Hydropower station underground powerhouse flood prevention linkage early warning method, device and equipment based on multi-source data fusion and medium

Abstract

The application discloses a hydropower station underground powerhouse flood prevention linkage early warning method, device, equipment and medium based on multi-source data fusion, comprising the following steps: optimizing a virtual liquid level monitoring point in a three-dimensional model according to the topographic features and protection height features of a target underground powerhouse; processing image data to obtain to-be-fused image features; obtaining to-be-processed liquid level data at an actual liquid level monitoring point, and performing Kalman filtering and liquid level data redundancy voting on the liquid level data to obtain to-be-fused liquid level data; fusing the to-be-fused liquid level data and the to-be-fused image features to obtain fusion features; and constructing a powerhouse flood prevention alarm neural network model, which is used for flood prevention early warning of the target underground powerhouse after training is completed. The application belongs to the field of underground powerhouse flood prevention. The application can realize early warning of flood prevention.

Description

Technical Field

[0001] This invention relates to the field of flood prevention in underground powerhouses, and more particularly to a method, device, equipment, and medium for flood prevention linkage early warning of underground powerhouses in hydropower stations based on multi-source data fusion. Background Technology

[0002] As an important component of the hydropower system, the underground powerhouse of a hydropower station is in a closed, humid, and densely packed operating environment for a long time. It is susceptible to various factors such as backflow of external floodwaters, seepage from the rock walls, rupture of the water diversion pressure steel pipes, drainage system failures, and equipment leakage, which can lead to water accumulation in the underground powerhouse or even flooding accidents.

[0003] Current flood monitoring methods for underground power plants mainly rely on level sensors or manual inspections. Level sensors have limited monitoring range and are susceptible to installation location and sensor malfunctions, making it difficult to comprehensively reflect the overall water accumulation status of the power plant. Manual inspections suffer from low efficiency, poor real-time performance, and blind spots, making it difficult to promptly detect leaks and abnormal water accumulation in hidden areas. Existing industrial television systems are mostly used for video surveillance and accident tracing, lacking intelligent analysis capabilities for image information and unable to coordinate with level monitoring data for judgment, resulting in low reliability of flood risk warnings. Therefore, improving the reliability of flood risk warnings is an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method, device, equipment, and medium for flood prevention and early warning of underground powerhouses in hydropower stations based on multi-source data fusion. This invention solves the technical problem of low reliability of flood risk early warning in the prior art and achieves the technical effect of improving the reliability of flood risk early warning.

[0005] In a first aspect, the present invention provides a method for flood prevention and early warning linkage of underground powerhouses in hydropower stations based on multi-source data fusion, including: A three-dimensional model of the target underground powerhouse is constructed, and the virtual liquid level monitoring points in the three-dimensional model are optimized according to the terrain features and protection height features of the target underground powerhouse. The virtual liquid level monitoring points are used to arrange actual liquid level monitoring points in the target underground powerhouse. The image data to be processed of the monitoring area of ​​the target underground plant is acquired and processed to obtain the image features to be fused. The target underground plant is divided into several monitoring areas, and there is at least one liquid level monitoring point in each monitoring area. The liquid level data to be processed is acquired at the actual liquid level monitoring point, and Kalman filtering and liquid level data redundancy voting are performed on the liquid level data to obtain the liquid level data to be fused. The liquid level data to be fused is fused with the image features to be fused to obtain the fused features; A neural network model for flood prevention alarm of a factory building is constructed, and the model is trained based on fused features and corresponding flood risk level samples. After training is completed, the neural network model for flood prevention alarm of a factory building is used for flood prevention early warning of the target underground factory building.

[0006] Furthermore, a three-dimensional model of the target underground plant is constructed, including: Acquire point cloud data, architectural design data, and equipment layout data of the target underground factory, and register the point cloud data based on a unified spatial coordinate system; Based on the registered point cloud data and architectural design data, a three-dimensional model of the target underground plant is constructed using BIM modeling. The three-dimensional model is used to characterize the terrain features and spatial distribution of equipment in the target underground plant.

[0007] Furthermore, the virtual liquid level monitoring points in the 3D model are optimized based on the terrain features and protection height features of the target underground plant, including: Terrain analysis was performed on the 3D model to obtain terrain risk features, which include water confluence nodes, low-lying waterlogged areas, and flooded areas. The protection height characteristics are determined based on the unit's accident shutdown requirements and the equipment's installation elevation.

[0008] in, For the first The protection height characteristics of each monitoring area The first one determined according to the unit emergency shutdown requirements The minimum safe elevation of equipment in each monitoring area. For the first Ground elevation of each monitoring area To preset a safety margin; Based on terrain risk characteristics and protection height characteristics, a set of candidate virtual liquid level monitoring points is established, and a genetic algorithm is used to optimize the set of candidate virtual liquid level monitoring points to obtain a layout scheme for virtual liquid level monitoring points.

[0009] Furthermore, a genetic algorithm is used to optimize the set of candidate virtual liquid level monitoring points to obtain a layout scheme for the virtual liquid level monitoring points, including: The 3D model is divided into multiple spatial mesh units, and the first... The center coordinates of each spatial grid cell are Each spatial grid unit belongs to a monitoring area; Based on the protection height characteristics and terrain risk characteristics of each spatial grid cell, the risk index of each spatial grid cell is determined, including:

[0010] in, For the first Risk index of each spatial grid cell All are weights. For the first The water accumulation and confluence nodes of each spatial grid cell are represented as follows: if they exist, they are denoted as 1; if they do not exist, they are denoted as 0. For the first The representation of a low-lying waterlogged area in a spatial grid cell is denoted as 1 if it exists and 0 if it does not. For the first The representation of the flooded region of each spatial grid cell, denoted as 1 if it exists and 0 if it does not. Preset height; Furthermore, a set of candidate virtual liquid level monitoring points was obtained. ,include:

[0011] in, The default value for the risk index; Using monitoring coverage as the optimization objective, a genetic algorithm is used to optimize the candidate set of virtual liquid level monitoring points, resulting in a layout scheme for the virtual liquid level monitoring points.

[0012] Further, the image data is processed to obtain the features of the image to be fused, including: Image data is preprocessed to obtain preprocessed images. Preprocessing includes video frame segmentation, resolution scaling, noise removal, color equalization, and ROI region extraction. The preprocessed image is input into the pre-trained YOLOv8 model for feature extraction, and the target detection results of each grid in each image frame are obtained. The target detection results include the target bounding box, the target confidence score and the target category. Based on the target detection results, extract the water level scale features, ground water features, and seepage morphology features of the corresponding grid. The image anomaly level of the corresponding grid is determined based on the characteristics of the water gauge scale, the characteristics of surface water, and the characteristics of seepage morphology. The target detection results and image anomaly levels are used as features of the image to be fused.

[0013] Furthermore, Kalman filtering and redundancy voting are performed on the liquid level data to obtain the liquid level data to be fused, including: The liquid level data is processed by Kalman filtering to obtain filtered liquid level data; Acquire the signal from the water level float switch node; If the filtered liquid level data simultaneously outputs an alarm signal greater than the first water level threshold and an alarm signal greater than the second water level threshold, and the water level float switch node signal outputs an alarm signal greater than the first water level threshold, then the liquid level data is output and used as the liquid level data to be fused, wherein the first water level threshold is less than the second water level threshold; If the water level float switch node signal simultaneously outputs an alarm signal greater than the first water level threshold and an alarm signal greater than the second water level threshold, and the filtered liquid level data outputs an alarm signal greater than the first water level threshold, then the liquid level data is output and used as the liquid level data to be merged. Otherwise, the previously output liquid level data will be used as the liquid level data to be fused.

[0014] Furthermore, a neural network model for flood prevention alarm of the factory building is constructed, and the model is trained based on fused features and corresponding flood risk level samples, including: Input the fused features into the factory flood prevention alarm neural network model and obtain the flood risk level prediction results; Based on the difference between the predicted flood risk level and the actual flood risk level, the network parameters of the factory flood prevention alarm neural network model are adjusted. When the preset number of iterations is met, the neural network model for flood prevention alarm of the factory building is used for flood prevention early warning of the target underground factory building.

[0015] Secondly, this invention provides a flood prevention and early warning device for underground powerhouses of hydropower stations based on multi-source data fusion, comprising: The model building module is used to build a three-dimensional model of the target underground plant and optimize the virtual liquid level monitoring points in the three-dimensional model according to the terrain features and protection height features of the target underground plant. The virtual liquid level monitoring points are used to arrange actual liquid level monitoring points in the target underground plant. The image feature acquisition module is used to acquire the image data to be processed of the monitoring area of ​​the target underground plant, and process the image data to obtain the image features to be fused. The target underground plant is divided into several monitoring areas, and there is at least one liquid level monitoring point in each monitoring area. The fusion level module is used to acquire the level data to be processed at the actual level monitoring point, and to perform Kalman filtering and redundancy voting on the level data to obtain the level data to be fused. The fusion feature acquisition module is used to fuse the liquid level data to be fused with the features of the image to be fused to obtain the fused features; The model training module is used to construct a neural network model for flood prevention alarm of factory buildings, and to train the neural network model for flood prevention alarm of factory buildings based on fused features and corresponding flood risk level samples. After training is completed, the neural network model for flood prevention alarm of factory buildings is used for flood prevention early warning of target underground factory buildings.

[0016] Thirdly, the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute a flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion, as provided in the first aspect.

[0017] Fourthly, the present invention provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as provided in the first aspect.

[0018] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a method for accurately sensing and intelligently warning of flooding risks in underground power plants by integrating multiple technologies such as 3D spatial modeling, liquid level monitoring, industrial television image analysis, and neural network intelligent recognition. Based on a 3D model of the underground power plant, virtual liquid level monitoring points are optimized by combining terrain features and equipment protection height characteristics, and the virtual liquid level monitoring points are used to guide the layout of actual liquid level monitoring points. This allows monitoring resources to prioritize coverage of water accumulation confluence nodes, low-lying areas, and critical equipment areas, improving the targeting and coverage of liquid level monitoring.

[0019] This invention improves the stability and reliability of liquid level monitoring data by performing Kalman filtering and redundant voting on the liquid level data. At the same time, it combines industrial television images to extract visual features such as water level gauge scale, water accumulation range and leakage pattern, so as to achieve complementary advantages between liquid level information and image information, and effectively overcome the problems of single sensors being susceptible to obstruction, noise and false alarms.

[0020] This invention utilizes neural networks to perform deep learning on multi-source fusion features, which can fully explore the correlation between liquid level changes, image anomalies, and regional attributes, enabling accurate prediction of the flood risk level of underground powerhouses. This improves the accuracy and robustness of flood risk identification under complex operating conditions. Ultimately, based on the prediction results, it achieves hierarchical linkage early warning and automatic handling, providing more timely, reliable, and intelligent flood prevention and safety assurance for underground powerhouses in hydropower stations, effectively reducing the risk of equipment flooding, and enhancing the safety and stability of hydropower station operation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating the flood prevention and early warning method for underground powerhouses in hydropower stations based on multi-source data fusion provided by this invention; Figure 2 A flowchart illustrating the redundant voting strategy provided by this invention. Detailed Implementation

[0023] This invention provides a flood prevention and early warning method for underground powerhouses in hydropower stations based on multi-source data fusion, which solves the technical problem of low reliability of flood risk early warning in the prior art.

[0024] The technical solution of this invention is to solve the above-mentioned technical problems, and the overall idea is as follows: A method for flood prevention and early warning of underground powerhouses in hydropower stations based on multi-source data fusion includes: constructing a three-dimensional model of the target underground powerhouse and optimizing virtual liquid level monitoring points in the three-dimensional model according to the terrain features and protection height features of the target underground powerhouse, wherein the virtual liquid level monitoring points are used to deploy actual liquid level monitoring points in the target underground powerhouse; acquiring image data to be processed in the monitoring area of ​​the target underground powerhouse and processing the image data to obtain image features to be fused, wherein the target underground powerhouse is divided into several monitoring areas, and each monitoring area has at least one liquid level monitoring point; acquiring liquid level data to be processed at the actual liquid level monitoring points and performing Kalman filtering and liquid level data redundancy voting on the liquid level data to obtain liquid level data to be fused; fusing the liquid level data to be fused with the image features to be fused to obtain fused features; constructing a powerhouse flood prevention alarm neural network model and training the powerhouse flood prevention alarm neural network model based on the fused features and corresponding flood risk level samples; after training is completed, using the powerhouse flood prevention alarm neural network model for flood prevention and early warning of the target underground powerhouse.

[0025] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0026] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0027] This invention provides, for example Figure 1 The flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion, as shown, includes steps S11-S15: Step S11: Construct a three-dimensional model of the target underground plant, and optimize the virtual liquid level monitoring points in the three-dimensional model according to the terrain features and protection height features of the target underground plant. The virtual liquid level monitoring points are used to arrange actual liquid level monitoring points in the target underground plant.

[0028] Constructing a 3D model of the target underground plant includes: acquiring point cloud data, architectural design data, and equipment layout data of the target underground plant, and registering the point cloud data based on a unified spatial coordinate system; based on the registered point cloud data and architectural design data, constructing a 3D model of the target underground plant using BIM modeling, and using the 3D model to characterize the terrain features and spatial distribution of equipment of the target underground plant.

[0029] Specifically, high-precision measuring equipment such as laser scanners and total stations can be used to collect on-site data of the target underground plant, and obtain high-density point cloud data of objects such as the main structure of the plant, corridors, water collection wells, drainage ditches, equipment foundations and electromechanical equipment. Simultaneously, the architectural design drawings, structural design parameters, and equipment layout data of the factory building are obtained. The architectural design data includes the structural dimensions, elevation information, and geometric parameters of the components, while the equipment layout data includes the installation location, external dimensions, and spatial layout relationship of each piece of equipment.

[0030] Using a unified spatial coordinate system as a benchmark, coordinate transformation, noise removal, and multi-site stitching and registration are performed on the point cloud data to ensure spatial consistency between the point cloud data and the architectural design data, providing a unified data foundation for subsequent 3D model construction.

[0031] Using architectural design data as the model framework, geometric reconstruction is performed on the main structure of the factory, internal corridors, floors, walls, equipment foundations, and drainage facilities based on the registered point cloud data. Combined with equipment layout data, parametric modeling of various power generation equipment, auxiliary equipment, and pipeline facilities is completed, so that the model can not only realistically reflect the architectural structure of the underground factory, but also accurately express the spatial relationship and installation height between equipment.

[0032] During the modeling process, BIM modeling software such as AutoCAD and Revit can be used to assign corresponding attribute information to various building components and equipment objects, thereby achieving integrated building structural information, equipment information and spatial information.

[0033] The 3D model can accurately reflect the elevation changes, corridor slopes, sump well inlet shapes, drainage ditch widths and connectivity of different areas inside the underground powerhouse, as well as other terrain features. It can also accurately describe the spatial location, elevation and spacing of generator sets, transformers, control cabinets and other key equipment.

[0034] Based on the terrain features and protection height characteristics of the target underground powerhouse, the virtual liquid level monitoring points in the 3D model are optimized, including: Terrain analysis was performed on the 3D model to obtain terrain risk features, which include water confluence nodes, low-lying waterlogged areas, and flooded areas. The protection height characteristics are determined based on the unit's accident shutdown requirements and the equipment's installation elevation.

[0035] in, For the first The protection height characteristics of each monitoring area The first one determined according to the unit emergency shutdown requirements The minimum safe elevation of equipment in each monitoring area. For the first Ground elevation of each monitoring area To preset a safety margin; Based on terrain risk characteristics and protection height characteristics, a set of candidate virtual liquid level monitoring points is established, and a genetic algorithm is used to optimize the set of candidate virtual liquid level monitoring points to obtain a layout scheme for virtual liquid level monitoring points.

[0036] A genetic algorithm is used to optimize the set of candidate virtual liquid level monitoring points, resulting in a layout scheme for the virtual liquid level monitoring points, including: The 3D model is divided into multiple spatial mesh units, and the first... The center coordinates of each spatial grid cell are Each spatial grid unit belongs to a monitoring area; Based on the protection height characteristics and terrain risk characteristics of each spatial grid cell, the risk index of each spatial grid cell is determined, including:

[0037] in, For the first Risk index of each spatial grid cell All are weights. For the first The water accumulation and confluence nodes of each spatial grid cell are represented as follows: if they exist, they are denoted as 1; if they do not exist, they are denoted as 0. For the first The representation of a low-lying waterlogged area in a spatial grid cell is denoted as 1 if it exists and 0 if it does not. For the first The representation of the flooded region of each spatial grid cell, denoted as 1 if it exists and 0 if it does not. Preset height; Furthermore, a set of candidate virtual liquid level monitoring points was obtained. ,include:

[0038] in, The default value for the risk index; Using monitoring coverage as the optimization objective, a genetic algorithm is used to optimize the candidate set of virtual liquid level monitoring points, resulting in a layout scheme for the virtual liquid level monitoring points.

[0039] Specifically, based on the completed 3D model of the underground plant, spatial topographic analysis is conducted on areas such as corridors, equipment floors, drainage ditches, water collection wells, and ecological water intake pipes inside the plant. By calculating the elevation changes, slope distribution, confluence direction, and local depression characteristics at each location, the natural confluence path of rainwater or seepage water in the underground plant and the trend of water accumulation and diffusion are identified, and topographic risk characteristics such as water accumulation nodes, low-lying water accumulation areas, and flooded areas are obtained.

[0040] Among them, the water confluence node is used to characterize the location where the water flow finally converges, the low-lying water accumulation area is used to characterize the location where continuous water accumulation is likely to occur, and the flooded area is used to characterize the area where equipment may be flooded first after the liquid level continues to rise.

[0041] Based on the hydropower station unit accident shutdown specifications, the minimum safe operating elevation of key equipment in different monitoring areas is obtained. Combined with the equipment installation elevation and ground elevation in the 3D model, the protection height characteristics corresponding to each monitoring area are calculated according to the formula.

[0042] The 3D model is divided into multiple spatial grid units. The risk index is calculated based on the characteristics of water accumulation confluence nodes, low-lying water accumulation areas, flooded areas and protection heights corresponding to each spatial grid unit. A set of candidate virtual liquid level monitoring points is obtained based on the risk index.

[0043] Subsequently, with the goal of maximizing monitoring coverage (minimizing monitoring blind spots), a genetic algorithm was used to encode candidate monitoring points. Through iterative processes such as population initialization, fitness evaluation, selection, crossover, and mutation, the location of monitoring points was globally optimized. This prioritized the distribution of liquid level monitoring points at water accumulation nodes, low-lying water accumulation areas, near equipment, and areas with low accident protection heights. At the same time, it also took into account areas prone to leakage, such as near unit equipment, cable trenches, water collection wells, drainage ditches, and ecological water intake pipes. Finally, the optimal layout scheme of virtual liquid level monitoring points was obtained, achieving full coverage monitoring of key areas of the underground plant.

[0044] To further verify the rationality of the virtual liquid level monitoring point layout scheme, actual liquid level monitoring points can be arranged in the field area corresponding to the optimized monitoring point locations, and real-time sampling verification can be performed by linking industrial television images with liquid level transmitter data.

[0045] Specifically, high-precision liquid level transmitters are added in areas prone to leakage or water accumulation, such as near the unit equipment, cable trenches, water collection wells, drainage ditches, and ecological water intake pipes. These transmitters are then compared spatiotemporally with image data collected by industrial television to verify the monitoring points' response to liquid level changes and the monitoring coverage effect. If monitoring blind spots or prediction errors are found in local areas, the candidate virtual liquid level monitoring points in the 3D model are re-optimized to improve the accuracy and reliability of the monitoring point layout scheme.

[0046] Step S12: Obtain the image data to be processed of the monitoring area of ​​the target underground plant, and process the image data to obtain the image features to be fused. The target underground plant is divided into several monitoring areas, and there is at least one liquid level monitoring point in each monitoring area.

[0047] Specifically, the process includes: preprocessing image data to obtain preprocessed images, including video frame segmentation, resolution scaling, noise removal, color equalization, and ROI region extraction; inputting the preprocessed images into a pre-trained YOLOv8 model for feature extraction to obtain target detection results for each grid in each image frame, including target bounding boxes, target confidence scores, and target categories; extracting water level scale features, surface water features, and seepage morphology features for the corresponding grids based on the target detection results; determining the image anomaly level for the corresponding grids based on the water level scale features, surface water features, and seepage morphology features; and using the target detection results and image anomaly levels as features for the image to be fused.

[0048] Specifically, industrial television cameras can be deployed near the unit equipment, cable trenches, water collection wells, drainage ditches, and other areas prone to leakage and water accumulation, based on the actual liquid level monitoring point locations corresponding to each monitoring area in the underground plant. At least one camera should be configured for each monitoring area to ensure that there are no blind spots in the video monitoring of the corresponding area.

[0049] The camera can be a high-resolution CMOS network camera, featuring low-light imaging, infrared illumination, waterproofing, dustproofing, and low network latency to meet the continuous video acquisition needs in the complex environment of an underground factory. The camera continuously acquires video data from various monitoring areas of the target underground factory and extracts video frames at preset time intervals to form image data to be processed.

[0050] The acquired video data is processed sequentially by video frame segmentation, resolution scaling, noise removal, color equalization, and ROI region extraction.

[0051] Among them, video framing is used to convert continuous video into continuous image frames; resolution scaling is used to unify the image input size and improve the efficiency of subsequent neural network inference; noise removal is used to eliminate image noise generated in the low-light environment of the underground factory; color equalization is used to enhance image contrast and improve the recognizability of target areas; ROI region extraction is used to extract key areas such as water level gauge areas, ground water areas, and seepage areas from the whole image, reduce background interference, improve the accuracy of subsequent target detection, and obtain a preprocessed image.

[0052] The preprocessed image is divided into multiple detection grids, each responsible for the target detection task. The convolutional neural network in the YOLOv8 network is used to perform multi-layer convolution and feature aggregation on the image to extract semantic features at different scales. The bounding box, target confidence and target category of the target corresponding to each detection grid are predicted to form the target detection result.

[0053] Based on the target detection results, the water level scale features, ground water features, and leakage morphology features of the corresponding grid are further extracted. Among them, the water level scale features are used to reflect the current liquid level height change, the ground water features are used to characterize the water accumulation area, coverage range, and diffusion trend, and the leakage morphology features are used to characterize the leakage location, leakage shape, and leakage degree.

[0054] The image anomaly level of the corresponding grid is determined based on the water level gauge scale characteristics, ground water features, and seepage morphology characteristics, and the target detection results and image anomaly levels are used as the features of the image to be fused.

[0055] Specifically, by comprehensively analyzing the changes in water level gauge readings, the extent of surface water accumulation, and the changes in leakage patterns, an image anomaly level assessment is conducted for each detection grid to reflect the risk level of abnormal liquid levels, water spread, or equipment leakage in the current monitoring area.

[0056] For example:

[0057] in, For the first Image anomaly scoring for each detection grid. ~ As weight; The normalized range of the water level gauge reading is used to characterize the degree of liquid level change. This represents the proportion of the area of ​​water accumulation on the ground to the area of ​​the corresponding detection grid, used to characterize the degree of water spread. The leakage morphology score is determined by comprehensively considering the leakage area, connectivity, and edge diffusion rate, and is used to characterize the severity of leakage.

[0058] Based on image anomaly scoring The correspondence between the image anomaly score and the preset level threshold is used to determine the image anomaly level of the corresponding detection grid. The higher the image anomaly score, the higher the risk of liquid level anomaly, water diffusion, or equipment leakage in the corresponding detection grid.

[0059] The target bounding box, target confidence, target category, and corresponding grid image anomaly level in the target detection results are combined to form the image features to be fused, providing a unified image feature input for subsequent multi-source information fusion with liquid level monitoring data and regional feature data.

[0060] Step S13: Obtain the liquid level data to be processed at the actual liquid level monitoring point, and perform Kalman filtering and liquid level data redundancy voting on the liquid level data to obtain the liquid level data to be fused. Specifically, this includes: performing Kalman filtering on the liquid level data to obtain filtered liquid level data; and acquiring the water level float switch node signal. If the filtered liquid level data simultaneously outputs an alarm signal greater than the first water level threshold and an alarm signal greater than the second water level threshold, and the water level float switch node signal outputs an alarm signal greater than the first water level threshold, then the liquid level data is output and used as the liquid level data to be fused, wherein the first water level threshold is less than the second water level threshold; If the water level float switch node signal simultaneously outputs an alarm signal greater than the first water level threshold and an alarm signal greater than the second water level threshold, and the filtered liquid level data outputs an alarm signal greater than the first water level threshold, then the liquid level data is output and used as the liquid level data to be merged. Otherwise, the previously output liquid level data will be used as the liquid level data to be fused.

[0061] Specifically, according to the virtual liquid level monitoring point layout plan, actual liquid level monitoring points are deployed in various monitoring areas of the underground plant. Submersible liquid level sensors are used in open areas such as sump pits and drainage ditches, while pressure-type liquid level sensors are used in pressure pipeline areas. Submersible liquid level transmitters are installed in each unit section of the plant corridor, and liquid level float switches are added to form redundant monitoring. The 4-20mA or 1-5V analog signals output by the liquid level sensors are converted into digital liquid level data by the PLC acquisition module and transmitted to the data fusion processing center via hardwired or wireless communication to form the liquid level data to be processed. Simultaneously, the calibration coefficients obtained from the periodic calibration of standard sensors are used to correct errors in the liquid level data, improving the accuracy of liquid level measurement.

[0062] The continuously collected liquid level time series can be used as the input of a Kalman filter. A liquid level state prediction model can be established based on the liquid level estimate of the previous moment to predict the liquid level value at the current moment. The state can then be updated by combining the liquid level measurement value at the current moment, thereby obtaining filtered liquid level data.

[0063] The first and second water level threshold alarm signals corresponding to the filtered liquid level data are acquired, as well as the first and second water level threshold alarm signals output by the water level float switch. When the filtered liquid level data simultaneously outputs alarm signals greater than the first and second water level thresholds, and the water level float switch node signal output is greater than the first water level threshold alarm signal, the abnormal liquid level state is determined to be established, and the current liquid level data is output as the liquid level data to be fused. Alternatively, when the water level float switch node signal simultaneously outputs an alarm signal greater than the first water level threshold and an alarm signal greater than the second water level threshold, and the filtered liquid level data output is greater than the first water level threshold alarm signal, the liquid level abnormality is also determined to be established, and the current liquid level data is output as the liquid level data to be merged.

[0064] The above judgment logic essentially adopts a three-to-two redundant voting strategy for liquid level information. By cross-verifying the alarm signals of the liquid level transmitter and the float switch, it improves the reliability of liquid level anomaly judgment and reduces the risk of misjudgment caused by single sensor failure or false alarm.

[0065] If none of the above redundant voting conditions are met, the current liquid level monitoring result is considered to be uncertain. Therefore, the currently collected data is not directly used; instead, the previously output liquid level data is used as the liquid level data to be fused. By retaining the most recently confirmed valid liquid level data through redundant voting, sudden changes in liquid level data due to transient noise, communication anomalies, or single sensor failures can be avoided. This ensures the continuity, stability, and reliability of the liquid level data input to the subsequent multi-source data fusion module, providing accurate liquid level information for flood prevention and early warning in underground power plants.

[0066] The process of the three-choice-two-redundant voting strategy can also be referenced. Figure 2 .

[0067] Step S14: The liquid level data to be fused and the image features to be fused are fused to obtain fused features.

[0068] The fused features can be in a weighted form or a vectorized form; there is no restriction here, and the choice can be made according to the actual situation.

[0069] Step S15: Construct a factory flood prevention alarm neural network model, and train the factory flood prevention alarm neural network model based on the fusion features and corresponding flood risk level samples. After training is completed, the factory flood prevention alarm neural network model is used for flood prevention early warning of the target underground factory.

[0070] Specifically, this includes: inputting fused features into the factory flood prevention alarm neural network model and obtaining the flood risk level prediction result; adjusting the network parameters of the factory flood prevention alarm neural network model based on the difference between the flood risk level prediction result and the actual flood risk level result; and when the preset number of iterations is met, using the factory flood prevention alarm neural network model for flood prevention early warning of the target underground factory.

[0071] A backpropagation (BP) neural network is constructed as the neural network model for flood prevention alarm in the factory. The BP neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to input fused features, which include liquid level data, image anomaly features, regional features, and information such as liquid level change rate and historical change trend. The hidden layer is used to learn the nonlinear mapping relationship between the features and can be set with 6 neurons to fully explore the correlation features between liquid level changes, image anomalies, and regional attributes. The output layer outputs the flood risk level prediction result for the corresponding monitoring area, which includes low risk, general risk, and high risk.

[0072] Among them, low risk corresponds to normal liquid level and only slight water seepage in the image; general risk corresponds to liquid level close to the warning threshold and continuously rising, while the image shows the area of ​​water accumulation on the ground continuously expanding; greater risk corresponds to liquid level two-out-of-three judgment signal triggered or liquid level rapidly exceeding the warning threshold, while the image shows obvious water accumulation or a large amount of leakage, and the flooded area continues to expand.

[0073] Input fused features into the factory flood prevention alarm neural network model and obtain the flood risk level prediction result; adjust the network parameters of the factory flood prevention alarm neural network model according to the difference between the flood risk level prediction result and the actual flood risk level result.

[0074] Specifically, using historical flooding event data and corresponding manually labeled risk levels as training samples, forward propagation is used to calculate the risk level prediction results for each training sample, and the error between the prediction results and the actual risk levels is calculated. The connection weights and bias parameters of each layer of neurons are continuously adjusted through the error backpropagation algorithm until the model error gradually converges. When the preset number of iterations is met or the model error reaches the preset threshold, the training of the factory flood prevention alarm neural network model is completed, and the trained model is used for real-time flood prevention early warning of the target underground factory.

[0075] In the actual early warning process, the fusion features acquired in real time are input into the trained factory flood prevention alarm neural network model, which outputs the flood risk level of the corresponding monitoring area and executes the corresponding graded linkage early warning strategy according to the risk level.

[0076] When the prediction result is low risk, a level one early warning is triggered, outputting monitoring alarm information and pushing the corresponding industrial television screen; when the prediction result is moderate risk, a level two early warning is triggered, starting the plant drainage equipment and issuing an audible and visual alarm; when the prediction result is significant risk, combined with the structural characteristics of the monitoring area and the time delay protection strategy, joint response measures such as emergency shutdown of the unit, closure of the accident gate, and voice alarm throughout the plant are implemented, thereby realizing intelligent identification, graded early warning, and automatic linkage control of underground plant flooding risk.

[0077] In summary, this invention provides a method for accurately perceiving and intelligently warning of flooding risks in underground power plants by integrating multiple technologies, including 3D spatial modeling, liquid level monitoring, industrial television image analysis, and neural network intelligent recognition. Based on a 3D model of the underground power plant, virtual liquid level monitoring points are optimized by combining terrain features and equipment protection height characteristics, and this optimization guides the layout of actual liquid level monitoring points. This ensures that monitoring resources can prioritize coverage of water accumulation nodes, low-lying areas, and critical equipment areas, thereby improving the targeting and coverage of liquid level monitoring.

[0078] This invention improves the stability and reliability of liquid level monitoring data by performing Kalman filtering and redundant voting on the liquid level data. At the same time, it combines industrial television images to extract visual features such as water level gauge scale, water accumulation range and leakage pattern, so as to achieve complementary advantages between liquid level information and image information, and effectively overcome the problems of single sensors being susceptible to obstruction, noise and false alarms.

[0079] This invention utilizes neural networks to perform deep learning on multi-source fusion features, which can fully explore the correlation between liquid level changes, image anomalies, and regional attributes, enabling accurate prediction of the flood risk level of underground powerhouses. This improves the accuracy and robustness of flood risk identification under complex operating conditions. Ultimately, based on the prediction results, it achieves hierarchical linkage early warning and automatic handling, providing more timely, reliable, and intelligent flood prevention and safety assurance for underground powerhouses in hydropower stations, effectively reducing the risk of equipment flooding, and enhancing the safety and stability of hydropower station operation.

[0080] Based on the same inventive concept, this invention provides a flood prevention and early warning device for underground powerhouses of hydropower stations based on multi-source data fusion, comprising: The model building module is used to build a three-dimensional model of the target underground plant and optimize the virtual liquid level monitoring points in the three-dimensional model according to the terrain features and protection height features of the target underground plant. The virtual liquid level monitoring points are used to arrange actual liquid level monitoring points in the target underground plant. The image feature acquisition module is used to acquire the image data to be processed of the monitoring area of ​​the target underground plant, and process the image data to obtain the image features to be fused. The target underground plant is divided into several monitoring areas, and there is at least one liquid level monitoring point in each monitoring area. The fusion level module is used to acquire the level data to be processed at the actual level monitoring point, and to perform Kalman filtering and redundancy voting on the level data to obtain the level data to be fused. The fusion feature acquisition module is used to fuse the liquid level data to be fused with the features of the image to be fused to obtain the fused features; The model training module is used to construct a neural network model for flood prevention alarm of factory buildings, and to train the neural network model for flood prevention alarm of factory buildings based on fused features and corresponding flood risk level samples. After training is completed, the neural network model for flood prevention alarm of factory buildings is used for flood prevention early warning of target underground factory buildings.

[0081] Based on the same inventive concept, the present invention also provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute a flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion, as described above.

[0082] Based on the same inventive concept, the present invention also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the aforementioned method for flood prevention and early warning of underground powerhouses in hydropower stations based on multi-source data fusion.

[0083] Since the electronic device described in this embodiment is an electronic device used to implement the information processing method in the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the information processing method described in the embodiments of the present invention. Therefore, how the electronic device implements the method in the embodiments of the present invention will not be described in detail here. Any electronic device used by those skilled in the art to implement the information processing method in the embodiments of the present invention falls within the scope of protection of the present invention.

[0084] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0085] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0086] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0087] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0088] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0089] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A flood prevention and early warning method for underground powerhouses in hydropower stations based on multi-source data fusion, characterized in that: include: A three-dimensional model of the target underground plant is constructed, and the virtual liquid level monitoring points in the three-dimensional model are optimized according to the terrain features and protection height features of the target underground plant. The virtual liquid level monitoring points are used to arrange actual liquid level monitoring points in the target underground plant. The image data to be processed of the monitoring area of ​​the target underground plant is acquired and processed to obtain the image features to be fused. The target underground plant is divided into several monitoring areas, and there is at least one liquid level monitoring point in each monitoring area. The liquid level data to be processed is acquired at the actual liquid level monitoring point, and Kalman filtering and liquid level data redundancy voting are performed on the liquid level data to obtain the liquid level data to be fused. The liquid level data to be fused is fused with the image features to be fused to obtain the fused features; A neural network model for flood prevention alarm of a factory building is constructed, and the model is trained based on fusion features and corresponding flood risk level samples. After training is completed, the neural network model for flood prevention alarm of a factory building is used for flood prevention early warning of the target underground factory building.

2. The flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as described in claim 1, characterized in that, Construct a 3D model of the target underground powerhouse, including: The point cloud data, architectural design data, and equipment layout data of the target underground factory are acquired, and the point cloud data is registered based on a unified spatial coordinate system. Based on the registered point cloud data and architectural design data, a three-dimensional model of the target underground factory is constructed using BIM modeling. The three-dimensional model is used to characterize the terrain features and spatial distribution of equipment in the target underground factory.

3. The flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as described in claim 1, characterized in that, Based on the terrain features and protection height characteristics of the target underground powerhouse, the virtual liquid level monitoring points in the 3D model are optimized, including: A terrain analysis is performed on the three-dimensional model to obtain terrain risk features, which include water confluence nodes, low-lying waterlogged areas, and flooded areas. The protection height characteristics are determined based on the unit's accident shutdown requirements and the equipment's installation elevation. in, For the first The protection height characteristics of each monitoring area The first one determined according to the unit emergency shutdown requirements The minimum safe elevation of equipment in each monitoring area. For the first Ground elevation of each monitoring area To preset a safety margin; Based on terrain risk characteristics and protection height characteristics, a set of candidate virtual liquid level monitoring points is established, and a genetic algorithm is used to optimize the set of candidate virtual liquid level monitoring points to obtain a layout scheme for virtual liquid level monitoring points.

4. The flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as described in claim 3, characterized in that, A genetic algorithm is used to optimize the candidate set of virtual liquid level monitoring points to obtain a layout scheme for the virtual liquid level monitoring points, including: The 3D model is divided into multiple spatial mesh units, and the first... The center coordinates of each spatial grid cell are Each spatial grid unit belongs to a monitoring area; Based on the protection height characteristics and terrain risk characteristics of each spatial grid cell, the risk index of each spatial grid cell is determined, including: in, For the first Risk index of each spatial grid cell All are weights. For the first The water accumulation and confluence nodes of each spatial grid cell are represented as follows: if they exist, they are denoted as 1; if they do not exist, they are denoted as 0. For the first The representation of a low-lying waterlogged area in a spatial grid cell is denoted as 1 if it exists and 0 if it does not. For the first The representation of the flooded region of each spatial grid cell, denoted as 1 if it exists and 0 if it does not. Preset height; Furthermore, a set of candidate virtual liquid level monitoring points was obtained. ,include: in, The default value for the risk index; Using monitoring coverage as the optimization objective, a genetic algorithm is used to optimize the candidate set of virtual liquid level monitoring points to obtain a layout scheme for the virtual liquid level monitoring points.

5. The flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as described in claim 1, characterized in that, The image data is processed to obtain the features of the image to be fused, including: The image data is preprocessed to obtain a preprocessed image. The preprocessing includes video frame splitting, resolution scaling, noise removal, color equalization, and ROI region extraction. The preprocessed image is input into a pre-trained YOLOv8 model for feature extraction to obtain the target detection result for each grid in each image frame. The target detection result includes the target bounding box, the target confidence score, and the target category. Based on the target detection results, extract the corresponding grid's water level scale features, ground water features, and seepage morphology features; The image anomaly level of the corresponding grid is determined based on the water gauge scale characteristics, the ground water accumulation characteristics, and the seepage morphology characteristics. The target detection results and the image anomaly level are used as the features of the image to be fused.

6. The flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as described in claim 1, characterized in that, The liquid level data is subjected to Kalman filtering and liquid level data redundancy voting to obtain the liquid level data to be fused, including: The liquid level data is processed by Kalman filtering to obtain filtered liquid level data; Acquire the signal from the water level float switch node; If the filtered liquid level data simultaneously outputs an alarm signal greater than the first water level threshold and an alarm signal greater than the second water level threshold, and the water level float switch node signal outputs an alarm signal greater than the first water level threshold, then the liquid level data is output and used as the liquid level data to be fused, wherein the first water level threshold is less than the second water level threshold; If the water level float switch node signal simultaneously outputs an alarm signal greater than the first water level threshold and an alarm signal greater than the second water level threshold, and the filtered liquid level data outputs an alarm signal greater than the first water level threshold, then the liquid level data is output and used as the liquid level data to be merged. Otherwise, the previously output liquid level data will be used as the liquid level data to be fused.

7. The flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as described in claim 1, characterized in that, A neural network model for flood prevention alarm of a factory building is constructed, and the model is trained based on fused features and corresponding flood risk level samples, including: Input the fused features into the factory flood prevention alarm neural network model and obtain the flood risk level prediction results; Based on the difference between the predicted flood risk level and the actual flood risk level, the network parameters of the factory flood prevention alarm neural network model are adjusted. When the preset number of iterations is met, the neural network model for flood prevention alarm of the factory building is used for flood prevention early warning of the target underground factory building.

8. A flood prevention and early warning device for underground powerhouses of hydropower stations based on multi-source data fusion, characterized in that, include: The model building module is used to build a three-dimensional model of the target underground plant and optimize the virtual liquid level monitoring points in the three-dimensional model according to the terrain features and protection height features of the target underground plant. The virtual liquid level monitoring points are used to arrange actual liquid level monitoring points in the target underground plant. The image feature acquisition module is used to acquire image data to be processed of the monitoring area of ​​the target underground plant, and process the image data to obtain image features to be fused. The target underground plant is divided into several monitoring areas, and there is at least one liquid level monitoring point in each monitoring area. The fusion liquid level module is used to acquire liquid level data to be processed at actual liquid level monitoring points, and to perform Kalman filtering and liquid level data redundancy voting on the liquid level data to obtain liquid level data to be fused. The fusion feature acquisition module is used to fuse the liquid level data to be fused with the features of the image to be fused to obtain the fused features; The model training module is used to construct a factory flood prevention alarm neural network model, and to train the factory flood prevention alarm neural network model based on fused features and corresponding flood risk level samples. After training is completed, the factory flood prevention alarm neural network model is used for flood prevention early warning of target underground factories.

9. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to execute the flood prevention and early warning method for underground powerhouses of hydropower stations based on multi-source data fusion as described in any one of claims 1 to 7.

Images