A method for predicting dam body erosion of a hydropower station

Abstract

The application discloses a kind of hydropower station dam body erosion prediction method, it is related to hydropower station engineering safety monitoring technical field, comprising the following steps: S1, obtains the historical image of dam body spillway apron at each period of history, and the historical image is preprocessed;S2, according to the image after preprocessing, erosion area is identified using deep learning instance segmentation model, and the erosion volume of each period is calculated in combination with digital elevation model difference method;S3, extract the discharge parameter synchronous with historical image from hydropower station monitoring system;The erosion prediction method of the application uses multiple factors such as discharge capacity, sediment scouring capacity and apron structure parameters, and combines historical erosion volume to predict the future erosion volume of hydropower station dam body, through erosion prediction and risk assessment, can reasonably arrange maintenance work, avoid excessive maintenance or insufficient maintenance, improve the operation efficiency and economic benefit of hydropower station.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology for hydropower station projects, and in particular to a method for predicting erosion of hydropower station dams. Background Technology

[0002] In hydropower station projects, the concrete structure below the spillway of the dam, namely the apron structure, plays a crucial role. When water flows down from the spillway of the dam, it carries enormous energy. The apron structure can effectively reduce the kinetic energy of the water flow, prevent excessive scouring of the downstream riverbed and banks, and ensure the safety and stability of the hydropower station and the surrounding area.

[0003] During the operation of hydropower stations, the spillway apron of the dam body is subjected to long-term scouring by water flow and silt erosion, leading to gradual structural damage and phenomena such as pitting and spalling on the concrete surface. If erosion problems are not monitored and predicted in a timely manner, they may cause serious engineering safety accidents. Traditional erosion monitoring methods mainly rely on regular manual inspections and simple experience-based estimations, which have disadvantages such as low monitoring frequency, poor accuracy, and inability to predict future erosion trends, increasing the risk of engineering safety accidents and raising maintenance costs. Therefore, a method for predicting erosion in hydropower station dam bodies is proposed. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the prior art by proposing a method for predicting erosion of hydropower station dams.

[0005] A method for predicting erosion of hydropower station dams includes the following steps:

[0006] S1. Obtain historical images of the spillway apron of the dam body at various historical periods, and preprocess the historical images;

[0007] S2. Based on the preprocessed image, a deep learning instance segmentation model is used to identify the erosion area, and the erosion volume at each stage is calculated using the digital elevation model difference method. ;

[0008] S3. Extract discharge parameters synchronized with historical images from the hydropower station monitoring system, wherein the discharge parameters include discharge volume. and sediment content Calculate the amount of sediment erosion. ;

[0009] S4. Construct and train an erosion prediction model, wherein the erosion prediction model uses the discharge volume of historical periods. and sediment erosion volume and tank structure parameters i As input, predict the erosion volume for different historical periods. For output;

[0010] S5. Input the predicted discharge volume for the future preset time period. and sediment content Calculate future erosion volume using an erosion prediction model ;

[0011] S6. Based on future erosion volume Assess the risk of dam erosion.

[0012] Preferably, in step S1, the historical images include at least two of the following: high-resolution satellite images, UAV oblique photography images, or ground three-dimensional laser scanning point clouds. The preprocessing of the historical images includes radiometric correction, geometric registration, and multi-temporal image matching.

[0013] Preferably, in step S2, the erosion volume at each stage The calculation formula is:

[0014]

[0015] in, For the first Pixels in historical images Elevation value, The pixel area.

[0016] Preferably, in step S3, the amount of sediment erosion is calculated. The formula is:

[0017] .

[0018] Preferably, the erosion prediction model adopts a temporal hybrid model architecture, including a spatial feature extraction module and a temporal feature fusion module;

[0019] The spatial feature extraction module processes multi-temporal erosion volume data based on a 3D convolutional neural network;

[0020] The temporal feature fusion module uses a Transformer encoder to capture the long-term dependency between discharge volume and sediment scour volume.

[0021] Preferably, in step S4, the loss function of the erosion prediction model... Designed as follows:

[0022]

[0023] in, Mean square error, For robust loss function, and These are the weighting coefficients. Erosion volume calculated from historical images , The predicted erosion volume output by the erosion prediction model .

[0024] Preferably, in step S4, the tank shield structural parameters i Including the angle of the tank α ,thickness d and concrete abrasion resistance coefficient k .

[0025] Preferably, in step S5, the predicted discharge volume... and sediment content Predictions are based on future rainfall, watershed evaporation, and historical discharge data.

[0026] Preferably, in step S6, the step of determining the future erosion volume... Assessing the risk of dam erosion includes:

[0027] Define the safe threshold for erosion volume Its design life is based on the dam revetment design life. Dynamic adjustment, the formula is:

[0028]

[0029] in, Overall safety erosion volume, The aging index of the tank's structural materials;

[0030] when < At that time, the erosion was judged to be of low risk;

[0031] when ≥ At that time, the risk of erosion was determined to be high, and an emergency warning was issued.

[0032] Compared with existing technologies, the advantages of this invention are:

[0033] The erosion prediction method of this invention utilizes multiple factors such as discharge volume, sediment erosion volume, and abutment structure parameters, combined with historical erosion volume, to predict the future erosion volume of the hydropower station dam. Through erosion prediction and risk assessment, maintenance work can be rationally arranged to avoid over-maintenance or under-maintenance, thereby improving the operating efficiency and economic benefits of the hydropower station. Attached Figure Description

[0034] Figure 1 This is a flowchart of the prediction method of the present invention. Detailed Implementation

[0035] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0036] Reference Figure 1 As shown, a method for predicting erosion of a hydropower station dam includes the following steps:

[0037] S1. Acquire historical images of the spillway apron of the dam body at various historical periods. The historical images include at least two of the following: high-resolution satellite imagery, UAV oblique photography images, or ground-based three-dimensional laser scanning point clouds. When taking images, avoid collecting them during the spillway period and at night to prevent the acquisition of unclear images. The historical images should contain data from multiple flood seasons and non-flood seasons of the hydropower station to comprehensively reflect various aspects of the erosion process. The historical images are then preprocessed, including radiometric correction, geometric registration, and multi-temporal image matching.

[0038] Images acquired at different times, by different sensors, or under different atmospheric conditions may have significantly different radiometric values. Radiometric correction is used to eliminate or reduce differences in image radiometric values ​​caused by factors such as differences in sensor response, changes in atmospheric conditions (such as aerosol and water vapor content), and changes in solar altitude and azimuth angles. Geometric registration is used to spatially align images acquired at different times, from different perspectives, or by different sensors so that they can be accurately mapped to the same geographic coordinate system, thereby ensuring the accuracy of subsequent erosion volume calculations. Multi-temporal image matching is used to align and match images acquired at different times in a time series to analyze the changes in the erosion process over time.

[0039] S2. Based on the preprocessed image, an erosion area is identified using a deep learning instance segmentation model. The deep learning instance segmentation model can be YOLOv8-Seg or SOLOv2. The erosion volume for each period is calculated using the digital elevation model (DEM) difference method. The calculation formula is:

[0040]

[0041] in, For the first Pixels in historical images Elevation value, The formula uses pixel area as a unit; it is based on the principle of calculating volume change based on changes in terrain elevation. By multiplying the elevation difference of each pixel at different times by the pixel area and summing the results for all pixels, the erosion volume of the entire eroded area at that time is obtained. This method makes full use of the high-precision spatial information of the image and can quantify the degree of erosion relatively accurately.

[0042] S3. Extract discharge parameters synchronized with historical images from the hydropower station monitoring system, wherein the discharge parameters include discharge volume. and sediment content (Sediment concentration), calculate sediment erosion volume The calculation formula is:

[0043]

[0044] The amount of sediment erosion is directly proportional to the sediment content and the discharge volume. This formula allows us to quantitatively understand the intensity of the erosion effect of sediment on the dam revetment under different discharge conditions. The amount of sediment erosion not only affects the rate and extent of erosion but is also closely related to the service life of the revetment structure. A larger amount of sediment erosion means that the revetment structure will suffer more severe wear in a shorter period of time, requiring more frequent maintenance and repair. Therefore, calculating the amount of sediment erosion is of great significance for assessing the erosion risk of the dam and formulating reasonable maintenance strategies.

[0045] S4. Construct and train an erosion prediction model, wherein the erosion prediction model uses the discharge volume of historical periods. and sediment erosion volume and tank structure parameters i As input, predict the erosion volume for different historical periods. For output;

[0046] The erosion prediction model adopts a temporal hybrid model architecture, including a spatial feature extraction module and a temporal feature fusion module;

[0047] The spatial feature extraction module uses a 3D convolutional neural network (3D-CNN) to process multi-temporal erosion volume data. 3D-CNN has the ability to extract features in three-dimensional space (which can be understood as a three-dimensional space composed of a time dimension and a two-dimensional spatial dimension). For erosion volume data, it performs convolution operations simultaneously on the temporal series and spatial distribution, enabling it to capture the changing characteristics of erosion volume at different times and spatial locations. For example, by sliding convolutional kernels of different sizes along the time-space dimension, it extracts feature patterns from local areas, thereby uncovering the evolutionary patterns of erosion volume over time and its spatial distribution characteristics.

[0048] The temporal feature fusion module employs a Transformer encoder to capture the long-term dependency between discharge volume and sediment erosion. Through its self-attention mechanism, the Transformer encoder considers information from all positions in the sequence, not just local information. In erosion prediction scenarios, discharge volume and sediment erosion may exhibit complex correlations at different times. For example, a large discharge volume in the early stages may lead to changes in sediment erosion later, thus affecting the erosion volume. The Transformer encoder can effectively learn this long-term, non-local dependency, fusing the temporal features of discharge volume and sediment erosion to provide more comprehensive information for the final erosion volume prediction.

[0049] Loss function of erosion prediction model Designed as follows:

[0050]

[0051] in, Mean square error, For robust loss function, and These are the weighting coefficients. Erosion volume calculated from historical images , The predicted erosion volume output by the erosion prediction model , It measures the average deviation between the predicted and the true values. By minimizing the MSE, the model's predicted values ​​can be made as close as possible to the true values, focusing on the overall prediction accuracy. The Huber loss function is less sensitive to outliers than the MSE loss function. When the error between the predicted and true values ​​is small, it resembles the MSE loss function; when the error is large, it resembles the absolute value loss function. This reduces the impact of outliers on model training, enabling the erosion prediction model to better handle outliers in the data while pursuing overall prediction accuracy, thus improving the model's robustness. Historical data is used to train the erosion prediction model, and by continuously adjusting the model parameters, the loss function value is optimized. Minimize, thereby improving the model's prediction accuracy.

[0052] The tank protection structure parameters i Including the angle of the tank α ,thickness d and concrete abrasion resistance coefficient k The angle of inclination of the protective tank α ,thickness d The concrete abrasion resistance coefficient can be obtained from the hydropower station's design and manufacturing company. kThe abrasion resistance coefficient can be obtained through experimental testing. Representative locations on the abutment structure are selected for on-site drilling and sampling. The retrieved concrete samples are used in experiments to simulate the scouring effect of water flow on the concrete. By measuring the abrasion amount of the concrete under certain scouring conditions, the abrasion resistance coefficient can be calculated. k .

[0053] S5. Input the predicted discharge volume for the future preset time period. and sediment content Calculate future erosion volume using an erosion prediction model The predicted discharge volume and sediment content Predictions are based on future rainfall, watershed evaporation, and historical discharge data;

[0054] Predicted discharge volume and sediment content Multiple factors need to be considered comprehensively. Future rainfall is a crucial natural factor affecting discharge volume, as the amount of rainfall directly determines the amount of water flowing into the hydropower station reservoir, thus influencing the discharge volume. Basin evaporation also impacts the reservoir's water balance from another perspective; higher evaporation rates may reduce the reservoir's storage capacity, indirectly affecting discharge volume. Furthermore, historical discharge data is of significant reference value. Analyzing historical data and predicting trends allows for a better understanding of future discharge patterns. Based on these factors, scientific meteorological forecasting models and hydrological analysis methods are employed to predict rainfall and basin evaporation for a predetermined period, and combined with historical discharge data, a comprehensive estimate of the predicted discharge volume is obtained. and sediment content ;

[0055] For example, suppose that according to meteorological forecasts, the rainfall in the basin where the hydropower station is located is expected to reach 200 mm in the next month, which is higher than the historical average of 150 mm for the same period. Basin evaporation also affects the water balance of the reservoir from another perspective; when evaporation is high, the reservoir's water storage may decrease, thus indirectly affecting the discharge. For instance, if the predicted evaporation for the next month is 80 mm, then historical discharge data is also of significant reference value. By analyzing historical data and predicting trends, we can better understand the patterns of future discharge. For example, the average discharge for the same period over the past five years is 10 million cubic meters. Based on these factors, using scientific meteorological forecasting and hydrological analysis methods, we can predict the rainfall and basin evaporation for the next preset period, and combine this with historical discharge data to comprehensively estimate the predicted discharge for the next month. ≈11 million cubic meters.

[0056] Regarding sediment content The prediction also needs to consider multiple factors. Sediment content is closely related to the geological conditions, vegetation cover, and rainfall intensity within the watershed. If geological disasters such as landslides have occurred in the watershed recently, or if vegetation cover has decreased due to natural disasters, it may lead to an increase in sediment content. Therefore, based on historical sediment content monitoring data, assuming a historical average level of 2.3 kg / m³, and after a landslide occurs, the predicted sediment content for the next month is as follows: =2.5 kg / m³.

[0057] S6. Based on future erosion volume Assess the risk of dam erosion;

[0058] Define the safe threshold for erosion volume Its design life is based on the dam revetment design life. Dynamic adjustment, the formula is:

[0059]

[0060] in, Overall safety erosion volume, The aging index of the tank structural materials is given in the formula. This factor takes into account the fact that, over time, due to material aging, the safe erosion volume that the tank can withstand will gradually decrease; This involves summing up the erosion volumes from various past periods, representing the total erosion the tank has experienced up to the current time t. minus This gives us the remaining safe erosion volume that the tank can withstand at the current time t. ;

[0061] when < At this point, the erosion is classified as low-risk, meaning that the erosion of the dam abutment is within acceptable limits for the predetermined period, and the abutment structure can maintain good stability and safety. However, even in a low-risk state, it is necessary to continuously monitor changes in erosion and conduct regular monitoring and inspections.

[0062] when ≥ When erosion is deemed high-risk, an emergency warning is issued, notifying relevant departments and personnel, including the hydropower station's management, engineering technicians, and safety monitoring personnel, to ensure they are aware of the serious erosion risks facing the abutment structure. On the other hand, an expert team should be quickly organized to conduct a comprehensive and detailed inspection and assessment of the abutment structure and develop an emergency reinforcement or repair plan to prevent further erosion from causing engineering safety accidents.

[0063] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A method for predicting erosion of hydropower station dams, characterized in that: Includes the following steps: S1. Obtain historical images of the spillway apron of the dam body at various historical periods, and preprocess the historical images; S2. Based on the preprocessed image, a deep learning instance segmentation model is used to identify the erosion area, and the erosion volume at each stage is calculated using the digital elevation model difference method. ; S3. Extract discharge parameters synchronized with historical images from the hydropower station monitoring system, wherein the discharge parameters include discharge volume. and sediment content Calculate the amount of sediment erosion. ; S4. Construct and train an erosion prediction model, wherein the erosion prediction model uses the discharge volume of historical periods. and sediment scour volume and tank structure parameters θ As input, predict the erosion volume for each historical period. For output; S5. Input the predicted discharge volume for the future preset time period. and sediment content Calculate future erosion volume using an erosion prediction model ; S6. Based on future erosion volume Assess the risk of dam erosion; In step S2, the erosion volume of each period The calculation formula is: in, For the first Pixels in historical images Elevation value, The pixel area; In step S3, the amount of sediment erosion is calculated. The formula is: ； In step S6, the step of determining the future erosion volume... Assessing the risk of dam erosion includes: Define the safe threshold for erosion volume Its design life is based on the dam revetment design life. Dynamic adjustment, the formula is: in, For the overall safe erosion volume, The aging index of the tank's structural materials; when < At that time, the erosion was judged to be of low risk; when ≥ At that time, the risk of erosion was determined to be high, and an emergency warning was issued.

2. The method for predicting erosion of a hydropower station dam body according to claim 1, characterized in that: In step S1, the historical images include at least two of the following: high-resolution satellite images, UAV oblique photography images, or ground three-dimensional laser scanning point clouds. The preprocessing of the historical images includes radiometric correction, geometric registration, and multi-temporal image matching.

3. The method for predicting erosion of a hydropower station dam body according to claim 1, characterized in that: In step S4, the erosion prediction model adopts a temporal hybrid model architecture, including a spatial feature extraction module and a temporal feature fusion module; The spatial feature extraction module processes multi-temporal erosion volume data based on a 3D convolutional neural network; The temporal feature fusion module uses a Transformer encoder to capture the long-term dependency between discharge volume and sediment scour volume.

4. The method for predicting erosion of a hydropower station dam body according to claim 1, characterized in that: In step S4, the loss function of the erosion prediction model Designed as follows: in, Mean square error, For robust loss function, and These are the weighting coefficients. Erosion volume calculated from historical images , The predicted erosion volume output by the erosion prediction model .

5. The method for predicting erosion of a hydropower station dam body according to claim 1, characterized in that: In step S4, the tank shield structure parameters θ Including the angle of the tank α ,thickness d and concrete impact resistance coefficient k .

6. The method for predicting erosion of a hydropower station dam body according to claim 1, characterized in that: In step S5, the predicted discharge volume and sediment content Predictions are based on future rainfall, watershed evaporation, and historical discharge data.

Images