Dam body monitoring and early warning method and system

Abstract

The present application relates to the fields of artificial intelligence and dam monitoring and early warning, in particular to a dam monitoring and early warning method and system. The present application collects key parameters of the dam through sensors, obtains a continuous time sequence dam image set covering key parts of the dam surface, and processes the continuous time sequence dam image set. Through a preset dam damage early warning model, joint damage evaluation processing is performed to generate damage detection results. Based on historical data in the dam damage early warning model and a preset threshold, a dam early warning instruction is triggered and sent to the monitoring system to trigger emergency response operations. The present application can trigger a dam early warning instruction according to dam aging or environmental changes, is suitable for various dam types such as earth-rock dams and concrete dams, has the characteristics of flexible deployment and fast response, and can effectively reduce the false positive rate and the false negative rate, thereby improving the timeliness and effectiveness of dam monitoring and early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and dam monitoring and early warning, in particular to a dam monitoring and early warning method and system. BACKGROUND

[0002] In the construction and operation management of water conservancy projects, the safety of the dam, as an important water retaining structure, is directly related to the safety of people's lives and property and social stability downstream. At present, the traditional dam monitoring method mainly relies on manual inspection and data collection of some fixed sensors. Manual inspection is not only inefficient, but also greatly affected by human factors, making it difficult to conduct comprehensive and detailed inspection of the dam surface and easily miss some early minor damage. Although fixed sensors can collect some physical quantity data such as displacement and stress in real time, the number and position of the sensors are limited and cannot cover all key parts of the dam, and the monitoring capability for some non-structural surface damage (such as cracks and peeling) is insufficient. In addition, most of the existing monitoring methods can only provide single-dimensional data, lack comprehensive evaluation of the time sequence analysis of dam damage change information and the spatial structure stability, and are difficult to accurately predict the development trend of dam damage and to timely issue effective early warning information, which brings potential risks to the safe operation of the dam. SUMMARY

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present application, the present application provides a dam monitoring and early warning method. The present application acquires key parameters of the dam through sensors, obtains a continuous time sequence dam image set covering key parts of the dam surface, and processes the continuous time sequence dam image set to generate damage detection results through joint damage evaluation processing by a dam damage early warning model. Based on historical data in the dam damage early warning model and a preset threshold, the position migration trajectory of damage in the continuous dam image unit is analyzed based on the damage detection results, a damage spatio-temporal evolution model is constructed and damage spatio-temporal distribution feature information is generated, a dam early warning instruction is triggered, and is sent to the monitoring system to trigger emergency response operations. The present application can trigger a dam early warning instruction according to some non-structural surface damage (such as cracks and peeling) caused by dam aging or environmental changes, is suitable for various dam types such as earth-rock dams and concrete dams, has the characteristics of flexible deployment and fast response, and can effectively reduce the false positive rate and the false negative rate, thereby improving the timeliness and effectiveness of dam monitoring and early warning.

[0004] The present application provides a dam monitoring and early warning method, which comprises: acquiring a continuous time sequence dam image set covering key parts of the dam surface, the continuous time sequence dam image set being composed of multiple frames of dam image units arranged in time sequence, each frame of dam image unit containing pixel information of dam surface texture, structural boundary and potential damage area;

[0005] performing a feature extraction operation on the continuous time sequence dam body image set to obtain a time sequence damage feature sequence reflecting damage change information and a spatial structure feature set reflecting dam body structure stability;

[0006] inputting the time sequence damage feature sequence and the spatial structure feature set into a pre-trained dam body damage early warning model for joint damage evaluation processing to generate a damage detection result containing damage type identification and damage area position;

[0007] analyzing a position migration trajectory of damage in a continuous dam body image unit based on the damage detection result, constructing a damage space-time evolution model and generating damage space-time distribution feature information;

[0008] generating a dam body early warning instruction containing damage starting position, extension path and response strategy according to the damage type identification and the damage space-time distribution feature information, and sending the dam body early warning instruction to a monitoring system to trigger an emergency response operation.

[0009] In still another aspect, the embodiment of the present application also provides a dam body monitoring and early warning system, comprising a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for running the programs, instructions or codes in the machine readable storage medium to realize the above method.

[0010] Based on the above aspects, the embodiment of the present application obtains a continuous time sequence dam body image set covering key positions on the dam body surface, performs a feature extraction operation on the continuous time sequence dam body image set, simultaneously obtains a time sequence damage feature sequence reflecting damage change information and a spatial structure feature set reflecting dam body structure stability, fully characterizes the dam body from two dimensions of time and space, makes the evaluation of the dam body state more comprehensive and in-depth, inputs the two kinds of features into a pre-trained dam body damage early warning model for joint damage evaluation processing, uses the powerful learning and analysis capability of the dam body damage early warning model to more accurately generate a damage detection result containing damage type identification and damage area position, improves the accuracy and reliability of damage identification, constructs a damage space-time evolution model based on the damage detection result and generates damage space-time distribution feature information, further reveals the development law and trend of damage, finally generates a dam body early warning instruction containing damage starting position, extension path and response strategy according to the damage type identification and the damage space-time distribution feature information, and timely sends the dam body early warning instruction to a monitoring system to trigger an emergency response operation, thereby greatly improving the timeliness and effectiveness of dam body monitoring and early warning. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1is an execution flow schematic diagram of the dam body monitoring and early warning method provided by the embodiment of the present application.

[0012] Figure 2 is a hardware architecture schematic diagram of the dam body monitoring and early warning system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0013] The present application will be described in detail below with reference to the accompanying drawings of the specification, Figure 1 is a flow schematic diagram of the dam body monitoring and early warning method provided by an embodiment of the present application, and the dam body monitoring and early warning method will be described in detail below.

[0014] Step S110: Obtain a continuous time sequence dam body image set covering key parts of the dam body surface, which is composed of multiple frames of dam body image units arranged in time sequence, and each frame of dam body image unit contains pixel information of dam body surface texture, structure boundary and potential damage area.

[0015] Step S111: Obtain initial image data by regularly shooting key parts of the dam body surface through image acquisition devices deployed around the dam body.

[0016] In this step, image acquisition devices can be deployed at appropriate positions around the dam body, and the selection of these positions fully considers the ability to comprehensively and clearly cover key parts of the dam body surface. For example, image acquisition devices are arranged at key positions such as different sides and top of the dam body to ensure that the dam body is shot from multiple angles. At the same time, attention should be paid to avoid external interference of the devices, such as avoiding setting in positions that are easily eroded by wind and rain, blocked or damaged by other physical factors.

[0017] When the image acquisition devices are deployed, the shooting program is started according to the pre-set time interval. During the shooting process, the image acquisition devices automatically record the relevant information of each frame of image, including the shooting time and position of the image. With the passage of time, the continuously accumulated image data constitutes the initial image data set, which contains the appearance information of the dam body at different time points, but may have some problems, such as inaccurate time stamp and containing unnecessary background information, which need to be further processed.

[0018] Step S112: Perform time stamp correction processing on the initial image data, and perform spatial cropping processing on the initial image data after time stamp correction processing according to the spatial range of the key parts determined according to the dam body design drawings, to obtain cropped image data.

[0019] After obtaining the initial image data, due to various reasons such as device clock error, data transmission delay, etc., the timestamp of the image may be inaccurate. Therefore, it is necessary to perform timestamp correction processing on the initial image data. First, collect the clock calibration information of the image acquisition device and related time synchronization data, and find out the deviation of the timestamp by comparing and analyzing these information. Then, adjust the timestamp of each frame of image according to the deviation value, to ensure the accuracy of the time sequence of the image.

[0020] At the same time, according to the dam body design drawing, the spatial range of the key position is determined. The dam body design drawing records the structure of the dam body and the position information of each part in detail. By analyzing the dam body design drawing, the specific position and boundary of the key position on the surface of the dam body can be accurately determined. For example, the spatial coordinate range of the key position such as the upstream face, the downstream face and the dam top of the dam body is determined.

[0021] After completing the timestamp correction processing, the spatial cropping processing is performed on the corrected initial image data. According to the determined spatial range of the key position, the part containing the key position is cropped from each frame of image. In the specific operation, the cropping tool in the image processing software can be used to accurately crop according to the spatial coordinate range. The cropped image data only contains the information of the key position on the surface of the dam body, and removes unnecessary background and other interference information, which is beneficial to subsequent analysis and processing.

[0022] Step S113: performing illumination compensation processing on the cropped image data, and performing resolution uniformization processing on the illumination compensated image data to generate a plurality of dam body image units arranged in time sequence, wherein the plurality of dam body image units constitute a continuous time sequence dam body image set.

[0023] In this embodiment, the cropped image data may be affected by different illumination conditions, resulting in differences in brightness and contrast of the image. In order to eliminate these differences, illumination compensation processing needs to be performed on the cropped image data. First, analyze the illumination of each frame of image, which can be evaluated by calculating the average brightness, histogram, etc. of the image. Then, according to the evaluation result, select a suitable illumination compensation algorithm, such as histogram equalization, adaptive illumination compensation, etc. Taking histogram equalization as an example, it adjusts the pixel value distribution of the image to make the brightness of the image more uniform. In the specific operation, the pixel value of the image is remapped so that the pixel value is uniformly distributed in the entire gray scale range, thereby improving the contrast and clarity of the image.

[0024] After the illumination compensation processing is completed, the image data after illumination compensation also needs to be subjected to resolution unification processing. Since the resolutions of different image acquisition devices can be different, or the resolutions of the same device under different shooting conditions can also be different, in order to facilitate subsequent analysis and processing, the resolutions of all images need to be unified to a standard value. A suitable resolution can be selected as a target resolution, and then an image scaling algorithm is used to process the image. For example, a bilinear interpolation, bicubic interpolation or the like is used to scale the image, and the resolution of the image is adjusted to the target resolution.

[0025] After the illumination compensation and resolution unification processing, a plurality of dam body image units arranged in time sequence are obtained. These plurality of dam body image units contain pixel information of dam body surface texture, structural boundary and potential damage area, which collectively constitute a continuous time sequence dam body image set.

[0026] Step S120: performing a feature extraction operation on the continuous time sequence dam body image set to obtain a time sequence damage feature sequence reflecting damage change information and a spatial structure feature set reflecting dam body structure stability.

[0027] In this embodiment, after obtaining the continuous time sequence dam body image set, features valuable for dam body damage monitoring and early warning need to be extracted therefrom. Through the feature extraction operation, the image data can be converted into feature information more representative and analyzable.

[0028] Step S121: performing adjacent frame difference analysis processing on the continuous time sequence dam body image set to extract a pixel change area between each pair of adjacent dam body image units, the pixel change area corresponding to a potential damage development area on the dam body surface.

[0029] Step S1211: performing grayscale processing on the continuous time sequence dam body image set to convert each dam body image unit into a grayscale image, and performing pixel value point-by-point subtraction processing on the adjacent dam body image units after grayscale processing to generate a pixel difference image, a region with a non-zero pixel value in the pixel difference image being a pixel change area.

[0030] In the process of adjacent frame difference analysis, the gray processing is first performed on the continuous time sequence dam body image set. Since the color image contains the information of three channels of red, green and blue, the processing is relatively complex, while the gray image only contains the information of one channel, which can simplify the processing process. There are many methods for gray processing, and the common one is the weighted average method. The corresponding gray value is obtained by performing weighted average on the pixel values of the red, green and blue channels of each pixel in the color image. Different weighting coefficients can be adjusted according to actual needs to achieve different gray processing effects. For example, the standard weighting coefficients 0.299, 0.587 and 0.114 are used to perform weighted average on the red, green and blue channels, and the obtained gray image can well preserve the brightness information of the image.

[0031] After converting each frame of dam body image unit into a gray image, the pixel value point-by-point subtraction processing is performed on the adjacent dam body image units after gray processing. In the specific operation, the corresponding pixel values of the adjacent two frames of gray images are subtracted to obtain a new image, i.e. a pixel difference image. In the pixel difference image, the area with pixel value of zero indicates that the pixel values of the adjacent two frames of images at this position have not changed, while the area with non-zero pixel value indicates that there is pixel change. These pixel change areas may be caused by damage development of the dam surface, light change, object movement, etc. By analyzing the pixel difference image, the area where the damage of the dam surface may exist can be found out.

[0032] Step S1212: performing binaryzation processing on the pixel difference image, and setting a fixed threshold to convert the pixel difference image into a binary image, wherein the area with pixel value greater than the threshold is marked as a potential damage development area.

[0033] After obtaining the pixel difference image, in order to more clearly identify the potential damage development area, the binaryzation processing needs to be performed on the pixel difference image. The basic idea of binaryzation processing is to set a fixed threshold, compare the pixel value in the image with the threshold, and divide the pixels into two categories according to the comparison result. In the specific operation, each pixel in the pixel difference image is traversed, the pixel with pixel value greater than the threshold is marked as 1, indicating that the area is a potential damage development area; and the pixel with pixel value less than or equal to the threshold is marked as 0, indicating that the area is a non-damage area.

[0034] In this embodiment, the selection of the threshold needs to consider multiple factors, such as the noise level of the image, the characteristics of the damage area, etc. If the threshold is set too high, some real damage areas may be missed; if the threshold is set too low, some noise areas may be misjudged as damage areas. The appropriate threshold can be determined through experiments and experience, or some adaptive threshold algorithm can be used to automatically adjust the threshold according to the local features of the image.

[0035] Step S1213: Perform morphological closing operation processing on the binary image to connect adjacent small area difference regions and eliminate isolated noise points, and generate a continuous pixel change region mask.

[0036] After obtaining the binary image, since there may be some isolated noise points and discontinuous small area difference regions in the binary image, morphological closing operation processing needs to be performed on the binary image. Morphological closing operation is an image processing method based on morphological operation, which consists of two operations: dilation and erosion.

[0037] First, perform dilation operation on the binary image. The basic idea of dilation operation is to expand the target region in the image to connect adjacent small area difference regions. In specific operation, a structure element (such as a circle, square, etc.) is used to scan the binary image, and when the structure element overlaps with the target region in the image, the overlapping region is marked as the target region. Through dilation operation, adjacent small area difference regions can be connected into a larger region.

[0038] Then, perform erosion operation on the dilated image. The basic idea of erosion operation is to shrink the target region in the image to remove some unnecessary edges and noise. In specific operation, the same structure element is used to scan the dilated image, and when the structure element is completely contained in the target region, the center position of the structure element is marked as the target region. Through erosion operation, some isolated noise points and unnecessary edges can be removed, making the target region more clear and continuous.

[0039] After morphological closing operation processing, the obtained image is a continuous pixel change region mask. This mask accurately identifies the pixel change region between each pair of adjacent dam body image cells.

[0040] Step S1214: Perform region extraction processing on the original dam body image cells based on the pixel change region mask to obtain the pixel change region between each pair of adjacent dam body image cells, which is the overlapping part of the original image and the mask region.

[0041] After obtaining the continuous pixel change region mask, perform region extraction processing on the original dam body image cells based on the pixel change region mask. In specific operation, compare the pixel change region mask with the original dam body image cells pixel by pixel to find the overlapping part of the original image and the mask region. For the pixel position marked as 1 in the mask, keep the pixel value at the corresponding position in the original image; for the pixel position marked as 0 in the mask, set the pixel value at the corresponding position in the original image to 0. In this way, the pixel change region between each pair of adjacent dam body image cells is extracted.

[0042] In this embodiment, the extracted pixel change region contains the part of the dam surface that changes at adjacent time points, which may be caused by damage development. By further analyzing these pixel change regions, the development of dam damage can be understood.

[0043] Step S122: Perform morphological feature extraction processing on the pixel change region to obtain the area expansion, shape regularity, and edge complexity of the change region as the basic features of damage change.

[0044] After extracting the pixel change region, morphological feature extraction processing needs to be performed on these pixel change regions to obtain the basic features reflecting the damage change. Morphological features are important parameters for describing the shape and structure of an object. By analyzing the morphological features of the pixel change region, the development of damage can be understood.

[0045] First, the area expansion of the change region is calculated. The area expansion reflects the area change of the damage region at adjacent time points. By comparing the area size of the pixel change region in the adjacent two frames of images, the area expansion can be calculated. Specifically, the number of pixels in the pixel change region in the adjacent two frames of images is counted, the number of pixels is converted into the actual area value, then the difference between the area values in the two frames of images is calculated to obtain the area expansion. The larger the area expansion, the faster the expansion speed of the damage region.

[0046] Secondly, the shape regularity of the change region is calculated. Shape regularity describes whether the shape of the damage region is regular. For regular shapes such as circles, squares, etc., the shape regularity is higher; for irregular shapes, the shape regularity is lower. The shape regularity can be evaluated by calculating the relationship between the perimeter and the area of the change region. For example, shape factors such as circularity, rectangularity, etc. are used to measure the shape regularity. Circularity can be obtained by calculating the ratio of the area of the change region to the square of the perimeter. The closer the ratio is to 1, the closer the shape is to a circle, and the higher the shape regularity.

[0047] Finally, the edge complexity of the change region is calculated. Edge complexity reflects the complexity of the edge of the damage region. The more complex the edge, the more irregular the boundary of the damage region, which may mean that the development of damage is more complex. The edge complexity can be evaluated by calculating parameters such as edge length and edge curvature of the change region. For example, the number of edge pixels of the change region is counted as an approximate value of the edge length; the complexity of the edge is evaluated by calculating the curvature change of the edge pixels.

[0048] Step S123: Arrange the basic features of damage change in time sequence to form an initial time sequence damage feature sequence, and perform time window smoothing processing on the initial time sequence damage feature sequence to obtain a time sequence damage feature sequence after time window smoothing processing.

[0049] After obtaining the damage change base features, these damage change base features are arranged in chronological order to form an initial time sequence damage feature sequence. The initial time sequence damage feature sequence contains damage change base features at different time points, reflecting the changes of damage over time.

[0050] Since there may be some noise and fluctuations in the initial time sequence damage feature sequence, in order to better analyze the development trend of damage, time window smoothing processing needs to be performed on the initial time sequence damage feature sequence. The basic idea of time window smoothing processing is to average or weighted average the feature values within a set time window to reduce the influence of noise and fluctuations. In specific operation, a suitable time window size is selected, such as a window containing several time points. For each time point in the window, the average or weighted average of the feature values of the time point and its adjacent time points before and after it is calculated as the smoothed feature value of the time point. By traversing the entire initial time sequence damage feature sequence, each time point is smoothed to obtain a time sequence damage feature sequence after time window smoothing processing. The time sequence damage feature sequence after time window smoothing processing is more stable and can better reflect the development trend of damage.

[0051] Step S124: Perform structure boundary detection processing on the single-frame dam image unit to extract the offset amount between the dam surface design structure boundary line and the actual image boundary line as the structure offset feature.

[0052] In order to evaluate the structural stability of the dam, structure boundary detection processing needs to be performed on the single-frame dam image unit. The purpose of structure boundary detection is to find out the actual structure boundary line of the dam surface and compare it with the design structure boundary line in the dam design drawing to extract the offset amount between them as the structure offset feature.

[0053] Firstly, an edge detection algorithm is used to process the single-frame dam image unit to find out the edge information in the image. The edge detection algorithm can detect edges according to the gray level changes of the image. Common edge detection algorithms include Sobel operator, Canny operator, etc. Taking the Canny operator as an example, it detects edges through a multi-stage processing process, including Gaussian smoothing, gradient calculation, non-maximum suppression and double threshold processing steps. Through these steps, the actual structure boundary line of the dam surface can be accurately detected.

[0054] Then, the position of the design structure boundary line is determined according to the dam design drawing. The dam design drawing records the structure and boundary position of each part of the dam in detail. By matching and aligning the boundary information in the design drawing with the boundary information in the actual image, the position of the design structure boundary line in the image can be determined.

[0055] After determining the position of the design structure boundary line in the image, it is compared with the actual detected structure boundary line. To ensure the accuracy of the comparison, the spatial position of the two boundary lines needs to be accurately matched. The design structure boundary line and the actual structure boundary line can be aligned in the image coordinate system through image registration methods, so that they have the same reference position in space.

[0056] Then, the distance between the corresponding points of the design structure boundary line and the actual structure boundary line is calculated point by point. In the image, each boundary line is composed of a series of pixel points. For each point on the design structure boundary line, find its corresponding point on the actual structure boundary line (which can be done through nearest neighbor search or other methods), and then calculate the pixel distance between the two corresponding points. These distance values are used as a measure of the offset, forming an offset set.

[0057] In order to obtain a comprehensive structure offset feature, the offset set needs to be further processed. The average value of the offset set can be calculated as a representative value of the structure offset feature. The average value is calculated by adding all the distance values in the offset set and dividing by the number of offsets. This average value reflects the overall offset between the design structure boundary line and the actual structure boundary line.

[0058] At the same time, the standard deviation of the offset set can also be calculated. The standard deviation describes the dispersion of the offset, the larger the standard deviation, the greater the fluctuation of the offset, and the more unstable the structure boundary. The calculation process of the standard deviation is to first calculate the square of the difference between each distance value in the offset set and the average value, then add these square values and divide by the number of offsets, and finally take the square root to get the standard deviation.

[0059] By considering the average value and the standard deviation comprehensively, the offset between the design structure boundary line and the actual structure boundary line can be described more comprehensively, and a more accurate structure offset feature can be obtained. This structure offset feature is of great significance for evaluating the structural stability of the dam body. If the value of the structure offset feature is large, it means that there is a large deviation between the actual structure and the design structure of the dam body, which may pose a risk of structural instability.

[0060] Step S125: Perform texture consistency analysis on the single-frame dam image unit to calculate the texture gradient distribution difference between the normal region and the potential damage region of the dam surface as a texture anomaly feature.

[0061] To further evaluate the structural stability of the dam, texture consistency analysis is performed on the single-frame dam image unit. First, the normal region and the potential damage region of the dam surface need to be determined. The normal region can be determined according to the design information and historical image data of the dam, which is usually the region without obvious damage on the surface of the dam. The potential damage region can be determined by the pixel change region extracted in the previous step.

[0062] For the normal region and the potential damage region, their texture gradient distributions are calculated respectively. Texture gradient reflects the change of texture in the image, and can be obtained by calculating the gradient of pixel value in the image. Sobel operator and other methods can be used to calculate the gradient of pixel value. Sobel operator is a commonly used edge detection operator, which calculates the gradient of pixel value in horizontal and vertical directions by convolution operation on the image.

[0063] For the normal region, it is divided into multiple small sub-regions, and the statistical characteristics of texture gradient, such as mean value and standard deviation, are calculated in each sub-region. These statistical characteristics can reflect the distribution of texture gradient in the normal region. Similarly, the same processing is also performed on the potential damage region, which is divided into multiple sub-regions and the statistical characteristics of texture gradient in each sub-region are calculated.

[0064] Then, the texture gradient distributions of the normal region and the potential damage region are compared. The texture anomaly feature can be obtained by calculating the difference of the statistical characteristics of texture gradient between the two regions. For example, the difference of the mean value and the difference of the standard deviation of the texture gradient between the normal region and the potential damage region are calculated. The larger the difference is, the greater the difference of the texture gradient distribution between the normal region and the potential damage region is, which may mean that there is a texture anomaly in the potential damage region.

[0065] In addition to the difference of the mean value and the standard deviation, the difference between the histograms of the texture gradient distributions of the two regions can also be calculated. Histogram can more comprehensively describe the distribution of texture gradient, and by comparing the similarity of the two histograms, more accurate texture anomaly feature can be obtained. Histogram intersection method, Bhattacharyya distance and other methods can be used to calculate the difference between the two histograms. Histogram intersection method is to calculate the intersection area of two histograms in each interval. The larger the intersection area is, the more similar the two histograms are, and the smaller the texture anomaly feature value is; on the contrary, the larger the texture anomaly feature value is.

[0066] By considering the difference of the statistical characteristics of texture gradient and the difference of the histograms comprehensively, a more comprehensive and accurate texture anomaly feature can be obtained. This texture anomaly feature can be used to detect the potential damage on the surface of the dam. When the texture anomaly feature value is large, it means that there may be damage on the surface of the dam, which needs further attention.

[0067] Step S126: Feature concatenation is performed on the structure offset feature and the texture anomaly feature to generate a spatial structure feature set reflecting the stability of the dam structure.

[0068] After obtaining the structure offset feature and the texture anomaly feature, in order to more comprehensively reflect the structural stability of the dam, feature concatenation is needed to be performed on the structure offset feature and the texture anomaly feature to form a more representative feature set.

[0069] For example, the normalized structure offset feature and the texture anomaly feature can be spliced in a set order. For example, the structure offset feature can be placed in front and the texture anomaly feature can be placed in back to form a spatial structure feature set. The spatial structure feature set can be an important input for subsequent dam damage assessment. By analyzing the spatial structure feature set, the structural stability of the dam and the potential risk of damage can be more accurately determined.

[0070] Step S127: Time dimension alignment is performed on the time window smoothed time sequence of damage features and the spatial structure feature set so that the time sequence of damage features at each time point has the same time identifier as the spatial structure feature at the corresponding time point.

[0071] After obtaining the time window smoothed time sequence of damage features and the spatial structure feature set, in order to effectively combine them for joint damage assessment, time dimension alignment is needed.

[0072] First, the time range of the time sequence of damage features and the spatial structure feature set is determined. The time sequence of damage features is a series of damage feature values arranged in chronological order, each feature value corresponding to a specific time point; the spatial structure feature set is a feature set reflecting the stability of the dam structure collected at different time points. The common time range of the two data sets needs to be found, i.e. the time interval in which they both contain data.

[0073] Then, for each time point in the time sequence of damage features, the corresponding time point in the spatial structure feature set is found. Since the time points of the two data sets may not be completely consistent during actual data collection, there may be a certain time deviation. The nearest neighbor search method can be used to find the time point closest in time to each time point in the time sequence of damage features as the corresponding time point in the spatial structure feature set.

[0074] After finding the corresponding time point, the damage feature of the time point in the time sequence damage feature sequence is associated with the spatial structure feature of the corresponding time point in the spatial structure feature set. They can be spliced in a set order to form a time sequence damage feature and a spatial structure feature containing the time point, which can more comprehensively reflect the damage and structural stability of the dam at the time point.

[0075] By performing the above processing on each time point in the time sequence damage feature sequence, a new feature set is obtained, wherein each feature vector contains the time sequence damage feature and the spatial structure feature of the corresponding time point, and has the same time identifier.

[0076] Step S130: inputting the time sequence damage feature sequence and the spatial structure feature set into a pre-trained dam damage early warning model for joint damage evaluation processing to generate a damage detection result containing damage type identifier and damage area position.

[0077] After completing the time dimension alignment processing, the obtained time sequence damage feature sequence and spatial structure feature set are input into a pre-trained dam damage early warning model for joint damage evaluation processing. The dam damage early warning model is trained by a large amount of sample data, which can learn the internal relationship between the time sequence damage feature and the spatial structure feature and the dam damage, thereby realizing accurate evaluation of the dam damage.

[0078] First, the time sequence damage feature sequence is input into the time sequence dependence modeling module of the dam damage early warning model. The main function of this module is to capture the forward and backward damage change dependence relationship in the time sequence damage feature sequence, thereby generating a damage change feature vector with time sequence context information.

[0079] Step S131: inputting the time sequence damage feature sequence into the time sequence dependence modeling module of the dam damage early warning model, wherein the time sequence dependence modeling module comprises a bidirectional long short-term memory network layer, and the bidirectional long short-term memory network layer is used to capture the forward and backward damage change dependence relationship in the time sequence damage feature sequence, thereby generating a damage change feature vector with time sequence context information.

[0080] Bidirectional long short-term memory network (Bi-LSTM) is a special recurrent neural network that can handle long-term dependence in sequence data. In the dam damage early warning model, the bidirectional long short-term memory network layer of the time sequence dependence modeling module is composed of a forward LSTM unit and a backward LSTM unit.

[0081] When the time-series damage feature sequence is input into the bidirectional LSTM layer, the forward LSTM unit processes the data from the start of the sequence in time order, which remembers the information of the previous time points in the time-series damage feature sequence and passes the information to the subsequent time points. The backward LSTM unit processes the data from the end of the sequence in the opposite time order, which remembers the information of the subsequent time points in the time-series damage feature sequence.

[0082] At each time step, the forward LSTM unit and the backward LSTM unit output a hidden state vector respectively. The forward hidden state vector contains the forward dependency information from the start of the sequence to the current time point, and the backward hidden state vector contains the backward dependency information from the end of the sequence to the current time point.

[0083] The forward hidden state vector and the backward hidden state vector are spliced to obtain a comprehensive hidden state vector containing forward and backward dependency information. The comprehensive hidden state vector can reflect the dependency relationship between the damage change at the current time point and the damage changes at the previous and subsequent time points in the time-series damage feature sequence, thereby generating a damage change feature vector with time context information.

[0084] Through the processing of the bidirectional LSTM layer, the dynamic changes in the time-series damage feature sequence can be better understood, and the trends and patterns of damage development can be captured.

[0085] Step S132: input the spatial structure feature set into the spatial constraint analysis module of the dam damage early warning model, the spatial constraint analysis module contains a convolutional neural network layer, and the spatial correlation pattern of the structure offset feature and the texture abnormal feature in the spatial structure feature set is extracted through the convolutional neural network layer to generate a structure stability feature vector with spatial context information.

[0086] The spatial structure feature set is input into the spatial constraint analysis module of the dam damage early warning model. The convolutional neural network layer of the module can automatically learn the spatial correlation pattern between the structure offset feature and the texture abnormal feature in the spatial structure feature set.

[0087] The convolutional neural network (CNN) is composed of multiple convolutional layers, pooling layers and fully connected layers. When the spatial structure feature set is input into the convolutional neural network layer, the convolutional layer can use a set of convolutional kernels to perform sliding convolution operation on the input features. Each convolutional kernel extracts a specific local pattern in the input features, such as a certain combination pattern of the structure offset feature and the texture abnormal feature in space.

[0088] During the convolution operation, the convolution kernel will perform a weighted sum on the input features, and then perform a nonlinear transformation through an activation function (such as the ReLU function) to obtain a convolution feature map. Different convolution kernels will extract different local patterns, generating multiple convolution feature maps.

[0089] The pooling layer will perform a down-sampling operation on the convolution feature map, reducing the dimension of the feature map while preserving important feature information. Common pooling operations include max pooling and average pooling. Max pooling selects the maximum value in each pooling window as the output, while average pooling calculates the average value in each pooling window as the output.

[0090] After processing through multiple convolution layers and pooling layers, the obtained feature map is flattened into a one-dimensional vector, which is then input into the fully connected layer. The fully connected layer will perform further nonlinear transformation on the one-dimensional vector, mapping it to a new feature space and generating a structure-stable feature vector with spatial context information.

[0091] The structure-stable feature vector contains spatial correlation information between structure offset features and texture abnormal features in the spatial structure feature set, and can reflect the spatial distribution of the stability of the dam structure.

[0092] Step S133: input the damage change feature vector and the structure-stable feature vector into the feature fusion layer of the dam damage early warning model, assign dynamic weights to the damage change feature vector and the structure-stable feature vector based on the attention mechanism, generate a fusion feature vector through weighted fusion operation, and the dynamic weights are dynamically adjusted according to the contribution of damage change features and structure-stable features to damage assessment.

[0093] After obtaining the damage change feature vector and the structure-stable feature vector, they are input into the feature fusion layer of the dam damage early warning model. The main function of the feature fusion layer is to fuse the two feature vectors from different sources to more comprehensively evaluate the damage of the dam.

[0094] The attention mechanism plays an important role in the feature fusion process, which can dynamically assign weights to the two feature vectors according to their contribution to damage assessment. Specifically, the attention mechanism will calculate an attention score, which reflects the importance of each feature vector in the current time for damage assessment.

[0095] First, the damage change feature vector and the structural stability feature vector are input into an attention calculation module. The attention calculation module performs linear transformation on the two feature vectors, and then calculates the attention scores through an activation function (such as the softmax function). The softmax function converts the input scores into a probability distribution, so that the sum of all attention scores is 1.

[0096] According to the calculated attention scores, dynamic weights are assigned to the damage change feature vector and the structural stability feature vector. The damage change feature vector is multiplied by its corresponding attention score, and the structural stability feature vector is multiplied by its corresponding attention score. Then, the two weighted feature vectors are spliced to obtain the fusion feature vector.

[0097] Since the attention scores are dynamically adjusted according to the contribution of the damage change feature and the structural stability feature to the damage assessment, the weights of the damage change feature vector and the structural stability feature vector will be different in different situations. For example, when the damage change is severe, the weight of the damage change feature vector may increase; when the structural stability becomes the main concern, the weight of the structural stability feature vector may increase.

[0098] Through the above weighting fusion operation, the fusion feature vector can more effectively integrate the information of the damage change feature and the structural stability feature.

[0099] Step S134: input the fusion feature vector into the damage classification layer of the dam damage early warning model. The damage classification layer includes a fully connected neural network layer, which performs nonlinear transformation processing on the fusion feature vector through the fully connected neural network layer, and outputs a confidence distribution vector containing different damage types.

[0100] The fusion feature vector is input into the damage classification layer of the dam damage early warning model. The fully connected neural network layer of the damage classification layer is composed of multiple neurons, and each neuron is connected to all neurons of the previous layer.

[0101] When the fusion feature vector is input into the fully connected neural network layer, linear transformation is performed first. Each neuron performs weighted summation on each element of the input fusion feature vector, and the weights are learned during model training. Then, the weighted summation result is nonlinearly transformed through an activation function (such as sigmoid function, tanh function or ReLU function, etc.).

[0102] The fully connected neural network layer usually includes multiple hidden layers, and each hidden layer performs nonlinear transformation on the input to learn the complex patterns and features in the fusion feature vector. After processing by multiple hidden layers, the neurons in the last layer output a vector, and each element of the vector represents the confidence of different damage types.

[0103] For example, if there are multiple different damage types (such as cracks, leaks, erosion, etc.), each element of the output vector corresponds to the confidence of a damage type. The confidence value ranges from 0 to 1, and the closer the value is to 1, the greater the likelihood of the presence of the damage type; the closer the value is to 0, the less the likelihood of the presence of the damage type.

[0104] Through the nonlinear transformation processing of the fully connected neural network layer, the fusion feature vector can be mapped to the confidence distribution space of different damage types, providing a basis for subsequent damage type judgment.

[0105] Step S135: input the confidence distribution vector into the damage positioning layer of the dam damage early warning model. The damage positioning layer performs damage region pixel-level classification processing on each frame of dam image unit based on the spatial position information in the fusion feature vector, and generates a damage positioning map containing a damage region mask.

[0106] After obtaining the vector containing the confidence distribution of different damage types, it is input into the damage positioning layer of the dam damage early warning model. The main task of the damage positioning layer is to determine the specific location of the damage region in the dam image.

[0107] The damage positioning layer uses the spatial position information in the fusion feature vector to perform pixel-level classification processing of the damage region. The fusion feature vector not only contains information about the damage type, but also contains position information related to the spatial structure features.

[0108] In the damage positioning layer, the fusion feature vector is first processed to extract the spatial position features. Then, these spatial position features are used to analyze each pixel in each frame of dam image unit. For each pixel in the dam image unit, it is determined whether the pixel belongs to the damage region according to the fusion feature vector and the confidence distribution vector.

[0109] A threshold can be set to make the determination. If the damage type confidence of a pixel exceeds the threshold, the pixel is marked as a pixel in the damage region; otherwise, it is marked as a pixel in the non-damage region.

[0110] After the pixel-by-pixel classification processing, a binary image is obtained, in which the pixel value of the damage region is 1 and the pixel value of the non-damage region is 0. The binary image is the damage positioning map containing the damage region mask. The damage positioning map accurately identifies the specific location of the damage region in the dam image.

[0111] Step S136: associate and integrate the confidence distribution vector and the damage positioning map to generate a damage detection result containing damage type identification and corresponding damage region location.

[0112] After obtaining the confidence distribution vector and the damage localization map, they need to be associated and integrated to generate the final damage detection result.

[0113] First, for each damage area in the damage localization map, the corresponding damage type confidence is found in the confidence distribution vector according to its position information. Since the damage localization map identifies the specific location of the damage area, and the confidence distribution vector contains the confidence information of different damage types, the most likely damage type corresponding to each damage area can be found by the above association. In specific operation, each damage area in the damage localization map is traversed, and for each damage area, its position information is extracted, and then the corresponding element in the confidence distribution vector is determined according to the position information, and the damage type with the highest confidence is selected as the damage type identifier of the damage area.

[0114] Next, the damage type identifier is integrated with the position information of the damage area. A unique identifier can be assigned to each damage area, and then the identifier is associated with the corresponding damage type identifier and the specific position information of the damage area (such as pixel coordinate range, etc.). In this way, an information set containing damage type identifier and corresponding damage area position is formed.

[0115] During the integration process, the accuracy and consistency of the damage type identifier and the position information of the damage area need to be ensured. For position information, its accuracy in the image coordinate system needs to be ensured so that the damage area can be accurately located subsequently. For damage type identifier, reasonable judgment needs to be made according to the confidence distribution vector to avoid misjudgment.

[0116] Finally, the integrated information set is arranged and formatted to generate a damage detection result containing damage type identifier and corresponding damage area position. The damage detection result presents the type and position information of the dam damage.

[0117] Step S140: Analyzing the position migration trajectory of damage in the continuous dam image unit based on the damage detection result, constructing a damage spatio-temporal evolution model and generating damage spatio-temporal distribution feature information.

[0118] After obtaining the damage detection result, in order to deeply understand the development process and trend of dam damage, the position migration trajectory of damage in the continuous dam image unit needs to be analyzed based on the result, a damage spatio-temporal evolution model needs to be constructed, and damage spatio-temporal distribution feature information needs to be generated.

[0119] Step S141: Extracting the damage type identifier and the corresponding damage area mask of each frame of dam image unit from the damage detection result.

[0120] First, the damage type identifier and the corresponding damage area mask of each frame dam body image unit are extracted from the damage detection result. The damage detection result contains the type and location information of the damage in each frame image. The damage type identifier specifies the specific type of the damage, and the damage area mask accurately identifies the location of the damage area in the form of a binary image.

[0121] In specific operation, each record in the damage detection result is traversed, and for each frame dam body image unit, the corresponding damage type identifier and damage area mask are extracted therefrom. The extraction can be performed in time sequence of the images to ensure the continuity of the extracted information in time. The extracted damage type identifier and damage area mask are stored in different sets for subsequent processing and analysis.

[0122] Step S142: Perform centroid coordinate calculation processing on the damage area mask to obtain the centroid coordinate point of the damage area in each frame dam body image unit, which reflects the position of the damage in the image space.

[0123] After obtaining the damage area mask of each frame dam body image unit, centroid coordinate calculation processing is performed on these masks. The centroid coordinate is an important parameter that can represent the position of the damage area in the image space.

[0124] For each damage area mask, it is regarded as a two-dimensional binary image, where the pixel value of the damage area is 1 and the pixel value of the non-damage area is 0. The basic idea of calculating the centroid coordinate is to perform weighted average on the coordinates of all pixels in the damage area. The specific calculation process is as follows: first, traverse each pixel in the damage area mask, and for the pixel with a pixel value of 1, record its coordinate value. Then, sum the x and y coordinates of all damage area pixels respectively. Finally, divide the sum of the x coordinates by the number of damage area pixels to obtain the x coordinate of the centroid, and divide the sum of the y coordinates by the number of damage area pixels to obtain the y coordinate of the centroid. In this way, the centroid coordinate point of the damage area is obtained.

[0125] The above centroid coordinate calculation processing is performed on the damage area mask in each frame dam body image unit to obtain the centroid coordinate point of the damage area in each frame image. These centroid coordinate points reflect the position of the damage in the image space, and by analyzing the change of the centroid coordinate points over time, the position migration of the damage can be understood.

[0126] Step S143: Connect the centroid coordinate points of consecutive dam body image units in time sequence to generate a damage position migration trajectory, which is composed of a time identifier sequence and a corresponding centroid coordinate sequence.

[0127] After obtaining the centroid coordinate points of the damage area in each frame of dam image unit, the centroid coordinate points are connected in the time sequence of the images. Since each frame of image has a corresponding time identifier, the time identifier and the centroid coordinate point can be corresponded one by one.

[0128] In the specific operation, the centroid coordinate points of each frame of image are taken out in the time sequence, and are arranged in sequence to form a centroid coordinate sequence. At the same time, the corresponding time identifiers are also arranged in sequence to form a time identifier sequence. The two sequences are combined together to generate the damage position migration trajectory. The damage position migration trajectory clearly shows the position change of the damage in the continuous dam image unit, and through the analysis of the damage position migration trajectory, the damage migration in the space can be directly observed.

[0129] Step S144: performing displacement analysis processing on the damage position migration trajectory, and calculating the centroid coordinate displacement amount and the displacement direction between adjacent time identifiers as the damage extension speed information and the extension direction information.

[0130] After obtaining the damage position migration trajectory, displacement analysis processing is performed on the damage position migration trajectory to obtain the damage extension speed information and the extension direction information. The displacement analysis processing mainly calculates the change of the centroid coordinate between adjacent time identifiers.

[0131] For each two adjacent time identifiers and the corresponding centroid coordinate points in the damage position migration trajectory, the centroid coordinate displacement amount between them is calculated. The displacement amount is obtained by calculating the coordinate difference of the two centroid coordinate points in the x direction and the y direction. Specifically, the x coordinate of the latter centroid coordinate point is subtracted from the x coordinate of the former centroid coordinate point to obtain the displacement amount in the x direction; the y coordinate of the latter centroid coordinate point is subtracted from the y coordinate of the former centroid coordinate point to obtain the displacement amount in the y direction. Then, the total displacement amount can be calculated according to the displacement amounts in the x direction and the y direction.

[0132] At the same time, the displacement direction is determined according to the displacement amounts in the x direction and the y direction. The displacement direction can be determined by comparing the size and the sign of the displacement amounts in the x direction and the y direction, for example, if the displacement amount in the x direction is positive and the displacement amount in the y direction is positive, the displacement direction is approximately right up; if the displacement amount in the x direction is negative and the displacement amount in the y direction is positive, the displacement direction is approximately left up, and so on.

[0133] The calculated centroid coordinate displacement amount between adjacent time identifiers is taken as a measure of the damage extension speed information, and the greater the displacement amount, the faster the damage extension speed; and the displacement direction is taken as the damage extension direction information.

[0134] Step S145: Based on the damage location migration trajectory, the damage extension speed information and the damage extension direction information, a damage spatiotemporal evolution model is constructed.

[0135] After obtaining the damage location migration trajectory, the damage extension speed information and the damage extension direction information, a damage spatiotemporal evolution model is constructed based on these information. The damage spatiotemporal evolution model aims to describe the development and change law of damage in time and space.

[0136] Firstly, the damage location migration trajectory is taken as the basic data of the damage spatiotemporal evolution model, which provides the spatial position information of damage at different time points. The damage extension speed information and the damage extension direction information are used to describe the change between adjacent time points.

[0137] In constructing the damage spatiotemporal evolution model, a state transition-based idea can be used. The damage state at each time point (including position, speed and direction information) is taken as a state node, and the state transition relationship between adjacent time points is determined by the damage extension speed information and the damage extension direction information. By integrating the state nodes and state transition relationships of multiple time points, a model reflecting the spatiotemporal evolution law of damage is constructed.

[0138] Specifically, a recursive method can be used to describe the model. Assuming that the damage state (position, speed and direction) at a certain time is known, the damage position at the next time can be predicted according to the damage extension speed information and the damage extension direction information. By continuously recursing this process, the spatiotemporal evolution of damage in the future can be simulated.

[0139] In addition, some correction mechanisms can be introduced into the model to improve the accuracy of the damage spatiotemporal evolution model. For example, the model is updated and corrected in real time according to the new damage data actually monitored, so that the model can better adapt to the actual situation of damage development.

[0140] Step S146: The historical damage location migration trajectory is fitted by the damage spatiotemporal evolution model, the centroid coordinate change trend of the damage area in the future time window is predicted, and an extended trajectory sequence containing historical trajectory points and predicted trajectory points is generated.

[0141] After the damage spatiotemporal evolution model is constructed, the historical damage location migration trajectory is fitted using the model. The purpose of fitting is to let the model learn the law of historical damage development, so as to more accurately predict the future damage.

[0142] The time identifier sequence and the mass center coordinate sequence in the historical damage location migration trajectory are input into a time sequence prediction sub-model of the damage spatio-temporal evolution model. The time sequence prediction sub-model comprises an autoregressive moving average model structure, which can analyze the internal rules in the time sequence data. By fitting the historical data of the time identifier sequence and the mass center coordinate sequence, the sub-model learns the trend and pattern of the damage location change over time.

[0143] Meanwhile, the spatial coordinate sequence in the historical damage location migration trajectory is input into a spatial extension prediction sub-model of the damage spatio-temporal evolution model. The spatial extension prediction sub-model comprises a Kalman filter structure, which can perform noise filtering processing on the spatial coordinate sequence to extract the main direction and speed characteristics of the damage extension. Through processing of the spatial coordinate sequence, the spatial extension prediction sub-model can remove the interference of noise and more accurately grasp the extension trend of the damage in space.

[0144] The output results of the time sequence prediction sub-model and the spatial extension prediction sub-model are fused. The fusion processing can adopt a weighted splicing method, different weights are assigned according to the importance of the output results of the time sequence prediction sub-model and the spatial extension prediction sub-model, and then they are spliced in a set order. Through the fusion processing, a comprehensive prediction result is obtained, i.e., the predicted mass center coordinates corresponding to each time identifier in the future time window.

[0145] Finally, the historical mass center coordinate sequence and the predicted mass center coordinate sequence are combined to generate an extended trajectory sequence comprising historical trajectory points and predicted trajectory points. The extended trajectory sequence shows the complete spatio-temporal evolution of the damage from history to future.

[0146] Step S147: Extracting the minimum enclosing region in the extended trajectory sequence as the damage spatial coverage range and extracting the displacement change rate of adjacent trajectory points in the extended trajectory sequence as the damage extension acceleration information.

[0147] After obtaining the extended trajectory sequence, the minimum enclosing region in the extended trajectory sequence is extracted as the damage spatial coverage range. The minimum enclosing region is the smallest region that can completely contain all the trajectory points in the extended trajectory sequence.

[0148] In specific operation, a boundary search-based method can be used to determine the minimum enclosing region. The maximum and minimum x coordinates and the maximum and minimum y coordinates of all the trajectory points in the extended trajectory sequence are found, and a rectangular region is constructed with these coordinate values as boundaries. The rectangular region is the minimum enclosing region. The minimum enclosing region reflects the spatial coverage range of the damage, and by analyzing its size and change, the extension degree of the damage in space can be understood.

[0149] Meanwhile, the displacement amount change rate of adjacent track points in the extended track sequence is extracted as damage extension acceleration information. The displacement amount change rate describes the change of damage extension speed, i.e., whether the damage extension is accelerating or decelerating.

[0150] When calculating the displacement amount change rate of adjacent track points, first, the displacement amount between adjacent track points (calculated in the previous step) is calculated, and then the difference between adjacent displacement amounts is calculated. The difference is divided by the time interval of adjacent displacement amounts to obtain the displacement amount change rate. By performing the above calculation on all adjacent track points in the extended track sequence, a series of displacement amount change rates are obtained, which reflect the change of damage extension speed at different time periods, i.e., damage extension acceleration information.

[0151] Step S148: Perform information integration processing on the damage type identification, the damage spatial coverage range, the damage extension speed information, the damage extension direction information, and the damage extension acceleration information to generate damage spatiotemporal distribution feature information.

[0152] After obtaining the damage type identification, the damage spatial coverage range, the damage extension speed information, the damage extension direction information, and the damage extension acceleration information, these information are integrated to generate damage spatiotemporal distribution feature information.

[0153] First, a unique identifier is assigned to each damage area, and the identifier is associated with the damage type identification, the damage spatial coverage range, the damage extension speed information, the damage extension direction information, and the damage extension acceleration information. A data structure (such as a list or a dictionary) can be used to store these associated information.

[0154] During the integration process, the accuracy and consistency of the information need to be ensured. For the damage type identification, it needs to be ensured that it is consistent with the actual damage situation; for the damage spatial coverage range, the damage extension speed information, the damage extension direction information, and the damage extension acceleration information, it needs to be ensured that the calculation is accurate and reasonable.

[0155] All the associated information of the damage areas are sorted and formatted to generate a set containing damage spatiotemporal distribution feature information. The set presents the distribution characteristics of damage in time and space in a clear, easy-to-understand, and easy-to-use manner.

[0156] Step S150: Generate a dam pre-warning instruction containing the damage starting position, extension path, and response strategy according to the damage type identification and the damage spatiotemporal distribution feature information, and send the dam pre-warning instruction to the monitoring system to trigger emergency response operations.

[0157] After obtaining the damage type identification and the damage spatiotemporal distribution characteristic information, a dam body early warning instruction containing a damage starting position, an extension path and a response strategy is generated according to the information, and the instruction is sent to a monitoring system to trigger an emergency response operation.

[0158] Step S151: Extracting, from the damage spatiotemporal distribution characteristic information, a starting time identification of a damage position migration trajectory and a corresponding starting mass center coordinate as damage starting position information.

[0159] First, the starting time identification of a damage position migration trajectory and the corresponding starting mass center coordinate are extracted from the damage spatiotemporal distribution characteristic information. The damage position migration trajectory records the position change of the damage at different time points, and the starting time identification and the corresponding starting mass center coordinate represent the time and position at which the damage begins to appear.

[0160] In specific operation, the starting time point of the damage position migration trajectory is found in the damage spatiotemporal distribution characteristic information, and the time identification corresponding to the time point is extracted. At the same time, the mass center coordinate corresponding to the time point is extracted, which is the spatial coordinate of the damage starting position. The starting time identification and the starting mass center coordinate are combined together to form the damage starting position information. The damage starting position information clearly indicates the specific time and spatial position at which the damage begins to appear.

[0161] Step S152: Extracting, from the damage spatiotemporal distribution characteristic information, an extension trajectory sequence, connecting historical trajectory points and predicted trajectory points to form a damage extension path, and the extension path is composed of a sequence of continuous coordinate points.

[0162] The extension trajectory sequence is extracted from the damage spatiotemporal distribution characteristic information. The extension trajectory sequence contains historical damage position migration trajectory points and predicted future damage trajectory points, which completely show the development of the damage in time and space.

[0163] The historical trajectory points and the predicted trajectory points in the extension trajectory sequence are sequentially connected in time order to form a damage extension path. The damage extension path is composed of a series of continuous coordinate point sequences, and each coordinate point represents the spatial position of the damage at different time points. By analyzing the damage extension path, it can be seen directly how the damage extends from the starting position to other positions and the predicted development direction of the damage in the future.

[0164] Step S153: Analyzing a preset dam body damage response rule library, determining a corresponding response strategy according to the damage type identification, the damage extension speed information and the damage extension acceleration information, and the response strategy includes on-site verification, device debugging and emergency reinforcement.

[0165] Step S1531: Extracting a basic response strategy set associated with the damage type identifier from the dam damage response rule base, the basic response strategy set containing general response measures for different damage types.

[0166] In determining the response strategy, first, a basic response strategy set associated with the damage type identifier is extracted from the dam damage response rule base. The dam damage response rule base is a pre-constructed knowledge system containing general response measures for various possible damage types.

[0167] When the damage type identifier is obtained, a search and match are performed in the rule base. Each rule in the rule base is associated with a specific damage type. By comparing the damage type identifier with the damage type definition in the rule base, all matching rules are found. The response measures corresponding to these matching rules constitute the basic response strategy set.

[0168] For example, for a crack damage type identifier, the rule base may contain general response measures for different degrees of cracks. For example, for a slight crack, possible response measures include regular observation, surface repair, etc.; for a serious crack, possible response measures include internal reinforcement, restriction of dam use, etc. By extracting these general response measures from the rule base, a basic response strategy set associated with the crack damage type identifier is formed.

[0169] Step S1532: Filtering a candidate response strategy matching the current extension speed from the basic response strategy set according to the damage extension speed information, the emergency degree of the candidate response strategy being positively correlated with the extension speed.

[0170] After obtaining the basic response strategy set, a candidate response strategy matching the current extension speed is filtered from the set according to the damage extension speed information. The damage extension speed information reflects the speed of damage expansion in space.

[0171] Each response strategy in the basic response strategy set has its applicable damage extension speed range. For each response strategy, its applicable extension speed condition is analyzed. For example, some response strategies may be applicable to slow damage extension speed conditions, such as regular observation, simple repair, etc.; for other response strategies, they may be applicable to fast damage extension speed conditions, such as emergency reinforcement, immediate evacuation, etc.

[0172] The damage extension speed information is compared with the applicable extension speed range of each response strategy in the basic response strategy set. If the damage extension speed is within the applicable extension speed range of a response strategy, the response strategy is filtered out as a candidate response strategy.

[0173] The urgency of the candidate response strategy is positively correlated with the extension speed, that is, the faster the damage extension speed, the higher the urgency of the candidate response strategy. Therefore, in the screening process, the response strategy suitable for a higher extension speed can be preferentially selected. For example, if the damage extension speed is fast, the response strategy that needs to take immediate action will be preferentially screened out.

[0174] Step S1533: priority adjustment processing is performed on the candidate response strategies according to the damage extension acceleration information. If the extension acceleration information shows that the damage extension speed is increasing, the emergency priority of the candidate response strategy is increased.

[0175] After the candidate response strategies are obtained, priority adjustment processing is performed on the strategies according to the damage extension acceleration information. The damage extension acceleration information reflects the trend of the damage extension speed, that is, whether the damage extension speed is increasing, decreasing, or remaining unchanged.

[0176] For each candidate response strategy, there is a default priority at the beginning. The priority is adjusted according to the damage extension acceleration information. If the damage extension acceleration information shows that the damage extension speed is increasing, it means that the development of the damage is more dangerous, and the emergency priority of the candidate response strategy needs to be increased.

[0177] The specific priority adjustment method can be set according to actual conditions. For example, an adjustment coefficient can be used to modify the initial priority. If the damage extension speed increases greatly, the adjustment coefficient can be set to be larger; if the damage extension speed increases slightly, the adjustment coefficient can be set to be smaller. The initial priority is multiplied by the adjustment coefficient to obtain the adjusted priority.

[0178] By adjusting the priority of the candidate response strategy, the response strategy can more accurately adapt to the actual development of the damage. For example, the emergency priority of a candidate response strategy is originally medium, but due to the increasing damage extension speed, the emergency priority is increased to high after priority adjustment, so that in the subsequent selection of the final response strategy, this strategy is more likely to be selected.

[0179] During the priority adjustment process, the change of the damage extension acceleration information needs to be monitored in real time. If the damage extension acceleration information has changed, the priority of the candidate response strategy is re-adjusted in time to ensure the timeliness and effectiveness of the response strategy.

[0180] Step S1534: The strategy with the highest priority is selected from the adjusted candidate response strategies as the final response strategy, and the final response strategy contains specific operation instructions and responsibility subject information.

[0181] After the priority adjustment of the candidate response strategies, the highest priority strategy is selected from the adjusted candidate response strategies as the final response strategy.

[0182] First, the adjusted candidate response strategies are sorted according to priority. Bubble sort, quicksort, and other sorting algorithms can be used to arrange the candidate response strategies from high to low priority.

[0183] After sorting, the first strategy is selected as the final response strategy. The final response strategy contains specific operation instructions and responsibility subject information. The operation instructions clearly specify the specific measures to be taken in the face of the current damage situation, such as the specific content of on-site inspection, the specific steps of reinforcement construction, etc. The responsibility subject information clearly specifies who will execute these operation instructions, such as the names and responsibilities of the inspection personnel, the information of the reinforcement construction team, etc.

[0184] When selecting the final response strategy, its feasibility and effectiveness need to be ensured. The operation instructions of the final response strategy are evaluated in detail to check whether they meet the actual situation of the dam body and the engineering requirements. At the same time, it is ensured that the responsibility subject has the ability and resources to execute the operation instructions.

[0185] After determining the final response strategy, it is associated with the damage starting position information, damage extension path, etc. to generate accurate and effective dam pre-warning instructions. And the final response strategy needs to be conveyed to the relevant responsibility subject in time to ensure that the emergency response operation can be quickly and accurately carried out.

[0186] Step S154: Perform information association processing on the damage starting position information, the damage extension path, and the response strategy to generate intermediate pre-warning data containing a damage positioning coordinate chain with time identifiers aligned. Each coordinate node in the damage positioning coordinate chain contains a corresponding time identifier, spatial coordinates, and a response strategy identifier.

[0187] After obtaining the damage starting position information, the damage extension path, and the response strategy, these information are associated to generate intermediate pre-warning data.

[0188] First, assign a corresponding time identifier to each coordinate point in the damage extension path, so that it matches the time identifier in the damage starting position information. Then, associate each coordinate point with the corresponding response strategy identifier. In this way, each coordinate point in the damage extension path forms a coordinate node containing a time identifier, spatial coordinates, and a response strategy identifier.

[0189] These coordinate nodes are arranged in time sequence to form a damage positioning coordinate chain. The damage positioning coordinate chain completely records the position of the damage at different time points and the corresponding response strategy, providing detailed information for subsequent pre-warning and processing.

[0190] The damage starting position information is integrated with the damage positioning coordinate chain to form intermediate warning data containing time mark aligned damage positioning coordinate chain. The intermediate warning data presents the starting position, extension path of the damage and corresponding response strategy in a structured manner.

[0191] Step S155: The intermediate warning data is subjected to format standardization processing and converted into a communication protocol format recognizable by the monitoring system to generate dam warning instructions containing damage starting position, extension path and response strategy.

[0192] After obtaining the intermediate warning data, it is subjected to format standardization processing and converted into a communication protocol format recognizable by the monitoring system.

[0193] Different monitoring systems may use different communication protocols, so the intermediate warning data needs to be converted according to the specific requirements of the monitoring system. First, determine the communication protocol format supported by the monitoring system and understand the specific requirements of the format, such as the definition of data fields, data encoding method, etc.

[0194] Then, the intermediate warning data is reorganized and encoded according to the requirements of the communication protocol format. For example, the damage starting position information, coordinate node information in the damage extension path and response strategy identification are rearranged and combined according to the fields specified in the communication protocol. At the same time, the data is encoded, such as using a set of character encoding (such as UTF-8) to convert the data into a format suitable for transmission.

[0195] During the format standardization processing, the integrity and accuracy of the data need to be ensured. Check whether each data field contains the necessary information and whether the format of the information meets the requirements of the communication protocol. For some possible data missing or format errors, corresponding supplement and correction need to be made.

[0196] After format standardization processing, dam warning instructions containing damage starting position, extension path and response strategy are generated. The dam warning instructions conform to the communication protocol format of the monitoring system and can be accurately recognized and processed by the monitoring system.

[0197] Step S156: The dam warning instructions are sent to the monitoring system through a pre-set communication interface, so that the monitoring system triggers corresponding emergency response operations according to the response strategy identification, including starting on-site inspection, adjusting monitoring equipment parameters and dispatching reinforcement resources.

[0198] After generating the dam early warning instruction, it is sent to the monitoring system through the preset communication interface. The preset communication interface is set in advance according to the communication requirements and characteristics of the monitoring system, which can be a wired communication interface (such as an Ethernet interface) or a wireless communication interface (such as a Wi-Fi interface, a 4G / 5G interface, etc.).

[0199] Before sending the dam early warning instruction, it is necessary to ensure that the communication interface is in normal working state and a stable communication connection is established with the monitoring system. Check whether the parameter settings of the communication interface are correct, such as communication protocol, baud rate, IP address, etc.

[0200] When the communication interface is ready, the dam early warning instruction is packaged and transmitted according to the requirements of the communication protocol. During transmission, attention should be paid to the security and reliability of the data. Encryption technology can be used to encrypt the dam early warning instruction to prevent data from being stolen or tampered with during transmission. At the same time, retransmission mechanism and error checking mechanism are adopted to ensure that the data can be accurately transmitted to the monitoring system.

[0201] When the monitoring system receives the dam early warning instruction, it parses the response strategy identifier in it. The monitoring system triggers the corresponding emergency response operation according to the response strategy identifier. If the response strategy identifier indicates that on-site inspection needs to be started, the monitoring system will send task instructions to the relevant inspection personnel and arrange them to go to the dam site for detailed inspection to confirm the actual situation of the damage. If the response strategy identifier indicates that the monitoring equipment parameters need to be adjusted, the monitoring system will automatically adjust the parameters of the relevant monitoring equipment, such as increasing the monitoring frequency, adjusting the monitoring range, etc., so as to obtain the status information of the dam more timely and accurately. If the response strategy identifier indicates that reinforcement resources need to be dispatched, the monitoring system will send a request to the resource management department to dispatch the corresponding reinforcement materials and equipment to carry out emergency reinforcement of the dam to prevent the damage from further expanding.

[0202] Obviously, the embodiment of the present application collects the key parameters of the dam body through the sensor, obtains a continuous time sequence dam body image set covering the key positions on the dam body surface, and generates a damage detection result through joint damage evaluation processing of the continuous time sequence dam body image set, the dam body damage early warning model, and the like. Based on the historical data and the preset threshold in the dam body damage early warning model, the position migration track of the damage in the continuous dam body image unit is analyzed based on the damage detection result, the damage spatiotemporal evolution model is constructed, and the damage spatiotemporal distribution feature information is generated. The dam body early warning instruction is triggered, and is sent to the monitoring system to trigger the emergency response operation. The embodiment of the present application can trigger the dam body early warning instruction according to some non-structural surface damages (such as cracks, peeling, etc.) caused by dam body aging or environmental changes, is suitable for various dam types such as earth-rock dams and concrete dams, has the characteristics of flexible deployment and fast response, and can effectively reduce the false positive rate and the false negative rate, thereby improving the timeliness and effectiveness of dam body monitoring and early warning.

[0203] Figure 2 The hardware structure of the dam body monitoring and early warning system 100 provided by the embodiment of the present application for implementing the dam body monitoring and early warning method is shown. Figure 2 As shown in the figure, the dam body monitoring and early warning system 100 can include a processor 110, a machine readable storage medium 120, a bus 130, and a communication unit 140.

[0204] In a possible design, the dam body monitoring and early warning system 100 can be a single server or a server group. The server group can be centralized or distributed (for example, the dam body monitoring and early warning system 100 can be a distributed system). In some embodiments, the dam body monitoring and early warning system 100 can be local or remote. For example, the dam body monitoring and early warning system 100 can access information and / or data stored in the machine readable storage medium 120 via a network. For another example, the dam body monitoring and early warning system 100 can be directly connected to the machine readable storage medium 120 to access the stored information and / or data. In some embodiments, the dam body monitoring and early warning system 100 can be implemented on a dam body monitoring and early warning system. For example only, the dam body monitoring and early warning system can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any aggregation thereof.

[0205] The machine readable storage medium 120 can store data and / or instructions. In some embodiments, the machine readable storage medium 120 can store data obtained from an external terminal. In some embodiments, the machine readable storage medium 120 can store data and / or instructions used by the dam body monitoring and early warning system 100 to perform or use to complete the exemplary methods described in the present application.

[0206] In the implementation process, the one or more processors 110 execute the computer executable instructions stored in the machine readable storage medium 120, so that the processor 110 can perform the dam monitoring and early warning method as in the above method embodiments. The processor 110, the machine readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the transceiving action of the communication unit 140.

[0207] The specific implementation process of the processor 110 can refer to the above various method embodiments executed by the dam monitoring and early warning system 100, which has similar implementation principles and technical effects, and will not be described here.

[0208] In addition, the embodiment of the present application also provides a readable storage medium, wherein computer executable instructions are set in the readable storage medium, and when the processor executes the computer executable instructions, the dam monitoring and early warning method as above is realized.

[0209] It should be noted that, in order to simplify the description of the present application and help understand one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, sometimes multiple features are combined into one embodiment, figure or description thereof. Similarly, it should be noted that, in order to simplify the description of the present application and help understand one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, sometimes multiple features are combined into one embodiment, figure or description thereof.

Claims

1. A dam monitoring and early warning method, characterized in that, The method comprises: acquiring a continuous time sequence dam body image set covering key positions of a dam body surface, the continuous time sequence dam body image set being composed of multiple frames of dam body image units arranged in time sequence, each frame of dam body image unit containing pixel information of a dam body surface texture, a structure boundary and a potential damage area; performing a feature extraction operation on the continuous time sequence dam body image set to obtain a time sequence damage feature sequence reflecting damage change information and a spatial structure feature set reflecting dam body structure stability; inputting the time sequence damage feature sequence and the spatial structure feature set into a pre-trained dam body damage early warning model for joint damage evaluation processing to generate a damage detection result containing a damage type identifier and a damage area position; analyzing a position migration trajectory of damage in continuous dam body image units based on the damage detection result, constructing a damage space-time evolution model and generating damage space-time distribution feature information; generating a dam body early warning instruction containing a damage starting position, an extension path and a response strategy according to the damage type identifier and the damage space-time distribution feature information, and sending the dam body early warning instruction to a monitoring system to trigger an emergency response operation; the inputting the time sequence damage feature sequence and the spatial structure feature set into a pre-trained dam body damage early warning model for joint damage evaluation processing to generate a damage detection result containing a damage type identifier and a damage area position comprises: inputting the time sequence damage feature sequence into a time sequence dependent modeling module of the dam body damage early warning model, the time sequence dependent modeling module containing a bidirectional long short-term memory network layer, through which a forward and backward damage change dependent relationship in the time sequence damage feature sequence is captured to generate a damage change feature vector with time sequence context information; inputting the spatial structure feature set into a spatial constraint analysis module of the dam body damage early warning model, the spatial constraint analysis module containing a convolutional neural network layer, through which a spatial correlation pattern of structure offset features and texture abnormal features in the spatial structure feature set is extracted to generate a structure stability feature vector with spatial context information; inputting the damage change feature vector and the structure stability feature vector into a feature fusion layer of the dam body damage early warning model, dynamically weighting the damage change feature vector and the structure stability feature vector based on an attention mechanism, generating a fusion feature vector through a weighted fusion operation, and dynamically adjusting the dynamic weight according to the contribution of the damage change feature and the structure stability feature to damage evaluation; inputting the fusion feature vector into a damage classification layer of the dam body damage early warning model, the damage classification layer containing a fully connected neural network layer, through which the fusion feature vector is subjected to nonlinear transformation processing, and a confidence distribution vector containing different damage types is output; inputting the confidence distribution vector into a damage positioning layer of the dam body damage early warning model, the damage positioning layer performing damage area pixel-level classification processing on each frame of dam body image unit based on spatial position information in the fusion feature vector to generate a damage positioning map containing a damage area mask; The confidence distribution vector and the damage location map are associated and integrated to generate a damage detection result containing damage type identification and corresponding damage area location; Based on the damage detection result, the position migration trajectory of the damage in the continuous dam image unit is analyzed, a damage spatio-temporal evolution model is constructed, and damage spatio-temporal distribution feature information is generated, including: From the damage detection result, the damage type identification and the corresponding damage area mask of each frame of dam image unit are extracted; The damage area mask is processed by centroid coordinate calculation to obtain the centroid coordinate point of the damage area in each frame of dam image unit, and the centroid coordinate point reflects the position of the damage in the image space; The centroid coordinate points of the continuous dam image unit are connected in time sequence to generate a damage position migration trajectory, which is composed of a time identification sequence and a corresponding centroid coordinate sequence; The damage position migration trajectory is analyzed by displacement analysis to calculate the centroid coordinate displacement and displacement direction between adjacent time identifications as damage extension speed information and extension direction information; Based on the damage position migration trajectory, the damage extension speed information and the damage extension direction information, a damage spatio-temporal evolution model is constructed; The historical damage position migration trajectory is fitted by the damage spatio-temporal evolution model to predict the centroid coordinate change trend of the damage area in the future time window, and an extended trajectory sequence containing historical trajectory points and predicted trajectory points is generated; The minimum enclosing region in the extended trajectory sequence is extracted as the damage spatial coverage range, and the displacement change rate of adjacent trajectory points in the extended trajectory sequence is extracted as the damage extension acceleration information; The damage type identification, the damage spatial coverage range, the damage extension speed information, the damage extension direction information and the damage extension acceleration information are integrated to generate damage spatio-temporal distribution feature information.

2. The dam monitoring and early warning method according to claim 1, characterized in that, The continuous time sequence dam image set covering the key parts of the dam surface is obtained, including: The key parts of the dam surface are photographed by the image acquisition device deployed around the dam to obtain initial image data; The initial image data is processed by timestamp correction, and the spatial range of the key parts is determined according to the dam design drawings. The initial image data after timestamp correction is processed by spatial cropping to obtain cropped image data; The cropped image data is processed by illumination compensation, and the image data after illumination compensation is processed by resolution uniformity to generate a plurality of dam image units arranged in time sequence, which constitute a continuous time sequence dam image set.

3. The dam monitoring and early warning method according to claim 1, characterized in that, The continuous time sequence dam image set is processed by feature extraction to obtain a time sequence damage feature sequence reflecting damage change information and a spatial structure feature set reflecting dam structure stability, including: The continuous time sequence dam image set is processed by adjacent frame difference analysis to extract the pixel change area between each pair of adjacent dam image units, which corresponds to the potential damage development area of the dam surface; extracting a change area expansion, shape regularity and edge complexity of the pixel change area as basic damage change features; arranging the basic damage change features in time sequence to form an initial time sequence damage feature sequence, and performing time window smoothing processing on the initial time sequence damage feature sequence to obtain a time sequence damage feature sequence after time window smoothing processing; performing structure boundary detection processing on a single frame dam image unit to extract a displacement between a design structure boundary line and an actual image boundary line on the dam surface as a structure displacement feature; performing texture consistency analysis processing on the single frame dam image unit to calculate a texture gradient distribution difference between a normal area and a potential damage area on the dam surface as a texture abnormality feature; performing feature concatenation processing on the structure displacement feature and the texture abnormality feature to generate a spatial structure feature set reflecting the stability of the dam structure; performing time dimension alignment processing on the time sequence damage feature sequence after time window smoothing processing and the spatial structure feature set, so that the time sequence damage feature at each time point and the spatial structure feature at the corresponding time point have the same time identifier.

4. The dam monitoring and early warning method according to claim 3, characterized in that, The adjacent frame difference analysis processing on the continuous time sequence dam image set includes: performing grayscale processing on the continuous time sequence dam image set to convert each frame dam image unit into a grayscale image, and performing pixel value point-by-point subtraction processing on the grayscale adjacent dam image units to generate a pixel difference image, wherein the area with non-zero pixel value in the pixel difference image is a pixel change area; performing binaryzation processing on the pixel difference image to convert the pixel difference image into a binary image by setting a fixed threshold, wherein the area with a pixel value greater than the threshold is marked as a potential damage development area; performing morphological closing operation processing on the binary image to connect adjacent small area difference areas and eliminate isolated noise points to generate a continuous pixel change area mask; performing area extraction processing on the original dam image unit based on the pixel change area mask to obtain the pixel change area between each pair of adjacent dam image units, wherein the pixel change area is the overlapping part of the original image and the mask area.

5. The dam monitoring and early warning method according to claim 1, characterized in that, The fitting processing on the historical damage position migration trajectory by the damage spatio-temporal evolution model includes: inputting a time identifier sequence and a mass center coordinate sequence in the historical damage position migration trajectory into a time sequence prediction sub-model of the damage spatio-temporal evolution model, wherein the time sequence prediction sub-model comprises an autoregressive moving average model structure; performing historical data fitting processing on the time identifier sequence and the mass center coordinate sequence by the time sequence prediction sub-model to learn the internal law of the change of the damage position with time; inputting a spatial coordinate sequence in the historical damage position migration trajectory into a spatial extension prediction sub-model of the damage spatio-temporal evolution model, wherein the spatial extension prediction sub-model comprises a Kalman filter structure; The spatial extension prediction sub-model is used for noise filtering processing on the spatial coordinate sequence, and main direction and speed characteristics of damage extension are extracted; The output result of the time sequence prediction sub-model is fused with the output result of the spatial extension prediction sub-model, and a predicted mass center coordinate corresponding to each time identifier in a future time window is generated; The historical mass center coordinate sequence and the predicted mass center coordinate sequence are combined to generate an extended trajectory sequence containing historical trajectory points and predicted trajectory points.

6. The dam monitoring and early warning method according to claim 1, characterized in that, The dam early warning instruction containing the damage starting position, the extension path and the response strategy is generated according to the damage type identifier and the damage spatiotemporal distribution characteristic information, and the dam early warning instruction is sent to the monitoring system to trigger an emergency response operation, including: The starting time identifier and the corresponding starting mass center coordinate of the damage position migration trajectory are extracted from the damage spatiotemporal distribution characteristic information as damage starting position information; The extended trajectory sequence is extracted from the damage spatiotemporal distribution characteristic information, and the historical trajectory points and the predicted trajectory points are connected to form a damage extension path, and the extension path is composed of a continuous coordinate point sequence; A preset dam damage response rule library is analyzed, and a corresponding response strategy is determined according to the damage type identifier, the damage extension speed information and the damage extension acceleration information, and the response strategy includes on-site verification, device debugging and emergency reinforcement; The damage starting position information, the damage extension path and the response strategy are information-associated to generate intermediate early warning data containing a damage positioning coordinate chain aligned with time identifiers, and each coordinate node in the damage positioning coordinate chain contains a corresponding time identifier, a spatial coordinate and a response strategy identifier; The intermediate early warning data is format-standardized to convert into a communication protocol format recognizable by the monitoring system, and the dam early warning instruction containing the damage starting position, the extension path and the response strategy is generated; The dam early warning instruction is sent to the monitoring system through a preset communication interface, so that the monitoring system triggers a corresponding emergency response operation according to the response strategy identifier, and the emergency response operation includes starting on-site inspection, adjusting monitoring device parameters and scheduling reinforcement resources.

7. The dam monitoring and early warning method according to claim 6, characterized in that, The preset dam damage response rule library is analyzed, and a corresponding response strategy is determined according to the damage type identifier, the damage extension speed information and the damage extension acceleration information, including: A basic response strategy set associated with the damage type identifier is extracted from the dam damage response rule library, and the basic response strategy set contains general response measures for different damage types; A candidate response strategy matching the current extension speed is selected from the basic response strategy set according to the damage extension speed information, and the emergency degree of the candidate response strategy is positively correlated with the extension speed; The priority of the candidate response strategy is adjusted according to the damage extension acceleration information, and if the damage extension acceleration information shows that the damage extension speed is increasing, the emergency priority of the candidate response strategy is increased. Select the highest priority strategy from the adjusted candidate response strategies as a final response strategy, the final response strategy containing specific operation instructions and responsibility subject information.

8. A dam monitoring and early warning system, characterized by, The dam monitoring and early warning system comprises a processor and a memory, the memory and the processor are connected, the memory is used for storing programs, instructions or codes, and the processor is used for running the programs, instructions or codes in the memory to realize the dam monitoring and early warning method in any one of claims 1-7.

Images