Risk assessment method and system for hydraulic structures based on deep learning and digital twins

Abstract

The present invention discloses a hydraulic structure risk assessment method and system based on deep learning and digital twins, comprising: firstly, constructing a target three-dimensional hydraulic structure model by acquiring hydraulic structure parameters, and using a key module identification model and a module type identification model configured on a digital twin platform for analysis, determining the current hydraulic structure module, and then performing abnormal identification, and finally obtaining a risk assessment result of the hydraulic structure. This design realizes intelligent and precise risk assessment of hydraulic structures, and improves assessment efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the field of hydraulic structures, and in particular to a hydraulic structure risk assessment method and system based on deep learning and digital twins. Background Art

[0002] In recent years, judging the overall safety status of hydraulic structures by examining the changes in deformation energy of hydraulic structures under various external and internal factors has become a hot topic in research. Exploring the integration of traditional hydraulic structure safety and stability assessment with information and intelligent technology to achieve comprehensive and accurate structural risk assessment and enhance the readability and scientificity of assessment results has become one of the important needs for the development of water conservancy engineering. Summary of the invention

[0003] The purpose of the present invention is to provide a hydraulic structure risk assessment method and system based on deep learning and digital twins.

[0004] In a first aspect, an embodiment of the present invention provides a hydraulic structure risk assessment method based on deep learning and digital twins, comprising:

[0005] Obtain hydraulic structure parameters of the hydraulic structure to be evaluated;

[0006] Inputting the hydraulic structure parameters into a preset digital twin platform, constructing a target three-dimensional hydraulic structure model corresponding to the hydraulic structure to be evaluated; the target three-dimensional hydraulic structure model is composed of multiple hydraulic structure modules;

[0007] Calling the key module identification model and the module type identification model configured on the preset digital twin platform to analyze the target three-dimensional hydraulic structure model, and obtaining a module type identification result of the target three-dimensional hydraulic structure model;

[0008] According to the module type identification result, determining a current hydraulic structure module from the multiple hydraulic structure modules;

[0009] Anomalies are identified on the current hydraulic structure module to obtain a hydraulic structure risk assessment result of the target three-dimensional hydraulic structure model.

[0010] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.

[0011] Compared with the prior art, the beneficial effects provided by the present invention include: adopting a hydraulic structure risk assessment method and system based on deep learning and digital twins disclosed by the present invention, constructing a target three-dimensional hydraulic structure model by acquiring hydraulic structure parameters, and using the key module identification model and module type identification model configured on the digital twin platform for analysis, determining the current hydraulic structure module, and then performing abnormal identification, and finally obtaining the risk assessment result of the hydraulic structure. Such a design realizes the intelligent and precise risk assessment of hydraulic structures, and improves the assessment efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0013] Figure 1 A schematic diagram of the steps of a hydraulic structure risk assessment method based on deep learning and digital twins provided in an embodiment of the present invention;

[0014] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0016] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0017] In order to solve the technical problems in the aforementioned background technology, Figure 1 A flow chart of a hydraulic structure risk assessment method based on deep learning and digital twins is provided in an embodiment of the present disclosure. The hydraulic structure risk assessment method based on deep learning and digital twins is introduced in detail below.

[0018] Step S201, obtaining hydraulic structure parameters of the hydraulic structure to be evaluated;

[0019] Step S202, inputting the hydraulic structure parameters into a preset digital twin platform to construct a target three-dimensional hydraulic structure model corresponding to the hydraulic structure to be evaluated; the target three-dimensional hydraulic structure model is composed of multiple hydraulic structure modules;

[0020] Step S203, calling the key module identification model and the module type identification model configured in the preset digital twin platform to analyze the target three-dimensional hydraulic structure model, and obtaining the module type identification result of the target three-dimensional hydraulic structure model;

[0021] Step S204, determining a current hydraulic structure module from the plurality of hydraulic structure modules according to the module type identification result;

[0022] Step S205 , performing abnormality identification on the current hydraulic structure module to obtain a hydraulic structure risk assessment result of the target three-dimensional hydraulic structure model.

[0023] It can be understood that the hydraulic structure may include various types of hydraulic structures such as dams, aqueducts, and spillways.

[0024] In an embodiment of the present invention, exemplarily, the server first obtains real-time hydraulic structure parameters from the dam body monitoring system, including the height, width, material type, geological structure information, etc. of the dam body. For example, the height of a hydropower station dam body is 150 meters and the width is 200 meters. It is mainly composed of concrete and rock, and the dam foundation is located in a complex geological fault area. The server inputs these hydraulic structure parameters into the preset digital twin platform, and uses the modeling tools of the platform to construct a three-dimensional geological model of the hydraulic structure to be evaluated. This model consists of multiple hydraulic structure modules, such as a dam foundation module, a dam body module, a dam shoulder module, etc. For example, based on the input hydraulic structure parameters, the digital twin platform automatically generates a detailed three-dimensional hydraulic structure model, which includes the rock layer distribution of the dam foundation, the concrete structure of the dam body, and the soil layer distribution of the dam shoulder. The server calls the key module identification model and the module type identification model configured in the digital twin platform to analyze the constructed three-dimensional hydraulic structure model. These models are based on deep learning algorithms and can automatically identify different geological structure modules and their types in the model. For example, through analysis, the server identified that the dam foundation module belongs to the complex rock layer type, the dam body module belongs to the concrete pouring type, and the dam shoulder module belongs to the soft soil layer type. Based on the module type identification results, the server determines the hydraulic structure module that currently needs to be focused on from multiple hydraulic structure modules. For example, based on the analysis results, the server determines that the dam foundation module is a geological structure module that currently needs to be focused on because it is located in a complex rock layer area. The server identifies anomalies in the current hydraulic structure module and uses a deep learning algorithm to analyze whether the module has geological structural anomalies or potential risks. For example, the server found that a part of the rock layer in the dam foundation module has obvious faults and cracks, which may affect the stability of the dam body. Therefore, the server obtains the hydraulic structure risk assessment results of the target three-dimensional hydraulic structure model, indicating that the dam foundation module has a high geological structure risk.

[0025] In the embodiment of the present invention, the aforementioned step S203 can be implemented through the following example execution.

[0026] According to the key module identification model, the target three-dimensional hydraulic structure model is subjected to key module identification, and the key module identification range in the target three-dimensional hydraulic structure model is identified; the key module identification model is a model trained based on a standard three-dimensional hydraulic structure model instance configured with a key module identification target value, and the standard three-dimensional hydraulic structure model instance is a three-dimensional model containing a general type of module;

[0027] Determine, according to the newly added three-dimensional hydraulic structure model instance including the standard three-dimensional hydraulic structure model instance and the preset three-dimensional hydraulic structure model, the module performance characteristics of the module type included in the newly added three-dimensional hydraulic structure model instance; the preset three-dimensional hydraulic structure model is a three-dimensional model including a module of a preset type;

[0028] According to the module basic features corresponding to each of the key module identification ranges and the module performance features of each of the module types, and according to the module type identification model, module types of the modules in the key module identification range are identified to obtain a module type identification result of the target three-dimensional hydraulic structure model;

[0029] Among them, the module type identification model is a model obtained by iteratively training the preset module type identification model according to the newly added three-dimensional hydraulic structure model instance; the newly added three-dimensional hydraulic structure model instance is configured with a module type target value and a key module identification target value.

[0030] In an embodiment of the present invention, exemplarily, the server uses a key module identification model to identify key modules of a target three-dimensional hydraulic structure model. This model is obtained by training a standard three-dimensional hydraulic structure model instance configured with a key module identification target value. The standard three-dimensional hydraulic structure model instance is a three-dimensional model containing a general type of module, such as a common dam foundation, dam body and dam shoulder module. For example, the server identifies that the key module identification range in the target three-dimensional hydraulic structure model includes a dam foundation module and a dam body module, because these modules are marked as key modules in the standard three-dimensional hydraulic structure model instance. The server determines the module performance characteristics of the module type included in the newly added three-dimensional hydraulic structure model instance based on the newly added three-dimensional hydraulic structure model instance including the standard three-dimensional hydraulic structure model instance and the preset three-dimensional hydraulic structure model. The preset three-dimensional hydraulic structure model is a three-dimensional model containing a preset type of module, such as a dam foundation module of a special geological structure. For example, the server analyzes the newly added three-dimensional hydraulic structure model instance and finds that the preset type of dam foundation module (such as a dam foundation located in a complex fault area) has unique rock layer distribution and structural characteristics, which are recorded as module performance characteristics. The server uses the module type identification model to identify the module type in the key module identification range according to the basic module features corresponding to each key module identification range and the module performance features of each module type. The module type identification model is a model obtained by iteratively training the preset module type identification model according to the newly added three-dimensional hydraulic structure model instance. For example, the server combines the basic features of the dam foundation module (such as shape and size) and the module performance features of the preset type of dam foundation module (such as rock layer distribution and structural characteristics), and uses the module type identification model to identify that the dam foundation module in the target three-dimensional hydraulic structure model belongs to the dam foundation module of the complex fault type. Through the above steps, the server obtains the module type identification result of the target three-dimensional hydraulic structure model, including the type and characteristics of each key module. For example, the module type identification result of the target three-dimensional hydraulic structure model finally obtained by the server shows that the dam foundation module is a complex fault type, the dam body module is a concrete casting type, and the dam shoulder module is a soft soil layer type. These results provide an important basis for the subsequent risk assessment of hydraulic structures.

[0031] In an embodiment of the present invention, the key module identification model includes a range selection model and a key range positioning unit; the key module identification is performed on the target three-dimensional hydraulic structure model according to the key module identification model, and the key module identification range in the target three-dimensional hydraulic structure model is identified, which can be implemented through the following examples.

[0032] Obtaining a pending selection range in the target three-dimensional hydraulic structure model through the range selection model, and obtaining a range selection probability value corresponding to the pending selection range through a range selection probability component of the range selection model;

[0033] Selecting the pending selection range according to the range selection probability value;

[0034] The key module identification range in the target three-dimensional hydraulic structure model is determined by performing key module identification processing on the selected pending selection range through the key range positioning unit.

[0035] In an embodiment of the present invention, exemplarily, the server first uses the range selection model in the key module identification model to analyze the target three-dimensional hydraulic structure model to obtain a pending selection range in the model. This range may include multiple potential key module areas. For example, the server analyzes the target three-dimensional hydraulic structure model through the range selection model to obtain a pending selection range including a dam foundation module and a partial area of ​​the dam body module. This range may cover a more important part of the dam structure. Then, the server calculates the range selection probability value corresponding to the pending selection range through the range selection probability component of the range selection model. This probability value indicates the possibility that the range is selected as the key module identification range. For example, the server calculates that the range selection probability value of the pending selection range (including the dam foundation module and a partial area of ​​the dam body module) is 0.8, indicating that this range is very likely to be the key module identification range. The server selects the pending selection range according to the range selection probability value. Usually, the server sets a threshold, and only when the range selection probability value is greater than or equal to this threshold will the pending selection range be selected as the key module identification range. For example, the server sets the threshold value to 0.7. Since the calculated range selection probability value is 0.8, which is greater than the threshold, the server selects the pending selection range (including the dam foundation module and part of the dam body module) as the key module identification range. The server performs key module identification processing on the selected pending selection range through the key range positioning unit in the key module identification model. This processing process will further analyze the geological structure characteristics within the selection range and determine the key modules therein. For example, the server uses the key range positioning unit to conduct a detailed analysis of the selection range (including the dam foundation module and part of the dam body module), and finally determines that the dam foundation module is the key module identification range, because the geological structure characteristics of the dam foundation module have an important influence on the stability of the hydraulic structure. Through the above steps, the server can accurately identify the key module identification range in the target three-dimensional hydraulic structure model, providing an important basis for subsequent module type identification and hydraulic structure risk assessment.

[0036] In the embodiments of the present invention, the following implementation modes are also provided.

[0037] Obtaining a general type three-dimensional hydraulic structure model instance configured with a key module identification target value and an initial scope selection model, wherein the initial scope selection model includes a module type identification component;

[0038] Replacing the module type identification component of the initial range selection model with a range selection probability component;

[0039] Constructing an initial key module identification model according to the initial range selection model after module replacement;

[0040] The initial key module identification model is trained by using the general type three-dimensional hydraulic structure model instance configured with the key module identification target value to obtain the key module identification model.

[0041] In an embodiment of the present invention, exemplarily, the server first obtains a general type three-dimensional hydraulic structure model instance configured with a key module identification target value. These instances are basic data for training the key module identification model, and they include common module types in hydraulic structures and their key module identification ranges. For example, the server obtains 100 general type three-dimensional hydraulic structure model instances configured with key module identification target values ​​from the database, and these instances cover hydraulic structures with different dam heights, dam types and geological conditions. At the same time, the server also obtains an initial range selection model, which includes a module type identification component for identifying different module types in the three-dimensional hydraulic structure model. The server replaces the module type identification component in the initial range selection model with a range selection probability component. The function of the range selection probability component is to calculate the probability that different areas in the three-dimensional hydraulic structure model are selected as the key module identification range. For example, the server replaces the module type identification component in the initial range selection model with the range selection probability component through programming, thereby realizing the conversion of the model function. According to the initial range selection model after the component is replaced, the server constructs an initial key module identification model. This model includes a range selection probability component and a key range positioning unit, which are used to identify the key module identification range in the three-dimensional hydraulic structure model. For example, the server uses a programming tool to combine the range selection probability component and the key range positioning unit to form an initial key module identification model. The server uses a general type three-dimensional hydraulic structure model instance configured with a key module identification target value to train the initial key module identification model. During the training process, the server will continuously adjust the parameters of the model so that the model can accurately identify the key module identification range in the three-dimensional hydraulic structure model. For example, the server inputs 100 general type three-dimensional hydraulic structure model instances configured with key module identification target values ​​into the initial key module identification model and trains the model through a machine learning algorithm. After multiple iterations and optimizations, the server obtains a key module identification model that can accurately identify the key module identification range. Through the above steps, the server successfully builds and trains a key module identification model, which can accurately identify the key module identification range in the three-dimensional hydraulic structure model, providing important technical support for subsequent hydraulic structure risk assessment.

[0042] In an embodiment of the present invention, the module type identification model includes a key range detection unit;

[0043] According to the module basic features corresponding to each of the key module identification ranges and the module performance features of each of the module types, according to the module type identification model, the module type identification is performed on the modules in the key module identification range to obtain the module type identification result of the target three-dimensional hydraulic structure model, including:

[0044] According to the module basic features corresponding to each of the key module identification ranges, the key module identification range is detected by a key range detection unit to obtain a key range detection feature;

[0045] Obtaining an inferred range position and a type score corresponding to the inferred range position according to the key range detection feature and the module performance feature of each module type;

[0046] According to the inferred range position and the type score corresponding to the inferred range position, a module type recognition result of the target three-dimensional hydraulic structure model is determined.

[0047] In the embodiment of the present invention, illustratively, the server first detects the key module identification range through the key range detection unit in the module type identification model according to the module basic features corresponding to each key module identification range.

[0048] For example, the server has determined that the dam foundation module and part of the dam body module are the key module identification ranges through the key module identification model. Now, the server uses the key range detection unit to detect these areas, analyze their geological structure, shape, size and other module basic features, and obtain key range detection features. The server then matches and compares the key range detection features and the module performance features of each module type.

[0049] For example, the server compares the key range detection features with the preset module type representation feature library to find the most similar module type and calculate an inferred range position and the type score corresponding to the position. The type score indicates the probability that the inferred range position belongs to a certain module type.

[0050] Finally, the server determines the module type recognition result of the target three-dimensional hydraulic structure model according to the inferred range position and the type score corresponding to the inferred range position.

[0051] For example, based on the calculated inference range position and its type score, the server finally determined that the key module identification range of the dam foundation module belongs to the "complex fault type" and the key module identification range of the dam body module belongs to the "concrete pouring type". These identification results provide important basis for subsequent hydraulic structure analysis and risk assessment.

[0052] Through the above steps, the server can accurately identify the module types within the identification range of each key module in the target three-dimensional hydraulic structure model, providing strong support for the refined analysis and management of the hydraulic structure.

[0053] In the embodiment of the present invention, the inferred range position and the type score corresponding to the inferred range position are obtained according to the key range detection feature and the module performance feature of each module type, which can be implemented through the following examples.

[0054] Determine, according to the key range detection feature and the module performance features of each module type, an inferred range position and a gap between inferred ranges corresponding to different inferred range positions;

[0055] Processing the gaps between the inferred ranges by a range comparison component in the key range detection unit to reduce the gaps between inferred ranges of the same type and increase the gaps between inferred ranges of different types;

[0056] According to the gaps between the processed inference ranges, the type scores corresponding to the positions of the inference ranges are obtained.

[0057] In an embodiment of the present invention, exemplarily, the server first matches and compares the key range detection features and the module performance features of each module type, thereby determining multiple inference range positions. For example, within the key module identification range of the dam foundation module, the server preliminarily determines three possible inference range positions based on the key range detection features such as geological structure and shape, and marks them as A, B, and C respectively. Then, the server calculates the gaps between the inference ranges corresponding to these inference range positions. These gaps reflect the similarities or differences between different inference ranges in terms of geological structure, shape and other features. For example, the server calculates the gaps between the inference range positions A and B, A and C, and B and C, and finds that the gap between A and B is small, while the gaps between A and C, and B and C are large. The server processes the gaps between the inference ranges through the range comparison component in the key range detection unit. The goal of the processing is to reduce the gaps between inference ranges of the same type and to increase the gaps between inference ranges of different types. For example, the server finds through the range comparison component analysis that the geological structure features of the inference range positions A and B are relatively similar and belong to the same type (such as "complex fault type"), so the gap between A and B is reduced. At the same time, the server found that the geological structure characteristics of the inference range position C were quite different from those of A and B, and belonged to different types (such as "rock and soil layer type"), so the gaps between A and C, and between B and C were increased. Finally, the server calculated the type scores corresponding to each inference range position based on the gaps between the processed inference ranges. The type score reflects the possibility that the inference range position belongs to a certain module type. For example, the inference range gap after processing shows that the inference range positions A and B are more likely to belong to the "complex fault type", so their corresponding type scores are also higher. The inference range position C is more likely to belong to the "rock and soil layer type", so its corresponding type score is higher in comparison with the "rock and soil layer type", but lower in comparison with the "complex fault type". Through the above steps, the server can more accurately determine the module type of each key module identification range in the target three-dimensional hydraulic structure model, and provide more accurate data support for subsequent hydraulic structure analysis and risk assessment.

[0058] In the embodiments of the present invention, the following implementation modes are also provided.

[0059] Obtaining a general type three-dimensional hydraulic structure model instance configured with a module type target value and a key module identification target value, a preset type three-dimensional hydraulic structure model instance configured with a module type target value and a key module identification target value, and a preset module type identification model;

[0060] Replacing the range selection probability component of the key range detection unit in the preset module type identification model with a module type identification component;

[0061] The key range detection unit after module replacement is iteratively trained by using the general type three-dimensional hydraulic structure model instance configured with module type target value and key module identification target value, and the preset type three-dimensional hydraulic structure model instance configured with module type target value and key module identification target value to obtain a module type identification model.

[0062] In an embodiment of the present invention, exemplarily, the server first obtains two types of three-dimensional hydraulic structure model instances: one type is a general type three-dimensional hydraulic structure model instance configured with a module type target value and a key module identification target value, and the other type is a preset type three-dimensional hydraulic structure model instance also configured with these two target values. For example, the server obtains 100 general type three-dimensional hydraulic structure model instances from the database, which cover dam bodies of different dam heights, dam types and geological conditions, and each instance is marked with a module type target value and a key module identification target value. At the same time, the server also obtains 50 preset type three-dimensional hydraulic structure model instances, which are targeted at preset dam body types (such as arch dams, gravity dams, etc.), and are also marked with module type target values ​​and key module identification target values. In addition, the server also obtains a preset module type identification model, which is a basic model for identifying module types in three-dimensional hydraulic structure models. The server replaces the range selection probability component of the key range detection unit in the preset module type identification model with a module type identification component. The function of the module type identification component is to identify different module types in the three-dimensional hydraulic structure model. For example, the server replaces the range selection probability component in the preset module type identification model with the module type identification component through programming, thereby realizing the conversion of the model function. The server uses the general type three-dimensional hydraulic structure model instance and the preset type three-dimensional hydraulic structure model instance configured with the module type target value and the key module identification target value to iteratively train the key range detection unit after the replacement component. For example, the server inputs 100 general type three-dimensional hydraulic structure model instances and 50 preset type three-dimensional hydraulic structure model instances into the key range detection unit after the replacement component, and iteratively trains the model through the machine learning algorithm. During the training process, the server continuously adjusts the parameters of the model so that the model can accurately identify the module type and the key module identification range in the three-dimensional hydraulic structure model. After iterative training, the server obtains a module type identification model that can accurately identify the module type and the key module identification range. For example, after multiple iterations and optimizations, the server obtained a module type recognition model that can accurately identify the module type (such as "complex fault type" and "concrete pouring type") and key module recognition range (such as the key areas of the dam foundation module and dam body module) in the three-dimensional hydraulic structure model. Through the above steps, the server successfully built and trained a module type recognition model that can accurately identify the module type and key module recognition range in the three-dimensional hydraulic structure model, providing important technical support for subsequent hydraulic structure analysis and risk assessment.

[0063] In an embodiment of the present invention, the method of determining the module performance characteristics of the module type included in the newly added three-dimensional hydraulic structure model instance based on the newly added three-dimensional hydraulic structure model instance including the standard three-dimensional hydraulic structure model instance and the preset three-dimensional hydraulic structure model can be implemented through the following examples.

[0064] Acquire a newly added three-dimensional hydraulic structure model instance including the standard three-dimensional hydraulic structure model instance and the preset three-dimensional hydraulic structure model, wherein the newly added three-dimensional hydraulic structure model instance is configured with a type target value;

[0065] Extracting multi-view features of the newly added three-dimensional hydraulic structure model instance;

[0066] According to the type target value, weighted averaging is performed on multi-view features of newly added three-dimensional hydraulic structure model instances of the same type to obtain module performance features.

[0067] In an embodiment of the present invention, exemplarily, the server first obtains newly added three-dimensional hydraulic structure model instances including standard three-dimensional hydraulic structure model instances and preset three-dimensional hydraulic structure models. These newly added instances are to further enrich and improve the training data set of the module type recognition model. For example, the server obtains 20 newly added three-dimensional hydraulic structure model instances from the database, which cover different types of dam bodies, and each instance is configured with a type target value to indicate the module type to which the instance belongs. The server then extracts the multi-view features of each newly added three-dimensional hydraulic structure model instance. The multi-view features are features that describe the geological model of the dam body from different angles and levels, including geological structure, shape, size, material, etc. For example, the server performs multi-view scanning and feature extraction on each newly added instance, and obtains a multi-view feature set including geological structure features, shape features, size features and material features. The server performs weighted average processing on the multi-view features of the newly added instances of the same type according to the type target value of each newly added three-dimensional hydraulic structure model instance. The purpose of this step is to obtain the module performance features of each module type, that is, the typical performance of the module of this type in the three-dimensional hydraulic structure model. For example, the server classified 20 newly added instances according to the type target value and obtained 5 different types of modules. Then, the server performed weighted averaging on the multi-view features of each type of module instance to obtain the module performance features of each module type. These features will be used in subsequent module type identification tasks. Through the above steps, the server successfully extracted the module performance features of the module types contained in the newly added three-dimensional hydraulic structure model instances. These features will serve as important inputs to the module type identification model to improve the model's recognition accuracy and generalization ability for module types.

[0068] In the embodiment of the present invention, the abnormality identification of the current hydraulic structure module to obtain the hydraulic structure risk assessment result of the target three-dimensional hydraulic structure model can be implemented through the following examples.

[0069] Performing feature extraction processing on the current hydraulic structure module to obtain features of the current hydraulic structure module, and performing feature extraction processing on the abnormal hydraulic structure module to obtain features of the abnormal hydraulic structure module; the abnormal hydraulic structure module is pre-set;

[0070] Performing feature analysis on the current hydraulic structure module feature to obtain the correlation coefficient of each current module sub-feature in the current hydraulic structure module feature, and performing feature analysis on the abnormal hydraulic structure module feature to obtain the correlation coefficient of each abnormal module sub-feature in the abnormal hydraulic structure module feature, wherein the correlation coefficient is used to indicate the influence on the hydraulic structure module correlation;

[0071] According to the correlation coefficients of the current module sub-features, the current hydraulic structure module feature is feature screened to obtain a first number of current module sub-features, and according to the correlation coefficients of the abnormal module sub-features, the abnormal hydraulic structure module feature is feature screened to obtain a first number of abnormal module sub-features;

[0072] Performing feature analysis on the first number of current module sub-features and the first number of abnormal module sub-features to obtain current association identifiers corresponding to the first number of current module sub-features and abnormal association identifiers corresponding to the first number of abnormal module sub-features, the current association identifiers being used to obtain a second number of current module sub-features having similar module regions to the first number of abnormal module sub-features from the first number of current module sub-features, and the abnormal association identifiers being used to enhance a third number of abnormal module sub-features having similar module regions to the first number of current module sub-features from the first number of abnormal module sub-features;

[0073] Performing feature comparison on the second number of current module sub-features and the third number of abnormal module sub-features to obtain a first association result between the current hydraulic structure module and the abnormal hydraulic structure module;

[0074] According to the first association result, a risk assessment result of the current hydraulic structure module is determined, and the risk assessment result is used as the hydraulic structure risk assessment result of the target three-dimensional hydraulic structure model.

[0075] In an embodiment of the present invention, exemplarily, the server performs feature extraction processing on the current hydraulic structure module to obtain a series of features describing the geological structure, shape, material, etc. of the module, which are called the current hydraulic structure module features. At the same time, the server also performs feature extraction processing on the pre-set abnormal hydraulic structure module to obtain a set of features describing the abnormal module, which is called the abnormal hydraulic structure module features. For example, the server obtains the geological structure data of the current dam body from the digital twin system, and extracts the current hydraulic structure module features including geological layer distribution, fault conditions, rock mass strength, etc. At the same time, the server also obtains the geological structure data of the known abnormal dam body from the abnormal dam body database, and extracts the corresponding abnormal hydraulic structure module features. The server performs feature analysis on the current hydraulic structure module features, and calculates the correlation coefficients between each current module sub-feature, which indicates the influence of different module sub-features on the association of the hydraulic structure module. Similarly, the server also performs feature analysis on the abnormal hydraulic structure module features to obtain the correlation coefficients of each abnormal module sub-feature. For example, the server analyzes the current hydraulic structure module features through a deep learning algorithm and finds that the geological layer distribution and fault conditions have a great influence on the stability of the dam body, so the correlation coefficient of these two features is high. Similarly, the server also performs a similar analysis on the abnormal hydraulic structure module features. According to the correlation coefficient, the server screens the current hydraulic structure module features and selects the first number (such as 10) of module sub-features that have the greatest impact on the current hydraulic structure. At the same time, the server also screens the abnormal hydraulic structure module features and selects the same number (such as 10) of module sub-features that have the greatest impact on the abnormal hydraulic structure. For example, the server selects the 10 most important features such as geological layer distribution, fault conditions, and rock mass strength from the current hydraulic structure module features according to the correlation coefficient. At the same time, the corresponding 10 most important features are also selected from the abnormal hydraulic structure module features. The server further analyzes the screened current module sub-features and abnormal module sub-features, and assigns a current correlation identifier to each current module sub-feature, which is used to identify the similar module area that may exist when the feature is compared with the abnormal module sub-feature. At the same time, the server also assigns an abnormal association identifier to each abnormal module sub-feature, which is used to enhance the feature importance of similar module areas when compared with the current module sub-feature. For example, the server compares and analyzes the current module sub-feature and the abnormal module sub-feature through a deep learning algorithm, and assigns a unique association identifier to each feature. The server uses the current association identifier and the abnormal association identifier to compare the screened current module sub-feature and the abnormal module sub-feature, finds the feature pairs with similar module areas, and calculates the similarity between them. Based on the degree of similarity, the server obtains the first association result between the current hydraulic structure module and the abnormal hydraulic structure module.For example, the server finds that the geological layer distribution of the current dam body is highly similar to that of a known abnormal dam body, so it regards them as a pair of similar module areas. Based on the first association result, the server performs a risk assessment on the current hydraulic structure module. If the current hydraulic structure module has similar module areas with multiple abnormal hydraulic structure modules, and the risk levels of these abnormal hydraulic structure modules are high, then the risk assessment result of the current hydraulic structure will also be correspondingly high. The server outputs the risk assessment result as the hydraulic structure risk assessment result of the target three-dimensional hydraulic structure model. For example, the server calculates that the risk assessment level of the current dam body is "high" based on the association results and the risk assessment algorithm, and outputs this result to the user interface of the digital twin system for display and alarm.

[0076] In an embodiment of the present invention, the feature screening of the current hydraulic structure module feature according to the correlation coefficient of each current module sub-feature to obtain a first number of current module sub-features, and the feature screening of the abnormal hydraulic structure module feature according to the correlation coefficient of each abnormal module sub-feature to obtain a first number of abnormal module sub-features can be implemented through the following examples.

[0077] Determining a first quantity according to a preset proportion;

[0078] According to the correlation coefficients of the various current module sub-features, according to a sequence of correlation coefficients from high to low, the first first number of current module sub-features in the current hydraulic structure module features are determined as the first number of current module sub-features, and according to the correlation coefficients of the various abnormal module sub-features, according to a sequence of correlation coefficients from high to low, the first first number of abnormal module sub-features in the abnormal hydraulic structure module features are determined as the first number of abnormal module sub-features.

[0079] In an embodiment of the present invention, exemplarily, the server first determines the first number according to a preset proportion. This proportion is set based on prior knowledge or experience of the importance of the hydraulic structure module feature, and is used to control the number of features screened out. For example, the server sets a proportion of 20%, which means that features accounting for 20% of the total number of features will be screened out from the current hydraulic structure module feature and the abnormal hydraulic structure module feature as the first number of features. The server sorts these features in order of correlation coefficient from high to low according to the correlation coefficient of each current module sub-feature. Then, the server selects the first first number (i.e., the number calculated according to the proportion) of current module sub-features from the sorted feature list as the important features screened out. For example, if there are a total of 50 features of the current hydraulic structure module, the first number is 10 according to the proportion of 20%. The server sorts these 50 features from high to low according to the correlation coefficient, and selects the first 10 as the first number of current module sub-features. The server also sorts these features in order of correlation coefficient from high to low according to the correlation coefficient of each abnormal module sub-feature. Next, the server selects the first number (the same number as the current module sub-feature) of abnormal module sub-features from the sorted abnormal feature list as the selected important features. For example, there are 50 abnormal hydraulic structure module features in total. The server also sorts them from high to low according to the correlation coefficient and selects the first 10 as the first number of abnormal module sub-features. Through the above steps, the server successfully selects the first number of the most important features from the current hydraulic structure module features and the abnormal hydraulic structure module features, respectively, for subsequent abnormality identification and risk assessment.

[0080] In the embodiment of the present invention, the determining of the first quantity according to the preset proportion may be implemented through the following examples.

[0081] The first quantity is determined according to the preset proportion and the module scale of the current hydraulic structure module characteristics, or the first quantity is determined according to the preset proportion and the module scale of the abnormal hydraulic structure module characteristics.

[0082] In an embodiment of the present invention, exemplarily, the server first obtains the module scale of the current hydraulic structure module feature, that is, the total number of features contained in the module feature. Then, the server calculates the first number according to the preset proportion (such as 20%) and the module scale. For example, if the module scale contains 100 features, then according to the proportion of 20%, the first number will be determined as 20 features. Therefore, the server will select the top 20 features with the highest correlation coefficient from the current hydraulic structure module feature as the first number of current module sub-features based on the number of these 20 features. Similarly, the server will also obtain the module scale of the abnormal hydraulic structure module feature, that is, the total number of features contained in the module feature. Then, the server also calculates the first number according to the preset proportion (such as 20%) and the abnormal module scale. For example, if the abnormal module scale also contains 100 features, then the first number will also be determined as 20 features. Therefore, based on the number of these 20 features, the server will select the top 20 features with the highest correlation coefficients from the abnormal hydraulic structure module features as the first number of abnormal module sub-features. Through the above steps, the server can dynamically determine the first number based on the actual module scale and the preset proportion, so as to more accurately select the most important features from the current hydraulic structure module features and the abnormal hydraulic structure module features for subsequent analysis. This method improves the flexibility and accuracy of feature screening, and helps to more accurately identify abnormal conditions of hydraulic structures and conduct risk assessments.

[0083] In an embodiment of the present invention, the feature analysis of the first number of current module sub-features and the first number of abnormal module sub-features to obtain current association identifiers corresponding to the first number of current module sub-features and abnormal association identifiers corresponding to the first number of abnormal module sub-features can be implemented through the following examples.

[0084] Determine current original association identifiers corresponding to the first number of current module sub-features, and determine abnormal original association identifiers corresponding to the first number of abnormal module sub-features;

[0085] adjusting the first number of current module sub-features according to the first number of current module sub-features, the first number of abnormal module sub-features, the current original association identifier, and the abnormal original association identifier to obtain a first number of current adjusted features; and adjusting the first number of abnormal module sub-features according to the first number of current module sub-features, the first number of abnormal module sub-features, the current original association identifier, and the abnormal original association identifier to obtain a first number of abnormal adjusted features;

[0086] According to the first number of current adjustment features, the current original association identifier is adjusted to obtain a current adjustment association identifier, and according to the first number of abnormal adjustment features, the abnormal original association identifier is adjusted to obtain an abnormal adjustment association identifier;

[0087] The current association identifier is determined according to the current adjustment association identifier, and the abnormal association identifier is determined according to the abnormal adjustment association identifier.

[0088] In an embodiment of the present invention, exemplarily, the server first allocates current original association identifiers for the first number of current module sub-features, which are initial and unadjusted, and are used to distinguish different current module sub-features. At the same time, the server also allocates abnormal original association identifiers for the first number of abnormal module sub-features, which are also used to distinguish different abnormal module sub-features. For example, the server allocates current original association identifiers of ID1 to ID10 for the 10 current module sub-features selected, and allocates abnormal original association identifiers of ID'1 to ID'10 for the 10 abnormal module sub-features selected. The server adjusts the current module sub-features according to the first number of current module sub-features, the first number of abnormal module sub-features, and their corresponding original association identifiers. This adjustment can be based on the similarity, difference or other predefined rules between the features. At the same time, the server also makes similar adjustments to the abnormal module sub-features. For example, the server finds that the current module sub-feature ID5 and the abnormal module sub-feature ID'3 have a high similarity in geological layer distribution, so it decides to adjust ID5 to enhance its association with ID'3. After adjusting the module sub-features, the server adjusts the current original association identifier according to the first number of current adjustment features to obtain the current adjustment association identifier. Similarly, the server also adjusts the abnormal original association identifier according to the first number of abnormal adjustment features to obtain the abnormal adjustment association identifier. For example, due to the adjustment of ID5, the server can update its current adjustment association identifier to ID5* to indicate that it is different from the original identifier ID5 and reflects the enhanced association with ID'3. Finally, the server determines the final current association identifier based on the current adjustment association identifier, which will be used for subsequent feature comparison and association analysis. At the same time, the server also determines the final abnormal association identifier based on the abnormal adjustment association identifier. For example, the server uses ID5* as one of the final current association identifiers and uses it for comparison and analysis with the abnormal association identifier. Through the above steps, the server can more accurately assign association identifiers to the current module sub-features and abnormal module sub-features. These identifiers reflect the similarities and differences between the features, which are helpful for subsequent abnormal identification and risk assessment.

[0089] In the embodiment of the present invention, before performing feature extraction processing on the current hydraulic structure module to obtain the features of the current hydraulic structure module and performing feature extraction processing on the abnormal hydraulic structure module to obtain the features of the abnormal hydraulic structure module, the following implementation mode is also provided.

[0090] If the module scale of the current hydraulic structure module and the module scale of the abnormal hydraulic structure module have not been unified, unify the module scales of the current hydraulic structure module and the abnormal hydraulic structure module to the same module scale;

[0091] The determining of the current original association identifiers corresponding to the first number of current module sub-features and the determining of the abnormal original association identifiers corresponding to the first number of abnormal module sub-features include:

[0092] Determine the current original association identifier according to the elements supplemented in the unified scale processing of the current hydraulic structure module, wherein the data corresponding to the features of the elements supplemented in the first number of current module sub-features in the current original association identifier is a reference value;

[0093] According to the elements supplemented in the unified scale processing of the abnormal hydraulic structure module, the abnormal original association identifier is determined, and in the abnormal original association identifier, the data corresponding to the characteristics of the elements supplemented in the first number of abnormal module sub-features is the reference value.

[0094] In an embodiment of the present invention, exemplarily, the server first checks whether the module scales of the current hydraulic structure module and the abnormal hydraulic structure module are consistent. If the scales are found to be inconsistent, the server will supplement the smaller module so that it has the same number of elements as the larger module. This supplementation can be achieved by adding zero values, average values, or other predefined values. For example, if the current hydraulic structure module contains 100 features, and the abnormal hydraulic structure module only contains 80 features, the server will add 20 zero values ​​or average values ​​to the abnormal module so that it also reaches 100 features. After the scale unification process is performed, the server begins to determine the original association identifiers corresponding to the first number of current module sub-features and the first number of abnormal module sub-features. For elements supplemented due to scale unification, the server will give special consideration when assigning original association identifiers. These supplemented elements will have corresponding data in the original association identifier, which are set as reference values ​​to distinguish whether they are supplemented elements or originally existing elements. For example, the server assigns current original association identifiers from ID1 to ID100 to the 100 features of the current hydraulic structure module. For the 20 features supplemented due to scale unification, the server sets the data in their current original association identifiers to a preset baseline value, such as "Supplement_01" to "Supplement_20", to distinguish them from the original 80 features. For the abnormal hydraulic structure module, the server also performs similar operations. It assigns abnormal original association identifiers to the elements supplemented due to scale unification and sets these data as baseline values. For example, for the 20 features also supplemented by the abnormal module, the server sets the data in their abnormal original association identifiers to "Abnormal_Supplement_01" to "Abnormal_Supplement_20". Through the above steps, the server ensures that the current hydraulic structure module and the abnormal hydraulic structure module have the same module scale, and specially considers the elements supplemented due to scale unification when assigning original association identifiers. This facilitates subsequent feature analysis and determination of association identifiers, thereby improving the accuracy of abnormal identification and risk assessment.

[0095] In an embodiment of the present invention, adjusting the first number of current module sub-features according to the first number of current module sub-features, the first number of abnormal module sub-features, the current original association identifier, and the abnormal original association identifier to obtain a first number of current adjusted features and a first number of abnormal adjusted features, and adjusting the first number of abnormal module sub-features according to the first number of current module sub-features, the first number of abnormal module sub-features, the current original association identifier, and the abnormal original association identifier to obtain a first number of abnormal adjusted features can be implemented through the following examples.

[0096] Applying the current original association identifier to the first number of current module sub-features to obtain a first number of current processing features, and applying the abnormal original association identifier to the first number of abnormal module sub-features to obtain a first number of abnormal processing features;

[0097] The first number of current processing features are dynamically weighted by using a first dynamic weight allocation network, and the first number of abnormal processing features and the first number of features output by the first dynamic weight allocation network are processed by using a first associated weight allocation network to obtain the first number of current adjustment features, and the first number of abnormal processing features are dynamically weighted by using a second dynamic weight allocation network, and the first number of current processing features and the first number of features output by the second dynamic weight allocation network are processed by using a second associated weight allocation network to obtain the first number of abnormal adjustment features.

[0098] In an embodiment of the present invention, exemplarily, the server applies the current original association identifier to the first number of current module sub-features to obtain the first number of current processing features. This step is achieved by associating each current module sub-feature with its corresponding current original association identifier. At the same time, the server also applies the abnormal original association identifier to the first number of abnormal module sub-features to obtain the first number of abnormal processing features. This step is similar to associating each abnormal module sub-feature with its corresponding abnormal original association identifier. For example, the server has 10 current module sub-features, each of which has a corresponding current original association identifier (such as ID1 to ID10). The server applies these identifiers to the features to obtain 10 current processing features. Similarly, the server also obtains 10 abnormal processing features. The server uses the first dynamic weight allocation network to perform dynamic weight allocation processing on the first number of current processing features. This network assigns a dynamic weight to each current processing feature according to the complexity and importance of the feature. At the same time, the server also uses the second dynamic weight allocation network to perform dynamic weight allocation processing on the first number of abnormal processing features. This network also assigns a dynamic weight to each abnormal processing feature. For example, the first dynamic weight allocation network may assign a higher weight to the current processing feature ID1 because it has a higher importance in the hydraulic structure. Similarly, the second dynamic weight allocation network may also assign a higher weight to a certain abnormal processing feature. The server uses the first association weight allocation network to process the first number of abnormal processing features and the first number of features output by the first dynamic weight allocation network to obtain the first number of current adjustment features. This network considers the correlation between the abnormal processing feature and the current processing feature, and adjusts the weight of the current processing feature according to this correlation. At the same time, the server also uses the second association weight allocation network to process the first number of current processing features and the first number of features output by the second dynamic weight allocation network to obtain the first number of abnormal adjustment features. This network also considers the correlation between the current processing feature and the abnormal processing feature, and adjusts the weight of the abnormal processing feature according to this correlation. For example, if the current processing feature ID1 has a high correlation with a certain abnormal processing feature, the first association weight allocation network may increase the weight of the abnormal processing feature and adjust the weight of ID1 accordingly to obtain the current adjustment feature. Similarly, the second association weight allocation network will also perform similar operations to obtain the abnormal adjustment feature. Through the above steps, the server can dynamically adjust the weights of these features according to the correlation between the current module sub-features and the abnormal module sub-features and their importance, so as to obtain more accurate and meaningful current adjustment features and abnormal adjustment features. This helps in subsequent feature comparison and correlation analysis to improve the accuracy of abnormal identification and risk assessment.

[0099] In an embodiment of the present invention, the current original association identifier is adjusted according to the first number of current adjustment features to obtain a current adjustment association identifier, and the abnormal original association identifier is adjusted according to the first number of abnormal adjustment features to obtain an abnormal adjustment association identifier. This can be implemented through the following examples.

[0100] Performing feature analysis on the first number of current adjustment features to obtain two confidences corresponding to each of the first number of current adjustment features, and performing feature analysis on the first number of abnormal adjustment features to obtain two confidences corresponding to each of the first number of abnormal adjustment features, the two confidences being used to indicate the confidence of features belonging to a similar module region and the confidence of features not belonging to a similar module region;

[0101] Determine a first number of first pointers corresponding to the first number of current adjustment features according to the two confidences corresponding to the respective current adjustment features, and determine a first number of second pointers corresponding to the first number of abnormal adjustment features according to the two confidences corresponding to the respective abnormal adjustment features, the first pointer being used to indicate a locator of a layer corresponding to a maximum confidence value among the two confidences corresponding to the current adjustment feature, and the second pointer being used to indicate a locator of a layer corresponding to a maximum confidence value among the two confidences corresponding to the abnormal adjustment feature;

[0102] Performing an inverse operation on the first number of first pointers to obtain a first number of first confidences, and performing an inverse operation on the first number of second pointers to obtain a first number of second confidences;

[0103] The current original association identifier is applied to the first number of first confidences to obtain the current adjusted association identifier, and the abnormal original association identifier is applied to the first number of second confidences to obtain the abnormal adjusted association identifier.

[0104] In an embodiment of the present invention, exemplarily, the server performs feature analysis on the first number of current adjustment features to obtain two confidences corresponding to each current adjustment feature. The two confidences are used to indicate the confidence that the feature belongs to the similar module area and the confidence that it does not belong to the similar module area, respectively. At the same time, the server also performs feature analysis on the first number of abnormal adjustment features to obtain two confidences corresponding to each abnormal adjustment feature. The two confidences are also used to indicate the confidence that the feature belongs to the similar module area and the confidence that it does not belong to the similar module area, respectively. For example, for the current adjustment feature ID1, the server can obtain a higher confidence (such as 0.8) that it belongs to the similar module area and a lower confidence (such as 0.2) that it does not belong to the similar module area. The server determines the first number of first pointers corresponding to the first number of current adjustment features according to the two confidences corresponding to each current adjustment feature. Each first pointer is used to indicate the locator of the level corresponding to the maximum confidence value of the two confidences corresponding to the current adjustment feature. At the same time, the server also determines the first number of second pointers corresponding to the first number of abnormal adjustment features according to the two confidences corresponding to each abnormal adjustment feature. Each second pointer is used to indicate the locator of the level corresponding to the maximum confidence value of the two confidences corresponding to the abnormal adjustment feature. For example, for the current adjustment feature ID1, since its confidence that it belongs to the similar module area is high (0.8), its corresponding first pointer may point to the locator of the level of "similar module area". The server performs inverse operation processing on the first number of first pointers to obtain the first number of first confidences. These first confidences are obtained after the original confidences are transformed or inversely transformed. At the same time, the server also performs inverse operation processing on the first number of second pointers to obtain the first number of second confidences. These second confidences are also obtained after the original confidences are transformed or inversely transformed. For example, for the first pointer corresponding to the current adjustment feature ID1, the inverse operation processing may obtain a more refined confidence belonging to the similar module area (such as 0.85). The server uses the current original association identifier to apply to the first number of first confidences to obtain the current adjustment association identifier. This step is to associate each first confidence with its corresponding current original association identifier to form a new association identifier. At the same time, the server also uses the abnormal original association identifier to apply to the first number of second confidences to obtain the abnormal adjustment association identifier. This step is similar to associating each second confidence with its corresponding abnormal original association identifier. For example, for the first confidence (0.85) of the current adjustment feature ID1, the server can associate it with the current original association identifier ID1 to obtain a new current adjustment association identifier (such as ID1_0.85). Similarly, the server will also obtain the abnormal adjustment association identifier.Through the above steps, the server can dynamically adjust these association identifiers according to the confidence of the current adjustment features and abnormal adjustment features and their association with the original association identifiers, so as to obtain more accurate and meaningful current adjustment association identifiers and abnormal adjustment association identifiers. This helps in subsequent feature comparison and association analysis to improve the accuracy of abnormality identification and risk assessment.

[0105] In an embodiment of the present invention, the current original association identifier is adjusted according to the first number of current adjustment features to obtain a current adjustment association identifier, and the abnormal original association identifier is adjusted according to the first number of abnormal adjustment features to obtain an abnormal adjustment association identifier. This can be implemented through the following examples.

[0106] According to the first number of current adjustment features, the first number of abnormal adjustment features, the current adjustment association identifier, and the abnormal adjustment association identifier, the first number of current adjustment features are adjusted to obtain a first number of current optimized adjustment features; and according to the first number of current adjustment features, the first number of abnormal adjustment features, the current adjustment association identifier, and the abnormal adjustment association identifier, the first number of abnormal adjustment features are adjusted to obtain a first number of abnormal optimized adjustment features;

[0107] According to the first number of current optimization adjustment features, the current adjustment association identifier is adjusted to obtain a current optimization adjustment association identifier, and according to the first number of abnormal optimization adjustment features, the abnormal adjustment association identifier is adjusted to obtain an abnormal optimization adjustment association identifier;

[0108] The current association identifier is determined according to the current optimization adjustment association identifier, and the abnormal association identifier is determined according to the abnormal optimization adjustment association identifier, wherein the current association identifier is an adjustment association identifier obtained when the optimization adjustment times of the current original association identifier exceeds a preset adjustment times threshold, and the abnormal association identifier is an adjustment association identifier obtained when the optimization adjustment times of the abnormal original association identifier exceeds the preset adjustment times threshold.

[0109] In an embodiment of the present invention, exemplarily, the server adjusts the first number of current adjustment features according to the first number of current adjustment features, the first number of abnormal adjustment features, the current adjustment association identifier and the abnormal adjustment association identifier to obtain the first number of current optimized adjustment features. At the same time, the server also adjusts the first number of abnormal adjustment features according to this information to obtain the first number of abnormal optimized adjustment features. For example, the server can fine-tune the feature value of ID1 according to the correlation between the current adjustment feature ID1 and a certain abnormal adjustment feature, and their corresponding adjustment association identifiers, to obtain a more accurate current optimized adjustment feature. Similarly, the server will also perform similar optimization adjustments on the abnormal adjustment features. The server adjusts the current adjustment association identifier according to the first number of current optimized adjustment features to obtain the current optimized adjustment association identifier. At the same time, the server also adjusts the abnormal adjustment association identifier according to the first number of abnormal optimized adjustment features to obtain the abnormal optimized adjustment association identifier. For example, for the current adjustment association identifier ID1_0.85, the server can adjust it to ID1_0.87 according to the value of the corresponding current optimized adjustment feature to more accurately reflect the correlation between the feature and the similar module area. Similarly, the server will also make similar adjustments to the abnormal adjustment association identifier. The server determines the current association identifier based on the current optimization adjustment association identifier. When the number of optimization adjustments of the current optimization adjustment association identifier exceeds the preset adjustment number threshold, the server uses the optimization adjustment association identifier as the current association identifier. At the same time, the server also determines the abnormal association identifier based on the abnormal optimization adjustment association identifier. When the number of optimization adjustments of the abnormal optimization adjustment association identifier exceeds the preset adjustment number threshold, the server uses the optimization adjustment association identifier as the abnormal association identifier. For example, if the server optimizes ID1_0.85 multiple times and the number of adjustments exceeds the preset threshold (such as 5 times), the server uses the last obtained optimization adjustment association identifier (such as ID1_0.90) as the current association identifier. Similarly, the server will also determine the abnormal association identifier. Through the above steps, the server can further optimize and adjust according to the current adjustment features, abnormal adjustment features and their corresponding adjustment association identifiers, so as to obtain more accurate and meaningful current association identifiers and abnormal association identifiers. This helps in subsequent feature comparison and association analysis to improve the accuracy of abnormal identification and risk assessment.

[0110] In the embodiment of the present invention, the feature extraction processing of the current hydraulic structure module to obtain the features of the current hydraulic structure module and the feature extraction processing of the abnormal hydraulic structure module to obtain the features of the abnormal hydraulic structure module can be implemented through the following examples.

[0111] Performing feature extraction processing on the current hydraulic structure module to obtain the current hydraulic structure module feature and the comparison hydraulic structure module feature, and performing feature extraction processing on the abnormal hydraulic structure module to obtain the abnormal hydraulic structure module feature and the reference hydraulic structure module feature, the module scale of the current hydraulic structure module feature is different from the module scale of the comparison hydraulic structure module feature, and the module scale of the abnormal hydraulic structure module feature is different from the module scale of the reference hydraulic structure module feature;

[0112] Wherein, after obtaining the first association result between the current hydraulic structure module and the abnormal hydraulic structure module, the method further includes:

[0113] According to the first association result, the module scale of the current hydraulic structure module feature, the module scale of the comparison hydraulic structure module feature, and the comparison hydraulic structure module feature, feature screening is performed on the comparison hydraulic structure module feature to obtain a second number of hydraulic structure sub-modules corresponding to the second number of current module sub-features, and according to the first association result, the module scale of the abnormal hydraulic structure module feature, the module scale of the reference hydraulic structure module feature, and the reference hydraulic structure module feature, feature screening is performed on the reference hydraulic structure module feature to obtain a third number of hydraulic structure sub-modules corresponding to the third number of abnormal module sub-features;

[0114] The second number of hydraulic structure sub-modules are adjusted according to the second number of hydraulic structure sub-modules and the third number of hydraulic structure sub-modules to obtain a second number of adjusted sub-modules, and the third number of hydraulic structure sub-modules are adjusted according to the second number of hydraulic structure sub-modules and the third number of hydraulic structure sub-modules to obtain a third number of adjusted sub-modules;

[0115] Identifying the geometric midpoints of the second number of adjustment submodules to obtain a second number of geometric midpoints, and adjusting the size of the third number of adjustment submodules according to the scale difference between the comparison hydraulic structure module feature and the reference hydraulic structure module feature to obtain a third number of adjusted submodules;

[0116] Performing spatial mapping processing on the third number of adjusted sub-modules to obtain a third number of spatial coordinates corresponding to the third number of adjusted sub-modules;

[0117] The second number of geometric midpoints and the third number of spatial coordinates are matched to obtain a second association result between the current hydraulic structure module and the abnormal hydraulic structure module.

[0118] In an embodiment of the present invention, exemplarily, the server performs feature extraction processing on the current hydraulic structure module to obtain the current hydraulic structure module features and the comparative hydraulic structure module features. At the same time, the server also performs feature extraction processing on the abnormal hydraulic structure module to obtain the abnormal hydraulic structure module features and the reference hydraulic structure module features. For example, the server can use a deep learning algorithm to perform feature extraction on the image data of the current hydraulic structure module to obtain a set of feature vectors describing the geological structure of the module as the current hydraulic structure module features. At the same time, the server can also extract a comparative hydraulic structure module feature of a different scale for subsequent comparison and screening. The server performs feature screening on the comparative hydraulic structure module feature based on the first association result, the module scale of the current hydraulic structure module feature, the module scale of the comparative hydraulic structure module feature, and the comparative hydraulic structure module feature to obtain a second number of hydraulic structure sub-modules corresponding to the second number of current module sub-features. At the same time, the server also performs feature screening on the reference hydraulic structure module feature according to the first association result, the module scale of the abnormal hydraulic structure module feature, the module scale of the reference hydraulic structure module feature, and the reference hydraulic structure module feature, and obtains a third number of hydraulic structure sub-modules corresponding to the third number of abnormal module sub-features. For example, the server can screen out the second number of sub-features that are most similar to the current hydraulic structure module feature from the comparison hydraulic structure module feature according to the similarity and difference indicated in the first association result, and extract their corresponding sub-modules. Similarly, the server will also screen out the third number of sub-modules that are most relevant to the abnormal hydraulic structure module feature from the reference hydraulic structure module feature. The server adjusts the second number of hydraulic structure sub-modules according to the second number of hydraulic structure sub-modules and the third number of hydraulic structure sub-modules to obtain the second number of adjusted sub-modules. At the same time, the server also adjusts the third number of hydraulic structure sub-modules according to the two sub-module sets to obtain the third number of adjusted sub-modules. For example, the server can fine-tune the second number of hydraulic structure sub-modules based on the similarities and differences between the two sub-module sets to make them closer to the third number of hydraulic structure sub-modules. Similarly, the server will also make similar adjustments to the third number of hydraulic structure sub-modules. The server identifies the geometric midpoints of the second number of adjusted sub-modules to obtain the second number of geometric midpoints. At the same time, the server also adjusts the size of the third number of adjusted sub-modules based on the scale difference between the comparison hydraulic structure module features and the reference hydraulic structure module features to obtain the third number of adjusted sub-modules. For example, the server can calculate the geometric center of each adjusted sub-module as the geometric midpoint. At the same time, the server can also perform size adjustments such as scaling or stretching on the third number of adjusted sub-modules based on the scale difference between the comparison module and the reference module.The server performs spatial mapping processing on the third number of adjusted sub-modules to obtain a third number of spatial coordinates corresponding to the third number of adjusted sub-modules. Then, the server matches the second number of geometric midpoints with the third number of spatial coordinates to obtain a second association result between the current hydraulic structure module and the abnormal hydraulic structure module. For example, the server can map each adjusted sub-module to a coordinate point in a three-dimensional space. Then, the server can use a matching algorithm (such as nearest neighbor matching) to match the second number of geometric midpoints with the third number of spatial coordinates to obtain the association relationship between them as a second association result. This second association result can be used in further risk assessment or anomaly detection tasks.

[0119] In the embodiment of the present invention, performing spatial mapping processing on the third number of adjusted sub-modules to obtain a third number of spatial coordinates corresponding to the third number of adjusted sub-modules includes:

[0120] Obtaining a spatial influence coefficient of each feature in a first adjusted submodule in the third number of adjusted submodules;

[0121] A linear superposition operation is performed on the spatial influence coefficient and initial coordinates of each feature in the first adjusted sub-module to obtain the spatial coordinates of the first adjusted sub-module.

[0122] In an embodiment of the present invention, exemplarily, the server first obtains the first adjusted submodule in the third number of adjusted submodules. This submodule is a set of geological structure data that has been feature extracted, screened and adjusted. Next, the server analyzes the spatial attributes of each feature in the first adjusted submodule and calculates their degree of influence on the overall spatial position, that is, the spatial influence coefficient. These features may include the thickness, lithology, inclination, etc. of the geological layer, which have different influence weights on the position of the submodule in space. The server performs a linear superposition operation on the spatial influence coefficient of each feature in the first adjusted submodule and a set of initial coordinates. This initial coordinate can be set according to the overall layout of the hydraulic structure or the reference coordinate system. For example, if the first adjusted submodule contains three features: thickness (spatial influence coefficient is 0.5), lithology (spatial influence coefficient is 0.3) and inclination (spatial influence coefficient is 0.2), and the initial coordinates are (100, 200, 300), then the server will calculate a new spatial coordinate, where the x coordinate is 100 plus the thickness effect, the y coordinate is 200 plus the lithology effect, and the z coordinate is 300 plus the inclination effect. Through linear superposition operation, the server obtains the exact coordinates of the first adjusted submodule in space. This coordinate reflects the relative position of the submodule in the hydraulic structure. The server repeats the above steps to calculate the corresponding spatial coordinates for each submodule in the third number of adjusted submodules. Finally, the server obtains a set of spatial coordinates corresponding to the adjusted submodules. In this way, the server completes the spatial mapping processing of the third number of adjusted submodules and obtains their corresponding spatial coordinates. These coordinates can be used in subsequent tasks such as geological structure analysis, risk assessment or visualization.

[0123] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned hydraulic structure risk assessment method based on deep learning and digital twins. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0124] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.

Claims

1. A hydraulic structure risk assessment method based on deep learning and digital twins, characterized in that: include: Obtain hydraulic structure parameters of the hydraulic structure to be evaluated; Inputting the hydraulic structure parameters into a preset digital twin platform, constructing a target three-dimensional hydraulic structure model corresponding to the hydraulic structure to be evaluated; the target three-dimensional hydraulic structure model is composed of multiple hydraulic structure modules; Calling the key module identification model and the module type identification model configured in the preset digital twin platform to analyze the target three-dimensional hydraulic structure model, and obtaining the module type identification result of the target three-dimensional hydraulic structure model; According to the module type identification result, determining a current hydraulic structure module from the multiple hydraulic structure modules; Performing abnormality identification on the current hydraulic structure module to obtain a hydraulic structure risk assessment result of the target three-dimensional hydraulic structure model; The key module identification model includes a range selection model and a key range positioning unit, and the module type identification model includes a key range detection unit; The calling of the key module identification model and the module type identification model configured in the preset digital twin platform to analyze the target three-dimensional hydraulic structure model to obtain the module type identification result of the target three-dimensional hydraulic structure model includes: Obtaining a pending selection range in the target three-dimensional hydraulic structure model through the range selection model, and obtaining a range selection probability value corresponding to the pending selection range through a range selection probability component of the range selection model; Selecting the pending selection range according to the range selection probability value; The key module identification processing is performed on the selected pending selection range by the key range positioning unit to determine the key module identification range in the target three-dimensional hydraulic structure model; the key module identification model is a model trained based on a standard three-dimensional hydraulic structure model instance configured with a key module identification target value, and the standard three-dimensional hydraulic structure model instance is a three-dimensional model containing a general type of module; Determine, according to the newly added three-dimensional hydraulic structure model instance including the standard three-dimensional hydraulic structure model instance and the preset three-dimensional hydraulic structure model, the module performance characteristics of the module type included in the newly added three-dimensional hydraulic structure model instance; the preset three-dimensional hydraulic structure model is a three-dimensional model including a module of a preset type; According to the module basic features corresponding to each of the key module identification ranges, the key module identification range is detected by a key range detection unit to obtain a key range detection feature; Determine, according to the key range detection features and the module performance features of each module type, the inferred range position and the gap between the inferred ranges corresponding to different inferred range positions; Processing the gaps between the inferred ranges by a range comparison component in the key range detection unit to reduce the gaps between inferred ranges of the same type and increase the gaps between inferred ranges of different types; According to the gaps between the processed inference ranges, the type scores corresponding to the positions of each inference range are obtained; Determining a module type recognition result of the target three-dimensional hydraulic structure model according to the inferred range position and the type score corresponding to the inferred range position; Among them, the module type identification model is a model obtained by iteratively training the preset module type identification model according to the newly added three-dimensional hydraulic structure model instance; the newly added three-dimensional hydraulic structure model instance is configured with a module type target value and a key module identification target value.

2. The method according to claim 1, characterized in that The method further comprises: Obtaining a general type three-dimensional hydraulic structure model instance configured with a key module identification target value and an initial scope selection model, wherein the initial scope selection model includes a module type identification component; Replacing the module type identification component of the initial range selection model with a range selection probability component; Constructing an initial key module identification model according to the initial range selection model after module replacement; The initial key module identification model is trained by using the general type three-dimensional hydraulic structure model instance configured with the key module identification target value to obtain the key module identification model.

3. The method according to claim 1, characterized in that The method further comprises: Obtaining a general type three-dimensional hydraulic structure model instance configured with a module type target value and a key module identification target value, a preset type three-dimensional hydraulic structure model instance configured with a module type target value and a key module identification target value, and a preset module type identification model; Replacing the range selection probability component of the key range detection unit in the preset module type identification model with a module type identification component; The key range detection unit after module replacement is iteratively trained by using the general type three-dimensional hydraulic structure model instance configured with module type target value and key module identification target value, and the preset type three-dimensional hydraulic structure model instance configured with module type target value and key module identification target value to obtain a module type identification model.

4. The method according to claim 1, characterized in that The method of determining the module performance characteristics of the module type included in the newly added three-dimensional hydraulic structure model instance according to the newly added three-dimensional hydraulic structure model instance including the standard three-dimensional hydraulic structure model instance and the preset three-dimensional hydraulic structure model includes: Acquire a newly added three-dimensional hydraulic structure model instance including the standard three-dimensional hydraulic structure model instance and the preset three-dimensional hydraulic structure model, wherein the newly added three-dimensional hydraulic structure model instance is configured with a type target value; Extracting multi-view features of the newly added three-dimensional hydraulic structure model instance; According to the type target value, weighted averaging is performed on multi-view features of newly added three-dimensional hydraulic structure model instances of the same type to obtain module performance features.

5. The method according to claim 1, characterized in that The abnormality identification of the current hydraulic structure module to obtain the hydraulic structure risk assessment result of the target three-dimensional hydraulic structure model includes: Performing feature extraction processing on the current hydraulic structure module to obtain current hydraulic structure module features and comparative hydraulic structure module features, and performing feature extraction processing on the abnormal hydraulic structure module to obtain abnormal hydraulic structure module features and reference hydraulic structure module features, the module scale of the current hydraulic structure module features is different from the module scale of the comparative hydraulic structure module features, and the module scale of the abnormal hydraulic structure module features is different from the module scale of the reference hydraulic structure module features; Performing feature analysis on the current hydraulic structure module feature to obtain the correlation coefficient of each current module sub-feature in the current hydraulic structure module feature, and performing feature analysis on the abnormal hydraulic structure module feature to obtain the correlation coefficient of each abnormal module sub-feature in the abnormal hydraulic structure module feature, wherein the correlation coefficient is used to indicate the influence on the hydraulic structure module correlation; Determine the first quantity according to a preset proportion and the module scale of the current hydraulic structure module characteristic, or determine the first quantity according to the preset proportion and the module scale of the abnormal hydraulic structure module characteristic; According to the correlation coefficients of the current module sub-features, according to the sequence of the correlation coefficients from high to low, the first first number of current module sub-features in the current hydraulic structure module feature are determined as the first number of current module sub-features, and according to the correlation coefficients of the abnormal module sub-features, according to the sequence of the correlation coefficients from high to low, the first first number of abnormal module sub-features in the abnormal hydraulic structure module feature are determined as the first number of abnormal module sub-features; Determine current original association identifiers corresponding to a first number of current module sub-features, and determine abnormal original association identifiers corresponding to a first number of abnormal module sub-features; Applying the current original association identifier to a first number of current module sub-features to obtain a first number of current processing features, and applying the abnormal original association identifier to a first number of abnormal module sub-features to obtain a first number of abnormal processing features; Performing dynamic weight allocation processing on a first number of current processing features using a first dynamic weight allocation network, and processing a first number of abnormal processing features and a first number of features output by the first dynamic weight allocation network using a first associated weight allocation network to obtain a first number of current adjustment features, and performing dynamic weight allocation processing on a first number of abnormal processing features using a second dynamic weight allocation network, and processing a first number of current processing features and a first number of features output by the second dynamic weight allocation network using a second associated weight allocation network to obtain a first number of abnormal adjustment features; According to the first number of current adjustment features, the current original association identifier is adjusted to obtain a current adjustment association identifier, and according to the first number of abnormal adjustment features, the abnormal original association identifier is adjusted to obtain an abnormal adjustment association identifier; Determine a current association identifier according to the current adjustment association identifier, and determine an abnormal association identifier according to the abnormal adjustment association identifier, wherein the current association identifier is used to obtain a second number of current module sub-features having similar module regions with the first number of abnormal module sub-features in the first number of current module sub-features, and the abnormal association identifier is used to enhance a third number of abnormal module sub-features having similar module regions with the first number of current module sub-features in the first number of abnormal module sub-features; Performing feature comparison on the second number of current module sub-features and the third number of abnormal module sub-features to obtain a first association result between the current hydraulic structure module and the abnormal hydraulic structure module; Determine a risk assessment result of the current hydraulic structure module according to the first association result, and use the risk assessment result as a hydraulic structure risk assessment result of the target three-dimensional hydraulic structure model; Wherein, after obtaining the first association result between the current hydraulic structure module and the abnormal hydraulic structure module, the method further includes: According to the first association result, the module scale of the current hydraulic structure module feature, the module scale of the comparison hydraulic structure module feature, and the comparison hydraulic structure module feature, feature screening is performed on the comparison hydraulic structure module feature to obtain a second number of hydraulic structure sub-modules corresponding to the second number of current module sub-features, and according to the first association result, the module scale of the abnormal hydraulic structure module feature, the module scale of the reference hydraulic structure module feature, and the reference hydraulic structure module feature, feature screening is performed on the reference hydraulic structure module feature to obtain a third number of hydraulic structure sub-modules corresponding to the third number of abnormal module sub-features; The second number of hydraulic structure sub-modules are adjusted according to the second number of hydraulic structure sub-modules and the third number of hydraulic structure sub-modules to obtain a second number of adjusted sub-modules, and the third number of hydraulic structure sub-modules are adjusted according to the second number of hydraulic structure sub-modules and the third number of hydraulic structure sub-modules to obtain a third number of adjusted sub-modules; Identifying the geometric midpoints of the second number of adjustment submodules to obtain the second number of geometric midpoints, and adjusting the size of the third number of adjustment submodules according to the scale difference between the comparison hydraulic structure module feature and the reference hydraulic structure module feature to obtain the third number of adjusted submodules; Obtaining a spatial influence coefficient of each feature in a first adjusted submodule in a third number of adjusted submodules; Performing a linear superposition operation on the spatial influence coefficient and initial coordinates of each feature in the first adjusted sub-module to obtain the spatial coordinates of the first adjusted sub-module; Matching the second number of geometric midpoints and the third number of spatial coordinates, a second association result between the current hydraulic structure module and the abnormal hydraulic structure module; the abnormal hydraulic structure module is preset.

6. The method according to claim 5, characterized in that Before performing feature extraction processing on the current hydraulic structure module to obtain the features of the current hydraulic structure module and the features of the comparison hydraulic structure module, and performing feature extraction processing on the abnormal hydraulic structure module to obtain the features of the abnormal hydraulic structure module and the features of the reference hydraulic structure module, the method further includes: If the module scale of the current hydraulic structure module and the module scale of the abnormal hydraulic structure module have not been unified, unify the module scales of the current hydraulic structure module and the abnormal hydraulic structure module to the same module scale; The determining of the current original association identifiers corresponding to the first number of current module sub-features and the determining of the abnormal original association identifiers corresponding to the first number of abnormal module sub-features include: Determine the current original association identifier according to the elements supplemented in the scale unified processing of the current hydraulic structure module, wherein the data corresponding to the features of the elements supplemented in the first number of current module sub-features in the current original association identifier is a reference value; According to the elements supplemented in the unified scale processing of the abnormal hydraulic structure module, the abnormal original association identifier is determined, and in the abnormal original association identifier, the data corresponding to the characteristics of the elements supplemented in the first number of abnormal module sub-features is the reference value.

7. The method according to claim 5, characterized in that The adjusting the current original association identifier according to the first number of current adjustment features to obtain the current adjustment association identifier, and adjusting the abnormal original association identifier according to the first number of abnormal adjustment features to obtain the abnormal adjustment association identifier, include: Performing feature analysis on the first number of current adjustment features to obtain two confidences corresponding to each of the first number of current adjustment features, and performing feature analysis on the first number of abnormal adjustment features to obtain two confidences corresponding to each of the first number of abnormal adjustment features, wherein the two confidences are used to indicate the confidence of the feature belonging to the similar module area and the confidence of the feature not belonging to the similar module area; Determine a first number of first pointers corresponding to a first number of current adjustment features according to the two confidences corresponding to each current adjustment feature, and determine a first number of second pointers corresponding to a first number of abnormal adjustment features according to the two confidences corresponding to each abnormal adjustment feature, wherein the first pointer is used to indicate a locator of a layer corresponding to a maximum confidence value among the two confidences corresponding to the current adjustment feature, and the second pointer is used to indicate a locator of a layer corresponding to a maximum confidence value among the two confidences corresponding to the abnormal adjustment feature; Performing an inverse operation on a first number of first pointers to obtain a first number of first confidences, and performing an inverse operation on a first number of second pointers to obtain a first number of second confidences; The current adjusted association identifier is obtained by applying the current original association identifier to a first number of first confidences, and the abnormal adjusted association identifier is obtained by applying the abnormal original association identifier to a first number of second confidences.

8. The method according to claim 5, characterized in that The adjusting the current original association identifier according to the first number of current adjustment features to obtain the current adjustment association identifier, and adjusting the abnormal original association identifier according to the first number of abnormal adjustment features to obtain the abnormal adjustment association identifier, include: According to the first number of current adjustment features, the first number of abnormal adjustment features, the current adjustment association identifier, and the abnormal adjustment association identifier, the first number of current adjustment features are adjusted to obtain a first number of current optimized adjustment features, and according to the first number of current adjustment features, the first number of abnormal adjustment features, the current adjustment association identifier, and the abnormal adjustment association identifier, the first number of abnormal adjustment features are adjusted to obtain a first number of abnormal optimized adjustment features; According to the first number of current optimization adjustment features, the current adjustment association identifier is adjusted to obtain a current optimization adjustment association identifier, and according to the first number of abnormal optimization adjustment features, the abnormal adjustment association identifier is adjusted to obtain an abnormal optimization adjustment association identifier; The current association identifier is determined according to the current optimization adjustment association identifier, and the abnormal association identifier is determined according to the abnormal optimization adjustment association identifier, wherein the current association identifier is an adjustment association identifier obtained when the optimization adjustment times of the current original association identifier exceeds a preset adjustment times threshold, and the abnormal association identifier is an adjustment association identifier obtained when the optimization adjustment times of the abnormal original association identifier exceeds the preset adjustment times threshold.

9. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 8.

Images