of the water accumulation position in the previous hours can be used to predict the water depth of the future one hour.

[0076] Specifically, the spatio-temporal graph neural network is trained by using the historical meteorological data and historical water depth of the target area to construct a water depth time series prediction model; then the short-range meteorological data of the target area and the real-time water depth of each water accumulation position are used to actually predict the predicted water depth of each prediction time in the prediction period, and the water depth time series of each water accumulation position in the prediction period is obtained.

[0077] Specifically, before training or prediction, the rainfall, water depth and time label data are extracted in combination with the meteorological data, and then the extracted data is cleaned and fused, and then the multivariate time series data is generated by data normalization, reconstruction into supervised data and other methods. The multivariate time series data is used as the input of the subsequent water depth time series prediction model training or prediction.

[0078] Specifically, the spatio-temporal graph neural network used in the embodiment is a multivariate time graph neural network (Multivariate Time Graph Neural Networks, MTGNN for short). MTGNN is a graph neural network model specially designed for multivariate time series prediction. MTGNN can automatically learn the potential correlation between variables by constructing a dynamic adaptive graph structure. It combines graph convolution and time series convolution, and fully utilizes the spatial information of graph structure and the dynamic characteristics of time series, so as to improve the prediction accuracy.

[0079] MTGNN mainly consists of an adaptive graph learning module, a time series convolution network module and a graph convolution neural network module:

[0080] 1. Adaptive graph learning module

[0081] The goal of this module is to automatically learn the spatial dependency between nodes.

[0082] Suppose the input multivariate time series data is where N is the number of nodes, T is the time step, and F is the feature dimension.

[0083] The adaptive graph learning module constructs the graph adjacency matrix by the following formula:

[0084] A ij = softmax(f(x i , x j )) (1)

[0085] In formula (1), A ij represents the association weight between node i and node j, x i represents the feature of node i, x j represents the feature of node j, and f(x i , x j ) represents a learnable similarity function based on x i and x j .

[0086] 2. Time series convolutional network module

[0087] This module is responsible for modeling in the time dimension, using dilated convolution to capture long-term dependencies, with the specific formula being:

[0088] H (l+1) = ReLU(W*H (l) +b (l) ) (2)

[0089] In formula (2), H (l+1) represents the hidden state of the l+1 layer, H (l) represents the hidden state of the l layer, W represents the convolution kernel, * represents the convolution operation, and b (l) represents the bias term of the l+1 layer.

[0090] 3. Graph convolutional neural network module

[0091] This module performs graph convolution operations on the adaptive graph structure, with the specific formula being:

[0092] H (l+1) = σ(AH (l) W (l) +b (l) ) (3)

[0093] In formula (3), H (l+1)denotes the hidden state of the l+1-th layer, A is the learned graph adjacency matrix, H (l) denotes the hidden state of the l-th layer, W (l) denotes the learned weight matrix, b (l) denotes the bias term of the l+1-th layer, and σ denotes the activation function.

[0094] The MTGNN is trained using an error-based loss function, and the formula of the loss function is as follows:

[0095]

[0096] In formula (4), L denotes the loss function, denotes the predicted value of the model, Y t denotes the true label, and T is the time step.

[0097] Through back propagation and a stochastic gradient descent algorithm (such as the Adam optimizer), the parameters of the model are constantly updated during the training process, and finally accurate time series prediction is achieved.

[0098] In the water depth time series prediction task of the present embodiment, the MTGNN can regard the waterlogging depths of different waterlogging monitoring stations as nodes in a graph, and construct a dynamic graph structure by implicitly learning the correlation between the waterlogging monitoring stations. At the same time, the changes in the time dimension are modeled by a time convolution network module. This joint modeling method can effectively capture the spatiotemporal characteristics of water depth changes, thereby achieving better water depth time series prediction results.

[0099] To further improve the prediction accuracy, the present embodiment introduces a correction unit in the water depth time series prediction model. The correction unit can correct the predicted water depth of the current prediction according to the real-time waterlogging depth, and use the corrected predicted water depth as the input of the next prediction, so as to correct the accuracy of the water depth time series prediction model in predicting the water depth of multiple future time steps, thereby enhancing the reliability of the prediction.

[0100] Specifically, the water depth time series prediction model includes a prediction unit and a correction unit. The present embodiment uses a spatiotemporal graph neural network to construct the prediction unit, and uses a time convolution graph neural network to construct the correction unit.

[0101] After introducing the correction unit, S22 can be specifically as follows:

[0102] According to the short-weather data of the target area and the real-time waterlogging depth of each waterlogging position, the prediction unit is used to determine the predicted water depth of each waterlogging position at the current prediction time in the prediction period;

[0103] The prediction water depth at the current prediction time is corrected by the correction unit, and the corrected prediction water depth is taken as the input of the prediction unit to perform water depth prediction at the next prediction time until the prediction water depths at all prediction times in the prediction period are determined, thereby obtaining the prediction water depth sequence of each waterlogging position in the prediction period.

[0104] Specifically, the prediction period includes a plurality of prediction times. The prediction unit can predict the water depth at the future prediction time based on the current real-time time, and the correction unit can receive the real-time waterlogging depth at the real-time time determined by S1 in real time.

[0105] For example, if the real-time time is m and the water depth after n times is predicted, the prediction time is (m+n), and if the real-time time is (m+1), the prediction time is (m+n+1). Assuming that the prediction time (m+n) is the first prediction time of the prediction period, the prediction and correction process of the embodiment is as follows:

[0106] First, the water depth at the prediction time (m+n) is predicted based on the real-time time m. At this time, the prediction unit predicts the water depth at the first prediction time (m+n) based on the short-term weather data and the real-time waterlogging depth at the real-time time m, thereby obtaining the prediction water depth at the first prediction time (m+n).

[0107] Then, the water depth at the prediction time (m+n+1) is predicted based on the real-time time (m+1). At this time, the correction unit compares the prediction water depth at the prediction time (m+n) with the real-time waterlogging depth at the real-time time (m+1), and corrects the prediction water depth at the prediction time (m+n) according to the comparison result, thereby obtaining the corrected prediction water depth; the corrected prediction water depth is taken as the input of the prediction unit, and the prediction unit performs this water depth prediction based on the corrected prediction water depth and the short-term weather data at the real-time time (m+1), thereby obtaining the prediction water depth at the prediction time (m+n+1).

[0108] Subsequently, the water depth at the prediction time (m+n+2) is predicted based on the real-time time (m+2). At this time, the correction unit compares the prediction water depth at the prediction time (m+n+1) with the real-time waterlogging depth at the real-time time (m+2), and corrects the prediction water depth at the prediction time (m+n+1) according to the comparison result, thereby obtaining the corrected prediction water depth; the corrected prediction water depth is taken as the input of the prediction unit, and the prediction unit performs this water depth prediction based on the corrected prediction water depth and the short-term weather data at the real-time time (m+2), thereby obtaining the prediction water depth at the prediction time (m+n+2).

[0109] The subsequent prediction and correction process is as described above, until the prediction water depths at all prediction times in the prediction period are obtained, thereby obtaining the prediction water depth sequence of the prediction period.

[0110] By introducing the correction module, the embodiment realizes dynamic feedback and accurate correction of the prediction result, which is beneficial to improve the accuracy and reliability of the prediction.

[0111] After obtaining the predicted water depth time sequence of each water accumulation position in the prediction period, the water depth data of each position in the target region can be completed, and the water depth data between adjacent time points in the prediction period can be completed, so as to realize comprehensive prediction of the waterlogging information in the target region.

[0112] S3 can be specifically:

[0113] S31, according to the predicted water depth time sequence of all water accumulation positions, determining the predicted water depth of any position in the target region corresponding to any time point in the prediction period;

[0114] S32, according to the predicted water depth of each position in the target region corresponding to each time point in the prediction period, determining the predicted waterlogging information of the target region in the prediction period.

[0115] Specifically, for a certain prediction time, the embodiment can complete the predicted water depth of any other position in the target region according to the predicted water depth of the above-mentioned dispersed multiple dispersed water accumulation positions; for a certain water accumulation position, the embodiment can complete the predicted water depth of any other time point in the prediction period according to the predicted water depth of multiple dispersed prediction time points in the prediction period. For example, if the predicted water depth time sequence of the water accumulation position only includes the predicted water depth of 3 o'clock, 4 o'clock, 5 o'clock and the like, the embodiment can determine the water depth data of any time point between 3:01, 3:05, 3:10, 4:45, 4:56 and the like according to the predicted water depth of these prediction time points. Similarly, if the predicted water depth time sequence of the water accumulation position only includes the predicted water depth of 6 minutes, 7 minutes, 8 minutes and the like, the embodiment can determine the water depth data of any time point between 5 minutes and 5 seconds, 7 minutes and 20 seconds, 8 minutes and 50 seconds and the like according to the predicted water depth of these prediction time points.

[0116] Further, S31 can be specifically:

[0117] According to the predicted water depth time sequence of all water accumulation positions, the predicted water depth of any position in the target region corresponding to any time point in the prediction period is determined by using an interpolation algorithm.

[0118] Specifically, this embodiment uses inverse distance weighting (IDW) and kriging to determine the predicted water depth at any location in the target area; and uses autoregressive integrated moving average (ARIMA) to determine the predicted water depth at any time within the prediction period.

[0119] In practice, each of the three interpolation methods has its limitations. For example, IDW and Kriging only use spatial information and not temporal information, while ARIMA mainly uses temporal information and ignores spatial distribution information. Therefore, this embodiment needs to use an ensemble learning method to integrate the three interpolation methods. This can both compensate for the shortcomings of each method and combine their advantages to achieve more accurate water depth data completion, thereby realizing accurate prediction of flooding information in the target area.

[0120] Based on the above method, the technical roadmap of the prediction method of the present invention is as follows: Figure 3 As shown, this invention identifies the real-time water depth at each waterlogged location based on images of multiple waterlogged locations. Then, it combines meteorological data and historical water depth data for the target area to determine the predicted water depth sequence for each location within a given time period. Finally, it predicts the overall urban flooding information for the entire target area during the predicted time period based on this predicted water depth sequence. Since the real-time water depth prediction is based on waterlogged image identification, it possesses authenticity and real-time accuracy, thus improving the accuracy of urban flooding prediction. Furthermore, this invention can predict the predicted water depth at any point within the predicted time period based on the predicted water depth sequence for each waterlogged location, and it can predict the predicted water depth at any location within the target area based on the predicted water depths of multiple waterlogged locations within the target area. In other words, this invention can predict the waterlogging situation over a continuous period based on the waterlogging situation at multiple discrete moments, and it can also predict the overall waterlogging situation of the target area based on the waterlogging situation at multiple dispersed locations, expanding the prediction space and time, and enabling a more comprehensive prediction of urban flooding information.

[0121] The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A method for predicting urban flooding based on visual recognition, characterized in that, The method includes: S1. Determine the real-time water depth at each water accumulation location based on water accumulation images of multiple water accumulation locations within the target area; S2. Based on the historical meteorological data and historical water depth of the target area, a water depth time series prediction model is constructed using a spatiotemporal graph neural network. The water depth time series prediction model includes a prediction unit and a correction unit. Based on the short-term meteorological data of the target area and the real-time water depth, the prediction unit determines the predicted water depth of each water accumulation location at the current prediction time within the prediction period. The correction unit corrects the predicted water depth at the current prediction time, and the corrected predicted water depth is used as the input of the prediction unit to predict the water depth at the next prediction time, until the predicted water depth at all prediction times within the prediction period is determined, thus obtaining the predicted water depth time series for each water accumulation location within the prediction period. S3. Based on the predicted water depth time series of all water accumulation locations, use an interpolation algorithm to determine the predicted water depth of any location in the target area corresponding to any moment within the predicted time period; based on the predicted water depth of each location in the target area corresponding to each moment within the predicted time period, determine the predicted flooding information of the target area during the predicted time period.

2. The method according to claim 1, characterized in that, Specifically, S1 is: S11. Based on the water accumulation images of multiple water accumulation locations within the target area, determine the real-time flooding level of each water accumulation location; S12. Determine the real-time water depth at each water accumulation location based on the real-time flood level.

3. The method according to claim 2, characterized in that, Specifically, S11 is: S111. Identify a preset reference object from the water accumulation image of each water accumulation location within the target area, and determine the real-time boundary information of the preset reference object in the water accumulation image; S112. Determine the real-time flooding level of each water accumulation location based on the real-time boundary information.

4. The method according to claim 3, characterized in that, Specifically, S112 is as follows: Construct a first correspondence between the flooding level and the boundary information of the preset reference object in the water accumulation image; Based on the first correspondence and the real-time boundary information, the real-time flooding level of each water accumulation location is determined.

5. The method according to claim 2, characterized in that, Specifically, S12 is: Establish a second correspondence between inundation levels and water depth; Based on the second correspondence and the real-time flood level, the real-time water depth at each water accumulation location is determined.

6. The method according to claim 1, characterized in that, Prior to S1, the method further includes: Based on image information from multiple monitoring points within the target area, a large language model is used to filter out multiple locations of water accumulation from these monitoring points.

7. The method according to claim 3, characterized in that, Specifically, S111 is: A semantic segmentation model is used to identify preset reference objects from the water accumulation images of each water accumulation location within the target area, and to determine the real-time boundary information of the preset reference objects in the water accumulation images.

Images