CN-121616590-B - Bridge scouring form generation method based on generation countermeasure network

CN121616590BCN 121616590 BCN121616590 BCN 121616590BCN-121616590-B

Abstract

The invention discloses a bridge scour morphology generation method based on a generated countermeasure network, which relates to the technical field of image processing and comprises the following steps of S01, obtaining a plurality of groups of bridge scour pit sample data, S02, constructing a generated countermeasure network model which at least comprises a generator and a discriminator, S03, performing countermeasure training on the generated countermeasure network model by using a training data set to obtain an optimized countermeasure network model, S04, inputting a bridge pile mask image to be predicted and corresponding environmental parameters into the optimized countermeasure network model to obtain a predicted two-dimensional depth distribution thermodynamic diagram of scour pits around a bridge pile, S05, performing image processing on the two-dimensional depth distribution thermodynamic diagram output in the step S04, and extracting and generating an isopipe map of the scour pit. The invention converts bridge and water flow information into a two-dimensional scouring pit morphology graph and an equal depth line containing depth distribution and range by generating an countermeasure network.

Inventors

GAO HAIFENG
CHEN ZHENTAO
Zhao Shuaikang
GAN JIANJUN
MA BINJIE
WANG YICHEN
Ding Shaochao
CHU HONGTAO
HOU SHIHUA
XIE DONGFENG
XU WEILI
YAN YUHAN

Assignees

浙江省水利河口研究院(浙江省海洋规划设计研究院)

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (9)

1. The bridge scouring morphology generating method based on the generation countermeasure network is characterized by comprising the following steps of: S01, acquiring a plurality of groups of bridge flushing pit sample data, wherein each group of data comprises water flow working condition parameters, bridge pile geometric parameters, sediment parameters, time parameters and a real flushing pit two-dimensional morphology map corresponding to the sediment parameters; S02, constructing and generating an countermeasure network model, wherein the model at least comprises a generator and a discriminator, the generator is used for generating a scouring pit morphology graph according to input conditions, and the discriminator is used for discriminating the authenticity of the input morphology graph; S03, performing resistance training on the generated resistance network model by using the sample data to obtain an optimized resistance network model; S04, inputting the bridge pile mask image to be predicted and the corresponding environmental parameters into the optimized countermeasure network model to obtain a predicted two-dimensional depth distribution thermodynamic diagram of the scour pit around the bridge pile; s05, performing image processing on the two-dimensional depth distribution thermodynamic diagram output in the step S04, and extracting and generating an isodepth line diagram of the flushing pit; the input of the generator is made up of two parts: the binary image of the bridge pile geometric mask, wherein the pixel value of the binary image is 1 to identify the area occupied by the bridge pile, and the pixel value is 0 to identify the riverbed area; The environment parameter vector comprises flow velocity, flow rate, pile diameter, sediment grain diameter and scouring duration, the vector firstly performs dimension transformation and feature extraction through a full connection layer, and then reconstructs the result into a multichannel feature map consistent with the space size of the mask image through a replication filling mode.
2. The bridge scour pattern generation method according to claim 1, wherein said generator in step S02 employs a U-Net structure encoder-decoder architecture with a jump connection, comprising the steps of: the encoder is formed by serially connecting four-stage downsampling modules, each stage comprises a convolution layer with a convolution kernel of 4 multiplied by 4 and a step length of 2, and is sequentially provided with a batch normalization layer and a ReLU activation function, and the number of channels of the feature map is doubled and the size is halved through each stage, so that the image features from low stages to high stages are gradually extracted; The decoder and the encoder are symmetrical and formed by serially connecting four stages of up-sampling modules, each stage comprises a deconvolution layer with a convolution kernel of 4 multiplied by 4 and a step length of 2, and the deconvolution layer is also provided with batch normalization and ReLU activation functions, and the number of channels of the feature map is halved and the size is doubled through each stage, so that a high-resolution image is reconstructed step by step; The jump connection in the U-Net is used for splicing the feature map after downsampling of each stage of the encoder with the feature map before upsampling of the corresponding stage of the decoder in the channel dimension, so that the decoder can fuse fine space structure information reserved by the encoder when reconstructing images; the output layer of the generator uses a Tanh activation function to restrict the pixel value to the [ -1,1] interval, and uses the input mask to forcedly reset the output pixel value corresponding to the bridge pile area to 0.
3. The method according to claim 1, wherein in the step S02, the discriminator adopts a conditional PatchGAN architecture, and the input is formed by splicing three parts in the channel dimension, namely a flushing pit morphology diagram to be discriminated, a bridge pile mask image as one of the conditions, and a feature diagram obtained by expanding an environmental parameter vector through a full connection layer.
4. The bridge scour pattern generation method based on generation of an countermeasure network according to claim 1, wherein the step S03 of performing the countermeasure training on the generation of the countermeasure network model using the sample data includes the steps of: S31, dividing the sample data into a training set and a verification set according to a preset proportion; s32, applying data enhancement operation to the bridge pile mask image and the real scour pit two-dimensional morphological image in the training set; S33, initializing the network weights of the generator and the discriminator respectively, and configuring an optimizer and super parameters thereof; S34, in each training iteration, updating the discriminator firstly, and updating the generator, so that the model gradually converges through alternate optimization; s35, in each updating, calculating the loss of the discriminator and the loss of the generator respectively, and updating parameters of the corresponding network through a back propagation algorithm.
5. The bridge flushing morphology generating method based on generating an countermeasure network according to claim 1, wherein in the bridge flushing pit sample data in the step S01: the real scouring pit two-dimensional morphology graph is a gray scale thermodynamic diagram expressed by 512×512 pixel resolution, and the gray scale value of each pixel in the graph linearly corresponds to the scouring depth value of the river bed; the bridge pile mask is a single-channel binary image which is completely consistent with the size of the morphological image, wherein a pixel value 1 clearly indicates an area which is occupied by the cross section of the bridge pile and can not be flushed, and a pixel value 0 indicates a riverbed area which can be flushed; The sample data is not less than 1000 sets of image data, and the image data includes, but is not limited to, horseshoe, oval, pear, and inverted cone scour pits.
6. A method of generating a bridge scour pattern based on a generation countermeasure network according to claim 3, wherein the objective function of the arbiter when trained Is specifically defined as binary cross entropy mean loss based on a local discriminant matrix: ; Wherein, the For the purpose of the training to be desired, Indicating the desire for a distribution, For the input of the condition(s), In the form of a figure, the figure is a form, Representing the mean of the binary cross entropy loss over the Patch matrix, Representing the output of the arbiter(s), Is the output of the generator for the condition, Representing the random noise of the generator.
7. The bridge scour pattern generation method based on generation of an countermeasure network according to claim 1, wherein a specific image processing flow of generating the isocenter map in step S05 is as follows: s51, taking a two-dimensional depth distribution thermodynamic diagram output by a generator as an input source; s52, setting one or more depth thresholds with definite engineering significance, sequentially thresholding the thermodynamic diagram to generate a series of binary images, and identifying areas with depth exceeding a specific threshold by each binary image; S53, applying a Canny edge detection algorithm to each obtained binary image to accurately extract a region boundary contour line corresponding to the depth threshold; s54, a complete multi-level isocenter line graph is formed by superposing a plurality of contour lines extracted by different depth thresholds.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the bridge scour pattern generation method of any one of claims 1 to 7 based on generating an countermeasure network when the program is executed by the processor.
9. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the method for generating a bridge flushing pattern based on a generation countermeasure network according to any one of claims 1 to 7.

Description

Bridge scouring form generation method based on generation countermeasure network Technical Field The invention relates to the technical field of image processing, in particular to a bridge scouring form generating method based on a generated countermeasure network. Background Local scour around bridge foundations is a main cause of bridge collapse and is one of the main risks faced by bridge safety operation. The relevant statistics indicate that about 60% of bridge collapse accidents are closely related to foundation washout. In recent years, with the frequent occurrence of extreme weather events such as extreme floods caused by climate change, the problem of flushing a bridge foundation is increasingly remarkable, and the long-term safety of a bridge structure is seriously threatened. During the flushing process, the water flow forms a complex three-dimensional flow structure around the bridge piles, so that the local river bed is corroded and a flushing pit is formed. The form and depth of the flushing pit directly determine the stability of the bridge foundation, wherein the shape of the flushing pit determines the exposure range of the foundation, and the flushing pit is particularly critical to evaluating the transverse bearing capacity of the pile foundation. For a long time, the focus of attention in the industry and the prior art approaches have focused mainly on predicting the maximum flush depth value, often simplifying complex flush risk assessment into a single depth indicator. For example, chinese issued patent, publication No. CN116861821B, discloses a method for rapidly predicting the maximum scouring depth of a bridge foundation based on artificial intelligence, and Chinese issued patent, publication No. CN120373205A, discloses a system and a method for evaluating the local scouring of a bridge pile foundation based on the flow rate strain rate, which are used for predicting the maximum scouring depth of a bridge pile by machine learning or building a water-sand model. However, in practical engineering applications, it is difficult to fully evaluate the threat of flushing to bridge structures by grasping the maximum depth only, because the three-dimensional morphology of the flushing pit is also a key factor affecting the foundation stability and bearing capacity. At present, if the detailed form distribution of the scour pit is to be obtained, a physical model test (such as Chinese patent application publication No. CN115935602B, which discloses a general scour form calculation method of a cross-sea bridge taking turbulence effect into consideration) or high-precision numerical simulation calculation (such as Chinese patent application publication No. CN117726757A, which discloses a three-dimensional form real-time inversion method of the scour pit of a bridge pier under the action of reciprocating tide) is mainly relied on. Although the methods have certain accuracy, the methods have the limitations of high cost, long calculation period or serious dependence on measured data, and are difficult to meet the requirements of rapid and batch evaluation in engineering practice. Therefore, how to quickly and accurately generate the three-dimensional morphological distribution of the bridge scour pit under the conditions of known environment and structural parameters breaks through the bottleneck of the prior art, and becomes an important technical problem to be solved urgently in the current field. Disclosure of Invention Aiming at the technical problems, the technical scheme adopted by the invention is a bridge scouring form generating method based on generating an antagonism network, which comprises the following steps: S01, acquiring a plurality of groups of bridge flushing pit sample data, wherein each group of data comprises water flow working condition parameters, bridge pile geometric parameters, sediment parameters, time parameters and a real flushing pit two-dimensional morphology map corresponding to the sediment parameters; S02, constructing and generating an countermeasure network model, wherein the model at least comprises a generator and a discriminator, the generator is used for generating a scouring pit morphology graph according to input conditions, and the discriminator is used for discriminating the authenticity of the input morphology graph; S03, performing resistance training on the generated resistance network model by using the training data set to obtain an optimized resistance network model; S04, inputting the bridge pile mask image to be predicted and the corresponding environmental parameters into the optimized countermeasure network model to obtain a predicted two-dimensional depth distribution thermodynamic diagram of the scour pit around the bridge pile; and S05, performing image processing on the two-dimensional depth distribution thermodynamic diagram output in the step S04, and extracting and generating an isodepth line diagram of the flushing pit. Preferably, t