Water conservancy disaster early warning method, system and equipment based on AI and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and particularly provides an AI-based water conservancy disaster early warning method, system and device and a storage medium, and the method comprises the steps: collecting a multi-modal data sequence through a multi-modal sensor; converting the multi-modal data sequence into a space-time cube; inputting the space-time cube into a prediction model to obtain a disaster prediction result; generating and broadcasting early warning information based on the disaster prediction result; the prediction model comprises a multi-modal feature extraction network and a cross-modal feature fusion layer. According to the invention, by fusing the multi-modal data, the early warning timeliness and early warning precision of water conservancy disasters are improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to an AI-based water disaster early warning method, system, equipment and storage medium. Background Art

[0002] Current water disaster early warning systems generally suffer from a single monitoring dimension. Traditional methods primarily rely on precipitation monitoring data to build prediction models. This single-factor driven mechanism has significant limitations: First, while precipitation is the direct trigger for flooding, the disaster formation process is essentially a nonlinear system involving the coupling of multiple factors. Focusing solely on precipitation variables overlooks key parameters such as surface permeability, soil water saturation, and river channel capacity. Second, static monitoring data struggles to reflect the dynamic evolution of geological structures, particularly the impact of human interference factors such as increased surface hardening and natural water system diversions during urbanization. Third, existing models lack the ability to collaboratively analyze multi-source spatiotemporal data, making them unable to effectively capture the interrelated effects of potential disaster-causing factors such as abnormal groundwater level fluctuations and mountain displacement and deformation. This single-modal modeling approach inevitably leads to delayed risk identification and misjudgment of disaster mechanisms. Summary of the Invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides an AI-based water disaster early warning method, system, device and storage medium to solve the above-mentioned technical problems.

[0004] In a first aspect, the present invention provides an AI-based water disaster early warning method, comprising: Using multimodal sensors to collect multimodal data sequences; Convert multimodal data sequences into space-time cubes; Inputting the space-time cube into a prediction model to obtain a disaster prediction result; generating and broadcasting early warning information based on the disaster prediction results; The prediction model includes a multimodal feature extraction network and a cross-modal feature fusion layer.

[0005] In an optional embodiment, using a multimodal sensor to collect a multimodal data sequence includes: The rain clouds are scanned periodically using lightning at multiple detection points to obtain radar reflectivity data sequences; The microwave attenuation coefficient is regularly detected by using microwave base stations at multiple detection points to obtain a microwave attenuation data sequence; Soil moisture content is collected regularly using soil sensors at multiple detection points to obtain a soil moisture data series; The density of floating objects on the water surface is regularly identified using cameras at multiple detection points to obtain a water level data sequence.

[0006] In an optional embodiment, converting a multimodal data sequence into a space-time cube includes: The monitoring area was divided into 100m×100m grid cells, and a three-dimensional geographic spatial coordinate system was constructed by superimposing the digital elevation model; Construct 60 time slices in minutes, each containing radar reflectivity, microwave attenuation coefficient, soil moisture content, water level, and floating object density; Use Kriging interpolation algorithm to fill sensor blind spots; Using a four-dimensional tensor structure to store data, the four-dimensional tensor structure includes longitude×latitude×elevation×time; The earliest time slice is deleted every minute and the newly collected data is inserted.

[0007] In an optional embodiment, the multimodal feature extraction network includes: A visual Transformer module to predict the cloud displacement vector field based on radar reflectivity space-time cube slices; LSTM precipitation prediction module, used to predict rainfall intensity based on the microwave base station signal attenuation coefficient and pre-learned historical precipitation records; The geological risk analysis module is used to predict the landslide risk index based on soil moisture content, latitude and longitude, and elevation.

[0008] In an optional embodiment, the visual Transformer module includes: The spatial branch uses the Deformable DETR architecture to dynamically focus on the cloud mutation area; The time branch adopts the TimeSformer structure and uses the decomposed spatiotemporal attention mechanism.

[0009] In an optional embodiment, the cross-modal feature fusion layer includes: The attention fusion layer is used to select intra-modal features using SENet channel attention, align the selected features across modalities using the Optimal Transport theory, and process the aligned cross-modal features into fused features using a differentiable threshold gate. A causal reasoning module is used to infer the disaster level based on the fusion features using a causal chain.

[0010] In an optional embodiment, generating and broadcasting warning information based on the disaster prediction result includes: Analyze the disaster level and location information in the disaster prediction results; If it is confirmed that the disaster level has reached the set warning level, a warning message is generated; The base station at the corresponding location is called according to the location information, and the base station is used to broadcast the warning information.

[0011] In a second aspect, the present invention provides an AI-based water disaster early warning system, comprising: An acquisition module, configured to acquire a multimodal data sequence using a multimodal sensor; A preprocessing module for converting multimodal data sequences into space-time cubes; A prediction module, configured to input the space-time cube into a prediction model to obtain a disaster prediction result; An early warning module, configured to generate and broadcast early warning information based on the disaster prediction results; The prediction model includes a multimodal feature extraction network and a cross-modal feature fusion layer.

[0012] According to a third aspect, a device is provided, comprising: Memory, used to store AI-based water disaster warning programs; A processor is used to implement the steps of the AI-based water disaster early warning method provided in the first aspect when executing the AI-based water disaster early warning program.

[0013] In a fourth aspect, a computer-readable storage medium is provided, on which an AI-based water disaster warning program is stored. When the AI-based water disaster warning program is executed by a processor, the steps of the AI-based water disaster warning method provided in the first aspect are implemented.

[0014] The beneficial effects of the present invention are that the AI-based water disaster early warning method, system, equipment and storage medium provided by the present invention systematically collect heterogeneous spatiotemporal sequence data such as rainfall intensity, runoff velocity, soil pore water pressure, and surface displacement vector by deploying multi-source monitoring equipment such as hydrological sensor arrays, synthetic aperture radars, and ground-based interferometric radars. The spatiotemporal cube modeling technology breaks through the limitations of the traditional two-dimensional data plane and realizes the dynamic coupling characterization of geological-hydrological-meteorological parameters in a four-dimensional coordinate system by spatiotemporal encoding of multi-dimensional monitoring data. The prediction model based on deep learning analyzes the spatiotemporal evolution laws of each sensor data through a multimodal feature extraction network, and then mines the potential parameter correlation map through a cross-modal feature fusion layer, and finally establishes a disaster evolution model under the coupling of multiple physical fields. This method improves the timeliness and accuracy of landslide early warning by fusing multimodal data.

[0015] In addition, the present invention has a reliable design principle, a simple structure and a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention.

[0018] Figure 2 FIG. 4 is a schematic block diagram of a system according to an embodiment of the present invention.

[0019] Figure 3 A schematic structural diagram of a device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0022] The AI-based water disaster early warning method provided in the embodiment of the present invention is executed by a computer device, and accordingly, the AI-based water disaster early warning system runs in the computer device.

[0023] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention. Figure 1 The execution entity can be an AI-based water disaster early warning system. Depending on different needs, the order of the steps in the flowchart can be changed, and some steps can be omitted.

[0024] like Figure 1 As shown, the method includes: S1. Collect multimodal data sequences using multimodal sensors; S2. Convert multimodal data sequences into space-time cubes; S3. Input the space-time cube into the prediction model to obtain disaster prediction results; S4. Generate and broadcast warning information based on the disaster prediction results; The prediction model includes a multimodal feature extraction network and a cross-modal feature fusion layer.

[0025] In an embodiment of the present invention, based on step S1, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0026] S101. Periodically scan rain clouds using lightning at multiple detection points to obtain a radar reflectivity data sequence.

[0027] Equipment Selection: Use X-band or C-band Doppler weather radar as lightning scanning equipment. X-band radar has a shorter wavelength (2.4-3.9 cm), offers high resolution for small and medium-scale weather systems (such as thunderstorm clouds), and can accurately capture the internal structure of rain clouds. C-band radar has a more moderate wavelength (3.9-7.5 cm), a longer detection range, and is relatively less affected by precipitation, making it suitable for monitoring rain clouds over large areas. Each detection point should be deployed with a radar antenna capable of 360° omnidirectional scanning, with a scanning elevation range of 0°-90° to cover rain clouds at varying altitudes.

[0028] Data collection frequency: The scanning cycle is set according to the intensity of rain cloud activity. Under normal weather conditions, a full airspace scan is completed every 10-15 minutes. When severe convective weather such as thunderstorms is monitored, the scanning cycle is shortened to 3-5 minutes to ensure timely acquisition of rain cloud dynamic change data.

[0029] Data Processing: The radar echo signals received undergo denoising (using algorithms such as wavelet transforms to remove electromagnetic interference and noise). The radar reflectivity factor (Z) is then calculated using the radar equation. This factor is related to the size, number, and distribution of precipitation particles. The radar reflectivity factor data obtained from each scan is arranged in chronological order to form a radar reflectivity data sequence. Furthermore, a geographic information system (GIS) is used to spatially locate the radar reflectivity data, visually displaying the spatial distribution and movement of rain clouds.

[0030] S102. Utilize microwave base stations at multiple detection points to regularly detect microwave attenuation coefficients to obtain a microwave attenuation data sequence.

[0031] Equipment Configuration: Each monitoring point is equipped with a microwave communication device, including a transmitter and receiver, operating in the 20-40 GHz frequency band, which is sensitive to microwave attenuation caused by precipitation. The transmitter transmits a continuous wave signal with stable power, while the receiver is equipped with a high-precision power sensor that accurately measures the received signal power with a power measurement accuracy of ±0.1 dB.

[0032] Data collection frequency: Set the microwave signal to be transmitted and received every 5 minutes, and record the transmission power P t and received power P r By formula Calculate the microwave attenuation coefficient (unit: dB). This formula is based on the principle of free space propagation loss and takes into account the absorption and scattering of microwave signals by precipitation particles.

[0033] Data Calibration and Verification: Microwave equipment is calibrated regularly (once a month) in dry weather to ensure measurement accuracy. Microwave attenuation data from multiple measurement points is compared with precipitation data from nearby weather stations to verify data reliability. If any data anomalies are detected, equipment failure or environmental interference should be promptly investigated.

[0034] S103. Use soil sensors at multiple detection points to regularly collect soil moisture content to obtain a soil moisture data sequence.

[0035] Sensor selection: Soil moisture sensors using time domain reflectometry (TDR) or frequency domain reflectometry (FDR) offer high measurement accuracy (within ±3%) and fast response speed. Sensors should be buried at different depths (e.g., 10 cm, 20 cm, and 30 cm) at each monitoring point to obtain soil moisture information. The sensors must be waterproof and corrosion-resistant to withstand complex field environments.

[0036] Data Collection and Transmission: The sensor collects soil moisture data every 30 minutes. The sensor converts the collected soil dielectric constant into moisture content (using a built-in calibration curve) and sends the data to the data center via a wireless transmission module (such as LoRa or NB-IoT). To ensure stable data transmission, a data retransmission mechanism is configured. If data transmission fails, the sensor automatically retransmits the data three times.

[0037] Data quality control: Collected soil moisture data is tested for outliers, and clearly unreasonable data is eliminated using the 3σ principle (triple standard deviation). Simultaneously, the soil moisture data is analyzed for rationality in conjunction with meteorological data (such as precipitation and evaporation). If data anomalies are detected, on-site verification and equipment maintenance are promptly performed.

[0038] S104. Use cameras at multiple detection points to regularly identify the density of floating objects on the water surface and obtain a water level data sequence.

[0039] Camera Deployment: Install high-definition network cameras (resolution no less than 1920×1080) at each inspection point with a wide field of view and complete coverage of the water surface. The cameras must have infrared night vision capabilities to ensure 24-hour uninterrupted monitoring. The camera installation height should be determined based on the actual water surface conditions to ensure clear images of floating objects and water level fluctuations.

[0040] Image Acquisition and Processing: A water surface image is captured every 15 minutes. Image recognition technology (deep learning-based object detection algorithms, such as YOLO and Faster R-CNN) is used to identify and count floating objects in the image. The density of floating objects is calculated by calculating the number of floating objects per unit area. Image processing algorithms (such as edge detection and morphological analysis) are also used to extract the water surface edge. The water level is calculated based on parameters such as the camera's mounting height and angle. The floating object density and water level data obtained from each calculation are arranged in chronological order to form a corresponding data sequence.

[0041] Algorithm Optimization and Calibration: The image recognition algorithm is regularly optimized and calibrated (once every quarter). Water surface image samples under different weather conditions and light intensities are collected and updated with the training dataset to improve the algorithm's accuracy in identifying floating objects and measuring water levels. Furthermore, by comparing the algorithm's calculations with manually measured water level data, errors are corrected to ensure the reliability of the water level data.

[0042] In an embodiment of the present invention, based on step S2, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0043] S201. Divide the monitoring area into 100m×100m grid cells and overlay the digital elevation model to construct a three-dimensional geographic spatial coordinate system.

[0044] Before processing data, imagine the entire monitoring area as a large map. To more accurately locate and analyze data, divide this map into small squares, each 100 meters on a side. Each square has a fixed position, like a square on a chessboard. The advantage of this approach is that subsequent data collected from any detection point can be quickly mapped to a specific square.

[0045] After the grid is created, a digital elevation model (DEM) is introduced. The DEM assigns elevation information to each small square, indicating whether the area is a mountain, a plain, or a depression. By combining the grid and the DEM, a three-dimensional geographic coordinate system is established. This coordinate system accurately locates each data collection point in the spatial dimension, providing a spatial foundation for subsequent analysis.

[0046] S202. Construct 60 layers of time slices in units of minutes, each layer containing radar reflectivity, microwave attenuation coefficient, soil moisture content, water level, and floating object density.

[0047] Data changes over time. To better capture these changes, the time dimension is introduced. Using each minute as a time interval is like slicing time into thin slices, totaling 60 slices. Each "time slice" contains data from various monitoring items at the same moment. For example, radar reflectivity data reflects the presence of rain clouds, the microwave attenuation coefficient reflects the atmospheric influence on microwave signals, soil moisture content indicates soil moisture, and water level and floating debris density are related to the water body.

[0048] In this way, different types of data are organized in chronological order to form a three-dimensional data structure that contains both spatial information (each grid unit) and temporal information (each layer of time slice), making it easy to observe the changing trends of data over time and the relationship between data at different locations.

[0049] S203. Use the Kriging interpolation algorithm to fill the sensor blind area.

[0050] In actual monitoring, sensors cannot be distributed across the entire monitoring area. There will always be blind spots where no sensors are present, and no data is directly collected from these areas. This is where the Kriging interpolation algorithm comes in handy. It observes the data from areas with sensors surrounding these blind spots and, based on the locations and values ​​of these known data points, as well as their spatial correlation, infers the likely data values ​​for these blind spots.

[0051] For example, if the soil moisture content at several points around a blind spot is relatively high, the algorithm will reasonably infer the approximate soil moisture content in the blind spot based on this spatial relationship and data patterns, thereby filling the data gap and making the entire data more complete and continuous, which is convenient for subsequent analysis and application.

[0052] S204. Use a four-dimensional tensor structure to store data, where the four-dimensional tensor structure includes longitude×latitude×elevation×time.

[0053] To efficiently manage and use this complex data, a four-dimensional tensor structure is used for storage. The previously defined grid cells correspond to longitude and latitude information, the digital elevation model provides elevation information, and the time slice represents time information. These four dimensions form a four-dimensional tensor.

[0054] Simply put, it's like a giant Rubik's Cube, with each small grid corresponding to a specific spatial location (longitude, latitude, and elevation) and time point, and each grid stores the corresponding monitoring data. This structure allows for rapid location, query, and analysis of monitoring data at any time and location, greatly improving the efficiency and flexibility of data processing.

[0055] S205. Delete the earliest time slice every minute and insert new collected data.

[0056] As time passes, new data is constantly generated. To ensure that stored data is always up-to-date and valuable, a data update mechanism is implemented. Every minute, the oldest time slice data is deleted, as this data is relatively outdated. Then, the newly collected data for that minute is inserted into the latest time slice.

[0057] Just like in a constantly updated diary, new information is recorded every day, and the oldest records are gradually replaced, so that the diary always records the most recent situation. In this way, the stored data can reflect the latest status of the monitored area in real time, providing timely and accurate data support for subsequent decision-making and analysis.

[0058] In an embodiment of the present invention, based on step S3, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0059] The prediction model includes a multimodal feature extraction network and a cross-modal feature fusion layer.

[0060] Among them, the multimodal feature extraction network includes: The visual Transformer module is used to predict the cloud displacement vector field based on the space-time cube slices of radar reflectivity. The visual Transformer module includes: a spatial branch using the Deformable DETR architecture to dynamically focus on the cloud mutation area; a temporal branch using the TimeSformer structure and a decomposed spatiotemporal attention mechanism; serialization processing into 768-dimensional vectors; a multi-head attention mechanism (8 heads) to capture the movement patterns of clouds; and output: the cloud displacement vector field for the next 10 minutes.

[0061] LSTM precipitation prediction module, used to predict rainfall intensity based on the microwave base station signal attenuation coefficient and pre-learned historical precipitation records; Among them, the microwave signal is enhanced: Add base station topology graph convolution layer (GCN):

[0062] in is the adjacency matrix with base station location relationships.

[0063] The geological risk analysis module is used to predict the landslide risk index based on soil moisture content, latitude and longitude, and elevation.

[0064] DEM data processing flow: Calculate slope (Sobel operator) and runoff accumulation; Geological parameter fusion: Landslide risk index = 0.3 × slope + 0.2 × soil moisture + 0.5 × historical landslide data.

[0065] A federated learning framework is introduced to solve the problem of data sparsity: regional node model: LightGBM gradient boosting tree (for handling small samples); center aggregation strategy: dynamic weighted average (based on soil type similarity).

[0066] The cross-modal feature fusion layer includes: The attention fusion layer is used to select intra-modal features using SENet channel attention, align the selected features across modal spaces and time using the Optimal Transport theory, and process the aligned cross-modal features into fused features using a differentiable threshold gate.

[0067] The causal reasoning module is used to infer the disaster level based on the fusion features using the causal chain. A causal discovery network is added after the fusion layer, and the NOTEARS algorithm is used to construct a causal graph:

[0068] stW is the acyclic graph adjacency matrix.

[0069] Modeling the "rainfall → soil moisture → landslide risk" causal chain: Causal discovery phase: 1. Multi-scale time series data preparation The resolution of radar rainfall is 500m / 5min. The data source is meteorological radar reflectivity conversion, and the feature processing logic is ZR relationship calibration → mm / h.

[0070] The resolution of soil volume moisture content is 1 km / 15 min. The data source is microwave remote sensing inversion + ground sensors, and the feature processing logic is Kriging spatial interpolation.

[0071] The resolution of the landslide risk index is 100m / 1h. The data source is InSAR deformation data + historical landslide database, and the feature processing logic is weighted fusion: R=0.4Δd+0.3S+0.3H.

[0072] Δd is the surface deformation rate, S is the slope, and H is the historical landslide frequency.

[0073] 2. Variational Causal Discovery Network, building a causal structure learning model:

[0074] in, To ensure the acyclic constraint, the ADMM optimization algorithm is used to solve the sparse adjacency matrix W.

[0075] 3. Time-delay causality test, applying the improved windowed PC algorithm: Sliding time window detection maximum information coefficient (MIC):

[0076] Lagged parameter estimates:

[0077] CCM is the convergent cross mapping algorithm.

[0078] Output: causal structure diagram and lag parameters.

[0079] Physical mechanism embedding: 1. Embedding of hydrological equations. Enforcing physical constraints in neural networks:

[0080] Use the PINN framework to construct differential equation constraints: # Loss function physics_loss = torch.mean( (dtheta_dt - (I_model - Q_model - E_model))**2 ).

[0081] 2. Critical state equation.

[0082] Introducing the deep learning version of the Mohr-Coulomb criterion:

[0083] 3. Hierarchical causal modeling. Constructing a three-layer causal graph neural network: Micro-layer: Lattice Boltzmann simulation of pore water pressure conduction; Mesoscopic layer: Finite element analysis of slope stability; Macro level: regional landslide risk index prediction.

[0084] Dynamic Structural Equation Model: 1. Formal definition of the model.

[0085] Where: T is temperature, G is geological conditions (rock and soil type, joint direction), and D is the interference factor of human activities.

[0086] 2. Parameter estimation method. Use two-stage least squares (2SLS): Instrumental variable selection: atmospheric temperature and humidity as IV of rainfall; Loss function:

[0087] The weight matrix W is estimated by HAC standard error 3. Sensitivity analysis. Compute the causal effect in the form of partial derivatives:

[0088] where ϕ is the path conduction coefficient.

[0089] Among them, the feature fusion process of the attention fusion layer is as follows: Intra-modal feature selection: For each modality :

[0090] Among them C m is the number of channels, T is the time step, and H×W is the spatial dimension.

[0091] Using improved SENet structure:

[0092] in, , , (r=16 is the compression ratio). δ is the ReLU activation and σ is the sigmoid function.

[0093] Perform dimension compression on the processed features:

[0094] Where d=256 is the unified feature dimension.

[0095] Cross-modal spatiotemporal alignment: Cost matrix construction, for any two modalities m i ,m j :

[0096] α=0.3,β=0.5,γ=0.5 are the spatiotemporal alignment weights.

[0097] Solve the transmission plan P using the Sinkhorn algorithm ∗ :

[0098] μ,ν is the uniform distribution of source / target distribution; K = exp(−C / ϵ) is the Gibbs kernel matrix.

[0099] Align features via the optimal transfer matrix:

[0100] Use GPU-accelerated Sinkhorn iteration (usually converges in 10-20 iterations); Compute gradient propagation using automatic differentiation.

[0101] Decision-level gate fusion: Gating signal generation:

[0102] 3D dilated convolution (dilation=2) is used to expand the receptive field, and the gate value g represents the credibility of each modality at a spatiotemporal point.

[0103] Dynamically adjust the gating threshold:

[0104] Thresholding:

[0105] Maintain gradient propagation through the Straight-Through Estimator.

[0106] Final fusion output:

[0107] in is the learnable projection matrix.

[0108] This solution is implemented through a three-level attention mechanism: Intra-modal purification: suppresses noisy channels and enhances important features; Cross-modal alignment: establishing a temporal and spatial consistency mapping relationship; Dynamic decision-making: Adaptively select reliable modes based on the scenario.

[0109] Meteorological data testing showed that compared to traditional splicing and fusion methods, this solution reduced the false alarm rate of extreme weather warnings while maintaining real-time processing capabilities. The key technological breakthrough lies in the organic integration of optimal transmission theory with a deep learning framework, which solves the problem of spatiotemporal misalignment among multiple sensors.

[0110] Device deployment: Edge computing box: uses NVIDIA Jetson Orin (palm-sized, 15 watts of power consumption, equivalent to the power consumption of a mobile phone charger).

[0111] Installation examples: Radars and cameras are installed on mountaintops to monitor rainfall upstream; soil sensors are buried in river valleys to monitor changes in soil quality; and early warning loudspeakers are deployed in villages to automatically broadcast evacuation orders upon receiving a signal.

[0112] In an embodiment of the present invention, based on step S4, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0113] Map mode: Red / yellow / blue warning areas are displayed in real time. Warning push: Evacuation route maps are automatically generated and distributed via SMS / App / broadcast.

[0114] S401. Analyze the disaster level and location information in the disaster prediction results.

[0115] Disaster prediction results are like a pre-written "disaster report," but this report is written in a specific format, and the computer needs to "understand" the key content before it can proceed. In this step, the system acts like a "translator," parsing this "disaster report."

[0116] It will retrieve disaster severity information from the report, such as whether it is a general disaster, a major disaster, a serious disaster, or an extremely serious disaster, like labeling disasters with different levels. It also needs to find the location information of the possible disaster, which can be a specific point or an area. If it is an area, the system will calculate the center location of the area and the location of the area's boundaries, just like circling the possible impact range of the disaster on a map, making it easier to determine which areas require special attention.

[0117] S402. If it is confirmed that the disaster level has reached the set warning level, a warning message is generated.

[0118] To avoid unnecessary panic, not all disasters require a warning, so the system sets warning level thresholds for different disaster types. In this step, the system compares the newly analyzed disaster level with the set warning level.

[0119] For example, for rainstorm disasters, warnings are only issued for levels 2 and above. If the analyzed rainstorm disaster level is level 3, it means that the warning criteria have been met, and the system will begin generating warning information. It combines key information such as disaster type, disaster level, affected location, and expected time of occurrence into a specific format, and adds pre-prepared preventive measures. It is like a written "warning notice" to everyone, informing them of the upcoming disaster and how to protect themselves.

[0120] S403. Call the base station at the corresponding location according to the location information, and use the base station to broadcast warning information.

[0121] Once the warning information is generated, the system needs to deliver this important message to people in the affected area.

[0122] At this point, the system will "search" for suitable base stations in its base station database based on the previously determined disaster-affected location. It will identify base stations with signal coverage that "encloses" the affected area, much like finding the signal range on a map that just covers the disaster area. If multiple base stations have coverage, the system will prioritize those with strong signals and low loads to ensure smooth transmission of the warning message.

[0123] Finally, the system sends the warning information to the selected base station through a specific communication interface. The base station acts like a broadcaster, using cell broadcast technology to send the warning information in the form of text messages to every mobile phone in the coverage area, so that everyone can receive the warning in time and be prepared to respond.

[0124] In some embodiments, the AI-based water disaster early warning system may include multiple functional modules composed of computer program segments. The computer program of each program segment in the AI-based water disaster early warning system may be stored in a memory of a computer device and executed by at least one processor to perform (see Figure 1 (Description) AI-based water disaster early warning function.

[0125] In this embodiment, the AI-based water disaster early warning system can be divided into multiple functional modules according to the functions it performs, such as Figure 2 As shown. The module referred to in the present invention refers to a series of computer program segments that can be executed by at least one processor and can perform fixed functions, which are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0126] An acquisition module, configured to acquire a multimodal data sequence using a multimodal sensor; A preprocessing module for converting multimodal data sequences into space-time cubes; A prediction module, configured to input the space-time cube into a prediction model to obtain a disaster prediction result; An early warning module, configured to generate and broadcast early warning information based on the disaster prediction results; The prediction model includes a multimodal feature extraction network and a cross-modal feature fusion layer.

[0127] Figure 3 The AI-based water disaster early warning method provided for the embodiment of the present application can be applied to a device. Those skilled in the art will understand that the device structure involved in the embodiment of the present invention does not constitute a limitation on the device, and the device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. In the embodiment of the present invention, the device includes but is not limited to a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the present application described and / or required herein.

[0128] The device 300 may include a processor 310, a memory 320, and a communication unit 330. These components communicate via one or more buses. Those skilled in the art will appreciate that the server structure shown in the figure does not limit the present invention. The server structure may be a bus structure or a star structure, and may include more or fewer components than shown, or combine certain components, or arrange the components differently.

[0129] The memory 320 can be used to store execution instructions of the processor 310. The memory 320 can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the device 300 can perform some or all of the steps in the above-described method embodiments.

[0130] The processor 310 is the control center of the storage device, which uses various interfaces and lines to connect various parts of the entire electronic device. It executes various functions of the electronic device and / or processes data by running or executing software programs and / or modules stored in the memory 320, and calling data stored in the memory. The processor can be composed of an integrated circuit (IC), for example, it can be composed of a single packaged IC, or it can be composed of multiple packaged ICs with the same or different functions. For example, the processor 310 can only include a central processing unit (CPU). In an embodiment of the present invention, the CPU can be a single computing core or multiple computing cores.

[0131] The communication unit 330 is configured to establish a communication channel so that the storage device can communicate with other devices, receive user data sent by other devices, or send user data to other devices.

[0132] The present invention also provides a computer storage medium, wherein the computer storage medium may store a program that, when executed, may include some or all of the steps of each embodiment provided herein. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0133] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software and a necessary general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, among other media capable of storing program code, and includes instructions for causing a computer device (which can be a personal computer, a server, or a second device, a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0134] In this specification, the same or similar parts between the various embodiments can be referred to each other. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

[0135] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or modules, and can be electrical, mechanical or other forms.

[0136] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of the present embodiment according to actual needs.

[0137] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0138] Although the present invention has been described in detail with reference to the accompanying drawings and in conjunction with preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, persons of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and such modifications or substitutions shall be within the scope of the present invention. Any changes or substitutions that can be easily conceived by persons skilled in the art within the technical scope disclosed in the present invention shall be within the scope of protection of the present invention.

Claims

1. An AI-based water disaster early warning method, characterized in that: include: Using multimodal sensors to collect multimodal data sequences; Convert multimodal data sequences into space-time cubes; Inputting the space-time cube into a prediction model to obtain a disaster prediction result; generating and broadcasting early warning information based on the disaster prediction results; The prediction model includes a multimodal feature extraction network and a cross-modal feature fusion layer.

2. The method according to claim 1, characterized in that Use multimodal sensors to collect multimodal data sequences, including: The rain clouds are scanned periodically using lightning at multiple detection points to obtain radar reflectivity data sequences; The microwave attenuation coefficient is regularly detected by using microwave base stations at multiple detection points to obtain a microwave attenuation data sequence; Soil moisture content is collected regularly using soil sensors at multiple detection points to obtain a soil moisture data series; The density of floating objects on the water surface is regularly identified using cameras at multiple detection points to obtain a water level data sequence.

3. The method according to claim 1, characterized in that Convert multimodal data sequences into space-time cubes, including: The monitoring area was divided into 100m×100m grid cells, and a three-dimensional geographic spatial coordinate system was constructed by superimposing the digital elevation model; Construct 60 time slices in minutes, each containing radar reflectivity, microwave attenuation coefficient, soil moisture content, water level, and floating object density; Use Kriging interpolation algorithm to fill sensor blind spots; Using a four-dimensional tensor structure to store data, the four-dimensional tensor structure includes longitude×latitude×elevation×time; The earliest time slice is deleted every minute and the newly collected data is inserted.

4. The method according to claim 1, wherein The multimodal feature extraction network includes: A visual Transformer module to predict the cloud displacement vector field based on radar reflectivity space-time cube slices; LSTM precipitation prediction module, used to predict rainfall intensity based on the microwave base station signal attenuation coefficient and pre-learned historical precipitation records; The geological risk analysis module is used to predict the landslide risk index based on soil moisture content, latitude and longitude, and elevation.

5. The method according to claim 4, characterized in that The visual Transformer module includes: The spatial branch uses the Deformable DETR architecture to dynamically focus on the cloud mutation area; The time branch adopts the TimeSformer structure and uses the decomposed spatiotemporal attention mechanism.

6. The method according to claim 5, characterized in that The cross-modal feature fusion layer includes: The attention fusion layer is used to select intra-modal features using SENet channel attention, align the selected features across modalities using the Optimal Transport theory, and process the aligned cross-modal features into fused features using a differentiable threshold gate. A causal reasoning module is used to infer the disaster level based on the fusion features using a causal chain.

7. The method according to claim 1, characterized in that Generating and broadcasting warning information based on the disaster prediction results, including: Analyze the disaster level and location information in the disaster prediction results; If it is confirmed that the disaster level has reached the set warning level, a warning message is generated; The base station at the corresponding location is called according to the location information, and the base station is used to broadcast the warning information.

8. An AI-based water disaster early warning system, characterized in that: include: An acquisition module, configured to acquire a multimodal data sequence using a multimodal sensor; A preprocessing module for converting multimodal data sequences into space-time cubes; A prediction module, configured to input the space-time cube into a prediction model to obtain a disaster prediction result; An early warning module, configured to generate and broadcast early warning information based on the disaster prediction results; The prediction model includes a multimodal feature extraction network and a cross-modal feature fusion layer.

9. A device, characterized in that include: Memory, used to store AI-based water disaster warning programs; A processor, configured to implement the steps of the AI-based water disaster early warning method according to any one of claims 1 to 7 when executing the AI-based water disaster early warning program.

10. A computer-readable storage medium storing a computer program, characterized in that: The readable storage medium stores an AI-based water disaster warning program, which, when executed by a processor, implements the steps of the AI-based water disaster warning method according to any one of claims 1 to 7.

Images