Method and system for evaluating the stability of a flexible apron structure impacted by a large rock-mud flow

Abstract

The application provides a method and system for evaluating the stability of a large boulder mudslide-impacted flexible apron structure. A feature anomaly detection unit generates a confidence coefficient between each structural feature vector and a feature cleaning unit. The confidence coefficient represents the correlation between the structural feature vector and the feature cleaning unit. Then, the confidence coefficient is used to select Y structural feature vectors from X structural feature vectors to generate a spliced feature vector. The spliced feature vector is loaded into the feature cleaning unit connected to the feature anomaly detection unit. The feature cleaning unit outputs a feature vector used to generate a structural stability evaluation result matching the X flexible apron structure incident description information. In this way, the accuracy and precision of the evaluation can be improved, and the flexible apron structure can withstand the impact of large boulder mudslides, ensuring the safety of people's lives and property.

Description

Technical Field

[0001] This application relates to the field of stability assessment technology, and more specifically, to a method and system for assessing the stability of a tough revetment structure subjected to impact from large rocks and debris flows. Background Technology

[0002] Resilient apron structures are protective structures in hydraulic engineering with "resilience" as their core design concept. They are mainly deployed in stilling basins, apron sections, or riverbed scour zones downstream of spillway structures (such as dams, sluices, and spillways). Through a synergistic mechanism of "damage resistance, energy absorption, and recoverability," they resist adverse factors such as high-speed water flow scouring, extreme hydrological events, and material degradation, ensuring the safety of the downstream riverbed and the foundations of the structures. Unlike traditional aprons that rely solely on strength or stiffness to resist damage, resilient aprons emphasize the ability to "survive damage and recover after disturbance," representing a typical application of the "safety and resilience" concept in modern hydraulic engineering.

[0003] In practice, assessing the stability of resilient tanker structures is crucial for ensuring the safety of people and property. However, how to assess resilient tanker structures remains a technical challenge that is currently difficult to resolve. Summary of the Invention

[0004] To address the technical problems existing in related technologies, this application provides a method and system for evaluating the stability of a tough revetment structure subjected to impact from large rocks and debris flows.

[0005] Firstly, a method for evaluating the stability of a tough revetment structure impacted by large debris flows is provided, the method comprising:

[0006] Obtain X resilient tank structure descriptions and load them into a stability analysis thread; the data types corresponding to the X resilient tank structure descriptions are inconsistent; the stability analysis thread includes a convolution unit, a feature cleaning unit, and a feature anomaly detection unit; X is an integer greater than 0;

[0007] The convolutional unit performs convolution processing on the description information of the X tough sluice structure items to obtain X structural feature vectors, and loads the X structural feature vectors into the feature cleaning unit connected to the convolutional unit.

[0008] The X structural feature vectors are loaded into the feature anomaly detection unit, and the feature anomaly detection unit generates confidence coefficients between each structural feature vector and the feature cleaning unit; the confidence coefficients are used to represent the correlation between the structural feature vectors and the feature cleaning unit.

[0009] Y structural feature vectors are selected from the X structural feature vectors using the confidence coefficient. A spliced ​​feature vector is generated from the Y structural feature vectors. The spliced ​​feature vector is loaded into a feature cleaning unit connected to the feature anomaly detection unit. The feature cleaning unit outputs a feature vector used to generate a structural stability assessment result that matches the structural description information of the X toughness tank structures. Y is an integer greater than 0 that is not greater than X.

[0010] In this application, the step of loading the X structural feature vectors into the feature anomaly detection unit, and generating confidence coefficients between each structural feature vector and the feature cleaning unit through the feature anomaly detection unit, includes:

[0011] The X structural feature vectors are loaded into the feature anomaly detection unit, and the X structural feature vectors are simplified to obtain X simplified feature vectors; the simplification methods of the X simplified feature vectors are all the same.

[0012] Obtain the anomaly detection input queue and anomaly detection output queue that match the feature cleaning unit in the feature anomaly detection unit. Perform function processing on the X simplified feature vectors with the anomaly detection input queue respectively to obtain X anomaly detection feature vectors. The anomaly detection input queue and the anomaly detection output queue are used together to represent the stability elements of the feature cleaning unit.

[0013] Using Leaky ReLU, adaptive data distribution vectors corresponding to the X anomaly detection feature vectors are generated respectively. The X adaptive data distribution vectors are then processed by the anomaly detection output queue to obtain X offset coefficients.

[0014] By projecting the X offset coefficients onto the decision function, the confidence coefficients between each structural feature vector and the feature cleaning unit are obtained.

[0015] In this application, the simplification process of the X structural feature vectors to obtain X simplified feature vectors includes:

[0016] Obtain the attribute description values ​​corresponding to the X structural feature vectors respectively, and based on the X attribute description values, obtain the attribute coefficient queues corresponding to the X structural feature vectors respectively;

[0017] By processing the corresponding structural feature vectors through X queues of attribute coefficients, X simplified feature vectors are obtained.

[0018] In this application, the confidence coefficient includes a first importance parameter, a second importance parameter, and a third importance parameter; the step of projecting the X offset coefficients through a decision function to obtain the confidence coefficient between each structural feature vector and the feature cleaning unit includes:

[0019] Obtain the first and second calculated specified values ​​corresponding to the decision function;

[0020] If the offset coefficient corresponding to the structural feature vector is less than the first calculated specified value, then the confidence coefficient corresponding to the structural feature vector is set as the first importance parameter;

[0021] If the offset coefficient corresponding to the structural feature vector is greater than the first calculated specified value and less than the second calculated specified value, then the second importance parameter is generated based on the offset coefficient, and the confidence coefficient corresponding to the structural feature vector is set as the second importance parameter.

[0022] If the offset coefficient corresponding to the structural feature vector is greater than the second calculated specified value, then the confidence coefficient corresponding to the structural feature vector is set as the third importance parameter;

[0023] The structural feature vectors represented by the first importance parameter, the second importance parameter, and the third importance parameter have different association relationships with the feature cleaning unit; the association relationship corresponding to the first importance parameter is less than the association relationship corresponding to the second importance parameter, and the association relationship corresponding to the second importance parameter is less than the association relationship corresponding to the third importance parameter.

[0024] In this application, the confidence coefficient includes importance parameters corresponding to the X structural feature vectors respectively; the step of selecting Y structural feature vectors from the X structural feature vectors using the confidence coefficient includes:

[0025] Among the X structural feature vectors, the structural feature vectors with confidence coefficients greater than the specified weight values ​​are taken as the Y structural feature vectors;

[0026] The step of generating a concatenated feature vector from the Y structural feature vectors includes:

[0027] A potential importance queue is generated using the importance parameters corresponding to the Y structural feature vectors, and simplified feature vectors corresponding to the Y structural feature vectors are obtained; the simplified feature vectors are obtained by simplifying the structural feature vectors.

[0028] The arrangement of Y simplified feature vectors is processed by a function with the potential importance queue to obtain a concatenated feature vector.

[0029] In this application, the number of feature cleaning units and feature anomaly detection units in the stability analysis thread is A. Each feature cleaning unit is connected to a feature anomaly detection unit, and the feature anomaly detection units connected to each feature cleaning unit are different from each other. The A feature cleaning units are connected serially. The input of each feature anomaly detection unit is the X structural feature vectors, and the output of each feature anomaly detection unit is used to load the connected feature cleaning unit. The A feature cleaning units include feature cleaning unit Lz and feature cleaning unit Lz-0, where z is an integer greater than 0 that is less than A. The A feature anomaly detection units include feature anomaly detection unit Kz connected to the feature cleaning unit Lz. The feature vectors output by the feature cleaning units to generate structural stability assessment results that match the description information of the X toughness shield structures include:

[0030] In the feature cleaning unit Lz, a feature vector is output based on the target feature vector corresponding to the feature cleaning unit Lz and the concatenated feature vector input to the feature anomaly detection unit Kz; if the feature cleaning unit Lz is a feature cleaning unit connected to the convolution unit, then the target feature vector is the X structural feature vectors; if the feature cleaning unit Lz is connected to the feature cleaning unit Lz-0, then the target feature vector is the feature vector output by the feature cleaning unit Lz-0.

[0031] The method further includes:

[0032] If the feature cleaning unit Lz is the last feature cleaning unit among the A feature cleaning units, then the feature vector output by the feature cleaning unit Lz generates a structural stability assessment result that matches the description information of the X tough abutment structures.

[0033] In this application, the stability analysis thread includes a fully connected layer and R prediction units, where R is an integer greater than 0. Each prediction unit includes a feature cleaning unit and a feature anomaly detection unit, and the data types corresponding to the R prediction units are inconsistent. Before loading the X structural feature vectors into the feature anomaly detection unit, the method further includes:

[0034] The X structural feature vectors are loaded into a fully connected layer, and the fully connected layer generates fully connected computation values ​​corresponding to each prediction unit; the fully connected computation values ​​are used to represent the association between the prediction unit and the X structural feature vectors;

[0035] Based on R fully connected computation values, the R prediction units are distributed, and a target prediction unit is determined from the distributed R prediction units. The undetermined feature cleaning unit in the target prediction unit is taken as the feature cleaning unit, and the undetermined feature anomaly detection unit in the target prediction unit is taken as the feature anomaly detection unit.

[0036] In this application, the step of outputting a feature vector through the feature cleaning unit to generate a structural stability assessment result that matches the description information of the X tough abutment structures includes:

[0037] In the feature cleaning unit, the feature vectors of historical debris flow impact contents corresponding to U historical debris flow impact contents in the historical debris flow impact database are obtained, and the feature vectors of the U historical debris flow impact contents are coupled to obtain the feature arrangement of historical debris flow impact contents; U is an integer greater than 0.

[0038] The spliced ​​feature vector and the X structural feature vectors are spliced ​​together to obtain a spliced ​​feature vector. The historical debris flow impact content feature arrangement and the spliced ​​feature vector are filtered to obtain a feature vector that covers the filtered result vector.

[0039] The method further includes:

[0040] Based on the filtering scores corresponding to the U historical debris flow impact contents in the filtering result vector, the estimated interaction coefficients corresponding to the U historical debris flow impact contents are generated. The estimated interaction coefficients corresponding to the U historical debris flow impact contents are used as the structural stability assessment results that match the description information of the X tough abutment structures.

[0041] In this application, the filtering process performed on the historical debris flow impact content feature arrangement and the spliced ​​feature vector to obtain a feature vector covering the filtered result vector includes:

[0042] Obtain the search coefficient queue, importance coefficient queue, and derivative queue in the feature cleaning unit; the search coefficient queue, importance coefficient queue, and derivative queue are all queues composed of learnable coefficients;

[0043] The spliced ​​feature vector is processed with the search coefficient queue to obtain the search vector. The historical debris flow impact content feature arrangement is processed with the importance coefficient queue to obtain the key description vector. The historical debris flow impact content feature arrangement is processed with the derivative queue to obtain the function processing result vector.

[0044] A filtering result vector is generated based on the search vector and the key description vector. The filtering result vector is simplified based on the vector data volume of the key description vector. The simplified filtering result vector is then simplified without dimensions to obtain a filtering weight vector. The filtering weight vector and the function processing result vector are then processed by a function to obtain a feature vector that covers the filtering result vector.

[0045] In this application, the process of concatenating the concatenated feature vector and the X structural feature vectors to obtain the concatenated feature vector includes:

[0046] If the concatenated feature vector and the X structural feature vectors meet the isomorphic feature requirement, then the concatenated feature vector and the X structural feature vectors are vector-wise added to obtain the concatenated feature vector; the isomorphic feature requirement means that the concatenated feature vector and the X structural feature vectors belong to the same semantic space;

[0047] If the concatenated feature vector does not meet the isomorphic feature requirement with the X structural feature vectors, then the concatenated feature vector and the X structural feature vectors are vector-coupled to obtain the concatenated feature vector.

[0048] Secondly, a stability assessment system for a tough revetment structure impacted by large rocks and debris flows is provided, comprising a processor and a memory that communicate with each other, wherein the processor is used to read a computer program from the memory and execute it to implement the above-mentioned method.

[0049] This application provides a method and system for assessing the stability of resilient abutment structures impacted by large boulders and debris flows. The method involves using a convolutional unit in a stability analysis thread to process X disparate descriptions of resilient abutment structure issues, resulting in X structural feature vectors. The stability analysis thread also includes a feature cleaning unit and a feature anomaly detection unit. The X structural feature vectors are loaded into the feature cleaning unit connected to the convolutional unit. A corresponding feature anomaly detection unit is configured for the feature cleaning unit. The feature anomaly detection unit generates confidence coefficients between each structural feature vector and the feature cleaning unit, representing the correlation between the structural feature vectors and the feature cleaning unit. Then, Y structural feature vectors are selected from the X structural feature vectors using the confidence coefficients to generate a concatenated feature vector. This concatenated feature vector is loaded into the feature cleaning unit connected to the feature anomaly detection unit. The feature cleaning unit outputs a feature vector used to generate a structural stability assessment result that matches the descriptions of the X resilient abutment structures. This will improve the accuracy and precision of the assessment, ensuring that the resilient abutment structure can withstand the impact of large rocks and debris flows, thus protecting people's lives and property. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart illustrating a method for evaluating the stability of a tough revetment structure subjected to impact from large rocks and debris flows, as provided in an embodiment of this application. Detailed Implementation

[0052] To better understand the above technical solutions, the technical solutions of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this application and the specific features in the embodiments are detailed descriptions of the technical solutions of this application, rather than limitations on the technical solutions of this application. In the absence of conflict, the embodiments of this application and the technical features in the embodiments can be combined with each other.

[0053] Please see Figure 1 This paper presents a method for evaluating the stability of a tough revetment structure subjected to impact from large rock debris flows. The method may include the technical solutions described in steps S101-S104.

[0054] Step S101: Obtain X resilient tank structure description information and load the X resilient tank structure description information into the stability analysis thread; the data types corresponding to the X resilient tank structure description information are inconsistent; the stability analysis thread includes a convolution unit, a feature cleaning unit, and a feature anomaly detection unit; X is an integer greater than 0;

[0055] For example, information describing X aspects of a toughened apron structure can be obtained through ultrasonic testing or other specialized equipment. The data types in this application include: material property information, mechanical and load response information, etc.

[0056] In this application, the convolutional unit refers to the feature extraction unit;

[0057] In this application, the feature cleaning unit removes unimportant information (such as: silt abrasion and minor environmental factors that are not directly affected).

[0058] The characteristic anomaly detection unit in this application includes: extreme natural environment anomalies, material and structural anomalies, external load anomalies, etc.

[0059] The description information of X toughness tank structure items is loaded into the stability analysis thread. The stability analysis thread can be trained on each feature cleaning unit in the Baseline Model under the premise of any data type Baseline Model. The stability analysis thread can include convolutional units, feature cleaning units and feature anomaly detection units.

[0060] Step S102: The X structural description information of the toughness tank structure is processed by the convolution unit to obtain X structural feature vectors. The X structural feature vectors are then loaded into the feature cleaning unit connected to the convolution unit.

[0061] The structural feature vector includes the foundation layer (e.g., bearing capacity characteristics, stable foundation characteristics), the transition layer (e.g., buffering loads, coordinating the deformation of the foundation and the protective layer), the tough protective layer (e.g., directly bearing water scouring and impact), and the auxiliary system (e.g., improving durability and adaptability).

[0062] Step S103: Load X structural feature vectors into the feature anomaly detection unit, and generate confidence coefficients between each structural feature vector and the feature cleaning unit through the feature anomaly detection unit; the confidence coefficients are used to represent the correlation between the structural feature vectors and the feature cleaning unit.

[0063] For example, X structural feature vectors are loaded into a feature anomaly detection unit. The multilayer perceptron and binary classification function within the feature anomaly detection unit generate confidence coefficients between each structural feature vector and the feature cleaning unit. These confidence coefficients represent the correlation between the structural feature vectors and the feature cleaning unit, indicating the importance of cross-connection. For instance, the confidence coefficient can be a value between [0,1]. A confidence coefficient of 0 indicates that the corresponding structural feature vector is unrelated to the feature cleaning unit, while a confidence coefficient greater than 0 indicates that the corresponding structural feature vector is related to the feature cleaning unit. A higher confidence coefficient indicates a stronger correlation between the structural feature vector and the feature cleaning unit.

[0064] Step S104: Select Y structural feature vectors from X structural feature vectors using confidence coefficients; generate a spliced ​​feature vector from the Y structural feature vectors; load the spliced ​​feature vector into the feature cleaning unit connected to the feature anomaly detection unit; and output a feature vector from the feature cleaning unit to generate a structural stability assessment result that matches the structural description information of X toughness shield structures; Y is an integer greater than 0 that is not greater than X.

[0065] For example, Y structural feature vectors are selected from X structural feature vectors using X confidence coefficients. For instance, structural feature vectors with confidence coefficients greater than 0 among the X structural feature vectors can be used as the Y structural feature vectors. A concatenated feature vector is generated from the Y structural feature vectors and loaded into a feature cleaning unit connected to the feature anomaly detection unit. The feature cleaning unit is used to generate feature vectors that match the structural stability assessment results of the X toughness shield structures.

[0066] For example, in the feature cleaning unit, feature vectors corresponding to U historical debris flow impacts from the historical debris flow impact database are obtained. Vector coupling is performed on the global historical debris flow impact feature vectors to obtain a feature arrangement of historical debris flow impacts. The spliced ​​feature vector and X structural feature vectors are then spliced ​​together to obtain a spliced ​​feature vector. The historical debris flow impact feature arrangement and the spliced ​​feature vector are then filtered to obtain a feature vector encompassing the filtered result vector. The filtering scores in the filtered result vector are used to generate estimated interaction coefficients corresponding to the U historical debris flow impacts. These estimated interaction coefficients are then used as the structural stability assessment results.

[0067] This embodiment of the application uses a convolutional unit in the stability analysis thread to perform convolutional processing on X resilient revetment structure descriptions that are inconsistent in data types, resulting in X structural feature vectors. The stability analysis thread also includes a feature cleaning unit and a feature anomaly detection unit. The X structural feature vectors are loaded into the feature cleaning unit connected to the convolutional unit. A corresponding feature anomaly detection unit is configured for the feature cleaning unit. The feature anomaly detection unit generates confidence coefficients between each structural feature vector and the feature cleaning unit, representing the correlation between the structural feature vectors and the feature cleaning unit. Then, Y structural feature vectors are selected from the X structural feature vectors using the confidence coefficients to generate a spliced ​​feature vector. This spliced ​​feature vector is loaded into the feature cleaning unit connected to the feature anomaly detection unit. The feature cleaning unit outputs a feature vector used to generate a structural stability assessment result that matches the X resilient revetment structure descriptions. This improves the accuracy and precision of the assessment, confirming that the resilient revetment structure can withstand the impact of large debris flows, thus protecting people's lives and property.

[0068] This application provides a flowchart of a method for assessing the stability of a tough revetment structure impacted by large boulders and debris flows. This method can be executed by a computer device. The following description uses the execution of this method by a computer device as an example. The method includes at least the following steps: S201-S206.

[0069] Step S201: Obtain description information for X toughness tank structure items;

[0070] Step S202: The X structural description information of the toughness tank structure is processed by the convolution unit to obtain X structural feature vectors. The X structural feature vectors are then loaded into the feature cleaning unit connected to the convolution unit.

[0071] For example, X structural descriptions of resilient tank structures are loaded into a stability analysis thread. This thread may include convolutional units, A feature cleaning units, A feature anomaly detection units, and another convolutional unit. The A feature cleaning units may include feature cleaning unit 1, feature cleaning unit 2, ..., feature cleaning unit A. The A feature anomaly detection units may include feature anomaly detection unit 1 corresponding to feature cleaning unit 1, feature anomaly detection unit 2 corresponding to feature cleaning unit 2, ..., feature anomaly detection unit A corresponding to feature cleaning unit A. Feature cleaning units may be network layers requiring multi-level feature processing, while feature anomaly detection units may include multilayer perceptrons and network layers with binary classification functions. Each feature cleaning unit is connected to a feature anomaly detection unit, and the feature anomaly detection units connected to each feature cleaning unit are different from each other. The A feature cleaning units are connected serially. The inputs to each feature anomaly detection unit are identical, and the outputs of each feature anomaly detection unit are used to load the connected feature cleaning units.

[0072] Step S203: Load X structural feature vectors into the feature anomaly detection unit, simplify the X structural feature vectors to obtain X simplified feature vectors; the simplification method of the X simplified feature vectors is the same; obtain the anomaly detection input queue and anomaly detection output queue that match the feature cleaning unit in the feature anomaly detection unit, and perform function processing on the X simplified feature vectors with the anomaly detection input queue respectively to obtain X anomaly detection feature vectors; the anomaly detection input queue and anomaly detection output queue are used together to represent the stability elements of the feature cleaning unit; generate adaptive data distribution vectors corresponding to the X anomaly detection feature vectors through LeakyReLU, and perform function processing on the X adaptive data distribution vectors with the anomaly detection output queue respectively to obtain X offset coefficients; project the X offset coefficients through the decision function to obtain the confidence coefficients between each structural feature vector and the feature cleaning unit.

[0073] For example, X structural feature vectors are loaded into each feature anomaly detection unit. Each feature anomaly detection unit can simplify the X structural feature vectors to obtain X simplified feature vectors. The simplification process can refer to converting the X structural feature vectors into feature dimensions that match the feature cleaning unit connected to that feature anomaly detection unit. The simplification method for the X simplified feature vectors is the same for all of them.

[0074] As can be understood, taking Feature Anomaly Detection Unit 1 out of A feature anomaly detection units as an example, Feature Anomaly Detection Unit 1 can obtain attribute description values ​​corresponding to X structural feature vectors, where the attribute description values ​​are the vector data size of the structural feature vectors. Based on the X attribute description values, the Feature Anomaly Detection Unit can obtain attribute coefficient queues corresponding to the X structural feature vectors, and process the corresponding structural feature vectors through the X attribute coefficient queues to obtain X simplified feature vectors.

[0075] The decision function can also be represented as a piecewise function. Obtain the first and second calculated specified values ​​corresponding to the decision function. If the offset coefficient corresponding to the structural feature vector is less than the first calculated specified value, then the confidence coefficient corresponding to the structural feature vector is set as the first importance parameter. If the offset coefficient corresponding to the structural feature vector is greater than the first calculated specified value and less than the second calculated specified value, then the second importance parameter is generated based on the offset coefficient, and the confidence coefficient corresponding to the structural feature vector is set as the second importance parameter. If the offset coefficient corresponding to the structural feature vector is greater than the second calculated specified value, then the confidence coefficient corresponding to the structural feature vector is set as the third importance parameter.

[0076] Step S204: Select the structural feature vectors among the X structural feature vectors whose confidence coefficients are greater than the specified weight values ​​as the Y structural feature vectors;

[0077] For example, the confidence coefficient includes importance parameters corresponding to X structural feature vectors. The structural feature vectors with confidence coefficients greater than a specified weight among the X structural feature vectors are selected as the Y structural feature vectors. The specified weight can be a pre-set value, such as 70%.

[0078] Step S205: Generate a potential importance queue using the importance parameters corresponding to the Y structural feature vectors, and obtain simplified feature vectors corresponding to the Y structural feature vectors. The simplified feature vectors are obtained by simplifying the structural feature vectors. The arrangement of the Y simplified feature vectors and the potential importance queue are processed by a function to obtain the concatenated feature vector.

[0079] For example, a potential importance queue is generated using the importance parameters corresponding to the Y structural feature vectors. The element values ​​of the potential importance queue and the structural feature vectors can be the importance parameters corresponding to those structural feature vectors. Simplified feature vectors corresponding to the Y structural feature vectors are obtained. A concatenated feature vector is obtained by processing the arrangement of the Y simplified feature vectors with the potential importance queue using a function. Specifically, simplified feature vectors not needed by the feature cleaning unit are excluded by processing with the element value 0 in the potential importance queue, thus selecting the simplified feature vectors that need to be cross-connected. Furthermore, the simplified feature vectors that need to be cross-connected are weighted using the importance parameters, representing the association between the simplified feature vectors and the feature anomaly detection unit.

[0080] Step S206: Load the spliced ​​feature vector into the feature cleaning unit connected to the feature anomaly detection unit, and output the feature vector through the feature cleaning unit to generate the structural stability assessment result that matches the description information of X tough abutment structures.

[0081] For example, the concatenated feature vectors are loaded into the feature cleaning unit connected to the feature anomaly detection unit. Each feature anomaly detection unit receives X structural feature vectors as input, and the output of each feature anomaly detection unit is loaded into the connected feature cleaning unit. The A feature cleaning units include feature cleaning unit Lz and feature cleaning unit Lz-0, where z is a positive integer less than A. The A feature anomaly detection units include feature anomaly detection unit Kz connected to feature cleaning unit Lz.

[0082] In the feature cleaning unit Lz, a feature vector is output based on the concatenated feature vector input from the feature anomaly detection unit Kz and the target feature vector corresponding to Lz. If Lz is a feature cleaning unit connected to a convolutional unit, the target feature vector consists of X structural feature vectors. If Lz is connected to Lz-0, the target feature vector is the feature vector output by Lz-0. If Lz is the last feature cleaning unit among A feature cleaning units, i.e., Lz is a feature cleaning unit connected to a convolutional unit, then the structural stability assessment result matching the description information of X toughness shield structures is generated through the feature vector output by Lz.

[0083] It is understandable that, taking A feature cleaning units, including feature cleaning unit 1, feature cleaning unit 2, ..., feature cleaning unit A, as an example, the concatenated feature vector output by feature anomaly detection unit 1 is loaded into feature cleaning unit 1, which is the feature cleaning unit connected to the convolutional unit. Feature cleaning unit 1 can generate feature vector 1 from the concatenated feature vector output by feature anomaly detection unit 1 and X structural feature vectors. Feature vector 1 is the vector output by feature cleaning unit 1.

[0084] Feature cleaning unit 1 can load feature vector 1 into feature cleaning unit 2. Feature cleaning unit 2 can generate feature vector 2 by using the concatenated feature vector output by feature anomaly detection unit 2 and feature vector 1 output by feature cleaning unit 1. Feature vector 2 can be loaded into the feature cleaning unit connected to feature cleaning unit 2 until feature vector A output by feature cleaning unit A is obtained. Feature vector A can be loaded into the convolution unit. Feature vector A can be decoded in the convolution unit to obtain the structural stability assessment result. Feature vector A is the feature vector used to generate the structural stability assessment result that matches the description information of X toughness shield structures.

[0085] Each feature cleaning unit can perform task processing on the input vector. To illustrate, let's take feature cleaning unit 1 out of A feature cleaning units as an example. In feature cleaning unit 1, feature cleaning unit 1 can obtain the feature vectors of the historical debris flow impact content corresponding to U historical debris flow impact content in the historical debris flow impact database. Vector coupling is performed on the feature vectors of the U historical debris flow impact content to obtain the feature arrangement of the historical debris flow impact content.

[0086] Feature cleaning unit 1 can concatenate the concatenated feature vector and X structural feature vectors to obtain a concatenated feature vector. The concatenation process can be as follows: if the concatenated feature vector and the X structural feature vectors meet the isomorphic feature requirement, then the concatenated feature vector and the X structural feature vectors are vector-added to obtain the concatenated feature vector; if the concatenated feature vector and the X structural feature vectors do not meet the isomorphic feature requirement, then the concatenated feature vector and the X structural feature vectors are vector-coupled to obtain the concatenated feature vector. Here, the isomorphic feature requirement means that the concatenated feature vector and the X structural feature vectors belong to the same semantic space. The specific process of concatenation processing in this embodiment is not limited.

[0087] Feature cleaning unit 1 can filter the arrangement and spliced ​​feature vectors of historical debris flow impact content features to obtain feature vectors covering the filtered result vectors. The filtering process can be as follows: obtaining the search coefficient queue, importance coefficient queue, and derivative queue in the feature cleaning unit; the search coefficient queue, importance coefficient queue, and derivative queue are all queues composed of learnable coefficients; performing function processing on the spliced ​​feature vector and the search coefficient queue to obtain the search vector; performing function processing on the arrangement of historical debris flow impact content features and the importance coefficient queue to obtain the key description vector; performing function processing on the arrangement of historical debris flow impact content features and the derivative queue to obtain the function processing result vector; generating the filtered result vector based on the search vector and the key description vector; simplifying the filtered result vector based on the vector data volume of the key description vector; performing dimensionless simplification on the simplified filtered result vector to obtain the filtering weight vector; and performing function processing on the filtering weight vector and the function processing result vector to obtain the feature vector covering the filtered result vector.

[0088] The concatenated feature vector is processed with the search coefficient queue WU to obtain the search vector U. The historical debris flow impact content feature arrangement is processed with the importance coefficient queue WK to obtain the key description vector K. The historical debris flow impact content feature arrangement is processed with the derivative queue WV to obtain the function processing result vector V. A filtering result vector is generated based on the search vector U and the key description vector K. The filtering result vector is simplified based on the dimension d of the key description vector. The simplified filtering result vector is then subjected to dimensionless simplification (RofAXax) to obtain the dimensionless simplified vector. The dimensionless simplified vector is then processed with the function processing result vector V to obtain the feature vector output by feature cleaning unit 1.

[0089] X structural feature vectors are loaded into a fully connected layer. The fully connected layer generates fully connected computational values ​​corresponding to each prediction unit, which represent the association between the prediction unit and the X structural feature vectors. Based on R fully connected computational values, the R prediction units are distributed. A target prediction unit is determined from these R distributed prediction units. The undetermined feature cleaning units and the undetermined feature anomaly detection units within the target prediction units are used as feature cleaning units. The processing of the target prediction unit can be found in steps S203 to S206 above, and will not be repeated here. The target prediction unit can output the original prediction result, which is used as the structural stability evaluation result. Optionally, the computer can also select and calculate several prediction units from the R prediction units. Each expert network can focus on different task domains, and each expert network can output the original prediction result. The original prediction results from each expert network are summarized to obtain the structural stability evaluation result, improving the accuracy of the business prediction structure.

[0090] This embodiment of the application uses a convolutional unit in the stability analysis thread to perform convolutional processing on X resilient revetment structure descriptions that are inconsistent in data types, resulting in X structural feature vectors. The stability analysis thread also includes a feature cleaning unit and a feature anomaly detection unit. The X structural feature vectors are loaded into the feature cleaning unit connected to the convolutional unit. A corresponding feature anomaly detection unit is configured for the feature cleaning unit. The feature anomaly detection unit generates confidence coefficients between each structural feature vector and the feature cleaning unit, representing the correlation between the structural feature vectors and the feature cleaning unit. Then, Y structural feature vectors are selected from the X structural feature vectors using the confidence coefficients to generate a spliced ​​feature vector. This spliced ​​feature vector is loaded into the feature cleaning unit connected to the feature anomaly detection unit. The feature cleaning unit outputs a feature vector used to generate a structural stability assessment result that matches the X resilient revetment structure descriptions. This improves the accuracy and precision of the assessment, confirming that the resilient revetment structure can withstand the impact of large debris flows, thus protecting people's lives and property.

[0091] This application provides a flowchart for evaluating the stability of a tough revetment structure impacted by large boulders and debris flows. This method can be executed by a computer. The following description uses the execution of this method by a computer as an example. The method includes at least the following steps: S301-S306.

[0092] Step S301: Obtain X example toughness tank structure description information and load the X example toughness tank structure description information into the original evaluation thread; the data types corresponding to the X example toughness tank structure description information are inconsistent; the original evaluation thread includes an original convolutional unit, an original feature cleaning unit, and an original feature anomaly detection unit; X is an integer greater than 0;

[0093] Step S302: The original convolutional unit performs convolution processing on the description information of the X example toughness shield structure, respectively, to obtain X example structure feature vectors, and loads the X example structure feature vectors into the original feature cleaning unit connected to the original convolutional unit.

[0094] Step S303: Load X example structure feature vectors into the original feature anomaly detection unit, and generate example confidence coefficients between each example structure feature vector and the original feature cleaning unit through the original feature anomaly detection unit; the example confidence coefficients are used to represent the correlation between the example structure feature vectors and the example feature cleaning unit.

[0095] Step S304: Select Y example structural feature vectors from X example structural feature vectors using the example confidence coefficient; generate example concatenation feature vectors using the Y example structural feature vectors; and load the example concatenation feature vectors into the original feature cleaning unit connected to the original feature anomaly detection unit; Y is an integer greater than 0 that is not greater than X.

[0096] The specific details of steps S204 and S205 in the exemplary embodiments will not be repeated here.

[0097] Step S305: In the original feature cleaning unit, the original prediction result that matches the description information of the X example toughness shield structure is generated by concatenating the example feature vector and the X example structural feature vectors.

[0098] Step S306: Based on the annotation results and original prediction results corresponding to the structural description information of X exemplary resilient bulwark structures, the model coefficients of the original convolutional units, original feature cleaning units, and original feature anomaly detection units are adjusted to obtain a stability analysis thread composed of convolutional units, feature cleaning units, and feature anomaly detection units; the stability analysis thread is used to generate the structural stability assessment results corresponding to the structural description information of the resilient bulwark structures.

[0099] This embodiment trains the original evaluation thread into a stability analysis thread, which may include a convolutional unit, a feature anomaly detection unit, and a feature cleaning unit. When the task domain handled by the feature cleaning unit in the stability analysis thread changes, the feature anomaly detection unit and the feature cleaning unit can be retrained to achieve adaptive changes in the feature anomaly detection unit and adjust the features connected to the feature cleaning unit. The convolutional unit in the stability analysis thread performs convolution processing on X resilient tank structure descriptions with inconsistent data types, obtaining X structural feature vectors. These X structural feature vectors are loaded into the feature cleaning unit connected to the convolutional unit. A corresponding feature anomaly detection unit is configured for the feature cleaning unit. The feature anomaly detection unit generates confidence coefficients between each structural feature vector and the feature cleaning unit, representing the correlation between the structural feature vectors and the feature cleaning unit. Then, by using confidence coefficients, Y structural feature vectors are selected from X structural feature vectors to generate a spliced ​​feature vector. This spliced ​​feature vector is then loaded into a feature cleaning unit connected to the feature anomaly detection unit. The feature cleaning unit outputs a feature vector used to generate a structural stability assessment result that matches the description information of the X resilient revetment structures. This improves the accuracy and precision of the assessment, ensuring that the resilient revetment structure can withstand the impact of large debris flows, thus protecting people's lives and property.

[0100] Based on the above, a stability assessment system for a tough revetment structure impacted by large rocks and debris flows is presented, comprising a processor and a memory that communicate with each other. The processor is used to read a computer program from the memory and execute it to implement the above method.

[0101] Based on the above, a computer-readable storage medium is also provided, on which a computer program stored implements the above method during runtime.

[0102] In summary, based on the above scheme, the convolutional unit in the stability analysis thread performs convolutional processing on X resilient revetment structure descriptions with inconsistent data types, resulting in X structural feature vectors. The stability analysis thread also includes a feature cleaning unit and a feature anomaly detection unit. The X structural feature vectors are loaded into the feature cleaning unit connected to the convolutional unit. A corresponding feature anomaly detection unit is configured for the feature cleaning unit. The feature anomaly detection unit generates confidence coefficients between each structural feature vector and the feature cleaning unit, representing the correlation between the structural feature vectors and the feature cleaning unit. Then, Y structural feature vectors are selected from the X structural feature vectors using the confidence coefficients to generate a concatenated feature vector. This concatenated feature vector is loaded into the feature cleaning unit connected to the feature anomaly detection unit. The feature cleaning unit outputs a feature vector used to generate a structural stability assessment result that matches the X resilient revetment structure descriptions. This improves the accuracy and precision of the assessment, confirming that the resilient revetment structure can withstand the impact of large debris flows, thus protecting people's lives and property.

[0103] It should be understood that the systems and modules described above can be implemented in various ways. For example, in some embodiments, the systems and modules can be implemented by hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the methods and systems described above can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The systems and modules of this application can be implemented not only by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., but also by software executed by various types of processors, or by a combination of the aforementioned hardware circuits and software (e.g., firmware).

[0104] It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects may be any one or a combination of the above, or any other possible beneficial effects.

Claims

1. A method for evaluating the stability of a flexible apron structure impacted by a large rock and mud flow, characterized by, The method comprises: X toughness armor structure matter description information is obtained, and the X toughness armor structure matter description information is loaded to a stability analysis thread; the data categories corresponding to the X toughness armor structure matter description information are inconsistent with each other; the stability analysis thread comprises a convolution unit, a feature cleaning unit and a feature anomaly detection unit; X is an integer greater than 0; The X toughness armor structure matter description information is respectively subjected to convolution processing by the convolution unit, X structure feature vectors are obtained, and the X structure feature vectors are loaded to the feature cleaning unit connected with the convolution unit; The X structure feature vectors are loaded to the feature anomaly detection unit, and a confidence coefficient between each structure feature vector and the feature cleaning unit is generated by the feature anomaly detection unit; the confidence coefficient is used to represent the association relationship between the structure feature vector and the feature cleaning unit; Y structure feature vectors are screened from the X structure feature vectors by the confidence coefficient, a splicing feature vector is generated by the Y structure feature vectors, the splicing feature vector is loaded to the feature cleaning unit connected with the feature anomaly detection unit, and a feature vector used to generate a structure stability evaluation result matched with the X toughness armor structure matter description information is output by the feature cleaning unit; Y is an integer greater than 0 and not greater than X.

2. The method of claim 1, wherein, The X structure feature vectors are loaded to the feature anomaly detection unit, and a confidence coefficient between each structure feature vector and the feature cleaning unit is generated by the feature anomaly detection unit, comprising: The X structure feature vectors are loaded to the feature anomaly detection unit, and X simplified feature vectors are obtained by simplifying the X structure feature vectors; the simplification modes of the X simplified feature vectors are the same; An anomaly detection input queue and an anomaly detection output queue matched with the feature cleaning unit in the feature anomaly detection unit are obtained, the X simplified feature vectors are respectively subjected to function processing with the anomaly detection input queue, and X anomaly detection feature vectors are obtained; the anomaly detection input queue and the anomaly detection output queue are used to represent the stability elements of the feature cleaning unit together; An adaptive data distribution vector corresponding to each of the X anomaly detection feature vectors is generated by Leaky ReLU, X adaptive data distribution vectors are respectively subjected to function processing with the anomaly detection output queue, and X offset coefficients are obtained; The X offset coefficients are projected by a decision function, and a confidence coefficient between each structure feature vector and the feature cleaning unit is obtained.

3. The method of claim 2, wherein, The X structure feature vectors are subjected to simplifying processing, and X simplified feature vectors are obtained, comprising: Attribute description values corresponding to the X structure feature vectors are obtained, and attribute coefficient queues corresponding to the X structure feature vectors are obtained based on the X attribute description values; The X structure feature vectors are respectively processed by the X attribute coefficient queues, and X simplified feature vectors are obtained.

4. The method of claim 2, wherein, The confidence coefficients include a first importance parameter, a second importance parameter, and a third importance parameter; the projecting the X offset coefficients by the decision function to obtain the confidence coefficients between each structural feature vector and the feature cleaning unit includes: obtaining a first calculation specified value and a second calculation specified value corresponding to the decision function; if the offset coefficient corresponding to the structural feature vector is less than the first calculation specified value, setting the confidence coefficient corresponding to the structural feature vector as the first importance parameter; if the offset coefficient corresponding to the structural feature vector is greater than the first calculation specified value and less than the second calculation specified value, generating the second importance parameter based on the offset coefficient, and setting the confidence coefficient corresponding to the structural feature vector as the second importance parameter; if the offset coefficient corresponding to the structural feature vector is greater than the second calculation specified value, setting the confidence coefficient corresponding to the structural feature vector as the third importance parameter; wherein the structural feature vectors represented by the first importance parameter, the second importance parameter, and the third importance parameter have mutually inconsistent association relationships with the feature cleaning unit; the association relationship corresponding to the first importance parameter is less than the association relationship corresponding to the second importance parameter, and the association relationship corresponding to the second importance parameter is less than the association relationship corresponding to the third importance parameter.

5. The method of claim 1, wherein, The confidence coefficients include importance parameters corresponding to the X structural feature vectors respectively; the filtering Y structural feature vectors from the X structural feature vectors by the confidence coefficients includes: structural feature vectors with confidence coefficients greater than a weight specified value in the X structural feature vectors are taken as the Y structural feature vectors; generating the splicing feature vector by the Y structural feature vectors includes: generating a potential importance queue by the importance parameters corresponding to the Y structural feature vectors respectively, and obtaining simplified feature vectors corresponding to the Y structural feature vectors respectively; the simplified feature vectors are obtained by simplifying the structural feature vectors; performing function processing on the arrangement of the Y simplified feature vectors and the potential importance queue to obtain the splicing feature vector.

6. The method of claim 1, wherein, The number of feature cleaning units and feature anomaly detection units in the stability analysis thread is A, each feature cleaning unit is connected with a feature anomaly detection unit, the feature anomaly detection units connected with each feature cleaning unit are inconsistent, and the A feature cleaning units are connected in series; the input of each feature anomaly detection unit is the X structure feature vectors, and the output of each feature anomaly detection unit is loaded into the connected feature cleaning unit; the A feature cleaning units include a feature cleaning unit Lz and a feature cleaning unit Lz-0, z is an integer greater than 0 and less than A; the A feature anomaly detection units include a feature anomaly detection unit Kz connected with the feature cleaning unit Lz; the feature vector output by the feature cleaning unit for generating a structure stability evaluation result matched with the X structure matter description information of the flexible armor, comprising: In the feature cleaning unit Lz, based on the target feature vector corresponding to the feature cleaning unit Lz and the spliced feature vector input by the feature anomaly detection unit Kz, a feature vector is output; if the feature cleaning unit Lz is connected with the convolution unit, the target feature vector is the X structure feature vector; if the feature cleaning unit Lz is connected with the feature cleaning unit Lz-0, the target feature vector is the feature vector output by the feature cleaning unit Lz-0; The method further comprises: If the feature cleaning unit Lz is the last feature cleaning unit in the A feature cleaning units, the feature vector output by the feature cleaning unit Lz generates a structure stability evaluation result matched with the X structure matter description information of the flexible armor.

7. The method of claim 1, wherein, The stability analysis thread includes a fully connected layer and R prediction units, R is an integer greater than 0, each prediction unit includes a pending feature cleaning unit and a pending feature anomaly detection unit, and the data categories corresponding to the R prediction units are inconsistent; Before loading the X structure feature vectors into the feature anomaly detection unit, the method further comprises: loading the X structure feature vectors into a fully connected layer, and generating a fully connected calculation value corresponding to each prediction unit through the fully connected layer; The fully connected calculation value is used to represent the association relationship between the prediction unit and the X structure feature vectors; Based on the R fully connected calculation values, the R prediction units are distributed, a target prediction unit is determined in the distributed R prediction units, the pending feature cleaning unit in the target prediction unit is taken as the feature cleaning unit, and the pending feature anomaly detection unit in the target prediction unit is taken as the feature anomaly detection unit.

8. The method of claim 1, wherein, The feature vector output by the feature cleaning unit for generating a structure stability evaluation result matched with the X structure matter description information of the flexible armor, comprising: In the feature cleaning unit, U historical debris flow impact content feature vectors corresponding to U historical debris flow impact contents in a historical debris flow impact database are obtained, and the U historical debris flow impact content feature vectors are coupled to obtain a historical debris flow impact content feature arrangement; U is an integer greater than 0; The spliced feature vector and the X structure feature vectors are spliced to obtain a spliced feature vector, and the historical debris flow impact content feature arrangement and the spliced feature vector are filtered to obtain a feature vector covering a filtering result vector; The method further comprises: Based on filtering scores corresponding to U historical debris flow impact contents in the filtering result vector, estimated interaction coefficients corresponding to the U historical debris flow impact contents are generated, and the estimated interaction coefficients corresponding to the U historical debris flow impact contents are taken as structure stability evaluation results matched with the X structure stability information.

9. The method of claim 8, wherein, The historical debris flow impact content feature arrangement and the spliced feature vector are filtered to obtain a feature vector covering a filtering result vector, comprising: Obtain a search coefficient queue, an important coefficient queue and a derivative queue in the feature cleaning unit; the search coefficient queue, the important coefficient queue and the derivative queue are all queues composed of learnable coefficients; The spliced feature vector is functionally processed with the search coefficient queue to obtain a search vector, the historical debris flow impact content feature arrangement is functionally processed with the important coefficient queue to obtain a key description vector, and the historical debris flow impact content feature arrangement is functionally processed with the derivative queue to obtain a function processing result vector; Based on the search vector and the key description vector, a filtering result vector is generated, the filtering result vector is simplified based on a vector data amount of the key description vector, the simplified filtering result vector is dimensionless simplified to obtain a filtering weight vector, the filtering weight vector is functionally processed with the function processing result vector to obtain a feature vector covering the filtering result vector; The spliced feature vector and the X structure feature vectors are spliced to obtain a spliced feature vector, comprising: If the spliced feature vector and the X structure feature vectors meet the same feature requirement, the spliced feature vector and the X structure feature vectors are added to obtain a spliced feature vector; the same feature requirement means that the spliced feature vector and the X structure feature vectors belong to the same meaning space; If the spliced feature vector and the X structure feature vectors do not meet the same feature requirement, the spliced feature vector and the X structure feature vectors are coupled to obtain a spliced feature vector.

10. A system for assessing the stability of a flexible apron structure impacted by a large rock debris flow, characterized in that, The processor and the memory that communicate with each other, the processor is used to read computer program from the memory and executes to realize the method of any one of claims 1-9.

Images